<%BANNER%>

Cross-Layer Design of Resource Aware Protocols for Heterogeneous Wireless Ad Hoc Networks

xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID E20101220_AAAABM INGEST_TIME 2010-12-20T14:23:18Z PACKAGE UFE0011611_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES
FILE SIZE 1872 DFID F20101220_AABJRJ ORIGIN DEPOSITOR PATH liu_w_Page_003.QC.jpg GLOBAL false PRESERVATION BIT MESSAGE_DIGEST ALGORITHM MD5
f70ff2e674f9c06ce1da6a05f1565ac9
SHA-1
69d77719e1eda5c1f04ba82354d835d233a849a5
2494 F20101220_AABJQU liu_w_Page_156.txt
bbd95de17ee26c093ba496b5a390a283
f37d9e768b7d4eb46fed03728b2448c5d7b21ec1
124218 F20101220_AABIOH liu_w_Page_025.jpg
4cc0c04403c3c9b05c4c29f1ab903c1d
b046c0b647e7c5eb4ea4a495f23ff4a6154df093
98272 F20101220_AABINS liu_w_Page_010.jpg
867437241f9f78366d5aa74ec46fedff
19d80292d0e82d2a07e281fab25c9057c8ccc45c
596 F20101220_AABJRK liu_w_Page_003thm.jpg
64edef2f035b1de006e07fb9e587824a
9675b4bad8fa3c3a4ab3b18471f89ac6e5806f88
2584 F20101220_AABJQV liu_w_Page_157.txt
7236dd26795c7611bf0a3980a9b2d4f0
fce0b277a72f3d0e5b5a5c7ca5e6620e013a82c6
111870 F20101220_AABIOI liu_w_Page_026.jpg
f1bf2ce6a74c083aff3cabb2ce35fbf1
31dbac2bdf84919fbf6832a1380d7d00d7ec8530
92336 F20101220_AABINT liu_w_Page_011.jpg
47546ff5d32cce6f48fdb2df8e742357
a08f1d74218a0d3cb8c08b08be143cb14c6cf3e5
28209 F20101220_AABJRL liu_w_Page_004.QC.jpg
479294b0e7935026842457564a4d8f95
e8625be3c0bf63c3e2f58913254be39207455a29
2641 F20101220_AABJQW liu_w_Page_158.txt
b2a3cdef16b6eb6c825a4ce8f4753e2f
032c608d186eed6be1e342fba23850a28b395f53
122385 F20101220_AABIOJ liu_w_Page_027.jpg
ed59352cae006c62d4fea7364ff3cd33
d26b58ab02f5d67beddf492d90478a6eb4770f3d
58949 F20101220_AABINU liu_w_Page_012.jpg
0a0a626c70ebf8ebfeaf9cba1f7f4753
927f53cb3be15608ff459ad8488fcde8de13aead
33635 F20101220_AABJSA liu_w_Page_013.QC.jpg
5c4f3d2cf31c39a32abb156440627dde
96ca1953d5954f6c21429fa96b19891929cc2cfc
2543 F20101220_AABJQX liu_w_Page_159.txt
dc8b31e449e8973efb0053d6c49ed63f
8b725175d8fc75e35eef73ca130dd64d08877729
104438 F20101220_AABINV liu_w_Page_013.jpg
cd2addd14ed84a723ab02ea64a0f90c0
8692a3b59b4fb84df6051fd01b0557074bec12fe
7463 F20101220_AABJSB liu_w_Page_013thm.jpg
5674e11a28ebcbf93982a7483272c250
a64640a6a27d4961a44cdcb36cd44b014b6ee7ee
6667 F20101220_AABJRM liu_w_Page_004thm.jpg
50fdd874fd2299672793982377f38c97
c93fc644875f2c14733b661251ba7e15752aa17f
2728 F20101220_AABJQY liu_w_Page_160.txt
5efc6e2cbaf06be0fa5e78dce34ea44e
fd200e8ac456062d31d1348c87bcdb63e3b165c9
122131 F20101220_AABIOK liu_w_Page_028.jpg
816e0e5db107d8e604fc1d291dc59775
86bdfa67d37d391508b4c460d4107161cf9ad1fa
110896 F20101220_AABINW liu_w_Page_014.jpg
af67f74ed97eb001c09c6e01d895e518
b66d1ea2b22047420e28c10fad8cd317086a587c
35678 F20101220_AABJSC liu_w_Page_014.QC.jpg
fbba53d00776ddd898f50e824b35def2
f338ada92a9996e80383a727ed66d1c6410700be
30512 F20101220_AABJRN liu_w_Page_005.QC.jpg
256decee89de2dd99112b83a52390929
d56423ee3ff909f80cfbbe5c0becbc6cbe3bcf81
2793 F20101220_AABJQZ liu_w_Page_161.txt
c763bcf01c64d85d3c359b5c6eea3c8d
9b5a1ada417db9e4268def870c560a4acacae530
88662 F20101220_AABIPA liu_w_Page_045.jpg
1a85779b6d45f117989c728dbab30125
594eeb840b6d8a7161302cfae45037aaaa1a0c12
117237 F20101220_AABIOL liu_w_Page_029.jpg
6f5cba0a8ecc4e76c45fd4ca14961378
c71dcbe219da74a4b5d29dcd36150d5220232103
125081 F20101220_AABINX liu_w_Page_015.jpg
0986532b9dbbe31b528823dc8db421c8
74d197bea37e8c84373e2d9fd98e3b117c3e889b
8130 F20101220_AABJSD liu_w_Page_014thm.jpg
d379f3d186b2bc8caf6f3fa356b00db1
5d3f427b6ccf31f33d87eea39e454288e023bb1f
6353 F20101220_AABJRO liu_w_Page_005thm.jpg
1277fad61b09a1b690f0efee37b3bd84
f0860afdfb74cd3ae0ea21d9b52a3aabbdb11748
63612 F20101220_AABIPB liu_w_Page_046.jpg
2385bae57949d77d4203c024973d60bc
efc608bcc817269cbe0b5e4b724158545269d6b2
97405 F20101220_AABIOM liu_w_Page_030.jpg
c7bf48ab57657cd6739c2bc813c6b660
3b5d8252d4874db0174b0414554dea47262705b4
124008 F20101220_AABINY liu_w_Page_016.jpg
894a3e4c3f4c402d8b61bbc179d20e5d
1eef0848b37775e70d90f9db7951b2fe89861612
39664 F20101220_AABJSE liu_w_Page_015.QC.jpg
4520763ca2abe2c534f1151468cb6d8c
23d8e42633cb06b877b4c3529ab2fa8d91654f31
38124 F20101220_AABJRP liu_w_Page_006.QC.jpg
a1b0be75fd87ed6234d18a50421384ae
a7987a764b5238a42759e5be32ab41b202b02353
114202 F20101220_AABIPC liu_w_Page_047.jpg
d3c5216bd57ac10f793fb1ee051275c6
ec4d6094f26e8f5f4837f9e5546261e6530c194a
120211 F20101220_AABION liu_w_Page_031.jpg
e516430b1e8dc05a5d707e8f89fe4def
dd4a9aa12d4e46cbd8567822c660283b0fa1d697
118731 F20101220_AABINZ liu_w_Page_017.jpg
d09cf29f88d905ff29c2e498964a6f02
73e54e2c4b1ceadb14d253c8ebda57044e077af6
9156 F20101220_AABJSF liu_w_Page_015thm.jpg
0f85803d236f52a8c8d35442c4066a48
49bbf4fd471be5b14f66215a22bd11c3e5129971
23058 F20101220_AABJRQ liu_w_Page_007.QC.jpg
b2f81422c9dafad0ebd48e2cf87a2eaa
9f6e9e632100041203c0042270cf95b74c48e2b6
113144 F20101220_AABIPD liu_w_Page_048.jpg
10b83154add554738e52884974130f7a
a8e610072f7ebcb006416cfcde295987967c7d71
114978 F20101220_AABIOO liu_w_Page_032.jpg
42cf84b45a9c461eee4ed6f1c7ef263b
2760c7d344113e9a350683079fba8f6af639d064
39483 F20101220_AABJSG liu_w_Page_016.QC.jpg
9d075d129c3d34467c34262864357434
76c45d1e72d7d31174c621e73e64e810c4426024
4804 F20101220_AABJRR liu_w_Page_007thm.jpg
e6097ca0525a73f5b9fbc89f42838dc6
24fce862c75b9f229d5b8f198fe03abf133b3eef
92435 F20101220_AABIPE liu_w_Page_049.jpg
16e91fd81f16bdb65e3e1262a26745c8
7aad9fb75a12ee4d7a50b33bcef58175c74a75ca
92700 F20101220_AABIOP liu_w_Page_034.jpg
83f2a2e7a3a1e3c965f13a8aeeac8b2c
3fa7839c0d56c5a9ff23214f1a3239ec6dc37477
8950 F20101220_AABJSH liu_w_Page_016thm.jpg
0a7b1df1450219d145e267c5be61d00e
d370832cc53423b63e809be14174d0f58992dfde
2846 F20101220_AABJRS liu_w_Page_008thm.jpg
401e51aa09305351b483ffdda7908e25
752aace48305de608fc797212cc0d77129144ab3
27451 F20101220_AABIPF liu_w_Page_050.jpg
96ccd31f127a69fd85f117e75ae2c88b
3f38a7f822ee21604f023b309f9c8b4823b5a5bc
70042 F20101220_AABIOQ liu_w_Page_035.jpg
9d1aa797ee17bca9b0f1f98c72261077
192f6557569d193a18de3c6f49f110acf5d75fdc
37597 F20101220_AABJSI liu_w_Page_017.QC.jpg
5cdfec3e7646deae2eded424be0dcee7
48383048d7b15107f786d832307ebe857191e3b4
31706 F20101220_AABJRT liu_w_Page_009.QC.jpg
4757278363e0e7661d93a2d6ed9117c2
e42900b501d5da7048725f2109bd66fbf89fad10
119462 F20101220_AABIPG liu_w_Page_051.jpg
2fbf61460acf96bf82cde78a743a30f9
5878d8a7c04544e3cec752910e2fe2e3b25a4d24
124705 F20101220_AABIOR liu_w_Page_036.jpg
efa3384c14efd8d7902d7df9bce4a3be
b04f3e629ac1979dd9db0644bda4ba3e0f807d8f
8786 F20101220_AABJSJ liu_w_Page_017thm.jpg
bfcd46b67f2a0d47793a126ae70f0c65
10013ad91fd414c3e986bd431787a26491df1019
7541 F20101220_AABJRU liu_w_Page_009thm.jpg
b8202cccb2a2d4a75e55c0c8a8163dd3
44ad64934d53b9e9976109948c2273d9faa60748
107936 F20101220_AABIPH liu_w_Page_052.jpg
f3e067e34f089ebc1bd3226c38851980
ca7b7701acdc05b5cd1888047937a3fa93fff27c
103768 F20101220_AABIOS liu_w_Page_037.jpg
f4a45b6b8335394cc078bc0c5b22bddd
90775bf176eb43c177ec7c63175c4c843cc5e790
39654 F20101220_AABJSK liu_w_Page_018.QC.jpg
5633e258d423e3a09996f21a5d4b1cbe
af07a3f6b80c87af81b88a3a9545a66fe98a67f4
30383 F20101220_AABJRV liu_w_Page_010.QC.jpg
4a5c67fd1ed1b57f7c00ab5e1cb24877
e2611592e3b42374129b36f85f5ce3450b42e980
114149 F20101220_AABIPI liu_w_Page_053.jpg
d2acd87c9db4e34471efc60dfdefd02b
929ae5006a604016380b04807653e352dfe8df1c
109987 F20101220_AABIOT liu_w_Page_038.jpg
87bce6cd14113cc333eb617db5533a92
1df3889c932fd56766c4bece27ff278c1e6f8041
8941 F20101220_AABJSL liu_w_Page_018thm.jpg
6680d44a08fb63fe550f25d9431a6699
da03f761a55159571d212ae949753f7cc3a5fb1e
6929 F20101220_AABJRW liu_w_Page_010thm.jpg
7c4383f519e9d56a9c06daacd0b19c23
3ec3469dd31e424a6eecd740b17529a23a73dc8e
120915 F20101220_AABIPJ liu_w_Page_054.jpg
ce6c7eb6a1127a2b30bd68233a12ea30
d888ab65f3628057832ed61cd469ce5fc6536e1d
98135 F20101220_AABIOU liu_w_Page_039.jpg
fe57432099c5dfbad3f08b305925cb1a
bc2774cb0a0b855b5c11c4ff99e733982da796fe
35470 F20101220_AABJTA liu_w_Page_026.QC.jpg
2744879c7c7f8bb098fb3588f99beffe
2ba30ea23189ecffb8b0271e5111b111bb9ead09
36452 F20101220_AABJSM liu_w_Page_019.QC.jpg
698db0ca9f1adb17ac0c27845971b813
ca663ab414443b9376c92c4616ba74599855a879
28558 F20101220_AABJRX liu_w_Page_011.QC.jpg
878b4e401de7392a2f92af9f44d7333a
4cf705764eb00190bfbf07cc9e9dfa60f82b9d0e
34513 F20101220_AABIPK liu_w_Page_055.jpg
8ecc4ea745e3e5cb26db2cf2ef6d0c8d
ae3751387e7eb3022223a081cd8ca7422d2185bc
101562 F20101220_AABIOV liu_w_Page_040.jpg
2d9df8cf019d44cd7f3a079ff525d3a4
ea7a3429aa00efc88bce4eb4a33b7d74ab9f453a
8397 F20101220_AABJTB liu_w_Page_026thm.jpg
ec9cd2e2cca94fbc3a9734911bf11715
6be4d3029876546d089be6880c697e36781745e9
18740 F20101220_AABJRY liu_w_Page_012.QC.jpg
ab4162a19403eb3e19125aba1b589846
938b8feb9a6bb42435e237fffa3239f5beba2d55
118392 F20101220_AABIOW liu_w_Page_041.jpg
4f02b143169ec0af414a5e23e7f41500
48b7c28bfaa9342e9bc3d92008566626f26bcf57
38354 F20101220_AABJTC liu_w_Page_027.QC.jpg
222ce41e9b91687b11804c2742070a95
0563b88fb1d544da8614fdc83a99f2cc8387e00c
8621 F20101220_AABJSN liu_w_Page_019thm.jpg
ac8b5bafb8c0058ac15467bf01c572d5
8b086aea3a2cdf2ff3d00f3e29d84a671f2ebad0
4466 F20101220_AABJRZ liu_w_Page_012thm.jpg
1debbf2bfb720e5b02dbb24267ea5240
d2e5c5f0623eddb029b3205a55eabfb7c032c0f6
119516 F20101220_AABIPL liu_w_Page_056.jpg
624c61cb3daa518db9abe66fe38aef2c
3bd00a5148368649745c8aece8ad7df606c2b5f1
100516 F20101220_AABIOX liu_w_Page_042.jpg
b448550aa2524434c8e2242158bc383d
fdba7cdfd1d14ae5741eb629476d3da63fe29451
96224 F20101220_AABIQA liu_w_Page_072.jpg
258f27ddfb62ec99c909607255a79aa0
3866d7ae68757268ad74aeee2f106696f52a6fa3
8521 F20101220_AABJTD liu_w_Page_027thm.jpg
85153968671b411c988b6c18836ad32b
129cb7ef0262c520c10ebc394cf91b8f29af2bbd
34917 F20101220_AABJSO liu_w_Page_020.QC.jpg
4903cb23e6ca3473fb86d107bf832055
ca9b6d8b569349884ffb914b9cd7d661f8188e75
36660 F20101220_AABIPM liu_w_Page_057.jpg
64eb38912f0ffc81d77ed30f8f3e4799
60e8e8ef796c89d218b087b39bb1762755d07ce2
121371 F20101220_AABIOY liu_w_Page_043.jpg
b6874dfab8b966d2431c41300e3df3a8
f66835546fbdb38e54f4a0d489900e277578e67f
99720 F20101220_AABIQB liu_w_Page_073.jpg
cac3d0452331a4cf9788a946571366b3
3168903b67f3407481c873e4c15ddd9235a6f569
38870 F20101220_AABJTE liu_w_Page_028.QC.jpg
10eec7a2bfe122aea5db102de368d1d9
e6378c8ab5e57d642fc9dbe6d591a2768a0b1d22
8333 F20101220_AABJSP liu_w_Page_020thm.jpg
3c25ecac80a80e7f5f805bedb25a514f
82c8a7f2f2524dfaa56239dfac587e6fe3d614b3
64466 F20101220_AABIPN liu_w_Page_058.jpg
bfe5f297d3ba28afde88e6d7cac05ee4
20d9abe58582c742beee78d39955d1a15a214093
90340 F20101220_AABIOZ liu_w_Page_044.jpg
c1f23af581b383fb8ba719407c2bd587
dca55a46925a17a88ac2390f6d79f6ae99fcb667
44279 F20101220_AABIQC liu_w_Page_074.jpg
5167f7947ee0c7dc9b008e7ce4cab515
1a868dfade73fe5a02c228d2b35bd60709407937
9086 F20101220_AABJTF liu_w_Page_028thm.jpg
456fc88da9d702ef39273811252ecf24
efd2deed3271b53967dcbe0c826bb04c3da6b690
38382 F20101220_AABJSQ liu_w_Page_021.QC.jpg
9c77e834394ffd38e73cc8827bbe6324
9ac69b2851f3d49f29edb9a9a574e7465d3584f9
83763 F20101220_AABIQD liu_w_Page_075.jpg
3d6e5717231dba7afa24a36393e69bee
dc15d13b1867eb2fcabb930322735b6dac38439d
109475 F20101220_AABIPO liu_w_Page_059.jpg
e97c072a607abd287b3439d0ff372d25
63054c430913da3e0db8b94f5bae1deea04e5166
37091 F20101220_AABJTG liu_w_Page_029.QC.jpg
716924611b729e701b75228b0810aa56
1826c675dec34bc7cdf98dc9c674a5bfce691837
8974 F20101220_AABJSR liu_w_Page_021thm.jpg
990a506446428afb8d7632a596efbf4f
6535da1be30f8d2f082c77f35587ca858440dc96
98981 F20101220_AABIQE liu_w_Page_076.jpg
56830a014b0412a831e1c8f6cb7b8995
dd9d1154d80a01cf0d7050f4a99bebb3a0ca009a
59709 F20101220_AABIPP liu_w_Page_060.jpg
3108191b5548ed275eed95f3d8264049
e59efedba6fcc8f21eecd4f5485d042c2c24f216
8445 F20101220_AABJTH liu_w_Page_029thm.jpg
c367e475b28a0754caa213391349677e
739f36b635f6c7054cc7b68d909d63c5ce3afca6
9453 F20101220_AABJSS liu_w_Page_022.QC.jpg
b275ad57f7a0644f96349312bde888fb
16de63429b783fa477e361c5ecd2718aac9f64d4
122644 F20101220_AABIQF liu_w_Page_077.jpg
55958bdbe833f04422c43c276a7fd40e
070b346f0a2b6ca761c82cf16ca923f01f30bf21
104193 F20101220_AABIPQ liu_w_Page_061.jpg
e7540daf65596984b0443e6fef8eaebd
20cccdb9fd0362c03c9f3d8eae0defff61ddd26b
7559 F20101220_AABJTI liu_w_Page_030thm.jpg
6fb41b9e7be67d393656b806ad4b286d
7f4477d6e7878c0f63b362b6ec5118c2c5e5fae7
2453 F20101220_AABJST liu_w_Page_022thm.jpg
1cafe819a2b45d4d5ffb07992f9a1d02
81ae36e0ee71d565b1545e774b52f2d10d05f2ca
108052 F20101220_AABIQG liu_w_Page_078.jpg
16202f9cc7f2194d160b079ef0926d87
21d56608b5f0d7cc44c3a78f47fb1fa2f81e56bc
117243 F20101220_AABIPR liu_w_Page_063.jpg
e605c6332b12d334be079757899a147b
102c150be9199428e58c369538d841012a0d4d1c
38339 F20101220_AABJTJ liu_w_Page_031.QC.jpg
274cba06b69bb73000419bf9d8223bd5
87409f75f0897538205b15a3d77fe044cf4065aa
31822 F20101220_AABJSU liu_w_Page_023.QC.jpg
77e7c6e3965d54bcabe065c8d3a4fe0e
ad6074845e7b1d9127a0daab7e71354d9a845179
95702 F20101220_AABIQH liu_w_Page_079.jpg
5e57afeff42d3c9f61c1b6ab8b2da51f
79b74476d75f28066b20e03ae546641de9eb8e36
118064 F20101220_AABIPS liu_w_Page_064.jpg
0eac703f8123b752d39503fbd860eff8
5f6322e2172c572a2f158e01acfe4548b3177df4
8569 F20101220_AABJTK liu_w_Page_031thm.jpg
faf6b84b7b2718d342181b7cf3803eb9
1f5022c88178f9e71f541a8c6d6869f9abccd405
7584 F20101220_AABJSV liu_w_Page_023thm.jpg
496cf178147d9ef259ecaef82057096c
594d5b1adfcb75cbd3d378003f6131d8f293e2fe
122472 F20101220_AABIQI liu_w_Page_080.jpg
1375f017bbe94c837a3d62227c12b6b5
4b2680def8ea9d9443a4055012d570bf55c3ab9e
125432 F20101220_AABIPT liu_w_Page_065.jpg
4f3bb1435689776054274cf9ea39d951
2123b2065a5bffa0bc55e595ab0758492e03779f
35271 F20101220_AABJTL liu_w_Page_032.QC.jpg
4482ae989ec03868d4c70dd9ffc122ff
276aeb4202b8431ba038e4d3bd9e66bb1cb24d88
36244 F20101220_AABJSW liu_w_Page_024.QC.jpg
86fb8f4d841de1eed5d7ee3f425842e5
3c1729a0d26f6e3f7910ad2b8fe7fdc3e5e96138
119855 F20101220_AABIQJ liu_w_Page_082.jpg
e6ba74787d0afe5dbc7a7afbad784c89
6c024f60f77453ca862c235f5e238fc74bca2d54
116845 F20101220_AABIPU liu_w_Page_066.jpg
b220920e4bca44639ad0cd6cd9f1fe54
a3ecf8c89dfb499d73d17a3094c6cfb06de0afaf
31714 F20101220_AABJUA liu_w_Page_040.QC.jpg
e33fc2707512f0678bf7f7553e0ac4eb
ecde140aee94f6ea8eb2c2ba240b1efe9c957c8c
39348 F20101220_AABJTM liu_w_Page_033.QC.jpg
72d08dc406e65168c41d4a3e8bf34800
a5f02448b5e41c6c8e3c6cf54c045f7d5a5601d5
8517 F20101220_AABJSX liu_w_Page_024thm.jpg
2af22b0f56ee4d83ae0b60baa3041889
07ac8c4ae5e156125411a4f69b03802ec4730eca
54694 F20101220_AABIQK liu_w_Page_083.jpg
0b3db6d75a42cc787c60057903a3c36a
f53bb2511220953c498191905ca9991bd72131ee
81393 F20101220_AABIPV liu_w_Page_067.jpg
34b2d8a7c84c288b88d04ddb4556d23a
9b9148e64ee6c33b3f500428d6133702ad15c478
7865 F20101220_AABJUB liu_w_Page_040thm.jpg
a271a7c4a6680c06024aa46049c57c7a
b224c8da84340317b225581015bb2a7b9dc513e1
9181 F20101220_AABJTN liu_w_Page_033thm.jpg
c797eab0e3fd4d44b653c1a30f1b7de5
de932d3d369a54b991ac393d217ae96a31107648
39421 F20101220_AABJSY liu_w_Page_025.QC.jpg
665193b9c0d5a77175e4e369a848ea62
115f512b0e49d93758a1e84ac04eddef9b3eedb0
105650 F20101220_AABIQL liu_w_Page_084.jpg
9ce9ede8452e7f523e4c2e3781181119
a2459b95c09beb504ecadaa9c9a04a99c0eae205
112925 F20101220_AABIPW liu_w_Page_068.jpg
09e581f77791a9da65eb8bd44b4339a1
78edc6731aaa0f302042a2b6c2d487ca7e4cef78
37420 F20101220_AABJUC liu_w_Page_041.QC.jpg
3acfc501c9f742ea1e10678c2c49055c
41820904344c5f6c7e8c519afb8b6e7192885f6b
9196 F20101220_AABJSZ liu_w_Page_025thm.jpg
1f8d6cf22b0d2c1912912f1634bb66d5
d8dca1ac46f75b809c23b09c92bc8854cbd0f8a2
129668 F20101220_AABIRA liu_w_Page_100.jpg
5b5ef7313b6580dde4c90e52d227e992
a76a4a9e2bfa0c1d817ef14a4f647356e2644fcd
123730 F20101220_AABIPX liu_w_Page_069.jpg
ccdecb8a8adaea49b637c67856a1393f
45e1b73579eb8c7ef6620fb4ed617bf455f60f55
8870 F20101220_AABJUD liu_w_Page_041thm.jpg
7a56cae639fb0c08ba0f075ce865b5aa
bbfc887f83f88c6624470e8a6e9ced414ebc22ac
30129 F20101220_AABJTO liu_w_Page_034.QC.jpg
787f7b1492b8d84aa168d23485387dad
e59bca8a6d431616bdc376831b577dfb19864fe4
122464 F20101220_AABIRB liu_w_Page_101.jpg
e55b7acd87d3ee6900f3cf7055540764
3216b65c7a64d23efed0337453e1a872f74f0ad7
117649 F20101220_AABIQM liu_w_Page_085.jpg
51625282dac64cb49d4eb1fec696e1ab
5c9484085d7f7476820416536a27448f3f486dbb
102645 F20101220_AABIPY liu_w_Page_070.jpg
fc6cb2a4a0df5f3be3e40455cdd494c9
fab56c53d2e36d7adec6e09cf626f030d34eee71
32514 F20101220_AABJUE liu_w_Page_042.QC.jpg
d880a89475fc476ac5303a6e03a82ba4
35681cfed729318df46c7811e78679fb9bdef801
7479 F20101220_AABJTP liu_w_Page_034thm.jpg
dbfcdd53647e4c053bd80a670fed40a9
661952b207a41fa6c251d09a4540d2ea3142904b
68825 F20101220_AABIRC liu_w_Page_102.jpg
31b8c208ec7060ed74e5cb15e423e56e
a67d4bac2c04ae7a2ee50033da72de1dec198155
124052 F20101220_AABIQN liu_w_Page_086.jpg
4db602107abd50c6315cd34fc0d5b5da
1dcee23c677a814c98067d5097328e2cb8b7967a
67195 F20101220_AABIPZ liu_w_Page_071.jpg
971b026b31b81c2588705bfe09cdc9ff
9cfcc2c540b1a5aba3a38bbc91e2778e0eb26d57
7922 F20101220_AABJUF liu_w_Page_042thm.jpg
a3c1cb13864e76987d4689d60f362876
d3a649a23b20b9ad1bc06b76516aab48a1be3d10
23209 F20101220_AABJTQ liu_w_Page_035.QC.jpg
d864a71a4256025113ac659bed1646e4
fd32febe076e2bb4404a37a70bc128e8daf8f9fe
115822 F20101220_AABIRD liu_w_Page_103.jpg
95209f3870f216376726b8c046147100
fad969e12fc52c883c597c832c7983512256d180
128931 F20101220_AABIQO liu_w_Page_087.jpg
ee66753f1eacc2b54d4906addc557178
ff029cc8dc62abae695e640777dbb58139d41372
8094 F20101220_AABKAA liu_w_Page_123thm.jpg
170654e6bc8036ec823c556386cf55e5
799910dd743a24cd05adcfb92ddf9632dda6e75e
38225 F20101220_AABJUG liu_w_Page_043.QC.jpg
17d167d678cc65ce1146e1ba3056928f
7a9310ce3b24b3f1cda3447d6792edb2ea861263
6178 F20101220_AABJTR liu_w_Page_035thm.jpg
512229cd54739216f823d4d1ae1c4dd8
fbc389aaf795cca207fd4731e98355700280f7a2
118911 F20101220_AABIRE liu_w_Page_104.jpg
2058a6bd010e9f2d6b43bcf01472adb5
ca627f58447afe6989bd7ee6685f128bfd18dc2e
120006 F20101220_AABIQP liu_w_Page_088.jpg
540cab669bb2b7fc4b9074d1d0b8be91
409421acfb3a7de32838bb0efa53b2145d26cf7b
34660 F20101220_AABKAB liu_w_Page_124.QC.jpg
91c53ca791b9217ec1cb7da7bc9ab2ab
bae6faaa689ff46b5457288ac46b80bfa219015c
28729 F20101220_AABJUH liu_w_Page_044.QC.jpg
2de131e8d799f817dd78d54efd15bb35
d9c1dc5803113a5a774f052da0089831e14487c3
39852 F20101220_AABJTS liu_w_Page_036.QC.jpg
1379461433b15e5fbc3455c57e99e378
2eb242c51a9078e33dfef8d40f161d6fdb594dd1
115050 F20101220_AABIRF liu_w_Page_105.jpg
72e7f7d643feb087c27212017cb71c7a
68efc139e263ccb712bbce2e2b891de8a43cf9db
96289 F20101220_AABIQQ liu_w_Page_089.jpg
2c8513fecfb564e14fc67a516a2ed8de
689ca704a0f2bebdaadbe7dac36197ee541ffa6f
8600 F20101220_AABKAC liu_w_Page_124thm.jpg
e3f24e86c76352289d2559cbc90536e6
089ef6bb6599897bc09b6fe534201841f2978a28
6845 F20101220_AABJUI liu_w_Page_044thm.jpg
6da7e39ca61898dd29ac3e4f76aafb7e
e2d9ec79319fa22a7270669031955f813c1b551b
9292 F20101220_AABJTT liu_w_Page_036thm.jpg
28512611440c44b61ab0a22bffd955b7
8bee477d19774cfb05c6cbd62155bcef91c9e8f2
109093 F20101220_AABIRG liu_w_Page_106.jpg
a390432555ee21dc3170af66b57aa459
8b2c24b8f20e97c9cbc1184990d75721fce39d0a
124064 F20101220_AABIQR liu_w_Page_090.jpg
a889479c3794b695823568e71af90a9c
b8b5a27b4f130119d9e9f4942de6682ffc76834f
39451 F20101220_AABKAD liu_w_Page_125.QC.jpg
8fe37e1045295d789c948f57cd876c1a
ec53abfd64f7fe74cfb416725d647eefe66532fe
27918 F20101220_AABJUJ liu_w_Page_045.QC.jpg
5ae17b0ad24dc11335a4ae12c8c75c11
32ff4ec774a0d479a47470d4c8a8290fc871aa05
33066 F20101220_AABJTU liu_w_Page_037.QC.jpg
70e197ce667791f78c1e5c69f9614649
1fa80e024748fb9d7e56f71fc7e173c16fc98b09
119370 F20101220_AABIRH liu_w_Page_107.jpg
80a23012d9df0d7dc543dccb23b74581
50f964c47572627e7221cbab215835ae64b85226
97932 F20101220_AABIQS liu_w_Page_091.jpg
3521fcf548ddb465528fb2df05f7d6be
017829358069873796c7bf8a21f4ad8ebf546c07
9000 F20101220_AABKAE liu_w_Page_125thm.jpg
f4836b807df19359e973cf195fef941b
3f9bcb727471e0f09d34251ed5b5c157f76acb14
6940 F20101220_AABJUK liu_w_Page_045thm.jpg
ae8cf639ec78c011592163c7768b4a40
f30f2db256c7e9c6387dd110b1d13f2a5c86aa90
7605 F20101220_AABJTV liu_w_Page_037thm.jpg
ec5fafc22d94e33c9315baf792dc722c
c6eb41bdb5065547b17798009bab1ea43e7f9a08
74639 F20101220_AABIRI liu_w_Page_108.jpg
ea38c8a0ca0715783198da9f030525cd
b8762b43c87577bcf1768fe1b0ecaf599743842d
92190 F20101220_AABIQT liu_w_Page_092.jpg
e7e1a8e52252334e2084fe98e37f10a6
11f60b8a19de7a96815ede396b4cf04dd186c57e
37491 F20101220_AABKAF liu_w_Page_126.QC.jpg
c511227b57cff37eb7295ef4f73a76e3
87a799ccafbdf731015c9e9e903f25f3d5f49917
21687 F20101220_AABJUL liu_w_Page_046.QC.jpg
6aae6a1d8eb4eeb6022c2fd8fd781212
072a4b8f8769920c3e5eada53526d34ac327d425
34442 F20101220_AABJTW liu_w_Page_038.QC.jpg
85b71d6f3172de5df7bc530894cc613b
d93d049aed03c1bd913ab21e3365680674006342
82755 F20101220_AABIRJ liu_w_Page_109.jpg
256a72b5a1a20a19b47efd40374c268a
8c54668620831664516e360eb005229c4069a645
68229 F20101220_AABIQU liu_w_Page_093.jpg
fc45e2a9cdd2bac7a39bfd345e5ff9fc
03b5f101d0e102188ef5db94acf7e6e03760b1ed
8646 F20101220_AABKAG liu_w_Page_126thm.jpg
74a5500dbfad37c687f71a17e2062cf6
f4c37b5af4e557379047a1547e7996a8dfda20c0
9223 F20101220_AABJVA liu_w_Page_054thm.jpg
d98d7017ea57a977445364b6c4e1a334
e72ee04241682e1aa15dc078005cdbbb877ba024
6172 F20101220_AABJUM liu_w_Page_046thm.jpg
b13426b1349bed20c83cd47ba02f9f92
bbd02f12f79fcb46df8f018e58c18cdd33dba947
8132 F20101220_AABJTX liu_w_Page_038thm.jpg
e8137618d6b826c0b8e640577d6f4059
964b91dad27161a0fe94a52e85e995e72ea772c2
118644 F20101220_AABIRK liu_w_Page_110.jpg
f16a92a7e622081c50592719bb61acc5
9526737a229858efd71a7a9c902d7a5ae20eb996
108994 F20101220_AABIQV liu_w_Page_094.jpg
de8b53b3c848e8be98f493ce30d614f6
e3e1151d35cffa85c706c14c771f245b819f8a15
36446 F20101220_AABKAH liu_w_Page_127.QC.jpg
cf1ab2348a9ee492fe3359c7a4838509
a4e26e24cf5e891311671fadd045e79c5f4c750c
11612 F20101220_AABJVB liu_w_Page_055.QC.jpg
f69e53920d127f4343581f4d4b912a4e
4c57c9f7e9ea9ea3dabb7e0778ecb53bba576d16
36120 F20101220_AABJUN liu_w_Page_047.QC.jpg
7bc2b7617631df8ccf6910746b919ad1
ad23988edaac7babbd240039bbd829bf3316b770
31733 F20101220_AABJTY liu_w_Page_039.QC.jpg
5825773258a14c44c0087471a8f7988d
68fb8f88bba68aeeaa55a9dac8e6f48f988d5d85
116424 F20101220_AABIRL liu_w_Page_111.jpg
f526e1ec6137db7e64266c57e579e9ff
c6ae641ef61773cedf4e4cae201ab699d8b4f970
88608 F20101220_AABIQW liu_w_Page_096.jpg
f23be8e21d1524c19a918eddac989a6a
b92faec3fde082d6d089f83ce83ff3cf90cd9f6c
9542 F20101220_AABKAI liu_w_Page_127thm.jpg
a958bcc005a5f01c3dc6ff6bdfd3875e
feabe201a591f4f0c98df90d768f5eb981cfa869
3395 F20101220_AABJVC liu_w_Page_055thm.jpg
222fc944c02a4ab0e67b92bb42c34547
ad314705d0b7c300d95d9ec511d909023ea45fbe
8860 F20101220_AABJUO liu_w_Page_047thm.jpg
3ea08c4c69e90176e0f24cb0eed868d6
0f1b951333df9a1dd4fe5cda0bedcccf89ddb156
7391 F20101220_AABJTZ liu_w_Page_039thm.jpg
de77f538e39c2555f38c4a1ee6439b4c
62cdc4aa0233ab06f830bafcc83a07944ffcb2d0
97195 F20101220_AABIRM liu_w_Page_112.jpg
08df815f3306cb4d199e153db6c5a24a
ac6bfcf83a16f41da77c994a2792ac26439340be
125183 F20101220_AABIQX liu_w_Page_097.jpg
593fad9014acb4a4cbd411a9e77a17c5
3a1e9ceff0cd150779a48bc3c23ce99add93ab9e
120388 F20101220_AABISA liu_w_Page_126.jpg
cf33689d354218619c8cbf4ff83bdeec
ede6a3ad6157de9ca785b6544214adf103594350
37089 F20101220_AABKAJ liu_w_Page_128.QC.jpg
3d7af597359ae33e2f3812a02ce8843f
a56f17f624f781ed1a480bef5cfd65f8fa16c76d
37803 F20101220_AABJVD liu_w_Page_056.QC.jpg
fac1ba752d9b7d8905a95039f32924d4
e9659144cffcdabc6bd1b5f4efe327a94503d1ca
124788 F20101220_AABIQY liu_w_Page_098.jpg
f75ff920d6d49819bdfa225751289206
e995c2baf27427572737852cfc8fe5255d8fd7fe
116095 F20101220_AABISB liu_w_Page_127.jpg
475648e687c478a8d2117a4e556eb70a
ca2864eafc24c4619cd89b03b6c754e43a74d2d3
8711 F20101220_AABKAK liu_w_Page_128thm.jpg
bb9ca0420f458a21599a681a8624abc8
b42eab172831bebcb3b7327eaed1f74a158e63d3
8979 F20101220_AABJVE liu_w_Page_056thm.jpg
a6b7563f82031dc2b10ecd31a689b11e
8beb85bd59788fda44f57941c328a65f2c72e223
8842 F20101220_AABJUP liu_w_Page_048thm.jpg
d66aa18e51af7bb3219e866aa6a89c87
7db4d19dc218e22969a687f9f3389cc84e0e963b
96755 F20101220_AABIRN liu_w_Page_113.jpg
6396d313093151cf1c9ade0fae952c59
8b267f63c5c51a18e0d4bd26036dce640d44100b
95556 F20101220_AABIQZ liu_w_Page_099.jpg
157fbf107860681778f5bb3dd9f1530f
4617b2ec3a8ac2986da939bd94f066319e07489d
117560 F20101220_AABISC liu_w_Page_128.jpg
1c810392f4979d4c573c05851d72ad17
63d485891c3e215213968f9ecf5dddf73055beef
35108 F20101220_AABKAL liu_w_Page_129.QC.jpg
42c1686b1b203a13c8cbf5f135caf1e2
ff7b696f9e1d4cf293c8cdf65c207bc1df9b8356
12300 F20101220_AABJVF liu_w_Page_057.QC.jpg
5b90b1b0dfa83b1b6bbedc2b1a86b99c
16bd1213fbe85d7ce920b5690450232d0c3ba1e0
29705 F20101220_AABJUQ liu_w_Page_049.QC.jpg
3b6e5e95574b291aae83874921d4489a
ea1a06766828cafa9ae7884604146e034f7279a5
42412 F20101220_AABIRO liu_w_Page_114.jpg
26a771b11ccb70f0074cc12ff6e1fe8e
dc78e6592d036b4f14f3b337cf0187ba88d53819
108715 F20101220_AABISD liu_w_Page_129.jpg
4c486c9d18742d61d3ebcead9d99d4df
0ce25f1cd395f353cd932de65abfd28b90a03a8f
8309 F20101220_AABKBA liu_w_Page_136thm.jpg
dbef3d8f8d9353516dc9b34b5b48cdc8
1b4ee0d86fe9d90342b0789e2a384acfc1b21ffb
7964 F20101220_AABKAM liu_w_Page_129thm.jpg
90dfc7c34158611a1afb085cac9a6987
10b59a0c45ad1b347c797914f250dbe9581ce677
3542 F20101220_AABJVG liu_w_Page_057thm.jpg
bc358c958bd3830fd211a133a858f4b7
77799435ae96cd4c7eb3d3749609ce57b66f9ad6
7429 F20101220_AABJUR liu_w_Page_049thm.jpg
a4a38f0b14457d6714723a46b702d69a
d1aaf757aee1cdba2758aff53b8c1509953839ab
125480 F20101220_AABIRP liu_w_Page_115.jpg
bff57ecf6c238733fe766b2bbf5f3e52
d308507d48364b1e51a7719f5a9b708234a46d14
126953 F20101220_AABISE liu_w_Page_130.jpg
42e49cac8b64cecc23fbc1ce4ccbbdd8
cf9f43444b43f135597c14cdba1b0855aa190bc9
39773 F20101220_AABKBB liu_w_Page_137.QC.jpg
bd6cdc80d211c5e34e590f9d326814f0
1daf94cf3de9901fe3c421d36e3d60be2322ba1a
40120 F20101220_AABKAN liu_w_Page_130.QC.jpg
69d105902d536df1234376d3c6316d84
ee882a2ee4ba3344a4fdb8281f01493675e4a37c
21012 F20101220_AABJVH liu_w_Page_058.QC.jpg
925166ea6f83f54355a080972bf37bdf
8db3708d3a28c755a3258a1a473df9212d557da8
2694 F20101220_AABJUS liu_w_Page_050thm.jpg
d213ef63f379fc0d8d1460bb671cf613
0c88faa503ee62a92e68f33c0e142bef11dbea37
119666 F20101220_AABIRQ liu_w_Page_116.jpg
abecbb1589df04a02e0e0ad48d24a948
bce098953bf9bde1ff4a551ac97ee7722d0a06b3
108828 F20101220_AABISF liu_w_Page_131.jpg
2357c38d20a719b213b147f757104efa
add35a919699c77423ee10ed3ee4196c999732e8
9381 F20101220_AABKBC liu_w_Page_137thm.jpg
8a0ee3a5418d8a23492c8e1d837aa57a
d0d28750d65f534c39bb41f4e4425dafc8e56248
9247 F20101220_AABKAO liu_w_Page_130thm.jpg
fa7ea5bbc70ad6adcc0b50f0f99c8a4a
146945094adfa093f85a1ff136ec4ec8dcfed4cb
5542 F20101220_AABJVI liu_w_Page_058thm.jpg
2e0da1176f055718228b96e54549f786
72c9bc16a71aaeb08b86e3a241e42e6bd3b14e5b
38248 F20101220_AABJUT liu_w_Page_051.QC.jpg
ad01ccd20371fce65edbfccaf847d2b9
2b2fddb730409bb2da98f44b4cfe8f0b2374f517
37922 F20101220_AABIRR liu_w_Page_117.jpg
3b99f9558a88f76c9a8a6106ec80ff38
dd1f46599f2cc8c050d4760dc95a970ebf1a6eed
47124 F20101220_AABISG liu_w_Page_132.jpg
3e2151439618692a5930dc25dd058967
aa4d15bb901af6041e35ab3f560535b0ad66df4b
37616 F20101220_AABKBD liu_w_Page_138.QC.jpg
62dd8ae28eedee3dc2c00fa2c6015045
720a13d0e1f5ab3dfd8bc0385fd2e5b189a47333
33910 F20101220_AABKAP liu_w_Page_131.QC.jpg
83b1f3b22039bafec052dbdd0837efad
640521ca5919b85ad7f1855be4114c16c13449d0
34621 F20101220_AABJVJ liu_w_Page_059.QC.jpg
ec5d440bd70ae654b3581d10c235a7d4
7387aec309bd1c2337579079753ee0b281c95058
8935 F20101220_AABJUU liu_w_Page_051thm.jpg
10f7157d26af76b544a988eec43b4059
4522f9818eb9fbdf44ee5759bf5d6a27e446ccbf
103673 F20101220_AABIRS liu_w_Page_118.jpg
0d4031c124f732d14eed25dbfaff334c
f31809ae6a0214c2bd1bdb8b2b58702bbac34c60
109979 F20101220_AABISH liu_w_Page_134.jpg
1cdb0cbce7d65e4475c7ca889240db2b
f42029a541c883d4e586952e4541df56eddc4cf9
8765 F20101220_AABKBE liu_w_Page_138thm.jpg
1eb04d442136deb5bd1dc1c4bf6a00c1
3c390e7914cc0895ce65da2a931e26dea8150b6e
8434 F20101220_AABKAQ liu_w_Page_131thm.jpg
d37354bd5aa2532e732fc9492546016e
af0180ab81eb4c370f1a09a6db36bf7a135f405d
8212 F20101220_AABJVK liu_w_Page_059thm.jpg
78a48ed746c304bee79be65531fd9c70
1a41d0cf0050612114820ac4264cb8623af0f7aa
34931 F20101220_AABJUV liu_w_Page_052.QC.jpg
ff8d4a260fbe9d73d9eda439436d33d7
7168ab655d71be9e35c338e3703c6a38c1518f6e
126289 F20101220_AABIRT liu_w_Page_119.jpg
25acd8279b0d3fc1a7600d84e0f913c1
8af2ae918f22b59ee766f7844079392656938582
117342 F20101220_AABISI liu_w_Page_135.jpg
ddcb61a99d661d7a4343c368c166052f
5401acb03a16f2c1071930fea06aaa921dae4a83
34983 F20101220_AABKBF liu_w_Page_139.QC.jpg
b841cc163e3242d75c8aefb1e0afce82
c67244fd2f0a71a5585aa03e68758698bf9af35e
15550 F20101220_AABKAR liu_w_Page_132.QC.jpg
f772e9f5151fb87589fbd46a706e951d
87a40596d3c1d890196077a08810b81b711a079f
18892 F20101220_AABJVL liu_w_Page_060.QC.jpg
78eff32d6be64eddc7b6cd03c2411fa5
829c2d759eea612b13b05d8492dc16516198953d
8633 F20101220_AABJUW liu_w_Page_052thm.jpg
2abaa5f429d0e3a0f4670dca23df98e4
0887aa53081620137c3dd9f13f7dddd45cdfe5ad
120771 F20101220_AABIRU liu_w_Page_120.jpg
4d05f3ac0fbbff55ca5a8259f371a101
121036934c33d69f04bd9e13fbf9cb90e3a20493
100645 F20101220_AABISJ liu_w_Page_136.jpg
f80a97e1580abf1133581c348dde9f8c
3f9300e2cc6e7c28b59b5596b18ae41fa36d431f
8504 F20101220_AABKBG liu_w_Page_139thm.jpg
6839c3ab2148e285b701201032d2cf58
f17d7b7f6ecbcfe3143538d8a32d552852778158
4412 F20101220_AABKAS liu_w_Page_132thm.jpg
9fd9f7fbd5bb9a68c2a45ab3672d0387
a401ed68dcad8f8ffc5c79b89ed3a260e46d265f
7288 F20101220_AABJWA liu_w_Page_067thm.jpg
037b864dbda9796a2a2c9286386c1278
e126638e030980552322e4beaa073c8f24b81802
4231 F20101220_AABJVM liu_w_Page_060thm.jpg
d57661e74f9300af64ca84c72072f38e
11f0c2ee6edf10cb935d673b17757950655792ae
36842 F20101220_AABJUX liu_w_Page_053.QC.jpg
75b2c287aaeb08114c4e3b5e87db0d89
73c73e1c0247da3f2ce1a8271bd00ceb04123d25
122168 F20101220_AABIRV liu_w_Page_121.jpg
5967a559e2a792593cee982ab6136e8f
889d4d34e919095ccac1a9108a7905a80689d6ce
124067 F20101220_AABISK liu_w_Page_137.jpg
870362910eb8b9e235fad49be548b2d4
db881928f84ba292b576baca30a0e01826123179
37461 F20101220_AABKBH liu_w_Page_140.QC.jpg
96853be008bb555ad189171d737e6bf0
aef5c341992ca4b85b59461c7db2306299c98b3b
37496 F20101220_AABKAT liu_w_Page_133.QC.jpg
8cea283f7d2fbdc982e85288934b7cfa
f8d0f422acd1eead08268f414b2a6f936ab4823a
35873 F20101220_AABJWB liu_w_Page_068.QC.jpg
bde16798dc6956d912e614b1c290b6d6
e994af7789b1bbfc441db302b8fcd4d499b54147
32972 F20101220_AABJVN liu_w_Page_061.QC.jpg
fb1236a34831e0917b9e769cb063e2ab
0f4d446fd5da8df74f7fb64748f0c137e455da2b
8461 F20101220_AABJUY liu_w_Page_053thm.jpg
af0f1db66a7dcb7162f443c093e7afa7
6ab4b542bb4718afa2cfdb519ad49bb97eb134a0
124893 F20101220_AABIRW liu_w_Page_122.jpg
7dcbd5eae78098dcee1ec648c610433d
e299934cd0cb8c3e717d62a7960069dd16842755
116522 F20101220_AABISL liu_w_Page_138.jpg
394960e2065f10cb515eec262f7c34ac
e285eba6fb33a68a838f519ee048e4bd31af6de8
F20101220_AABKBI liu_w_Page_140thm.jpg
3b109af63cde3c322a4d93b313b17f7c
548a4abfa66d5bc032f55e9192dc20e3797899af
9029 F20101220_AABKAU liu_w_Page_133thm.jpg
e5bfc3613694654b105faab13d26ecb3
7751c4a5c08ecbde4a5dc8537c5cd68104c4f964
8577 F20101220_AABJWC liu_w_Page_068thm.jpg
b80628d537e85284c882b5926dd4a8cf
a26b70b847189afb2e52dd4a9d5998ef876ab18e
8069 F20101220_AABJVO liu_w_Page_061thm.jpg
646d3df3142b359fe1769ab178c39634
b908c11c7eec93bcb325e34fb1bdd31129cae544
38709 F20101220_AABJUZ liu_w_Page_054.QC.jpg
ef1f68266a089f2565b90997ea8d6002
801979dfd289033446e84f15e6be7fa5b2bd1e59
102913 F20101220_AABIRX liu_w_Page_123.jpg
ea254fc8ea9642ee01ea0c55663666a5
4689b69b72e9f01f0552bd89f77b2595fff659e3
112918 F20101220_AABITA liu_w_Page_153.jpg
81c817fdf0aee110dbee981b5abc1603
97807879c41f1131f00b76714a8e75c33f6914b1
112128 F20101220_AABISM liu_w_Page_139.jpg
8c4dd11a5336e53555014e032b66ba81
a652d774baa9fbbd08b16500d92188805f4f8f0c
38092 F20101220_AABKBJ liu_w_Page_141.QC.jpg
5718e4b56fb4ea769f47217b879f1a94
f77e43649dea44fce6c4ca5a4d5c36b315e5a2fe
35076 F20101220_AABKAV liu_w_Page_134.QC.jpg
af0d4cebcee4e26d961e253b02fb1491
1e1a52fb46d99cb041828f7a3d09e5f3b24ed24b
39432 F20101220_AABJWD liu_w_Page_069.QC.jpg
80be6f807a29b29434339677ad9837cc
368f0ca7f33b94e5ba44bed0683ca803e642e999
37506 F20101220_AABJVP liu_w_Page_062.QC.jpg
89f97bc252e65c6e4b8eac7394a64d8a
7eb89d4c89d3eb74e996a10d24c1bb7ba381b9a7
109512 F20101220_AABIRY liu_w_Page_124.jpg
2381f4eef574cf37b9f0ff8d09887faa
c436d0024eccb3aaf164a99326a3af31d907c77b
122919 F20101220_AABITB liu_w_Page_154.jpg
2adf7e2caf332763007d5e5e56e6cde5
b345a6c34229b5f37fb8244eebc426fb4f74b00f
118428 F20101220_AABISN liu_w_Page_140.jpg
c9a18af4f9aaafbaee053211081c383d
3bef4b5f2a87357d6428ea32d7982480111f785d
9051 F20101220_AABKBK liu_w_Page_141thm.jpg
609df50260435667d7f5dee6a7225b3f
a18e7a77cef88b9f372119e29fe9d2e5d620d810
8407 F20101220_AABKAW liu_w_Page_134thm.jpg
de6f06f8e7c96d881c4a9b2162572024
9d19008feb951d760493ee18ca782e601ce77a6e
8932 F20101220_AABJWE liu_w_Page_069thm.jpg
30bcbd7184dd09cf39035b430ea4449b
e3e04e716886894917df237244324d905ef8c6a6
124804 F20101220_AABIRZ liu_w_Page_125.jpg
1212c1d6ab9ce9e886be0569713343f3
0594cf99d038314875a9b2a3a60548032452b2b0
122743 F20101220_AABITC liu_w_Page_156.jpg
61899d14873b106c48e6c12f15924d36
a90df46059dfd27ba6bd8fa870952f3abe5c5a1f
33442 F20101220_AABKBL liu_w_Page_142.QC.jpg
b5097ed3a51173b5d4e122cb6e605258
e4eebb0db2985c09804e9ace4ab24266689ff105
37801 F20101220_AABKAX liu_w_Page_135.QC.jpg
17d74a75bf126b6e803e8dd276e1a449
557ab5e2d8107b65ef9f1849ae26a68afab4fca0
33041 F20101220_AABJWF liu_w_Page_070.QC.jpg
7ad808a23cfaddd768d32fec3dbc344d
102cd94c9b48d8cd44a9dbae7998df369d2eadc8
8951 F20101220_AABJVQ liu_w_Page_062thm.jpg
1153d6b7d9aa360e776f76bd36836358
b613874677eafb43dea5147efc120cbaee62e834
131554 F20101220_AABITD liu_w_Page_157.jpg
09b35cf3b1d0d4c3ec24fdc1098dde80
fdfd33d78bbcc08a87ad445f37a30b55f342cd90
119066 F20101220_AABISO liu_w_Page_141.jpg
1cdffa3f1127f7943ff6e14d10ec51ab
b1e84a344b4466130ebd98bb00fad8a18fce6666
7471 F20101220_AABKCA liu_w_Page_149thm.jpg
dbe0ee800d47c0acc37f175240a9bc22
334ac26520c028677236e68ad9fb68ced4bd3158
7935 F20101220_AABKBM liu_w_Page_142thm.jpg
6a6ac54fba55c3de9cd1a39ffe1d0dfd
8c2f200b1de8f6842080c40ec646fa15f967f474
9058 F20101220_AABKAY liu_w_Page_135thm.jpg
d89ec3b9f275c6669dedf302139273a5
8d10b2720ae5f2632e0e0b426c289c0a6a4105f9
8013 F20101220_AABJWG liu_w_Page_070thm.jpg
90fc9958b8336d040541373e26f0a48f
83effc4cb867fe991ca898ba7e9e3242624fabc1
37201 F20101220_AABJVR liu_w_Page_063.QC.jpg
1057bf4a6e7bc6227d0f084afbb9912d
dcb79fbe7c0ee9c96a7a53d3bf4fb9d1046600f4
133354 F20101220_AABITE liu_w_Page_158.jpg
8dab4fd500967b71366dcee66dae3805
fb813366754b1079cde14c346610607a78f63401
109633 F20101220_AABISP liu_w_Page_142.jpg
90ecbc1e68197e6903e63faf65d155b3
c75aaf502658dac95b874a108974adffa1ca8cf0
33928 F20101220_AABKCB liu_w_Page_151.QC.jpg
431f2ca18304672bc7a700c8ac663f44
5c07fb39eb4947d858c23458c711f862e68e81c5
19723 F20101220_AABKBN liu_w_Page_143.QC.jpg
2a04e40d33416ec413210b83568e848b
74f80fdb089ad9a86ea8061e3622d07dc170f901
31947 F20101220_AABKAZ liu_w_Page_136.QC.jpg
3f91993fbf284b4793d58d6e5ea594de
8dd2a4c9961949d9192a058b787d5e3b8a761d3f
21496 F20101220_AABJWH liu_w_Page_071.QC.jpg
838ec8b219c149b58c47773e0482bc8a
a1a91c858f6d462d4112577393cb91eb9c218734
8911 F20101220_AABJVS liu_w_Page_063thm.jpg
e7cebe61629a79457f269209720586d3
0868d8af4ed023077974275eaea9664493746c71
134584 F20101220_AABITF liu_w_Page_161.jpg
22334d215f28d0355fa800e6d6b8e0b1
096c605ac453be20c2d6826a80e9c3520b206b2e
61679 F20101220_AABISQ liu_w_Page_143.jpg
d89366d9b311d841291625c4dd1bfb38
2f24649f7b12850d523f595204d7d048e70afd6e
8408 F20101220_AABKCC liu_w_Page_151thm.jpg
f7b1fcf3f4e7f3f45fcd985013cd89d0
28315774e3f34a501760c237188426161a1ac4d0
5506 F20101220_AABKBO liu_w_Page_143thm.jpg
5ecfacc2ac8c7cc05ebab1538c4ad667
3b31c2df42a946bbbf78c15bdc98b5fc9518d7d1
6113 F20101220_AABJWI liu_w_Page_071thm.jpg
d2a8219a88ac8bbc26da4512938db5ea
c202053e30de8e4f06b9115566339cb785ecec04
37702 F20101220_AABJVT liu_w_Page_064.QC.jpg
177da8bcd2127ebfb76f53af04742114
349b5d0911b2cbca0310f86662694be701cc6962
79765 F20101220_AABITG liu_w_Page_162.jpg
e159a3b23c26277877256282b0cf48a6
ce6bfdecb904cb718a5b01c601c1883d3a438fdf
71476 F20101220_AABISR liu_w_Page_144.jpg
ce072f3a42a29cc94c21ddee65513bef
00a36d08647211b05a209f19ce715287b7fe6521
7047 F20101220_AABKCD liu_w_Page_152.QC.jpg
9d7594f0a8f6f0c650bc1e69f2bb9fe1
d5adcce658dac0c7e6ef0f7a0630efded9fcc599
21427 F20101220_AABKBP liu_w_Page_144.QC.jpg
dbfe432c8ef46ae9c516ddb7bb10b017
8c835451f40ce4e88c53038187124ef169516881
30981 F20101220_AABJWJ liu_w_Page_072.QC.jpg
3d70fe78be0bc485bc814f78fd8dfad2
fc00e11d7f23c17ae90ca212694070dce6705a13
8774 F20101220_AABJVU liu_w_Page_064thm.jpg
b20435408b0d48f83a7c90372ed8eda1
7e0e3c94349a4a3142eebf9dbefef403053c90c1
32194 F20101220_AABITH liu_w_Page_163.jpg
d3d52a53065a086ef5486b3a1976f4f0
c9d09d742a045acea299b8d3233bb9f3a6af9184
121558 F20101220_AABISS liu_w_Page_145.jpg
069e065c4638b50295a26d84ed27d0db
ca8d1be485b6a461d3ebfff8ad95e7e00bbce756
1962 F20101220_AABKCE liu_w_Page_152thm.jpg
a808f7cf7f2a0097a3f4c08681449a97
22e313b50784183cbc40c5fdab473c966c666a6d
5588 F20101220_AABKBQ liu_w_Page_144thm.jpg
5bc2a9d078bd457a70bcac3f4d73ab8f
782acab600447088d6205183fc08b63dad0fa510
7393 F20101220_AABJWK liu_w_Page_072thm.jpg
ed67f0319cb66a1b1420001f2aff6817
383b859772920646a289d4944e75c2f52babf6d5
39901 F20101220_AABJVV liu_w_Page_065.QC.jpg
1f6464877843dc19be0549754d6ff4d2
2630a11a57db8b70a254c3c9b2ff421619065337
21449 F20101220_AABITI liu_w_Page_165.jpg
0d4174b8dc615b12bbcbd51888d3f2f5
a51e3fdf6e631d3c380b98f661bf8a626cebd784
83889 F20101220_AABIST liu_w_Page_146.jpg
456837cf18638743f3aaa1fb637fb2a0
26f40105c614f9496078dc6388aba32c0a1177b3
31878 F20101220_AABKCF liu_w_Page_153.QC.jpg
21d393c0591cf2d0428689ed7b920a3c
3ffb34164f6d4dc16d7f4f51263a39fd0d883b3d
38088 F20101220_AABKBR liu_w_Page_145.QC.jpg
699c1d208717f725453e85440f1a767d
0dbce76815f26cf96047c6a8786090da58519e88
29869 F20101220_AABJWL liu_w_Page_073.QC.jpg
6a76704963a2f8780a2dead0fb2cdf83
511184a82dbd9d29c604f486a0b503baf06a6458
9190 F20101220_AABJVW liu_w_Page_065thm.jpg
8958c78ac0d871b24cdceb745e6224df
9edd46571b541c731bbe1e508fc2f42b9465dabe
69750 F20101220_AABITJ liu_w_Page_166.jpg
712c8c72e0c2d91ba69ad6095f3fe1c1
b48bfe473d47e67ac4f6fb8952ae835fcd859c7b
50794 F20101220_AABISU liu_w_Page_147.jpg
db25ffaf18566c628edb31971c039fa9
f1fb746a57af1687e00b89f1766c014d8f956db1
7823 F20101220_AABKCG liu_w_Page_153thm.jpg
ac4366c63cbfc37e247f7538db5f9add
769fd6626db1b1db8947dea1f20777b29ed6a48a
8906 F20101220_AABKBS liu_w_Page_145thm.jpg
5c593b0eaae21563598c17085eb16cc7
6ff3f6fb1680df52c73a0fcb8a55ab2940a577f1
4815 F20101220_AABJXA liu_w_Page_081thm.jpg
bfbba50f457e01577e94475b97b08830
f158ce69ec8d161b5d2418b896aa7eeccdffe72f
7415 F20101220_AABJWM liu_w_Page_073thm.jpg
020326a162d2f0bd57940417cd0beb48
50878631fb5a0246e98079b0d05ac096c5fafc22
37000 F20101220_AABJVX liu_w_Page_066.QC.jpg
e57ecc2fbe126c5a22e70effcc1ab290
5f112d387264e0ab97a28e649aeb01cd4426bb7a
252636 F20101220_AABITK liu_w_Page_001.jp2
4683c71ce91fac9dcf49d09a4d6e93c4
a4a2553503be0de4559873d9e4f1e226a078c063
125629 F20101220_AABISV liu_w_Page_148.jpg
b423d70e6d584546bc0a700098cd8e2e
4a208d682846c7d9aa065b34abe0b428673702de
34869 F20101220_AABKCH liu_w_Page_154.QC.jpg
042161400d7c5c86d3ac29f4de512981
09faa0eb5cf7c197d97f9dc244286102e2e358a2
26301 F20101220_AABKBT liu_w_Page_146.QC.jpg
b89e8f7e01bcc9b25b6100f8eabf1b61
10c3679f1d48056e97e00fa0eb8348f4c593f7dd
37644 F20101220_AABJXB liu_w_Page_082.QC.jpg
920fa7f5e7191c1098575eef57ad6a55
5f2ebc80222b7aa060f0aee3986fb479f74ff660
13225 F20101220_AABJWN liu_w_Page_074.QC.jpg
d635f025b0f6a2f54f3baed050d8632e
eee0ebd964d1272657d17510649f429ce8751264
8658 F20101220_AABJVY liu_w_Page_066thm.jpg
827a34bdf9797a0d68d6a718a4d8033c
533fec473774f260d33561e34a07033e5f7d9a0a
26066 F20101220_AABITL liu_w_Page_002.jp2
e1f41e254447bb657492539c05235856
3af2acd7087bad85756e884b7236d9aac8a780d0
100806 F20101220_AABISW liu_w_Page_149.jpg
5ae37fd9db8889eed36cc7dfce57940e
c0d2a3d6c7c37f901b446b9676e1d4b026aadd0d
8820 F20101220_AABKCI liu_w_Page_154thm.jpg
ab9fe6cb2a24b1a98b6151159969f59d
2a898adcf7cfc3ff113c62bf391bfee9a952fa90
6814 F20101220_AABKBU liu_w_Page_146thm.jpg
17666777249de4492594af7977fb34be
6d51c012875b682ba8cd5aa703daf4d82054f197
8867 F20101220_AABJXC liu_w_Page_082thm.jpg
a532f3be2191f3471df7c733662389a2
5d2fea0c0a5448654898dc6d6eb2be24143e9770
3624 F20101220_AABJWO liu_w_Page_074thm.jpg
e9240ba9ec78e93d411444e954c95f7e
1d7d9f7630f676ee5b5c70c2f9beb5a34da5f291
26212 F20101220_AABJVZ liu_w_Page_067.QC.jpg
1e38b648f9a76ffdbd461eda8531197e
e9a0003b221dbb2cc32ae98136d5c2d97aedecb8
1051974 F20101220_AABIUA liu_w_Page_019.jp2
4e3fa9f262957007564c041342cd9579
fc5d4d683f5983409b1581cf02adcd2ce2c59860
38230 F20101220_AABITM liu_w_Page_003.jp2
192b84b4535fa8334daeca5cea77ab31
4806e181c61b8e350ac3a4f6489bea2a619ff0f9
99369 F20101220_AABISX liu_w_Page_150.jpg
1ce23516f48430f363935c1116bc37ac
ada8c374d2e6c4b644040fdf3f597598d32856f4
35352 F20101220_AABKCJ liu_w_Page_155.QC.jpg
7e607b71037b08b8b9fa40a9b4c03ede
e513302518b6b38ede668179086890e8ab7f9646
17943 F20101220_AABKBV liu_w_Page_147.QC.jpg
4fbbb1cce910b6c055226f41cc7a85bf
dc58554f24a91ae8886ea4294d57da406e6be0b4
3802 F20101220_AABJXD liu_w_Page_083thm.jpg
6d8bab26d3a04cbb9864b360167d6f7e
84346aa1f1d628e79b180009f1497cbed414f480
26506 F20101220_AABJWP liu_w_Page_075.QC.jpg
acb4f47b9f00f2b39613bbab5cc1020f
b311375224c5530ce778b004ac0da221d075335c
1051986 F20101220_AABIUB liu_w_Page_020.jp2
823cc8ffaa405abeb08d677be64ca0a8
ee1474b954a3fb236a832bb18539bd34ddf4a4a5
963965 F20101220_AABITN liu_w_Page_004.jp2
850acb970e5660c1504df7c730020771
72371e25984348b992c8f60ce6fdbc2a2bb5073d
107802 F20101220_AABISY liu_w_Page_151.jpg
5c1fbfa938c7bb9586b9dfe9a943decb
a37de10a389e96711f1e89ca908f4af0c5e8c0f7
8640 F20101220_AABKCK liu_w_Page_155thm.jpg
b09e9fe513ebf0d40541222124e48323
3a945a8d9d1f48932bf2b3ad04d1aabb6d1f18f9
5739 F20101220_AABKBW liu_w_Page_147thm.jpg
62d1389c672a479cf6c1eb33f331ac56
b6b15166620b8e5d88fa5e7060747a76e291a052
8162 F20101220_AABJXE liu_w_Page_084thm.jpg
9add0e313ebe6b6488879e151f41ce99
06d08f1cde2aa8b172dd6e1bdb1ae04463fafdf3
7092 F20101220_AABJWQ liu_w_Page_075thm.jpg
7b8155553e63c50fb1572ce669f59ff6
a582ab28f37893e8b6b1539feabbf30e999919ce
1051969 F20101220_AABIUC liu_w_Page_021.jp2
970b053363ff3ffa632923c07cdc40d9
f2695e2d242d87dcca7882ce81b02db387f91f51
1051982 F20101220_AABITO liu_w_Page_005.jp2
d3e518d662a7302bcd36db1f1cad1055
7175a9ded362e7144443488b71c9460ac3366463
20630 F20101220_AABISZ liu_w_Page_152.jpg
a5a1087e0d0638107d28efda9783e227
7e8c9eb7fc63be10201f89e6bb35d017c2ad9c5a
34623 F20101220_AABKCL liu_w_Page_156.QC.jpg
6d9f889b79448948ada425c75b59259d
7750677f41645bb7b5074430774a0426869e424c
39593 F20101220_AABKBX liu_w_Page_148.QC.jpg
9f88b2dab907c28c80e66ec1745cd4d4
4a372d797e9a4b882edde9cf15a7b25bc32ce850
36424 F20101220_AABJXF liu_w_Page_085.QC.jpg
ef00ef0270a19a8402402faa41f092c5
9dcac06888eda63e34bffc373952f17d4b42f717
282201 F20101220_AABIUD liu_w_Page_022.jp2
ae462386e260d8c295edefc47c66619b
63759b04382b4b6f8214630e01916894a84cc98d
23017 F20101220_AABKDA liu_w_Page_164.QC.jpg
7153957a034e922bcfe82e04afcd6bc0
4342d85f28c23f74103ff6940520ed556d8c9fc6
8620 F20101220_AABKCM liu_w_Page_156thm.jpg
e369f0523a6700884d8b58cc2b44dad1
a9a8cbc694228b1c5d80f9786a6b5d4da53042da
9403 F20101220_AABKBY liu_w_Page_148thm.jpg
00cbd9e1791f36b1bd7f5675c78bc34c
a2ce69040e9fd07e80632e2bef75e7239b009c37
39298 F20101220_AABJXG liu_w_Page_086.QC.jpg
d8405542c00637ea13b5908801b0ec28
88721d028eb16ef38a673ea1bef0e7263119b52e
7847 F20101220_AABJWR liu_w_Page_076thm.jpg
b596ccf1e279999803e2a25d31595c00
d4c4c2eeab9b90a9ea4fee199c882a87748cf6d6
1051935 F20101220_AABIUE liu_w_Page_023.jp2
e43d903cf80cbbbfdd90de80ac6109b6
61b973e217e9668054f4b85163a94c6f2bfd8fd9
F20101220_AABITP liu_w_Page_007.jp2
ab400fc02b8b3fe3a3c6a3234361ed44
d1ffa31ce2d493f650ae81cf3297ad17d6590777
6006 F20101220_AABKDB liu_w_Page_164thm.jpg
c265d2e2acdb76c9a1cd207b38d73c6f
5856d69018cecfbb7d4cb910747fbaa54f3cbd17
37295 F20101220_AABKCN liu_w_Page_157.QC.jpg
62247e02061ec9d2b021e5a4a140b42e
b5cc6b3757efc14367fde9c4fa8f869bc5faa565
32307 F20101220_AABKBZ liu_w_Page_149.QC.jpg
f7562be1a6aff2d0dc7d9e52ef06e5f4
03ff9b2943ba543fc17da9d17689f7f89c03f792
9134 F20101220_AABJXH liu_w_Page_086thm.jpg
7bc424378b69804341f6f9c02177a814
e35d2324e2f165362cae4a71e5b671231c133d43
38282 F20101220_AABJWS liu_w_Page_077.QC.jpg
cd4d72a59e3bff89923a75c235f70edc
3ae9db38853d98a9e88b83fd91591a4d86d35923
1051972 F20101220_AABIUF liu_w_Page_025.jp2
ee93b5b84910507ddacb7933d83b4506
f9e093c43f5f659af787c270192877314d4b4b69
447921 F20101220_AABITQ liu_w_Page_008.jp2
f2af7480a2077367a8308640ed885ead
595bc5362c86985505d29b426ccb5867dd25733a
F20101220_AABKDC liu_w_Page_165.QC.jpg
a6ae9550f94acb21ad112ef480bf8f2e
6e835b752bea66bbf3cba4e6be9eacb1c6aa49b2
9168 F20101220_AABKCO liu_w_Page_157thm.jpg
abc4dbc810ef81225ca2abc95cb2bc02
1e8d1910e5aacfd21e8cb9db328d5a731fe77098
40823 F20101220_AABJXI liu_w_Page_087.QC.jpg
1ce57aedb395605e1b772ebc0ca7ac41
8418455a88d2bb4a8eaf95c25033e730562c956e
F20101220_AABJWT liu_w_Page_077thm.jpg
1f9f42c00426e6bbf18b6ad6ce6544d3
c16067c028d1665838babc4d214e8a05e61d5782
F20101220_AABIUG liu_w_Page_026.jp2
76db266cb2d10de63597c8d1e258dac0
937d06fae168128aa8fd508fe4bf0502de200e89
1051980 F20101220_AABITR liu_w_Page_009.jp2
e97c45158db9c4b1dffa19d6be0aa9da
a4e4b1d0feaf090b92329a5db3f08e425d5783aa
25271604 F20101220_AABJAA liu_w_Page_019.tif
52bf8810e841293fc61c8a038240e90f
51e954935ef7efe02302d9ed3a00d0429b16e841
1965 F20101220_AABKDD liu_w_Page_165thm.jpg
d7be016767b6c1258e7d1370b32aa186
4e08e41ea28066db6b2852a5889313372f7ff6ac
37549 F20101220_AABKCP liu_w_Page_158.QC.jpg
aeeee2e219786ffff33a421250b859f3
f1fee2739e2446f1e6dd3fcbf4bfe5492ba16503
9285 F20101220_AABJXJ liu_w_Page_087thm.jpg
7b3a2fc0d7a1c5fef04e857df2cde9dc
00b6f928f42d7248b35fd536b53116e1da9b7277
34346 F20101220_AABJWU liu_w_Page_078.QC.jpg
a4ea2ec84cd4c06e05ae84474787d937
f221f7848f90e607543ff2be78ed273a115b7fab
1051985 F20101220_AABIUH liu_w_Page_027.jp2
6c4038e3366a4544636733e653657afc
3ff25517218a819f4db68ee061c4a1cd99b8ee87
1051979 F20101220_AABITS liu_w_Page_010.jp2
ad052525d8b227c1fb6011cc180bf51b
47fd23b159879c16c0133907bfe3243eb0e91c0f
F20101220_AABJAB liu_w_Page_020.tif
b3dd5c81f62ef626bd4a35212785adef
ca0f7d89d6c06eb820543adb71d65e53005b41a0
21351 F20101220_AABKDE liu_w_Page_166.QC.jpg
8ebd80e9a512618e7b99dc23f02e912b
16cc26d469fa92906b907bf39d09cbc8f07e57c7
9150 F20101220_AABKCQ liu_w_Page_158thm.jpg
e263866db4c59f97e815aa75cbd16afb
02556f645dc39eaa1967bf6b3d7421b240f820b3
38556 F20101220_AABJXK liu_w_Page_088.QC.jpg
d407ae685569b13cc7f0053a5f3ccfa6
d4b0bb07efb6aa582da23e6bf72c60f7c7e0f07d
8118 F20101220_AABJWV liu_w_Page_078thm.jpg
9443e3eb9cf3fbcb3ef011f95fd09675
07ada47c2d173779ae6d926298170f1c50d26a48
F20101220_AABIUI liu_w_Page_029.jp2
4939d3ad506cbada10d5dd461fe65f14
317fc0aba0a9e9bb264823ca552fbd366c6fa41d
1011858 F20101220_AABITT liu_w_Page_011.jp2
49bd4d75dd42130f379684b5bf733cab
0c7ab67aacc2cbf8413f8060ed6caa56cc5642aa
F20101220_AABJAC liu_w_Page_021.tif
bd0e9ea6fc1409163c9e1cb57b408bab
88f72b00cb55eecf27053aa0025d2449149c1603
4983 F20101220_AABKDF liu_w_Page_166thm.jpg
539d5467b26a5240e5a7c3f3e730a41c
4ddc7b1c6b98d1819133feb9f8d099212169e861
35975 F20101220_AABKCR liu_w_Page_159.QC.jpg
c5d964a53f3c5119e624d0f5641a8927
315c789875fa91f19100768263d727f3949c1b00
9138 F20101220_AABJXL liu_w_Page_088thm.jpg
031d2d59a1bf4efbfbf2fe0e82bfca3d
7576e3751bbffcd69104a65d6344c7b74bf2452f
7306 F20101220_AABJWW liu_w_Page_079thm.jpg
bb4ada82de278bd30420e7747ac293c3
f7c5ec4f070a148e5bb1501956d9ced8ca2a7d10
1049962 F20101220_AABIUJ liu_w_Page_030.jp2
02ea1e686cc60f1d2953c1c60e19effe
826adc3ee63eb1e4241f0d41895268e46e1aeed1
630696 F20101220_AABITU liu_w_Page_012.jp2
f7a0a49307c7f7ff329265079a94ce52
381d71637cab2db798cb6c83ab0a7491c22f3e09
F20101220_AABJAD liu_w_Page_022.tif
ef54a7f56b20a4647baadc59d7f659e5
4cd24f41abbe5428a30ad89ce1fec6dccca16eac
189144 F20101220_AABKDG UFE0011611_00001.mets FULL
7a7c40f5965a279b542ba54e790b3ed8
e38240b45dfe3df365e5dd9ec638056969973660
8744 F20101220_AABKCS liu_w_Page_159thm.jpg
a72e65570af4ea4ec30efbf3caf9548b
eb3ab9d8ea1269f69b887c46519afb30febfa92b
26002 F20101220_AABJYA liu_w_Page_096.QC.jpg
55d6f7b6265eaedb668dc41a55f5b185
35201b6ec86198dc2aaefc1cd091663c83554efd
31198 F20101220_AABJXM liu_w_Page_089.QC.jpg
7730594ff18cc1d2834fe648e993649a
8a4fd325d1629c842108ce40c3f46eab4e6a4f3e
39350 F20101220_AABJWX liu_w_Page_080.QC.jpg
c73b5b4f46d9754dff61e27bf452e8a6
8b887c04af31f63f2d6d46277e1b06bd05f2488e
1051961 F20101220_AABIUK liu_w_Page_031.jp2
d0fc2c417b007e7b6096ebaa0f1b7ef5
8e0768d2a16023b58dd487c0b5ab21d4a7c65455
1051968 F20101220_AABITV liu_w_Page_013.jp2
a433d31480eabad2f9a81fd661f3bc51
9818358cdd4c4f3efaddc56e0452cff73fc15717
F20101220_AABJAE liu_w_Page_023.tif
4890ddd06b44c6681e4fa9c9125097a4
54bf11a2726e74c63b96d3758fd03296bf638e7b
38067 F20101220_AABKCT liu_w_Page_160.QC.jpg
99b9f2fb7a314bdac35c19e5eec4be86
38754ce22aac488ad257f00b1fc8502cea186355
6624 F20101220_AABJYB liu_w_Page_096thm.jpg
b685e8b86095389ac05ee08ac6e46866
27745e157fa6f9c47b44c2e56670f56d3790e650
7330 F20101220_AABJXN liu_w_Page_089thm.jpg
0cd98b78b84c0e11a1c9d355b59240dd
1ad288e4a178982a02690ec92214b5c1f6f46aa5
9158 F20101220_AABJWY liu_w_Page_080thm.jpg
29582244af4cc68ca7d2ec220ba47801
0db4f10559d137f493adf8636f41b296c6c67f70
F20101220_AABIUL liu_w_Page_032.jp2
3d459ed3c6e30e18de8eba5e219268a8
6a4fd84bfb3a1aec651404c64d80b24ea6f18bec
F20101220_AABITW liu_w_Page_014.jp2
4822eb2d52ff9a674676b19b47a3ff4f
423d78e6711a12fb50c0076b349fb8bda45fe1ad
F20101220_AABJAF liu_w_Page_024.tif
4dbb81c9f96c2baec60343291313432d
e560144e7dc8c053a1e0ab224acaf9acc8ee780d
9008 F20101220_AABKCU liu_w_Page_160thm.jpg
74ede7eb62ede46193c36ff395113f62
2e0f35b749525e34a42b82252e07e8b73f3a0d39
39556 F20101220_AABJYC liu_w_Page_097.QC.jpg
7109b70d7be87f51842fbec4a6e10863
46b2ea844655a004890091c93dfc086df6c8a7fb
39174 F20101220_AABJXO liu_w_Page_090.QC.jpg
3a0f961e28110222f89e09221b2ece38
eb70dde7de66cf35eccd809bf6a38cbb7641df66
16965 F20101220_AABJWZ liu_w_Page_081.QC.jpg
1477fc0b3e7a6cb92b7e1246bdf10d9e
ab998927744ae0cc511efaed80373576a5c9144c
1051977 F20101220_AABITX liu_w_Page_015.jp2
d52bf564c69c49064a85d0674c0108d6
9e0b4449acc21f2d7cc3a33b4c94280c458a8276
F20101220_AABJAG liu_w_Page_025.tif
eb63bd7093f0ad75036edf5b7a5b3fee
665d8ef76db4d3b81ac0c8cf7be81d99b37b439a
1051966 F20101220_AABIVA liu_w_Page_048.jp2
2dab8e680bd0572cc5d82d3c796b31b4
3a8507fa07957ad919c09efb5819becd279bdffa
1051981 F20101220_AABIUM liu_w_Page_033.jp2
974817c979c13b00a7d0480ad619a9fa
9434ca5144a20f39189ccdd9e94ea637798eda61
37602 F20101220_AABKCV liu_w_Page_161.QC.jpg
0891b9b09448facc5d5cb7ab55935ad3
e68feaf507f45551ca9caca51f615bd7615f672b
F20101220_AABJYD liu_w_Page_097thm.jpg
b75ecd4fabf00c8b9b45b8db8ced192e
49f939464c6711199461dc8f6d037e45a0d5d08e
9182 F20101220_AABJXP liu_w_Page_090thm.jpg
74dcbf64392e401fce4233b3a7bb6a26
d2a701bdeec346a9d16b70897eb7ad15f36ecfa5
1051967 F20101220_AABITY liu_w_Page_017.jp2
04eb06670ce93aa30b67ce0d3b0e94b8
e32565fe80b45fe0638277367ef07a3392ce8685
F20101220_AABJAH liu_w_Page_026.tif
d5c367a166873c7823ab8a25a85cf4b6
c351640e444776ea0ef8074b6887ff65d32cec72
964001 F20101220_AABIVB liu_w_Page_049.jp2
f9538520cbbf6281afd5c2e5fc4fa5b9
3cc920d96c70ef91d5cee229d9167ac2d92dd50f
1033474 F20101220_AABIUN liu_w_Page_034.jp2
14295c84677f85e988303d7f9ab4d2be
5463dc95741e7982fdb6d5bb5636b58109d488b0
9195 F20101220_AABKCW liu_w_Page_161thm.jpg
363d64dd51d27e48b720490d7185fdff
430c456c47065c813184cfe2cdcbf2e470980e43
39310 F20101220_AABJYE liu_w_Page_098.QC.jpg
ec213b51cf7ecdeef48a7b84cfb00534
3398c8302748174df8bac4b1eb2444d8eae7301f
31526 F20101220_AABJXQ liu_w_Page_091.QC.jpg
a43ea611ce38748153fdc96707971d13
8aec96d25400a4cad610606dadf86f13f327253e
1051971 F20101220_AABITZ liu_w_Page_018.jp2
b67e1a56fb47853b37ce6079b1b19304
17fa7707a9a72375e895ea55459bdecd8a9dea76
F20101220_AABJAI liu_w_Page_027.tif
af22505df5046aaec4d40cf7e6b341dc
163ef4ed265f730f7ef15158cd1738babdb27388
298677 F20101220_AABIVC liu_w_Page_050.jp2
181687bea2d5e083385f6e85548e5971
0275bdb49254a5e460596f6527d580df3c21da78
766955 F20101220_AABIUO liu_w_Page_035.jp2
bfb3cd7aef39fedc576dec70af15d7ef
8864315a243376755d043e05c31622eeb89104d0
6003 F20101220_AABKCX liu_w_Page_162thm.jpg
2ae00d0dc865f3f9fc5d660ded18f8f3
1e28e08b0f6fdeab9edf506c31b7358775b266c2
29185 F20101220_AABJYF liu_w_Page_099.QC.jpg
4890a415ed093923c4297fa26be80e0e
3462ad859cf45eaea2b9c658c24cddd9e21ca33d
7778 F20101220_AABJXR liu_w_Page_091thm.jpg
57bd2ebbccf991ae2c50e5cb6f2dda65
0b0626740b7eca15b36563f7b40a671ac431744d
F20101220_AABJAJ liu_w_Page_029.tif
f7b0119c58e59b9686d7cdf452e69a9c
c49b1547787d2c6d6ba8daa4d2ef6f229e5a85a4
1051976 F20101220_AABIVD liu_w_Page_051.jp2
b063d892f78d91380b7955fc31218645
abead04b919aa503e005d8878e2191a37709cbf2
1051944 F20101220_AABIUP liu_w_Page_036.jp2
1764389d1f5690eecd11fb1388bf3143
5e1eb020a049b264610caa9a216f52669d63ff29
10895 F20101220_AABKCY liu_w_Page_163.QC.jpg
bf4d4ca4cf656853358dfb1e14f7fe5f
cb6e03c266c87e17f1205e9388029f12ef787c21
7648 F20101220_AABJYG liu_w_Page_099thm.jpg
793aed4eca59aeb0d598ebdbd3545935
a08a67d28dae7ef004185ac1231ee285ebb0c0c1
F20101220_AABJAK liu_w_Page_030.tif
db88947c526e060dacf4da76f63e7c97
d42cbf7cbe6b942fdd288e2b5a83534db0cb6bec
F20101220_AABIVE liu_w_Page_052.jp2
88301c7c3c82268b9ec72212754f1315
9d6469abccb3a826af145064b73d13a3520f6011
2698 F20101220_AABKCZ liu_w_Page_163thm.jpg
83d0d319a34a9fef7f03dd0b594e1ead
60659dc7fafa38f608b3a4ef9caa926e50bdaff1
40985 F20101220_AABJYH liu_w_Page_100.QC.jpg
781f97c7aef2edf549a3cad1050070b9
4c803cdaaa18cc8f78c3a12748852b6e38eca8c9
28846 F20101220_AABJXS liu_w_Page_092.QC.jpg
d84c3a743fb7ee222df5accd34f94c89
fd9544670a6eba67d30092152916245b9d7dd596
F20101220_AABJAL liu_w_Page_031.tif
ff931330164762ecc160da4b44f3d587
a075f4641e419271d33d370f1774c1f3904c3cfc
F20101220_AABIVF liu_w_Page_053.jp2
8e58a876d9842034bd45f1a7e4308865
5c2e00aeddaa965434476104ea04b931f8b41bd2
1051978 F20101220_AABIUQ liu_w_Page_037.jp2
e98048f9a801c57eeded4911ec67f241
264cf386c1fb516cb05ca6b52bed266eedc50f0e
9358 F20101220_AABJYI liu_w_Page_100thm.jpg
d9c7f107621b6725e35902a776333c2a
d9586ec4c5ba288a40a6a8eef38c4eac9a8185c2
7791 F20101220_AABJXT liu_w_Page_092thm.jpg
b44ea570e00a926b3da3994e18f10811
eff1a6c40d6dbf948728ed89fc033ef288ca979a
F20101220_AABJBA liu_w_Page_047.tif
3d5ff4e970d0580f395aff1044c18eec
d3470d5317afde80c69e0d493def8871c89976ab
F20101220_AABJAM liu_w_Page_032.tif
7f6a24abf15b774ff4a92c4283abb60a
0fd80daf17de2c8a63a93c5778c9b2062f7cee03
F20101220_AABIVG liu_w_Page_054.jp2
421f858cbe361a089a229ee25b3600e7
2ef32e356506382808465a09a8961c0dbde6be12
1051965 F20101220_AABIUR liu_w_Page_038.jp2
3e2f2e153b355cd684ae7333539d03ba
4ac0f71c94a1ddc349c3e1c79118469402a2bd9e
39166 F20101220_AABJYJ liu_w_Page_101.QC.jpg
14db10d4c721bd1688e13c69d9056d7b
482d21d5f0b0f0629d411b819a45e18f5348a3c4
22289 F20101220_AABJXU liu_w_Page_093.QC.jpg
c8cfbeb8774c5679ee7636f9369e5661
245a49cf72e433e99204cb489d4d9f91704ea6a9
F20101220_AABJBB liu_w_Page_048.tif
00e69eae36bec52962811131e6105bd2
fd8f8c4f8eff279ca4ccc8d8100b001c804f1ee7
F20101220_AABJAN liu_w_Page_034.tif
377240b7beb12fcb62f6731bc11ecb20
67d4069283aea995bffe9175f09b8744cb1489e3
396351 F20101220_AABIVH liu_w_Page_055.jp2
1e0e7aa1343257717a13bc94424be865
9c029baf06da997c18d50cf9356f9779586f55d3
1051959 F20101220_AABIUS liu_w_Page_039.jp2
f3f0a590a1da45081dc9e23873b07b6c
a54d98534252ba1c954d1892e12cb76631642950
9035 F20101220_AABJYK liu_w_Page_101thm.jpg
cffcc6e1832a7a3a3cf6148dec3fa01b
2a2df18013b3a02502355f8418cf3ad4e04e1667
5949 F20101220_AABJXV liu_w_Page_093thm.jpg
7a09f197af90885d0aeb562cbe0f225a
2ab8d238f74009b1fe020fabd434de223ca68349
F20101220_AABJBC liu_w_Page_049.tif
15c00b8c37d9cffb31fc05cc559d4686
e9c2e274dc79623a05c94d37d862e283babaacc2
F20101220_AABJAO liu_w_Page_035.tif
376829b034c2f635bd52409c88f26920
5e44d75ed6dce44bb1fa0b8f28ee6f70f2106eee
1051932 F20101220_AABIVI liu_w_Page_056.jp2
28a2974c379a137c733e04436c308a3c
dde6cf968bb8e207a3db8fb5aab7d1ebafca4840
1051953 F20101220_AABIUT liu_w_Page_040.jp2
004c2fd9523000b903f492c31ce5243a
221cf636ac737e820bc85d11c6b12dba0a7ae628
21347 F20101220_AABJYL liu_w_Page_102.QC.jpg
541e242a4e8cee684c9ba2b0082ec967
116e56f99df517a9dcacf8f68acdd0274999062e
34826 F20101220_AABJXW liu_w_Page_094.QC.jpg
b1502c84404a93930c6615a49c529b1f
bf860af57cb3331b4e7c6e4d6ec5278b1e7f4463
F20101220_AABJBD liu_w_Page_050.tif
5d8256c2f579dec7a07ae0268c42bc6d
a95552be98c3b07bb4350607d28da569569236e5
F20101220_AABJAP liu_w_Page_036.tif
b0627920a49ee3db9a018d6abbfffe1d
67a1aa3579626fe494964469afa8924a7146e98e
408736 F20101220_AABIVJ liu_w_Page_057.jp2
aad5079b69097ed2a603300ddd2ca907
40cb7bb377523427ff699d7f19c350832ca3452a
F20101220_AABIUU liu_w_Page_041.jp2
a7ab99d1859dc530da04d1a9c9786461
dcf0b102d903fd5dd4a38210287c729ab39ef171
6670 F20101220_AABJZA liu_w_Page_109thm.jpg
5b0a42735d58c7899d2d50461696f9e3
6f0410c12dc8c3df3924dae5eb7237f81702f859
5450 F20101220_AABJYM liu_w_Page_102thm.jpg
c09e5bbceb5c3f42305f40a5f6eb8778
65842bf264ddeb99f571b20d24b62e15cb8ce9cb
F20101220_AABJXX liu_w_Page_094thm.jpg
20c29929a093ed8ec7f91de14051fa8c
a26afea0f67528320b4effcaff706bd495407c5e
F20101220_AABJBE liu_w_Page_051.tif
d609a8e46b9b786e7bc2e871e99d1146
c96a8a38a4eacf0c0e3d2c38e12a4a1c33943117
F20101220_AABJAQ liu_w_Page_037.tif
26e00a57b21cbeedebd6ce322c0d8354
8686cfb2cda2713403dda3239c1291cf828efdc2
707418 F20101220_AABIVK liu_w_Page_058.jp2
8fef2b231a58c3e041e635bd7876320e
d08f2bd2bff17d8ae0c37be3b405ad7a0696731b
F20101220_AABIUV liu_w_Page_042.jp2
8e30c4d6d9fe406bd1ce6ee16f93f637
3cd5ea86968762f636af1c92fc85d7f147165182
37764 F20101220_AABJZB liu_w_Page_110.QC.jpg
9dfac47aaaacf172bac5ca9cf2b19d3d
bff8eb7274545fafdfb162d83b02167fae8b5b54
36898 F20101220_AABJYN liu_w_Page_103.QC.jpg
1799716b47ab7f4073fd487cf9c10e35
9cedd20ef39541fe8d9c1e304ec08301175c1222
35270 F20101220_AABJXY liu_w_Page_095.QC.jpg
abaa7f26858a5c95c8d52461aaa51cc9
a542c435abb01cb53fb6a96f3c1afcf48e2f552e
F20101220_AABJBF liu_w_Page_052.tif
bd99f69d29694ab3088a1cc2706b0cbb
e9a5fa5aaf48f0f38fde76122ea56566d1fe3659
F20101220_AABJAR liu_w_Page_038.tif
1930a4788d0f9ece8f209fe31b3dd4ae
ddad74716fd26848d6f3898cc9d2190130db040a
F20101220_AABIVL liu_w_Page_059.jp2
c5f2dfa4424d511f0a43098695789930
e78a9f0bb19de16decd64607d630722df460fbb6
1002656 F20101220_AABIUW liu_w_Page_044.jp2
cdf3f3487a27e63c0c591c38705ba984
d6a865a665a4b61f07cb4d6a6831ecd37414a040
8790 F20101220_AABJZC liu_w_Page_110thm.jpg
e802d4e26fb2e2079c74a6ed23104518
25bdf997204b0fe5c990633523c8f8be07d975b0
8900 F20101220_AABJYO liu_w_Page_103thm.jpg
6e75728cb3fdde180387dbc5bdc520c4
5a6c53381d7eb83c5f009daa0bb4e2ebed3b19ab
8357 F20101220_AABJXZ liu_w_Page_095thm.jpg
fd18647836daf748fc1bcab2bf64e3b9
077b019ca7a6e3503cc729f3ddb6da2e236ac1af
F20101220_AABJBG liu_w_Page_053.tif
301868ec58b928fa48bc4ce11678570e
120cf00a0231f1291ad6b12ef1d02426f498954b
451974 F20101220_AABIWA liu_w_Page_074.jp2
6b1daf5a9103088b13cad5b04e298448
c2d00e34f72e7ef26e369be6da395d89aea02c6d
F20101220_AABJAS liu_w_Page_039.tif
49151f4037035411ac229b380946a8af
c3216562c32807eb69948a70ad5cdb4c902b43f7
626484 F20101220_AABIVM liu_w_Page_060.jp2
b0d3084163c86804db0adfbfe0b2f45b
a5eb11aa1fb5495f9b2a5158e95b84fa13a856e8
945257 F20101220_AABIUX liu_w_Page_045.jp2
1a8803eb9b268cbc5f82da1caabc6631
b0d23baf717b47b7b11dc001b03e3b78abd11e9a
37398 F20101220_AABJZD liu_w_Page_111.QC.jpg
6051b5d020d61a9f00641240139cb776
a52117afd72377790b9338bf8338349110033d71
37587 F20101220_AABJYP liu_w_Page_104.QC.jpg
7f4d8133fd8b954070ea906a49c48074
f6d96b9806931134a005ba9f744d768a85cf1411
F20101220_AABJBH liu_w_Page_054.tif
6e517ba44f235c6dfa22b4091f4a864c
ce6c518f672eb713d290a42c2c5606207dddb245
910495 F20101220_AABIWB liu_w_Page_075.jp2
b77308f13f3cf43f027a89a4b5c0bb9c
7147e379f28b86fc30c732445c69f1ae74efa3ef
F20101220_AABJAT liu_w_Page_040.tif
75cfc8c1a2f8a359b9dbbb691d04938a
da542b492b8609c644cb0588141e436f85fab6ec
F20101220_AABIVN liu_w_Page_061.jp2
1fc419f00a35edba2a253063bfda3a80
518b96921ff8fd37df5a16e8bf584a49f134586a
692617 F20101220_AABIUY liu_w_Page_046.jp2
a26e2faf24261610e00b525b39cfead7
6b5c7bd8f39ba4128c48e5dd294542bf627dcf0f
8933 F20101220_AABJZE liu_w_Page_111thm.jpg
3e9aa0dd60680293c4909612708f6318
0c4b96ec316593482a9d96d22b96547f89bacbf1
8881 F20101220_AABJYQ liu_w_Page_104thm.jpg
91794229e3306ad7e02ddb36e6144930
836758d25860285589dd2d3b2262ccd33f55e3ad
F20101220_AABJBI liu_w_Page_055.tif
c4208e313e4bbe2242313d99a995cf0e
31b6b2227ed39f2ec9d282ede65089ca515717e2
1051973 F20101220_AABIWC liu_w_Page_076.jp2
a5f1aed2a28e44c60149f6afab3dafcf
dff7b84930cc160969ead38ea3d17b66f1424b53
F20101220_AABJAU liu_w_Page_041.tif
a5dd01c604260ae1870192a3dba65b90
3117f484b7964a32b941889191b1cd215ef7dcdf
1051984 F20101220_AABIVO liu_w_Page_062.jp2
1ac8b77a4fe7be093e6faa7ecaa5e40c
63bfa00247d5bb2bac64267c640340b8e4a17c47
F20101220_AABIUZ liu_w_Page_047.jp2
10ef3cd5e804c22478851db7a52ce9cc
fb22cd9f460e89c0f0b1315e8a13f88039964b33
32225 F20101220_AABJZF liu_w_Page_112.QC.jpg
187781f3b351389f7164e17519f82c78
66fc558ac2485ce8d67c72103ee73c7695d5b7b8
36223 F20101220_AABJYR liu_w_Page_105.QC.jpg
0549fb4880ccbefddada3081129e6dd1
16b6ce7d9a50572998f7fea843953edc11b12743
F20101220_AABIWD liu_w_Page_077.jp2
b96677ce9b92aff01a98a8149372cf34
15abf5e86a5b75241b7d88bad897ec778cceb864
F20101220_AABJAV liu_w_Page_042.tif
024b7658646a00c69b5fc62c3d8c62f3
df0875105c29ca219385e81fbd3d0d5d236dd464
F20101220_AABIVP liu_w_Page_063.jp2
ca357a2ab8443c55ebc0cde02fbae661
460ce0b856a75c516c8fa23b88a7b739e3d769e9
F20101220_AABJBJ liu_w_Page_056.tif
9a7afedba4eaa01d55b0d5489dc376bb
497d820a9f2a60a36d7dbd33290fc4a2b83c1fa3
7963 F20101220_AABJZG liu_w_Page_112thm.jpg
c53e3562648803bfe6d21c25b2b36f96
15eb84f6fecccf8720be80009c1329f339b2f986
8652 F20101220_AABJYS liu_w_Page_105thm.jpg
cd0f19e86c52e3df789c3f0a1cd45884
777a03951e3ab8078f60e6219cbae22bff7b6bc9
1051952 F20101220_AABIWE liu_w_Page_078.jp2
a3f3d86b439bc4e7f4b76f6f98604b99
cd0b4bfb07190088bb5dd5c364412a7ca9a80152
F20101220_AABJAW liu_w_Page_043.tif
e4cdfc666b6f0a6171b3a8a754c1e24a
7bca63707ff69cd95e6c0a742e4115c2dad024a2
1051975 F20101220_AABIVQ liu_w_Page_064.jp2
4b5ffb0fde82e4e970e22904c330e98f
c5f6470a28806cfdbe74e0e2110997dc8a9732ab
F20101220_AABJBK liu_w_Page_057.tif
68a85e7c92d45b90b5b28ccb46a7bbc9
8e446f1b51ee7d730fbeaa5e4d86f1ee0a92eda2
31211 F20101220_AABJZH liu_w_Page_113.QC.jpg
19aa14cc031d44efe50e36a57a7a9c86
0772b3ad40ae32a1d8f4d5939e90d70237171692
1029765 F20101220_AABIWF liu_w_Page_079.jp2
972ea7f695dfaa92a564919beb277504
a8ef0562ee462c41e066f35b6f0db070d92a2a6a
F20101220_AABJAX liu_w_Page_044.tif
cd12768e0512c779e503d763f4caac58
a6faabb74b2520e08411be71156d8ef26c1ad54c
F20101220_AABJBL liu_w_Page_058.tif
7d0fdc6b6cdd8407003190b907d1bb32
016a93279e6f8fb9c89963c568eb052893f74ba6
7821 F20101220_AABJZI liu_w_Page_113thm.jpg
77927fcf3f65cd2ca8c827bdce8378a4
803d7fb15d68e4fe7f4349dbdad378b27d274c9e
33252 F20101220_AABJYT liu_w_Page_106.QC.jpg
9fd0959c86ad6fd0e048a2d10f50d0c4
d539eb1b84bb00306ce95fafa310dacb4b6a056e
F20101220_AABIWG liu_w_Page_080.jp2
07093e5fe04aced9aa9711e1d83e0e44
46108bce0e706172b8a740f08fd3895e333cfdc4
F20101220_AABJAY liu_w_Page_045.tif
1d83f01491731cf957a58460db259460
4987f0a4bea95d7e07c55226a0ac761a0858093e
1051949 F20101220_AABIVR liu_w_Page_065.jp2
bc74b13562d926c07a8ab70d5cd12688
b9b24112c4c6f701eff02a272c32e8057ea3d561
F20101220_AABJCA liu_w_Page_074.tif
0a9ea831f2e06284592987893399556e
8eac83a61cfad980a27993c80ea04883955fb6b8
F20101220_AABJBM liu_w_Page_059.tif
5215849b9ed038bcb893b1ad962dca02
06805f220d0fd44d5064bbb4e9cfa2c11b77e074
13590 F20101220_AABJZJ liu_w_Page_114.QC.jpg
58353e25e620ca3af53af0dafb25e02e
e00ae68997d45705aeef49d758eda8175cd166a2
8091 F20101220_AABJYU liu_w_Page_106thm.jpg
cda0a3b0692e5750fd84a565910aeafc
0c8b09d07b85f6d5c2f0ce6a48a6d159cc4e9b60
553567 F20101220_AABIWH liu_w_Page_081.jp2
7cd5bd20ff03594f3e163f022055497b
8540845964df40a925b1a592d386e4989e607537
F20101220_AABJAZ liu_w_Page_046.tif
6b281309e620572e2e11a5bffa31e856
8112c4ae6d28dccaadec2087c1a63f77d46fc616
F20101220_AABIVS liu_w_Page_066.jp2
3feb140bc090bf00a527500b35ba6739
9accc846115818f3482150ef9d1224c37f14321c
F20101220_AABJCB liu_w_Page_075.tif
f9608e3cb331b9e5c80ab17df53363d3
274c72aa4a344e9b9dff8b9d4c3db9afd829c50b
F20101220_AABJBN liu_w_Page_060.tif
a37ff6a1ccaf134a7812cb904f9f1be8
d381ba94bb98439843152edf67f7b1c9402656e2
39530 F20101220_AABJZK liu_w_Page_115.QC.jpg
3289396cde676455b2a72686d0f6ec36
7f3f1d6305998c180d4e26336d1b1e629af3f093
38193 F20101220_AABJYV liu_w_Page_107.QC.jpg
25b8ccefdbca6958de0c8ed52abb9fae
2f8977209c388459d53bbaf391a7ceddbf8194f2
1051955 F20101220_AABIWI liu_w_Page_082.jp2
fca32af498510ba0e1749caa3da9a99c
c333d297391e446ba64a2dec211e08c96c5d8fdd
882534 F20101220_AABIVT liu_w_Page_067.jp2
85df7895cd867aa56f6997c7a92f3faf
525fa0b1587fe0018cce33f8cb0127e8c73ea168
F20101220_AABJCC liu_w_Page_076.tif
0a476c5794deca63c824a62f9259df44
b094bfca8dddb56162a10c99e1a030fd7a6996e9
F20101220_AABJBO liu_w_Page_061.tif
5ec04e740ceecb252491dd2c4910124f
0769d61c91981a9ec59d9c16d13630e80cab2b2f
9249 F20101220_AABJZL liu_w_Page_115thm.jpg
54fe944eaf7a3d69f69161cad5600b11
2df88a8811ffa52a046da2106d2a0eb45880a7a3
9055 F20101220_AABJYW liu_w_Page_107thm.jpg
32d8e3b9befe6122c1ea6ff3ca0e8b26
5dffcf68519f4b75f900ec69f1662c103540d5e8
558530 F20101220_AABIWJ liu_w_Page_083.jp2
945010c5e3da5bfff2c81eead362d175
3bb0a3993ab355ededd8ab8a0c4458c5c8dc2f61
F20101220_AABIVU liu_w_Page_068.jp2
e4c2864719d549f0c125acb67a9baad6
cd9aa23f4e0599253de81239bcd13dc41ee60c37
F20101220_AABJCD liu_w_Page_077.tif
794640c6dc30bca7f7c7c661f780de0d
ef39b6f6e83f25384a3af154ce69618ce1e3ce69
F20101220_AABJBP liu_w_Page_062.tif
90fc176b35dce39a9f34898ba12a8ebe
89d2b53bf367c6773e027747a41e5ac1ff697c29
38176 F20101220_AABJZM liu_w_Page_116.QC.jpg
da0efb26970e7b803c0ddd6760e88b06
9df245374f2675f0896c051540a244668b0d3dfc
25030 F20101220_AABJYX liu_w_Page_108.QC.jpg
e968fd5189220116b262e6c566e3a9f5
b6b12d5eba0371d541a189d122de246816739d3c
F20101220_AABIWK liu_w_Page_086.jp2
638d2b82a59479ba35a296b5be79570e
eb2758749361a7cf3191582062ad86ab5e6bd2a4
F20101220_AABIVV liu_w_Page_069.jp2
7766c1709c91514af0fbd7756b670dbe
431ff4f4c7832c4a9119853e3ee523b9b4b8a4d1
F20101220_AABJCE liu_w_Page_078.tif
1039cf62600ecf060eb3cbd4fb1d8529
693439dacb515e0cbb18c223dab60df27860f3a7
F20101220_AABJBQ liu_w_Page_063.tif
aeffd351ac1e0707d8857cbba4233ff1
10db41d5b2af8d6ab7599d2bbf12b18ddf2e205c
8816 F20101220_AABJZN liu_w_Page_116thm.jpg
d34e2e98d33fb55b5ceda77d47e34021
b93a0a8f992ac16054c188daabc90688d9353e68
6246 F20101220_AABJYY liu_w_Page_108thm.jpg
630fa206e4ed96fe03b1b0da788f5bd7
71d989ec1cde64ee89998f578894ddaf2a4f811b
1051940 F20101220_AABIWL liu_w_Page_087.jp2
57682c70c43cd24daacead0c61ef9de4
d01681bc75571dbcc85a78a6c11eec809203de5f
1051954 F20101220_AABIVW liu_w_Page_070.jp2
05eb7158bac6933daa9bdbfd699c067f
635e27653bf49ad71276cc77752186686b7b44b2
F20101220_AABJCF liu_w_Page_079.tif
50f8d7a5cadb9c632559472c8d390b76
bcde06438ae0b2e89bb7f1c6e87609a0f2728ef3
F20101220_AABJBR liu_w_Page_065.tif
6d3435f00e2b6287f0e73751aad7afa4
d77d47750d2ba09a815eda06585e4870307a33d7
12105 F20101220_AABJZO liu_w_Page_117.QC.jpg
8a75b118d7f0f552d2724a3724644478
eaf87a1b901ff8a173763c61426dd03b4036577c
26114 F20101220_AABJYZ liu_w_Page_109.QC.jpg
d395184db5c8e695da995d3677522548
fa87eb8cb996cd0a6a5277e1e007053bfa5527f8
F20101220_AABIWM liu_w_Page_088.jp2
e5030f9646cd75faf39d06fcdb988056
5e9683b768a85e27babe6ede8539765c07739836
726109 F20101220_AABIVX liu_w_Page_071.jp2
302bab153bb505375425694fc3ed110f
14d4d711f687f081accb47b8924559c6751f4bdd
F20101220_AABJCG liu_w_Page_080.tif
7059f020ea44a35bac9608fb4b42f75d
7a41fd56891b229139eae227433b6d7a8dcde4ca
734063 F20101220_AABIXA liu_w_Page_102.jp2
aa6ee82d67080205df1a24d1cfa16e6c
16b57b8dc228a99ca614f7104fbd3c858d739354
F20101220_AABJBS liu_w_Page_066.tif
57cd32b56232aa51a91e627a7380fa78
38efc5037f96c72f83437a7fc7c18438f9b9d3e0
2815 F20101220_AABJZP liu_w_Page_117thm.jpg
231970b25002379c52ebd1a18cf20f35
142c676213a7aeb205fe7f53c4ba041c8c052a04
1024883 F20101220_AABIWN liu_w_Page_089.jp2
eacb51666ffa229d671e435833c48a1d
715d5efb37e4492d9e1a17b6be55b90d573ba308
1024616 F20101220_AABIVY liu_w_Page_072.jp2
a44797b050c022422976463275193b64
b8cb5d8c242e4bd63943ef0b4a5ff6aac4822805
F20101220_AABJCH liu_w_Page_081.tif
65797824d9b4d3552e6fae8a9f7236db
65c9424103de903b97abc898a360103f3b4187dd
F20101220_AABIXB liu_w_Page_103.jp2
cc66259f624fa555a902ec1b3f1264f8
271d7f3e8d38396521a47171abcf6f23b65e1d1a
F20101220_AABJBT liu_w_Page_067.tif
a108d0c8169e5496d3eaa971b1d5e5cf
c4532ab742abdb140e5337e6c0fe6103643478e7
32516 F20101220_AABJZQ liu_w_Page_118.QC.jpg
fd0216ebd15d35f33ec0d96bbd3f609b
3d1a8dbe3919451f8dccdfd9dc4ded86d0e04464
F20101220_AABIWO liu_w_Page_090.jp2
b6100e752e874d002dad44ccf09d121c
9c307a131aca09fd797df692b75afd2b3de01ecb
1033605 F20101220_AABIVZ liu_w_Page_073.jp2
8f71f1258488c571db256f7aad9e153f
76ea9f1fd49134bd3687e6b18a8d56656e847a7c
F20101220_AABJCI liu_w_Page_082.tif
a5f58cb2232c18ee563ef77ce2e8adb8
9733d8b88004ce19f9939fc15d47fcc25fde44ea
F20101220_AABIXC liu_w_Page_104.jp2
8d0ea42147b61c830b8350903367e729
6242ba11210752d3359d075ba1fc50eb5c015a0a
F20101220_AABJBU liu_w_Page_068.tif
d7f4db4b075d47ef7dbf8bbf3548d322
20f5e63936c8a8d8821e490f94e43d1d34078ec0
7862 F20101220_AABJZR liu_w_Page_118thm.jpg
14ab7e049749911713d374e4b435a20e
b9b6c67ce63e87185a2b0c3f955f6325ebea43dd
1021773 F20101220_AABIWP liu_w_Page_091.jp2
371310f93bb8dcb54af40c2e888ca7d2
b5b7fdca676cf1987a670c4448b2d93bb5deeecc
F20101220_AABJCJ liu_w_Page_083.tif
5cadd07e2d1e863fedcd3987b0eaac0e
24e4ae7aa20800cb24721d48f19f733d79845951
1051963 F20101220_AABIXD liu_w_Page_105.jp2
d5dfb7a22a75b77be54e7513bc0ea660
1c34f86a0fd84dcf019e899184516585c7157126
F20101220_AABJBV liu_w_Page_069.tif
ff82478807298d74529b59b0ba60b94b
4b69d011d9c1b022538c8a86da747bdf523fe77f
39548 F20101220_AABJZS liu_w_Page_119.QC.jpg
436a2647ad2176b39a9fe72cb980925e
62ad831919bac3092427ec6dfdb266d3cc66e66a
947242 F20101220_AABIWQ liu_w_Page_092.jp2
1be40bc82ba6c926a2e8a69579899f0e
479e7c99310bf06dbdc65ebcea4e54f0fe7f9cca
F20101220_AABJCK liu_w_Page_084.tif
f533fb4d69435d01f563c368b16309d8
eecd5d7d6554eca2d53feed8fe50631df3493520
F20101220_AABIXE liu_w_Page_106.jp2
49d80bcd0cec8d6a14f376038e8507d3
2633f3441bf7bf8a8bb88a865efc812605ba2239
F20101220_AABJBW liu_w_Page_070.tif
5340019758ca2a9083e63800dc3bbe38
f409b7321e14ada57179fb82b6f3d2608d945a27
38680 F20101220_AABJZT liu_w_Page_120.QC.jpg
1d62d717b0dc334b186ca0fffd3e14b0
96c145223899d715f377c622963bbd2826ef950c
697486 F20101220_AABIWR liu_w_Page_093.jp2
5dc2ae7666e5939323c4ee9172b668a6
5e658eeb0fde5fcd8f2672d45fb55710bfd5d71b
F20101220_AABJCL liu_w_Page_085.tif
f9443ecf763a0e0c7eacbd35fca2b1ae
9647e80af3f67be7d2cfc7e243f26d4b4bbd7668
1051983 F20101220_AABIXF liu_w_Page_107.jp2
29c411d17da5d300c947df048a6154f7
babf70e1515ebc9ae611208f96275465b22fb700
F20101220_AABJBX liu_w_Page_071.tif
7cf4626afa0cfebcf01c1b2e08b7cca7
ce5aa5cf5f0591b5ff48e0b3c0572f3a519a396c
F20101220_AABJDA liu_w_Page_102.tif
82ad785235b16d27d41926c5015f8b84
abd73c4990157e3cc8841bff1688472e8a30517a
F20101220_AABJCM liu_w_Page_086.tif
e25d8d7ac187de658d66a3756e7e80cc
2e4b310b59297669510abd5173ae91842e953464
777119 F20101220_AABIXG liu_w_Page_108.jp2
179462f762ab40d42960959408faf915
a7f5283ca5563957fec068b0da994a9aaa7dfc0a
F20101220_AABJBY liu_w_Page_072.tif
f6c287510c3cb7902bb0a040cb7dfd81
4b4134792d928c35f750215ee8a25e88032a278b
9103 F20101220_AABJZU liu_w_Page_120thm.jpg
da6fd3c1bcd56e38a7f2b2b19786763b
d5576087bd59e21478f087db794ae7a898e812d6
F20101220_AABIWS liu_w_Page_094.jp2
e313fb1bf814f7ed0dae71d68956ae31
0864d2136959278eb94b96e5705043b0aa3d0270
F20101220_AABJDB liu_w_Page_103.tif
034d602f7fcb2b10e0fee299563c7cab
6481fddced0d261f66ada0beee415ea813d6740f
F20101220_AABJCN liu_w_Page_087.tif
62ce23fa6a448e2e46a773482dc6278a
8cfcbef1f2df8059395baf01627d27605e12baa8
906817 F20101220_AABIXH liu_w_Page_109.jp2
754787ef4e8c320146dd2909c544dd15
bba615a070e533cb6c8b2999e10f12c3629c5432
F20101220_AABJBZ liu_w_Page_073.tif
c80f96d7bbe21c6637d822704d30fb36
d52b0098359895000f90564c234fb8fd4a330773
38660 F20101220_AABJZV liu_w_Page_121.QC.jpg
47fc5abdcfd065a58594183d252810d9
f732b2c4f6f28683de81b3257dbabdffec0d5a57
F20101220_AABIWT liu_w_Page_095.jp2
58dda7ea78a56a10e06d53a9e09a1b22
44f542e2e98773f1375978b2eea89991b1eb1fd5
F20101220_AABJDC liu_w_Page_104.tif
842a48294ebb384aa45adca19bad4afe
7bcd98767b025842936105c2a2a2a4f5fa0f38e4
F20101220_AABJCO liu_w_Page_089.tif
daae6c54a46192465de354825e31eee1
f3631dad7c3cf35a8c2bc47e0738e5c652f07d31
1051914 F20101220_AABIXI liu_w_Page_110.jp2
b4692c8d97f6c1541036b13f58d43cf2
1254b87a7473c3425964bfb1b22b4cb8c94a1089
8813 F20101220_AABJZW liu_w_Page_121thm.jpg
9fbf4dc51df1dff3a1f300e31ce31f85
8c9d96f0fddc9d8ce5ca39dae8f156c4219c207f
F20101220_AABIWU liu_w_Page_096.jp2
3d0c9cc329348be19611ee56c45da864
04ee251f4e7feac01e62ddd50189969c7cf467a5
F20101220_AABJDD liu_w_Page_105.tif
c999aade963e7c1cbdeae747ecb4d0ee
a0c479987c89f4b26f389b678e2f6f96ed7ea9e3
F20101220_AABJCP liu_w_Page_090.tif
c1b0537e2ed19e466cb77a84abf4f91e
6c8839a8ea879fba6a84d75be6808764a3c63b12
F20101220_AABIXJ liu_w_Page_111.jp2
d8b20501dc1c332031ac05341df5e2b1
66ffaac2bc2d2d92848308181e6189171b5f34a0
39235 F20101220_AABJZX liu_w_Page_122.QC.jpg
eba64d2890a667838e7cfdbce15cf0e5
f9c8b57801cc9ba4e54b5b937fe48d49c8d0d3cc
F20101220_AABIWV liu_w_Page_097.jp2
c32de7977264a334e5fa05b2421d0095
ebe7683e7513316dc14e1901493f76ca688b7c29
F20101220_AABJDE liu_w_Page_106.tif
b5ccea87c20ff6ca22a6ee174d0a0ae2
ea16f587dac4ae829fe117f8198cd15d4605b634
F20101220_AABJCQ liu_w_Page_092.tif
32bc09950b4480559d68be5ed5bab107
1d6d722f137f566356f38e93e6d8742caaee2a78
1051960 F20101220_AABIXK liu_w_Page_112.jp2
3e027a57eef9cfec90ad62d104008d3b
06c834d001bf185704b7c49221f2784e0954c308
9275 F20101220_AABJZY liu_w_Page_122thm.jpg
1bfa6cadd2f798b63c27802f6e6ede69
5573f67cf0cff54db7dbe4a59f099737d18cfb8f
F20101220_AABIWW liu_w_Page_098.jp2
cb89df245ed48a670d11192cf6fb275f
e72766800e9d7173d16ed000a069c8ddd9e9d7b2
F20101220_AABJDF liu_w_Page_107.tif
2e62abd5de7e445a5641d1eef13bcfe2
b3e985c6bf089dc2d8b59599ff175fd36cf3a1ce
F20101220_AABJCR liu_w_Page_093.tif
7335542b8d2b4e82305b5dcb0ec9d358
7041ee0345f9dc7c0ef15d20494f3c76daca5ea3
1042137 F20101220_AABIXL liu_w_Page_113.jp2
84d2e5cc10e52c6eae108ffdcca60d57
38e25ea0a4f8d108b4b8388f789d6bff2cf250bf
32735 F20101220_AABJZZ liu_w_Page_123.QC.jpg
35921c9ef7187c34a6ba47991c64e18d
3bed1e780ebd2efa885d5ed9e84dfe9c769fb9b7
1028639 F20101220_AABIWX liu_w_Page_099.jp2
3603a14e5c80a6eb5ae46b78f9304b73
e13e3e2e40c5367859bf4f494ed20f0f1173a581
F20101220_AABJDG liu_w_Page_108.tif
d8eabacb4db2788d37446ab17eb0a57f
9f964ede501d6d252390f9dc9a34b3d77ad6a14e
F20101220_AABIYA liu_w_Page_130.jp2
f0a0c4de1a7e301919da8f0ef066e785
bf4f6da1f2d39a91e2122d2c35bf4c2ff7232666
F20101220_AABJCS liu_w_Page_094.tif
e6ce93f7ef24075d617bd8908dfe85eb
48d6af3f6b7377c612bab2bddffc089a6ec55b00
597906 F20101220_AABIXM liu_w_Page_114.jp2
e2fcfaa47bbb424bbebdc8096933eab2
91be820affe28c836338c28396aed09c275f92a9
F20101220_AABIWY liu_w_Page_100.jp2
0d0270f6809d73047c57d596d7a2bb2d
57c376c98cd7f298262d40d1ba5a725964f41b99
F20101220_AABJDH liu_w_Page_109.tif
ea66eec1b0704135531aac63040a4871
9c1537bca82d0b593e9b8c59595f1731ba936a97
F20101220_AABIYB liu_w_Page_131.jp2
b9c90fb304fe7da9d711540528e8af9d
d809acae5ee58c9832314d13268b2f47d143e206
F20101220_AABJCT liu_w_Page_095.tif
27bb1382fff91ed289cc474abf799d53
b124e307de51ccde69de4fe02c8e8d607be5834c
1051945 F20101220_AABIXN liu_w_Page_115.jp2
a1a8679830eec178a4f435a1c2d4cc7b
f69e7eda344eac756c4814b1484fbf08312a9126
1051937 F20101220_AABIWZ liu_w_Page_101.jp2
3475692cb086b1a2836bd08c0886dbb9
b132575d4e4ce582dd781f437a612917c32c5762
F20101220_AABJDI liu_w_Page_110.tif
40c8c3c18865ae1fd996a05c2b4436ec
352f07b1db02124e390e05b1f8414d3368f6ddf8
578994 F20101220_AABIYC liu_w_Page_132.jp2
de31816f1c0ebe089e5008e06cbd3ce0
09696adf441dca8dbad85e7823c2464b7d1b21ec
F20101220_AABJCU liu_w_Page_096.tif
abd8ac1ed88e492a02a89b19d3668270
a95c8a77b10e5419f4270c4eefc6888ba1397baa
1051964 F20101220_AABIXO liu_w_Page_116.jp2
44e433360388fdc19f5a1e0fcab9d0c6
4c860eaf22dc3f8da1b8b26abe3c162719850566
F20101220_AABJDJ liu_w_Page_111.tif
8b039b753016a9de5875e597d0f1b41a
8562d594fed0b4eed2419e83a9cb4b68bb95ed45
1051873 F20101220_AABIYD liu_w_Page_133.jp2
e6df3d910ecd564dba09704b0a442d4e
5bb62d25385f01b39712c151b833f81690a884c3
F20101220_AABJCV liu_w_Page_097.tif
fd916399a7b49637779f4e1e43dac0be
10dd88353c9a26319a7a7fb3b44c709df586f0f4
399042 F20101220_AABIXP liu_w_Page_117.jp2
d3d65909d6c66dfc69e4cd7f60fd8001
529dec3842f1be5f09dadd88bd957bf8cd32cd4a
F20101220_AABJDK liu_w_Page_112.tif
fb21ddac280f7eefb0f8fd4fc2568605
278c91524ed60aac9b7c7f105463867cfcfa6903
F20101220_AABIYE liu_w_Page_134.jp2
f97111ff7fc0e5d34342c3f00034da9a
a6002aa18c0c568f525b560d05379032f36c7fc2
F20101220_AABJCW liu_w_Page_098.tif
ae8d3e0844da8a7d7aca5f760521a953
e5ea970de18d73b4842acc3afbd4e744bfaba9a8
1051909 F20101220_AABIXQ liu_w_Page_118.jp2
f9c8bdb10a6967667175327ce9652cd7
950afaded65d5f7c22f19ba1306fb58dedfc4f84
F20101220_AABJDL liu_w_Page_114.tif
5c5a2ea7cd77b684abe6f0adf0c32d5a
8fb10b5dbe095bd3cccbe6fa7381b243fac902e7
1051941 F20101220_AABIYF liu_w_Page_135.jp2
14db785be7377872fc77d3beac31361b
7e738dd7e56e7d427076bf479b53e837d7e654db
F20101220_AABJCX liu_w_Page_099.tif
15d1db7bdecad4e3d15efef0b7b9e350
d442337be16c40b2cb46d7297959f24806afbfb2
1051970 F20101220_AABIXR liu_w_Page_119.jp2
b9d737de8d924528fae090574ff1d9fa
1a0f7b558fbb81d2a4c107f02aa3f50147f8f4d8
F20101220_AABJDM liu_w_Page_116.tif
44cef1a9b33307aa6b2bc07f9cfc7f09
6be019cac9920cc363c3419de28c76b5a9d2b547
1051922 F20101220_AABIYG liu_w_Page_136.jp2
f1c4ae1277381e91d542fddd66068a63
2edc03e0b6dcee9f183847dbbfbfa88280098d31
F20101220_AABJCY liu_w_Page_100.tif
32e036e5c0ed2676f34311bf6e32bbf7
7d53b81b8de9998209d1662959300e3e22c31afd
1051921 F20101220_AABIXS liu_w_Page_120.jp2
7903de4ab7e9ea5cbe2dbd2f5f4530b1
323507404dc244bc9071ffa32283398986620bf3
F20101220_AABJEA liu_w_Page_130.tif
bd806a72a6750588165a25655ab1ed57
52c3f570818b76d011d97157854480ea3cb01292
F20101220_AABJDN liu_w_Page_117.tif
6d8a156b92bfd0114712a2202bb9091f
d0a40efd8b6328faaf3539857a86615311dc4b10
F20101220_AABIYH liu_w_Page_137.jp2
4cfc5d3f60765f48d6ae266c9f2a4f34
349f6c62e96675cdeff45e01a61d0dfd16f356ab
F20101220_AABJCZ liu_w_Page_101.tif
58491b978e98d05192767794ee74a5c2
3cbfa83a71fc9aa9c14a38771c5386ee9b224bc9
F20101220_AABJEB liu_w_Page_131.tif
add1989a963677e4b209a08b29426f21
fc7c7d3d4219fa4bddef8c8ca3416bd64a12e5c6
F20101220_AABJDO liu_w_Page_118.tif
91c3695f02120f6772e770c4dbb77938
c32879a5f2ea1ffec907b57c47cef56107a1f6dd
F20101220_AABIYI liu_w_Page_138.jp2
86943cba6461ae7d44394a4e19c4e722
e9e903c91c7bc7b27970a0ac85fab739fcb206b8
F20101220_AABIXT liu_w_Page_121.jp2
66080df7d25b6b2293f8f34e699d8c8a
1443bc68844cc6c278d49a8d000c77f8e745cada
F20101220_AABJEC liu_w_Page_132.tif
4e26f1385448c72050969b9f24f74aa9
8e4b8f3733bafb7b2be0af8468d4b51d3cdad1aa
F20101220_AABJDP liu_w_Page_119.tif
dd63aec64c76528aec4adffd92a41c45
1247b31b2eab4650898d2f9c15518bd0c47bc4da
1051929 F20101220_AABIYJ liu_w_Page_139.jp2
f72a8614fab117aa4bd8db8edc302970
9032619532640f4cacc674d5c630eaee51a35227
F20101220_AABIXU liu_w_Page_122.jp2
6fb05b6aa1e47438be5bb9a0652fc706
83e3d801cf6f6312e015a081ec48be7d3dce209d
F20101220_AABJED liu_w_Page_133.tif
86c7bf605838e99c3ae3200999b92fdd
65e33fd9dc2f152c0d0f8536c73ce54441854d0e
F20101220_AABJDQ liu_w_Page_120.tif
c4f36d4675b00a055402666012f69c06
8684e0d5c3957af5552e8ec9130160fa5c2e9484
F20101220_AABIYK liu_w_Page_140.jp2
95b3c906c05d6f59eb74df29ca96b615
0cbd6c00189dfa113d41bfee7e055d6bb7bcc8de
F20101220_AABIXV liu_w_Page_123.jp2
6a9c97e44b250cfe10d9bdc1d7bf1e41
dfce63702f5f2ca8795ce0149c7874ff443f99c6
F20101220_AABJEE liu_w_Page_134.tif
d432f7b4af7ad9ec9a376df67f975d9d
886545eceed02e1f0d68047d6299a11a24d724b1
F20101220_AABJDR liu_w_Page_121.tif
22f399504970ecee2a0ddcef8c6a14c4
fec52387ef2e80bc722dd0b6c8afc96877064857
F20101220_AABIYL liu_w_Page_141.jp2
85a95e4bd8fc041eeec95cc44d246f11
e426c194290c4509f11ba60d67f68c753a7a6870
F20101220_AABIXW liu_w_Page_125.jp2
fc9b7aa5a972347448572ea5f4bc5c5c
8130bcc397c5f6dc6ad74af56dafb1916b697c32
F20101220_AABJEF liu_w_Page_135.tif
dad03959a19dc35d3a4d76aedd6c2e0d
1234a3dd3208aa4cedbaa782a5947ef55a7cfa75
F20101220_AABIZA liu_w_Page_156.jp2
b710dee4d8d7328276495815bd75fa74
00067889e49f1765eb517161204c71ae713fb101
F20101220_AABJDS liu_w_Page_122.tif
3c0d3418815d6a5b8c14108722bfc1a8
8bafc0de4ca8ce1adaf608532e6e6a96e956e08a
1051901 F20101220_AABIYM liu_w_Page_142.jp2
ae4fb559d24cabc3dcbffd1ac57097bc
217c5ecd95e1173d7515638a1289984e585b0f4e
F20101220_AABIXX liu_w_Page_126.jp2
abfbab9e489438cff42034a8c4a8be90
01d583872869383e59ba1963ac917bafed2b1c6d
F20101220_AABJEG liu_w_Page_136.tif
1acb8c61be7f7872e764bbce8cdba953
35a88c50c0ca38568d38c68cf67247e91910f98b
F20101220_AABIZB liu_w_Page_157.jp2
a70be324cd38650f59393e271f2d55df
a60e046ddc7cc2c2836e2357a865ab2d500f1d38
F20101220_AABJDT liu_w_Page_123.tif
876205099b8e227965b1413415cff9bc
9767e66d2205f0f82be2322279d024c3fd65a3ab
682693 F20101220_AABIYN liu_w_Page_143.jp2
6d34fa143f91eb32514588793ee26fd5
dea50f391edc1d29ae1ab76a0be7d71c62c5c6ee
F20101220_AABIXY liu_w_Page_127.jp2
9b66f9712bb21a44283fdce94ec26447
bcb1af8728235dbb870d442a4d7315b8b34fa071
F20101220_AABJEH liu_w_Page_137.tif
57c3afed9a02e39f47302636b53f15ca
38c05426b66fc7a94aee531649d50e835c7a92bc
F20101220_AABIZC liu_w_Page_158.jp2
cc8f3fc3f74ec3a9dddd2e2c2ef6ca3f
f12283fe6b71f0ed630db84706e7631fd384af0f
F20101220_AABJDU liu_w_Page_124.tif
10cd8d90c2e6e79a22846cd0372b66ca
845668bdd9109337f8153b31f066d9b1ee8770c2
781530 F20101220_AABIYO liu_w_Page_144.jp2
71b51509ecacc61e36ce1c5ffb2bcdb6
301702369d6c9777ed9eed87c80d2a5a62045cd5
F20101220_AABIXZ liu_w_Page_129.jp2
786b8e2763f1483fb9d28a5fc228a015
37476bc9e4b80a2c4d13b8248c1e8af0cc6656c8
F20101220_AABJEI liu_w_Page_138.tif
ea8953d541e0e50e96043bb4df27ae6f
7d2de5b007821d7004aa1db6fcce6f40abb7b5df
F20101220_AABIZD liu_w_Page_159.jp2
039177a5f1c4a53bbcdf195ae4a12278
1445710369435da99e752952d52ea10132e1c53c
F20101220_AABJDV liu_w_Page_125.tif
2f6923ef90d73b32d814c80960ebb652
009d53e265d6e8e824de2db2e3485e09e045b633
F20101220_AABIYP liu_w_Page_145.jp2
ef7fb50439974cde65cad135d1351bc4
66187ccc02ec9981b3c386d7af2fb7b9a9937eaf
F20101220_AABJEJ liu_w_Page_139.tif
4cb4ff820b6d4b1569d2db50b2a1d166
348081ef80dd74565d34e8019e31ed8883872199
1051958 F20101220_AABIZE liu_w_Page_160.jp2
5a501a85a35a25aabf4605eb9ef85db5
8d439178a6d2ba0c42c0e1abbf97302ae41957df
F20101220_AABJDW liu_w_Page_126.tif
81b8338844ecf9df6abff566d8094c72
18659477a2ab7107b6a97239188a869dc0476857
914058 F20101220_AABIYQ liu_w_Page_146.jp2
b76ed61b4fbea1e67efe7dc3059cb86f
a7a52365a9ae1ae79bb9ec5b66548498be95bdd0
F20101220_AABJEK liu_w_Page_140.tif
4c8ec46f656164005210ff8bf72e3989
9cd94ae2e9aa33e36f6566a3f70077ef504c8648
F20101220_AABIZF liu_w_Page_161.jp2
6ec657e98038d6cf24b00ff244749bc1
761e9dfa8ea0483b1e8358b62641416f698192db
F20101220_AABJDX liu_w_Page_127.tif
2185accb5f290b5d4c6c6bea118565c7
b29a9f3d2067dba9729e12b6b254ab7fe1c7901b
659327 F20101220_AABIYR liu_w_Page_147.jp2
f9e2d614693a7de0f9ede972b303c99d
d19bfdbe4304d9c234159f2489847731fabd00ef
F20101220_AABJEL liu_w_Page_141.tif
67cd04126f66245d1b01c8f25fda5679
3514b66f94c6e2f62e39599823ec5f315948b5e5
892097 F20101220_AABIZG liu_w_Page_162.jp2
56f30f1f18b7048b3100f45eba2451d2
0097a8d4a826d3ab0549a204fe0df798b5074195
F20101220_AABJDY liu_w_Page_128.tif
3a263c7b16bf1e19042adfdcaa499804
8b387633f5d37d91380dc976d705c36aae38fb9e
F20101220_AABIYS liu_w_Page_148.jp2
c314ededa0933354259796b36f4e84e0
52e9adcfa211a4f85455fae287b89c45c2eea966
F20101220_AABJFA liu_w_Page_157.tif
5d04969285c1e2479f81274323126353
88cc128adb350f28f078ba3bd447ca56f9d79a97
F20101220_AABJEM liu_w_Page_142.tif
64e3a671ac47a957b0702530ff16efd9
fce9f4b151646fc8176cd95e65a55ac2156037ad
348551 F20101220_AABIZH liu_w_Page_163.jp2
53d72ed8dda596a82859b16fd8f36c56
57ca1047798604d92eecc8df7c80ab85413a864a
F20101220_AABJDZ liu_w_Page_129.tif
2441fadd570c6a52061bdd50f0cd6aba
fffd8ee6b39dc7f0a835079de55015bf7e5a7480
1051950 F20101220_AABIYT liu_w_Page_149.jp2
0f25117bc9d75fa3d821ae1c9b16e5e2
f5d1700f59c98eed8e1cd2e5375d0df282585b3b
F20101220_AABJFB liu_w_Page_158.tif
0a6986c3f2bcc0bf4b727ada54d8daa9
ef32852b206823937d5490a6d701433cc8a6c79d
F20101220_AABJEN liu_w_Page_144.tif
809c61e9a2923ec1b27d9169e9b54f2b
dfe22b44479bf2107cb4f34a9f744916b890f03f
205724 F20101220_AABIZI liu_w_Page_165.jp2
d10abe15cbb97f93eb8ec3867839eade
1f448346ce921dd9c9271a3ddbc0ad044f53c07b
F20101220_AABJFC liu_w_Page_159.tif
35d3b90110105973e242938a22911268
e38756cedfc1d93bc41d50bd8891251a1b869688
F20101220_AABJEO liu_w_Page_145.tif
0466de7fec7f3c7cabeeb519692cf9b3
68fdfd79b56ba999a1f51b3832e9d9d16b2e25d5
F20101220_AABIYU liu_w_Page_150.jp2
2f8bf6f905d1b732ab8740af22a370a0
fbff7767f185b37d40829af211dde57dd615528b
F20101220_AABJFD liu_w_Page_161.tif
8f2938d65fd4077fccc30823dfeb3d06
724550178832a439878c7cd677c624ec97fa3c4c
F20101220_AABJEP liu_w_Page_146.tif
1c36e0f90ecff282e6940e2ae33c1da3
bb7d55bb2551a598f7551758ff6d0407b21784c0
736414 F20101220_AABIZJ liu_w_Page_166.jp2
bcb9e1009f41a72915655a2dbc6bab43
77516107acae99da5e289fd49837ff3743678193
F20101220_AABIYV liu_w_Page_151.jp2
5e634ae38beba5406fc569dadc6d8e5a
1de7f462407eb9ba5ecba749951524f72e5fd324
F20101220_AABJFE liu_w_Page_162.tif
444476dbdb6aa9a09f641618847da8bb
1f410608efe8c70750291f873d5b544bd36b8a84
F20101220_AABJEQ liu_w_Page_147.tif
e8bbc655c64e1d61da1b5702f87d158d
34b7f6dbc4830c3c1f00f45e933b3dfc2e9f22cd
F20101220_AABIZK liu_w_Page_001.tif
98ac9d9a560db85cc22899c02325875e
8d457ea509ceb53866817a003fcc92169cacf04b
218794 F20101220_AABIYW liu_w_Page_152.jp2
1f4adcdb7174a2bffdf64367c2c3dc14
8f69732628531cd93a6d66617c9c6d87815e8d4d
F20101220_AABJFF liu_w_Page_163.tif
bfae05e9ba21d8db1f988f07bebb27e6
6c2a9c23fe2755323044f16a253bec6d8fcdf71c
F20101220_AABJER liu_w_Page_148.tif
d854d523d127b57e7377a3e2ee58a104
be7ffe7d99a5b2b1d5ec3151b54a7c7b657ba00b
F20101220_AABIZL liu_w_Page_002.tif
23903c3e45789739e0f9cf3b16312613
31fad2452901909ceb4a55010cc9ebba2ed46739
F20101220_AABIYX liu_w_Page_153.jp2
fb37fd70354f9c3c39d7ed0a76011a74
2f80e300ba9946e335b1608163ff064f1ce1abfe
F20101220_AABJFG liu_w_Page_164.tif
24f238aeebf924105a9bd069266990e5
4ddb868f3d6657d58a57bee23d7ea5f072a0071c
F20101220_AABJES liu_w_Page_149.tif
94673d70a3432f193ce57b8dcd5aeaae
8ccaaf5af6c1744b07d29be5691ab85df55f311b
F20101220_AABIZM liu_w_Page_003.tif
a060f2392f116089e5fb24aa38ef8569
79e26ade81153b2c2e1efcca6ae235ac80134ea7
F20101220_AABIYY liu_w_Page_154.jp2
9fd26926a2da7a96063495e20638ff49
b31061f0dc8562df97ceda1f962938addf05c4ee
F20101220_AABJFH liu_w_Page_165.tif
127b1121526903c8ad176a1a6305cab4
b67fd014b568f28614e17db28c7362ce2baac294
F20101220_AABJET liu_w_Page_150.tif
43870a59c5ccdb4e2360e7a8fd59520d
623d7b665b3da161b593e78a3c77ba0e2b7f7ebd
F20101220_AABIZN liu_w_Page_004.tif
8cf25c8ff3be9d658ff4430b60257f89
87ec6e15e6c86793052b83a293056aeb2abcf2a6
F20101220_AABIYZ liu_w_Page_155.jp2
1f505c2dc59fdb7b5e48ebfae1d3eb7b
b27cfafb6ff96f309dabb3bfbe3bf4ff3336e7b7
F20101220_AABJFI liu_w_Page_166.tif
16f75bd577df39d1adb79b898e210695
1892d5cb99d324d9d84b3c10c24fb5130b12ec1f
F20101220_AABJEU liu_w_Page_151.tif
800143e894b88e84564db10947485de1
e815c06b3bd5677b3b042cf7461e7e5f01572026
F20101220_AABIZO liu_w_Page_005.tif
f1206e76311727d3d03966cf86276839
a7e01504e68e24e94116215554900a92abfb2c11
8351 F20101220_AABJFJ liu_w_Page_001.pro
0fd75935ea2324188a52b554aae1553e
9ed8a4c13ccd921d0247d49c0ce9940e075e4e08
F20101220_AABJEV liu_w_Page_152.tif
71870fae87d0948ba4e85aab4667614b
d0751af6364e49e4ad0737bb065b89fbe9c7433c
F20101220_AABIZP liu_w_Page_006.tif
812daeab7fc4b1c06716a399595942c0
ecc930fd59ad2557077a646b357d343c826e667a
961 F20101220_AABJFK liu_w_Page_002.pro
5ec8088f9cfe6b64f9f739598f9e51bd
82c72190802e30ae92ef9d2ab03b508095ab99c0
F20101220_AABJEW liu_w_Page_153.tif
dc40a1f9fc24e894a7720dcbc58b0d05
4acd23250f953d394c393eb8bef450634f5b21e5
F20101220_AABIZQ liu_w_Page_007.tif
a36e51a9e7e9809357fb4940f4483b79
33f5dada532de7de0fc76204d8302f8cf31988d4
62060 F20101220_AABJGA liu_w_Page_018.pro
34e69ef94be760dae7cea7971c097fe1
b498721971233f40ffc316b74930cb61459560b7
1529 F20101220_AABJFL liu_w_Page_003.pro
011a51985b2d01544e0b1d4b05c81402
ac78c0f0164adc77e6b854e4a73863cf06c7c2ef
F20101220_AABJEX liu_w_Page_154.tif
ca9dd8cc6e77dfea7791024c633128d0
e9a5bff597896e7c9435c8d9e284068289dc9dda
F20101220_AABIZR liu_w_Page_008.tif
735c1412005807d94902d15aaf60eb93
ba0dd3f830bbe0726d533a88d56b69c77008757b
43361 F20101220_AABJFM liu_w_Page_004.pro
38e515c3724a84559c2ba75bf94eb1b1
a2d5e1381589ff8a49d106ec022bb62d0c4fde8a
F20101220_AABJEY liu_w_Page_155.tif
9832628e5f5bd5751be54276ee56b5e0
b76dfe5e91ee408d81f3358cf8cdbb2ed334cfd1
F20101220_AABIZS liu_w_Page_010.tif
4a6a80d7e290f1f1228be1cb7546ae76
082c8b2e708f6a7730dbd544761580bac149d794
59383 F20101220_AABJGB liu_w_Page_019.pro
cbf3f9b167f6978b2b54f5c1522948ca
c5ca78321464d6a834fef4b1908d8601a8211402
28709 F20101220_AABJFN liu_w_Page_005.pro
6bf919ef64c76ebc9c0d5ddecce79457
7a7fe1b5f2f397779011c7293082b21abc790ccb
F20101220_AABJEZ liu_w_Page_156.tif
ad68a9660e6614425817a1e5da835e21
54542942a757ea5802dea5c323d48f5e1c2bc966
F20101220_AABIZT liu_w_Page_011.tif
b3f701bc909761115c3084630e373586
6ccbaba9cafe78330f947b0ab086428764fcac1f
54945 F20101220_AABJGC liu_w_Page_020.pro
1a904ffa08646761863d18c86a9375d4
93a52d724b04a9035f0a582fc7a890d92ac15ec8
61618 F20101220_AABJFO liu_w_Page_006.pro
34bcfe0cc875451ac12c3ddc94b405df
92aa1407d269fa95e00e15ae191009428fbc7386
F20101220_AABIZU liu_w_Page_012.tif
ffcb37b49ee92d04992f1bf6bcf47102
48d523f2fa940b878b47a2669999ed3aef70a94e
59975 F20101220_AABJGD liu_w_Page_021.pro
682e6889be8d55ebdf3d86c9de2e68ad
d6b047725c0c9220735e7ca7cc2833feca6d84a1
34041 F20101220_AABJFP liu_w_Page_007.pro
3c23537126500880eeead91c92373cc4
66954e51a96f194255393733bdbe2ad6a12936f3
12575 F20101220_AABJGE liu_w_Page_022.pro
346eef59c9af7bcf7d9da122642f27a1
750f900bbb02e821203c9883ae18b7b2642c0719
12350 F20101220_AABJFQ liu_w_Page_008.pro
3a45db985ee1aa7387664fdeb65de1d7
63ac9a129a29382c0d37ce7c95815e0dc37b2d92
F20101220_AABIZV liu_w_Page_013.tif
1822516de1b3babef44f2a97fa60a4f6
43775704cbf437493a997f35ecb39103bf909c51
54881 F20101220_AABJGF liu_w_Page_024.pro
3258d5fdab39300a0dc7c0d15a48398d
9b69a5bf1ca9ddad5add2dbd4162949bdd7b2158
47936 F20101220_AABJFR liu_w_Page_009.pro
38f32478d5fcd6dcac5d313aaedc54d0
74336afe7db580392ce95c9b34660811e27dfb74
F20101220_AABIZW liu_w_Page_014.tif
e5b9a93852b6333463d384e985576e2a
c33937fb4c84f5a04f04507509366cf8dd5c4c43
44969 F20101220_AABJFS liu_w_Page_010.pro
9995f758e99709b843e43e7610890d87
4d38857681465eb870a164740ede5a0c64db644a
F20101220_AABIZX liu_w_Page_015.tif
95fefe8a448a862d9d68f6a8d817af78
38b9b3e7896192338aba4e96d812a4e4ae5287b3
60563 F20101220_AABJGG liu_w_Page_025.pro
1b685659cf0987769d1b59845d86e0a9
47d1b83840ccacc9ebecfbeb031e239d483b231d
44201 F20101220_AABJFT liu_w_Page_011.pro
6d5410adb8d6d4efc9c48c1ba3826bb5
f22cb980dabd0729416b0c205a3458976b29a597
F20101220_AABIZY liu_w_Page_016.tif
f4a4e57d17f0633fd7d4a16ba9b8c60b
6dd97fdd4a70b08e26eed31e4422f510dfbcd07f
54628 F20101220_AABJGH liu_w_Page_026.pro
460f3a8a2b837818f5e67445647cd132
b1e8c84809cd7ea881b823f8a9dab9f620d079a2
28256 F20101220_AABJFU liu_w_Page_012.pro
69e1ac51c6959ec738041ac5780435bd
fa7e199f3a80a668bf2881ee25658597abddb2cb
F20101220_AABIZZ liu_w_Page_017.tif
a81d36f4348a84a032a949ad9a0c3061
6af17eba8dc8075d101847c8060412f71323d7a0
60326 F20101220_AABJGI liu_w_Page_027.pro
10a604de6c92d99eec61c3b5cdbe8ceb
c0bf0be14c91eeb9a30e022c65662a2724460cc6
53180 F20101220_AABJFV liu_w_Page_013.pro
e38f0381e98ed04a770c3062582ac262
df9a365442c36bac67865932b97e5975f19f2347
60557 F20101220_AABJGJ liu_w_Page_028.pro
472b8ecadba9fcbd70685159f7fd953f
4503bcd5947fbeaf9abf2b6438c639fbd5dccc76
57014 F20101220_AABJFW liu_w_Page_014.pro
eb1b3575a885178a7b45c098351c9202
f3fa806e898c46d4f86e5a63943f13e3e66f941d
59147 F20101220_AABJGK liu_w_Page_029.pro
da9b0160b75c45523fef709b4c770fee
b230565bb8effc99981d6b5f65b48949b12c3344
61101 F20101220_AABJFX liu_w_Page_015.pro
81be07d97066ef7c50422d6882d91be6
c5f6248b659f9b7bbdd09078ab1be00799d4b445
28843 F20101220_AABJHA liu_w_Page_046.pro
e4dc279c621bbc7bb0008f76a717b150
801c2d8f6df9569c4f2ab48b850c56ad1f377280
47696 F20101220_AABJGL liu_w_Page_030.pro
bd68429b638cb9384e8e9151272c666b
61fa1ed042dc5a3aa632cdde85246ca5c625ffa4
62355 F20101220_AABJFY liu_w_Page_016.pro
a4a74a5255534388e0c550377680a929
a7255a683bc6a55ae5a7f4b56f612b5456f3c3fa
54566 F20101220_AABJHB liu_w_Page_047.pro
e0208de2a966475f9defa1bc26019147
f9a98536a5162ffcd791bf993fa9489e0ea8bdbc
61022 F20101220_AABJGM liu_w_Page_031.pro
6b9431469e6248d0cb806d3cda4c12f7
e6031a0f0cf481e1a803f2b4cb354a516edfc58c
61719 F20101220_AABJFZ liu_w_Page_017.pro
1e562828b3e9c0055f19903312511c68
9c02c15373c169dc023097e357bb859161f8a281
55004 F20101220_AABJGN liu_w_Page_032.pro
5e69e778ff6117f3a36ecec3e9402e2a
826d733e95f8c241538af642f0614f6eb043881b
56511 F20101220_AABJHC liu_w_Page_048.pro
1bd8fb586b536b51c69340e8962b5169
51bc24f369668fdec15fbd534e1b66a105b5affd
38069 F20101220_AABJGO liu_w_Page_034.pro
8eb0dc3efc84e0793428514986163c35
f1f3006a099937c2bba4f8ba50d59faf6e2df107
44167 F20101220_AABJHD liu_w_Page_049.pro
ee06dbe54c2813029620da47189e467c
b068b440f9bd7f2af88446961c6b3d1517c2e024
29908 F20101220_AABJGP liu_w_Page_035.pro
602da74fc76b9fed1d8b8d695adf2739
37cce2a80fa6280412ecc6cdbf5dd1c0929f336b
6823 F20101220_AABJHE liu_w_Page_050.pro
d5670480a8f7f35a44b29d68de84570b
a7772dfaadcd183e16be990a7d3cbb3b703ecf0d
61368 F20101220_AABJGQ liu_w_Page_036.pro
dc36308adde5c86be31c09824a49c791
b8d3921f13c271f800b6e5665aa0e55bd1ca9bd9
58724 F20101220_AABJHF liu_w_Page_051.pro
0ee880e266f83f1db1b9c7be13a8f323
ea5627514ce66acd8cb6bef684cbc850d101a680
50527 F20101220_AABJGR liu_w_Page_037.pro
bfbbd8a5e07254b50c7e18f64c1ea9ec
0fbd2d33fed45484da4a371653d573e028225173
54019 F20101220_AABJHG liu_w_Page_052.pro
825a527b1ee75c2bff5d5e77fc8d4f7e
e483223f8e2ee7283ff076e10aa00ad02fbe7e54
54930 F20101220_AABJGS liu_w_Page_038.pro
1dc627e7290cafe182ee09e4c3473c20
267a7fe2c924466251733a9c0aa8bbaf02cf73e3
56050 F20101220_AABJHH liu_w_Page_053.pro
972c34745f58eec0723cbabb4acf5017
77667034e1a121a480e34217d022282219436b1a
46677 F20101220_AABJGT liu_w_Page_039.pro
14b255d8243cc190cb62058770561417
e80b8f25fe427e51dcc5ef19c5b2806f87ccdbf1
59287 F20101220_AABJHI liu_w_Page_054.pro
123078d3618666bfc34e803fdb8bc0fa
55b7ae666ce6485751897e8e85147b5deb097f3a
43959 F20101220_AABJGU liu_w_Page_040.pro
647d9162605977260da01614c02131ee
b660db9fa4e2bbea747914ae9c755fffca38f07d
14806 F20101220_AABJHJ liu_w_Page_055.pro
69a78d95f32dfc4c0160b96cbdf087cf
58ee4020a739e91396434570935aaf606bc03485
56918 F20101220_AABJGV liu_w_Page_041.pro
9feb796655d18e6fb6907a5b6af45897
4b40241c7e3cfb7a6da0673b89d0d570a855db1f
57334 F20101220_AABJHK liu_w_Page_056.pro
e1af18bcc43d81411ff450e28d5ffe64
71c7163d77d7c947f27e99987d38a18f364e5023
45573 F20101220_AABJGW liu_w_Page_042.pro
dd96f8ed09c47f6751b122d8051d1bf7
c061c21c35c09f05b6f5ad471932ddc42fa46bfd
48161 F20101220_AABJIA liu_w_Page_073.pro
2132571fc4358db39f3157037c5a3540
147d892382827af546f012cafffa8dc7c5a5a0f0
16388 F20101220_AABJHL liu_w_Page_057.pro
57484d6d2a4f107f78b07d687dce5cd3
c76dc74848e7e71030f8af92bec909db6d17d9d4
58934 F20101220_AABJGX liu_w_Page_043.pro
708a3045a66d993b3fca5fc19323f520
b194b06d2fde0934c8ad6017570d074830bc137e
40239 F20101220_AABJIB liu_w_Page_075.pro
3ee1516abc8135b24684c1f965d3dc4c
5264af0763bd403893714f14b0f60e0cd3c9447a
30269 F20101220_AABJHM liu_w_Page_058.pro
1fc40a7aaece070c92aa7ea97a1ab692
95a2140a053295bd3c42517d87c87ae331b0614f
44717 F20101220_AABJGY liu_w_Page_044.pro
340ea775bb609138fc55ac2971e4b54e
dd264571f002d9ae152fbeed88c2349da919466d
46131 F20101220_AABJIC liu_w_Page_076.pro
9b80249e24a0d7aba25aa7b383892a9b
3cd3bf3ccb68e705c3756e5ed60b6b1a04ee93fb
27511 F20101220_AABJHN liu_w_Page_060.pro
d5bb44b2cc945aca241c808cc55f5484
abffd10797af0149cae465c3d3f02a4b8ae100a5
41813 F20101220_AABJGZ liu_w_Page_045.pro
6c1e95866f251a5103c372808119bad0
1c1abdfbfd767dec240b8e256dbc770eaa14056c
50700 F20101220_AABJHO liu_w_Page_061.pro
f213d710eef57d7ab4fa9670486a7e9d
9ae6f9b4bd2c940d543bb97664519e9bcaa31037
59946 F20101220_AABJID liu_w_Page_077.pro
4060bd05a8a9de30bebf35a77587db0f
1fc6ed0bd4bd4c454ab8b02cb8586030845e1ac6
58715 F20101220_AABJHP liu_w_Page_062.pro
e7ed96085b47f90777141eb3bc7b6290
270a4b76cdee6fafa170d01d1f447553155cf40d
50735 F20101220_AABJIE liu_w_Page_078.pro
5db4d4dc7aae308c9c96bd07002ea403
f817b42647c73f56064a640c27c77d5a6a4543e6
57594 F20101220_AABJHQ liu_w_Page_063.pro
18eb2c425755a948e06cbe7103bdd62b
e0ef201f1b3c41f07c744eb3377ddecbc50f2644
44979 F20101220_AABJIF liu_w_Page_079.pro
0037833af3af97e9ed08626ca3577537
b5a6a427066890c3d3131b0d1582a6229bb025b5
58755 F20101220_AABJHR liu_w_Page_064.pro
aecf4d826c22aeae238be0e064c9a750
b2b35b076bf3a572a343570f5e58f03ce741658a
58331 F20101220_AABJIG liu_w_Page_080.pro
5e1b1e83035bfbc05569c8b7ec1704d2
77f2d66a75d77327fef1798c8b3e60785edd6058
63286 F20101220_AABJHS liu_w_Page_065.pro
ba013ea7046c368455cd39130f1d4be2
b550800fc2772742d64f4465168a4cd2e04e69d9
16479 F20101220_AABJIH liu_w_Page_081.pro
d33b24fd726b23eb4acbcef73cc8fd24
735ee1564c2fe67de93c12b7f37346e8c4941252
57376 F20101220_AABJHT liu_w_Page_066.pro
516cb0235ae29d02cdf9614bff4c4c41
6d9c4ff36f8b56f0708431ff6fba4f772c89ee5f
57352 F20101220_AABJII liu_w_Page_082.pro
7093b0c1c43b4c0f693302a7267c137c
c8f8519ce809292fc439b86ae2a133ec9a0e8bcc
36123 F20101220_AABJHU liu_w_Page_067.pro
17da4a6fb95a7ebff4c001212e68fc07
0dd3704b70412806e6ff38946fbb426175f9884d
24599 F20101220_AABJIJ liu_w_Page_083.pro
c20bd60dbb539c7cf43e799f465a13ae
066d465ee26ba2b22c9b8fb73c3a689a4e084cb1
55614 F20101220_AABJHV liu_w_Page_068.pro
0b3ab4dce420b901f94e59a5bd0f34df
69c2e88765660fe68cca4841cfa83a69b8f4c8b7
51407 F20101220_AABJIK liu_w_Page_084.pro
79a6a747822d5fed6ac040358f78072e
6204f4ca8068472d69e9bd605b9dc230562f2627
63257 F20101220_AABJHW liu_w_Page_069.pro
8fc97a23adfd7a4e5954066972f96530
928caa7efeee702f40b68a7dc86a2e23781d3277
58354 F20101220_AABJIL liu_w_Page_085.pro
e6561ae85cfdbcaebaf13e11f8db8c93
0f1d9a2b3a5d42117f8ea731805a0d923dbe5252
47577 F20101220_AABJHX liu_w_Page_070.pro
80d2dbb4a7f1b9a810aececdafba57ed
6b56ef866b413e9bcfe4a22e364149c024edabce
61160 F20101220_AABJJA liu_w_Page_101.pro
fd77bc6eee8000dd7b93036011a2c832
f5308012285cd6c985f90112d0d01899be078829
59964 F20101220_AABJIM liu_w_Page_086.pro
1ae34604fdec1bdd9b11130ce40a1417
1f0758120ebd48cb01e10c9591af42b690e82f61
29893 F20101220_AABJHY liu_w_Page_071.pro
17e30e059f708913da5c5a76c65e689a
50b9f780247fabd30682645e6652258877cdd65c
56693 F20101220_AABJJB liu_w_Page_103.pro
212e7057ae0bb58486948abfd4881e26
51f20c53a60d7612e43da6498f70c0439fcf40bc
60069 F20101220_AABJIN liu_w_Page_088.pro
b6dc20304ec4c4d34e51a6ee370151ca
74284932d5ca9cdac25e99eb4467cba5fcd2f779
46967 F20101220_AABJHZ liu_w_Page_072.pro
c6e85dd66ccd6386609957709cfa9323
0187da7933d22c0424d64dda881b7ae7e8aef3db
59661 F20101220_AABJJC liu_w_Page_104.pro
e21f2f7420258c5a78b6fcd59bc6b060
730ef545ea8afd9afa7f0813dd440e4dd4a72ad6
46879 F20101220_AABJIO liu_w_Page_089.pro
50448a05555b201ff39b8ce140c3c490
4d3e7919cd06c43e4c066007f20112fe072f6011
52113 F20101220_AABJJD liu_w_Page_106.pro
720a2f6b9480f3ce6a7350174eef18b8
e5b13476f1644bb495fb471bc464b0fb4dea5e5d
61792 F20101220_AABJIP liu_w_Page_090.pro
26a2e8cce9b7e43e94340fb9b628ab85
33439d1c5a87bbee2d325da4250e312402b650f2
45136 F20101220_AABJIQ liu_w_Page_091.pro
383b2d31c04e88e7746270c4e89feb2f
6a7cc7b7007b6c09a2196111e4d816c1e9acd967
59039 F20101220_AABJJE liu_w_Page_107.pro
6de4844b201747de0dcabbcddb0c3072
73b092198f20aa363ecebaa5e14cc2a603030665
41778 F20101220_AABJIR liu_w_Page_092.pro
3933ca66d16149711526d7b821bf447e
63584382ccf9e47984b5b837a7af27ef7b22798b
16465 F20101220_AABJJF liu_w_Page_108.pro
a1b0d5509290b0af2ae34bb1d6e1a278
f1b8e7ddcaed153c88b28e0955a79534fe710a9f
24682 F20101220_AABJIS liu_w_Page_093.pro
97386d4630cbd02e1b506cb77147658f
fcdd1e0543545e61360ab3793ee76c6418f134ea
39417 F20101220_AABJJG liu_w_Page_109.pro
2bb30ef701e2fa8b0b5b0105ea2f58d8
dd9bff7fb367a92c417fa05f3b2f6de2540fc7c7
52522 F20101220_AABJIT liu_w_Page_094.pro
4bb27182e48087939e69a5006f7c6774
4483eb482d833b2092e4dda959a35b03819cc8b2
59539 F20101220_AABJJH liu_w_Page_110.pro
a0612aadd9093e3f0e8cd9da62c41acf
93db4bce4fcdb6ec145e42995f5182b7adc099e0
53075 F20101220_AABJIU liu_w_Page_095.pro
04f0fe9874c5d38dfa5e695ec0c65145
cf97a6dd3d3798269dae72f8d54e9dac926f5fe1
46577 F20101220_AABJJI liu_w_Page_112.pro
939b582d1230165bca09f369adde54b9
6341fb1b99386e4116b341a0c55436244910a19e
33006 F20101220_AABJIV liu_w_Page_096.pro
a2046e1cb79c9899c75986edfd4c538f
0b85945e54d21fbbf4368fb6a41c88f67f96777e
46643 F20101220_AABJJJ liu_w_Page_113.pro
997cc74152f85b3869473696515fa320
806f85d4eb3792e97a484230e6a78a1a0d8c1ef5
62737 F20101220_AABJIW liu_w_Page_097.pro
9fbae23232ebb2308511d2f3410e167f
c867c6cf710cd8f1b18a3308dd442b82c213e96d
17612 F20101220_AABJJK liu_w_Page_114.pro
7bd5f813745ee22c1afffc09bf8f9ecf
3905fa4fc92ec1e66699dcc8ad178c11744674ef
62539 F20101220_AABJIX liu_w_Page_098.pro
09da2368b262730a429fc93e5f3cd6a1
f2ac4b6e175ba23a0e89aae2d51fe01ff32cc40a
64962 F20101220_AABJKA liu_w_Page_130.pro
5487374c17a836fad932aeaccdd0ea05
3caf733c3ca5166005dec72a46c0e54011d8d403
62918 F20101220_AABJJL liu_w_Page_115.pro
e0a2cb7c17a2f6c6a7f557f51c97a477
9a5391ea3345e7c4095c2ece78bf9811421db827
46316 F20101220_AABJIY liu_w_Page_099.pro
8debfc292dcd234c87e59621b6f7444c
c5a094fc26d658b269551d9ecadf44658c81413c
54679 F20101220_AABJKB liu_w_Page_131.pro
aee368039e52fa448c68f73a3d7c0304
07cb8a6146a93d053463a02baa3eca80bd8d9cfe
60322 F20101220_AABJJM liu_w_Page_116.pro
b4161ec39a4cd3065249a0d38f51ec38
f8d65a392109479068ef362d29d578daac3a5c93
63661 F20101220_AABJIZ liu_w_Page_100.pro
e957ea4a0216d463943b25f38d5e6a76
6088956cccc994440dd17c62bc50d7724d1cb24a
21522 F20101220_AABJKC liu_w_Page_132.pro
651b399ff19323f73ba3fe43e642fda0
0c9020f154f6b401e01a13a1de49d4b62c5b22e3
17498 F20101220_AABJJN liu_w_Page_117.pro
9412b043088aa4960e3769ae0b7c4714
28d3e9b2825f1b35492369103e13b84b2f7c0082
53739 F20101220_AABJKD liu_w_Page_134.pro
34ecfbea397cb9892a2f8d65242e3e8c
1fdfc6e262a2cf98cf951d5e733e4b2902f85c55
50433 F20101220_AABJJO liu_w_Page_118.pro
1d26a2a254840bd675a6275c36f3a429
0cce3238db7ddf28f6ae73dac10b1e78cae4bdae
59060 F20101220_AABJKE liu_w_Page_135.pro
1bc042eea13a2d7438504023dfc4fa80
103d2e261c64f85f559899d622c1c8d5f6865ee6
63205 F20101220_AABJJP liu_w_Page_119.pro
26a1c203d464e847e944439c19912e42
a44ef721e0ee69700663ffd5f41608fbfadd2939
60335 F20101220_AABJJQ liu_w_Page_120.pro
b1a6206f28bef34ff19eb6f8fe22ce98
5579598b5c2615f7aec04b7f683d4c9327b57be7
42907 F20101220_AABJKF liu_w_Page_136.pro
dba048a9315642912c9f734259bf7665
1abee7ef5f7263240363643fde4bc1a1cc27e9f7
60446 F20101220_AABJJR liu_w_Page_121.pro
d9b775c6da6a6fa2c1fc68db63ffd987
a5f75bf287bb43bcdf6621e73b1034f96a377144
61210 F20101220_AABJKG liu_w_Page_137.pro
859221977fccaed26c957ead105260c1
b6a9674001e2c39630541b5c28ab947e98a9d4db
62034 F20101220_AABJJS liu_w_Page_122.pro
838c5750aa8943302f86f9119cdcd41d
0853d34d16ba7a60b2fe0c479fee8be341f08975
58604 F20101220_AABJKH liu_w_Page_138.pro
fab0213fa37ac39e9747853ea66ad514
94e0f36a72dc5d2c5d4ea51a92b50c3580590a3f
47944 F20101220_AABJJT liu_w_Page_123.pro
bda66b273ce5bae5093ae84c62903b48
34cf9aebf8b99919e885b05a0d757a188d4bddf4
56056 F20101220_AABJKI liu_w_Page_139.pro
ee4578d8fc7ef226ea6a4935eead7ab8
d5003f03f4918c6370d5abb95ce45e6daee0cd03
55108 F20101220_AABJJU liu_w_Page_124.pro
e32b4e26bc457ed04a8945f6dfdf38f4
d94d60b7a47fdabaf83a3846f194db04b34eed6c
58111 F20101220_AABJKJ liu_w_Page_140.pro
deb59425b09d1c4bbf0a189bbc9ad14c
11ad544d09c3a73cafd8f2a4378bcde8fbcc8523
63083 F20101220_AABJJV liu_w_Page_125.pro
03def4b3e72c4cf61e131590b010a568
c14e7267951727c9539e47e08246d82859311cd3
53418 F20101220_AABJKK liu_w_Page_141.pro
3e55b7cc05fc49f8fc39594fb77744cb
c2e63fe7d6edcf416346893709d90ee3e948f96c
59158 F20101220_AABJJW liu_w_Page_126.pro
e4950c29970447367cb39d03b772449a
3ed47aca6ab2048be1aec7f809c4d4fc4ff5ccc3
65617 F20101220_AABJLA liu_w_Page_158.pro
ae4b976954aee2577e1ecd4520dfbefd
92127d108190f0be65d9fcec99486d69e3d627d1
49268 F20101220_AABJKL liu_w_Page_142.pro
561840e31adea17770c4fd0d0fadb821
801085e5b92de6f63e96f55def5335d24b1f9da8
51336 F20101220_AABJJX liu_w_Page_127.pro
e0d0c2633f57ebefdd6c3cfdd263930a
710302ce72081e4cb903cfbf4003e4b9fc87d0de
62900 F20101220_AABJLB liu_w_Page_159.pro
39e4ab8fdd1ca54e2d9d9bd7d0bd6332
f2fc8d004f2ce09e42b6a87f1f9e5569f8ce0124
24109 F20101220_AABJKM liu_w_Page_143.pro
b9f587a472a0ef4bbdbf67fabdc7f361
c776f5c8f7d1b0b45f93d0fc15555df34cbd5ef3
59444 F20101220_AABJJY liu_w_Page_128.pro
eb5ba76649b2f97404276f296092c809
01f8cb32bcc5b7461864aaf9f6135b57666359ea
67490 F20101220_AABJLC liu_w_Page_160.pro
be03bc8d9e2907fe28a726ed91a8bce8
efe387e20d1d4b46ad25e7892eac7741ca28f760
32039 F20101220_AABJKN liu_w_Page_144.pro
4345c699d58dba027a61739fc52d5e28
b85ff0f596308952fbd082d04a344ea5051aa3e1
53653 F20101220_AABJJZ liu_w_Page_129.pro
fa70f546cab20f871e02c4055540b7ff
5cdd2bafb13864c8bcf3b822bb21fa6c5a22a463
68451 F20101220_AABJLD liu_w_Page_161.pro
14dff930b14decc92651d3541f3efdfb
86465413021addf07ff893141c06cabfd5495d4d
58891 F20101220_AABJKO liu_w_Page_145.pro
30c5eb4c37d05c12d885b1151463750a
48a6e855105c865d6ab249692629d39129a622f1
37635 F20101220_AABJKP liu_w_Page_146.pro
227209900fe30d3f6374d973d6040a66
c3f6f4d5038c8e622bc463857a8a07e783191b7f
38881 F20101220_AABJLE liu_w_Page_162.pro
50e86e02baecccc87adbb59831ad198a
de2c88b60f66eecfee8c49b34d186101b0cd823b
10620 F20101220_AABJKQ liu_w_Page_147.pro
b5b9d9b96fa714ebb04b735f725d0485
178db072b3ba39fbedac8f81a002a89ae9e7d369
14582 F20101220_AABJLF liu_w_Page_163.pro
58b60580026bb1aabb3a9eb7f6a92d1e
cd1f654a18d4c40e1ad51f970a4c104126723a27
61899 F20101220_AABJKR liu_w_Page_148.pro
638c105a79f880cd6f0a8dfece256195
75d54c54be1868b7b8bb3d3ce0fd6483026ef51c
50621 F20101220_AABJKS liu_w_Page_149.pro
9e72ad5666e88734583012d71b48ed24
daa95165057618722d09fa6b37a8dc1d50d40027
40616 F20101220_AABJLG liu_w_Page_164.pro
057695494713611df4f2f3342ba32855
12b325e7b3822cb4f05fc47e5572e8cbda783278
48595 F20101220_AABJKT liu_w_Page_150.pro
c052691ff89265b58a121721a52976e3
2b19d2cbb5aaa298c0c48981b94f17170e74f6bc
8758 F20101220_AABJLH liu_w_Page_165.pro
8b84c08808034e576ffc385dda092ff8
0c9e90ceffff77c14c38ce09da6cd403010fb02d
53426 F20101220_AABJKU liu_w_Page_151.pro
cc49c5d7b157f8019134bae668e8344c
38637fe589e83227b871262933a5fd304682d82a
31517 F20101220_AABJLI liu_w_Page_166.pro
d8edced5dbf285ab8c9d050eb4c77e37
23505af54c8a4148e0b3ae09a8c8c0d77d6c24ce
9993 F20101220_AABJKV liu_w_Page_152.pro
99b51524b2486bbb9b234b754cd281e5
68e836df211e79ea2214752d0dfcc662856a4626
525 F20101220_AABJLJ liu_w_Page_001.txt
9d6e985abfb9af9c234d6390f3156525
6fcf2c5c2542cfb29e7f36d3fc8497ddc788cc0c
52887 F20101220_AABJKW liu_w_Page_153.pro
4924730616730fa39569e47e4a7a9995
14a9c0bc239bfa80ef81acfe9610c7759607f61a
81 F20101220_AABJLK liu_w_Page_003.txt
f16ad725e0cebb955f4cb18d93ddc308
89b3744f051068e43622f6a6f58f72f2a6a8c8eb
59319 F20101220_AABJKX liu_w_Page_154.pro
3a0a1a66037dcaa575eb1c03a8d6c56d
4077c0e5ace28d851ed06b2ee4b406a487a6dc5e
509 F20101220_AABJMA liu_w_Page_022.txt
7c4f930c2dd20d0a10391d4c0add1e7c
34bfec0748a52278f8ad3751e7ffa8b761d6d50e
1192 F20101220_AABJLL liu_w_Page_005.txt
e1e0d91bca032e3742c883952536f9bb
574a6a4d5367a410d23670f1b4e3d0f1c5b9adaf
60458 F20101220_AABJKY liu_w_Page_155.pro
17028886d9039ecf4c4da3fa54e8f656
7edb900f5b55f3e4d613932969cd9c1d4b933e1f
2036 F20101220_AABJMB liu_w_Page_023.txt
9f10d50a569c84586b6c7a99499ab5f4
359298120675a57f73d416ddc56ff3fd8cb15bdf
2766 F20101220_AABJLM liu_w_Page_006.txt
17de25a9b15353c0b70c757d58398399
0385fd06f5f9d1354a10f42647fe99e799c4648f
61636 F20101220_AABJKZ liu_w_Page_156.pro
8f9772e52608d39326a3483d76604f0a
6b77fbb90d136b693e932a4a77b566173eac3f12
2205 F20101220_AABJMC liu_w_Page_024.txt
8721d6020fe229d654e6c0f06c01e510
922ab80cce7b68bbf230d8e7f2203e53fe68433f
1563 F20101220_AABJLN liu_w_Page_007.txt
800dfadc537341352e56757767db612a
0751b0fe8aa81939ff2e2a61454b1ed91c434a6c
2381 F20101220_AABJMD liu_w_Page_025.txt
9ab5eba9cde1b0d39d0dd0ed4f2e256d
c0a77b3c453c632af38c90ee8b746925e25573ef
584 F20101220_AABJLO liu_w_Page_008.txt
9c9e92eb6edde1883d14e369280d6bda
3dcd429866d832b9c35fafb5b21cef9acdbf61e4
2193 F20101220_AABJME liu_w_Page_026.txt
96a5b31ef97c4a560e5be66ab0ce9052
c2f8457c7c13fabbfa9f37d3275a6063a1069a71
2040 F20101220_AABJLP liu_w_Page_009.txt
2f7e4db6b1a05755b7a5f50105382d4b
f0f3faa22e420239e46d73c5f856cd3ceaea8000
2367 F20101220_AABJMF liu_w_Page_027.txt
e4f49ead74958060e8db2f80731a1427
e0e9a9331eea7ab948cfdd1226ccf0fe00deab2a
1819 F20101220_AABJLQ liu_w_Page_010.txt
60c7209b2a8c2f6ecee203fca70a4516
c1aa4ef3338a79d8e8d8222307eef4e045bd3bab
2376 F20101220_AABJMG liu_w_Page_028.txt
046adecb67f2d8067240d2b8d226f30f
6594c5f3ce7165c2e7e46d6d0b91e5da48afc778
2188 F20101220_AABJLR liu_w_Page_013.txt
a0745e343163179f671d5d73617019d7
2698eccab22eb31fed76d6b1f15c9930b362c0f3
2270 F20101220_AABJLS liu_w_Page_014.txt
c4795046cd48e4e918b8c12b270e8051
6212b4f01e99ac9e3a109f33f523361bdb563b47
2338 F20101220_AABJMH liu_w_Page_029.txt
27b4c9a1c7c678e3678d6a8840a6ea05
cab79f2928261258df7d009559b972f0eadf0233
2392 F20101220_AABJLT liu_w_Page_015.txt
2b8b1bf02b2f17d71cd2af23e3020a70
06e16d0acd9145354b1235fbe70d9b50144f5087
2002 F20101220_AABJMI liu_w_Page_030.txt
b748402fb140aff7a528bda1f372c137
3443894fb74d330c3f99a18f236552deb62491de
2439 F20101220_AABJLU liu_w_Page_016.txt
0b64e90acf0c1d696ecf25fa7bbb3bd4
a41695d8e9965c0a72b2b6f01b0f20233eba1cfb
2398 F20101220_AABJMJ liu_w_Page_031.txt
3354b9912b576de1b6813a3f3cbe3d31
cbcc027a9058b99cfcea201bc8f8da638c020c21
2421 F20101220_AABJLV liu_w_Page_017.txt
cc27eff858f057d06c941cf26ee99ddb
93444781b7e24fed413ca07af7aeb1df0ae9ab36
2211 F20101220_AABJMK liu_w_Page_032.txt
802d61880b022c61aac4e3bf8899aa9e
a000a6c6364fa56e8862c4de84d7ef89a3ecc202
2433 F20101220_AABJLW liu_w_Page_018.txt
656217badc786a15ca5b92c3793b1ecd
2250a7eb90a381d250c98566d38ece41ccfaf42c
429 F20101220_AABJNA liu_w_Page_050.txt
3618fe036ff8c7318c5b93699790ebb5
b1a81c9bcc4760a70fa77052818e25dc9af2a682
2343 F20101220_AABJML liu_w_Page_033.txt
3253ce842b6c58d03716537ccdeff2af
f5b5b2e4627974db0a0eac692072b8d0b9e78abc
2364 F20101220_AABJLX liu_w_Page_019.txt
6335a1cc86802c97f4ad7f6b43768d6c
c885e9a81a44c673e5a0f58c771750a1cbbf457d
2311 F20101220_AABJNB liu_w_Page_051.txt
0912457a60b756788758a0ea15f7a407
098909302d2c91e47ea1028bc9ad58de343dda78
1623 F20101220_AABJMM liu_w_Page_034.txt
75bd98a7179ce09aa1409c577582c592
75d509d29d6f44ebf3c4f69eeff9103f96c429a4
2206 F20101220_AABJLY liu_w_Page_020.txt
9fc125b76cf39f1d9f29fddc66c9145a
739ff6cb55e424c64855c5643952c9a1a5923aa8
2165 F20101220_AABJNC liu_w_Page_052.txt
eb91b6f681b5139a8547640900aa8765
b669ab69bb306a48a3394a20b793f4fb9c1aa552
1490 F20101220_AABJMN liu_w_Page_035.txt
192a3ab2c4f0f2cc6d5d46a71510741b
9c0444fdff23b3322e411ee1708151f33be0f316
F20101220_AABJLZ liu_w_Page_021.txt
4074aaca33aae1f700c3fbbae8ab3ded
e55754180bb93b0cd9381a21db6eb61314ec1419
2216 F20101220_AABJND liu_w_Page_053.txt
25c59d4a574c4cdb778d9ccf608ba7c6
fd5bff60a753bfe6bdcdb806d5da7325f8d8c394
2413 F20101220_AABJMO liu_w_Page_036.txt
a32cd7d1367284c9c263bf546c0a8ec8
b2f07b45e09158ffeb4691fa6ad6220710e9fab0
2333 F20101220_AABJNE liu_w_Page_054.txt
e33f331da450da9c09a9cbd540811c7e
f05534805d4982307a8c4842504eec3fe113ff87
2019 F20101220_AABJMP liu_w_Page_037.txt
cf9f1108b8b496c13858134ade1b4e80
8f956746f6bb22293811e0ad7a2ab469fa33d790
756 F20101220_AABJNF liu_w_Page_055.txt
a3714708d1da0da835e007e60594efbb
b7f15ab61340ce68a1319e6cb3061a1985ec47df
2181 F20101220_AABJMQ liu_w_Page_038.txt
4f77741f9b26b449e8b6e3329bfaae6c
4eb72ff313088f555a83412d06b8bd7e8aeba302
2257 F20101220_AABJNG liu_w_Page_056.txt
c7b1efdd684191c1da1af5be0596925a
e6a793ef0134b3262686fb7706dbc6202e8206d9
1948 F20101220_AABJMR liu_w_Page_039.txt
76af400b50bfb384f6291e858aa2aa06
ec4a5932c2c3fbb966d5305b243437dec8ea97ea
1987 F20101220_AABJNH liu_w_Page_058.txt
81904ff74a0cbc78707b8bafba7fe07e
3aba8e3580c82514517bd6e3154b81d5cb93c616
1825 F20101220_AABJMS liu_w_Page_040.txt
9895d4db626889066b6b8d4c33749db8
dbe9f95d5eebb580e36fb6ffc297fff1fcd3c8b3
2340 F20101220_AABJMT liu_w_Page_041.txt
8c3d108f320bf80f82e75e9ed28a37ad
5acd4ee77cf25498cce974f0b29d93213d580420
2139 F20101220_AABJNI liu_w_Page_059.txt
c3bbc0f217e26dfa57214785ca048a7e
97c6af4b02725741a52183877c63dc400896786a
1841 F20101220_AABJMU liu_w_Page_042.txt
fd306d18e49c65b14d1f0c9975bc3fca
82761e58fcb6a37eaca33b8604f4a413b29a7f87
1093 F20101220_AABJNJ liu_w_Page_060.txt
231627fa53303d2c9cedd852a79d7943
a25f19ad8cdc550a1596616540da07825f9f6562
2323 F20101220_AABJMV liu_w_Page_043.txt
c8482716f81f3a953b012297760f75cf
8b3e6165be736a76b6440f3272299b33c9d8d49d
2106 F20101220_AABJNK liu_w_Page_061.txt
6918f450b49d810cce857025e4fa6513
05ce2837cc0d61e1fc0431afac682febe3cacdd3
1834 F20101220_AABJMW liu_w_Page_044.txt
38501629dc2ae0b13b64febefa627479
ee8006e4e715607d41aadd2ab438979b3505a0b4
2348 F20101220_AABJNL liu_w_Page_062.txt
a0502a568490170a3ced041b695fc53a
34a4e220322e55fabd9585c4f50cabdbce9e53bc
1851 F20101220_AABJMX liu_w_Page_045.txt
40ac249a25ee828d9325050eaf95e72e
8bd8df7dbd5b8f178517e4ba272aa30fb0968750
2295 F20101220_AABJOA liu_w_Page_080.txt
7eb851007f35a7ec8d3d4e7cda475171
863e331b73172e96a952223fc29364df664b18f6
F20101220_AABIKK liu_w_Page_028.jp2
76ec45434651ff2ec64e6a6fe0604dd9
fade380a71f10073bbac1eb5d6ce0942d69bb81e
2484 F20101220_AABJNM liu_w_Page_065.txt
a7783ecd8b7188af9800197138fa9dbe
8be4db4f2ed16c37cda98beb6fd475fe2864f670
2255 F20101220_AABJMY liu_w_Page_048.txt
487ae96dd8566958d745c48ab0ac9362
9b4ed70b24875d3da8ae677d84e49f5557c2de63
1130 F20101220_AABJOB liu_w_Page_081.txt
92a4aaadf9aa00d58454c2f9dc3b3689
ec5124989d338e0737a1f7793d6bfaf339a72ab2
F20101220_AABIKL liu_w_Page_043.jp2
cd540649b0e03b73847c0e3e3ed814bf
ff9639b5f8412cf9c8ab0b4d7585243a1e839ccd
2260 F20101220_AABJNN liu_w_Page_066.txt
f4a78b5e9670093355e6954a72f657f9
0e0213618110c37ba92b7f3029e941f36e4446f0
2012 F20101220_AABJMZ liu_w_Page_049.txt
051846b153c929ea95e5e9228f7614fb
680ef761ca3421f7cbb63ed6452edaee7f41926b
81762 F20101220_AABILA liu_w_Page_164.jpg
02ce39d9213f69df26296f2048c646e1
c991c11251fc75b180f66d448a95de19fe0e7730
2268 F20101220_AABJOC liu_w_Page_082.txt
23890852deadb002aa233540823c6ab4
a588a8c1f73750dfd6a54ae62367e4be278d42a1
F20101220_AABIKM liu_w_Page_024.jp2
cc0e8ffeac522f9e0724640f83ced9f5
77464769d1f420722895ace8c0e5d576a75cf6e3
1690 F20101220_AABJNO liu_w_Page_067.txt
3e1a9585b1a28c656122716a990b28e9
354ba0388046808d595e4f28360d60871c252553
101 F20101220_AABILB liu_w_Page_002.txt
0cbe95ab8f05fb68e4b241a881eef9df
f4f5fb3e43bd3fb1e03e8e5dcef3a7cab97e8cff
1020 F20101220_AABJOD liu_w_Page_083.txt
da7567d1f84be5c2ecbce14da6ffb972
abe4d79de6978de940cb2bb09e3c12ac8670f616
8949 F20101220_AABIKN liu_w_Page_043thm.jpg
5e751274bc5d7f6201633d8655bb63af
11e5d8ada4e6b73240754b25d15ddb86b214b140
2225 F20101220_AABJNP liu_w_Page_068.txt
dc2df7eca6f705c9ca07c815a12b63f5
3a28f2d0e43f1881ec86c3c7f2f615287c2fae7d
8912 F20101220_AABILC liu_w_Page_098thm.jpg
86f72f2015431f6bf53dda20ba43dc8a
c4bc2b4984880e6febedc5a7c15181a112c0ab3b
2102 F20101220_AABJOE liu_w_Page_084.txt
9c169a577b3df0f2379c79ee6a1d362a
b662775256e25d584e880a874fd62e348d149950
F20101220_AABIKO liu_w_Page_064.tif
160501761dc122ed02292fff860043e4
af3583ac08deb50c663b56b9855b7669afa1c647
2481 F20101220_AABJNQ liu_w_Page_069.txt
e6d867153a0fc7eb85df81caa4ac1d22
20e96d6d87b89cecda7d0085354f4ad5a457c610
62779 F20101220_AABILD liu_w_Page_087.pro
529aa65aaf1dcddeecc2a6151b8a2b43
2d1937be031f9c567cfebda1661cb6dcf3e03c66
2387 F20101220_AABJOF liu_w_Page_086.txt
eacae06ff522609d7e2d9389023599e2
ad393d7b3687da465ee7ea66a5e3b3a19ceb0c16
111865 F20101220_AABIKP liu_w_Page_095.jpg
7ea764ba8895e69a9502564ebd7ec453
5f4761d456e3a1b77404cda32f669726b9a5a218
1903 F20101220_AABJNR liu_w_Page_070.txt
aba79fa04f027752b30e6c1c81ce31b6
a490f242d609db73e86db3a88c8d2c356280979e
F20101220_AABILE liu_w_Page_033.tif
49e341b628c778efbf00348b32b2ce2e
52c248a29c9f00d3e1a8cf3c0ceee2ab86b39c21
2457 F20101220_AABJOG liu_w_Page_087.txt
414978cf7a6095aa83f37032412fe9f5
445dc4bfa1e6a34389234d68dcd5c35af3bee63d
58694 F20101220_AABIKQ liu_w_Page_111.pro
e62f5ba7d5782fab5988f9c18ba3eb1d
332e4648966753eadff70a52deaa3da08c05404b
1452 F20101220_AABJNS liu_w_Page_071.txt
5daaa3de00d2d426fbfdf2470787ac86
eb3311c50e3ff24ba9ce3ee467eb6a6a4a707fa5
F20101220_AABILF liu_w_Page_084.jp2
252eb1f718b6594e7cd1fb6f42cbcffc
275c8ada570b109f57b3deec5c1c9b676326e7a2
2391 F20101220_AABJOH liu_w_Page_088.txt
c91e3e2b8af268fb94563a3038155891
aeb41f8e03507862f7b9ace6f17bccbe99146936
2183 F20101220_AABJNT liu_w_Page_073.txt
7390cd25ed20c45793106926c8f295ff
3d29cf3a208731d24ab783218102a07801a792ca
F20101220_AABILG liu_w_Page_018.tif
f5ad3a30f6d0da8a2fe952b5bdf06d35
b127cdb51948946c39335092b6e689bdec76cbc4
1942 F20101220_AABJOI liu_w_Page_089.txt
6a1e024741ec4484a41577e90aaf8b12
d21f542ff1051bdf6caef5a798f3fda7b005e72a
2308 F20101220_AABIKR liu_w_Page_064.txt
53eecc302eddeaae8f8066cb73272359
7efb17d1b9539d35583fc5744c9b85057f62a526
798 F20101220_AABJNU liu_w_Page_074.txt
f4c808458fda573b283aac409fbdc6de
9ed814048b17ae82d9856805e58f76ca3a509abc
F20101220_AABIKS liu_w_Page_128.jp2
da75619eeda9d9ea656c1765fdfe9006
bfb35351213bb6660b72e8968b20991fe824c829
1731 F20101220_AABJNV liu_w_Page_075.txt
8fc46a28c25f4df53f34bd76c2bfd319
d1225c398ba358ca56e04ddb5385ae53cf72c4a9
22970 F20101220_AABILH liu_w_Page_162.QC.jpg
2e642cc1fb2096c36a155475cff7004c
0a1fab1056b140d8cb2a21eb224bb6673db43275
2429 F20101220_AABJOJ liu_w_Page_090.txt
774483928097659f769d24d08033a348
e1a8bc8d91a10e4cc0a7c15591bac5e29b02dcab
3845 F20101220_AABIKT liu_w_Page_114thm.jpg
d9ccf585e253295aac4ab983b166ee0d
058b25f79655f283e102d86f2be12d768124b6bc
2214 F20101220_AABJNW liu_w_Page_076.txt
4895cf70f39b3b747c050663249097eb
0630f5e13e96956b82d5e55096d7a4e7eab870fc
F20101220_AABILI liu_w_Page_091.tif
1d39fd0ab70246a56afadd67d361ad92
98e4619999ec5d4895f3dc419a08e77edf9d13a2
1866 F20101220_AABJOK liu_w_Page_091.txt
99e7cbdf177c2d61d98ab27c33cf7b79
301593e6ff3e71acde2391503fe4185104b06ccb
29967 F20101220_AABIKU liu_w_Page_079.QC.jpg
33c1384e53554aad33ef04d53953e828
8cc1c0ddee5a206165e554de0ae848260effb301
2213 F20101220_AABJPA liu_w_Page_109.txt
40028c6c08379871bb5772af44bd71df
1ee04e6752d800078294406a14aad9fae5c6626b
2359 F20101220_AABJNX liu_w_Page_077.txt
ab98b99fbaf5c686fba25abf07cb4489
ed67b6a45b65d8dfb24aee08bfda58321f7ccb0c
8467 F20101220_AABILJ liu_w_Page_032thm.jpg
6cdf60a70611e7f0cb6ea05d41a2e343
780a07d8257e18dacac15d6e291c3c75f069c186
1816 F20101220_AABJOL liu_w_Page_092.txt
0b166f8e3a2532d80c91c3c534985419
77b8fd3fd9752d1c84f7d552f5a89f7d9a18f9af
2296 F20101220_AABIKV liu_w_Page_085.txt
d87c0310b55cc6d442e76081143329e1
061fbde50a56196273f1730425db70da8feb5cf5
2345 F20101220_AABJPB liu_w_Page_110.txt
266938c48adfef1760158a5228a9a040
6888e4dab5c03328b1c3da224386ff8ae8064086
2048 F20101220_AABJNY liu_w_Page_078.txt
1d78663ff6185862ca18cbbb66d87576
703aeb0701740d4b38fa4dd8ac7832b23b0446c9
2243 F20101220_AABILK liu_w_Page_047.txt
06590a0cbfc8116987e3829d2c2759f3
85504435be59565fbc3764b47fbdf267c5bc94c5
1300 F20101220_AABJOM liu_w_Page_093.txt
64ce34c70a5a68bf768c5d62bdad0845
3a44799d3f1b899c62fdaeb57e57d46ed7bd230e
8623 F20101220_AABIKW liu_w_Page_085thm.jpg
935a3567157bb05ecfd43071ef736ef8
b6f81a27eaf59ca39adc8a72e567e7a2662c633c
F20101220_AABJPC liu_w_Page_111.txt
ca5d312faabeb5f45e539cd96526bf74
0b342b3b5b2a0c900227c9c7cef1b7c66a0c9a64
2034 F20101220_AABJNZ liu_w_Page_079.txt
a86ae9f6b993ef6d5a3425b98847f937
c1b49e910ea25e6f1b6f5b0fa49879e2c4a48303
122518 F20101220_AABIMA liu_w_Page_033.jpg
dc253443e7386aa5268ee2474bc2dc4d
0bdab7183ebf638f93c9d6b357800a9ca9425f2f
9065 F20101220_AABILL liu_w_Page_050.QC.jpg
d4209f2ddb8f36f1ff10aa0ad2fd27a6
fb17adcccba9ddbb4fa806925e704f8593b95966
2134 F20101220_AABJON liu_w_Page_095.txt
fedc71b5ffb619dfd06dce087a0b273a
8ba5dbc51913e431dcc8a7c0718fe1a77b8b5056
2114 F20101220_AABIKX liu_w_Page_094.txt
f5d5e46b45c0a2cef808e5b5e674bf91
6c5405d4da974517b774b20ba98eee75815c9d50
2251 F20101220_AABJPD liu_w_Page_112.txt
c16a06f8ec84c129794f4d7d9c88d40f
a55f9704ca8a03284371840d1d74a4ca5ef5f05f
125071 F20101220_AABIMB liu_w_Page_155.jpg
10cfd7af0c5426cd46b99c9b9800bb06
99d2322a410f9fbfd9501e1d7b125ab6af52cab0
F20101220_AABILM liu_w_Page_011thm.jpg
42bffafcbb5cf69e5d654a9ec500ceb9
08e34076f05db8e49295912a4fec36b2b8473122
1471 F20101220_AABJOO liu_w_Page_096.txt
6199368c5cc5d4d9cd1d071a3707fa1f
ec6bdedf18195e29b3f3d816816fd14f5afcd1de
33229 F20101220_AABIKY liu_w_Page_084.QC.jpg
5b8b0b54f29a22a421932471ed5eb16f
efb155e070f0e5dff920b7fb8f1e22be6cf2f295
2294 F20101220_AABJPE liu_w_Page_113.txt
cb58677e17016bde051de89623d66d86
5da6237c8867668180c01a9b25933f40010daaa0
136022 F20101220_AABIMC liu_w_Page_160.jpg
ec0944a0a23a62de919409c3bb6b4b1f
38f53649d60b12480cd8c7ccd10dc1bcf93f8b80
3447570 F20101220_AABILN liu_w.pdf
2ed73a9b3f98f89c2968fb8e516f467d
d2f347fd4406c62efd672465283cae01b1f277e1
BROKEN_LINK
www.proxim.com/learn/library/datasheet/11abgcombocard.pdf
2461 F20101220_AABJOP liu_w_Page_097.txt
000abbca0a115ad8daa25eb20ad89939
2082da8899b07ba2ea18c24c9dbab8157dbf6f72
1927 F20101220_AABIKZ liu_w_Page_011.txt
0b39c6dc37eeb4ae72b0345d643f7d46
72eeb0df6b8cc6d85a491af5bc56868904331908
1110 F20101220_AABJPF liu_w_Page_114.txt
db19d2561d2c26de3372303492eb5652
b4baf2b9702a44c2336ae77711c491cc5c857ea7
2446 F20101220_AABJOQ liu_w_Page_098.txt
df975f094f2c3e52f7d5e08db43b805a
e7089d3027bedd39ac73b600ca53211edb053f24
4052 F20101220_AABIMD liu_w_Page_002.jpg
6f7c9147f542389e5bc301e32003d12f
04743030d4cff2877af68fce948114f100f1ffe4
117708 F20101220_AABILO liu_w_Page_062.jpg
b98e48410bc62202d97d5ff738d03a8b
e804e1cdb429ec0d64133c66c972f442c691a4b0
2466 F20101220_AABJPG liu_w_Page_115.txt
a68415563a7814995b02710b42ecddc7
952ec0aae26f22176b9f32cba71dc7140a89b653
2092 F20101220_AABJOR liu_w_Page_099.txt
242cffab177bc9c42a89d58c2a6dddc8
94ff7872fe6b3427a6a357df8a965ae030031dc8
2322 F20101220_AABIME liu_w_Page_135.txt
0d53b7a8e2ad14b7789650d5875032f6
1edf8a3b100b1a69512a1787397528d1c69ef7eb
1134 F20101220_AABILP liu_w_Page_012.txt
20665bbe5edb2dfc2128b5e0702411db
f30dada4057eea74a0507da24675f6d43bf64095
2396 F20101220_AABJPH liu_w_Page_116.txt
2dc9408de50fd7a9f77bfb9dba4ef130
e8363d753280225a1fad8d945cb23c9aa7d52983
2492 F20101220_AABJOS liu_w_Page_100.txt
165d38c47823326c4876ff1e304b5200
bbafb20c7ccb4767bb501bb95e2e4d665394a792
F20101220_AABIMF liu_w_Page_016.jp2
d6eaaf56d785abf6c460e5ff9ee228b6
8d637d25fd90d62cadbb6dc88c517417016ff1fa
31003 F20101220_AABILQ liu_w_Page_102.pro
822ff68e7310b28dece9a6bd9674bc3f
dc10a1139f52dd634fb8e3a4623bd06bc24a5138
736 F20101220_AABJPI liu_w_Page_117.txt
5b9c4e86ebaa174808ee07e78c072089
bce8d72df8b1070d88d254359ce95f6017eaaebe
2394 F20101220_AABJOT liu_w_Page_101.txt
7d52b4342ca4c7879263b9fbb919c95a
bc462385add390592d184076f659e7fda7a12c15
53107 F20101220_AABIMG liu_w_Page_059.pro
518d9702ea0dbc891306653adfe1775d
954634a004d217561958d41d41bc3688594ee5cf
8948 F20101220_AABILR liu_w_Page_119thm.jpg
bb6541e13f6e09510775999ea13bd0b5
f7699385a4c93286b33e3c29a6e7de1c8d3c8e9b
2090 F20101220_AABJPJ liu_w_Page_118.txt
a26aff2c27e82233a2635335bc44c7c0
c164fe92f6a016cb14309259a189d72d9183541b
1526 F20101220_AABJOU liu_w_Page_102.txt
f9f5260dd47eb27b334b1bf0fe71882e
aa5bcfbc59ba9c75a3202bd00da7ca3c77dbb83d
63930 F20101220_AABIMH liu_w_Page_157.pro
fed929c1eb3780b9b5e8f41d95674c8e
deec0b3c11dcdd10cdb019e15da09ee130ed5ae0
50820 F20101220_AABILS liu_w_Page_081.jpg
7dc687341fa7a674bc8876f4d5184d98
d3aac833bbb2821c942daf8113f3288489bc8c52
F20101220_AABJOV liu_w_Page_103.txt
3aa6301e6207360a16a5763a03e1e79a
3bd618e8038d73a3b8a2d69d085744b1eef57b17
F20101220_AABILT liu_w_Page_028.tif
4b7515cf6ec44043edc843cba201eb2c
0bada3cfb7e4770847b45a2308bae2fc25c99d41
F20101220_AABJPK liu_w_Page_119.txt
1759864cce78f15fa13a6740d12b6391
e42079cb1650c911f19382e3f1a9447fa29039c4
2347 F20101220_AABJOW liu_w_Page_104.txt
b73c3314d8f9d4d4eca227d2b95e7fe4
fccb2e270a58a252c85c3e1dcb81938406f93d0b
59061 F20101220_AABIMI liu_w_Page_133.pro
22f44d453922041b9c74ea7d63ec1ec5
79dcb6a0a44ab21107e387052baad90cefc04995
1051933 F20101220_AABILU liu_w_Page_085.jp2
cd933ccbb82bf1a4f16ce46a87e1f45b
ddcae33b1c941808fde1b65d34ac448e6c89f35a
1916 F20101220_AABJQA liu_w_Page_136.txt
9a45638099805e512eba89e2c70cb56c
fedd1dd7c71b99ed658f567d44bdcfc8676f4b27
2400 F20101220_AABJPL liu_w_Page_120.txt
5e03b29e2b0fe22153d73f83b004853f
41faa1f5318d16eb0bbc323abd353824e8f0b03a
F20101220_AABJOX liu_w_Page_105.txt
22d7c3bbdbc46acc5b7af02555d5324e
cd16b00d376b7e6849ca97fec873d4656fcfa66f
F20101220_AABIMJ liu_w_Page_088.tif
9818744a55c349d89df030b61e263eec
3253dd0df8d41f38236676591d1e20646aaf248e
2302 F20101220_AABILV liu_w_Page_063.txt
6379d1580775c89eb72c21f15fe832a2
64d8c198f0e5ed8bcdd97743a299465bfed80dd4
2397 F20101220_AABJQB liu_w_Page_137.txt
03dc39be11f63c83099f7388c5fec345
7bfe515571e300b5124d347de6b6457db0354b14
2377 F20101220_AABJPM liu_w_Page_121.txt
b3e0e887ac023d11fd2c8513c276410e
5c9df4dc6bfb6aa02493775ce3d7771dc37d4fad
2232 F20101220_AABJOY liu_w_Page_106.txt
4ed3ce1787cc1f085da8d39f4930cc35
854ced96d816b25bc2e1043adb38550b60b43e5a
952 F20101220_AABIMK liu_w_Page_057.txt
cdfae46a2f5029d8f9076f847d55e61b
33af3f27b562c5d8e6b28f091d0be4535a86c523
49313 F20101220_AABILW liu_w_Page_023.pro
32b6f5205889aeacf9d3d6ded16b90cd
9cf6fb3423d4f481f765c81c8ab741bba87b94b7
2300 F20101220_AABJQC liu_w_Page_138.txt
1c74ed317c8fb0350f93b5c263b60b53
b9386e6377cbc21e20d5312015fd147200a779b6
2425 F20101220_AABJPN liu_w_Page_122.txt
3f3f98bdb58d02938ce557b8c3b84ca8
72cb948d6299ddef5a7f4db7c6492c1f5f99e596
770 F20101220_AABJOZ liu_w_Page_108.txt
312acaf70a7a0a2e9740f462d42ac738
20b466ef45efc70b1db8c90a370384579fd7be0d
F20101220_AABIML liu_w_Page_113.tif
1e3a745ce1c5ed90e1669122b21d6962
e350888c68006fbd2127a0c2898a0c160fd84c54
119311 F20101220_AABILX liu_w_Page_133.jpg
971dd2ab544b9bf3bd156cd6594652f0
9ee697e5d4155be3f9d2eae58900971a5f9f8e9d
31395 F20101220_AABINA liu_w_Page_030.QC.jpg
4c92f0112e7fcfd78a2d470687be08d7
24ab8bce106b92cff154a15495bf5b80226a6baf
2212 F20101220_AABJQD liu_w_Page_139.txt
28abe80d19a0099b3cb0420aa9927d9d
940d74ee454e87fb94aec7d1730b61fa4d4e7646
2088 F20101220_AABJPO liu_w_Page_123.txt
501e5fe56df2b5c8d68a3eade4b1d036
290eb015d2e37419a7f6975e1d3847a36b3e3d81
127004 F20101220_AABIMM liu_w_Page_159.jpg
aecf9e96589270e86e481bf92061046c
8b7e4e996eac2dfe13f61229af873df0396c171f
54069 F20101220_AABILY liu_w_Page_105.pro
e2ee0efd5d32734547ec7881433b186e
de39f9757b4d9a79a2fbb27165c81e5fdb0603dd
36870 F20101220_AABINB liu_w_Page_048.QC.jpg
b672bd8c98ebd0dc26c3de3649c2f541
df32f6aa12f9d8aea9b7bf6564cb3f3ebd1da773
2321 F20101220_AABJQE liu_w_Page_140.txt
73910f33a259202f78214a8b5484c3ac
70df044eb0e9aa1d300cdf204e1f0837884ef47d
2324 F20101220_AABJPP liu_w_Page_124.txt
3bbbb10086377e7c525da30925c7cef0
03767ca7e67cec34533bac71a2bf5777f7385f5a
59395 F20101220_AABIMN liu_w_Page_033.pro
d485bb95dd07c982160ffc8104e185c6
de75920aded466812983f07a03a6aef8f962d2d6
2482 F20101220_AABILZ liu_w_Page_107.txt
7f6a81a81c058bc92f6608e24dd28da5
2c2e992db6aebdd6631e1d78d9c5e8349ce10483
31470 F20101220_AABINC liu_w_Page_150.QC.jpg
fc5a43c802bb9ac2b1a3a2f054d144f1
1bb91d4e34dc43208ad0397c81a45451040f1697
2175 F20101220_AABJQF liu_w_Page_141.txt
47ee7fd9b81d244777b406175fffa346
8c404f7d7ba1eac3200b72631ddaffcc5c36a4bc
2468 F20101220_AABJPQ liu_w_Page_125.txt
89bb51d0acd2baa27227986338e4d02d
f187d1c73c3396c4206bd0cb2bcc11c4117afa68
19606 F20101220_AABIMO liu_w_Page_074.pro
2e53861457f6f08ad6dbc9b37346a348
c71b3f77e4a4ebff3bc6ef65f908d7cdf5539d37
878569 F20101220_AABIND liu_w_Page_164.jp2
d9bc9cb14a16fec01e79240ca61d1a5a
f56e5b5589a034342977f3cbd5b8b2f8e4997573
2007 F20101220_AABJQG liu_w_Page_142.txt
a2167bd8bc05edd8d943bf662e29ac8f
71bf500a259bd85a19949a55c1baf9cc1f73369a
2406 F20101220_AABJPR liu_w_Page_126.txt
42c368f9a696161230697de367ee59b1
32a73024f80caca185b55f10e1d77bb1778e1018
30785 F20101220_AABIMP liu_w_Page_076.QC.jpg
dc7951abe2dcf1c9e707519f10b5b624
375780a2cec65757f5c3073fc81fe9db7551bd45
F20101220_AABINE liu_w_Page_115.tif
870bb11522501610aa3853cd2c2e8dfd
ef5a859486c8581c8131a8a8e640820961d19240
991 F20101220_AABJQH liu_w_Page_143.txt
dc1a7b8a4c1d914be73c80958f8db70c
707a0a3659e425ff456140384e9b1927a77c1b06
2616 F20101220_AABJPS liu_w_Page_127.txt
5fd89acf9c9fd04df533ca615e14ff5d
a8877ca807c7c19430a7d56694295267754e8dc7
17015 F20101220_AABIMQ liu_w_Page_083.QC.jpg
a2ab6e118b8f7a80e33365748d56e357
5327ee3368dfb44ecc57b7066d77963a66e9631d
10114 F20101220_AABINF liu_w_Page_008.QC.jpg
e01c3ca5d4735fe5756b5922cbdd3a70
adf88f116b29bfe6a19985a412508652c593c728
1793 F20101220_AABJQI liu_w_Page_144.txt
8ae6c48a5e803fd9baf9a7b222f6c740
c4e5cdff66b8129a578df2bc4d845ece3c6895eb
2349 F20101220_AABJPT liu_w_Page_128.txt
a187e710212bfd817febc0f3183dba85
861ed74f84f2b3d0c6a15ae87b3482f0f2f7762b
F20101220_AABIMR liu_w_Page_072.txt
a46ffb2cb7c8dbcfe54f7e11ed54a4e7
4964a86f3ad65f79e43c4ddcf0f0c2aa4d175e88
7711 F20101220_AABING liu_w_Page_006thm.jpg
bc7e4000c85f8fb2ad7082e28d6c567b
c086e1e654c8fe9e04972ea2354dba75c8b3dc2f
2317 F20101220_AABJQJ liu_w_Page_145.txt
5664e8ba0cce789dccaa2ed037825ab8
9ace8e395fe0dd98f843f4a48cf679c5c288e150
2130 F20101220_AABJPU liu_w_Page_129.txt
05988a5e06229bb12672e8e2b903d8f9
ba254002ce2ad65965d1a3deb4d893992131b655
1777 F20101220_AABIMS liu_w_Page_004.txt
ab42bcac5debe961e5463c09a8660c20
9eda6fe9ec35a0ff75c4edfd18901cf3a57492ac
244618 F20101220_AABINH UFE0011611_00001.xml
5f178ee63086f15c49e1703677e6bdc0
03f812fee2d95db4e28ca328f16e36560d6a8e2c
1534 F20101220_AABJQK liu_w_Page_146.txt
1125001d1d4de992f57c9fa96fadce51
284b8c1e52244eddafa4db5c536157bbb2ac78ab
2545 F20101220_AABJPV liu_w_Page_130.txt
c11c2b557ac5c65fed298cc836dda440
d8ac1c95f824fd3865f2555df8e1a5422cbcd2d5
F20101220_AABIMT liu_w_Page_160.tif
be99e2f09b3538f4716bf0a37588ee39
d218e1b5337b1d9350ca145082e0c077810c808a
2174 F20101220_AABJPW liu_w_Page_131.txt
f7f5a6a55141581a50181a8c21a9cb21
4074d9d563063911006a8ea7df6f522f958cfda2
F20101220_AABIMU liu_w_Page_009.tif
0cc1a263d3af9277b28f8ad98f43541f
3558d0e4c5a9d19621a46b9ac4f55cf20f983d57
1603 F20101220_AABJRA liu_w_Page_162.txt
ae5af0767f59f0c25709628ec7296c3c
0258d48a680b2b1b9d0f8070bcd818fa18b994d6
412 F20101220_AABJQL liu_w_Page_147.txt
eaff16b62a73e3a1b5cc234a82c33da1
a0e67b72a2ec62ed67a46877a57d7c72c877c9e9
1112 F20101220_AABJPX liu_w_Page_132.txt
4b3106489018cf9d850778d535b65481
c85aa902e2796c75fa89adccf3f780419de04b5e
F20101220_AABIMV liu_w_Page_143.tif
9ceb796784831b9247cfd40f6479a99c
d2e2dd9b3f860919b4730cc7f7f898c5942c23cd
627 F20101220_AABJRB liu_w_Page_163.txt
5d5d46e992c41ebdb19c51149e802f7e
73954a3257d279a50c155870c83ba8e21c84d838
2432 F20101220_AABJQM liu_w_Page_148.txt
05bb5800c71767948577e0303c7808fe
e55fb36bf7c6d161c0fdac5b1d051c1a86a4a674
2328 F20101220_AABJPY liu_w_Page_133.txt
183085321537126cd21f0279cbf0a792
1d287b6e396c0bb53382ec1e908421c6a77e7843
7443 F20101220_AABIMW liu_w_Page_150thm.jpg
3ae09fad57a6ec70ff83eb75560e56aa
eb1b2cb9465501afa061a90932532149ba3e03a3
25371 F20101220_AABINK liu_w_Page_001.jpg
c1285026a23978a462210df841de0f21
67fbb84f28d8a9d3cfbca41be928bf7389f42caf
2271 F20101220_AABJRC liu_w_Page_164.txt
e4b5f430795942ed63dca7f4f0bf3077
cd8dc4d42a4d21514339763aacd8f11486053e0b
F20101220_AABJQN liu_w_Page_149.txt
65acade08b833c3b111d049848a2b9bb
0d707c8abee220721dd3c0ca2897621b0e48bb93
2172 F20101220_AABJPZ liu_w_Page_134.txt
b571f00a20914da784a047eba73d35bd
d6eee67b3d5bb6b3da398fd74af00ece50d4d5b6
F20101220_AABIMX liu_w_Page_006.jp2
8ca48ca8679021570a1fa67ed9fae21c
81f5b8c9ee4da8950da2c1ad255836e940fb815d
125060 F20101220_AABIOA liu_w_Page_018.jpg
e23852d76cdf7185bac64365715b1bd5
3ce5919134935e09ae824283d89d1e204331d856
5121 F20101220_AABINL liu_w_Page_003.jpg
3ea836a4cd44e8a2ee11f429677da865
a70dbe20e2d53e805dde47195b8cae610e08873a
520 F20101220_AABJRD liu_w_Page_165.txt
719f9a8ec0b514f354878d9794c24f0a
5c053327f1023e007e0d58be89d27364ecc62fea
2021 F20101220_AABJQO liu_w_Page_150.txt
d12c738e6de7e9a11d5c0562a49bd398
50d27928406a11d77bfac2c1c50f338fe69da414
F20101220_AABIMY liu_w_Page_124.jp2
fe7206bd8c23d7d6eb6b4cdce62fb4d5
c9f85e9ddbb97e1f6ef4c12f8d253519dcc695bf
113717 F20101220_AABIOB liu_w_Page_019.jpg
355769216924d8ed14fe6458cfe3096e
e6d72fde8b84b94af8ddb410f731196c5da64d5c
86639 F20101220_AABINM liu_w_Page_004.jpg
6a15e3328a5da08d3dca23dd8df613eb
e12370878e5b33a4bfb62816e1f67cf14d6e333c
1254 F20101220_AABJRE liu_w_Page_166.txt
6c22be45411d7576bd1fcce7efff2c0a
f0c07e3ace3b791aa635f4fbc1c021df1ef9c158
2219 F20101220_AABJQP liu_w_Page_151.txt
87680422193d98578d28f7d9b4e90718
ee9010ac0eca6d6332d1d761508866a39971c1a3
1701 F20101220_AABIMZ liu_w_Page_046.txt
5ec0b861bf954d657fd303f2baa04fa8
b488b1ccb53c9fc5d571dc31fb82dd8b51ad2617
111780 F20101220_AABIOC liu_w_Page_020.jpg
5eea1d6729f967592b1ad95a1d64ab57
9191bb71782dd6f0d03630be6a8bab27f721e88e
107143 F20101220_AABINN liu_w_Page_005.jpg
fb5c37d9a298b1e5b86132481b64960b
bff37c0b32542bbeed9b4f883f3ce55417ad3c28
F20101220_AABJRF liu_w_Page_001thm.jpg
d342753bc0c2909f48d26f41de87c544
049258c916e57c7a2c0aad9408446c14d0e45f89
446 F20101220_AABJQQ liu_w_Page_152.txt
4183dfdd985fdbd5c127fef3a661a097
18e64bdc1fe5407abd847a116901ce40efea0d86
122500 F20101220_AABIOD liu_w_Page_021.jpg
f02f6bd05e1f3aa33c7fccb60279744c
31e428fc8518a51d87cae0d4a50efc05ef476a58
136269 F20101220_AABINO liu_w_Page_006.jpg
6d64cec1c4ff70f073a7c10c25450841
068b82ecd8b705f5f8de8472d49c5879cce22ef3
6896 F20101220_AABJRG liu_w_Page_001.QC.jpg
7a1ebd0b93932e20ed628fa581f6e989
36ebadbcf9770ff77362ae12a317d15922213323
2185 F20101220_AABJQR liu_w_Page_153.txt
1aae6c0f6b34d1bc861aec392af78b8b
dc943c2e7558476c160fc6a2b52d6b8dc5629997
27301 F20101220_AABIOE liu_w_Page_022.jpg
f0e81ca819d5481b29e6ede5c5900f94
030875ac1373dd2526b4fc7fb8b124ba26f03c1a
79170 F20101220_AABINP liu_w_Page_007.jpg
29cbdf4994a4211593d39088a4546822
ebd7b8b806f236e6f7eefdb3a5fd2febb30bde2a
1264 F20101220_AABJRH liu_w_Page_002.QC.jpg
45ca883e8b53e6d27b5a49da721e379a
d6c5f14604cd4743b26c3b39e104f50b9d803621
F20101220_AABJQS liu_w_Page_154.txt
8a84389873adf34fa17fd171312f7f24
de8e5f3780115618df909d952a29e8840f4efd01
100472 F20101220_AABIOF liu_w_Page_023.jpg
f566b40a84ae9db521001205dbab30b9
4b25fcad76252325a6e619e12182d4f6ab80ebdd
32427 F20101220_AABINQ liu_w_Page_008.jpg
7563e40b43ecea60efab38fd70b965a3
4667fce7128763834d31e3a60465f1690e86fd90
561 F20101220_AABJRI liu_w_Page_002thm.jpg
8c33c52a01ed634c1a5379db54c91833
edadd57fdae4f5db7e8dada8599ee4a8e62cbf39
2445 F20101220_AABJQT liu_w_Page_155.txt
66afb93988dd189a2a5c846a4b6d93d4
8b85d99264ceea75aa9fb9113799135a5b0f5fa7
114363 F20101220_AABIOG liu_w_Page_024.jpg
c882882c4bf3d0ced7c39a466a934ae9
e1a13b404908822b44947c899914cbb0a348d3ba
107537 F20101220_AABINR liu_w_Page_009.jpg
f425a1355ea3f672ceb0ef0c4d01423b
3e3e9240e0a3469adb22258f1c902e8f1bb662ce


xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID E20110409_AAAAIP INGEST_TIME 2011-04-10T02:46:14Z PACKAGE UFE0011611_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES
FILE SIZE 8911 DFID F20110409_AACBXC ORIGIN DEPOSITOR PATH liu_w_Page_063thm.jpg GLOBAL false PRESERVATION BIT MESSAGE_DIGEST ALGORITHM MD5
e7cebe61629a79457f269209720586d3
SHA-1
0868d8af4ed023077974275eaea9664493746c71
8842 F20110409_AACBWN liu_w_Page_048thm.jpg
d66aa18e51af7bb3219e866aa6a89c87
7db4d19dc218e22969a687f9f3389cc84e0e963b
9181 F20110409_AACBVZ liu_w_Page_033thm.jpg
c797eab0e3fd4d44b653c1a30f1b7de5
de932d3d369a54b991ac393d217ae96a31107648
798 F20110409_AACATL liu_w_Page_074.txt
f4c808458fda573b283aac409fbdc6de
9ed814048b17ae82d9856805e58f76ca3a509abc
2257 F20110409_AACASX liu_w_Page_056.txt
c7b1efdd684191c1da1af5be0596925a
e6a793ef0134b3262686fb7706dbc6202e8206d9
2429 F20110409_AACAUA liu_w_Page_090.txt
774483928097659f769d24d08033a348
e1a8bc8d91a10e4cc0a7c15591bac5e29b02dcab
8774 F20110409_AACBXD liu_w_Page_064thm.jpg
b20435408b0d48f83a7c90372ed8eda1
7e0e3c94349a4a3142eebf9dbefef403053c90c1
7429 F20110409_AACBWO liu_w_Page_049thm.jpg
a4a38f0b14457d6714723a46b702d69a
d1aaf757aee1cdba2758aff53b8c1509953839ab
1731 F20110409_AACATM liu_w_Page_075.txt
8fc46a28c25f4df53f34bd76c2bfd319
d1225c398ba358ca56e04ddb5385ae53cf72c4a9
1987 F20110409_AACASY liu_w_Page_058.txt
81904ff74a0cbc78707b8bafba7fe07e
3aba8e3580c82514517bd6e3154b81d5cb93c616
1866 F20110409_AACAUB liu_w_Page_091.txt
99e7cbdf177c2d61d98ab27c33cf7b79
301593e6ff3e71acde2391503fe4185104b06ccb
9190 F20110409_AACBXE liu_w_Page_065thm.jpg
8958c78ac0d871b24cdceb745e6224df
9edd46571b541c731bbe1e508fc2f42b9465dabe
2694 F20110409_AACBWP liu_w_Page_050thm.jpg
d213ef63f379fc0d8d1460bb671cf613
0c88faa503ee62a92e68f33c0e142bef11dbea37
2214 F20110409_AACATN liu_w_Page_076.txt
4895cf70f39b3b747c050663249097eb
0630f5e13e96956b82d5e55096d7a4e7eab870fc
2139 F20110409_AACASZ liu_w_Page_059.txt
c3bbc0f217e26dfa57214785ca048a7e
97c6af4b02725741a52183877c63dc400896786a
1816 F20110409_AACAUC liu_w_Page_092.txt
0b166f8e3a2532d80c91c3c534985419
77b8fd3fd9752d1c84f7d552f5a89f7d9a18f9af
8658 F20110409_AACBXF liu_w_Page_066thm.jpg
827a34bdf9797a0d68d6a718a4d8033c
533fec473774f260d33561e34a07033e5f7d9a0a
8935 F20110409_AACBWQ liu_w_Page_051thm.jpg
10f7157d26af76b544a988eec43b4059
4522f9818eb9fbdf44ee5759bf5d6a27e446ccbf
2359 F20110409_AACATO liu_w_Page_077.txt
ab98b99fbaf5c686fba25abf07cb4489
ed67b6a45b65d8dfb24aee08bfda58321f7ccb0c
1300 F20110409_AACAUD liu_w_Page_093.txt
64ce34c70a5a68bf768c5d62bdad0845
3a44799d3f1b899c62fdaeb57e57d46ed7bd230e
7288 F20110409_AACBXG liu_w_Page_067thm.jpg
037b864dbda9796a2a2c9286386c1278
e126638e030980552322e4beaa073c8f24b81802
8633 F20110409_AACBWR liu_w_Page_052thm.jpg
2abaa5f429d0e3a0f4670dca23df98e4
0887aa53081620137c3dd9f13f7dddd45cdfe5ad
2048 F20110409_AACATP liu_w_Page_078.txt
1d78663ff6185862ca18cbbb66d87576
703aeb0701740d4b38fa4dd8ac7832b23b0446c9
2134 F20110409_AACAUE liu_w_Page_095.txt
fedc71b5ffb619dfd06dce087a0b273a
8ba5dbc51913e431dcc8a7c0718fe1a77b8b5056
8577 F20110409_AACBXH liu_w_Page_068thm.jpg
b80628d537e85284c882b5926dd4a8cf
a26b70b847189afb2e52dd4a9d5998ef876ab18e
8461 F20110409_AACBWS liu_w_Page_053thm.jpg
af0f1db66a7dcb7162f443c093e7afa7
6ab4b542bb4718afa2cfdb519ad49bb97eb134a0
2034 F20110409_AACATQ liu_w_Page_079.txt
a86ae9f6b993ef6d5a3425b98847f937
c1b49e910ea25e6f1b6f5b0fa49879e2c4a48303
1471 F20110409_AACAUF liu_w_Page_096.txt
6199368c5cc5d4d9cd1d071a3707fa1f
ec6bdedf18195e29b3f3d816816fd14f5afcd1de
8932 F20110409_AACBXI liu_w_Page_069thm.jpg
30bcbd7184dd09cf39035b430ea4449b
e3e04e716886894917df237244324d905ef8c6a6
9223 F20110409_AACBWT liu_w_Page_054thm.jpg
d98d7017ea57a977445364b6c4e1a334
e72ee04241682e1aa15dc078005cdbbb877ba024
2295 F20110409_AACATR liu_w_Page_080.txt
7eb851007f35a7ec8d3d4e7cda475171
863e331b73172e96a952223fc29364df664b18f6
60069 F20110409_AACBAA liu_w_Page_088.pro
b6dc20304ec4c4d34e51a6ee370151ca
74284932d5ca9cdac25e99eb4467cba5fcd2f779
2461 F20110409_AACAUG liu_w_Page_097.txt
000abbca0a115ad8daa25eb20ad89939
2082da8899b07ba2ea18c24c9dbab8157dbf6f72
8013 F20110409_AACBXJ liu_w_Page_070thm.jpg
90fc9958b8336d040541373e26f0a48f
83effc4cb867fe991ca898ba7e9e3242624fabc1
3395 F20110409_AACBWU liu_w_Page_055thm.jpg
222fc944c02a4ab0e67b92bb42c34547
ad314705d0b7c300d95d9ec511d909023ea45fbe
1130 F20110409_AACATS liu_w_Page_081.txt
92a4aaadf9aa00d58454c2f9dc3b3689
ec5124989d338e0737a1f7793d6bfaf339a72ab2
46879 F20110409_AACBAB liu_w_Page_089.pro
50448a05555b201ff39b8ce140c3c490
4d3e7919cd06c43e4c066007f20112fe072f6011
2446 F20110409_AACAUH liu_w_Page_098.txt
df975f094f2c3e52f7d5e08db43b805a
e7089d3027bedd39ac73b600ca53211edb053f24
8979 F20110409_AACBWV liu_w_Page_056thm.jpg
a6b7563f82031dc2b10ecd31a689b11e
8beb85bd59788fda44f57941c328a65f2c72e223
2268 F20110409_AACATT liu_w_Page_082.txt
23890852deadb002aa233540823c6ab4
a588a8c1f73750dfd6a54ae62367e4be278d42a1
61792 F20110409_AACBAC liu_w_Page_090.pro
26a2e8cce9b7e43e94340fb9b628ab85
33439d1c5a87bbee2d325da4250e312402b650f2
6113 F20110409_AACBXK liu_w_Page_071thm.jpg
d2a8219a88ac8bbc26da4512938db5ea
c202053e30de8e4f06b9115566339cb785ecec04
3542 F20110409_AACBWW liu_w_Page_057thm.jpg
bc358c958bd3830fd211a133a858f4b7
77799435ae96cd4c7eb3d3749609ce57b66f9ad6
1020 F20110409_AACATU liu_w_Page_083.txt
da7567d1f84be5c2ecbce14da6ffb972
abe4d79de6978de940cb2bb09e3c12ac8670f616
45136 F20110409_AACBAD liu_w_Page_091.pro
383b2d31c04e88e7746270c4e89feb2f
6a7cc7b7007b6c09a2196111e4d816c1e9acd967
2092 F20110409_AACAUI liu_w_Page_099.txt
242cffab177bc9c42a89d58c2a6dddc8
94ff7872fe6b3427a6a357df8a965ae030031dc8
9138 F20110409_AACBYA liu_w_Page_088thm.jpg
031d2d59a1bf4efbfbf2fe0e82bfca3d
7576e3751bbffcd69104a65d6344c7b74bf2452f
7393 F20110409_AACBXL liu_w_Page_072thm.jpg
ed67f0319cb66a1b1420001f2aff6817
383b859772920646a289d4944e75c2f52babf6d5
5542 F20110409_AACBWX liu_w_Page_058thm.jpg
2e0da1176f055718228b96e54549f786
72c9bc16a71aaeb08b86e3a241e42e6bd3b14e5b
2102 F20110409_AACATV liu_w_Page_084.txt
9c169a577b3df0f2379c79ee6a1d362a
b662775256e25d584e880a874fd62e348d149950
41778 F20110409_AACBAE liu_w_Page_092.pro
3933ca66d16149711526d7b821bf447e
63584382ccf9e47984b5b837a7af27ef7b22798b
2492 F20110409_AACAUJ liu_w_Page_100.txt
165d38c47823326c4876ff1e304b5200
bbafb20c7ccb4767bb501bb95e2e4d665394a792
7330 F20110409_AACBYB liu_w_Page_089thm.jpg
0cd98b78b84c0e11a1c9d355b59240dd
1ad288e4a178982a02690ec92214b5c1f6f46aa5
7415 F20110409_AACBXM liu_w_Page_073thm.jpg
020326a162d2f0bd57940417cd0beb48
50878631fb5a0246e98079b0d05ac096c5fafc22
8212 F20110409_AACBWY liu_w_Page_059thm.jpg
78a48ed746c304bee79be65531fd9c70
1a41d0cf0050612114820ac4264cb8623af0f7aa
2387 F20110409_AACATW liu_w_Page_086.txt
eacae06ff522609d7e2d9389023599e2
ad393d7b3687da465ee7ea66a5e3b3a19ceb0c16
24682 F20110409_AACBAF liu_w_Page_093.pro
97386d4630cbd02e1b506cb77147658f
fcdd1e0543545e61360ab3793ee76c6418f134ea
2394 F20110409_AACAUK liu_w_Page_101.txt
7d52b4342ca4c7879263b9fbb919c95a
bc462385add390592d184076f659e7fda7a12c15
9182 F20110409_AACBYC liu_w_Page_090thm.jpg
74dcbf64392e401fce4233b3a7bb6a26
d2a701bdeec346a9d16b70897eb7ad15f36ecfa5
3624 F20110409_AACBXN liu_w_Page_074thm.jpg
e9240ba9ec78e93d411444e954c95f7e
1d7d9f7630f676ee5b5c70c2f9beb5a34da5f291
4231 F20110409_AACBWZ liu_w_Page_060thm.jpg
d57661e74f9300af64ca84c72072f38e
11f0c2ee6edf10cb935d673b17757950655792ae
2457 F20110409_AACATX liu_w_Page_087.txt
414978cf7a6095aa83f37032412fe9f5
445dc4bfa1e6a34389234d68dcd5c35af3bee63d
52522 F20110409_AACBAG liu_w_Page_094.pro
4bb27182e48087939e69a5006f7c6774
4483eb482d833b2092e4dda959a35b03819cc8b2
2090 F20110409_AACAVA liu_w_Page_118.txt
a26aff2c27e82233a2635335bc44c7c0
c164fe92f6a016cb14309259a189d72d9183541b
1526 F20110409_AACAUL liu_w_Page_102.txt
f9f5260dd47eb27b334b1bf0fe71882e
aa5bcfbc59ba9c75a3202bd00da7ca3c77dbb83d
7778 F20110409_AACBYD liu_w_Page_091thm.jpg
57bd2ebbccf991ae2c50e5cb6f2dda65
0b0626740b7eca15b36563f7b40a671ac431744d
7092 F20110409_AACBXO liu_w_Page_075thm.jpg
7b8155553e63c50fb1572ce669f59ff6
a582ab28f37893e8b6b1539feabbf30e999919ce
2391 F20110409_AACATY liu_w_Page_088.txt
c91e3e2b8af268fb94563a3038155891
aeb41f8e03507862f7b9ace6f17bccbe99146936
53075 F20110409_AACBAH liu_w_Page_095.pro
04f0fe9874c5d38dfa5e695ec0c65145
cf97a6dd3d3798269dae72f8d54e9dac926f5fe1
2481 F20110409_AACAVB liu_w_Page_119.txt
1759864cce78f15fa13a6740d12b6391
e42079cb1650c911f19382e3f1a9447fa29039c4
2225 F20110409_AACAUM liu_w_Page_103.txt
3aa6301e6207360a16a5763a03e1e79a
3bd618e8038d73a3b8a2d69d085744b1eef57b17
7791 F20110409_AACBYE liu_w_Page_092thm.jpg
b44ea570e00a926b3da3994e18f10811
eff1a6c40d6dbf948728ed89fc033ef288ca979a
7847 F20110409_AACBXP liu_w_Page_076thm.jpg
b596ccf1e279999803e2a25d31595c00
d4c4c2eeab9b90a9ea4fee199c882a87748cf6d6
33006 F20110409_AACBAI liu_w_Page_096.pro
a2046e1cb79c9899c75986edfd4c538f
0b85945e54d21fbbf4368fb6a41c88f67f96777e
2400 F20110409_AACAVC liu_w_Page_120.txt
5e03b29e2b0fe22153d73f83b004853f
41faa1f5318d16eb0bbc323abd353824e8f0b03a
2347 F20110409_AACAUN liu_w_Page_104.txt
b73c3314d8f9d4d4eca227d2b95e7fe4
fccb2e270a58a252c85c3e1dcb81938406f93d0b
1942 F20110409_AACATZ liu_w_Page_089.txt
6a1e024741ec4484a41577e90aaf8b12
d21f542ff1051bdf6caef5a798f3fda7b005e72a
5949 F20110409_AACBYF liu_w_Page_093thm.jpg
7a09f197af90885d0aeb562cbe0f225a
2ab8d238f74009b1fe020fabd434de223ca68349
8640 F20110409_AACBXQ liu_w_Page_077thm.jpg
1f9f42c00426e6bbf18b6ad6ce6544d3
c16067c028d1665838babc4d214e8a05e61d5782
62737 F20110409_AACBAJ liu_w_Page_097.pro
9fbae23232ebb2308511d2f3410e167f
c867c6cf710cd8f1b18a3308dd442b82c213e96d
2377 F20110409_AACAVD liu_w_Page_121.txt
b3e0e887ac023d11fd2c8513c276410e
5c9df4dc6bfb6aa02493775ce3d7771dc37d4fad
F20110409_AACAUO liu_w_Page_105.txt
22d7c3bbdbc46acc5b7af02555d5324e
cd16b00d376b7e6849ca97fec873d4656fcfa66f
F20110409_AACBYG liu_w_Page_094thm.jpg
20c29929a093ed8ec7f91de14051fa8c
a26afea0f67528320b4effcaff706bd495407c5e
8118 F20110409_AACBXR liu_w_Page_078thm.jpg
9443e3eb9cf3fbcb3ef011f95fd09675
07ada47c2d173779ae6d926298170f1c50d26a48
62539 F20110409_AACBAK liu_w_Page_098.pro
09da2368b262730a429fc93e5f3cd6a1
f2ac4b6e175ba23a0e89aae2d51fe01ff32cc40a
2425 F20110409_AACAVE liu_w_Page_122.txt
3f3f98bdb58d02938ce557b8c3b84ca8
72cb948d6299ddef5a7f4db7c6492c1f5f99e596
2232 F20110409_AACAUP liu_w_Page_106.txt
4ed3ce1787cc1f085da8d39f4930cc35
854ced96d816b25bc2e1043adb38550b60b43e5a
8357 F20110409_AACBYH liu_w_Page_095thm.jpg
fd18647836daf748fc1bcab2bf64e3b9
077b019ca7a6e3503cc729f3ddb6da2e236ac1af
7306 F20110409_AACBXS liu_w_Page_079thm.jpg
bb4ada82de278bd30420e7747ac293c3
f7c5ec4f070a148e5bb1501956d9ced8ca2a7d10
46316 F20110409_AACBAL liu_w_Page_099.pro
8debfc292dcd234c87e59621b6f7444c
c5a094fc26d658b269551d9ecadf44658c81413c
2088 F20110409_AACAVF liu_w_Page_123.txt
501e5fe56df2b5c8d68a3eade4b1d036
290eb015d2e37419a7f6975e1d3847a36b3e3d81
770 F20110409_AACAUQ liu_w_Page_108.txt
312acaf70a7a0a2e9740f462d42ac738
20b466ef45efc70b1db8c90a370384579fd7be0d
6624 F20110409_AACBYI liu_w_Page_096thm.jpg
b685e8b86095389ac05ee08ac6e46866
27745e157fa6f9c47b44c2e56670f56d3790e650
9158 F20110409_AACBXT liu_w_Page_080thm.jpg
29582244af4cc68ca7d2ec220ba47801
0db4f10559d137f493adf8636f41b296c6c67f70
63661 F20110409_AACBAM liu_w_Page_100.pro
e957ea4a0216d463943b25f38d5e6a76
6088956cccc994440dd17c62bc50d7724d1cb24a
2324 F20110409_AACAVG liu_w_Page_124.txt
3bbbb10086377e7c525da30925c7cef0
03767ca7e67cec34533bac71a2bf5777f7385f5a
2213 F20110409_AACAUR liu_w_Page_109.txt
40028c6c08379871bb5772af44bd71df
1ee04e6752d800078294406a14aad9fae5c6626b
17498 F20110409_AACBBA liu_w_Page_117.pro
9412b043088aa4960e3769ae0b7c4714
28d3e9b2825f1b35492369103e13b84b2f7c0082
9247 F20110409_AACBYJ liu_w_Page_097thm.jpg
b75ecd4fabf00c8b9b45b8db8ced192e
49f939464c6711199461dc8f6d037e45a0d5d08e
4815 F20110409_AACBXU liu_w_Page_081thm.jpg
bfbba50f457e01577e94475b97b08830
f158ce69ec8d161b5d2418b896aa7eeccdffe72f
61160 F20110409_AACBAN liu_w_Page_101.pro
fd77bc6eee8000dd7b93036011a2c832
f5308012285cd6c985f90112d0d01899be078829
2468 F20110409_AACAVH liu_w_Page_125.txt
89bb51d0acd2baa27227986338e4d02d
f187d1c73c3396c4206bd0cb2bcc11c4117afa68
2345 F20110409_AACAUS liu_w_Page_110.txt
266938c48adfef1760158a5228a9a040
6888e4dab5c03328b1c3da224386ff8ae8064086
50433 F20110409_AACBBB liu_w_Page_118.pro
1d26a2a254840bd675a6275c36f3a429
0cce3238db7ddf28f6ae73dac10b1e78cae4bdae
7648 F20110409_AACBYK liu_w_Page_099thm.jpg
793aed4eca59aeb0d598ebdbd3545935
a08a67d28dae7ef004185ac1231ee285ebb0c0c1
8867 F20110409_AACBXV liu_w_Page_082thm.jpg
a532f3be2191f3471df7c733662389a2
5d2fea0c0a5448654898dc6d6eb2be24143e9770
56693 F20110409_AACBAO liu_w_Page_103.pro
212e7057ae0bb58486948abfd4881e26
51f20c53a60d7612e43da6498f70c0439fcf40bc
2406 F20110409_AACAVI liu_w_Page_126.txt
42c368f9a696161230697de367ee59b1
32a73024f80caca185b55f10e1d77bb1778e1018
2308 F20110409_AACAUT liu_w_Page_111.txt
ca5d312faabeb5f45e539cd96526bf74
0b342b3b5b2a0c900227c9c7cef1b7c66a0c9a64
63205 F20110409_AACBBC liu_w_Page_119.pro
26a1c203d464e847e944439c19912e42
a44ef721e0ee69700663ffd5f41608fbfadd2939
3802 F20110409_AACBXW liu_w_Page_083thm.jpg
6d8bab26d3a04cbb9864b360167d6f7e
84346aa1f1d628e79b180009f1497cbed414f480
59661 F20110409_AACBAP liu_w_Page_104.pro
e21f2f7420258c5a78b6fcd59bc6b060
730ef545ea8afd9afa7f0813dd440e4dd4a72ad6
2251 F20110409_AACAUU liu_w_Page_112.txt
c16a06f8ec84c129794f4d7d9c88d40f
a55f9704ca8a03284371840d1d74a4ca5ef5f05f
60335 F20110409_AACBBD liu_w_Page_120.pro
b1a6206f28bef34ff19eb6f8fe22ce98
5579598b5c2615f7aec04b7f683d4c9327b57be7
8816 F20110409_AACBZA liu_w_Page_116thm.jpg
d34e2e98d33fb55b5ceda77d47e34021
b93a0a8f992ac16054c188daabc90688d9353e68
9358 F20110409_AACBYL liu_w_Page_100thm.jpg
d9c7f107621b6725e35902a776333c2a
d9586ec4c5ba288a40a6a8eef38c4eac9a8185c2
8162 F20110409_AACBXX liu_w_Page_084thm.jpg
9add0e313ebe6b6488879e151f41ce99
06d08f1cde2aa8b172dd6e1bdb1ae04463fafdf3
52113 F20110409_AACBAQ liu_w_Page_106.pro
720a2f6b9480f3ce6a7350174eef18b8
e5b13476f1644bb495fb471bc464b0fb4dea5e5d
2616 F20110409_AACAVJ liu_w_Page_127.txt
5fd89acf9c9fd04df533ca615e14ff5d
a8877ca807c7c19430a7d56694295267754e8dc7
2294 F20110409_AACAUV liu_w_Page_113.txt
cb58677e17016bde051de89623d66d86
5da6237c8867668180c01a9b25933f40010daaa0
60446 F20110409_AACBBE liu_w_Page_121.pro
d9b775c6da6a6fa2c1fc68db63ffd987
a5f75bf287bb43bcdf6621e73b1034f96a377144
2815 F20110409_AACBZB liu_w_Page_117thm.jpg
231970b25002379c52ebd1a18cf20f35
142c676213a7aeb205fe7f53c4ba041c8c052a04
9035 F20110409_AACBYM liu_w_Page_101thm.jpg
cffcc6e1832a7a3a3cf6148dec3fa01b
2a2df18013b3a02502355f8418cf3ad4e04e1667
9134 F20110409_AACBXY liu_w_Page_086thm.jpg
7bc424378b69804341f6f9c02177a814
e35d2324e2f165362cae4a71e5b671231c133d43
59039 F20110409_AACBAR liu_w_Page_107.pro
6de4844b201747de0dcabbcddb0c3072
73b092198f20aa363ecebaa5e14cc2a603030665
2349 F20110409_AACAVK liu_w_Page_128.txt
a187e710212bfd817febc0f3183dba85
861ed74f84f2b3d0c6a15ae87b3482f0f2f7762b
1110 F20110409_AACAUW liu_w_Page_114.txt
db19d2561d2c26de3372303492eb5652
b4baf2b9702a44c2336ae77711c491cc5c857ea7
62034 F20110409_AACBBF liu_w_Page_122.pro
838c5750aa8943302f86f9119cdcd41d
0853d34d16ba7a60b2fe0c479fee8be341f08975
7862 F20110409_AACBZC liu_w_Page_118thm.jpg
14ab7e049749911713d374e4b435a20e
b9b6c67ce63e87185a2b0c3f955f6325ebea43dd
5450 F20110409_AACBYN liu_w_Page_102thm.jpg
c09e5bbceb5c3f42305f40a5f6eb8778
65842bf264ddeb99f571b20d24b62e15cb8ce9cb
9285 F20110409_AACBXZ liu_w_Page_087thm.jpg
7b3a2fc0d7a1c5fef04e857df2cde9dc
00b6f928f42d7248b35fd536b53116e1da9b7277
16465 F20110409_AACBAS liu_w_Page_108.pro
a1b0d5509290b0af2ae34bb1d6e1a278
f1b8e7ddcaed153c88b28e0955a79534fe710a9f
2130 F20110409_AACAVL liu_w_Page_129.txt
05988a5e06229bb12672e8e2b903d8f9
ba254002ce2ad65965d1a3deb4d893992131b655
2466 F20110409_AACAUX liu_w_Page_115.txt
a68415563a7814995b02710b42ecddc7
952ec0aae26f22176b9f32cba71dc7140a89b653
47944 F20110409_AACBBG liu_w_Page_123.pro
bda66b273ce5bae5093ae84c62903b48
34cf9aebf8b99919e885b05a0d757a188d4bddf4
2317 F20110409_AACAWA liu_w_Page_145.txt
5664e8ba0cce789dccaa2ed037825ab8
9ace8e395fe0dd98f843f4a48cf679c5c288e150
9103 F20110409_AACBZD liu_w_Page_120thm.jpg
da6fd3c1bcd56e38a7f2b2b19786763b
d5576087bd59e21478f087db794ae7a898e812d6
8900 F20110409_AACBYO liu_w_Page_103thm.jpg
6e75728cb3fdde180387dbc5bdc520c4
5a6c53381d7eb83c5f009daa0bb4e2ebed3b19ab
2545 F20110409_AACAVM liu_w_Page_130.txt
c11c2b557ac5c65fed298cc836dda440
d8ac1c95f824fd3865f2555df8e1a5422cbcd2d5
2396 F20110409_AACAUY liu_w_Page_116.txt
2dc9408de50fd7a9f77bfb9dba4ef130
e8363d753280225a1fad8d945cb23c9aa7d52983
55108 F20110409_AACBBH liu_w_Page_124.pro
e32b4e26bc457ed04a8945f6dfdf38f4
d94d60b7a47fdabaf83a3846f194db04b34eed6c
1534 F20110409_AACAWB liu_w_Page_146.txt
1125001d1d4de992f57c9fa96fadce51
284b8c1e52244eddafa4db5c536157bbb2ac78ab
8813 F20110409_AACBZE liu_w_Page_121thm.jpg
9fbf4dc51df1dff3a1f300e31ce31f85
8c9d96f0fddc9d8ce5ca39dae8f156c4219c207f
8881 F20110409_AACBYP liu_w_Page_104thm.jpg
91794229e3306ad7e02ddb36e6144930
836758d25860285589dd2d3b2262ccd33f55e3ad
39417 F20110409_AACBAT liu_w_Page_109.pro
2bb30ef701e2fa8b0b5b0105ea2f58d8
dd9bff7fb367a92c417fa05f3b2f6de2540fc7c7
2174 F20110409_AACAVN liu_w_Page_131.txt
f7f5a6a55141581a50181a8c21a9cb21
4074d9d563063911006a8ea7df6f522f958cfda2
736 F20110409_AACAUZ liu_w_Page_117.txt
5b9c4e86ebaa174808ee07e78c072089
bce8d72df8b1070d88d254359ce95f6017eaaebe
63083 F20110409_AACBBI liu_w_Page_125.pro
03def4b3e72c4cf61e131590b010a568
c14e7267951727c9539e47e08246d82859311cd3
412 F20110409_AACAWC liu_w_Page_147.txt
eaff16b62a73e3a1b5cc234a82c33da1
a0e67b72a2ec62ed67a46877a57d7c72c877c9e9
9275 F20110409_AACBZF liu_w_Page_122thm.jpg
1bfa6cadd2f798b63c27802f6e6ede69
5573f67cf0cff54db7dbe4a59f099737d18cfb8f
8652 F20110409_AACBYQ liu_w_Page_105thm.jpg
cd0f19e86c52e3df789c3f0a1cd45884
777a03951e3ab8078f60e6219cbae22bff7b6bc9
59539 F20110409_AACBAU liu_w_Page_110.pro
a0612aadd9093e3f0e8cd9da62c41acf
93db4bce4fcdb6ec145e42995f5182b7adc099e0
1112 F20110409_AACAVO liu_w_Page_132.txt
4b3106489018cf9d850778d535b65481
c85aa902e2796c75fa89adccf3f780419de04b5e
59158 F20110409_AACBBJ liu_w_Page_126.pro
e4950c29970447367cb39d03b772449a
3ed47aca6ab2048be1aec7f809c4d4fc4ff5ccc3
2432 F20110409_AACAWD liu_w_Page_148.txt
05bb5800c71767948577e0303c7808fe
e55fb36bf7c6d161c0fdac5b1d051c1a86a4a674
8094 F20110409_AACBZG liu_w_Page_123thm.jpg
170654e6bc8036ec823c556386cf55e5
799910dd743a24cd05adcfb92ddf9632dda6e75e
8091 F20110409_AACBYR liu_w_Page_106thm.jpg
cda0a3b0692e5750fd84a565910aeafc
0c8b09d07b85f6d5c2f0ce6a48a6d159cc4e9b60
46577 F20110409_AACBAV liu_w_Page_112.pro
939b582d1230165bca09f369adde54b9
6341fb1b99386e4116b341a0c55436244910a19e
2328 F20110409_AACAVP liu_w_Page_133.txt
183085321537126cd21f0279cbf0a792
1d287b6e396c0bb53382ec1e908421c6a77e7843
51336 F20110409_AACBBK liu_w_Page_127.pro
e0d0c2633f57ebefdd6c3cfdd263930a
710302ce72081e4cb903cfbf4003e4b9fc87d0de
F20110409_AACAWE liu_w_Page_149.txt
65acade08b833c3b111d049848a2b9bb
0d707c8abee220721dd3c0ca2897621b0e48bb93
8600 F20110409_AACBZH liu_w_Page_124thm.jpg
e3f24e86c76352289d2559cbc90536e6
089ef6bb6599897bc09b6fe534201841f2978a28
9055 F20110409_AACBYS liu_w_Page_107thm.jpg
32d8e3b9befe6122c1ea6ff3ca0e8b26
5dffcf68519f4b75f900ec69f1662c103540d5e8
2172 F20110409_AACAVQ liu_w_Page_134.txt
b571f00a20914da784a047eba73d35bd
d6eee67b3d5bb6b3da398fd74af00ece50d4d5b6
59444 F20110409_AACBBL liu_w_Page_128.pro
eb5ba76649b2f97404276f296092c809
01f8cb32bcc5b7461864aaf9f6135b57666359ea
2021 F20110409_AACAWF liu_w_Page_150.txt
d12c738e6de7e9a11d5c0562a49bd398
50d27928406a11d77bfac2c1c50f338fe69da414
46643 F20110409_AACBAW liu_w_Page_113.pro
997cc74152f85b3869473696515fa320
806f85d4eb3792e97a484230e6a78a1a0d8c1ef5
9000 F20110409_AACBZI liu_w_Page_125thm.jpg
f4836b807df19359e973cf195fef941b
3f9bcb727471e0f09d34251ed5b5c157f76acb14
6246 F20110409_AACBYT liu_w_Page_108thm.jpg
630fa206e4ed96fe03b1b0da788f5bd7
71d989ec1cde64ee89998f578894ddaf2a4f811b
1916 F20110409_AACAVR liu_w_Page_136.txt
9a45638099805e512eba89e2c70cb56c
fedd1dd7c71b99ed658f567d44bdcfc8676f4b27
32039 F20110409_AACBCA liu_w_Page_144.pro
4345c699d58dba027a61739fc52d5e28
b85ff0f596308952fbd082d04a344ea5051aa3e1
53653 F20110409_AACBBM liu_w_Page_129.pro
fa70f546cab20f871e02c4055540b7ff
5cdd2bafb13864c8bcf3b822bb21fa6c5a22a463
2219 F20110409_AACAWG liu_w_Page_151.txt
87680422193d98578d28f7d9b4e90718
ee9010ac0eca6d6332d1d761508866a39971c1a3
17612 F20110409_AACBAX liu_w_Page_114.pro
7bd5f813745ee22c1afffc09bf8f9ecf
3905fa4fc92ec1e66699dcc8ad178c11744674ef
8646 F20110409_AACBZJ liu_w_Page_126thm.jpg
74a5500dbfad37c687f71a17e2062cf6
f4c37b5af4e557379047a1547e7996a8dfda20c0
6670 F20110409_AACBYU liu_w_Page_109thm.jpg
5b0a42735d58c7899d2d50461696f9e3
6f0410c12dc8c3df3924dae5eb7237f81702f859
2397 F20110409_AACAVS liu_w_Page_137.txt
03dc39be11f63c83099f7388c5fec345
7bfe515571e300b5124d347de6b6457db0354b14
58891 F20110409_AACBCB liu_w_Page_145.pro
30c5eb4c37d05c12d885b1151463750a
48a6e855105c865d6ab249692629d39129a622f1
64962 F20110409_AACBBN liu_w_Page_130.pro
5487374c17a836fad932aeaccdd0ea05
3caf733c3ca5166005dec72a46c0e54011d8d403
446 F20110409_AACAWH liu_w_Page_152.txt
4183dfdd985fdbd5c127fef3a661a097
18e64bdc1fe5407abd847a116901ce40efea0d86
62918 F20110409_AACBAY liu_w_Page_115.pro
e0a2cb7c17a2f6c6a7f557f51c97a477
9a5391ea3345e7c4095c2ece78bf9811421db827
9542 F20110409_AACBZK liu_w_Page_127thm.jpg
a958bcc005a5f01c3dc6ff6bdfd3875e
feabe201a591f4f0c98df90d768f5eb981cfa869
8790 F20110409_AACBYV liu_w_Page_110thm.jpg
e802d4e26fb2e2079c74a6ed23104518
25bdf997204b0fe5c990633523c8f8be07d975b0
2300 F20110409_AACAVT liu_w_Page_138.txt
1c74ed317c8fb0350f93b5c263b60b53
b9386e6377cbc21e20d5312015fd147200a779b6
37635 F20110409_AACBCC liu_w_Page_146.pro
227209900fe30d3f6374d973d6040a66
c3f6f4d5038c8e622bc463857a8a07e783191b7f
54679 F20110409_AACBBO liu_w_Page_131.pro
aee368039e52fa448c68f73a3d7c0304
07cb8a6146a93d053463a02baa3eca80bd8d9cfe
2185 F20110409_AACAWI liu_w_Page_153.txt
1aae6c0f6b34d1bc861aec392af78b8b
dc943c2e7558476c160fc6a2b52d6b8dc5629997
60322 F20110409_AACBAZ liu_w_Page_116.pro
b4161ec39a4cd3065249a0d38f51ec38
f8d65a392109479068ef362d29d578daac3a5c93
8711 F20110409_AACBZL liu_w_Page_128thm.jpg
bb9ca0420f458a21599a681a8624abc8
b42eab172831bebcb3b7327eaed1f74a158e63d3
8933 F20110409_AACBYW liu_w_Page_111thm.jpg
3e9aa0dd60680293c4909612708f6318
0c4b96ec316593482a9d96d22b96547f89bacbf1
2212 F20110409_AACAVU liu_w_Page_139.txt
28abe80d19a0099b3cb0420aa9927d9d
940d74ee454e87fb94aec7d1730b61fa4d4e7646
10620 F20110409_AACBCD liu_w_Page_147.pro
b5b9d9b96fa714ebb04b735f725d0485
178db072b3ba39fbedac8f81a002a89ae9e7d369
21522 F20110409_AACBBP liu_w_Page_132.pro
651b399ff19323f73ba3fe43e642fda0
0c9020f154f6b401e01a13a1de49d4b62c5b22e3
F20110409_AACAWJ liu_w_Page_154.txt
8a84389873adf34fa17fd171312f7f24
de8e5f3780115618df909d952a29e8840f4efd01
7963 F20110409_AACBYX liu_w_Page_112thm.jpg
c53e3562648803bfe6d21c25b2b36f96
15eb84f6fecccf8720be80009c1329f339b2f986
2321 F20110409_AACAVV liu_w_Page_140.txt
73910f33a259202f78214a8b5484c3ac
70df044eb0e9aa1d300cdf204e1f0837884ef47d
61899 F20110409_AACBCE liu_w_Page_148.pro
638c105a79f880cd6f0a8dfece256195
75d54c54be1868b7b8bb3d3ce0fd6483026ef51c
53739 F20110409_AACBBQ liu_w_Page_134.pro
34ecfbea397cb9892a2f8d65242e3e8c
1fdfc6e262a2cf98cf951d5e733e4b2902f85c55
7964 F20110409_AACBZM liu_w_Page_129thm.jpg
90dfc7c34158611a1afb085cac9a6987
10b59a0c45ad1b347c797914f250dbe9581ce677
7821 F20110409_AACBYY liu_w_Page_113thm.jpg
77927fcf3f65cd2ca8c827bdce8378a4
803d7fb15d68e4fe7f4349dbdad378b27d274c9e
2175 F20110409_AACAVW liu_w_Page_141.txt
47ee7fd9b81d244777b406175fffa346
8c404f7d7ba1eac3200b72631ddaffcc5c36a4bc
50621 F20110409_AACBCF liu_w_Page_149.pro
9e72ad5666e88734583012d71b48ed24
daa95165057618722d09fa6b37a8dc1d50d40027
59060 F20110409_AACBBR liu_w_Page_135.pro
1bc042eea13a2d7438504023dfc4fa80
103d2e261c64f85f559899d622c1c8d5f6865ee6
2445 F20110409_AACAWK liu_w_Page_155.txt
66afb93988dd189a2a5c846a4b6d93d4
8b85d99264ceea75aa9fb9113799135a5b0f5fa7
F20110409_AACBZN liu_w_Page_130thm.jpg
fa7ea5bbc70ad6adcc0b50f0f99c8a4a
146945094adfa093f85a1ff136ec4ec8dcfed4cb
9249 F20110409_AACBYZ liu_w_Page_115thm.jpg
54fe944eaf7a3d69f69161cad5600b11
2df88a8811ffa52a046da2106d2a0eb45880a7a3
2007 F20110409_AACAVX liu_w_Page_142.txt
a2167bd8bc05edd8d943bf662e29ac8f
71bf500a259bd85a19949a55c1baf9cc1f73369a
48595 F20110409_AACBCG liu_w_Page_150.pro
c052691ff89265b58a121721a52976e3
2b19d2cbb5aaa298c0c48981b94f17170e74f6bc
28709 F20110409_AACAXA liu_w_Page_005.pro
6bf919ef64c76ebc9c0d5ddecce79457
7a7fe1b5f2f397779011c7293082b21abc790ccb
42907 F20110409_AACBBS liu_w_Page_136.pro
dba048a9315642912c9f734259bf7665
1abee7ef5f7263240363643fde4bc1a1cc27e9f7
2494 F20110409_AACAWL liu_w_Page_156.txt
bbd95de17ee26c093ba496b5a390a283
f37d9e768b7d4eb46fed03728b2448c5d7b21ec1
8434 F20110409_AACBZO liu_w_Page_131thm.jpg
d37354bd5aa2532e732fc9492546016e
af0180ab81eb4c370f1a09a6db36bf7a135f405d
991 F20110409_AACAVY liu_w_Page_143.txt
dc1a7b8a4c1d914be73c80958f8db70c
707a0a3659e425ff456140384e9b1927a77c1b06
53426 F20110409_AACBCH liu_w_Page_151.pro
cc49c5d7b157f8019134bae668e8344c
38637fe589e83227b871262933a5fd304682d82a
61618 F20110409_AACAXB liu_w_Page_006.pro
34bcfe0cc875451ac12c3ddc94b405df
92aa1407d269fa95e00e15ae191009428fbc7386
61210 F20110409_AACBBT liu_w_Page_137.pro
859221977fccaed26c957ead105260c1
b6a9674001e2c39630541b5c28ab947e98a9d4db
2584 F20110409_AACAWM liu_w_Page_157.txt
7236dd26795c7611bf0a3980a9b2d4f0
fce0b277a72f3d0e5b5a5c7ca5e6620e013a82c6
4412 F20110409_AACBZP liu_w_Page_132thm.jpg
9fd9f7fbd5bb9a68c2a45ab3672d0387
a401ed68dcad8f8ffc5c79b89ed3a260e46d265f
1793 F20110409_AACAVZ liu_w_Page_144.txt
8ae6c48a5e803fd9baf9a7b222f6c740
c4e5cdff66b8129a578df2bc4d845ece3c6895eb
9993 F20110409_AACBCI liu_w_Page_152.pro
99b51524b2486bbb9b234b754cd281e5
68e836df211e79ea2214752d0dfcc662856a4626
34041 F20110409_AACAXC liu_w_Page_007.pro
3c23537126500880eeead91c92373cc4
66954e51a96f194255393733bdbe2ad6a12936f3
2641 F20110409_AACAWN liu_w_Page_158.txt
b2a3cdef16b6eb6c825a4ce8f4753e2f
032c608d186eed6be1e342fba23850a28b395f53
9029 F20110409_AACBZQ liu_w_Page_133thm.jpg
e5bfc3613694654b105faab13d26ecb3
7751c4a5c08ecbde4a5dc8537c5cd68104c4f964
52887 F20110409_AACBCJ liu_w_Page_153.pro
4924730616730fa39569e47e4a7a9995
14a9c0bc239bfa80ef81acfe9610c7759607f61a
12350 F20110409_AACAXD liu_w_Page_008.pro
3a45db985ee1aa7387664fdeb65de1d7
63ac9a129a29382c0d37ce7c95815e0dc37b2d92
58604 F20110409_AACBBU liu_w_Page_138.pro
fab0213fa37ac39e9747853ea66ad514
94e0f36a72dc5d2c5d4ea51a92b50c3580590a3f
2543 F20110409_AACAWO liu_w_Page_159.txt
dc8b31e449e8973efb0053d6c49ed63f
8b725175d8fc75e35eef73ca130dd64d08877729
8407 F20110409_AACBZR liu_w_Page_134thm.jpg
de6f06f8e7c96d881c4a9b2162572024
9d19008feb951d760493ee18ca782e601ce77a6e
59319 F20110409_AACBCK liu_w_Page_154.pro
3a0a1a66037dcaa575eb1c03a8d6c56d
4077c0e5ace28d851ed06b2ee4b406a487a6dc5e
47936 F20110409_AACAXE liu_w_Page_009.pro
38f32478d5fcd6dcac5d313aaedc54d0
74336afe7db580392ce95c9b34660811e27dfb74
56056 F20110409_AACBBV liu_w_Page_139.pro
ee4578d8fc7ef226ea6a4935eead7ab8
d5003f03f4918c6370d5abb95ce45e6daee0cd03
2728 F20110409_AACAWP liu_w_Page_160.txt
5efc6e2cbaf06be0fa5e78dce34ea44e
fd200e8ac456062d31d1348c87bcdb63e3b165c9
9058 F20110409_AACBZS liu_w_Page_135thm.jpg
d89ec3b9f275c6669dedf302139273a5
8d10b2720ae5f2632e0e0b426c289c0a6a4105f9
60458 F20110409_AACBCL liu_w_Page_155.pro
17028886d9039ecf4c4da3fa54e8f656
7edb900f5b55f3e4d613932969cd9c1d4b933e1f
44969 F20110409_AACAXF liu_w_Page_010.pro
9995f758e99709b843e43e7610890d87
4d38857681465eb870a164740ede5a0c64db644a
58111 F20110409_AACBBW liu_w_Page_140.pro
deb59425b09d1c4bbf0a189bbc9ad14c
11ad544d09c3a73cafd8f2a4378bcde8fbcc8523
2793 F20110409_AACAWQ liu_w_Page_161.txt
c763bcf01c64d85d3c359b5c6eea3c8d
9b5a1ada417db9e4268def870c560a4acacae530
8309 F20110409_AACBZT liu_w_Page_136thm.jpg
dbef3d8f8d9353516dc9b34b5b48cdc8
1b4ee0d86fe9d90342b0789e2a384acfc1b21ffb
61636 F20110409_AACBCM liu_w_Page_156.pro
8f9772e52608d39326a3483d76604f0a
6b77fbb90d136b693e932a4a77b566173eac3f12
44201 F20110409_AACAXG liu_w_Page_011.pro
6d5410adb8d6d4efc9c48c1ba3826bb5
f22cb980dabd0729416b0c205a3458976b29a597
53418 F20110409_AACBBX liu_w_Page_141.pro
3e55b7cc05fc49f8fc39594fb77744cb
c2e63fe7d6edcf416346893709d90ee3e948f96c
1603 F20110409_AACAWR liu_w_Page_162.txt
ae5af0767f59f0c25709628ec7296c3c
0258d48a680b2b1b9d0f8070bcd818fa18b994d6
1872 F20110409_AACBDA liu_w_Page_003.QC.jpg
f70ff2e674f9c06ce1da6a05f1565ac9
69d77719e1eda5c1f04ba82354d835d233a849a5
9381 F20110409_AACBZU liu_w_Page_137thm.jpg
8a0ee3a5418d8a23492c8e1d837aa57a
d0d28750d65f534c39bb41f4e4425dafc8e56248
65617 F20110409_AACBCN liu_w_Page_158.pro
ae4b976954aee2577e1ecd4520dfbefd
92127d108190f0be65d9fcec99486d69e3d627d1
28256 F20110409_AACAXH liu_w_Page_012.pro
69e1ac51c6959ec738041ac5780435bd
fa7e199f3a80a668bf2881ee25658597abddb2cb
49268 F20110409_AACBBY liu_w_Page_142.pro
561840e31adea17770c4fd0d0fadb821
801085e5b92de6f63e96f55def5335d24b1f9da8
627 F20110409_AACAWS liu_w_Page_163.txt
5d5d46e992c41ebdb19c51149e802f7e
73954a3257d279a50c155870c83ba8e21c84d838
86639 F20110409_AACBDB liu_w_Page_004.jpg
6a15e3328a5da08d3dca23dd8df613eb
e12370878e5b33a4bfb62816e1f67cf14d6e333c
8765 F20110409_AACBZV liu_w_Page_138thm.jpg
1eb04d442136deb5bd1dc1c4bf6a00c1
3c390e7914cc0895ce65da2a931e26dea8150b6e
62900 F20110409_AACBCO liu_w_Page_159.pro
39e4ab8fdd1ca54e2d9d9bd7d0bd6332
f2fc8d004f2ce09e42b6a87f1f9e5569f8ce0124
53180 F20110409_AACAXI liu_w_Page_013.pro
e38f0381e98ed04a770c3062582ac262
df9a365442c36bac67865932b97e5975f19f2347
24109 F20110409_AACBBZ liu_w_Page_143.pro
b9f587a472a0ef4bbdbf67fabdc7f361
c776f5c8f7d1b0b45f93d0fc15555df34cbd5ef3
2271 F20110409_AACAWT liu_w_Page_164.txt
e4b5f430795942ed63dca7f4f0bf3077
cd8dc4d42a4d21514339763aacd8f11486053e0b
28209 F20110409_AACBDC liu_w_Page_004.QC.jpg
479294b0e7935026842457564a4d8f95
e8625be3c0bf63c3e2f58913254be39207455a29
8504 F20110409_AACBZW liu_w_Page_139thm.jpg
6839c3ab2148e285b701201032d2cf58
f17d7b7f6ecbcfe3143538d8a32d552852778158
67490 F20110409_AACBCP liu_w_Page_160.pro
be03bc8d9e2907fe28a726ed91a8bce8
efe387e20d1d4b46ad25e7892eac7741ca28f760
57014 F20110409_AACAXJ liu_w_Page_014.pro
eb1b3575a885178a7b45c098351c9202
f3fa806e898c46d4f86e5a63943f13e3e66f941d
520 F20110409_AACAWU liu_w_Page_165.txt
719f9a8ec0b514f354878d9794c24f0a
5c053327f1023e007e0d58be89d27364ecc62fea
107143 F20110409_AACBDD liu_w_Page_005.jpg
fb5c37d9a298b1e5b86132481b64960b
bff37c0b32542bbeed9b4f883f3ce55417ad3c28
F20110409_AACBZX liu_w_Page_140thm.jpg
3b109af63cde3c322a4d93b313b17f7c
548a4abfa66d5bc032f55e9192dc20e3797899af
68451 F20110409_AACBCQ liu_w_Page_161.pro
14dff930b14decc92651d3541f3efdfb
86465413021addf07ff893141c06cabfd5495d4d
61101 F20110409_AACAXK liu_w_Page_015.pro
81be07d97066ef7c50422d6882d91be6
c5f6248b659f9b7bbdd09078ab1be00799d4b445
1254 F20110409_AACAWV liu_w_Page_166.txt
6c22be45411d7576bd1fcce7efff2c0a
f0c07e3ace3b791aa635f4fbc1c021df1ef9c158
30512 F20110409_AACBDE liu_w_Page_005.QC.jpg
256decee89de2dd99112b83a52390929
d56423ee3ff909f80cfbbe5c0becbc6cbe3bcf81
9051 F20110409_AACBZY liu_w_Page_141thm.jpg
609df50260435667d7f5dee6a7225b3f
a18e7a77cef88b9f372119e29fe9d2e5d620d810
38881 F20110409_AACBCR liu_w_Page_162.pro
50e86e02baecccc87adbb59831ad198a
de2c88b60f66eecfee8c49b34d186101b0cd823b
8351 F20110409_AACAWW liu_w_Page_001.pro
0fd75935ea2324188a52b554aae1553e
9ed8a4c13ccd921d0247d49c0ce9940e075e4e08
136269 F20110409_AACBDF liu_w_Page_006.jpg
6d64cec1c4ff70f073a7c10c25450841
068b82ecd8b705f5f8de8472d49c5879cce22ef3
7935 F20110409_AACBZZ liu_w_Page_142thm.jpg
6a6ac54fba55c3de9cd1a39ffe1d0dfd
8c2f200b1de8f6842080c40ec646fa15f967f474
55004 F20110409_AACAYA liu_w_Page_032.pro
5e69e778ff6117f3a36ecec3e9402e2a
826d733e95f8c241538af642f0614f6eb043881b
14582 F20110409_AACBCS liu_w_Page_163.pro
58b60580026bb1aabb3a9eb7f6a92d1e
cd1f654a18d4c40e1ad51f970a4c104126723a27
62355 F20110409_AACAXL liu_w_Page_016.pro
a4a74a5255534388e0c550377680a929
a7255a683bc6a55ae5a7f4b56f612b5456f3c3fa
961 F20110409_AACAWX liu_w_Page_002.pro
5ec8088f9cfe6b64f9f739598f9e51bd
82c72190802e30ae92ef9d2ab03b508095ab99c0
38124 F20110409_AACBDG liu_w_Page_006.QC.jpg
a1b0be75fd87ed6234d18a50421384ae
a7987a764b5238a42759e5be32ab41b202b02353
38069 F20110409_AACAYB liu_w_Page_034.pro
8eb0dc3efc84e0793428514986163c35
f1f3006a099937c2bba4f8ba50d59faf6e2df107
40616 F20110409_AACBCT liu_w_Page_164.pro
057695494713611df4f2f3342ba32855
12b325e7b3822cb4f05fc47e5572e8cbda783278
61719 F20110409_AACAXM liu_w_Page_017.pro
1e562828b3e9c0055f19903312511c68
9c02c15373c169dc023097e357bb859161f8a281
1529 F20110409_AACAWY liu_w_Page_003.pro
011a51985b2d01544e0b1d4b05c81402
ac78c0f0164adc77e6b854e4a73863cf06c7c2ef
79170 F20110409_AACBDH liu_w_Page_007.jpg
29cbdf4994a4211593d39088a4546822
ebd7b8b806f236e6f7eefdb3a5fd2febb30bde2a
29908 F20110409_AACAYC liu_w_Page_035.pro
602da74fc76b9fed1d8b8d695adf2739
37cce2a80fa6280412ecc6cdbf5dd1c0929f336b
8758 F20110409_AACBCU liu_w_Page_165.pro
8b84c08808034e576ffc385dda092ff8
0c9e90ceffff77c14c38ce09da6cd403010fb02d
62060 F20110409_AACAXN liu_w_Page_018.pro
34e69ef94be760dae7cea7971c097fe1
b498721971233f40ffc316b74930cb61459560b7
43361 F20110409_AACAWZ liu_w_Page_004.pro
38e515c3724a84559c2ba75bf94eb1b1
a2d5e1381589ff8a49d106ec022bb62d0c4fde8a
23058 F20110409_AACBDI liu_w_Page_007.QC.jpg
b2f81422c9dafad0ebd48e2cf87a2eaa
9f6e9e632100041203c0042270cf95b74c48e2b6
61368 F20110409_AACAYD liu_w_Page_036.pro
dc36308adde5c86be31c09824a49c791
b8d3921f13c271f800b6e5665aa0e55bd1ca9bd9
59383 F20110409_AACAXO liu_w_Page_019.pro
cbf3f9b167f6978b2b54f5c1522948ca
c5ca78321464d6a834fef4b1908d8601a8211402
32427 F20110409_AACBDJ liu_w_Page_008.jpg
7563e40b43ecea60efab38fd70b965a3
4667fce7128763834d31e3a60465f1690e86fd90
50527 F20110409_AACAYE liu_w_Page_037.pro
bfbbd8a5e07254b50c7e18f64c1ea9ec
0fbd2d33fed45484da4a371653d573e028225173
31517 F20110409_AACBCV liu_w_Page_166.pro
d8edced5dbf285ab8c9d050eb4c77e37
23505af54c8a4148e0b3ae09a8c8c0d77d6c24ce
54945 F20110409_AACAXP liu_w_Page_020.pro
1a904ffa08646761863d18c86a9375d4
93a52d724b04a9035f0a582fc7a890d92ac15ec8
107537 F20110409_AACBDK liu_w_Page_009.jpg
f425a1355ea3f672ceb0ef0c4d01423b
3e3e9240e0a3469adb22258f1c902e8f1bb662ce
54930 F20110409_AACAYF liu_w_Page_038.pro
1dc627e7290cafe182ee09e4c3473c20
267a7fe2c924466251733a9c0aa8bbaf02cf73e3
25371 F20110409_AACBCW liu_w_Page_001.jpg
c1285026a23978a462210df841de0f21
67fbb84f28d8a9d3cfbca41be928bf7389f42caf
59975 F20110409_AACAXQ liu_w_Page_021.pro
682e6889be8d55ebdf3d86c9de2e68ad
d6b047725c0c9220735e7ca7cc2833feca6d84a1
31706 F20110409_AACBDL liu_w_Page_009.QC.jpg
4757278363e0e7661d93a2d6ed9117c2
e42900b501d5da7048725f2109bd66fbf89fad10
46677 F20110409_AACAYG liu_w_Page_039.pro
14b255d8243cc190cb62058770561417
e80b8f25fe427e51dcc5ef19c5b2806f87ccdbf1
6896 F20110409_AACBCX liu_w_Page_001.QC.jpg
7a1ebd0b93932e20ed628fa581f6e989
36ebadbcf9770ff77362ae12a317d15922213323
12575 F20110409_AACAXR liu_w_Page_022.pro
346eef59c9af7bcf7d9da122642f27a1
750f900bbb02e821203c9883ae18b7b2642c0719
118731 F20110409_AACBEA liu_w_Page_017.jpg
d09cf29f88d905ff29c2e498964a6f02
73e54e2c4b1ceadb14d253c8ebda57044e077af6
98272 F20110409_AACBDM liu_w_Page_010.jpg
867437241f9f78366d5aa74ec46fedff
19d80292d0e82d2a07e281fab25c9057c8ccc45c
43959 F20110409_AACAYH liu_w_Page_040.pro
647d9162605977260da01614c02131ee
b660db9fa4e2bbea747914ae9c755fffca38f07d
1264 F20110409_AACBCY liu_w_Page_002.QC.jpg
45ca883e8b53e6d27b5a49da721e379a
d6c5f14604cd4743b26c3b39e104f50b9d803621
54881 F20110409_AACAXS liu_w_Page_024.pro
3258d5fdab39300a0dc7c0d15a48398d
9b69a5bf1ca9ddad5add2dbd4162949bdd7b2158
37597 F20110409_AACBEB liu_w_Page_017.QC.jpg
5cdfec3e7646deae2eded424be0dcee7
48383048d7b15107f786d832307ebe857191e3b4
30383 F20110409_AACBDN liu_w_Page_010.QC.jpg
4a5c67fd1ed1b57f7c00ab5e1cb24877
e2611592e3b42374129b36f85f5ce3450b42e980
56918 F20110409_AACAYI liu_w_Page_041.pro
9feb796655d18e6fb6907a5b6af45897
4b40241c7e3cfb7a6da0673b89d0d570a855db1f
5121 F20110409_AACBCZ liu_w_Page_003.jpg
3ea836a4cd44e8a2ee11f429677da865
a70dbe20e2d53e805dde47195b8cae610e08873a
60563 F20110409_AACAXT liu_w_Page_025.pro
1b685659cf0987769d1b59845d86e0a9
47d1b83840ccacc9ebecfbeb031e239d483b231d
125060 F20110409_AACBEC liu_w_Page_018.jpg
e23852d76cdf7185bac64365715b1bd5
3ce5919134935e09ae824283d89d1e204331d856
92336 F20110409_AACBDO liu_w_Page_011.jpg
47546ff5d32cce6f48fdb2df8e742357
a08f1d74218a0d3cb8c08b08be143cb14c6cf3e5
45573 F20110409_AACAYJ liu_w_Page_042.pro
dd96f8ed09c47f6751b122d8051d1bf7
c061c21c35c09f05b6f5ad471932ddc42fa46bfd
54628 F20110409_AACAXU liu_w_Page_026.pro
460f3a8a2b837818f5e67445647cd132
b1e8c84809cd7ea881b823f8a9dab9f620d079a2
39654 F20110409_AACBED liu_w_Page_018.QC.jpg
5633e258d423e3a09996f21a5d4b1cbe
af07a3f6b80c87af81b88a3a9545a66fe98a67f4
28558 F20110409_AACBDP liu_w_Page_011.QC.jpg
878b4e401de7392a2f92af9f44d7333a
4cf705764eb00190bfbf07cc9e9dfa60f82b9d0e
58934 F20110409_AACAYK liu_w_Page_043.pro
708a3045a66d993b3fca5fc19323f520
b194b06d2fde0934c8ad6017570d074830bc137e
60326 F20110409_AACAXV liu_w_Page_027.pro
10a604de6c92d99eec61c3b5cdbe8ceb
c0bf0be14c91eeb9a30e022c65662a2724460cc6
113717 F20110409_AACBEE liu_w_Page_019.jpg
355769216924d8ed14fe6458cfe3096e
e6d72fde8b84b94af8ddb410f731196c5da64d5c
58949 F20110409_AACBDQ liu_w_Page_012.jpg
0a0a626c70ebf8ebfeaf9cba1f7f4753
927f53cb3be15608ff459ad8488fcde8de13aead
44717 F20110409_AACAYL liu_w_Page_044.pro
340ea775bb609138fc55ac2971e4b54e
dd264571f002d9ae152fbeed88c2349da919466d
60557 F20110409_AACAXW liu_w_Page_028.pro
472b8ecadba9fcbd70685159f7fd953f
4503bcd5947fbeaf9abf2b6438c639fbd5dccc76
36452 F20110409_AACBEF liu_w_Page_019.QC.jpg
698db0ca9f1adb17ac0c27845971b813
ca663ab414443b9376c92c4616ba74599855a879
18740 F20110409_AACBDR liu_w_Page_012.QC.jpg
ab4162a19403eb3e19125aba1b589846
938b8feb9a6bb42435e237fffa3239f5beba2d55
59147 F20110409_AACAXX liu_w_Page_029.pro
da9b0160b75c45523fef709b4c770fee
b230565bb8effc99981d6b5f65b48949b12c3344
111780 F20110409_AACBEG liu_w_Page_020.jpg
5eea1d6729f967592b1ad95a1d64ab57
9191bb71782dd6f0d03630be6a8bab27f721e88e
27511 F20110409_AACAZA liu_w_Page_060.pro
d5bb44b2cc945aca241c808cc55f5484
abffd10797af0149cae465c3d3f02a4b8ae100a5
104438 F20110409_AACBDS liu_w_Page_013.jpg
cd2addd14ed84a723ab02ea64a0f90c0
8692a3b59b4fb84df6051fd01b0557074bec12fe
41813 F20110409_AACAYM liu_w_Page_045.pro
6c1e95866f251a5103c372808119bad0
1c1abdfbfd767dec240b8e256dbc770eaa14056c
47696 F20110409_AACAXY liu_w_Page_030.pro
bd68429b638cb9384e8e9151272c666b
61fa1ed042dc5a3aa632cdde85246ca5c625ffa4
34917 F20110409_AACBEH liu_w_Page_020.QC.jpg
4903cb23e6ca3473fb86d107bf832055
ca9b6d8b569349884ffb914b9cd7d661f8188e75
50700 F20110409_AACAZB liu_w_Page_061.pro
f213d710eef57d7ab4fa9670486a7e9d
9ae6f9b4bd2c940d543bb97664519e9bcaa31037
33635 F20110409_AACBDT liu_w_Page_013.QC.jpg
5c4f3d2cf31c39a32abb156440627dde
96ca1953d5954f6c21429fa96b19891929cc2cfc
28843 F20110409_AACAYN liu_w_Page_046.pro
e4dc279c621bbc7bb0008f76a717b150
801c2d8f6df9569c4f2ab48b850c56ad1f377280
61022 F20110409_AACAXZ liu_w_Page_031.pro
6b9431469e6248d0cb806d3cda4c12f7
e6031a0f0cf481e1a803f2b4cb354a516edfc58c
122500 F20110409_AACBEI liu_w_Page_021.jpg
f02f6bd05e1f3aa33c7fccb60279744c
31e428fc8518a51d87cae0d4a50efc05ef476a58
58715 F20110409_AACAZC liu_w_Page_062.pro
e7ed96085b47f90777141eb3bc7b6290
270a4b76cdee6fafa170d01d1f447553155cf40d
110896 F20110409_AACBDU liu_w_Page_014.jpg
af67f74ed97eb001c09c6e01d895e518
b66d1ea2b22047420e28c10fad8cd317086a587c
54566 F20110409_AACAYO liu_w_Page_047.pro
e0208de2a966475f9defa1bc26019147
f9a98536a5162ffcd791bf993fa9489e0ea8bdbc
38382 F20110409_AACBEJ liu_w_Page_021.QC.jpg
9c77e834394ffd38e73cc8827bbe6324
9ac69b2851f3d49f29edb9a9a574e7465d3584f9
57594 F20110409_AACAZD liu_w_Page_063.pro
18eb2c425755a948e06cbe7103bdd62b
e0ef201f1b3c41f07c744eb3377ddecbc50f2644
35678 F20110409_AACBDV liu_w_Page_014.QC.jpg
fbba53d00776ddd898f50e824b35def2
f338ada92a9996e80383a727ed66d1c6410700be
56511 F20110409_AACAYP liu_w_Page_048.pro
1bd8fb586b536b51c69340e8962b5169
51bc24f369668fdec15fbd534e1b66a105b5affd
27301 F20110409_AACBEK liu_w_Page_022.jpg
f0e81ca819d5481b29e6ede5c5900f94
030875ac1373dd2526b4fc7fb8b124ba26f03c1a
58755 F20110409_AACAZE liu_w_Page_064.pro
aecf4d826c22aeae238be0e064c9a750
b2b35b076bf3a572a343570f5e58f03ce741658a
44167 F20110409_AACAYQ liu_w_Page_049.pro
ee06dbe54c2813029620da47189e467c
b068b440f9bd7f2af88446961c6b3d1517c2e024
9453 F20110409_AACBEL liu_w_Page_022.QC.jpg
b275ad57f7a0644f96349312bde888fb
16de63429b783fa477e361c5ecd2718aac9f64d4
63286 F20110409_AACAZF liu_w_Page_065.pro
ba013ea7046c368455cd39130f1d4be2
b550800fc2772742d64f4465168a4cd2e04e69d9
125081 F20110409_AACBDW liu_w_Page_015.jpg
0986532b9dbbe31b528823dc8db421c8
74d197bea37e8c84373e2d9fd98e3b117c3e889b
6823 F20110409_AACAYR liu_w_Page_050.pro
d5670480a8f7f35a44b29d68de84570b
a7772dfaadcd183e16be990a7d3cbb3b703ecf0d
97405 F20110409_AACBFA liu_w_Page_030.jpg
c7bf48ab57657cd6739c2bc813c6b660
3b5d8252d4874db0174b0414554dea47262705b4
100472 F20110409_AACBEM liu_w_Page_023.jpg
f566b40a84ae9db521001205dbab30b9
4b25fcad76252325a6e619e12182d4f6ab80ebdd
57376 F20110409_AACAZG liu_w_Page_066.pro
516cb0235ae29d02cdf9614bff4c4c41
6d9c4ff36f8b56f0708431ff6fba4f772c89ee5f
39664 F20110409_AACBDX liu_w_Page_015.QC.jpg
4520763ca2abe2c534f1151468cb6d8c
23d8e42633cb06b877b4c3529ab2fa8d91654f31
58724 F20110409_AACAYS liu_w_Page_051.pro
0ee880e266f83f1db1b9c7be13a8f323
ea5627514ce66acd8cb6bef684cbc850d101a680
120211 F20110409_AACBFB liu_w_Page_031.jpg
e516430b1e8dc05a5d707e8f89fe4def
dd4a9aa12d4e46cbd8567822c660283b0fa1d697
31822 F20110409_AACBEN liu_w_Page_023.QC.jpg
77e7c6e3965d54bcabe065c8d3a4fe0e
ad6074845e7b1d9127a0daab7e71354d9a845179
36123 F20110409_AACAZH liu_w_Page_067.pro
17da4a6fb95a7ebff4c001212e68fc07
0dd3704b70412806e6ff38946fbb426175f9884d
124008 F20110409_AACBDY liu_w_Page_016.jpg
894a3e4c3f4c402d8b61bbc179d20e5d
1eef0848b37775e70d90f9db7951b2fe89861612
54019 F20110409_AACAYT liu_w_Page_052.pro
825a527b1ee75c2bff5d5e77fc8d4f7e
e483223f8e2ee7283ff076e10aa00ad02fbe7e54
38339 F20110409_AACBFC liu_w_Page_031.QC.jpg
274cba06b69bb73000419bf9d8223bd5
87409f75f0897538205b15a3d77fe044cf4065aa
114363 F20110409_AACBEO liu_w_Page_024.jpg
c882882c4bf3d0ced7c39a466a934ae9
e1a13b404908822b44947c899914cbb0a348d3ba
55614 F20110409_AACAZI liu_w_Page_068.pro
0b3ab4dce420b901f94e59a5bd0f34df
69c2e88765660fe68cca4841cfa83a69b8f4c8b7
39483 F20110409_AACBDZ liu_w_Page_016.QC.jpg
9d075d129c3d34467c34262864357434
76c45d1e72d7d31174c621e73e64e810c4426024
56050 F20110409_AACAYU liu_w_Page_053.pro
972c34745f58eec0723cbabb4acf5017
77667034e1a121a480e34217d022282219436b1a
114978 F20110409_AACBFD liu_w_Page_032.jpg
42cf84b45a9c461eee4ed6f1c7ef263b
2760c7d344113e9a350683079fba8f6af639d064
36244 F20110409_AACBEP liu_w_Page_024.QC.jpg
86fb8f4d841de1eed5d7ee3f425842e5
3c1729a0d26f6e3f7910ad2b8fe7fdc3e5e96138
63257 F20110409_AACAZJ liu_w_Page_069.pro
8fc97a23adfd7a4e5954066972f96530
928caa7efeee702f40b68a7dc86a2e23781d3277
59287 F20110409_AACAYV liu_w_Page_054.pro
123078d3618666bfc34e803fdb8bc0fa
55b7ae666ce6485751897e8e85147b5deb097f3a
35271 F20110409_AACBFE liu_w_Page_032.QC.jpg
4482ae989ec03868d4c70dd9ffc122ff
276aeb4202b8431ba038e4d3bd9e66bb1cb24d88
124218 F20110409_AACBEQ liu_w_Page_025.jpg
4cc0c04403c3c9b05c4c29f1ab903c1d
b046c0b647e7c5eb4ea4a495f23ff4a6154df093
47577 F20110409_AACAZK liu_w_Page_070.pro
80d2dbb4a7f1b9a810aececdafba57ed
6b56ef866b413e9bcfe4a22e364149c024edabce
39348 F20110409_AACBFF liu_w_Page_033.QC.jpg
72d08dc406e65168c41d4a3e8bf34800
a5f02448b5e41c6c8e3c6cf54c045f7d5a5601d5
39421 F20110409_AACBER liu_w_Page_025.QC.jpg
665193b9c0d5a77175e4e369a848ea62
115f512b0e49d93758a1e84ac04eddef9b3eedb0
29893 F20110409_AACAZL liu_w_Page_071.pro
17e30e059f708913da5c5a76c65e689a
50b9f780247fabd30682645e6652258877cdd65c
14806 F20110409_AACAYW liu_w_Page_055.pro
69a78d95f32dfc4c0160b96cbdf087cf
58ee4020a739e91396434570935aaf606bc03485
92700 F20110409_AACBFG liu_w_Page_034.jpg
83f2a2e7a3a1e3c965f13a8aeeac8b2c
3fa7839c0d56c5a9ff23214f1a3239ec6dc37477
111870 F20110409_AACBES liu_w_Page_026.jpg
f1bf2ce6a74c083aff3cabb2ce35fbf1
31dbac2bdf84919fbf6832a1380d7d00d7ec8530
46967 F20110409_AACAZM liu_w_Page_072.pro
c6e85dd66ccd6386609957709cfa9323
0187da7933d22c0424d64dda881b7ae7e8aef3db
57334 F20110409_AACAYX liu_w_Page_056.pro
e1af18bcc43d81411ff450e28d5ffe64
71c7163d77d7c947f27e99987d38a18f364e5023
30129 F20110409_AACBFH liu_w_Page_034.QC.jpg
787f7b1492b8d84aa168d23485387dad
e59bca8a6d431616bdc376831b577dfb19864fe4
35470 F20110409_AACBET liu_w_Page_026.QC.jpg
2744879c7c7f8bb098fb3588f99beffe
2ba30ea23189ecffb8b0271e5111b111bb9ead09
16388 F20110409_AACAYY liu_w_Page_057.pro
57484d6d2a4f107f78b07d687dce5cd3
c76dc74848e7e71030f8af92bec909db6d17d9d4
70042 F20110409_AACBFI liu_w_Page_035.jpg
9d1aa797ee17bca9b0f1f98c72261077
192f6557569d193a18de3c6f49f110acf5d75fdc
122385 F20110409_AACBEU liu_w_Page_027.jpg
ed59352cae006c62d4fea7364ff3cd33
d26b58ab02f5d67beddf492d90478a6eb4770f3d
48161 F20110409_AACAZN liu_w_Page_073.pro
2132571fc4358db39f3157037c5a3540
147d892382827af546f012cafffa8dc7c5a5a0f0
30269 F20110409_AACAYZ liu_w_Page_058.pro
1fc40a7aaece070c92aa7ea97a1ab692
95a2140a053295bd3c42517d87c87ae331b0614f
23209 F20110409_AACBFJ liu_w_Page_035.QC.jpg
d864a71a4256025113ac659bed1646e4
fd32febe076e2bb4404a37a70bc128e8daf8f9fe
38354 F20110409_AACBEV liu_w_Page_027.QC.jpg
222ce41e9b91687b11804c2742070a95
0563b88fb1d544da8614fdc83a99f2cc8387e00c
40239 F20110409_AACAZO liu_w_Page_075.pro
3ee1516abc8135b24684c1f965d3dc4c
5264af0763bd403893714f14b0f60e0cd3c9447a
124705 F20110409_AACBFK liu_w_Page_036.jpg
efa3384c14efd8d7902d7df9bce4a3be
b04f3e629ac1979dd9db0644bda4ba3e0f807d8f
122131 F20110409_AACBEW liu_w_Page_028.jpg
816e0e5db107d8e604fc1d291dc59775
86bdfa67d37d391508b4c460d4107161cf9ad1fa
46131 F20110409_AACAZP liu_w_Page_076.pro
9b80249e24a0d7aba25aa7b383892a9b
3cd3bf3ccb68e705c3756e5ed60b6b1a04ee93fb
39852 F20110409_AACBFL liu_w_Page_036.QC.jpg
1379461433b15e5fbc3455c57e99e378
2eb242c51a9078e33dfef8d40f161d6fdb594dd1
59946 F20110409_AACAZQ liu_w_Page_077.pro
4060bd05a8a9de30bebf35a77587db0f
1fc6ed0bd4bd4c454ab8b02cb8586030845e1ac6
103768 F20110409_AACBFM liu_w_Page_037.jpg
f4a45b6b8335394cc078bc0c5b22bddd
90775bf176eb43c177ec7c63175c4c843cc5e790
38870 F20110409_AACBEX liu_w_Page_028.QC.jpg
10eec7a2bfe122aea5db102de368d1d9
e6378c8ab5e57d642fc9dbe6d591a2768a0b1d22
50735 F20110409_AACAZR liu_w_Page_078.pro
5db4d4dc7aae308c9c96bd07002ea403
f817b42647c73f56064a640c27c77d5a6a4543e6
90340 F20110409_AACBGA liu_w_Page_044.jpg
c1f23af581b383fb8ba719407c2bd587
dca55a46925a17a88ac2390f6d79f6ae99fcb667
33066 F20110409_AACBFN liu_w_Page_037.QC.jpg
70e197ce667791f78c1e5c69f9614649
1fa80e024748fb9d7e56f71fc7e173c16fc98b09
117237 F20110409_AACBEY liu_w_Page_029.jpg
6f5cba0a8ecc4e76c45fd4ca14961378
c71dcbe219da74a4b5d29dcd36150d5220232103
44979 F20110409_AACAZS liu_w_Page_079.pro
0037833af3af97e9ed08626ca3577537
b5a6a427066890c3d3131b0d1582a6229bb025b5
28729 F20110409_AACBGB liu_w_Page_044.QC.jpg
2de131e8d799f817dd78d54efd15bb35
d9c1dc5803113a5a774f052da0089831e14487c3
109987 F20110409_AACBFO liu_w_Page_038.jpg
87bce6cd14113cc333eb617db5533a92
1df3889c932fd56766c4bece27ff278c1e6f8041
37091 F20110409_AACBEZ liu_w_Page_029.QC.jpg
716924611b729e701b75228b0810aa56
1826c675dec34bc7cdf98dc9c674a5bfce691837
58331 F20110409_AACAZT liu_w_Page_080.pro
5e1b1e83035bfbc05569c8b7ec1704d2
77f2d66a75d77327fef1798c8b3e60785edd6058
88662 F20110409_AACBGC liu_w_Page_045.jpg
1a85779b6d45f117989c728dbab30125
594eeb840b6d8a7161302cfae45037aaaa1a0c12
34442 F20110409_AACBFP liu_w_Page_038.QC.jpg
85b71d6f3172de5df7bc530894cc613b
d93d049aed03c1bd913ab21e3365680674006342
16479 F20110409_AACAZU liu_w_Page_081.pro
d33b24fd726b23eb4acbcef73cc8fd24
735ee1564c2fe67de93c12b7f37346e8c4941252
27918 F20110409_AACBGD liu_w_Page_045.QC.jpg
5ae17b0ad24dc11335a4ae12c8c75c11
32ff4ec774a0d479a47470d4c8a8290fc871aa05
98135 F20110409_AACBFQ liu_w_Page_039.jpg
fe57432099c5dfbad3f08b305925cb1a
bc2774cb0a0b855b5c11c4ff99e733982da796fe
57352 F20110409_AACAZV liu_w_Page_082.pro
7093b0c1c43b4c0f693302a7267c137c
c8f8519ce809292fc439b86ae2a133ec9a0e8bcc
63612 F20110409_AACBGE liu_w_Page_046.jpg
2385bae57949d77d4203c024973d60bc
efc608bcc817269cbe0b5e4b724158545269d6b2
31733 F20110409_AACBFR liu_w_Page_039.QC.jpg
5825773258a14c44c0087471a8f7988d
68fb8f88bba68aeeaa55a9dac8e6f48f988d5d85
24599 F20110409_AACAZW liu_w_Page_083.pro
c20bd60dbb539c7cf43e799f465a13ae
066d465ee26ba2b22c9b8fb73c3a689a4e084cb1
21687 F20110409_AACBGF liu_w_Page_046.QC.jpg
6aae6a1d8eb4eeb6022c2fd8fd781212
072a4b8f8769920c3e5eada53526d34ac327d425
101562 F20110409_AACBFS liu_w_Page_040.jpg
2d9df8cf019d44cd7f3a079ff525d3a4
ea7a3429aa00efc88bce4eb4a33b7d74ab9f453a
51407 F20110409_AACAZX liu_w_Page_084.pro
79a6a747822d5fed6ac040358f78072e
6204f4ca8068472d69e9bd605b9dc230562f2627
114202 F20110409_AACBGG liu_w_Page_047.jpg
d3c5216bd57ac10f793fb1ee051275c6
ec4d6094f26e8f5f4837f9e5546261e6530c194a
31714 F20110409_AACBFT liu_w_Page_040.QC.jpg
e33fc2707512f0678bf7f7553e0ac4eb
ecde140aee94f6ea8eb2c2ba240b1efe9c957c8c
58354 F20110409_AACAZY liu_w_Page_085.pro
e6561ae85cfdbcaebaf13e11f8db8c93
0f1d9a2b3a5d42117f8ea731805a0d923dbe5252
36120 F20110409_AACBGH liu_w_Page_047.QC.jpg
7bc2b7617631df8ccf6910746b919ad1
ad23988edaac7babbd240039bbd829bf3316b770
59964 F20110409_AACAZZ liu_w_Page_086.pro
1ae34604fdec1bdd9b11130ce40a1417
1f0758120ebd48cb01e10c9591af42b690e82f61
113144 F20110409_AACBGI liu_w_Page_048.jpg
10b83154add554738e52884974130f7a
a8e610072f7ebcb006416cfcde295987967c7d71
118392 F20110409_AACBFU liu_w_Page_041.jpg
4f02b143169ec0af414a5e23e7f41500
48b7c28bfaa9342e9bc3d92008566626f26bcf57
92435 F20110409_AACBGJ liu_w_Page_049.jpg
16e91fd81f16bdb65e3e1262a26745c8
7aad9fb75a12ee4d7a50b33bcef58175c74a75ca
37420 F20110409_AACBFV liu_w_Page_041.QC.jpg
3acfc501c9f742ea1e10678c2c49055c
41820904344c5f6c7e8c519afb8b6e7192885f6b
29705 F20110409_AACBGK liu_w_Page_049.QC.jpg
3b6e5e95574b291aae83874921d4489a
ea1a06766828cafa9ae7884604146e034f7279a5
100516 F20110409_AACBFW liu_w_Page_042.jpg
b448550aa2524434c8e2242158bc383d
fdba7cdfd1d14ae5741eb629476d3da63fe29451
27451 F20110409_AACBGL liu_w_Page_050.jpg
96ccd31f127a69fd85f117e75ae2c88b
3f38a7f822ee21604f023b309f9c8b4823b5a5bc
32514 F20110409_AACBFX liu_w_Page_042.QC.jpg
d880a89475fc476ac5303a6e03a82ba4
35681cfed729318df46c7811e78679fb9bdef801
64466 F20110409_AACBHA liu_w_Page_058.jpg
bfe5f297d3ba28afde88e6d7cac05ee4
20d9abe58582c742beee78d39955d1a15a214093
119462 F20110409_AACBGM liu_w_Page_051.jpg
2fbf61460acf96bf82cde78a743a30f9
5878d8a7c04544e3cec752910e2fe2e3b25a4d24
21012 F20110409_AACBHB liu_w_Page_058.QC.jpg
925166ea6f83f54355a080972bf37bdf
8db3708d3a28c755a3258a1a473df9212d557da8
38248 F20110409_AACBGN liu_w_Page_051.QC.jpg
ad01ccd20371fce65edbfccaf847d2b9
2b2fddb730409bb2da98f44b4cfe8f0b2374f517
121371 F20110409_AACBFY liu_w_Page_043.jpg
b6874dfab8b966d2431c41300e3df3a8
f66835546fbdb38e54f4a0d489900e277578e67f
109475 F20110409_AACBHC liu_w_Page_059.jpg
e97c072a607abd287b3439d0ff372d25
63054c430913da3e0db8b94f5bae1deea04e5166
107936 F20110409_AACBGO liu_w_Page_052.jpg
f3e067e34f089ebc1bd3226c38851980
ca7b7701acdc05b5cd1888047937a3fa93fff27c
38225 F20110409_AACBFZ liu_w_Page_043.QC.jpg
17d167d678cc65ce1146e1ba3056928f
7a9310ce3b24b3f1cda3447d6792edb2ea861263
34621 F20110409_AACBHD liu_w_Page_059.QC.jpg
ec5d440bd70ae654b3581d10c235a7d4
7387aec309bd1c2337579079753ee0b281c95058
34931 F20110409_AACBGP liu_w_Page_052.QC.jpg
ff8d4a260fbe9d73d9eda439436d33d7
7168ab655d71be9e35c338e3703c6a38c1518f6e
59709 F20110409_AACBHE liu_w_Page_060.jpg
3108191b5548ed275eed95f3d8264049
e59efedba6fcc8f21eecd4f5485d042c2c24f216
114149 F20110409_AACBGQ liu_w_Page_053.jpg
d2acd87c9db4e34471efc60dfdefd02b
929ae5006a604016380b04807653e352dfe8df1c
18892 F20110409_AACBHF liu_w_Page_060.QC.jpg
78eff32d6be64eddc7b6cd03c2411fa5
829c2d759eea612b13b05d8492dc16516198953d
36842 F20110409_AACBGR liu_w_Page_053.QC.jpg
75b2c287aaeb08114c4e3b5e87db0d89
73c73e1c0247da3f2ce1a8271bd00ceb04123d25
104193 F20110409_AACBHG liu_w_Page_061.jpg
e7540daf65596984b0443e6fef8eaebd
20cccdb9fd0362c03c9f3d8eae0defff61ddd26b
120915 F20110409_AACBGS liu_w_Page_054.jpg
ce6c7eb6a1127a2b30bd68233a12ea30
d888ab65f3628057832ed61cd469ce5fc6536e1d
32972 F20110409_AACBHH liu_w_Page_061.QC.jpg
fb1236a34831e0917b9e769cb063e2ab
0f4d446fd5da8df74f7fb64748f0c137e455da2b
38709 F20110409_AACBGT liu_w_Page_054.QC.jpg
ef1f68266a089f2565b90997ea8d6002
801979dfd289033446e84f15e6be7fa5b2bd1e59
37506 F20110409_AACBHI liu_w_Page_062.QC.jpg
89f97bc252e65c6e4b8eac7394a64d8a
7eb89d4c89d3eb74e996a10d24c1bb7ba381b9a7
34513 F20110409_AACBGU liu_w_Page_055.jpg
8ecc4ea745e3e5cb26db2cf2ef6d0c8d
ae3751387e7eb3022223a081cd8ca7422d2185bc
117243 F20110409_AACBHJ liu_w_Page_063.jpg
e605c6332b12d334be079757899a147b
102c150be9199428e58c369538d841012a0d4d1c
11612 F20110409_AACBGV liu_w_Page_055.QC.jpg
f69e53920d127f4343581f4d4b912a4e
4c57c9f7e9ea9ea3dabb7e0778ecb53bba576d16
37201 F20110409_AACBHK liu_w_Page_063.QC.jpg
1057bf4a6e7bc6227d0f084afbb9912d
dcb79fbe7c0ee9c96a7a53d3bf4fb9d1046600f4
119516 F20110409_AACBGW liu_w_Page_056.jpg
624c61cb3daa518db9abe66fe38aef2c
3bd00a5148368649745c8aece8ad7df606c2b5f1
118064 F20110409_AACBHL liu_w_Page_064.jpg
0eac703f8123b752d39503fbd860eff8
5f6322e2172c572a2f158e01acfe4548b3177df4
37803 F20110409_AACBGX liu_w_Page_056.QC.jpg
fac1ba752d9b7d8905a95039f32924d4
e9659144cffcdabc6bd1b5f4efe327a94503d1ca
37702 F20110409_AACBHM liu_w_Page_064.QC.jpg
177da8bcd2127ebfb76f53af04742114
349b5d0911b2cbca0310f86662694be701cc6962
36660 F20110409_AACBGY liu_w_Page_057.jpg
64eb38912f0ffc81d77ed30f8f3e4799
60e8e8ef796c89d218b087b39bb1762755d07ce2
21496 F20110409_AACBIA liu_w_Page_071.QC.jpg
838ec8b219c149b58c47773e0482bc8a
a1a91c858f6d462d4112577393cb91eb9c218734
125432 F20110409_AACBHN liu_w_Page_065.jpg
4f3bb1435689776054274cf9ea39d951
2123b2065a5bffa0bc55e595ab0758492e03779f
96224 F20110409_AACBIB liu_w_Page_072.jpg
258f27ddfb62ec99c909607255a79aa0
3866d7ae68757268ad74aeee2f106696f52a6fa3
39901 F20110409_AACBHO liu_w_Page_065.QC.jpg
1f6464877843dc19be0549754d6ff4d2
2630a11a57db8b70a254c3c9b2ff421619065337
12300 F20110409_AACBGZ liu_w_Page_057.QC.jpg
5b90b1b0dfa83b1b6bbedc2b1a86b99c
16bd1213fbe85d7ce920b5690450232d0c3ba1e0
30981 F20110409_AACBIC liu_w_Page_072.QC.jpg
3d70fe78be0bc485bc814f78fd8dfad2
fc00e11d7f23c17ae90ca212694070dce6705a13
116845 F20110409_AACBHP liu_w_Page_066.jpg
b220920e4bca44639ad0cd6cd9f1fe54
a3ecf8c89dfb499d73d17a3094c6cfb06de0afaf
99720 F20110409_AACBID liu_w_Page_073.jpg
cac3d0452331a4cf9788a946571366b3
3168903b67f3407481c873e4c15ddd9235a6f569
37000 F20110409_AACBHQ liu_w_Page_066.QC.jpg
e57ecc2fbe126c5a22e70effcc1ab290
5f112d387264e0ab97a28e649aeb01cd4426bb7a
29869 F20110409_AACBIE liu_w_Page_073.QC.jpg
6a76704963a2f8780a2dead0fb2cdf83
511184a82dbd9d29c604f486a0b503baf06a6458
81393 F20110409_AACBHR liu_w_Page_067.jpg
34b2d8a7c84c288b88d04ddb4556d23a
9b9148e64ee6c33b3f500428d6133702ad15c478
44279 F20110409_AACBIF liu_w_Page_074.jpg
5167f7947ee0c7dc9b008e7ce4cab515
1a868dfade73fe5a02c228d2b35bd60709407937
26212 F20110409_AACBHS liu_w_Page_067.QC.jpg
1e38b648f9a76ffdbd461eda8531197e
e9a0003b221dbb2cc32ae98136d5c2d97aedecb8
13225 F20110409_AACBIG liu_w_Page_074.QC.jpg
d635f025b0f6a2f54f3baed050d8632e
eee0ebd964d1272657d17510649f429ce8751264
112925 F20110409_AACBHT liu_w_Page_068.jpg
09e581f77791a9da65eb8bd44b4339a1
78edc6731aaa0f302042a2b6c2d487ca7e4cef78
83763 F20110409_AACBIH liu_w_Page_075.jpg
3d6e5717231dba7afa24a36393e69bee
dc15d13b1867eb2fcabb930322735b6dac38439d
35873 F20110409_AACBHU liu_w_Page_068.QC.jpg
bde16798dc6956d912e614b1c290b6d6
e994af7789b1bbfc441db302b8fcd4d499b54147
26506 F20110409_AACBII liu_w_Page_075.QC.jpg
acb4f47b9f00f2b39613bbab5cc1020f
b311375224c5530ce778b004ac0da221d075335c
123730 F20110409_AACBHV liu_w_Page_069.jpg
ccdecb8a8adaea49b637c67856a1393f
45e1b73579eb8c7ef6620fb4ed617bf455f60f55
98981 F20110409_AACBIJ liu_w_Page_076.jpg
56830a014b0412a831e1c8f6cb7b8995
dd9d1154d80a01cf0d7050f4a99bebb3a0ca009a
39432 F20110409_AACBHW liu_w_Page_069.QC.jpg
80be6f807a29b29434339677ad9837cc
368f0ca7f33b94e5ba44bed0683ca803e642e999
122644 F20110409_AACBIK liu_w_Page_077.jpg
55958bdbe833f04422c43c276a7fd40e
070b346f0a2b6ca761c82cf16ca923f01f30bf21
102645 F20110409_AACBHX liu_w_Page_070.jpg
fc6cb2a4a0df5f3be3e40455cdd494c9
fab56c53d2e36d7adec6e09cf626f030d34eee71
38282 F20110409_AACBIL liu_w_Page_077.QC.jpg
cd4d72a59e3bff89923a75c235f70edc
3ae9db38853d98a9e88b83fd91591a4d86d35923
33041 F20110409_AACBHY liu_w_Page_070.QC.jpg
7ad808a23cfaddd768d32fec3dbc344d
102cd94c9b48d8cd44a9dbae7998df369d2eadc8
128931 F20110409_AACBJA liu_w_Page_087.jpg
ee66753f1eacc2b54d4906addc557178
ff029cc8dc62abae695e640777dbb58139d41372
108052 F20110409_AACBIM liu_w_Page_078.jpg
16202f9cc7f2194d160b079ef0926d87
21d56608b5f0d7cc44c3a78f47fb1fa2f81e56bc
67195 F20110409_AACBHZ liu_w_Page_071.jpg
971b026b31b81c2588705bfe09cdc9ff
9cfcc2c540b1a5aba3a38bbc91e2778e0eb26d57
40823 F20110409_AACBJB liu_w_Page_087.QC.jpg
1ce57aedb395605e1b772ebc0ca7ac41
8418455a88d2bb4a8eaf95c25033e730562c956e
34346 F20110409_AACBIN liu_w_Page_078.QC.jpg
a4ea2ec84cd4c06e05ae84474787d937
f221f7848f90e607543ff2be78ed273a115b7fab
120006 F20110409_AACBJC liu_w_Page_088.jpg
540cab669bb2b7fc4b9074d1d0b8be91
409421acfb3a7de32838bb0efa53b2145d26cf7b
95702 F20110409_AACBIO liu_w_Page_079.jpg
5e57afeff42d3c9f61c1b6ab8b2da51f
79b74476d75f28066b20e03ae546641de9eb8e36
38556 F20110409_AACBJD liu_w_Page_088.QC.jpg
d407ae685569b13cc7f0053a5f3ccfa6
d4b0bb07efb6aa582da23e6bf72c60f7c7e0f07d
122472 F20110409_AACBIP liu_w_Page_080.jpg
1375f017bbe94c837a3d62227c12b6b5
4b2680def8ea9d9443a4055012d570bf55c3ab9e
96289 F20110409_AACBJE liu_w_Page_089.jpg
2c8513fecfb564e14fc67a516a2ed8de
689ca704a0f2bebdaadbe7dac36197ee541ffa6f
39350 F20110409_AACBIQ liu_w_Page_080.QC.jpg
c73b5b4f46d9754dff61e27bf452e8a6
8b887c04af31f63f2d6d46277e1b06bd05f2488e
31198 F20110409_AACBJF liu_w_Page_089.QC.jpg
7730594ff18cc1d2834fe648e993649a
8a4fd325d1629c842108ce40c3f46eab4e6a4f3e
16965 F20110409_AACBIR liu_w_Page_081.QC.jpg
1477fc0b3e7a6cb92b7e1246bdf10d9e
ab998927744ae0cc511efaed80373576a5c9144c
124064 F20110409_AACBJG liu_w_Page_090.jpg
a889479c3794b695823568e71af90a9c
b8b5a27b4f130119d9e9f4942de6682ffc76834f
119855 F20110409_AACBIS liu_w_Page_082.jpg
e6ba74787d0afe5dbc7a7afbad784c89
6c024f60f77453ca862c235f5e238fc74bca2d54
39174 F20110409_AACBJH liu_w_Page_090.QC.jpg
3a0f961e28110222f89e09221b2ece38
eb70dde7de66cf35eccd809bf6a38cbb7641df66
37644 F20110409_AACBIT liu_w_Page_082.QC.jpg
920fa7f5e7191c1098575eef57ad6a55
5f2ebc80222b7aa060f0aee3986fb479f74ff660
97932 F20110409_AACBJI liu_w_Page_091.jpg
3521fcf548ddb465528fb2df05f7d6be
017829358069873796c7bf8a21f4ad8ebf546c07
54694 F20110409_AACBIU liu_w_Page_083.jpg
0b3db6d75a42cc787c60057903a3c36a
f53bb2511220953c498191905ca9991bd72131ee
31526 F20110409_AACBJJ liu_w_Page_091.QC.jpg
a43ea611ce38748153fdc96707971d13
8aec96d25400a4cad610606dadf86f13f327253e
105650 F20110409_AACBIV liu_w_Page_084.jpg
9ce9ede8452e7f523e4c2e3781181119
a2459b95c09beb504ecadaa9c9a04a99c0eae205
92190 F20110409_AACBJK liu_w_Page_092.jpg
e7e1a8e52252334e2084fe98e37f10a6
11f60b8a19de7a96815ede396b4cf04dd186c57e
117649 F20110409_AACBIW liu_w_Page_085.jpg
51625282dac64cb49d4eb1fec696e1ab
5c9484085d7f7476820416536a27448f3f486dbb
28846 F20110409_AACBJL liu_w_Page_092.QC.jpg
d84c3a743fb7ee222df5accd34f94c89
fd9544670a6eba67d30092152916245b9d7dd596
36424 F20110409_AACBIX liu_w_Page_085.QC.jpg
ef00ef0270a19a8402402faa41f092c5
9dcac06888eda63e34bffc373952f17d4b42f717
40985 F20110409_AACBKA liu_w_Page_100.QC.jpg
781f97c7aef2edf549a3cad1050070b9
4c803cdaaa18cc8f78c3a12748852b6e38eca8c9
68229 F20110409_AACBJM liu_w_Page_093.jpg
fc45e2a9cdd2bac7a39bfd345e5ff9fc
03b5f101d0e102188ef5db94acf7e6e03760b1ed
124052 F20110409_AACBIY liu_w_Page_086.jpg
4db602107abd50c6315cd34fc0d5b5da
1dcee23c677a814c98067d5097328e2cb8b7967a
122464 F20110409_AACBKB liu_w_Page_101.jpg
e55b7acd87d3ee6900f3cf7055540764
3216b65c7a64d23efed0337453e1a872f74f0ad7
22289 F20110409_AACBJN liu_w_Page_093.QC.jpg
c8cfbeb8774c5679ee7636f9369e5661
245a49cf72e433e99204cb489d4d9f91704ea6a9
39298 F20110409_AACBIZ liu_w_Page_086.QC.jpg
d8405542c00637ea13b5908801b0ec28
88721d028eb16ef38a673ea1bef0e7263119b52e
39166 F20110409_AACBKC liu_w_Page_101.QC.jpg
14db10d4c721bd1688e13c69d9056d7b
482d21d5f0b0f0629d411b819a45e18f5348a3c4
108994 F20110409_AACBJO liu_w_Page_094.jpg
de8b53b3c848e8be98f493ce30d614f6
e3e1151d35cffa85c706c14c771f245b819f8a15
68825 F20110409_AACBKD liu_w_Page_102.jpg
31b8c208ec7060ed74e5cb15e423e56e
a67d4bac2c04ae7a2ee50033da72de1dec198155
34826 F20110409_AACBJP liu_w_Page_094.QC.jpg
b1502c84404a93930c6615a49c529b1f
bf860af57cb3331b4e7c6e4d6ec5278b1e7f4463
21347 F20110409_AACBKE liu_w_Page_102.QC.jpg
541e242a4e8cee684c9ba2b0082ec967
116e56f99df517a9dcacf8f68acdd0274999062e
35270 F20110409_AACBJQ liu_w_Page_095.QC.jpg
abaa7f26858a5c95c8d52461aaa51cc9
a542c435abb01cb53fb6a96f3c1afcf48e2f552e
115822 F20110409_AACBKF liu_w_Page_103.jpg
95209f3870f216376726b8c046147100
fad969e12fc52c883c597c832c7983512256d180
88608 F20110409_AACBJR liu_w_Page_096.jpg
f23be8e21d1524c19a918eddac989a6a
b92faec3fde082d6d089f83ce83ff3cf90cd9f6c
36898 F20110409_AACBKG liu_w_Page_103.QC.jpg
1799716b47ab7f4073fd487cf9c10e35
9cedd20ef39541fe8d9c1e304ec08301175c1222
26002 F20110409_AACBJS liu_w_Page_096.QC.jpg
55d6f7b6265eaedb668dc41a55f5b185
35201b6ec86198dc2aaefc1cd091663c83554efd
118911 F20110409_AACBKH liu_w_Page_104.jpg
2058a6bd010e9f2d6b43bcf01472adb5
ca627f58447afe6989bd7ee6685f128bfd18dc2e
125183 F20110409_AACBJT liu_w_Page_097.jpg
593fad9014acb4a4cbd411a9e77a17c5
3a1e9ceff0cd150779a48bc3c23ce99add93ab9e
37587 F20110409_AACBKI liu_w_Page_104.QC.jpg
7f4d8133fd8b954070ea906a49c48074
f6d96b9806931134a005ba9f744d768a85cf1411
39556 F20110409_AACBJU liu_w_Page_097.QC.jpg
7109b70d7be87f51842fbec4a6e10863
46b2ea844655a004890091c93dfc086df6c8a7fb
115050 F20110409_AACBKJ liu_w_Page_105.jpg
72e7f7d643feb087c27212017cb71c7a
68efc139e263ccb712bbce2e2b891de8a43cf9db
124788 F20110409_AACBJV liu_w_Page_098.jpg
f75ff920d6d49819bdfa225751289206
e995c2baf27427572737852cfc8fe5255d8fd7fe
36223 F20110409_AACBKK liu_w_Page_105.QC.jpg
0549fb4880ccbefddada3081129e6dd1
16b6ce7d9a50572998f7fea843953edc11b12743
39310 F20110409_AACBJW liu_w_Page_098.QC.jpg
ec213b51cf7ecdeef48a7b84cfb00534
3398c8302748174df8bac4b1eb2444d8eae7301f
109093 F20110409_AACBKL liu_w_Page_106.jpg
a390432555ee21dc3170af66b57aa459
8b2c24b8f20e97c9cbc1184990d75721fce39d0a
95556 F20110409_AACBJX liu_w_Page_099.jpg
157fbf107860681778f5bb3dd9f1530f
4617b2ec3a8ac2986da939bd94f066319e07489d
33252 F20110409_AACBKM liu_w_Page_106.QC.jpg
9fd0959c86ad6fd0e048a2d10f50d0c4
d539eb1b84bb00306ce95fafa310dacb4b6a056e
29185 F20110409_AACBJY liu_w_Page_099.QC.jpg
4890a415ed093923c4297fa26be80e0e
3462ad859cf45eaea2b9c658c24cddd9e21ca33d
31211 F20110409_AACBLA liu_w_Page_113.QC.jpg
19aa14cc031d44efe50e36a57a7a9c86
0772b3ad40ae32a1d8f4d5939e90d70237171692
119370 F20110409_AACBKN liu_w_Page_107.jpg
80a23012d9df0d7dc543dccb23b74581
50f964c47572627e7221cbab215835ae64b85226
129668 F20110409_AACBJZ liu_w_Page_100.jpg
5b5ef7313b6580dde4c90e52d227e992
a76a4a9e2bfa0c1d817ef14a4f647356e2644fcd
42412 F20110409_AACBLB liu_w_Page_114.jpg
26a771b11ccb70f0074cc12ff6e1fe8e
dc78e6592d036b4f14f3b337cf0187ba88d53819
38193 F20110409_AACBKO liu_w_Page_107.QC.jpg
25b8ccefdbca6958de0c8ed52abb9fae
2f8977209c388459d53bbaf391a7ceddbf8194f2
13590 F20110409_AACBLC liu_w_Page_114.QC.jpg
58353e25e620ca3af53af0dafb25e02e
e00ae68997d45705aeef49d758eda8175cd166a2
74639 F20110409_AACBKP liu_w_Page_108.jpg
ea38c8a0ca0715783198da9f030525cd
b8762b43c87577bcf1768fe1b0ecaf599743842d
1051968 F20110409_AACAIB liu_w_Page_028.jp2
76ec45434651ff2ec64e6a6fe0604dd9
fade380a71f10073bbac1eb5d6ce0942d69bb81e
125480 F20110409_AACBLD liu_w_Page_115.jpg
bff57ecf6c238733fe766b2bbf5f3e52
d308507d48364b1e51a7719f5a9b708234a46d14
25030 F20110409_AACBKQ liu_w_Page_108.QC.jpg
e968fd5189220116b262e6c566e3a9f5
b6b12d5eba0371d541a189d122de246816739d3c
1051977 F20110409_AACAIC liu_w_Page_043.jp2
cd540649b0e03b73847c0e3e3ed814bf
ff9639b5f8412cf9c8ab0b4d7585243a1e839ccd
39530 F20110409_AACBLE liu_w_Page_115.QC.jpg
3289396cde676455b2a72686d0f6ec36
7f3f1d6305998c180d4e26336d1b1e629af3f093
1051961 F20110409_AACAID liu_w_Page_024.jp2
cc0e8ffeac522f9e0724640f83ced9f5
77464769d1f420722895ace8c0e5d576a75cf6e3
119666 F20110409_AACBLF liu_w_Page_116.jpg
abecbb1589df04a02e0e0ad48d24a948
bce098953bf9bde1ff4a551ac97ee7722d0a06b3
82755 F20110409_AACBKR liu_w_Page_109.jpg
256a72b5a1a20a19b47efd40374c268a
8c54668620831664516e360eb005229c4069a645
8949 F20110409_AACAIE liu_w_Page_043thm.jpg
5e751274bc5d7f6201633d8655bb63af
11e5d8ada4e6b73240754b25d15ddb86b214b140
38176 F20110409_AACBLG liu_w_Page_116.QC.jpg
da0efb26970e7b803c0ddd6760e88b06
9df245374f2675f0896c051540a244668b0d3dfc
26114 F20110409_AACBKS liu_w_Page_109.QC.jpg
d395184db5c8e695da995d3677522548
fa87eb8cb996cd0a6a5277e1e007053bfa5527f8
25271604 F20110409_AACAIF liu_w_Page_064.tif
160501761dc122ed02292fff860043e4
af3583ac08deb50c663b56b9855b7669afa1c647
37922 F20110409_AACBLH liu_w_Page_117.jpg
3b99f9558a88f76c9a8a6106ec80ff38
dd1f46599f2cc8c050d4760dc95a970ebf1a6eed
118644 F20110409_AACBKT liu_w_Page_110.jpg
f16a92a7e622081c50592719bb61acc5
9526737a229858efd71a7a9c902d7a5ae20eb996
111865 F20110409_AACAIG liu_w_Page_095.jpg
7ea764ba8895e69a9502564ebd7ec453
5f4761d456e3a1b77404cda32f669726b9a5a218
12105 F20110409_AACBLI liu_w_Page_117.QC.jpg
8a75b118d7f0f552d2724a3724644478
eaf87a1b901ff8a173763c61426dd03b4036577c
37764 F20110409_AACBKU liu_w_Page_110.QC.jpg
9dfac47aaaacf172bac5ca9cf2b19d3d
bff8eb7274545fafdfb162d83b02167fae8b5b54
58694 F20110409_AACAIH liu_w_Page_111.pro
e62f5ba7d5782fab5988f9c18ba3eb1d
332e4648966753eadff70a52deaa3da08c05404b
103673 F20110409_AACBLJ liu_w_Page_118.jpg
0d4031c124f732d14eed25dbfaff334c
f31809ae6a0214c2bd1bdb8b2b58702bbac34c60
116424 F20110409_AACBKV liu_w_Page_111.jpg
f526e1ec6137db7e64266c57e579e9ff
c6ae641ef61773cedf4e4cae201ab699d8b4f970
F20110409_AACAII liu_w_Page_064.txt
53eecc302eddeaae8f8066cb73272359
7efb17d1b9539d35583fc5744c9b85057f62a526
32516 F20110409_AACBLK liu_w_Page_118.QC.jpg
fd0216ebd15d35f33ec0d96bbd3f609b
3d1a8dbe3919451f8dccdfd9dc4ded86d0e04464
37398 F20110409_AACBKW liu_w_Page_111.QC.jpg
6051b5d020d61a9f00641240139cb776
a52117afd72377790b9338bf8338349110033d71
1051983 F20110409_AACAIJ liu_w_Page_128.jp2
da75619eeda9d9ea656c1765fdfe9006
bfb35351213bb6660b72e8968b20991fe824c829
126289 F20110409_AACBLL liu_w_Page_119.jpg
25acd8279b0d3fc1a7600d84e0f913c1
8af2ae918f22b59ee766f7844079392656938582
97195 F20110409_AACBKX liu_w_Page_112.jpg
08df815f3306cb4d199e153db6c5a24a
ac6bfcf83a16f41da77c994a2792ac26439340be
37491 F20110409_AACBMA liu_w_Page_126.QC.jpg
c511227b57cff37eb7295ef4f73a76e3
87a799ccafbdf731015c9e9e903f25f3d5f49917
3845 F20110409_AACAIK liu_w_Page_114thm.jpg
d9ccf585e253295aac4ab983b166ee0d
058b25f79655f283e102d86f2be12d768124b6bc
39548 F20110409_AACBLM liu_w_Page_119.QC.jpg
436a2647ad2176b39a9fe72cb980925e
62ad831919bac3092427ec6dfdb266d3cc66e66a
32225 F20110409_AACBKY liu_w_Page_112.QC.jpg
187781f3b351389f7164e17519f82c78
66fc558ac2485ce8d67c72103ee73c7695d5b7b8
116095 F20110409_AACBMB liu_w_Page_127.jpg
475648e687c478a8d2117a4e556eb70a
ca2864eafc24c4619cd89b03b6c754e43a74d2d3
29967 F20110409_AACAIL liu_w_Page_079.QC.jpg
33c1384e53554aad33ef04d53953e828
8cc1c0ddee5a206165e554de0ae848260effb301
120771 F20110409_AACBLN liu_w_Page_120.jpg
4d05f3ac0fbbff55ca5a8259f371a101
121036934c33d69f04bd9e13fbf9cb90e3a20493
96755 F20110409_AACBKZ liu_w_Page_113.jpg
6396d313093151cf1c9ade0fae952c59
8b267f63c5c51a18e0d4bd26036dce640d44100b
8467 F20110409_AACAJA liu_w_Page_032thm.jpg
6cdf60a70611e7f0cb6ea05d41a2e343
780a07d8257e18dacac15d6e291c3c75f069c186
36446 F20110409_AACBMC liu_w_Page_127.QC.jpg
cf1ab2348a9ee492fe3359c7a4838509
a4e26e24cf5e891311671fadd045e79c5f4c750c
2296 F20110409_AACAIM liu_w_Page_085.txt
d87c0310b55cc6d442e76081143329e1
061fbde50a56196273f1730425db70da8feb5cf5
38680 F20110409_AACBLO liu_w_Page_120.QC.jpg
1d62d717b0dc334b186ca0fffd3e14b0
96c145223899d715f377c622963bbd2826ef950c
2243 F20110409_AACAJB liu_w_Page_047.txt
06590a0cbfc8116987e3829d2c2759f3
85504435be59565fbc3764b47fbdf267c5bc94c5
117560 F20110409_AACBMD liu_w_Page_128.jpg
1c810392f4979d4c573c05851d72ad17
63d485891c3e215213968f9ecf5dddf73055beef
8623 F20110409_AACAIN liu_w_Page_085thm.jpg
935a3567157bb05ecfd43071ef736ef8
b6f81a27eaf59ca39adc8a72e567e7a2662c633c
122168 F20110409_AACBLP liu_w_Page_121.jpg
5967a559e2a792593cee982ab6136e8f
889d4d34e919095ccac1a9108a7905a80689d6ce
9065 F20110409_AACAJC liu_w_Page_050.QC.jpg
d4209f2ddb8f36f1ff10aa0ad2fd27a6
fb17adcccba9ddbb4fa806925e704f8593b95966
37089 F20110409_AACBME liu_w_Page_128.QC.jpg
3d7af597359ae33e2f3812a02ce8843f
a56f17f624f781ed1a480bef5cfd65f8fa16c76d
2114 F20110409_AACAIO liu_w_Page_094.txt
f5d5e46b45c0a2cef808e5b5e674bf91
6c5405d4da974517b774b20ba98eee75815c9d50
38660 F20110409_AACBLQ liu_w_Page_121.QC.jpg
47fc5abdcfd065a58594183d252810d9
f732b2c4f6f28683de81b3257dbabdffec0d5a57
6814 F20110409_AACAJD liu_w_Page_011thm.jpg
42bffafcbb5cf69e5d654a9ec500ceb9
08e34076f05db8e49295912a4fec36b2b8473122
108715 F20110409_AACBMF liu_w_Page_129.jpg
4c486c9d18742d61d3ebcead9d99d4df
0ce25f1cd395f353cd932de65abfd28b90a03a8f
33229 F20110409_AACAIP liu_w_Page_084.QC.jpg
5b8b0b54f29a22a421932471ed5eb16f
efb155e070f0e5dff920b7fb8f1e22be6cf2f295
124893 F20110409_AACBLR liu_w_Page_122.jpg
7dcbd5eae78098dcee1ec648c610433d
e299934cd0cb8c3e717d62a7960069dd16842755
3447570 F20110409_AACAJE liu_w.pdf
2ed73a9b3f98f89c2968fb8e516f467d
d2f347fd4406c62efd672465283cae01b1f277e1
BROKEN_LINK
www.proxim.com/learn/library/datasheet/11abgcombocard.pdf
35108 F20110409_AACBMG liu_w_Page_129.QC.jpg
42c1686b1b203a13c8cbf5f135caf1e2
ff7b696f9e1d4cf293c8cdf65c207bc1df9b8356
1927 F20110409_AACAIQ liu_w_Page_011.txt
0b39c6dc37eeb4ae72b0345d643f7d46
72eeb0df6b8cc6d85a491af5bc56868904331908
39235 F20110409_AACBLS liu_w_Page_122.QC.jpg
eba64d2890a667838e7cfdbce15cf0e5
f9c8b57801cc9ba4e54b5b937fe48d49c8d0d3cc
126953 F20110409_AACBMH liu_w_Page_130.jpg
42e49cac8b64cecc23fbc1ce4ccbbdd8
cf9f43444b43f135597c14cdba1b0855aa190bc9
117708 F20110409_AACAJF liu_w_Page_062.jpg
b98e48410bc62202d97d5ff738d03a8b
e804e1cdb429ec0d64133c66c972f442c691a4b0
81762 F20110409_AACAIR liu_w_Page_164.jpg
02ce39d9213f69df26296f2048c646e1
c991c11251fc75b180f66d448a95de19fe0e7730
102913 F20110409_AACBLT liu_w_Page_123.jpg
ea254fc8ea9642ee01ea0c55663666a5
4689b69b72e9f01f0552bd89f77b2595fff659e3
40120 F20110409_AACBMI liu_w_Page_130.QC.jpg
69d105902d536df1234376d3c6316d84
ee882a2ee4ba3344a4fdb8281f01493675e4a37c
1134 F20110409_AACAJG liu_w_Page_012.txt
20665bbe5edb2dfc2128b5e0702411db
f30dada4057eea74a0507da24675f6d43bf64095
101 F20110409_AACAIS liu_w_Page_002.txt
0cbe95ab8f05fb68e4b241a881eef9df
f4f5fb3e43bd3fb1e03e8e5dcef3a7cab97e8cff
32735 F20110409_AACBLU liu_w_Page_123.QC.jpg
35921c9ef7187c34a6ba47991c64e18d
3bed1e780ebd2efa885d5ed9e84dfe9c769fb9b7
108828 F20110409_AACBMJ liu_w_Page_131.jpg
2357c38d20a719b213b147f757104efa
add35a919699c77423ee10ed3ee4196c999732e8
31003 F20110409_AACAJH liu_w_Page_102.pro
822ff68e7310b28dece9a6bd9674bc3f
dc10a1139f52dd634fb8e3a4623bd06bc24a5138
8912 F20110409_AACAIT liu_w_Page_098thm.jpg
86f72f2015431f6bf53dda20ba43dc8a
c4bc2b4984880e6febedc5a7c15181a112c0ab3b
109512 F20110409_AACBLV liu_w_Page_124.jpg
2381f4eef574cf37b9f0ff8d09887faa
c436d0024eccb3aaf164a99326a3af31d907c77b
33910 F20110409_AACBMK liu_w_Page_131.QC.jpg
83b1f3b22039bafec052dbdd0837efad
640521ca5919b85ad7f1855be4114c16c13449d0
8948 F20110409_AACAJI liu_w_Page_119thm.jpg
bb6541e13f6e09510775999ea13bd0b5
f7699385a4c93286b33e3c29a6e7de1c8d3c8e9b
62779 F20110409_AACAIU liu_w_Page_087.pro
529aa65aaf1dcddeecc2a6151b8a2b43
2d1937be031f9c567cfebda1661cb6dcf3e03c66
34660 F20110409_AACBLW liu_w_Page_124.QC.jpg
91c53ca791b9217ec1cb7da7bc9ab2ab
bae6faaa689ff46b5457288ac46b80bfa219015c
47124 F20110409_AACBML liu_w_Page_132.jpg
3e2151439618692a5930dc25dd058967
aa4d15bb901af6041e35ab3f560535b0ad66df4b
50820 F20110409_AACAJJ liu_w_Page_081.jpg
7dc687341fa7a674bc8876f4d5184d98
d3aac833bbb2821c942daf8113f3288489bc8c52
F20110409_AACAIV liu_w_Page_033.tif
49e341b628c778efbf00348b32b2ce2e
52c248a29c9f00d3e1a8cf3c0ceee2ab86b39c21
124804 F20110409_AACBLX liu_w_Page_125.jpg
1212c1d6ab9ce9e886be0569713343f3
0594cf99d038314875a9b2a3a60548032452b2b0
118428 F20110409_AACBNA liu_w_Page_140.jpg
c9a18af4f9aaafbaee053211081c383d
3bef4b5f2a87357d6428ea32d7982480111f785d
15550 F20110409_AACBMM liu_w_Page_132.QC.jpg
f772e9f5151fb87589fbd46a706e951d
87a40596d3c1d890196077a08810b81b711a079f
F20110409_AACAJK liu_w_Page_028.tif
4b7515cf6ec44043edc843cba201eb2c
0bada3cfb7e4770847b45a2308bae2fc25c99d41
1051984 F20110409_AACAIW liu_w_Page_084.jp2
252eb1f718b6594e7cd1fb6f42cbcffc
275c8ada570b109f57b3deec5c1c9b676326e7a2
39451 F20110409_AACBLY liu_w_Page_125.QC.jpg
8fe37e1045295d789c948f57cd876c1a
ec53abfd64f7fe74cfb416725d647eefe66532fe
37461 F20110409_AACBNB liu_w_Page_140.QC.jpg
96853be008bb555ad189171d737e6bf0
aef5c341992ca4b85b59461c7db2306299c98b3b
37496 F20110409_AACBMN liu_w_Page_133.QC.jpg
8cea283f7d2fbdc982e85288934b7cfa
f8d0f422acd1eead08268f414b2a6f936ab4823a
1051933 F20110409_AACAJL liu_w_Page_085.jp2
cd933ccbb82bf1a4f16ce46a87e1f45b
ddcae33b1c941808fde1b65d34ac448e6c89f35a
F20110409_AACAIX liu_w_Page_018.tif
f5ad3a30f6d0da8a2fe952b5bdf06d35
b127cdb51948946c39335092b6e689bdec76cbc4
120388 F20110409_AACBLZ liu_w_Page_126.jpg
cf33689d354218619c8cbf4ff83bdeec
ede6a3ad6157de9ca785b6544214adf103594350
119066 F20110409_AACBNC liu_w_Page_141.jpg
1cdffa3f1127f7943ff6e14d10ec51ab
b1e84a344b4466130ebd98bb00fad8a18fce6666
109979 F20110409_AACBMO liu_w_Page_134.jpg
1cdb0cbce7d65e4475c7ca889240db2b
f42029a541c883d4e586952e4541df56eddc4cf9
2302 F20110409_AACAJM liu_w_Page_063.txt
6379d1580775c89eb72c21f15fe832a2
64d8c198f0e5ed8bcdd97743a299465bfed80dd4
22970 F20110409_AACAIY liu_w_Page_162.QC.jpg
2e642cc1fb2096c36a155475cff7004c
0a1fab1056b140d8cb2a21eb224bb6673db43275
F20110409_AACAKA liu_w_Page_088.tif
9818744a55c349d89df030b61e263eec
3253dd0df8d41f38236676591d1e20646aaf248e
38092 F20110409_AACBND liu_w_Page_141.QC.jpg
5718e4b56fb4ea769f47217b879f1a94
f77e43649dea44fce6c4ca5a4d5c36b315e5a2fe
35076 F20110409_AACBMP liu_w_Page_134.QC.jpg
af0d4cebcee4e26d961e253b02fb1491
1e1a52fb46d99cb041828f7a3d09e5f3b24ed24b
49313 F20110409_AACAJN liu_w_Page_023.pro
32b6f5205889aeacf9d3d6ded16b90cd
9cf6fb3423d4f481f765c81c8ab741bba87b94b7
F20110409_AACAIZ liu_w_Page_091.tif
1d39fd0ab70246a56afadd67d361ad92
98e4619999ec5d4895f3dc419a08e77edf9d13a2
952 F20110409_AACAKB liu_w_Page_057.txt
cdfae46a2f5029d8f9076f847d55e61b
33af3f27b562c5d8e6b28f091d0be4535a86c523
109633 F20110409_AACBNE liu_w_Page_142.jpg
90ecbc1e68197e6903e63faf65d155b3
c75aaf502658dac95b874a108974adffa1ca8cf0
117342 F20110409_AACBMQ liu_w_Page_135.jpg
ddcb61a99d661d7a4343c368c166052f
5401acb03a16f2c1071930fea06aaa921dae4a83
119311 F20110409_AACAJO liu_w_Page_133.jpg
971dd2ab544b9bf3bd156cd6594652f0
9ee697e5d4155be3f9d2eae58900971a5f9f8e9d
F20110409_AACAKC liu_w_Page_113.tif
1e3a745ce1c5ed90e1669122b21d6962
e350888c68006fbd2127a0c2898a0c160fd84c54
33442 F20110409_AACBNF liu_w_Page_142.QC.jpg
b5097ed3a51173b5d4e122cb6e605258
e4eebb0db2985c09804e9ace4ab24266689ff105
37801 F20110409_AACBMR liu_w_Page_135.QC.jpg
17d74a75bf126b6e803e8dd276e1a449
557ab5e2d8107b65ef9f1849ae26a68afab4fca0
54069 F20110409_AACAJP liu_w_Page_105.pro
e2ee0efd5d32734547ec7881433b186e
de39f9757b4d9a79a2fbb27165c81e5fdb0603dd
127004 F20110409_AACAKD liu_w_Page_159.jpg
aecf9e96589270e86e481bf92061046c
8b7e4e996eac2dfe13f61229af873df0396c171f
61679 F20110409_AACBNG liu_w_Page_143.jpg
d89366d9b311d841291625c4dd1bfb38
2f24649f7b12850d523f595204d7d048e70afd6e
100645 F20110409_AACBMS liu_w_Page_136.jpg
f80a97e1580abf1133581c348dde9f8c
3f9300e2cc6e7c28b59b5596b18ae41fa36d431f
2482 F20110409_AACAJQ liu_w_Page_107.txt
7f6a81a81c058bc92f6608e24dd28da5
2c2e992db6aebdd6631e1d78d9c5e8349ce10483
59395 F20110409_AACAKE liu_w_Page_033.pro
d485bb95dd07c982160ffc8104e185c6
de75920aded466812983f07a03a6aef8f962d2d6
19723 F20110409_AACBNH liu_w_Page_143.QC.jpg
2a04e40d33416ec413210b83568e848b
74f80fdb089ad9a86ea8061e3622d07dc170f901
31947 F20110409_AACBMT liu_w_Page_136.QC.jpg
3f91993fbf284b4793d58d6e5ea594de
8dd2a4c9961949d9192a058b787d5e3b8a761d3f
122518 F20110409_AACAJR liu_w_Page_033.jpg
dc253443e7386aa5268ee2474bc2dc4d
0bdab7183ebf638f93c9d6b357800a9ca9425f2f
19606 F20110409_AACAKF liu_w_Page_074.pro
2e53861457f6f08ad6dbc9b37346a348
c71b3f77e4a4ebff3bc6ef65f908d7cdf5539d37
71476 F20110409_AACBNI liu_w_Page_144.jpg
ce072f3a42a29cc94c21ddee65513bef
00a36d08647211b05a209f19ce715287b7fe6521
124067 F20110409_AACBMU liu_w_Page_137.jpg
870362910eb8b9e235fad49be548b2d4
db881928f84ba292b576baca30a0e01826123179
125071 F20110409_AACAJS liu_w_Page_155.jpg
10cfd7af0c5426cd46b99c9b9800bb06
99d2322a410f9fbfd9501e1d7b125ab6af52cab0
30785 F20110409_AACAKG liu_w_Page_076.QC.jpg
dc7951abe2dcf1c9e707519f10b5b624
375780a2cec65757f5c3073fc81fe9db7551bd45
21427 F20110409_AACBNJ liu_w_Page_144.QC.jpg
dbfe432c8ef46ae9c516ddb7bb10b017
8c835451f40ce4e88c53038187124ef169516881
39773 F20110409_AACBMV liu_w_Page_137.QC.jpg
bd6cdc80d211c5e34e590f9d326814f0
1daf94cf3de9901fe3c421d36e3d60be2322ba1a
136022 F20110409_AACAJT liu_w_Page_160.jpg
ec0944a0a23a62de919409c3bb6b4b1f
38f53649d60b12480cd8c7ccd10dc1bcf93f8b80
17015 F20110409_AACAKH liu_w_Page_083.QC.jpg
a2ab6e118b8f7a80e33365748d56e357
5327ee3368dfb44ecc57b7066d77963a66e9631d
121558 F20110409_AACBNK liu_w_Page_145.jpg
069e065c4638b50295a26d84ed27d0db
ca8d1be485b6a461d3ebfff8ad95e7e00bbce756
116522 F20110409_AACBMW liu_w_Page_138.jpg
394960e2065f10cb515eec262f7c34ac
e285eba6fb33a68a838f519ee048e4bd31af6de8
4052 F20110409_AACAJU liu_w_Page_002.jpg
6f7c9147f542389e5bc301e32003d12f
04743030d4cff2877af68fce948114f100f1ffe4
2019 F20110409_AACAKI liu_w_Page_072.txt
a46ffb2cb7c8dbcfe54f7e11ed54a4e7
4964a86f3ad65f79e43c4ddcf0f0c2aa4d175e88
31878 F20110409_AACBOA liu_w_Page_153.QC.jpg
21d393c0591cf2d0428689ed7b920a3c
3ffb34164f6d4dc16d7f4f51263a39fd0d883b3d
38088 F20110409_AACBNL liu_w_Page_145.QC.jpg
699c1d208717f725453e85440f1a767d
0dbce76815f26cf96047c6a8786090da58519e88
37616 F20110409_AACBMX liu_w_Page_138.QC.jpg
62dd8ae28eedee3dc2c00fa2c6015045
720a13d0e1f5ab3dfd8bc0385fd2e5b189a47333
2322 F20110409_AACAJV liu_w_Page_135.txt
0d53b7a8e2ad14b7789650d5875032f6
1edf8a3b100b1a69512a1787397528d1c69ef7eb
1777 F20110409_AACAKJ liu_w_Page_004.txt
ab42bcac5debe961e5463c09a8660c20
9eda6fe9ec35a0ff75c4edfd18901cf3a57492ac
83889 F20110409_AACBNM liu_w_Page_146.jpg
456837cf18638743f3aaa1fb637fb2a0
26f40105c614f9496078dc6388aba32c0a1177b3
112128 F20110409_AACBMY liu_w_Page_139.jpg
8c4dd11a5336e53555014e032b66ba81
a652d774baa9fbbd08b16500d92188805f4f8f0c
1051952 F20110409_AACAJW liu_w_Page_016.jp2
d6eaaf56d785abf6c460e5ff9ee228b6
8d637d25fd90d62cadbb6dc88c517417016ff1fa
F20110409_AACAKK liu_w_Page_160.tif
be99e2f09b3538f4716bf0a37588ee39
d218e1b5337b1d9350ca145082e0c077810c808a
122919 F20110409_AACBOB liu_w_Page_154.jpg
2adf7e2caf332763007d5e5e56e6cde5
b345a6c34229b5f37fb8244eebc426fb4f74b00f
26301 F20110409_AACBNN liu_w_Page_146.QC.jpg
b89e8f7e01bcc9b25b6100f8eabf1b61
10c3679f1d48056e97e00fa0eb8348f4c593f7dd
34983 F20110409_AACBMZ liu_w_Page_139.QC.jpg
b841cc163e3242d75c8aefb1e0afce82
c67244fd2f0a71a5585aa03e68758698bf9af35e
53107 F20110409_AACAJX liu_w_Page_059.pro
518d9702ea0dbc891306653adfe1775d
954634a004d217561958d41d41bc3688594ee5cf
F20110409_AACAKL liu_w_Page_009.tif
0cc1a263d3af9277b28f8ad98f43541f
3558d0e4c5a9d19621a46b9ac4f55cf20f983d57
34869 F20110409_AACBOC liu_w_Page_154.QC.jpg
042161400d7c5c86d3ac29f4de512981
09faa0eb5cf7c197d97f9dc244286102e2e358a2
50794 F20110409_AACBNO liu_w_Page_147.jpg
db25ffaf18566c628edb31971c039fa9
f1fb746a57af1687e00b89f1766c014d8f956db1
63930 F20110409_AACAJY liu_w_Page_157.pro
fed929c1eb3780b9b5e8f41d95674c8e
deec0b3c11dcdd10cdb019e15da09ee130ed5ae0
F20110409_AACAKM liu_w_Page_143.tif
9ceb796784831b9247cfd40f6479a99c
d2e2dd9b3f860919b4730cc7f7f898c5942c23cd
35352 F20110409_AACBOD liu_w_Page_155.QC.jpg
7e607b71037b08b8b9fa40a9b4c03ede
e513302518b6b38ede668179086890e8ab7f9646
17943 F20110409_AACBNP liu_w_Page_147.QC.jpg
4fbbb1cce910b6c055226f41cc7a85bf
dc58554f24a91ae8886ea4294d57da406e6be0b4
59061 F20110409_AACAJZ liu_w_Page_133.pro
22f44d453922041b9c74ea7d63ec1ec5
79dcb6a0a44ab21107e387052baad90cefc04995
F20110409_AACALB liu_w_Page_001.tif
98ac9d9a560db85cc22899c02325875e
8d457ea509ceb53866817a003fcc92169cacf04b
7443 F20110409_AACAKN liu_w_Page_150thm.jpg
3ae09fad57a6ec70ff83eb75560e56aa
eb1b2cb9465501afa061a90932532149ba3e03a3
122743 F20110409_AACBOE liu_w_Page_156.jpg
61899d14873b106c48e6c12f15924d36
a90df46059dfd27ba6bd8fa870952f3abe5c5a1f
125629 F20110409_AACBNQ liu_w_Page_148.jpg
b423d70e6d584546bc0a700098cd8e2e
4a208d682846c7d9aa065b34abe0b428673702de
F20110409_AACALC liu_w_Page_002.tif
23903c3e45789739e0f9cf3b16312613
31fad2452901909ceb4a55010cc9ebba2ed46739
1051982 F20110409_AACAKO liu_w_Page_006.jp2
8ca48ca8679021570a1fa67ed9fae21c
81f5b8c9ee4da8950da2c1ad255836e940fb815d
34623 F20110409_AACBOF liu_w_Page_156.QC.jpg
6d9f889b79448948ada425c75b59259d
7750677f41645bb7b5074430774a0426869e424c
39593 F20110409_AACBNR liu_w_Page_148.QC.jpg
9f88b2dab907c28c80e66ec1745cd4d4
4a372d797e9a4b882edde9cf15a7b25bc32ce850
F20110409_AACALD liu_w_Page_003.tif
a060f2392f116089e5fb24aa38ef8569
79e26ade81153b2c2e1efcca6ae235ac80134ea7
1051986 F20110409_AACAKP liu_w_Page_124.jp2
fe7206bd8c23d7d6eb6b4cdce62fb4d5
c9f85e9ddbb97e1f6ef4c12f8d253519dcc695bf
131554 F20110409_AACBOG liu_w_Page_157.jpg
09b35cf3b1d0d4c3ec24fdc1098dde80
fdfd33d78bbcc08a87ad445f37a30b55f342cd90
100806 F20110409_AACBNS liu_w_Page_149.jpg
5ae37fd9db8889eed36cc7dfce57940e
c0d2a3d6c7c37f901b446b9676e1d4b026aadd0d
F20110409_AACALE liu_w_Page_004.tif
8cf25c8ff3be9d658ff4430b60257f89
87ec6e15e6c86793052b83a293056aeb2abcf2a6
1701 F20110409_AACAKQ liu_w_Page_046.txt
5ec0b861bf954d657fd303f2baa04fa8
b488b1ccb53c9fc5d571dc31fb82dd8b51ad2617
37295 F20110409_AACBOH liu_w_Page_157.QC.jpg
62247e02061ec9d2b021e5a4a140b42e
b5cc6b3757efc14367fde9c4fa8f869bc5faa565
32307 F20110409_AACBNT liu_w_Page_149.QC.jpg
f7562be1a6aff2d0dc7d9e52ef06e5f4
03ff9b2943ba543fc17da9d17689f7f89c03f792
F20110409_AACALF liu_w_Page_005.tif
f1206e76311727d3d03966cf86276839
a7e01504e68e24e94116215554900a92abfb2c11
31395 F20110409_AACAKR liu_w_Page_030.QC.jpg
4c92f0112e7fcfd78a2d470687be08d7
24ab8bce106b92cff154a15495bf5b80226a6baf
133354 F20110409_AACBOI liu_w_Page_158.jpg
8dab4fd500967b71366dcee66dae3805
fb813366754b1079cde14c346610607a78f63401
99369 F20110409_AACBNU liu_w_Page_150.jpg
1ce23516f48430f363935c1116bc37ac
ada8c374d2e6c4b644040fdf3f597598d32856f4
F20110409_AACALG liu_w_Page_006.tif
812daeab7fc4b1c06716a399595942c0
ecc930fd59ad2557077a646b357d343c826e667a
36870 F20110409_AACAKS liu_w_Page_048.QC.jpg
b672bd8c98ebd0dc26c3de3649c2f541
df32f6aa12f9d8aea9b7bf6564cb3f3ebd1da773
37549 F20110409_AACBOJ liu_w_Page_158.QC.jpg
aeeee2e219786ffff33a421250b859f3
f1fee2739e2446f1e6dd3fcbf4bfe5492ba16503
107802 F20110409_AACBNV liu_w_Page_151.jpg
5c1fbfa938c7bb9586b9dfe9a943decb
a37de10a389e96711f1e89ca908f4af0c5e8c0f7
F20110409_AACALH liu_w_Page_007.tif
a36e51a9e7e9809357fb4940f4483b79
33f5dada532de7de0fc76204d8302f8cf31988d4
31470 F20110409_AACAKT liu_w_Page_150.QC.jpg
fc5a43c802bb9ac2b1a3a2f054d144f1
1bb91d4e34dc43208ad0397c81a45451040f1697
35975 F20110409_AACBOK liu_w_Page_159.QC.jpg
c5d964a53f3c5119e624d0f5641a8927
315c789875fa91f19100768263d727f3949c1b00
33928 F20110409_AACBNW liu_w_Page_151.QC.jpg
431f2ca18304672bc7a700c8ac663f44
5c07fb39eb4947d858c23458c711f862e68e81c5
F20110409_AACALI liu_w_Page_008.tif
735c1412005807d94902d15aaf60eb93
ba0dd3f830bbe0726d533a88d56b69c77008757b
878569 F20110409_AACAKU liu_w_Page_164.jp2
d9bc9cb14a16fec01e79240ca61d1a5a
f56e5b5589a034342977f3cbd5b8b2f8e4997573
F20110409_AACBPA liu_w_Page_005.jp2
d3e518d662a7302bcd36db1f1cad1055
7175a9ded362e7144443488b71c9460ac3366463
38067 F20110409_AACBOL liu_w_Page_160.QC.jpg
99b9f2fb7a314bdac35c19e5eec4be86
38754ce22aac488ad257f00b1fc8502cea186355
20630 F20110409_AACBNX liu_w_Page_152.jpg
a5a1087e0d0638107d28efda9783e227
7e8c9eb7fc63be10201f89e6bb35d017c2ad9c5a
F20110409_AACALJ liu_w_Page_010.tif
4a6a80d7e290f1f1228be1cb7546ae76
082c8b2e708f6a7730dbd544761580bac149d794
F20110409_AACAKV liu_w_Page_115.tif
870bb11522501610aa3853cd2c2e8dfd
ef5a859486c8581c8131a8a8e640820961d19240
F20110409_AACBPB liu_w_Page_007.jp2
ab400fc02b8b3fe3a3c6a3234361ed44
d1ffa31ce2d493f650ae81cf3297ad17d6590777
134584 F20110409_AACBOM liu_w_Page_161.jpg
22334d215f28d0355fa800e6d6b8e0b1
096c605ac453be20c2d6826a80e9c3520b206b2e
7047 F20110409_AACBNY liu_w_Page_152.QC.jpg
9d7594f0a8f6f0c650bc1e69f2bb9fe1
d5adcce658dac0c7e6ef0f7a0630efded9fcc599
F20110409_AACALK liu_w_Page_011.tif
b3f701bc909761115c3084630e373586
6ccbaba9cafe78330f947b0ab086428764fcac1f
10114 F20110409_AACAKW liu_w_Page_008.QC.jpg
e01c3ca5d4735fe5756b5922cbdd3a70
adf88f116b29bfe6a19985a412508652c593c728
37602 F20110409_AACBON liu_w_Page_161.QC.jpg
0891b9b09448facc5d5cb7ab55935ad3
e68feaf507f45551ca9caca51f615bd7615f672b
112918 F20110409_AACBNZ liu_w_Page_153.jpg
81c817fdf0aee110dbee981b5abc1603
97807879c41f1131f00b76714a8e75c33f6914b1
F20110409_AACALL liu_w_Page_012.tif
ffcb37b49ee92d04992f1bf6bcf47102
48d523f2fa940b878b47a2669999ed3aef70a94e
7711 F20110409_AACAKX liu_w_Page_006thm.jpg
bc7e4000c85f8fb2ad7082e28d6c567b
c086e1e654c8fe9e04972ea2354dba75c8b3dc2f
447921 F20110409_AACBPC liu_w_Page_008.jp2
f2af7480a2077367a8308640ed885ead
595bc5362c86985505d29b426ccb5867dd25733a
79765 F20110409_AACBOO liu_w_Page_162.jpg
e159a3b23c26277877256282b0cf48a6
ce6bfdecb904cb718a5b01c601c1883d3a438fdf
F20110409_AACAMA liu_w_Page_029.tif
f7b0119c58e59b9686d7cdf452e69a9c
c49b1547787d2c6d6ba8daa4d2ef6f229e5a85a4
F20110409_AACALM liu_w_Page_013.tif
1822516de1b3babef44f2a97fa60a4f6
43775704cbf437493a997f35ecb39103bf909c51
264394 F20110409_AACAKY UFE0011611_00001.xml FULL
c80d1438f853efe30164e63983305275
87969c4f1a74ebdb1cc3d4de3e0c1baa9c17a2a6
1051980 F20110409_AACBPD liu_w_Page_009.jp2
e97c45158db9c4b1dffa19d6be0aa9da
a4e4b1d0feaf090b92329a5db3f08e425d5783aa
32194 F20110409_AACBOP liu_w_Page_163.jpg
d3d52a53065a086ef5486b3a1976f4f0
c9d09d742a045acea299b8d3233bb9f3a6af9184
F20110409_AACAMB liu_w_Page_030.tif
db88947c526e060dacf4da76f63e7c97
d42cbf7cbe6b942fdd288e2b5a83534db0cb6bec
F20110409_AACALN liu_w_Page_014.tif
e5b9a93852b6333463d384e985576e2a
c33937fb4c84f5a04f04507509366cf8dd5c4c43
1051979 F20110409_AACBPE liu_w_Page_010.jp2
ad052525d8b227c1fb6011cc180bf51b
47fd23b159879c16c0133907bfe3243eb0e91c0f
10895 F20110409_AACBOQ liu_w_Page_163.QC.jpg
bf4d4ca4cf656853358dfb1e14f7fe5f
cb6e03c266c87e17f1205e9388029f12ef787c21
F20110409_AACAMC liu_w_Page_031.tif
ff931330164762ecc160da4b44f3d587
a075f4641e419271d33d370f1774c1f3904c3cfc
F20110409_AACALO liu_w_Page_015.tif
95fefe8a448a862d9d68f6a8d817af78
38b9b3e7896192338aba4e96d812a4e4ae5287b3
1011858 F20110409_AACBPF liu_w_Page_011.jp2
49bd4d75dd42130f379684b5bf733cab
0c7ab67aacc2cbf8413f8060ed6caa56cc5642aa
23017 F20110409_AACBOR liu_w_Page_164.QC.jpg
7153957a034e922bcfe82e04afcd6bc0
4342d85f28c23f74103ff6940520ed556d8c9fc6
F20110409_AACALP liu_w_Page_016.tif
f4a4e57d17f0633fd7d4a16ba9b8c60b
6dd97fdd4a70b08e26eed31e4422f510dfbcd07f
F20110409_AACAMD liu_w_Page_032.tif
7f6a24abf15b774ff4a92c4283abb60a
0fd80daf17de2c8a63a93c5778c9b2062f7cee03
630696 F20110409_AACBPG liu_w_Page_012.jp2
f7a0a49307c7f7ff329265079a94ce52
381d71637cab2db798cb6c83ab0a7491c22f3e09
21449 F20110409_AACBOS liu_w_Page_165.jpg
0d4174b8dc615b12bbcbd51888d3f2f5
a51e3fdf6e631d3c380b98f661bf8a626cebd784
F20110409_AACALQ liu_w_Page_017.tif
a81d36f4348a84a032a949ad9a0c3061
6af17eba8dc8075d101847c8060412f71323d7a0
F20110409_AACAME liu_w_Page_034.tif
377240b7beb12fcb62f6731bc11ecb20
67d4069283aea995bffe9175f09b8744cb1489e3
F20110409_AACBPH liu_w_Page_013.jp2
a433d31480eabad2f9a81fd661f3bc51
9818358cdd4c4f3efaddc56e0452cff73fc15717
6940 F20110409_AACBOT liu_w_Page_165.QC.jpg
a6ae9550f94acb21ad112ef480bf8f2e
6e835b752bea66bbf3cba4e6be9eacb1c6aa49b2
F20110409_AACALR liu_w_Page_019.tif
52bf8810e841293fc61c8a038240e90f
51e954935ef7efe02302d9ed3a00d0429b16e841
F20110409_AACAMF liu_w_Page_035.tif
376829b034c2f635bd52409c88f26920
5e44d75ed6dce44bb1fa0b8f28ee6f70f2106eee
F20110409_AACBPI liu_w_Page_014.jp2
4822eb2d52ff9a674676b19b47a3ff4f
423d78e6711a12fb50c0076b349fb8bda45fe1ad
69750 F20110409_AACBOU liu_w_Page_166.jpg
712c8c72e0c2d91ba69ad6095f3fe1c1
b48bfe473d47e67ac4f6fb8952ae835fcd859c7b
F20110409_AACALS liu_w_Page_020.tif
b3dd5c81f62ef626bd4a35212785adef
ca0f7d89d6c06eb820543adb71d65e53005b41a0
F20110409_AACAMG liu_w_Page_036.tif
b0627920a49ee3db9a018d6abbfffe1d
67a1aa3579626fe494964469afa8924a7146e98e
F20110409_AACBPJ liu_w_Page_015.jp2
d52bf564c69c49064a85d0674c0108d6
9e0b4449acc21f2d7cc3a33b4c94280c458a8276
21351 F20110409_AACBOV liu_w_Page_166.QC.jpg
8ebd80e9a512618e7b99dc23f02e912b
16cc26d469fa92906b907bf39d09cbc8f07e57c7
F20110409_AACALT liu_w_Page_021.tif
bd0e9ea6fc1409163c9e1cb57b408bab
88f72b00cb55eecf27053aa0025d2449149c1603
F20110409_AACAMH liu_w_Page_037.tif
26e00a57b21cbeedebd6ce322c0d8354
8686cfb2cda2713403dda3239c1291cf828efdc2
1051967 F20110409_AACBPK liu_w_Page_017.jp2
04eb06670ce93aa30b67ce0d3b0e94b8
e32565fe80b45fe0638277367ef07a3392ce8685
252636 F20110409_AACBOW liu_w_Page_001.jp2
4683c71ce91fac9dcf49d09a4d6e93c4
a4a2553503be0de4559873d9e4f1e226a078c063
F20110409_AACALU liu_w_Page_022.tif
ef54a7f56b20a4647baadc59d7f659e5
4cd24f41abbe5428a30ad89ce1fec6dccca16eac
F20110409_AACAMI liu_w_Page_038.tif
1930a4788d0f9ece8f209fe31b3dd4ae
ddad74716fd26848d6f3898cc9d2190130db040a
766955 F20110409_AACBQA liu_w_Page_035.jp2
bfb3cd7aef39fedc576dec70af15d7ef
8864315a243376755d043e05c31622eeb89104d0
1051971 F20110409_AACBPL liu_w_Page_018.jp2
b67e1a56fb47853b37ce6079b1b19304
17fa7707a9a72375e895ea55459bdecd8a9dea76
26066 F20110409_AACBOX liu_w_Page_002.jp2
e1f41e254447bb657492539c05235856
3af2acd7087bad85756e884b7236d9aac8a780d0
F20110409_AACALV liu_w_Page_023.tif
4890ddd06b44c6681e4fa9c9125097a4
54bf11a2726e74c63b96d3758fd03296bf638e7b
F20110409_AACAMJ liu_w_Page_039.tif
49151f4037035411ac229b380946a8af
c3216562c32807eb69948a70ad5cdb4c902b43f7
1051944 F20110409_AACBQB liu_w_Page_036.jp2
1764389d1f5690eecd11fb1388bf3143
5e1eb020a049b264610caa9a216f52669d63ff29
1051974 F20110409_AACBPM liu_w_Page_019.jp2
4e3fa9f262957007564c041342cd9579
fc5d4d683f5983409b1581cf02adcd2ce2c59860
38230 F20110409_AACBOY liu_w_Page_003.jp2
192b84b4535fa8334daeca5cea77ab31
4806e181c61b8e350ac3a4f6489bea2a619ff0f9
F20110409_AACALW liu_w_Page_024.tif
4dbb81c9f96c2baec60343291313432d
e560144e7dc8c053a1e0ab224acaf9acc8ee780d
F20110409_AACAMK liu_w_Page_040.tif
75cfc8c1a2f8a359b9dbbb691d04938a
da542b492b8609c644cb0588141e436f85fab6ec
1051978 F20110409_AACBQC liu_w_Page_037.jp2
e98048f9a801c57eeded4911ec67f241
264cf386c1fb516cb05ca6b52bed266eedc50f0e
F20110409_AACBPN liu_w_Page_020.jp2
823cc8ffaa405abeb08d677be64ca0a8
ee1474b954a3fb236a832bb18539bd34ddf4a4a5
963965 F20110409_AACBOZ liu_w_Page_004.jp2
850acb970e5660c1504df7c730020771
72371e25984348b992c8f60ce6fdbc2a2bb5073d
F20110409_AACALX liu_w_Page_025.tif
eb63bd7093f0ad75036edf5b7a5b3fee
665d8ef76db4d3b81ac0c8cf7be81d99b37b439a
F20110409_AACANA liu_w_Page_056.tif
9a7afedba4eaa01d55b0d5489dc376bb
497d820a9f2a60a36d7dbd33290fc4a2b83c1fa3
F20110409_AACAML liu_w_Page_041.tif
a5dd01c604260ae1870192a3dba65b90
3117f484b7964a32b941889191b1cd215ef7dcdf
1051969 F20110409_AACBPO liu_w_Page_021.jp2
970b053363ff3ffa632923c07cdc40d9
f2695e2d242d87dcca7882ce81b02db387f91f51
F20110409_AACALY liu_w_Page_026.tif
d5c367a166873c7823ab8a25a85cf4b6
c351640e444776ea0ef8074b6887ff65d32cec72
F20110409_AACAMM liu_w_Page_042.tif
024b7658646a00c69b5fc62c3d8c62f3
df0875105c29ca219385e81fbd3d0d5d236dd464
1051965 F20110409_AACBQD liu_w_Page_038.jp2
3e2f2e153b355cd684ae7333539d03ba
4ac0f71c94a1ddc349c3e1c79118469402a2bd9e
282201 F20110409_AACBPP liu_w_Page_022.jp2
ae462386e260d8c295edefc47c66619b
63759b04382b4b6f8214630e01916894a84cc98d
F20110409_AACALZ liu_w_Page_027.tif
af22505df5046aaec4d40cf7e6b341dc
163ef4ed265f730f7ef15158cd1738babdb27388
F20110409_AACANB liu_w_Page_057.tif
68a85e7c92d45b90b5b28ccb46a7bbc9
8e446f1b51ee7d730fbeaa5e4d86f1ee0a92eda2
F20110409_AACAMN liu_w_Page_043.tif
e4cdfc666b6f0a6171b3a8a754c1e24a
7bca63707ff69cd95e6c0a742e4115c2dad024a2
1051959 F20110409_AACBQE liu_w_Page_039.jp2
f3f0a590a1da45081dc9e23873b07b6c
a54d98534252ba1c954d1892e12cb76631642950
1051935 F20110409_AACBPQ liu_w_Page_023.jp2
e43d903cf80cbbbfdd90de80ac6109b6
61b973e217e9668054f4b85163a94c6f2bfd8fd9
F20110409_AACANC liu_w_Page_058.tif
7d0fdc6b6cdd8407003190b907d1bb32
016a93279e6f8fb9c89963c568eb052893f74ba6
F20110409_AACAMO liu_w_Page_044.tif
cd12768e0512c779e503d763f4caac58
a6faabb74b2520e08411be71156d8ef26c1ad54c
1051953 F20110409_AACBQF liu_w_Page_040.jp2
004c2fd9523000b903f492c31ce5243a
221cf636ac737e820bc85d11c6b12dba0a7ae628
1051972 F20110409_AACBPR liu_w_Page_025.jp2
ee93b5b84910507ddacb7933d83b4506
f9e093c43f5f659af787c270192877314d4b4b69
F20110409_AACAND liu_w_Page_059.tif
5215849b9ed038bcb893b1ad962dca02
06805f220d0fd44d5064bbb4e9cfa2c11b77e074
F20110409_AACAMP liu_w_Page_045.tif
1d83f01491731cf957a58460db259460
4987f0a4bea95d7e07c55226a0ac761a0858093e
F20110409_AACBQG liu_w_Page_041.jp2
a7ab99d1859dc530da04d1a9c9786461
dcf0b102d903fd5dd4a38210287c729ab39ef171
F20110409_AACBPS liu_w_Page_026.jp2
76db266cb2d10de63597c8d1e258dac0
937d06fae168128aa8fd508fe4bf0502de200e89
F20110409_AACANE liu_w_Page_060.tif
a37ff6a1ccaf134a7812cb904f9f1be8
d381ba94bb98439843152edf67f7b1c9402656e2
F20110409_AACAMQ liu_w_Page_046.tif
6b281309e620572e2e11a5bffa31e856
8112c4ae6d28dccaadec2087c1a63f77d46fc616
F20110409_AACBQH liu_w_Page_042.jp2
8e30c4d6d9fe406bd1ce6ee16f93f637
3cd5ea86968762f636af1c92fc85d7f147165182
1051985 F20110409_AACBPT liu_w_Page_027.jp2
6c4038e3366a4544636733e653657afc
3ff25517218a819f4db68ee061c4a1cd99b8ee87
F20110409_AACANF liu_w_Page_061.tif
5ec04e740ceecb252491dd2c4910124f
0769d61c91981a9ec59d9c16d13630e80cab2b2f
F20110409_AACAMR liu_w_Page_047.tif
3d5ff4e970d0580f395aff1044c18eec
d3470d5317afde80c69e0d493def8871c89976ab
1002656 F20110409_AACBQI liu_w_Page_044.jp2
cdf3f3487a27e63c0c591c38705ba984
d6a865a665a4b61f07cb4d6a6831ecd37414a040
F20110409_AACBPU liu_w_Page_029.jp2
4939d3ad506cbada10d5dd461fe65f14
317fc0aba0a9e9bb264823ca552fbd366c6fa41d
F20110409_AACANG liu_w_Page_062.tif
90fc176b35dce39a9f34898ba12a8ebe
89d2b53bf367c6773e027747a41e5ac1ff697c29
F20110409_AACAMS liu_w_Page_048.tif
00e69eae36bec52962811131e6105bd2
fd8f8c4f8eff279ca4ccc8d8100b001c804f1ee7
945257 F20110409_AACBQJ liu_w_Page_045.jp2
1a8803eb9b268cbc5f82da1caabc6631
b0d23baf717b47b7b11dc001b03e3b78abd11e9a
1049962 F20110409_AACBPV liu_w_Page_030.jp2
02ea1e686cc60f1d2953c1c60e19effe
826adc3ee63eb1e4241f0d41895268e46e1aeed1
F20110409_AACANH liu_w_Page_063.tif
aeffd351ac1e0707d8857cbba4233ff1
10db41d5b2af8d6ab7599d2bbf12b18ddf2e205c
F20110409_AACAMT liu_w_Page_049.tif
15c00b8c37d9cffb31fc05cc559d4686
e9c2e274dc79623a05c94d37d862e283babaacc2
692617 F20110409_AACBQK liu_w_Page_046.jp2
a26e2faf24261610e00b525b39cfead7
6b5c7bd8f39ba4128c48e5dd294542bf627dcf0f
F20110409_AACBPW liu_w_Page_031.jp2
d0fc2c417b007e7b6096ebaa0f1b7ef5
8e0768d2a16023b58dd487c0b5ab21d4a7c65455
F20110409_AACANI liu_w_Page_065.tif
6d3435f00e2b6287f0e73751aad7afa4
d77d47750d2ba09a815eda06585e4870307a33d7
F20110409_AACAMU liu_w_Page_050.tif
5d8256c2f579dec7a07ae0268c42bc6d
a95552be98c3b07bb4350607d28da569569236e5
F20110409_AACBRA liu_w_Page_062.jp2
1ac8b77a4fe7be093e6faa7ecaa5e40c
63bfa00247d5bb2bac64267c640340b8e4a17c47
1051973 F20110409_AACBQL liu_w_Page_047.jp2
10ef3cd5e804c22478851db7a52ce9cc
fb22cd9f460e89c0f0b1315e8a13f88039964b33
F20110409_AACBPX liu_w_Page_032.jp2
3d459ed3c6e30e18de8eba5e219268a8
6a4fd84bfb3a1aec651404c64d80b24ea6f18bec
F20110409_AACANJ liu_w_Page_066.tif
57cd32b56232aa51a91e627a7380fa78
38efc5037f96c72f83437a7fc7c18438f9b9d3e0
F20110409_AACAMV liu_w_Page_051.tif
d609a8e46b9b786e7bc2e871e99d1146
c96a8a38a4eacf0c0e3d2c38e12a4a1c33943117
F20110409_AACBRB liu_w_Page_063.jp2
ca357a2ab8443c55ebc0cde02fbae661
460ce0b856a75c516c8fa23b88a7b739e3d769e9
1051966 F20110409_AACBQM liu_w_Page_048.jp2
2dab8e680bd0572cc5d82d3c796b31b4
3a8507fa07957ad919c09efb5819becd279bdffa
1051981 F20110409_AACBPY liu_w_Page_033.jp2
974817c979c13b00a7d0480ad619a9fa
9434ca5144a20f39189ccdd9e94ea637798eda61
F20110409_AACANK liu_w_Page_067.tif
a108d0c8169e5496d3eaa971b1d5e5cf
c4532ab742abdb140e5337e6c0fe6103643478e7
F20110409_AACAMW liu_w_Page_052.tif
bd99f69d29694ab3088a1cc2706b0cbb
e9a5fa5aaf48f0f38fde76122ea56566d1fe3659
1051975 F20110409_AACBRC liu_w_Page_064.jp2
4b5ffb0fde82e4e970e22904c330e98f
c5f6470a28806cfdbe74e0e2110997dc8a9732ab
964001 F20110409_AACBQN liu_w_Page_049.jp2
f9538520cbbf6281afd5c2e5fc4fa5b9
3cc920d96c70ef91d5cee229d9167ac2d92dd50f
1033474 F20110409_AACBPZ liu_w_Page_034.jp2
14295c84677f85e988303d7f9ab4d2be
5463dc95741e7982fdb6d5bb5636b58109d488b0
F20110409_AACAOA liu_w_Page_083.tif
5cadd07e2d1e863fedcd3987b0eaac0e
24e4ae7aa20800cb24721d48f19f733d79845951
F20110409_AACANL liu_w_Page_068.tif
d7f4db4b075d47ef7dbf8bbf3548d322
20f5e63936c8a8d8821e490f94e43d1d34078ec0
F20110409_AACAMX liu_w_Page_053.tif
301868ec58b928fa48bc4ce11678570e
120cf00a0231f1291ad6b12ef1d02426f498954b
1051949 F20110409_AACBRD liu_w_Page_065.jp2
bc74b13562d926c07a8ab70d5cd12688
b9b24112c4c6f701eff02a272c32e8057ea3d561
298677 F20110409_AACBQO liu_w_Page_050.jp2
181687bea2d5e083385f6e85548e5971
0275bdb49254a5e460596f6527d580df3c21da78
F20110409_AACAOB liu_w_Page_084.tif
f533fb4d69435d01f563c368b16309d8
eecd5d7d6554eca2d53feed8fe50631df3493520
F20110409_AACANM liu_w_Page_069.tif
ff82478807298d74529b59b0ba60b94b
4b69d011d9c1b022538c8a86da747bdf523fe77f
F20110409_AACAMY liu_w_Page_054.tif
6e517ba44f235c6dfa22b4091f4a864c
ce6c518f672eb713d290a42c2c5606207dddb245
1051976 F20110409_AACBQP liu_w_Page_051.jp2
b063d892f78d91380b7955fc31218645
abead04b919aa503e005d8878e2191a37709cbf2
F20110409_AACANN liu_w_Page_070.tif
5340019758ca2a9083e63800dc3bbe38
f409b7321e14ada57179fb82b6f3d2608d945a27
F20110409_AACAMZ liu_w_Page_055.tif
c4208e313e4bbe2242313d99a995cf0e
31b6b2227ed39f2ec9d282ede65089ca515717e2
F20110409_AACBRE liu_w_Page_066.jp2
3feb140bc090bf00a527500b35ba6739
9accc846115818f3482150ef9d1224c37f14321c
F20110409_AACBQQ liu_w_Page_052.jp2
88301c7c3c82268b9ec72212754f1315
9d6469abccb3a826af145064b73d13a3520f6011
F20110409_AACAOC liu_w_Page_085.tif
f9443ecf763a0e0c7eacbd35fca2b1ae
9647e80af3f67be7d2cfc7e243f26d4b4bbd7668
F20110409_AACANO liu_w_Page_071.tif
7cf4626afa0cfebcf01c1b2e08b7cca7
ce5aa5cf5f0591b5ff48e0b3c0572f3a519a396c
882534 F20110409_AACBRF liu_w_Page_067.jp2
85df7895cd867aa56f6997c7a92f3faf
525fa0b1587fe0018cce33f8cb0127e8c73ea168
F20110409_AACBQR liu_w_Page_053.jp2
8e58a876d9842034bd45f1a7e4308865
5c2e00aeddaa965434476104ea04b931f8b41bd2
F20110409_AACAOD liu_w_Page_086.tif
e25d8d7ac187de658d66a3756e7e80cc
2e4b310b59297669510abd5173ae91842e953464
F20110409_AACANP liu_w_Page_072.tif
f6c287510c3cb7902bb0a040cb7dfd81
4b4134792d928c35f750215ee8a25e88032a278b
F20110409_AACBRG liu_w_Page_068.jp2
e4c2864719d549f0c125acb67a9baad6
cd9aa23f4e0599253de81239bcd13dc41ee60c37
F20110409_AACBQS liu_w_Page_054.jp2
421f858cbe361a089a229ee25b3600e7
2ef32e356506382808465a09a8961c0dbde6be12
F20110409_AACAOE liu_w_Page_087.tif
62ce23fa6a448e2e46a773482dc6278a
8cfcbef1f2df8059395baf01627d27605e12baa8
F20110409_AACANQ liu_w_Page_073.tif
c80f96d7bbe21c6637d822704d30fb36
d52b0098359895000f90564c234fb8fd4a330773
F20110409_AACBRH liu_w_Page_069.jp2
7766c1709c91514af0fbd7756b670dbe
431ff4f4c7832c4a9119853e3ee523b9b4b8a4d1
396351 F20110409_AACBQT liu_w_Page_055.jp2
1e0e7aa1343257717a13bc94424be865
9c029baf06da997c18d50cf9356f9779586f55d3
F20110409_AACAOF liu_w_Page_089.tif
daae6c54a46192465de354825e31eee1
f3631dad7c3cf35a8c2bc47e0738e5c652f07d31
F20110409_AACANR liu_w_Page_074.tif
0a9ea831f2e06284592987893399556e
8eac83a61cfad980a27993c80ea04883955fb6b8
1051954 F20110409_AACBRI liu_w_Page_070.jp2
05eb7158bac6933daa9bdbfd699c067f
635e27653bf49ad71276cc77752186686b7b44b2
1051932 F20110409_AACBQU liu_w_Page_056.jp2
28a2974c379a137c733e04436c308a3c
dde6cf968bb8e207a3db8fb5aab7d1ebafca4840
F20110409_AACAOG liu_w_Page_090.tif
c1b0537e2ed19e466cb77a84abf4f91e
6c8839a8ea879fba6a84d75be6808764a3c63b12
F20110409_AACANS liu_w_Page_075.tif
f9608e3cb331b9e5c80ab17df53363d3
274c72aa4a344e9b9dff8b9d4c3db9afd829c50b
726109 F20110409_AACBRJ liu_w_Page_071.jp2
302bab153bb505375425694fc3ed110f
14d4d711f687f081accb47b8924559c6751f4bdd
408736 F20110409_AACBQV liu_w_Page_057.jp2
aad5079b69097ed2a603300ddd2ca907
40cb7bb377523427ff699d7f19c350832ca3452a
F20110409_AACAOH liu_w_Page_092.tif
32bc09950b4480559d68be5ed5bab107
1d6d722f137f566356f38e93e6d8742caaee2a78
F20110409_AACANT liu_w_Page_076.tif
0a476c5794deca63c824a62f9259df44
b094bfca8dddb56162a10c99e1a030fd7a6996e9
1024616 F20110409_AACBRK liu_w_Page_072.jp2
a44797b050c022422976463275193b64
b8cb5d8c242e4bd63943ef0b4a5ff6aac4822805
707418 F20110409_AACBQW liu_w_Page_058.jp2
8fef2b231a58c3e041e635bd7876320e
d08f2bd2bff17d8ae0c37be3b405ad7a0696731b
F20110409_AACAOI liu_w_Page_093.tif
7335542b8d2b4e82305b5dcb0ec9d358
7041ee0345f9dc7c0ef15d20494f3c76daca5ea3
F20110409_AACANU liu_w_Page_077.tif
794640c6dc30bca7f7c7c661f780de0d
ef39b6f6e83f25384a3af154ce69618ce1e3ce69
F20110409_AACBSA liu_w_Page_090.jp2
b6100e752e874d002dad44ccf09d121c
9c307a131aca09fd797df692b75afd2b3de01ecb
1033605 F20110409_AACBRL liu_w_Page_073.jp2
8f71f1258488c571db256f7aad9e153f
76ea9f1fd49134bd3687e6b18a8d56656e847a7c
F20110409_AACBQX liu_w_Page_059.jp2
c5f2dfa4424d511f0a43098695789930
e78a9f0bb19de16decd64607d630722df460fbb6
F20110409_AACAOJ liu_w_Page_094.tif
e6ce93f7ef24075d617bd8908dfe85eb
48d6af3f6b7377c612bab2bddffc089a6ec55b00
F20110409_AACANV liu_w_Page_078.tif
1039cf62600ecf060eb3cbd4fb1d8529
693439dacb515e0cbb18c223dab60df27860f3a7
1021773 F20110409_AACBSB liu_w_Page_091.jp2
371310f93bb8dcb54af40c2e888ca7d2
b5b7fdca676cf1987a670c4448b2d93bb5deeecc
451974 F20110409_AACBRM liu_w_Page_074.jp2
6b1daf5a9103088b13cad5b04e298448
c2d00e34f72e7ef26e369be6da395d89aea02c6d
626484 F20110409_AACBQY liu_w_Page_060.jp2
b0d3084163c86804db0adfbfe0b2f45b
a5eb11aa1fb5495f9b2a5158e95b84fa13a856e8
F20110409_AACAOK liu_w_Page_095.tif
27bb1382fff91ed289cc474abf799d53
b124e307de51ccde69de4fe02c8e8d607be5834c
F20110409_AACANW liu_w_Page_079.tif
50f8d7a5cadb9c632559472c8d390b76
bcde06438ae0b2e89bb7f1c6e87609a0f2728ef3
947242 F20110409_AACBSC liu_w_Page_092.jp2
1be40bc82ba6c926a2e8a69579899f0e
479e7c99310bf06dbdc65ebcea4e54f0fe7f9cca
910495 F20110409_AACBRN liu_w_Page_075.jp2
b77308f13f3cf43f027a89a4b5c0bb9c
7147e379f28b86fc30c732445c69f1ae74efa3ef
F20110409_AACBQZ liu_w_Page_061.jp2
1fc419f00a35edba2a253063bfda3a80
518b96921ff8fd37df5a16e8bf584a49f134586a
F20110409_AACAOL liu_w_Page_096.tif
abd8ac1ed88e492a02a89b19d3668270
a95c8a77b10e5419f4270c4eefc6888ba1397baa
F20110409_AACANX liu_w_Page_080.tif
7059f020ea44a35bac9608fb4b42f75d
7a41fd56891b229139eae227433b6d7a8dcde4ca
F20110409_AACAPA liu_w_Page_111.tif
8b039b753016a9de5875e597d0f1b41a
8562d594fed0b4eed2419e83a9cb4b68bb95ed45
697486 F20110409_AACBSD liu_w_Page_093.jp2
5dc2ae7666e5939323c4ee9172b668a6
5e658eeb0fde5fcd8f2672d45fb55710bfd5d71b
F20110409_AACBRO liu_w_Page_076.jp2
a5f1aed2a28e44c60149f6afab3dafcf
dff7b84930cc160969ead38ea3d17b66f1424b53
F20110409_AACAOM liu_w_Page_097.tif
fd916399a7b49637779f4e1e43dac0be
10dd88353c9a26319a7a7fb3b44c709df586f0f4
F20110409_AACANY liu_w_Page_081.tif
65797824d9b4d3552e6fae8a9f7236db
65c9424103de903b97abc898a360103f3b4187dd
F20110409_AACAPB liu_w_Page_112.tif
fb21ddac280f7eefb0f8fd4fc2568605
278c91524ed60aac9b7c7f105463867cfcfa6903
F20110409_AACBSE liu_w_Page_094.jp2
e313fb1bf814f7ed0dae71d68956ae31
0864d2136959278eb94b96e5705043b0aa3d0270
F20110409_AACBRP liu_w_Page_077.jp2
b96677ce9b92aff01a98a8149372cf34
15abf5e86a5b75241b7d88bad897ec778cceb864
F20110409_AACAON liu_w_Page_098.tif
ae8d3e0844da8a7d7aca5f760521a953
e5ea970de18d73b4842acc3afbd4e744bfaba9a8
F20110409_AACANZ liu_w_Page_082.tif
a5f58cb2232c18ee563ef77ce2e8adb8
9733d8b88004ce19f9939fc15d47fcc25fde44ea
F20110409_AACAPC liu_w_Page_114.tif
5c5a2ea7cd77b684abe6f0adf0c32d5a
8fb10b5dbe095bd3cccbe6fa7381b243fac902e7
F20110409_AACBRQ liu_w_Page_078.jp2
a3f3d86b439bc4e7f4b76f6f98604b99
cd0b4bfb07190088bb5dd5c364412a7ca9a80152
F20110409_AACAOO liu_w_Page_099.tif
15d1db7bdecad4e3d15efef0b7b9e350
d442337be16c40b2cb46d7297959f24806afbfb2
F20110409_AACBSF liu_w_Page_095.jp2
58dda7ea78a56a10e06d53a9e09a1b22
44f542e2e98773f1375978b2eea89991b1eb1fd5
1029765 F20110409_AACBRR liu_w_Page_079.jp2
972ea7f695dfaa92a564919beb277504
a8ef0562ee462c41e066f35b6f0db070d92a2a6a
F20110409_AACAOP liu_w_Page_100.tif
32e036e5c0ed2676f34311bf6e32bbf7
7d53b81b8de9998209d1662959300e3e22c31afd
F20110409_AACAPD liu_w_Page_116.tif
44cef1a9b33307aa6b2bc07f9cfc7f09
6be019cac9920cc363c3419de28c76b5a9d2b547
F20110409_AACBSG liu_w_Page_096.jp2
3d0c9cc329348be19611ee56c45da864
04ee251f4e7feac01e62ddd50189969c7cf467a5
F20110409_AACBRS liu_w_Page_080.jp2
07093e5fe04aced9aa9711e1d83e0e44
46108bce0e706172b8a740f08fd3895e333cfdc4
F20110409_AACAOQ liu_w_Page_101.tif
58491b978e98d05192767794ee74a5c2
3cbfa83a71fc9aa9c14a38771c5386ee9b224bc9
F20110409_AACAPE liu_w_Page_117.tif
6d8a156b92bfd0114712a2202bb9091f
d0a40efd8b6328faaf3539857a86615311dc4b10
F20110409_AACBSH liu_w_Page_097.jp2
c32de7977264a334e5fa05b2421d0095
ebe7683e7513316dc14e1901493f76ca688b7c29
553567 F20110409_AACBRT liu_w_Page_081.jp2
7cd5bd20ff03594f3e163f022055497b
8540845964df40a925b1a592d386e4989e607537
F20110409_AACAOR liu_w_Page_102.tif
82ad785235b16d27d41926c5015f8b84
abd73c4990157e3cc8841bff1688472e8a30517a
F20110409_AACAPF liu_w_Page_118.tif
91c3695f02120f6772e770c4dbb77938
c32879a5f2ea1ffec907b57c47cef56107a1f6dd
1051960 F20110409_AACBSI liu_w_Page_098.jp2
cb89df245ed48a670d11192cf6fb275f
e72766800e9d7173d16ed000a069c8ddd9e9d7b2
1051955 F20110409_AACBRU liu_w_Page_082.jp2
fca32af498510ba0e1749caa3da9a99c
c333d297391e446ba64a2dec211e08c96c5d8fdd
F20110409_AACAOS liu_w_Page_103.tif
034d602f7fcb2b10e0fee299563c7cab
6481fddced0d261f66ada0beee415ea813d6740f
F20110409_AACAPG liu_w_Page_119.tif
dd63aec64c76528aec4adffd92a41c45
1247b31b2eab4650898d2f9c15518bd0c47bc4da
1028639 F20110409_AACBSJ liu_w_Page_099.jp2
3603a14e5c80a6eb5ae46b78f9304b73
e13e3e2e40c5367859bf4f494ed20f0f1173a581
558530 F20110409_AACBRV liu_w_Page_083.jp2
945010c5e3da5bfff2c81eead362d175
3bb0a3993ab355ededd8ab8a0c4458c5c8dc2f61
F20110409_AACAOT liu_w_Page_104.tif
842a48294ebb384aa45adca19bad4afe
7bcd98767b025842936105c2a2a2a4f5fa0f38e4
F20110409_AACAPH liu_w_Page_120.tif
c4f36d4675b00a055402666012f69c06
8684e0d5c3957af5552e8ec9130160fa5c2e9484
F20110409_AACBSK liu_w_Page_100.jp2
0d0270f6809d73047c57d596d7a2bb2d
57c376c98cd7f298262d40d1ba5a725964f41b99
F20110409_AACBRW liu_w_Page_086.jp2
638d2b82a59479ba35a296b5be79570e
eb2758749361a7cf3191582062ad86ab5e6bd2a4
F20110409_AACAOU liu_w_Page_105.tif
c999aade963e7c1cbdeae747ecb4d0ee
a0c479987c89f4b26f389b678e2f6f96ed7ea9e3
F20110409_AACAPI liu_w_Page_121.tif
22f399504970ecee2a0ddcef8c6a14c4
fec52387ef2e80bc722dd0b6c8afc96877064857
1051964 F20110409_AACBTA liu_w_Page_116.jp2
44e433360388fdc19f5a1e0fcab9d0c6
4c860eaf22dc3f8da1b8b26abe3c162719850566
1051937 F20110409_AACBSL liu_w_Page_101.jp2
3475692cb086b1a2836bd08c0886dbb9
b132575d4e4ce582dd781f437a612917c32c5762
1051940 F20110409_AACBRX liu_w_Page_087.jp2
57682c70c43cd24daacead0c61ef9de4
d01681bc75571dbcc85a78a6c11eec809203de5f
F20110409_AACAOV liu_w_Page_106.tif
b5ccea87c20ff6ca22a6ee174d0a0ae2
ea16f587dac4ae829fe117f8198cd15d4605b634
F20110409_AACAPJ liu_w_Page_122.tif
3c0d3418815d6a5b8c14108722bfc1a8
8bafc0de4ca8ce1adaf608532e6e6a96e956e08a
399042 F20110409_AACBTB liu_w_Page_117.jp2
d3d65909d6c66dfc69e4cd7f60fd8001
529dec3842f1be5f09dadd88bd957bf8cd32cd4a
734063 F20110409_AACBSM liu_w_Page_102.jp2
aa6ee82d67080205df1a24d1cfa16e6c
16b57b8dc228a99ca614f7104fbd3c858d739354
F20110409_AACBRY liu_w_Page_088.jp2
e5030f9646cd75faf39d06fcdb988056
5e9683b768a85e27babe6ede8539765c07739836
F20110409_AACAOW liu_w_Page_107.tif
2e62abd5de7e445a5641d1eef13bcfe2
b3e985c6bf089dc2d8b59599ff175fd36cf3a1ce
F20110409_AACAPK liu_w_Page_123.tif
876205099b8e227965b1413415cff9bc
9767e66d2205f0f82be2322279d024c3fd65a3ab
1051909 F20110409_AACBTC liu_w_Page_118.jp2
f9c8bdb10a6967667175327ce9652cd7
950afaded65d5f7c22f19ba1306fb58dedfc4f84
F20110409_AACBSN liu_w_Page_103.jp2
cc66259f624fa555a902ec1b3f1264f8
271d7f3e8d38396521a47171abcf6f23b65e1d1a
1024883 F20110409_AACBRZ liu_w_Page_089.jp2
eacb51666ffa229d671e435833c48a1d
715d5efb37e4492d9e1a17b6be55b90d573ba308
F20110409_AACAOX liu_w_Page_108.tif
d8eabacb4db2788d37446ab17eb0a57f
9f964ede501d6d252390f9dc9a34b3d77ad6a14e
F20110409_AACAQA liu_w_Page_139.tif
4cb4ff820b6d4b1569d2db50b2a1d166
348081ef80dd74565d34e8019e31ed8883872199
F20110409_AACAPL liu_w_Page_124.tif
10cd8d90c2e6e79a22846cd0372b66ca
845668bdd9109337f8153b31f066d9b1ee8770c2
1051970 F20110409_AACBTD liu_w_Page_119.jp2
b9d737de8d924528fae090574ff1d9fa
1a0f7b558fbb81d2a4c107f02aa3f50147f8f4d8
F20110409_AACBSO liu_w_Page_104.jp2
8d0ea42147b61c830b8350903367e729
6242ba11210752d3359d075ba1fc50eb5c015a0a
F20110409_AACAOY liu_w_Page_109.tif
ea66eec1b0704135531aac63040a4871
9c1537bca82d0b593e9b8c59595f1731ba936a97
F20110409_AACAQB liu_w_Page_140.tif
4c8ec46f656164005210ff8bf72e3989
9cd94ae2e9aa33e36f6566a3f70077ef504c8648
F20110409_AACAPM liu_w_Page_125.tif
2f6923ef90d73b32d814c80960ebb652
009d53e265d6e8e824de2db2e3485e09e045b633
1051921 F20110409_AACBTE liu_w_Page_120.jp2
7903de4ab7e9ea5cbe2dbd2f5f4530b1
323507404dc244bc9071ffa32283398986620bf3
1051963 F20110409_AACBSP liu_w_Page_105.jp2
d5dfb7a22a75b77be54e7513bc0ea660
1c34f86a0fd84dcf019e899184516585c7157126
F20110409_AACAOZ liu_w_Page_110.tif
40c8c3c18865ae1fd996a05c2b4436ec
352f07b1db02124e390e05b1f8414d3368f6ddf8
F20110409_AACAQC liu_w_Page_141.tif
67cd04126f66245d1b01c8f25fda5679
3514b66f94c6e2f62e39599823ec5f315948b5e5
F20110409_AACAPN liu_w_Page_126.tif
81b8338844ecf9df6abff566d8094c72
18659477a2ab7107b6a97239188a869dc0476857
F20110409_AACBTF liu_w_Page_121.jp2
66080df7d25b6b2293f8f34e699d8c8a
1443bc68844cc6c278d49a8d000c77f8e745cada
F20110409_AACBSQ liu_w_Page_106.jp2
49d80bcd0cec8d6a14f376038e8507d3
2633f3441bf7bf8a8bb88a865efc812605ba2239
F20110409_AACAQD liu_w_Page_142.tif
64e3a671ac47a957b0702530ff16efd9
fce9f4b151646fc8176cd95e65a55ac2156037ad
F20110409_AACAPO liu_w_Page_127.tif
2185accb5f290b5d4c6c6bea118565c7
b29a9f3d2067dba9729e12b6b254ab7fe1c7901b
F20110409_AACBSR liu_w_Page_107.jp2
29c411d17da5d300c947df048a6154f7
babf70e1515ebc9ae611208f96275465b22fb700
F20110409_AACAPP liu_w_Page_128.tif
3a263c7b16bf1e19042adfdcaa499804
8b387633f5d37d91380dc976d705c36aae38fb9e
F20110409_AACBTG liu_w_Page_122.jp2
6fb05b6aa1e47438be5bb9a0652fc706
83e3d801cf6f6312e015a081ec48be7d3dce209d
777119 F20110409_AACBSS liu_w_Page_108.jp2
179462f762ab40d42960959408faf915
a7f5283ca5563957fec068b0da994a9aaa7dfc0a
F20110409_AACAQE liu_w_Page_144.tif
809c61e9a2923ec1b27d9169e9b54f2b
dfe22b44479bf2107cb4f34a9f744916b890f03f
F20110409_AACAPQ liu_w_Page_129.tif
2441fadd570c6a52061bdd50f0cd6aba
fffd8ee6b39dc7f0a835079de55015bf7e5a7480
F20110409_AACBTH liu_w_Page_123.jp2
6a9c97e44b250cfe10d9bdc1d7bf1e41
dfce63702f5f2ca8795ce0149c7874ff443f99c6
906817 F20110409_AACBST liu_w_Page_109.jp2
754787ef4e8c320146dd2909c544dd15
bba615a070e533cb6c8b2999e10f12c3629c5432
F20110409_AACAQF liu_w_Page_145.tif
0466de7fec7f3c7cabeeb519692cf9b3
68fdfd79b56ba999a1f51b3832e9d9d16b2e25d5
F20110409_AACAPR liu_w_Page_130.tif
bd806a72a6750588165a25655ab1ed57
52c3f570818b76d011d97157854480ea3cb01292
F20110409_AACBTI liu_w_Page_125.jp2
fc9b7aa5a972347448572ea5f4bc5c5c
8130bcc397c5f6dc6ad74af56dafb1916b697c32
1051914 F20110409_AACBSU liu_w_Page_110.jp2
b4692c8d97f6c1541036b13f58d43cf2
1254b87a7473c3425964bfb1b22b4cb8c94a1089
F20110409_AACAQG liu_w_Page_146.tif
1c36e0f90ecff282e6940e2ae33c1da3
bb7d55bb2551a598f7551758ff6d0407b21784c0
F20110409_AACAPS liu_w_Page_131.tif
add1989a963677e4b209a08b29426f21
fc7c7d3d4219fa4bddef8c8ca3416bd64a12e5c6
F20110409_AACBTJ liu_w_Page_126.jp2
abfbab9e489438cff42034a8c4a8be90
01d583872869383e59ba1963ac917bafed2b1c6d
F20110409_AACBSV liu_w_Page_111.jp2
d8b20501dc1c332031ac05341df5e2b1
66ffaac2bc2d2d92848308181e6189171b5f34a0
F20110409_AACAQH liu_w_Page_147.tif
e8bbc655c64e1d61da1b5702f87d158d
34b7f6dbc4830c3c1f00f45e933b3dfc2e9f22cd
F20110409_AACAPT liu_w_Page_132.tif
4e26f1385448c72050969b9f24f74aa9
8e4b8f3733bafb7b2be0af8468d4b51d3cdad1aa
1051945 F20110409_AACBTK liu_w_Page_127.jp2
9b66f9712bb21a44283fdce94ec26447
bcb1af8728235dbb870d442a4d7315b8b34fa071
F20110409_AACBSW liu_w_Page_112.jp2
3e027a57eef9cfec90ad62d104008d3b
06c834d001bf185704b7c49221f2784e0954c308
F20110409_AACAQI liu_w_Page_148.tif
d854d523d127b57e7377a3e2ee58a104
be7ffe7d99a5b2b1d5ec3151b54a7c7b657ba00b
F20110409_AACAPU liu_w_Page_133.tif
86c7bf605838e99c3ae3200999b92fdd
65e33fd9dc2f152c0d0f8536c73ce54441854d0e
781530 F20110409_AACBUA liu_w_Page_144.jp2
71b51509ecacc61e36ce1c5ffb2bcdb6
301702369d6c9777ed9eed87c80d2a5a62045cd5
F20110409_AACBTL liu_w_Page_129.jp2
786b8e2763f1483fb9d28a5fc228a015
37476bc9e4b80a2c4d13b8248c1e8af0cc6656c8
1042137 F20110409_AACBSX liu_w_Page_113.jp2
84d2e5cc10e52c6eae108ffdcca60d57
38e25ea0a4f8d108b4b8388f789d6bff2cf250bf
F20110409_AACAQJ liu_w_Page_149.tif
94673d70a3432f193ce57b8dcd5aeaae
8ccaaf5af6c1744b07d29be5691ab85df55f311b
F20110409_AACAPV liu_w_Page_134.tif
d432f7b4af7ad9ec9a376df67f975d9d
886545eceed02e1f0d68047d6299a11a24d724b1
F20110409_AACBUB liu_w_Page_145.jp2
ef7fb50439974cde65cad135d1351bc4
66187ccc02ec9981b3c386d7af2fb7b9a9937eaf
F20110409_AACBTM liu_w_Page_130.jp2
f0a0c4de1a7e301919da8f0ef066e785
bf4f6da1f2d39a91e2122d2c35bf4c2ff7232666
597906 F20110409_AACBSY liu_w_Page_114.jp2
e2fcfaa47bbb424bbebdc8096933eab2
91be820affe28c836338c28396aed09c275f92a9
F20110409_AACAQK liu_w_Page_150.tif
43870a59c5ccdb4e2360e7a8fd59520d
623d7b665b3da161b593e78a3c77ba0e2b7f7ebd
F20110409_AACAPW liu_w_Page_135.tif
dad03959a19dc35d3a4d76aedd6c2e0d
1234a3dd3208aa4cedbaa782a5947ef55a7cfa75
914058 F20110409_AACBUC liu_w_Page_146.jp2
b76ed61b4fbea1e67efe7dc3059cb86f
a7a52365a9ae1ae79bb9ec5b66548498be95bdd0
F20110409_AACBTN liu_w_Page_131.jp2
b9c90fb304fe7da9d711540528e8af9d
d809acae5ee58c9832314d13268b2f47d143e206
F20110409_AACBSZ liu_w_Page_115.jp2
a1a8679830eec178a4f435a1c2d4cc7b
f69e7eda344eac756c4814b1484fbf08312a9126
525 F20110409_AACARA liu_w_Page_001.txt
9d6e985abfb9af9c234d6390f3156525
6fcf2c5c2542cfb29e7f36d3fc8497ddc788cc0c
F20110409_AACAQL liu_w_Page_151.tif
800143e894b88e84564db10947485de1
e815c06b3bd5677b3b042cf7461e7e5f01572026
F20110409_AACAPX liu_w_Page_136.tif
1acb8c61be7f7872e764bbce8cdba953
35a88c50c0ca38568d38c68cf67247e91910f98b
659327 F20110409_AACBUD liu_w_Page_147.jp2
f9e2d614693a7de0f9ede972b303c99d
d19bfdbe4304d9c234159f2489847731fabd00ef
578994 F20110409_AACBTO liu_w_Page_132.jp2
de31816f1c0ebe089e5008e06cbd3ce0
09696adf441dca8dbad85e7823c2464b7d1b21ec
F20110409_AACAQM liu_w_Page_152.tif
71870fae87d0948ba4e85aab4667614b
d0751af6364e49e4ad0737bb065b89fbe9c7433c
F20110409_AACAPY liu_w_Page_137.tif
57c3afed9a02e39f47302636b53f15ca
38c05426b66fc7a94aee531649d50e835c7a92bc
81 F20110409_AACARB liu_w_Page_003.txt
f16ad725e0cebb955f4cb18d93ddc308
89b3744f051068e43622f6a6f58f72f2a6a8c8eb
F20110409_AACBUE liu_w_Page_148.jp2
c314ededa0933354259796b36f4e84e0
52e9adcfa211a4f85455fae287b89c45c2eea966
1051873 F20110409_AACBTP liu_w_Page_133.jp2
e6df3d910ecd564dba09704b0a442d4e
5bb62d25385f01b39712c151b833f81690a884c3
F20110409_AACAQN liu_w_Page_153.tif
dc40a1f9fc24e894a7720dcbc58b0d05
4acd23250f953d394c393eb8bef450634f5b21e5
F20110409_AACAPZ liu_w_Page_138.tif
ea8953d541e0e50e96043bb4df27ae6f
7d2de5b007821d7004aa1db6fcce6f40abb7b5df
1192 F20110409_AACARC liu_w_Page_005.txt
e1e0d91bca032e3742c883952536f9bb
574a6a4d5367a410d23670f1b4e3d0f1c5b9adaf
1051950 F20110409_AACBUF liu_w_Page_149.jp2
0f25117bc9d75fa3d821ae1c9b16e5e2
f5d1700f59c98eed8e1cd2e5375d0df282585b3b
F20110409_AACBTQ liu_w_Page_134.jp2
f97111ff7fc0e5d34342c3f00034da9a
a6002aa18c0c568f525b560d05379032f36c7fc2
F20110409_AACAQO liu_w_Page_154.tif
ca9dd8cc6e77dfea7791024c633128d0
e9a5bff597896e7c9435c8d9e284068289dc9dda
2766 F20110409_AACARD liu_w_Page_006.txt
17de25a9b15353c0b70c757d58398399
0385fd06f5f9d1354a10f42647fe99e799c4648f
5506 F20110409_AACCAA liu_w_Page_143thm.jpg
5ecfacc2ac8c7cc05ebab1538c4ad667
3b31c2df42a946bbbf78c15bdc98b5fc9518d7d1
F20110409_AACBUG liu_w_Page_150.jp2
2f8bf6f905d1b732ab8740af22a370a0
fbff7767f185b37d40829af211dde57dd615528b
1051941 F20110409_AACBTR liu_w_Page_135.jp2
14db785be7377872fc77d3beac31361b
7e738dd7e56e7d427076bf479b53e837d7e654db
F20110409_AACAQP liu_w_Page_155.tif
9832628e5f5bd5751be54276ee56b5e0
b76dfe5e91ee408d81f3358cf8cdbb2ed334cfd1
1563 F20110409_AACARE liu_w_Page_007.txt
800dfadc537341352e56757767db612a
0751b0fe8aa81939ff2e2a61454b1ed91c434a6c
5588 F20110409_AACCAB liu_w_Page_144thm.jpg
5bc2a9d078bd457a70bcac3f4d73ab8f
782acab600447088d6205183fc08b63dad0fa510
1051922 F20110409_AACBTS liu_w_Page_136.jp2
f1c4ae1277381e91d542fddd66068a63
2edc03e0b6dcee9f183847dbbfbfa88280098d31
F20110409_AACAQQ liu_w_Page_156.tif
ad68a9660e6614425817a1e5da835e21
54542942a757ea5802dea5c323d48f5e1c2bc966
8906 F20110409_AACCAC liu_w_Page_145thm.jpg
5c593b0eaae21563598c17085eb16cc7
6ff3f6fb1680df52c73a0fcb8a55ab2940a577f1
F20110409_AACBUH liu_w_Page_151.jp2
5e634ae38beba5406fc569dadc6d8e5a
1de7f462407eb9ba5ecba749951524f72e5fd324
F20110409_AACBTT liu_w_Page_137.jp2
4cfc5d3f60765f48d6ae266c9f2a4f34
349f6c62e96675cdeff45e01a61d0dfd16f356ab
F20110409_AACAQR liu_w_Page_157.tif
5d04969285c1e2479f81274323126353
88cc128adb350f28f078ba3bd447ca56f9d79a97
584 F20110409_AACARF liu_w_Page_008.txt
9c9e92eb6edde1883d14e369280d6bda
3dcd429866d832b9c35fafb5b21cef9acdbf61e4
F20110409_AACCAD liu_w_Page_146thm.jpg
17666777249de4492594af7977fb34be
6d51c012875b682ba8cd5aa703daf4d82054f197
218794 F20110409_AACBUI liu_w_Page_152.jp2
1f4adcdb7174a2bffdf64367c2c3dc14
8f69732628531cd93a6d66617c9c6d87815e8d4d
F20110409_AACBTU liu_w_Page_138.jp2
86943cba6461ae7d44394a4e19c4e722
e9e903c91c7bc7b27970a0ac85fab739fcb206b8
F20110409_AACAQS liu_w_Page_158.tif
0a6986c3f2bcc0bf4b727ada54d8daa9
ef32852b206823937d5490a6d701433cc8a6c79d
2040 F20110409_AACARG liu_w_Page_009.txt
2f7e4db6b1a05755b7a5f50105382d4b
f0f3faa22e420239e46d73c5f856cd3ceaea8000
5739 F20110409_AACCAE liu_w_Page_147thm.jpg
62d1389c672a479cf6c1eb33f331ac56
b6b15166620b8e5d88fa5e7060747a76e291a052
F20110409_AACBUJ liu_w_Page_153.jp2
fb37fd70354f9c3c39d7ed0a76011a74
2f80e300ba9946e335b1608163ff064f1ce1abfe
1051929 F20110409_AACBTV liu_w_Page_139.jp2
f72a8614fab117aa4bd8db8edc302970
9032619532640f4cacc674d5c630eaee51a35227
F20110409_AACAQT liu_w_Page_159.tif
35d3b90110105973e242938a22911268
e38756cedfc1d93bc41d50bd8891251a1b869688
1819 F20110409_AACARH liu_w_Page_010.txt
60c7209b2a8c2f6ecee203fca70a4516
c1aa4ef3338a79d8e8d8222307eef4e045bd3bab
9403 F20110409_AACCAF liu_w_Page_148thm.jpg
00cbd9e1791f36b1bd7f5675c78bc34c
a2ce69040e9fd07e80632e2bef75e7239b009c37
F20110409_AACBUK liu_w_Page_154.jp2
9fd26926a2da7a96063495e20638ff49
b31061f0dc8562df97ceda1f962938addf05c4ee
F20110409_AACBTW liu_w_Page_140.jp2
95b3c906c05d6f59eb74df29ca96b615
0cbd6c00189dfa113d41bfee7e055d6bb7bcc8de
F20110409_AACAQU liu_w_Page_161.tif
8f2938d65fd4077fccc30823dfeb3d06
724550178832a439878c7cd677c624ec97fa3c4c
2188 F20110409_AACARI liu_w_Page_013.txt
a0745e343163179f671d5d73617019d7
2698eccab22eb31fed76d6b1f15c9930b362c0f3
7471 F20110409_AACCAG liu_w_Page_149thm.jpg
dbe0ee800d47c0acc37f175240a9bc22
334ac26520c028677236e68ad9fb68ced4bd3158
6353 F20110409_AACBVA liu_w_Page_005thm.jpg
1277fad61b09a1b690f0efee37b3bd84
f0860afdfb74cd3ae0ea21d9b52a3aabbdb11748
F20110409_AACBUL liu_w_Page_155.jp2
1f505c2dc59fdb7b5e48ebfae1d3eb7b
b27cfafb6ff96f309dabb3bfbe3bf4ff3336e7b7
F20110409_AACBTX liu_w_Page_141.jp2
85a95e4bd8fc041eeec95cc44d246f11
e426c194290c4509f11ba60d67f68c753a7a6870
F20110409_AACAQV liu_w_Page_162.tif
444476dbdb6aa9a09f641618847da8bb
1f410608efe8c70750291f873d5b544bd36b8a84
2270 F20110409_AACARJ liu_w_Page_014.txt
c4795046cd48e4e918b8c12b270e8051
6212b4f01e99ac9e3a109f33f523361bdb563b47
8408 F20110409_AACCAH liu_w_Page_151thm.jpg
f7b1fcf3f4e7f3f45fcd985013cd89d0
28315774e3f34a501760c237188426161a1ac4d0
4804 F20110409_AACBVB liu_w_Page_007thm.jpg
e6097ca0525a73f5b9fbc89f42838dc6
24fce862c75b9f229d5b8f198fe03abf133b3eef
F20110409_AACBUM liu_w_Page_156.jp2
b710dee4d8d7328276495815bd75fa74
00067889e49f1765eb517161204c71ae713fb101
1051901 F20110409_AACBTY liu_w_Page_142.jp2
ae4fb559d24cabc3dcbffd1ac57097bc
217c5ecd95e1173d7515638a1289984e585b0f4e
F20110409_AACAQW liu_w_Page_163.tif
bfae05e9ba21d8db1f988f07bebb27e6
6c2a9c23fe2755323044f16a253bec6d8fcdf71c
2392 F20110409_AACARK liu_w_Page_015.txt
2b8b1bf02b2f17d71cd2af23e3020a70
06e16d0acd9145354b1235fbe70d9b50144f5087
1962 F20110409_AACCAI liu_w_Page_152thm.jpg
a808f7cf7f2a0097a3f4c08681449a97
22e313b50784183cbc40c5fdab473c966c666a6d
2846 F20110409_AACBVC liu_w_Page_008thm.jpg
401e51aa09305351b483ffdda7908e25
752aace48305de608fc797212cc0d77129144ab3
F20110409_AACBUN liu_w_Page_157.jp2
a70be324cd38650f59393e271f2d55df
a60e046ddc7cc2c2836e2357a865ab2d500f1d38
682693 F20110409_AACBTZ liu_w_Page_143.jp2
6d34fa143f91eb32514588793ee26fd5
dea50f391edc1d29ae1ab76a0be7d71c62c5c6ee
F20110409_AACAQX liu_w_Page_164.tif
24f238aeebf924105a9bd069266990e5
4ddb868f3d6657d58a57bee23d7ea5f072a0071c
2398 F20110409_AACASA liu_w_Page_031.txt
3354b9912b576de1b6813a3f3cbe3d31
cbcc027a9058b99cfcea201bc8f8da638c020c21
2439 F20110409_AACARL liu_w_Page_016.txt
0b64e90acf0c1d696ecf25fa7bbb3bd4
a41695d8e9965c0a72b2b6f01b0f20233eba1cfb
7823 F20110409_AACCAJ liu_w_Page_153thm.jpg
ac4366c63cbfc37e247f7538db5f9add
769fd6626db1b1db8947dea1f20777b29ed6a48a
7541 F20110409_AACBVD liu_w_Page_009thm.jpg
b8202cccb2a2d4a75e55c0c8a8163dd3
44ad64934d53b9e9976109948c2273d9faa60748
F20110409_AACBUO liu_w_Page_158.jp2
cc8f3fc3f74ec3a9dddd2e2c2ef6ca3f
f12283fe6b71f0ed630db84706e7631fd384af0f
F20110409_AACAQY liu_w_Page_165.tif
127b1121526903c8ad176a1a6305cab4
b67fd014b568f28614e17db28c7362ce2baac294
2211 F20110409_AACASB liu_w_Page_032.txt
802d61880b022c61aac4e3bf8899aa9e
a000a6c6364fa56e8862c4de84d7ef89a3ecc202
2421 F20110409_AACARM liu_w_Page_017.txt
cc27eff858f057d06c941cf26ee99ddb
93444781b7e24fed413ca07af7aeb1df0ae9ab36
8820 F20110409_AACCAK liu_w_Page_154thm.jpg
ab9fe6cb2a24b1a98b6151159969f59d
2a898adcf7cfc3ff113c62bf391bfee9a952fa90
6929 F20110409_AACBVE liu_w_Page_010thm.jpg
7c4383f519e9d56a9c06daacd0b19c23
3ec3469dd31e424a6eecd740b17529a23a73dc8e
F20110409_AACBUP liu_w_Page_159.jp2
039177a5f1c4a53bbcdf195ae4a12278
1445710369435da99e752952d52ea10132e1c53c
F20110409_AACAQZ liu_w_Page_166.tif
16f75bd577df39d1adb79b898e210695
1892d5cb99d324d9d84b3c10c24fb5130b12ec1f
2343 F20110409_AACASC liu_w_Page_033.txt
3253ce842b6c58d03716537ccdeff2af
f5b5b2e4627974db0a0eac692072b8d0b9e78abc
2433 F20110409_AACARN liu_w_Page_018.txt
656217badc786a15ca5b92c3793b1ecd
2250a7eb90a381d250c98566d38ece41ccfaf42c
F20110409_AACCAL liu_w_Page_155thm.jpg
b09e9fe513ebf0d40541222124e48323
3a945a8d9d1f48932bf2b3ad04d1aabb6d1f18f9
4466 F20110409_AACBVF liu_w_Page_012thm.jpg
1debbf2bfb720e5b02dbb24267ea5240
d2e5c5f0623eddb029b3205a55eabfb7c032c0f6
1051958 F20110409_AACBUQ liu_w_Page_160.jp2
5a501a85a35a25aabf4605eb9ef85db5
8d439178a6d2ba0c42c0e1abbf97302ae41957df
1623 F20110409_AACASD liu_w_Page_034.txt
75bd98a7179ce09aa1409c577582c592
75d509d29d6f44ebf3c4f69eeff9103f96c429a4
2364 F20110409_AACARO liu_w_Page_019.txt
6335a1cc86802c97f4ad7f6b43768d6c
c885e9a81a44c673e5a0f58c771750a1cbbf457d
8620 F20110409_AACCAM liu_w_Page_156thm.jpg
e369f0523a6700884d8b58cc2b44dad1
a9a8cbc694228b1c5d80f9786a6b5d4da53042da
7463 F20110409_AACBVG liu_w_Page_013thm.jpg
5674e11a28ebcbf93982a7483272c250
a64640a6a27d4961a44cdcb36cd44b014b6ee7ee
F20110409_AACBUR liu_w_Page_161.jp2
6ec657e98038d6cf24b00ff244749bc1
761e9dfa8ea0483b1e8358b62641416f698192db
1490 F20110409_AACASE liu_w_Page_035.txt
192a3ab2c4f0f2cc6d5d46a71510741b
9c0444fdff23b3322e411ee1708151f33be0f316
2206 F20110409_AACARP liu_w_Page_020.txt
9fc125b76cf39f1d9f29fddc66c9145a
739ff6cb55e424c64855c5643952c9a1a5923aa8
9168 F20110409_AACCAN liu_w_Page_157thm.jpg
abc4dbc810ef81225ca2abc95cb2bc02
1e8d1910e5aacfd21e8cb9db328d5a731fe77098
8130 F20110409_AACBVH liu_w_Page_014thm.jpg
d379f3d186b2bc8caf6f3fa356b00db1
5d3f427b6ccf31f33d87eea39e454288e023bb1f
892097 F20110409_AACBUS liu_w_Page_162.jp2
56f30f1f18b7048b3100f45eba2451d2
0097a8d4a826d3ab0549a204fe0df798b5074195
2413 F20110409_AACASF liu_w_Page_036.txt
a32cd7d1367284c9c263bf546c0a8ec8
b2f07b45e09158ffeb4691fa6ad6220710e9fab0
F20110409_AACARQ liu_w_Page_021.txt
4074aaca33aae1f700c3fbbae8ab3ded
e55754180bb93b0cd9381a21db6eb61314ec1419
9150 F20110409_AACCAO liu_w_Page_158thm.jpg
e263866db4c59f97e815aa75cbd16afb
02556f645dc39eaa1967bf6b3d7421b240f820b3
348551 F20110409_AACBUT liu_w_Page_163.jp2
53d72ed8dda596a82859b16fd8f36c56
57ca1047798604d92eecc8df7c80ab85413a864a
509 F20110409_AACARR liu_w_Page_022.txt
7c4f930c2dd20d0a10391d4c0add1e7c
34bfec0748a52278f8ad3751e7ffa8b761d6d50e
8744 F20110409_AACCAP liu_w_Page_159thm.jpg
a72e65570af4ea4ec30efbf3caf9548b
eb3ab9d8ea1269f69b887c46519afb30febfa92b
9156 F20110409_AACBVI liu_w_Page_015thm.jpg
0f85803d236f52a8c8d35442c4066a48
49bbf4fd471be5b14f66215a22bd11c3e5129971
205724 F20110409_AACBUU liu_w_Page_165.jp2
d10abe15cbb97f93eb8ec3867839eade
1f448346ce921dd9c9271a3ddbc0ad044f53c07b
F20110409_AACASG liu_w_Page_037.txt
cf9f1108b8b496c13858134ade1b4e80
8f956746f6bb22293811e0ad7a2ab469fa33d790
2036 F20110409_AACARS liu_w_Page_023.txt
9f10d50a569c84586b6c7a99499ab5f4
359298120675a57f73d416ddc56ff3fd8cb15bdf
9008 F20110409_AACCAQ liu_w_Page_160thm.jpg
74ede7eb62ede46193c36ff395113f62
2e0f35b749525e34a42b82252e07e8b73f3a0d39
8950 F20110409_AACBVJ liu_w_Page_016thm.jpg
0a7b1df1450219d145e267c5be61d00e
d370832cc53423b63e809be14174d0f58992dfde
736414 F20110409_AACBUV liu_w_Page_166.jp2
bcb9e1009f41a72915655a2dbc6bab43
77516107acae99da5e289fd49837ff3743678193
2181 F20110409_AACASH liu_w_Page_038.txt
4f77741f9b26b449e8b6e3329bfaae6c
4eb72ff313088f555a83412d06b8bd7e8aeba302
2205 F20110409_AACART liu_w_Page_024.txt
8721d6020fe229d654e6c0f06c01e510
922ab80cce7b68bbf230d8e7f2203e53fe68433f
9195 F20110409_AACCAR liu_w_Page_161thm.jpg
363d64dd51d27e48b720490d7185fdff
430c456c47065c813184cfe2cdcbf2e470980e43
8786 F20110409_AACBVK liu_w_Page_017thm.jpg
bfcd46b67f2a0d47793a126ae70f0c65
10013ad91fd414c3e986bd431787a26491df1019
F20110409_AACBUW liu_w_Page_001thm.jpg
d342753bc0c2909f48d26f41de87c544
049258c916e57c7a2c0aad9408446c14d0e45f89
1948 F20110409_AACASI liu_w_Page_039.txt
76af400b50bfb384f6291e858aa2aa06
ec4a5932c2c3fbb966d5305b243437dec8ea97ea
2381 F20110409_AACARU liu_w_Page_025.txt
9ab5eba9cde1b0d39d0dd0ed4f2e256d
c0a77b3c453c632af38c90ee8b746925e25573ef
7479 F20110409_AACBWA liu_w_Page_034thm.jpg
dbfcdd53647e4c053bd80a670fed40a9
661952b207a41fa6c251d09a4540d2ea3142904b
8941 F20110409_AACBVL liu_w_Page_018thm.jpg
6680d44a08fb63fe550f25d9431a6699
da03f761a55159571d212ae949753f7cc3a5fb1e
561 F20110409_AACBUX liu_w_Page_002thm.jpg
8c33c52a01ed634c1a5379db54c91833
edadd57fdae4f5db7e8dada8599ee4a8e62cbf39
1825 F20110409_AACASJ liu_w_Page_040.txt
9895d4db626889066b6b8d4c33749db8
dbe9f95d5eebb580e36fb6ffc297fff1fcd3c8b3
2193 F20110409_AACARV liu_w_Page_026.txt
96a5b31ef97c4a560e5be66ab0ce9052
c2f8457c7c13fabbfa9f37d3275a6063a1069a71
6003 F20110409_AACCAS liu_w_Page_162thm.jpg
2ae00d0dc865f3f9fc5d660ded18f8f3
1e28e08b0f6fdeab9edf506c31b7358775b266c2
6178 F20110409_AACBWB liu_w_Page_035thm.jpg
512229cd54739216f823d4d1ae1c4dd8
fbc389aaf795cca207fd4731e98355700280f7a2
8621 F20110409_AACBVM liu_w_Page_019thm.jpg
ac8b5bafb8c0058ac15467bf01c572d5
8b086aea3a2cdf2ff3d00f3e29d84a671f2ebad0
596 F20110409_AACBUY liu_w_Page_003thm.jpg
64edef2f035b1de006e07fb9e587824a
9675b4bad8fa3c3a4ab3b18471f89ac6e5806f88
2340 F20110409_AACASK liu_w_Page_041.txt
8c3d108f320bf80f82e75e9ed28a37ad
5acd4ee77cf25498cce974f0b29d93213d580420
2367 F20110409_AACARW liu_w_Page_027.txt
e4f49ead74958060e8db2f80731a1427
e0e9a9331eea7ab948cfdd1226ccf0fe00deab2a
2698 F20110409_AACCAT liu_w_Page_163thm.jpg
83d0d319a34a9fef7f03dd0b594e1ead
60659dc7fafa38f608b3a4ef9caa926e50bdaff1
9292 F20110409_AACBWC liu_w_Page_036thm.jpg
28512611440c44b61ab0a22bffd955b7
8bee477d19774cfb05c6cbd62155bcef91c9e8f2
8333 F20110409_AACBVN liu_w_Page_020thm.jpg
3c25ecac80a80e7f5f805bedb25a514f
82c8a7f2f2524dfaa56239dfac587e6fe3d614b3
6667 F20110409_AACBUZ liu_w_Page_004thm.jpg
50fdd874fd2299672793982377f38c97
c93fc644875f2c14733b661251ba7e15752aa17f
1093 F20110409_AACATA liu_w_Page_060.txt
231627fa53303d2c9cedd852a79d7943
a25f19ad8cdc550a1596616540da07825f9f6562
1841 F20110409_AACASL liu_w_Page_042.txt
fd306d18e49c65b14d1f0c9975bc3fca
82761e58fcb6a37eaca33b8604f4a413b29a7f87
2376 F20110409_AACARX liu_w_Page_028.txt
046adecb67f2d8067240d2b8d226f30f
6594c5f3ce7165c2e7e46d6d0b91e5da48afc778
6006 F20110409_AACCAU liu_w_Page_164thm.jpg
c265d2e2acdb76c9a1cd207b38d73c6f
5856d69018cecfbb7d4cb910747fbaa54f3cbd17
7605 F20110409_AACBWD liu_w_Page_037thm.jpg
ec5fafc22d94e33c9315baf792dc722c
c6eb41bdb5065547b17798009bab1ea43e7f9a08
8974 F20110409_AACBVO liu_w_Page_021thm.jpg
990a506446428afb8d7632a596efbf4f
6535da1be30f8d2f082c77f35587ca858440dc96
2106 F20110409_AACATB liu_w_Page_061.txt
6918f450b49d810cce857025e4fa6513
05ce2837cc0d61e1fc0431afac682febe3cacdd3
2323 F20110409_AACASM liu_w_Page_043.txt
c8482716f81f3a953b012297760f75cf
8b3e6165be736a76b6440f3272299b33c9d8d49d
2338 F20110409_AACARY liu_w_Page_029.txt
27b4c9a1c7c678e3678d6a8840a6ea05
cab79f2928261258df7d009559b972f0eadf0233
1965 F20110409_AACCAV liu_w_Page_165thm.jpg
d7be016767b6c1258e7d1370b32aa186
4e08e41ea28066db6b2852a5889313372f7ff6ac
8132 F20110409_AACBWE liu_w_Page_038thm.jpg
e8137618d6b826c0b8e640577d6f4059
964b91dad27161a0fe94a52e85e995e72ea772c2
2453 F20110409_AACBVP liu_w_Page_022thm.jpg
1cafe819a2b45d4d5ffb07992f9a1d02
81ae36e0ee71d565b1545e774b52f2d10d05f2ca
2348 F20110409_AACATC liu_w_Page_062.txt
a0502a568490170a3ced041b695fc53a
34a4e220322e55fabd9585c4f50cabdbce9e53bc
1834 F20110409_AACASN liu_w_Page_044.txt
38501629dc2ae0b13b64febefa627479
ee8006e4e715607d41aadd2ab438979b3505a0b4
2002 F20110409_AACARZ liu_w_Page_030.txt
b748402fb140aff7a528bda1f372c137
3443894fb74d330c3f99a18f236552deb62491de
4983 F20110409_AACCAW liu_w_Page_166thm.jpg
539d5467b26a5240e5a7c3f3e730a41c
4ddc7b1c6b98d1819133feb9f8d099212169e861
7391 F20110409_AACBWF liu_w_Page_039thm.jpg
de77f538e39c2555f38c4a1ee6439b4c
62cdc4aa0233ab06f830bafcc83a07944ffcb2d0
7584 F20110409_AACBVQ liu_w_Page_023thm.jpg
496cf178147d9ef259ecaef82057096c
594d5b1adfcb75cbd3d378003f6131d8f293e2fe
2484 F20110409_AACATD liu_w_Page_065.txt
a7783ecd8b7188af9800197138fa9dbe
8be4db4f2ed16c37cda98beb6fd475fe2864f670
1851 F20110409_AACASO liu_w_Page_045.txt
40ac249a25ee828d9325050eaf95e72e
8bd8df7dbd5b8f178517e4ba272aa30fb0968750
189144 F20110409_AACCAX UFE0011611_00001.mets
7a7c40f5965a279b542ba54e790b3ed8
e38240b45dfe3df365e5dd9ec638056969973660
7865 F20110409_AACBWG liu_w_Page_040thm.jpg
a271a7c4a6680c06024aa46049c57c7a
b224c8da84340317b225581015bb2a7b9dc513e1
8517 F20110409_AACBVR liu_w_Page_024thm.jpg
2af22b0f56ee4d83ae0b60baa3041889
07ac8c4ae5e156125411a4f69b03802ec4730eca
2260 F20110409_AACATE liu_w_Page_066.txt
f4a78b5e9670093355e6954a72f657f9
0e0213618110c37ba92b7f3029e941f36e4446f0
2255 F20110409_AACASP liu_w_Page_048.txt
487ae96dd8566958d745c48ab0ac9362
9b4ed70b24875d3da8ae677d84e49f5557c2de63
8870 F20110409_AACBWH liu_w_Page_041thm.jpg
7a56cae639fb0c08ba0f075ce865b5aa
bbfc887f83f88c6624470e8a6e9ced414ebc22ac
9196 F20110409_AACBVS liu_w_Page_025thm.jpg
1f8d6cf22b0d2c1912912f1634bb66d5
d8dca1ac46f75b809c23b09c92bc8854cbd0f8a2
1690 F20110409_AACATF liu_w_Page_067.txt
3e1a9585b1a28c656122716a990b28e9
354ba0388046808d595e4f28360d60871c252553
2012 F20110409_AACASQ liu_w_Page_049.txt
051846b153c929ea95e5e9228f7614fb
680ef761ca3421f7cbb63ed6452edaee7f41926b
7922 F20110409_AACBWI liu_w_Page_042thm.jpg
a3c1cb13864e76987d4689d60f362876
d3a649a23b20b9ad1bc06b76516aab48a1be3d10
8397 F20110409_AACBVT liu_w_Page_026thm.jpg
ec9cd2e2cca94fbc3a9734911bf11715
6be4d3029876546d089be6880c697e36781745e9
F20110409_AACATG liu_w_Page_068.txt
dc2df7eca6f705c9ca07c815a12b63f5
3a28f2d0e43f1881ec86c3c7f2f615287c2fae7d
429 F20110409_AACASR liu_w_Page_050.txt
3618fe036ff8c7318c5b93699790ebb5
b1a81c9bcc4760a70fa77052818e25dc9af2a682
8521 F20110409_AACBVU liu_w_Page_027thm.jpg
85153968671b411c988b6c18836ad32b
129cb7ef0262c520c10ebc394cf91b8f29af2bbd
2311 F20110409_AACASS liu_w_Page_051.txt
0912457a60b756788758a0ea15f7a407
098909302d2c91e47ea1028bc9ad58de343dda78
6845 F20110409_AACBWJ liu_w_Page_044thm.jpg
6da7e39ca61898dd29ac3e4f76aafb7e
e2d9ec79319fa22a7270669031955f813c1b551b
9086 F20110409_AACBVV liu_w_Page_028thm.jpg
456fc88da9d702ef39273811252ecf24
efd2deed3271b53967dcbe0c826bb04c3da6b690
F20110409_AACATH liu_w_Page_069.txt
e6d867153a0fc7eb85df81caa4ac1d22
20e96d6d87b89cecda7d0085354f4ad5a457c610
2165 F20110409_AACAST liu_w_Page_052.txt
eb91b6f681b5139a8547640900aa8765
b669ab69bb306a48a3394a20b793f4fb9c1aa552
F20110409_AACBWK liu_w_Page_045thm.jpg
ae8cf639ec78c011592163c7768b4a40
f30f2db256c7e9c6387dd110b1d13f2a5c86aa90
8445 F20110409_AACBVW liu_w_Page_029thm.jpg
c367e475b28a0754caa213391349677e
739f36b635f6c7054cc7b68d909d63c5ce3afca6
1903 F20110409_AACATI liu_w_Page_070.txt
aba79fa04f027752b30e6c1c81ce31b6
a490f242d609db73e86db3a88c8d2c356280979e
2216 F20110409_AACASU liu_w_Page_053.txt
25c59d4a574c4cdb778d9ccf608ba7c6
fd5bff60a753bfe6bdcdb806d5da7325f8d8c394
8069 F20110409_AACBXA liu_w_Page_061thm.jpg
646d3df3142b359fe1769ab178c39634
b908c11c7eec93bcb325e34fb1bdd31129cae544
6172 F20110409_AACBWL liu_w_Page_046thm.jpg
b13426b1349bed20c83cd47ba02f9f92
bbd02f12f79fcb46df8f018e58c18cdd33dba947
7559 F20110409_AACBVX liu_w_Page_030thm.jpg
6fb41b9e7be67d393656b806ad4b286d
7f4477d6e7878c0f63b362b6ec5118c2c5e5fae7
1452 F20110409_AACATJ liu_w_Page_071.txt
5daaa3de00d2d426fbfdf2470787ac86
eb3311c50e3ff24ba9ce3ee467eb6a6a4a707fa5
2333 F20110409_AACASV liu_w_Page_054.txt
e33f331da450da9c09a9cbd540811c7e
f05534805d4982307a8c4842504eec3fe113ff87
8951 F20110409_AACBXB liu_w_Page_062thm.jpg
1153d6b7d9aa360e776f76bd36836358
b613874677eafb43dea5147efc120cbaee62e834
8860 F20110409_AACBWM liu_w_Page_047thm.jpg
3ea08c4c69e90176e0f24cb0eed868d6
0f1b951333df9a1dd4fe5cda0bedcccf89ddb156
8569 F20110409_AACBVY liu_w_Page_031thm.jpg
faf6b84b7b2718d342181b7cf3803eb9
1f5022c88178f9e71f541a8c6d6869f9abccd405
2183 F20110409_AACATK liu_w_Page_073.txt
7390cd25ed20c45793106926c8f295ff
3d29cf3a208731d24ab783218102a07801a792ca
756 F20110409_AACASW liu_w_Page_055.txt
a3714708d1da0da835e007e60594efbb
b7f15ab61340ce68a1319e6cb3061a1985ec47df



PAGE 1

CR OSS-LA YER DESIGN OF RESOURCE A W ARE PR O T OCOLS FOR HETER OGENEOUS WIRELESS AD HOC NETW ORKS By W ei Liu A DISSER T A TION PRESENTED T O THE GRADU A TE SCHOOL OF THE UNIVERSITY OF FLORID A IN P AR TIAL FULFILLMENT OF THE REQ UIREMENTS FOR THE DEGREE OF DOCT OR OF PHILOSOPHY UNIVERSITY OF FLORID A 2005

PAGE 2

Cop yright 2005 by W ei Liu

PAGE 3

T o my wife, Shu, and my f amily members in China.

PAGE 4

A CKNO WLEDGMENTS First and foremost, I w ould lik e to e xpress my sincere gratitude to my advisor Prof. Y uguang F ang, who leaded me to this e xciting research area, for his in v aluable advice, encouragement and moti v ation during these years of my Ph.D. studies. I also thank him for his philosophical advice on both my academic and nonacademic life. As a mentor he has helped me become more mature, scholastically and personally I also w ant to thank my committee members, Prof. Shig ang Chen, Prof. Sartaj K. Sahni, Prof. John. M. Shea, and Prof. Dapeng W u, for their attention and advices. I w ould lik e to e xtend my thanks to all my colleagues in W ireless Netw orks Laboratory (WINET) at Uni v ersity of Florida for pro viding me such a w arm f amily-lik e en vironment. The constructi v e discussions with them either indi vidually or together deserv e special ackno wledgments. Finally on a more personal le v el, I am particularly indebted to my wife Shu, for her enduring so man y late nights, and for countless sacrices made to me. Her lo v e, support and encouragement ha v e made it possible for me to carry out this w ork. I am also v ery grateful to all my f amily members in China for their endless encouragement and continued support in one w ay or the other This w ork is dedicated to all of them. This material is based upon w ork supported in part by the U.S. Of ce of Na v al Research and National Science F oundation. An y opinions, ndings, and conclusions or recommendations e xpressed in this material are those of the authors and do not necessarily reect the vie ws of the U.S. Of ce of Na v al Research and National Science F oundation. i v

PAGE 5

T ABLE OF CONTENTS pageA CKNO WLEDGMENTS. . . . . . . . . . . . . . . . i vLIST OF T ABLES. . . . . . . . . . . . . . . . . . viiiLIST OF FIGURES. . . . . . . . . . . . . . . . . . ixABSTRA CT. . . . . . . . . . . . . . . . . . . . xiCHAPTERS 1 INTR ODUCTION. . . . . . . . . . . . . . . . 11.1 W ireless Ad Hoc Netw orks. . . . . . . . . . . . 11.2 Current Research. . . . . . . . . . . . . . . 21.2.1 Reliable Communications. . . . . . . . . . . 21.2.2 Ener gy Conserv ation. . . . . . . . . . . . 31.2.3 Quality of Service. . . . . . . . . . . . . 41.2.4 Cross-Layer Design. . . . . . . . . . . . 51.2.5 Heterogeneity in W ireless Ad Hoc Netw orks. . . . . . 61.3 Our Research. . . . . . . . . . . . . . . . 71.4 Outline. . . . . . . . . . . . . . . . . . 82 A DEVICE-ENERGY -LO AD A W ARE RELA YING FRAMEW ORK IN HETER OGENEOUS WIRELESS AD HOC NETW ORKS. . . . 112.1 Introduction. . . . . . . . . . . . . . . . 112.2 Related W ork. . . . . . . . . . . . . . . . 122.3 Ov ervie w of DELAR. . . . . . . . . . . . . . 142.3.1 Problem Statement. . . . . . . . . . . . 152.3.2 Our Solution: DELAR. . . . . . . . . . . . 162.4 Design of DELAR. . . . . . . . . . . . . . . 202.4.1 P-nodes' Neighbor -Selection Criteria. . . . . . . . 202.4.2 Routing Component of DELAR. . . . . . . . . 222.4.3 Hybrid T ransmission Scheduling. . . . . . . . . 252.4.4 Asymmetric Media Access Control Protocol (A-MA C). . . 272.5 Multiple-P ack ets T ransmission in DELAR. . . . . . . . 302.5.1 Multiple-pack ets T ransmission. . . . . . . . . 312.5.2 Hierarchical Modulation. . . . . . . . . . . 322.6 Discussion. . . . . . . . . . . . . . . . . 362.6.1 The Existence of Backw ard P aths. . . . . . . . . 36 v

PAGE 6

2.6.2 DELAR and ZRP. . . . . . . . . . . . . 392.6.3 The Choice of m and n. . . . . . . . . . . 392.6.4 Benets of the T ime Di vision Scheduling. . . . . . . 402.7 Performance Ev aluation. . . . . . . . . . . . . 402.7.1 Simulation Setup. . . . . . . . . . . . . 402.7.2 Impact of The Number of P-nodes. . . . . . . . 422.7.3 Impact of Node Mobility. . . . . . . . . . . 442.7.4 Impact of T raf c Load. . . . . . . . . . . . 472.8 Summary. . . . . . . . . . . . . . . . . 473 RESOURCE A W ARE MO VEMENT IN HETER OGENEOUS MOBILE AD HOC NETW ORKS. . . . . . . . . . . . 493.1 Introduction. . . . . . . . . . . . . . . . 493.2 Related W ork. . . . . . . . . . . . . . . . 503.3 Problem F ormulation. . . . . . . . . . . . . . 513.3.1 General Mobility Model (GMM). . . . . . . . . 523.3.2 Resource A w are Mo v ement. . . . . . . . . . 543.4 W aterhunter Mo v ement. . . . . . . . . . . . . 563.4.1 Simplied W aterhunter Mo v ement Problem. . . . . . 563.4.2 RAM-DCLC Algorithm. . . . . . . . . . . 613.4.3 A Routing Delay Dif ferentiation Mechanism. . . . . . 653.4.4 Incorporate RAM into DELAR. . . . . . . . . 663.5 Performance Ev aluation. . . . . . . . . . . . . 663.5.1 Simulation Setup. . . . . . . . . . . . . 663.5.2 Simulation Results. . . . . . . . . . . . . 683.6 Summary. . . . . . . . . . . . . . . . . 714 SUPPOR T DIFFERENTIA TED SER VICES IN MOBILE AD HOC NETW ORKS724.1 Introduction. . . . . . . . . . . . . . . . 724.2 Related W ork. . . . . . . . . . . . . . . . 744.3 Moti v ation. . . . . . . . . . . . . . . . . 764.4 P ack et-length-based Channel Model. . . . . . . . . . 794.5 Courtesy Piggybacking. . . . . . . . . . . . . 824.5.1 System Assumptions. . . . . . . . . . . . 824.5.2 The Courtesy Piggybacking Scheme. . . . . . . . 834.5.3 Discussion. . . . . . . . . . . . . . . 894.6 Performance Analysis. . . . . . . . . . . . . . 944.7 Performance Ev aluation. . . . . . . . . . . . . 974.7.1 Simulation Setup. . . . . . . . . . . . . 974.7.2 Impact of Channel Characteristics. . . . . . . . 994.7.3 Impact of T raf c Load. . . . . . . . . . . 1004.7.4 Impact of Node Mobility. . . . . . . . . . . 1014.7.5 Impact of Piggybacking Rules. . . . . . . . . 1034.8 Summary. . . . . . . . . . . . . . . . . 104 vi

PAGE 7

5 R OB UST AND ENERGY -EFFICIENT D A T A DISSEMIN A TION IN WIRELESS SENSOR NETW ORKS. . . . . . . . . . 1065.1 Introduction. . . . . . . . . . . . . . . . 1065.2 Related W ork. . . . . . . . . . . . . . . . 1085.3 Modelling Sensor Netw orks as Supply Chains. . . . . . . 1115.3.1 Introduction to Supply Chains. . . . . . . . . . 1115.3.2 Ho w Could Supply Chains Help Us. . . . . . . . 1135.4 A Hybrid Data Dissemination Frame w ork for W ireless Sensor Netw orks. . . . . . . . . . . 1155.4.1 System Model. . . . . . . . . . . . . . 1155.4.2 Manuf acture Area. . . . . . . . . . . . . 1175.4.3 T ransportation Area. . . . . . . . . . . . 1195.4.4 W arehouse Area and Service Area. . . . . . . . 1235.4.5 Discussion. . . . . . . . . . . . . . . 1265.5 Performance Ev aluation. . . . . . . . . . . . . 1285.5.1 Methodology and Metrics. . . . . . . . . . . 1285.5.2 Simulation Results. . . . . . . . . . . . . 1315.6 Summary. . . . . . . . . . . . . . . . . 1376 CONCLUSIONS AND FUTURE W ORK. . . . . . . . . . 1386.1 Conclusions. . . . . . . . . . . . . . . . 1386.2 Future W ork. . . . . . . . . . . . . . . . 139REFERENCES. . . . . . . . . . . . . . . . . . . 141BIOGRAPHICAL SKETCH. . . . . . . . . . . . . . . 151 vii

PAGE 8

LIST OF T ABLES T able page3 RAM: resource a w are mo v ement. . . . . . . . . . . . 613 RAM-DCLC: a DCLC routing algorithm for the W aterhunter Mo v ement. 623 Ener gy consumption parameters.. . . . . . . . . . . . 674 Channel model statistics. . . . . . . . . . . . . . 975 Analogue between supply chains and wireless sensor netw orks. . . . 1145 Simulation conguration. . . . . . . . . . . . . . 130 viii

PAGE 9

LIST OF FIGURES Figure page2 The structure of a Super Frame.. . . . . . . . . . . . 182 An e xample of the neighbor determination process (m=4, n=2, T=3.).. . 222 The topology in homogeneous and heterogeneous cases.. . . . . . 232 A unidirectional link between A and B.. . . . . . . . . . 272 The A-MA C operation procedure.. . . . . . . . . . . 282 The multiple-pack ets transmission enhancement to DELAR.. . . . 302 General hierarchical 2/4-PSK constellation.. . . . . . . . . 322 Implement multiple-pack ets transmission with hierarchical modulation. . 332 The e xistence of backw ard path.. . . . . . . . . . . . 382 The a v erage number of nodes e xisting in the shaded area.. . . . . 382 Simulation results with dif ferent number of P-nodes.. . . . . . . 432 Simulation results with dif ferent maximum node speed.. . . . . . 452 Simulation results with dif ferent traf c load.. . . . . . . . . 463 Multiple paths between tw o consecuti v e stops.. . . . . . . . 553 An e x emplary complete graph.. . . . . . . . . . . . 583 An e x emplary resource-a w are mo v ement.. . . . . . . . . 593 Resource-a w are mo v ement (RAM) vs. random w aypoint mo v ement. A Bnode should consecuti v ely visit S 1 S 2 S 3 S 4 S 5 and the solid lines are labelled by the sequences the y were passed through.. . . . . . 643 Simulation results.. . . . . . . . . . . . . . . . 694 The Whittier T unnel scenario.. . . . . . . . . . . . . 774 The Gilbert-Elliott channel model.. . . . . . . . . . . 804 The optimal pack et length ( P L ) vs. SNR ( r ), h=128.. . . . . . 814 P ack et-Length-Based nite-state Mark o v channel model.. . . . . 81 ix

PAGE 10

4 A fragmentation e xample.. . . . . . . . . . . . . 834 The total o v erhead with P K max and F T .. . . . . . . . . . 844 Illustration of the courtesy piggybacking scheme.. . . . . . . 874 Three piggybacking cases.. . . . . . . . . . . . . 904 An alternati v e piggybacking method.. . . . . . . . . . . 944 The queue model for piggybacking.. . . . . . . . . . . 964 A v erage w aiting time.. . . . . . . . . . . . . . . 974 Simulation results with dif ferent channel settings.. . . . . . . 1004 Simulation results with dif ferent pack et arri v al rates.. . . . . . . 1014 Simulation results with dif ferent pause time.. . . . . . . . . 1025 An e x emplary supply chain.. . . . . . . . . . . . . 1115 Supply chain strate gies.. . . . . . . . . . . . . . 1125 A system architecture for habitat monitoring. . . . . . . . . 1155 The forw arding-decision-making process of nodes in the transportation area.1205 An e x emplary pack et structure.. . . . . . . . . . . . 1205 The routing process in the w arehouse area.. . . . . . . . . 1245 The simulated sensor eld.. . . . . . . . . . . . . 1295 Ev ent deli v ery ratio vs. pack et error rate.. . . . . . . . . . 1315 Normalized ener gy consumption vs. pack et error rate.. . . . . . 1315 A v erage e v ent end-to-end delay vs. pack et error rate.. . . . . . 1325 A v erage routing o v erhead vs. pack et error rate.. . . . . . . . 1325 Ev ent deli v ery ratio vs. pack et error rate.. . . . . . . . . . 1345 Ener gy consumption and e v ent end-to-end delay of RRP with dif ferent d. 135 x

PAGE 11

Abstract of Dissertation Presented to the Graduate School of the Uni v ersity of Florida in P artial Fulllment of the Requirements for the De gree of Doctor of Philosoph y CR OSS-LA YER DESIGN OF RESOURCE A W ARE PR O T OCOLS FOR HETER OGENEOUS WIRELESS AD HOC NETW ORKS By W ei Liu August 2005 Chair: Y uguang F ang Major Department: Electrical and Computer Engineering Their features of rapid establishment and self-or g anization ha v e rendered wireless ad hoc netw orks to be identied as an indispensable component to support ubiquitous communications. Ho we v er the viable deplo yment of such netw orks f aces man y challenges resulting from some innate unf a v orable features such as the constrained resources, for e xample, nite ener gy supply and limited bandwidth. In this dissertation we in v estig ate the cross-layer design of resource-a w are protocols that conserv e ener gy and support qualityof-service (QoS) in resource-constrained wireless ad hoc netw orks. W e rst study the issue of ener gy conserv ation in heterogenous wireless ad hoc netw orks, where most nodes are po wered by batteries with small capacity while some others are more po werful, for e xample, po wered by batteries with lar ge capacity W e propose a cross-layer designed de vice-ener gy-load a w are relaying frame w ork, called DELAR to capitalize such po werful nodes to conserv e ener gy Dif ferent from the pre vious w ork that addresses the ener gy conserv ation issue at one layer DELAR is a joint design of scheduling, po wer control and routing that can conserv e ener gy thus prolong the netw ork lifetime. W e further propose a resource-a w are mo v ement mechanism to utilize such po werful nodes xi

PAGE 12

for data relaying. Such a resource-a w are mo v ement mechanism pro vides a ne w w ay to conserv e ener gy in wireless ad hoc netw orks. W e then study the issue of ho w to support QoS in wireless ad hoc netw orks. W e propose a scheme called Courtesy Pig gybac king to utilize system dynamics including timev arying channel condition and stochastic traf c characteristics, to support dif ferentiated services in such netw orks. The basic idea of Courtesy Piggybacking is to let high priority traf c help the lo w priority traf c by sharing the unused residual bandwidth. As a result, the scheme impro v es the o v erall performance for dif ferent prioritized services. Finally we propose an ener gy ef cient, reliable and scalable data dissemination scheme for wireless sensor netw orks. Based on supply chain management methodology for each sensing task, the sensor eld is conceptually partitioned into se v eral functional areas, and then dif ferent routing schemes are applied in dif ferent areas. This scheme addresses ener gyef cienc y reliability and scalability at the same time. xii

PAGE 13

CHAPTER 1 INTR ODUCTION 1.1 W ireless Ad Hoc Netw orks A wireless ad hoc netw ork is a collection of autonomous communication de vices, sometimes called nodes, which can communicate with each other by forming a multi-hop radio netw ork and maintaining connecti vity in a decentralized manner There are tw o major types of wireless ad hoc netw orks that are of particular interest to academia, industry and go v ernment: mobile ad hoc netw orks (MANETs) and wireless sensor netw orks (WSNs). Both types of wireless ad hoc netw orks share se v eral typical features in common. W ireless ad hoc netw orks can be deplo yed quickly and on-demand. The y are self-or g anized and independent of infrastructure. Moreo v er each node in a wireless ad hoc netw ork functions as both a host and a router helping relay pack ets for other nodes. These f a v orable features mak e wireless ad hoc netw orks v ery attracti v e in military and ci vil applications for which x ed infrastructures are una v ailable or unreliable, yet f ast netw ork establishment and constant reconguration are required. As an instance of mobile ad hoc netw orks, after an earthquak e, a rescue team sets up a temporary wireless netw orks and shares information to coordinate the rescuing w ork. As an e xample of wireless sensor netw orks, a lar ge number of small sensor nodes are deplo yed in a eld to monitor the habitat beha vior of animals. Ho we v er there are some unf a v orable features impeding wireless ad hoc netw orks from widespread deplo yment. Due to the time-v arying and error -prone wireless channel, nodes ha v e to contend with the adv erse ef fects of radio communications, such as noise, f ading and interference. Since nodes may depart, join, or mo v e in the eld, or e v en perish due to ener gy depletion, the netw ork topology is in general dynamic and may change rapidly and unpredictable o v er time. In such hostile, dynamic en vironment, reliable communications are of great importance. Additionally the wireless channel has limited bandwidth. 1

PAGE 14

2 Therefore, it is a challenging task to support heterogeneous traf c with v arious qualityof-service requirements. Moreo v er most nodes, especially those small sensors deplo yed in wireless sensor netw orks, are battery-po wered thus with limited lifetime. These nodes w ould become useless once depleting the batteries. Some adv erse consequences of such node diminution include de gradation of netw ork performance and unf a v orable netw ork partition. Thus, ener gy conserv ation, that is, ho w to e xpend the ener gy resources more frug ally and more e v enly so as to prolong the netw ork lifetime, becomes a crucial issue for MANETs. The broadcast nature of wireless communications also mak es it easy for enemies to snoop the ongoing transmissions, thus communications in wireless ad hoc netw orks ha v e security hazard. Depending on the applications running in such netw orks, the netw orks range from small-scale netw orks with a small number of nodes to lar ge-scale netw orks with thousands e v en millions of nodes, for e xample, lar ge-scale wireless sensor netw orks. The abo v e unf a v orable f actors, especially v ariable wireless links, topology changes, limited bandwidth, and constrained ener gy supply mak e the protocol design a tremendous challenge. P articulary when heterogenous applications are nding their niches in wireless ad hoc netw orks, specially designed protocols that can o v ercome the abo v e adv erse f actors are desired. Such protocols, scalable themselv es, should pro vide r eliable QoS-supporting ener gy-ef cient and secur e communications for those e ye-catching applications. In the follo wing we gi v e a nutshell of the current research for the issues of reliable communications, ener gy conserv ation and QoS pro visioning. 1.2 Current Research 1.2.1 Reliable Communications The primary goal of protocol design for wireless ad hoc netw orks is to support reliable communications which is the basic requirement for an y applications, and all the other solutions for the issues of ener gy conserv ation and quality of service should be based on reliable communications. In such hostile and dynamic settings, pack et losses may be caused

PAGE 15

3 for v arious reasons including transmission errors, collisions, or route changes due to mobility Thus, there are man y w orks addressing reliable communications at dif ferent layers. F or e xample, at the ph ysical layer modulation and error coding schemes are proposed to combat the error -prone feature of wireless communications [1]. In order to reduce the interference and transmission collisions, at the MA C layer there are man y MA C protocols that are de vised to to ef ciently control the channel access [2]. These protocols can be classied into three cate gories: contention-based (for e xample, MA CA W [3], F AMA [4] and IEEE 802.11 [5]), contention-free (for e xample, TDMA, FDMA and CDMA) and h ybrid. The most f amous one is IEEE 802.11, and it uses CSMA/CA and four w ay handshak e (R TS/CTS/D A T A/A CK) to coordinate the channel access and pro vide reliable communications [5]. At the netw ork layer man y routing protocols ha v e been proposed to deal with the dynamic topology and support reliable communications [6]. The y can be proacti v e in that the y determine routes independent of traf c patterns, for e xample, DSD V [7] and OLSR [8]; the y can be reacti v e in that the y maintain routes on demand, such as DSR [9] and A OD V [10]; the y e v en can be h ybrid of the abo v e tw o, for e xample, ZRP [11] and LANMAR [12]. Though these schemes function at dif ferent layers, the y all w ork for the same goal: supporting reliable communications in wireless ad hoc netw orks. 1.2.2 Ener gy Conserv ation Ho w to lengthen the lifetime of communication de vices is a crucial issue for the wireless ad hoc netw orks and man y po wer conserv ation techniques are applied when hardw are and softw are including the protocol stack are specically designed. Man y ef forts ha v e been made at the ph ysical layer due to the f actor that the system hardw are is the primary consumption object in a mobile de vice. T w o dif ferent perspecti v es are commonly adopted to approach the problem. Though no breakthrough has been e xperienced in the past 30 years, the direct w ay is to increase the battery capacity while k eeping the weight of the battery tolerati v e. Another w ay is, hard yet attainable, to apply lo w-po wer techniques to decrease the ener gy dissipation when wireless terminals are designed. Considerable ener gy sa vings

PAGE 16

4 are resulted from the hardw are design; ho we v er it is pertinent to e xplore other v enues to further impro v e the ener gy ef cienc y that is, to design the protocol stack with ener gy ef cienc y in mind [13]. F or e xample, at the data-link layer transmission po wer control can be used to reduce the interference. Po wer -sa ving mode also can be used to conserv e ener gy by putting the netw ork interf ace into the sleep mode when no communication is needed. At the netw ork layer po wer a w are routing is used to nd ener gy-ef cient paths and route pack ets o v er the ener gy-ef cient paths to sa v e ener gy At the application layer ener gyef cient OS/middle w are and applications (for e xample, po wer -a w are video processing) are proposed to sa v e ener gy Because the k e y to ener gy conserv ation in wireless ad hoc netw orks lies within the higher le v els of the wireless protocol stack [13], it w ould be benecial if these multiple layers can jointly function to conserv e ener gy 1.2.3 Quality of Service It is hard to agree on a common denition of QoS, b ut a QoS enabled netw ork shall ensure that its applications and/or their users ha v e their QoS parameters fullled, while at the same time ensuring an ef cient resource usage, for e xample the bandwidth, and also ensure that the most important traf c still has its QoS parameters fullled during netw ork o v erload. The most important QoS parameters may include: throughput, a v ailability delay jitter and pack et loss. The netw ork can serv e the applications with hard-QoS, soft-QoS or adapti v e QoS [2,14]. Se v eral prominent service models or o v erall architectural frame w orks, within which certain types of services can be pro vided in the netw ork, are suggested for ad hoc netw orks, such as e xible quality of service model for MANETs (FQMM) [15] and stateless model for wireless ad-hoc netw orks (SW AN) [16]. Since the resource is limited and dif ferent traf c has v ery dif ferent quality of service requirements, usually Dif ferentiated Services (Dif fServ) architecture [17] is used in which traf c is classied into dif ferent classes or priority le v els, and the wireless netw orks serv e dif ferent classes of traf c with dif ferent processing rules or policies. There are se v eral mechanisms proposed for supporting QoS [14], such as QoS routing, scheduling, admission control,

PAGE 17

5 resource reserv ation, and signaling techniques. These mechanisms can coe xist in a netw ork. F or e xample, QoS routing is used to nd a path satisfying certain QoS requirements, then signalling techniques and resource reserv ation are used to reserv e resources along the found path. W e belie v e that in order to ef ciently utilize the resources and support QoS, v arious mechanisms implemented at dif ferent layers should w ork together 1.2.4 Cross-Layer Design As a matter of f act, the issues of ener gy conserv ation and QoS pro visioning are direct results of constrained resources: ener gy supply and bandwidth. If nodes and netw orks are endo wed with plenty of bandwidth and ener gy supply for e xample, unbounded frequenc y band and hea vy duty po wer supply with light wight, these tw o issues may ne v er come to the fore. Since the tough reality of constrained resources is there and the chance to en vision breakthrough for these tw o resources is rather slim, the natural response to these tw o issues is to check: Do all the current protocols proposed for wireless ad hoc netw orks mak e full use of such precious resources? The answer to the question is ne g ati v e. There do ha v e lots of resources thar are w asted without an y yield, o wing to lo w ef cienc y or their o v erlook of these protocols. F or e xample, in netw orks with contention-based medium access control mechanisms, lots of ener gy and bandwidth are w asted due to collisions. One of the reasons for resource underutilization is that v ast majority of the protocols are based on traditional layered-design principle, which lacks the interaction and information sharing among multiple layers. The lack of interaction causes lo w ef cienc y of those protocols. On one hand, man y resources at one layer are w asted due to undesirable decisions made at another layer F or e xample, poor routing decisions cause congestion and collisions at the ph ysical layer thus w aste precious resources, for e xample, bandwidth and ener gy On the other hand, man y a v ailable resources are ne glected by those protocols. F or e xample, time-v arying channel condition leads to time-v arying channel capacity ho we v er due to the lack of interaction between the MA C layer and the ph ysical layer man y MA C protocols cannot recognize such a resource b ut ne glect it. Hence, in recent years a ne w

PAGE 18

6 cr oss-layer design philosoph y emer ges [18] [19]. In contrast to the layered-design philosoph y cross-layer design philosoph y stresses on the information sharing and interaction among multiple layers. Cross-layer design can be performed in tw o w ays [20]. In the rst w ay parameters or information in other protocol layers are considered to impro v e the performance of a protocol layer The rst w ay of practice does not totally abandon the transparenc y between protocol layers. In contrast, the second w ay blurs this transparenc y by mer ging se v eral layers into one component to achie v e much better performance. The cross-layer design philosoph y attempts to mak e use of the inter -layer coupling to use the resources more ef ciently W ith this philosoph y the decision for resource utilization can be made more advisably based on the information from multiple layers. Though the cross-layer design may mak e the design issue more complicated than the layered design, the appealing benets mak e cross-layer design a strong candidate as the design principle in resource-constrained wireless ad hoc netw orks. 1.2.5 Heterogeneity in W ireless Ad Hoc Netw orks Most of the pre vious research assumes that netw orks are homogeneous or nodes are in fully symmetrical en vironment. F or e xample, all nodes ha v e identical capabilities and responsibilities. In reality ho we v er the heterogeneity of mobile de vices seems to be inher ent and has been commonly observ ed in MANETs [21]. Strictly speaking, heterogeneity in a netw ork usually means the de vice heter o g eneity re grading the ph ysical hardw are or softw are nodes use. Nodes in the same netw ork may dif fer in their communication, processing capabilities, which can be reected by CPU speed, memory size, battery life, transmission range and radio, netw ork interf ace, security le v el, operating system or protocol stack, and so on. Nodes in the same netw ork may dif fer in their responsibilities as well. The y may undertak e dif ferent roles, for e xample, routers, serv ers, cluster heads, and authentication centers, thus operate with dif ferent policies. Owing to such dif ferences in capabilities and responsibilities, along with the v arious applications running on each indi vidual node, nodes may also ha v e dif ferent traf c characteristics and mobility characteristics. F or e xample,

PAGE 19

7 nodes may ha v e dif ferent bit rate, timeliness constraints, reliability requirements, speed, direction, and predictable/unpredictable mo v ement patterns for each indi vidual application. Broadly speaking, heterogeneity in a netw ork also means that nodes may e xperience heterogeneous en vironment characterized by time-v arying channel and netw ork conditions. Such en vir onment heter o g eneity may ha v e dif ferent properties in dif ferent settings. F or e xample, dif ferent nodes at dif ferent locations and at dif ferent time may percei v e dif ferent channel conditions. The heterogeneity either de vice heterogeneity or en vironment heterogeneity coupling with the aforementioned unf a v orable f actors, mak es the protocol design much more complicated. Ho we v er such heterogeneity also introduces opportunities to design more efcient protocols. In f act, some types of heterogeneity are resource-related when the y themselv es are resources or can be mapped into certain type(s) of resources. Moreo v er de vice heterogeneity and en vironment heterogeneity can jointly af fect the usage of constrained resources. F or e xample, nodes with less a v ailable battery reserv e w ould refrain from transmissions in poor channel condition. Thus, to recognize the inherent heterogeneity and further to ef ciently utilize the resource-related heterogeneity to o v ercome the impact stemming from those unf a v orable features, for e xample, limited ener gy supply and constrained bandwidth, is a rather challenging task. In this dissertation, we mak e our efforts along this line to address the issues of ener gy conserv ation and QoS pro visioning in resource-constrained wireless ad hoc netw orks. 1.3 Our Research As mentioned abo v e, there is a big v olume of w ork studying ho w to support reliable communications, from the ph ysical layer to the data link layer and to the routing layer Ne v ertheless, the research on ener gy conserv ation and QoS pro visioning is still thin on the ground. Either of the tw o issues is quite interesting, ho we v er on top of the pre vious research on reliability it is pretty hard, if not impossible, to design a one-size-t-all protocol that can simultaneously resolv e these tw o issues. Thus, in this dissertation, based on

PAGE 20

8 the pre vious w ork for reliable communications, we attempt to study these tw o issues and resolv e them separately b ut bear in mind the possibility of future inte gration of these tw o issues. The cross-layer design philosoph y serv es as our design principle for its potential in better utilizing the a v ailable resources and pro viding desired ener gy ef cienc y and QoS pro visioning to applications. More important, the cross-layer design philosoph y will help us recognize the inherent heterogeneity in wireless ad hoc netw orks, and further help us utilize it to address the issues we are interested in. More specically we utilize the heterogeneity in terms of po wer supply or battery life to address the issues of ener gy conserv ation. Such heterogeneity is used to design more ef cient routing, po wer control and transmission scheduling. W e also utilize the time-v arying channel condition and stochastic traf c char acteristics to address issues associated with QoS pro visioning in wireless ad hoc netw orks. F or wireless sensor netw orks, we particularly recognize the dif ferent roles undertak en by each indi vidual sensor node, and utilize such heterogeneity to design more ef cient communication protocols. 1.4 Outline This dissertation is or g anized as follo ws. Chapter2concerns the issue of ener gy conserv ation in heterogeneous wireless ad hoc netw orks in terms of po wer supply In such netw orks, most nodes called B-nodes are battery-po wered, and some nodes called P-nodes ha v e relati v ely unlimited ener gy supply such as solar cell or dynamos. W e propose a de vice-ener gy-load a w are relaying frame w ork, called DELAR to capitalize such P-nodes to conserv e ener gy The basic idea of DELAR is to di vide time into Super F r ames each of which is composed by se v eral periods, and to enable P-nodes to transmit with dif ferent po wer in dif ferent periods. In this frame w ork, de vice heterogeneity nodal residual ener gy and local load status are mapped into a routing cost metric and incorporated into routing protocols to nd ener gy-ef cient paths. W e also

PAGE 21

9 propose a h ybrid medium access control mechanism to schedule the transmission acti vities of P-node and B-nodes. Moreo v er we introduce the concept of mini-routing into the MA C layer and propose an Asymmetric MA C (A-MA C) to support reliable communications o v er unidirectional links resulted from the asymmetric po wer capabilities between P-node and B-nodes. Further we propose a multiple-pack ets transmission scheme as an enhancement to impro v e the o v erall performance of DELAR. Chapter3discusses a resource-a w are mo v ement scheme that can be incorporated into DELAR. In this scheme, a node is enabled to determine its mo v ement characteristic (speed, direction, and so on) based on the netw ork en vironment and its o wn residual resources. Specically such kinds of mo v ement can be mapped into a path-constrained path-optimization problem. A heuristic algorithm is proposed to solv e this NP-complete problem. W ith this resource-a w are mo v ement scheme in place, B-nodes ha v e more oppor tunities to utilize P-nodes and thus conserv e ener gy and prolong the netw ork lifetime. Chapter4focuses on the issue of QoS pro visioning. W e consider heterogeneous mobile as hoc netw orks on ho w to utilize system dynamics, a combination of en vironment and de vice heterogeneity including time-v arying channel condition and stochastic traf c char acteristics. A scheme called Courtesy Pig gybac king is proposed to support dif ferentiated services in mobile ad hoc netw orks. The basic idea is to let the high priority traf c help the lo w priority traf c by sharing the unused residual bandwidth with courtesy Courtesy Piggybacking is able to utilize such system dynamics to impro v e the o v erall system per formance for dif ferent prioritized services: impro v ed pack et deli v ery ratio and shortened end-to-end delay More important, the piggybacking scheme can be readily incorporated into the DELAR frame w ork. In Charter5, we particularly address the issues of ener gy ef cienc y and reliability in wireless sensor netw orks, a special type of wireless ad hoc netw orks. W e introduce the concept of supply chain into wireless sensor netw orks and model wireless sensor netw orks as supply chains. Based on supply chain management strate gies, we propose an ener gy

PAGE 22

10 ef cient, reliable and scalable data dissemination scheme. Basically for each sensing task, the sensor eld is conceptually partitioned into se v eral functional areas, and dif ferent routing schemes are applied in dif ferent areas. Such partition strate gy h ybrid strate gy and cooperation strate gy help us address the issues of ener gy conserv ation and reliable communications at the same time. Chapter6concludes this dissertation and outlines the future research.

PAGE 23

CHAPTER 2 A DEVICE-ENERGY -LO AD A W ARE RELA YING FRAMEW ORK IN HETER OGENEOUS WIRELESS AD HOC NETW ORKS 2.1 Introduction In this chapter we focus on heterogeneous wireless ad hoc netw orks, where most nodes, denoted as B-nodes are equipped with limited po wer sources lik e batteries, while some other nodes, denoted as P-nodes ha v e relati v ely unlimited po wer supplies, for e xample, po wer sca v enging units such as solar cells, or dynamos when the y are installed in mobile v ehicles, and so on. Our goal is to de v elop more ener gy conscious protocols by taking adv antage of the heterogeneity of mobile de vices, that is, being generous in using the P-nodes while conserv ati v e in using the B-nodes. More specically the contrib utions are mainly fourfold. First, follo wing the cross-layer protocol design philosoph y we propose a De vice-Ener gy-Load A w are Relaying frame w ork, named DELAR, to achie v e ener gy conserv ation by utilizing the inherent heterogeneity of nodal po wer capabilities. Second, we design a h ybrid transmission scheduling scheme, which is a combination of reserv ationbased and contention-based medium access control schemes, to coordinate the transmission acti vities among P-nodes and B-nodes. Such scheduling can mak e full use of po werful nodes while reducing the interference and collisions. Third, we de v elop the mini-routing and Asymmetric MA C ( A-MA C ) protocols to support the MA C layer ackno wledgements o v er unidirectional links resulting from the asymmetric transmission po wer le v els at Pnodes and B-nodes. T o the best of our kno wledge, this is the rst ef fort to address this issue at the MA C layer Last, we present a multiple-pack ets transmission technique to fur ther impro v e the delay performance. Detailed simulation studies are carried out to justify the ef fecti v eness and ef cienc y of the proposed frame w ork. 11

PAGE 24

12 The proposed DELAR can serv e as a general frame w ork where v arious ener gy conserv ation techniques such as po wer sa ving modes, transmission po wer control and po wer a w are routing can be inte grated to jointly achie v e better ener gy conserv ation. In addition, it also of fers a platform to study other challenging issues, for e xample, quality of service (QoS) pro visioning and security support. F or instance, P-nodes can act as distrib uted admission controllers to coordinate the access to limited netw ork resources such as a v ailable bandwidth. As another e xample, in security-sensiti v e MANET applications, P-nodes can help B-nodes perform resource-hungry public-k e y operations. As f ar as we kno w no similar frame w ork has appeared else where in the literature. In the rest of this chapter we start with the re vie w of related w ork. W e then introduce the system model and the o v erall frame w ork of DELAR in Section2.3. In Section2.4, we elaborate the netw ork layer components of DELAR and a h ybrid transmission scheduling scheme, and present a no v el Asymmetric MA C protocol called A-MA C follo wed by a multiple-pack et transmission scheme to further impro v e the performance of DELAR. In Section2.7, we e v aluate the performance of DELAR through simulations. Finally we summarize this chapter in Section2.8. 2.2 Related W ork Recent years ha v e seen a gro wing body of research concerned with ener gy conserv ation in wireless ad hoc netw orks, among which man y ef forts ha v e been made at the ph ysical layer to impro v e the hardw are design of mobile de vices [13]. Though important, it is still pertinent to e xplore other v enues to further ameliorate the ener gy ef cienc y of mobile devices. F or the lack of space, here we only re vie w w orks performed at the MA C or netw ork layer of ad hoc netw orks, which are closely related to our study in this chapter Based on the mechanisms used, the solutions seen in the literature can be roughly classied into three cate gories: Po wer -Sa ving Modes (PSM), T ransmission Po wer Control (TPC) [22], and Po wer -A w are Routing (P AR).

PAGE 25

13 PSM is usually implemented at the MA C layer and the basic idea is to put the netw ork interf ace into the sleep mode when no communication is needed. One fundamental issue in PSM is when to enter the sleep mode and for ho w long to stay in this mode. Some w ork along this line includes P AMAS [23] and S-MA C [24]. In addition, Tseng et al. proposed three asynchronous protocols, namely Dominating-A w ak e-Interv al, Quorum-based, and Periodical-Fully-A w ak e-Interv al protocols [25]. These proposals stri v e to ef ciently and intelligently control nodes' sleep and w ak e schedules and at the same time deal with such f actors as clock synchronization, neighbor disco v ery and netw ork partitions which are inherent in multi-hop ad hoc netw orks [25]. TPC adapts the transmission po wer to the propag ation and interference characteristics e xperienced by the link [26,27,28,29,30,31,32,33,34,35]. TPC sometimes is called topology control when it attempts to control the transmission po wer or e v en turn of f the radio so that a desirable topology or connecti vity can be maintained for sa ving ener gy Man y proposals in this cate gory are concerned with maintaining a dominant set of nodes or forming some virtual backbone with certain clustering mechanisms. In addition, PCM [36] w as proposed to use dif ferent transmission po wer le v els for R TS/CTS and D A T A/A CK frames on a per -pack et basis. In contrast to TPC protocols aiming at making each link as ener gy-ef cient as possible, a P AR protocol determines which of these links to be used for end-to-end paths so that additional ener gy sa vings can be obtained by routing pack ets o v er ener gy-ef cient paths. Singh et al. proposed v e po wer -a w are routing metrics that could be incorporated into routing protocols to reduce the per -pack et ener gy consumption, to prolong the per -node lifetime, or to prolong the o v erall netw ork lifetime [37]. Besides, Chang and T assiulas [38] proposed a routing cost metric that is a combination of the transmission po wer le v el and the residual ener gy T oh [39] studied both the minimum ener gy and the netw ork lifetime issues and proposed a conditional min-max battery capacity routing protocol which tries to strik e a balance between these tw o competing objecti v es. While the link quality has been

PAGE 26

14 suggested as a routing metric to reduce queuing delays and data loss rates, Banerjee and Misra proposed a routing cost formulation that considers not only the residual po wer b ut also the link characteristics, for e xample, the cost for potential retransmissions, to capture the ef fects of both the link distance and the link error rate [40]. As an addition, P AR O [41] is another notable approach designed for scenarios where nodes can dynamically adjust their transmission po wer In P AR O, a candidate intermediate node monitors an ongoing direct communication between tw o nodes and inserts itself in the forw arding path if its action can lead to some ener gy sa vings. Most pre vious proposals do not tak e into account de vice heterogeneity inherent in MANETs to achie v e better ener gy conserv ation. Ho w to tak e full adv antage of P-nodes to prolong the netw ork lifetime as much as possible has been pre viously addressed in [21,42,43]. Ho we v er all of them focus on the netw ork layer and man y challenging issues relating to the MA C layer are either left untouched or o v erlook ed. F or e xample, none of them consider ho w to support the MA C-layer ackno wledgements o v er unidirectional links caused by dif ferent transmission po wer at P-nodes and B-nodes (cf. Section2.4.4). By contrast, our DELAR addresses the ener gy conserv ation from multiple f acets, including routing, transmission scheduling and po wer control. In brief, although all the aforementioned proposals can achie v e certain le v el of ener gy conserv ation, ho w to design a comprehensi v e, practical frame w ork that not only inte grates PSM, TPC, P AR, and transmission scheduling, b ut also mak es full use of inherent de vice heterogeneity remains an open, challenging problem. Our DELAR frame w ork is proposed to address this crucial issue. 2.3 Ov ervie w of DELAR In this section, we rst mak e a brief introduction to the problems we intend to address. W e then pro vide a high-le v el o v ervie w of our solution, DELAR which stands for De viceEner gy-Load A w are Relaying frame w ork for heterogeneous ad hoc netw orks.

PAGE 27

15 2.3.1 Problem Statement In this chapter we focus on heterogeneous ad hoc netw orks comprised of mobile nodes with dif ferent ener gy supplies, though heterogeneity may ha v e other meanings in dif ferent settings. W e assume that besides a majority of battery-po wered B-nodes, there e xist some po werful P-nodes ha ving relati v ely unlimited ener gy supplies (cf. Section2.1). Our objecti v e is to de v elop ener gy conserv ation protocols by utilizing such heterogeneity in ener gy resources. Intuiti v ely speaking, since P-nodes are of relati v ely innite ener gy reserv oir as opposed to the B-nodes with usually irreplaceable batteries, data communications should try to utilize these P-nodes as much as possible in order to prolong the netw ork lifetime. Therefore, on the one hand, a pack et should be forw arded to a P-node whene v er an ener gy sa ving can be e xpected. On the other hand, communications in the netw orks should a v oid using B-nodes if possible. F or these purposes, it is desirable to allo w P-nodes to ha v e higher transmission po wer so as to co v er a lar ger transmission area, which can statistically reduce the number of B-nodes in v olv ed in pack et forw arding. Ho we v er this straightforw ard proposal may pose lots of challenges to the protocol design Ho w can a B-node be a w are of the e xistence of P-nodes in its vicinity? If there e xist multiple paths through P-nodes to the destination, which path should be chosen? Should the transmission range of a P-node be as lar ge as possible, or be k ept within certain optimal ranges? Ho w can the protocol support reliable communications o v er the unidirectional links caused by asymmetric transmission po wer at P-nodes and B-nodes along with error -prone and time-v arying wireless channels? In addition, higher transmission po wer often implies more reachable neighbors, decreased spatial reuse, and increased local contention for the shared wireless medium, then ho w can the protocol schedule the transmission acti vities so that a good balance can be struck between ener gy sa vings and other system performance f actors such as pack et deli v ery ratio and end-to-end delay? These are all non-tri vial questions and need to be answered before we can indeed mak e full use of the aforementioned heterogeneity in ad hoc netw orks.

PAGE 28

16 After a careful in v estig ation on these interw o v en issues, we belie v e that the y are closely related to routing, transmission scheduling, and po wer control. F or e xample, with the adjustment of the transmission po wer of P-nodes and B-nodes, the topology and neighborhood, and thus the routing information, w ould change accordingly So w ould the schedule of transmission acti vities. Moreo v er there e xists a strong interaction between routing and MA C layers. Apparently our original design objecti v e can be boiled do wn to designing a joint routing, scheduling, and po wer control scheme, which should be addressed across the whole protocol stack, especially at the routing and MA C layers [44][45]. T o achie v e this, a cross-layer designed frame w ork is demanded. In this frame w ork, po wer control should be implemented to optimize the transmission po wer of each node (both P-nodes and B-nodes) to achie v e optimal ener gy utilization and maintain a reasonable netw ork topology; routing should be designed to inform all the nodes of the e xistence of P-nodes and nd the optimal ener gy-ef cient routes; and transmission scheduling should be able to adjust the transmission acti vities so that the ener gy spent on channel contentions and collisions can be minimized. In addition, an appropriate scheduling scheme should be capable of striking a good balance between ener gy ef cienc y and other system performance f actors such as end-to-end delay and pack et deli v ery ratio. 2.3.2 Our Solution: DELAR W e consider a mobile ad hoc netw ork consisting of N p P-nodes and N b B-nodes, where N p and N b are system design parameters. W e assume a single wireless broadcast channel shared by all the nodes, though our DELAR can be easily e xtended to multichannel cases. W e also adopt a simple po wer control scheme as follo ws. Each B-node transmits omni-directionally and can maintain a circular transmission range B T R (basic transmission range) before using up its battery which can be properly set to achie v e a good tradeof f between ener gy ef cienc y and netw ork connecti vity [46]. In addition, we postulate that P-nodes are able to adjust their transmission po wer so as to co v er lar ger areas than B-nodes if needed. Moreo v er all the P-nodes are assumed to ha v e identical maximum

PAGE 29

17 transmission range of P T R max = M B T R where M is a positi v e inte ger greater than 1. As re v ealed in [46], using common transmission po wer between the same type of nodes can ensure bidirectional links and thus the correct operations of e xisting routing and MA C protocols. W ith this simple yet ef cient po wer control scheme, a unidirectional link only e xists between a P-node and a B-node when the y use dif ferent transmission po wer instead of between an y tw o B-nodes or P-nodes1. According to [35], such simple po wer control is belie v ed to be more practical than other e xpensi v e transmission po wer control schemes, either making unrealistic assumptions or ha ving e xtra hardw are requirements. As mentioned before, DELAR arises from the follo wing intuition: the P-nodes should be utilized as much as possible. In other w ords, we should attempt to reduce the use of Bnodes if we cannot a v oid using them at all. Thus it is adv antageous to enable a P-node to directly communicate with other P-nodes nearby or in distance by using higher transmission po wer so that the number of B-nodes in v olv ed in the data forw arding can be much reduced. Ho we v er higher transmission po wer or lar ger transmission co v erage usually implies more neighbors and increased local contention for the shared wireless channel. Therefore, instead of granting these P-nodes unlimited pri vile ges of reaching an y other node at an y time at will, it mak es more sense to constrain P-nodes' transmission po wer under certain bound and to limit their transmission acti vities within some pre-planned periods in order to reduce the collisions with other ongoing transmissions and thus maintain good channel utilization. In order to better schedule the transmissions of P-nodes and B-nodes, we adopt a time-di vision multiple xing method. W e di vide time into equal length time slots called Superfr ames In each of the superframes, some time periods are e xclusi v ely designated to P-nodes, while the rest are shared by all P-nodes and B-nodes in the netw ork. More 1 In this chapter we only consider asymmetric transmission po wer as the primary cause for unidirectional links and omit others such as v arious collision/noise/interference le v els at dif ferent nodes.

PAGE 30

18 P n o d e 0 P n o d e 1 . P n o d e k P t o B p e r i o d 0 T i m e F r a m e D u r a t i o n P t o B p e r i o d 1 P t o B p e r i o d k P t o P p e r i o d B t o B p e r i o d m i n i s l o t Figure 2: The structure of a Super Frame. specically during one c ycle of the Superfr ame (see Fig.2), there is a P-to-P period with length t pp in which only P-nodes are allo wed to communicate with each other by using transmission range T R pp = m B T R (1 < m M ) while all B-nodes just k eep silent, as if the netw ork were merely formed by these mobile core P-nodes. Additionally in one Superfr ame each P-node has its o wn e xclusi v e period called P-to-B period with equal length t pb in which it can boost its transmission po wer to co v er a range of T R pb = n B T R (1 < n M )2. The rest of one Superfr ame is called B-to-B period with length t bb in which all the nodes in the netw ork can contend for the channel and initiate transmissions to w ards nodes in their T R bb = B T R Ob viously all the P-nodes should act as common B-nodes in the B-to-B period by adjusting their transmission range back to T R bb Notice that during one P-to-B period, since the P-node o wning this period and the B-nodes it intends to communicate with ha v e dif ferent transmission po wer unidirectional links between them may be formed. Therefore, in contrast to the P-to-P and B-to-B periods where some common contention-based MA C protocols such as the IEEE 802.11 can be used, the P-to-B period(s) demands an enhanced MA C protocol to support reliable communications o v er unidirectional links. Our Asymmetric MA C protocol A-MA C is e xactly de v eloped for this purpose. Such rendezv ous of reserv ation-based and contention-based MA C schemes is able to schedule the pack et transmission more ef ciently which we will see shortly 2 T o pro vide reliable communications during P-to-B periods, usually n is less than m .

PAGE 31

19 In DELAR, the heterogeneity of mobile nodes are also incorporated into the construction of routing tables. Routes are disco v ered based on routing metrics which tak e consideration of the residual ener gy and the load status of mobile nodes. Once generating a data pack et, a node looks up its routing table and sends the data pack et to the ne xt hop as it does in common ad hoc routing protocols. If residing in the forw arding path and ha ving recei v ed a forw arding request, a node will forw ard the data in an appropriate time period to the ne xt hop according to its o wn routing table. More specically for a B-node, when the ne xt hop is in its T R bb range, it can only forw ard the data pack et during the B-to-B period. While for a P-node, if the ne xt hop is another P-node located in this P-node' s T R pp the P-node can forw ard the pack et to the ne xt hop in the P-to-P period; if the ne xt hop is a B-node located inside its T R pb the P-node can forw ard the pack et to the ne xt hop in its e xclusi v e P-to-B period. In summary with such time-di vision scheduling and a de vice-ener gy-load a w are routing metric, we intend to utilize P-nodes as much as possible in an ef cient and cautious w ay while reducing collisions and interference to an acceptable le v el, so that we can achie v e the e xpected ener gy conserv ation without harming other system performance f actors. Se v eral research challenges remain in supporting the seemingly simple operations of DELAR as described abo v e. Gi v en a B-node (P-node) X located in a P-node P 0 s transmission range T R pb ( T R pp ), for instance, what criteria should P adopt to determine if X is a neighbor (in one hop range) or not, that is, forw arding a pack et to this node X in a one-hop manner or a multiple-hop manner? What kind of routing metric should be adopted to reect the heterogeneity in de vice types, nodal residual ener gy and local load status when setting up routing paths? Ho w can we di vide time into Superfr ames and ho w can one P-node re gister a P-to-B period without conicting with others' P-to-B periods? Ho w does node X send MA C layer ackno wledgements back to P in the presence of unidirectional links resulting from the asymmetry in transmission po wer? The remainder of this chapter will address these questions one by one in more detail.

PAGE 32

20 2.4 Design of DELAR In this section, we will rst discuss the neighbor -selection criteria of P-nodes. W e then illustrate the routing component of DELAR. Ne xt, we introduce in more detail the h ybrid transmission scheduling of DELAR. Then, we present the Asymmetric Media Access Control Protocol (A-MA C) and the multiple-pack et transmission scheme. Last we gi v e some further discussions. 2.4.1 P-nodes' Neighbor -Selection Criteria In the literature, tw o nodes are usually considered as neighboring nodes of each other when the y are one hop a w ay and the y can directly communicate with each other Ho we v er in heterogeneous netw orks, we ha v e to change the criteria to cope with the e xistence of P-nodes whose transmission ranges are much lar ger than those of B-nodes. In this case, an y node in a P-node' s T R pb could be a neighbor candidate of the P-node3. Ne v ertheless, in order to support the MA C layer ackno wledgements, not all the candidates can be nally chosen as neighbor s or be ne xt hops in the routing table. Before presenting the rules that guide P-nodes to mak e selection decisions, we rst introduce the notions of F orwar d P ath and Bac kwar d P ath F or an y node pair s and t a F orwar d P ath indicates the path deri v ed from normal routing tables. F or e xample, the F orwar d P ath ( s t ) can be represented as s N 1 ::: N k t where f N i g ( 1 i k ) denote the k intermediate nodes between s and t F or a gi v en P-node P and an y B-node X located in P 0 s transmission range T R pb the Bac kwar d P ath ( P X ) is dened as the minimum-hop F orwar d P ath ( X P ) when all the nodes ha v e a transmission range of B T R It is w orth noting that the minimum-hop F orwar d P ath ( X ; P ) is not necessarily the same as the F orwar d P ath ( X ; P ) Although F or war d P aths are dened for an y node pairs in the netw ork, Bac kwar d P aths are only v alid between a P-node and the B-nodes located within the P-node' s T R pb range. Furthermore, 3 Ho we v er due to the asymmetric transmission po wer the discussed P-node may not be a neighbor of an indi vidual node of those neighbor candidates.

PAGE 33

21 for an y neighbor candidate X of a gi v en P-node P this B-node X can be considered as P 0 s neighbor only when the Bac kwar d P ath( P X ) satises the follo wing criteria: All the intermediate nodes along the Bac kwar d P ath( P X ) should be in P 0 s T R pb range. In other w ords, a neighbor candidate X can be considered as a P-node P 0 s neighbor if and only if all the intermediate nodes along the Bac kwar d P ath( P X ) are P 0 s neighbor s as well. W ith the abo v e denitions, the remaining issue is ho w to set up these Bac kwar d P aths A simple w ay is to let a P-node broadcast a query message with a certain transmission po wer that is, co v ering all the B-nodes in its T R pb = n B T R range. Once seeing such a query each node broadcasts a special reply with the TTL v alue set to T4. Each node appends its o wn ID in the reply when relaying such a special reply The querying P-node will w ait some time until collecting enough replies. The initiator of a reply w ould be considered as a neighbor if and only if the querying P-node also recei v es replies initiated from all the relaying nodes of that reply W e will see later in order to f acilitate the operation of A-MA C, the path length of a Bac kwar d P ath should be limited. W e need to point out that, e v en when a P-node, say P 1 recei v es a query message initiated by another P-node, say P 2 P 1 should reply lik e common B-nodes with a transmission range of BTR Since our scheme is tar geted for netw orks with lo w or moderate mobility P-nodes can e x ecute this process infrequently or in their respecti v e P-to-B periods when topology changes are detected by the MA C protocol. Therefore, the resulting o v erhead is e xpected to be af fordable. P-nodes also need to disco v er the neighboring relationship among themselv es. T o achie v e this, during the P-to-P period a P-node may send a query with appropriate transmission po wer that is set to co v er a range of T R pp = m B T R P-nodes recei ving this query may directly send replies back to the requesting P-node. Fig.2gi v es an e xample of the neighbor selection process. Suppose A is a P-node with T R pb = 2 B T R and T R pp = 4 B T R and the Backw ard P aths for neighbor 4 T can assume an inte ger v alue slightly lar ger than n to allo w more replies.

PAGE 34

22 A B C D E G F I H J B T R T R p b T R p p P n o d e B n o d e Figure 2: An e xample of the neighbor determination process (m=4, n=2, T=3.). candidates C F G and I are C B A F E A G F E A and I H C B A respecti v ely Since node H does not initiate an reply to A only C D and G are considered as A 0 s neighbor s Of course, B D and E are A 0 s neighbor s as well. In this e xample, another P-node J is also a neighbor of node A because J is in A 0 s T R pp and the y can directly communicate with each other 2.4.2 Routing Component of DELAR In homogeneous ad hoc netw orks, a node can only communicate with other nodes in its B T R range, while in heterogeneous ad hoc netw orks, a P-node is able to reach an y other node within a lar ger transmission range, for e xample, T R pb and T R pp Therefore, the resulting topology and routing strate gy may be quite dif ferent from that in homogeneous netw orks. As an e xample, a netw ork topology without P-nodes is depicted in Fig.2 3.a, where all the links are bi-directional and labelled with equal or unequal costs on both directions. F or instance, a 1 =b 1 indicates that the link cost from A to B is a 1 while b 1 from B to A. In contrast, if one node, say A is identied as a P-node who can reach much further in the netw ork, more unidirectional links may be added as sho wn in Fig.2.b W e label unidirectional links from P-node A to its neighbors with cost 0 to represent node A 0 s unlimited po wer supply .

PAGE 35

23 A B C D E F 0 / b 1 0 / c 1 c 2 / e 1 b 2 / d 1 b 3 / c 3 d 3 / e 3 d 2 / f 1 e 2 / f 2 0 0 0 A B C D E F a 1 / b 1 a 2 / c 1 c 2 / e 1 b 2 / d 1 b 3 / c 3 d 3 / e 3 d 2 / f 1 e 2 / f 2 a T o p o l o g y i n H o m o g e n e o u s c a s e b T o p o l o g y i n H e t e r o g e n e o u s c a s e Figure 2: The topology in homogeneous and heterogeneous cases. T o cope with such heterogeneous netw orks, each P-node needs to maintain an internal neighbor table recording its chosen neighbors within T R pb and the corresponding Bac kwar d P aths of those neighbors. In addition, each node in the netw ork, either a P-node or a B-node, needs to maintain a forwar ding r outing table similar to that in a normal tabledri v en routing protocol such as DSD V [7]. F or each node i we dene f ( i ) = r esidual ener g y ( i ) )Tj/T1_3 11.95509 Tf11.94411 0 Td( q ueue l en ( i ) ; where r esidual ener g y ( i ) indicates current remaining ener gy le v el at node i q ueue l en ( i ) represents the current load status at node i and is a parameter representing the ener gy consumption per unit data transmission5. Then the de vice-ener gy-load a w are routing cost metric we adopt is gi v en in Eq.2.1, though other cost metrics are applicable in DELAR as well. cost ( i ) = 8 > < > : 1 =f ; f > r a; f r (2.1) 5 In our simulation, for e xample, is equal to the a v erage ener gy consumption per pack et transmission.

PAGE 36

24 In the abo v e cost metric, cost ( i ) is the cost of using node i as a relay it could be used as the cost of all directional links (arcs) starting from node i and directed to an y of its neighbors; r is a parameter used to adjust the weight of the a w areness of load and ener gy in the o v erall cost metric; constant a assumes a relati v ely lar ge v alue to a v oid using the nodes short of ener gy Dif ferent types of de vices may assume dif ferent v alues of these parameters: ; r and F or e xample, in this chapter to represent a P-node' s unique po wer capability or de vice type, a P-node assumes a zero link cost6for all the links tow ard its B-node neighbors within T R pb or P-node neighbors within T R pp Ideally in order to nd ener gy-ef cient paths, each node should be informed about the routing costs of other nodes as accurate and prompt as possible which may lead to e xcessi v e o v erhead. In practice, ho we v er the emplo yed routing protocol should strik e a good tradeof f between ener gy ef cienc y and o v erhead. Proacti v e routing protocols are kno wn for their capability of propag ating netw ork conditions through the whole netw ork in due course so that appropriate QoS decisions, for e xample, admission control and route selection, can be made intelligently Thus, we adopt a proacti v e routing protocol, for e xample, DSD V as the underlying routing protocol to periodically e xchange the routing information. W e note that other types of routing protocols can also be used in this frame w ork. After g athering enough routing information, a node can emplo y a shortest path algorithm to decide the minimum cost paths and the related costs to all the other nodes in the netw ork. Here the path cost is actually the sum of the cost dened in Eq.2.1of all the B-nodes along a forw arding path. Similar to those ener gy-a w are cost metrics proposed in the literature, by choosing the proper v alues of and r the cost function dened in Eq.2.1can help prolong the netw ork lifetime by distrib uting the traf c more e v enly throughout the netw ork, a v oiding the o v eruse of a small set of nodes, and consuming nodal ener gy resources in a more balanced manner [37,39]. Moreo v er DELAR spontaneously incorporates P-nodes' unique 6 In practice, a v ery small v alue can be used to a v oid possible routing loops.

PAGE 37

25 po wer capabilities or de vice types, residual ener gy information, and local load status into the routing protocol without using redirect tables in DEAR [43] an y more. 2.4.3 Hybrid T ransmission Scheduling In order to reduce the interference a P-node' s communication may impose on other ongoing transmissions, it is reasonable to only allo w a P-node to boost its transmission po wer during some e xclusi v ely reserv ed periods. F or this purpose, time is di vided into equal length time periods called Superfr ames in which each P-node is assigned an e xclusi v e time interv al, called a P-to-B period, to communicate with its B-node neighbors. In addition, a P-to-P period is allocated to allo w P-nodes to communicate with each other if the y are within each other' s T R pp range. The rest of a Superfr ame called a B-to-B period is shared by all the nodes. Therefore, as sho wn in Fig.2, a P-node has three dif ferent transmission ranges in each Superfr ame : T R pp in P-to-P period, T R pb in its o wn P-to-B period, and T R bb in the rest of the Superfr ame In contrast, a B-node only has one transmission range T R bb = B T R and it can only initiate transmissions during the B-to-B period. Fig.2gi v es an instance of such a Superfr ame structure including multiple reserv ed P-to-B periods, one for each P-node. The one-minislot-length paddings between consecuti v e P-to-B periods are used to further reduce the possible interference. The period allocation of the Superfr ame can be fullled as follo ws. During the netw ork startup phase, Pnodes use high transmission po wer to communicate and ne gotiate with each other deciding the lengths of the P-to-P period, the P-to-B period and the B-to-B period, also associating each P-node with a P-to-B period. After nishing the ne gotiation, the P-nodes broadcast such allocation information to all the B-nodes in their o wn T R pb In this w ay ultimately

PAGE 38

26 all the nodes are informed the Superfr ame allocation, and can synchronize to such allocation7. In our current design, one P-to-B period is not allo wed to share by multiple P-nodes for simplicity only Ho we v er a P-to-B period can be shared among P-nodes f ar a w ay from each other if such sharing can ensure the transmissions conict-free. Since a P-node can communicate with other P-nodes in the P-to-P periods and communicate with the B-nodes within its B T R range in the B-to-B periods, it is natural that in its o wn P-to-B period, a P-node should gi v e priority to pack ets intended to its B-node neighbors which are outside its BTR range b ut inside its T R pb range. Thus, pack et scheduling is needed at a P-node to determine the appropriate transmission schedule for the pack ets to be relayed or initiated by itself. One may notice that the aforementioned transmission scheduling may need time synchronization among the nodes, we ar gue that nodes equipped with GPS de vices can easily fulll this task. This requirement is quite realistic today since such de vices are ine xpensi v e and can pro vide reasonable precision. In this chapter we assume that nodes ha v e perfect time synchronization and lea v e the synchronization problem in the netw orks without GPS de vices as our future w ork. In f act, the time-di vision scheduling, essentially a reserv ation-based access control mechanism, and the MA C protocols emplo yed in the three types of periods in each Super fr ame either contention-based or reserv ation-based, form a h ybrid transmission scheduling for DELAR. Moreo v er each type of periods may ha v e dif ferent MA C protocol. F or e xample, the con v entional contention-based MA C protocols, such as the IEEE 802.11 MA C protocol, can be used during P-to-P periods and B-to-B periods. Since unidirectional links 7 F or simplicity we assume a x ed allocation scheme is used in this chapter ho we v er more adapti v e allocation is possible when P-nodes periodically e xchange local load infor mation and ne gotiate a ne w allocation scheme during the P-to-P periods.

PAGE 39

27 A B T r a n s m i s s i o n R a n g e Figure 2: A unidirectional link between A and B. are basically una v oidable in P-to-B periods, special measures are needed to deal with them. In what follo ws, we will delineate the A-MA C protocol de v eloped for P-to-B periods. 2.4.4 Asymmetric Media Access Control Protocol (A-MA C) The presence of unidirectional links is pretty common in heterogeneous netw orks especially when emplo yed po wer control schemes cause unequal transmission po wers among nodes. As an e xample, node A in Fig.2has a lar ger transmission range than node B Thus, node B can hear node A s transmission, ho we v er node A cannot detect node B s transmission. Ob viously a unidirectional link e xists between node A and node B The dilemma is that the stop-and-w ait ARQ (Automatic Repeat Request) scheme [47] emplo yed in cur rent contention-based MA C protocols only w orks well with bidirectional links. In the f ace of unidirectional links, the recei v er B (Fig.2) has no w ay to directly and successfully send the ackno wledgements back to the transmitter A which means that the transmitter w ould continuously transmit the same frame before timeout no matter whether the recei v er has recei v ed it or not. Moreo v er the unidirectional links may se v erely af fect the functionalities of ad hoc netw orks at v arious layers [48,49,50]. F or e xample, man y routing protocols such as DSR and A OD V rely on hop-wise ackno wledgments for disco v ering route errors. Therefore, ho w to support the MA C layer ackno wledgements o v er the unidirectional links is v ery important [51,52] and has not yet been well addressed. F ortunately we can mak e use of the aforementioned Bac kwar d P aths and the follo wing mini-r outing method to tackle this problem in an ele g ant w ay .

PAGE 40

28 C A ( P n o d e ) 1 2 3 4 5 6 A D I F S P R T S P D A T A S I F S P C T S C 1 4 2 B S I F S S I F S P C T S 3 S I F S P A C K 5 S I F S P A C K 6 B Figure 2: The A-MA C operation procedure. In current contention-based MA C protocols such as the IEEE 802.11, a recei v er can only transmit an ackno wledgement frame to its one-hop-a w ay transmitter W ith the crosslayer design methodology we introduce a ne w concept of mini-r outing intoT rack your eBay acti vities the MA C layer which requests intermediate nodes to relay the recei v er' s ackno wledgement frames, that is, CTS/A CK frames, along the established Bac kwar d P ath(tr ansmitter r eceiver) in a multi-hop f ashion to the transmitter (a P-node) at the MA C layer Here the routing information is no longer e xclusi v ely used by the netw ork layer b ut shared by the MA C and netw ork layers. Based on the IEEE 802.11 [5], we introduce into A-MA C four special frames: PR TS, P-CTS, P-D A T A, and P-A CK, all of which can only be transmitted in P-to-B periods. When a P-to-B period of a P-node comes and the P-node happens to ha v e some pack ets to transmit, it rst boosts its transmission po wer to co v er the range of T R pb = n B T R W ith the scheduling described in Section2.4.3, all the other nodes should refrain from initiating a transmission and temporarily stop transmitting usual frames, that is, R TS/CTS/D A T A/A CK. The P-node associated with this P-to-B period can send pack ets to an y neighboring B-node in the range of T R pb through P-R TS/P-CTS/P-D A T A/P-A CK e xchanges. Ne xt, we illustrate the A-MA C by using Fig.2, where we assume n = 2 and P-node A intends to send a pack et to C one of its B-node neighbors. The location relationship among A B and C is also depicted in Fig.2. First, A sends the P-R TS with T R pb =

PAGE 41

29 2 B T R containing the Bac kwar d P ath(A,C) ( C B A ). Then according to the length of the Bac kwar d P ath(A,C) in this e xample which is 2, A sets its w aiting timer for the PCTS to be 2( S I F S + T P )Tj/T1_5 7.97011 Tf6.5833 0 Td(C T S + T pr op )8, where T P )Tj/T1_5 7.97011 Tf6.5833 0 Td(C T S and T pr op are the transmission time and air propag ation time for one P-CTS respecti v ely Upon recei ving the P-R TS destined to it, and w aiting for a SIFS period, node C will send node B a P-CTS including the addresses of A and C F or an intermediate node residing on the backw ard path, it needs to set its w aiting timer according to its order in the Bac kwar d P ath F or e xample, the i th node (the intended recei v er lik e C is assumed to be the 0 th node on the path) on the path should set its timer to be i ( S I F S + T P )Tj/T1_5 7.97011 Tf6.5833 0 Td(C T S + T pr op ) In this e xample, node B when o v erhearing the abo v e P-R TS from A it starts a timer equal to S I F S + T P )Tj/T1_5 7.97011 Tf6.5833 0 Td(C T S + T pr op since it is the 1 st intermediate node on the backw ard path. Once recei ving a P-CTS from node C before timeout, B simply appends its address and relays the modied P-CTS to the ne xt hop which is the P-node A in this e xample. Otherwise, node B sends a P-CTS containing its o wn address to A after the timer e xpires, and the reason for doing this will be e xplained later If node A does not recei v e an y P-CTS before timeout, it can retransmit the P-R TS until reaching an admissible number of retries. If the same situation happens, A temporarily sa v es this pack et for future transmission and switches to another pack et with a dif ferent ne xt hop. When A successfully recei v es a P-CTS from B containing both B' s and C' s addresses, the P-R TS/P-CTS e xchange nishes. After a SIFS A can send a P-D A T A frame to node C and set the timer to 2( S I F S + T P )Tj/T1_5 7.97011 Tf6.5833 0 Td(AC K + T pr op ) Then the similar procedures apply After recei ving the P-A CK from node C relayed by the intermediate node B, the P-node A can start transmitting a ne w pack et after a DIFS in the same manner When its P-to-B period e xpires, A lo wers its transmission po wer and acts as a B-node in other P-nodes' P-to-B periods and in the B-to-B period. 8 SIFS stands for Short Inter -frame Space, and DIFS stands for DCF Inter -frame Space.

PAGE 42

30 C A ( P n o d e ) 1 2 3 4 5 6 A D I F S P R T S P D A T A S I F S P C T S C 1 4 2 B S I F S S I F S P C T S 3 S I F S P A C K 5 S I F S P A C K 6 B A s s e m l e d M P D U M A C h e a d e r M S D U C R C M A C h e a d e r M S D U C R C M P D U f o r B M P D U f o r C Figure 2: The multiple-pack ets transmission enhancement to DELAR. Besides the purpose of resolving the well-kno wn hidden/e xposed terminal problems, the P-R TS/P-CTS e xchange is also used to eliminate possible errors resulting from stale routes or nodes' mobility F or e xample, in the abo v e e xample, if node C mo v es out of P-node A 0 s 2 B T R range while B is still in A s B T R range, node C will not hear the PR TS from node A and hence A could only recei v e from node B a P-CTS including only B' s address. In this case, A will think that node C is currently unreachable and may temporarily sa v e the pack ets to C for future transmissions. Another situation may happen that node C is still in A 0 s 2 B T R range while node B mo v es out of A s B T R range, in which the P-node A will delete node C from its neighbor table. Moreo v er when node C mo v es into A s BTR range, A will hear the P-CTS from node C directly Hence A can optimize the future transmissions to C without the help of node B an y more. 2.5 Multiple-P ack ets T ransmission in DELAR In the basic DELAR design, time-di vision transmission scheduling is used to coordinate the transmission acti vities of P-nodes and B-nodes. One undesirable consequence is the e xcessi v e delay one pack et may e xperience because it may be b uf fered at intermediate nodes to w ait for appropriate transmission periods. On the other hand, since DELAR is ener gy a w are and it costs P-nodes almost nothing to transmit a pack et, man y data packets may sw arm to P-nodes. This may mak e P-nodes the bottlenecks of the netw ork and further increase the delay that pack ets accumulated at P-nodes w ould e xperience. In what follo ws, we seek a w ay to alle viate this phenomenon.

PAGE 43

31 2.5.1 Multiple-pack ets T ransmission In the basic design (see Fig.2), for e xample, A can only transmit pack ets to either B or C each time. When P-node A transmits a pack et to B-node C node B is only in v olv ed in forw arding the control frames (for e xample, P-CTS/P-A CK). Ho we v er since the channel is reserv ed for P-node A during its P-to-B periods and node B is also in P-node A s transmission range, node B has the capability to recei v e and demodulate an y signal transmitted by node A if it is allo wed. Therefor if node A can recognize B' s capability and utilize it, that is, transmitting pack ets to node B and C at the same time, the system performance could be impro v ed. This multiple-pack ets transmission mechanism can be illustrated in Fig.2. Suppose P-node A has some pack ets to B and some pack ets to C Node A w ould put one pack et for C and another for B together and send them in a single transmission from which nodes B and C can acquire their o wn part, respecti v ely In this w ay we e xpect to see the impro v ement of the end-to-end delay performance of DELAR. T o do this, node A rst mak es sure that both B and C are within its ef fecti v e range T R pb after the P-R TS/P-CTS e xchange as before. Then A can pull out from the w aiting queue one pack et to w ards C and another pack et to w ards B and send them in one P-D A T A frame as depicted in Fig.2. When seeing such a frame, nodes B and C can e xtract their o wn parts and dump the rest. The same procedures as specied in A-MA C are e x ecuted with the e xception that node B also needs to indicate in the P-A CK that it has successfully recei v ed the pack et to itself. The similar procedure can be applied to the cases with more than one interlineate nodes on the bac kwar d paths W ith the abo v e multiple-pack ets transmission mechanism, one may e xpect that the total number of pack ets that can be pack ed and transmitted at one time is bounded by the length of the Bac kwar d P ath Considering the possibility of adopting high-data rate modulation schemes with a higher po wer le v el so as not to de grade the recei v ed signal strength, more pack ets to w ards dif ferent recei v ers on the Bac kwar d P ath can be assembled together and transmitted at the same time.

PAGE 44

32 A 0 0 D 0 1 B 1 0 C 1 1 0 1 q Figure 2: General hierarchical 2/4-PSK constellation. 2.5.2 Hierarchical Modulation T o support the multiple-pack ets transmission, hierarchical modulation (or nonuniform modulation) schemes can be use to ensure all the recei v ers ha v e enough recei v ed signal strength to demodulate the useful information. In hierarchical modulation schemes, the constellations consist of nonuniformly spaced symbols and allo w for unequal error protection, that is, dif ferent de grees of protection for transmitted bits within a symbol are allo wed according to the importance of the information. F or e xample, suppose that there are tw o streams of data, each of which has a priority (a tar get BER), and QPSK hierarchical modulation (see Fig.2) is used. One bit from each stream is tak en to form a symbol of tw o bits. The bit from bitstream with high priority (lo wer allo wed BER) is assigned to the most signicant position, for e xample, the rst bit in the constellation in Fig.2. The bit from the bitstream with lo w priority is assigned to the less signicant position. By adjusting the angle of the constellation can be determined to achie v e the unequal error protection for each bitstream. Such unequal error protection has mak e hierarchical modulation v ery attracti v e in multimedia services [53,54,55]. W e notice that in the pre vious research, the hierarchical modulation is used in the cases that the transmission po wer is x ed and the constellation is chosen in a w ay to allocate the po wer among multiple data streams to achie v e unequal error protection. In addition, in the pre vious research, bitstreams usually are destined to the same recei v er Ho we v er in our

PAGE 45

33 0 0 0 1 1 0 1 1 q C A ( P n o d e ) d 1 d 2 d 1 2 BFigure 2: Implement multiple-pack ets transmission with hierarchical modulation DELAR scheme, the transmission po wer of P-nodes is not x ed. Moreo v er when multiplepack ets transmission is used, the bitstreams are destined to dif ferent recei v ers. As sho wn in Fig.2, multiple recei v ers in v olv ed in the multiple-pack ets transmission ha v e dif ferent distances to the transmitter Therefore, in order to use hierarchical modulation to support multiple-pack ets transmission at a P-node, more po wer ought to be allocated to f ar a w ay recei v ers and less po wer for close recei v ers. More specic, we need to nd out the constellation and the minimum o v erall transmission po wer for a P-node so that all the recei v ers can successfully get their o wn pack ets. In the follo wing we use QPSK hierarchical modulation to implement the multiple-pack ets transmission when tw o recei v ers are in v olv ed (Fig.2). Suppose B E R max is the BER requirement for both recei v ers B ( d 2 a w ay from transmitter A ) and C ( d 1 a w ay from transmitter A ) to demodulate the recei v ed signal. Moreo v er we assume tw o-ray ground model with = 4 is used to model the radio propag ation: P r = P t G t G r h 2 t h 2 r d 4 L where L is the system loss f actor P t is the transmission po wer P r is the recei v ed po wer G t ( G r ) is the antenna g ain at the transmitter(recei v er), and h t ( h r ) is the height of antenna at the transmitter(recei v er). F or simplicity we assume all the nodes ha v e the same antenna g ain and height. W e assume that is less than 45 o so that the inphase signal to w ards f ar

PAGE 46

34 a w ay recei v er C is assigned with more po wer and quadrature signal to w ards close recei v er B with less po wer It is easy to get the BER at C as p 1 = Q ( s 2 P r 1 W N f cos ) = Q ( v u u t 2 P t G t G r h 2 t h 2 r d 4 1 L W N f cos ) (2.2) where N is the noise signal po wer f is the transmission bit rate, W is the channel bandwidth (in H z ), and Q ( ) is Gaussian Q-function [54,55]. Similarly we can get the BER at the recei v er B as p 2 = Q ( s 2 P r 2 W N f sin ) = Q ( v u u t 2 P t G t G r h 2 t h 2 r d 4 2 L W N f sin ) : (2.3) Thus, in order to correctly demodulate the recei v ed signal, the BER should satisfy the follo wing conditions: p i B E R max where i = 1 ; 2 Since we are interested in the minimum transmission po wer of node A when equality is satised: p i = B E R max where i = 1 ; 2 we dene r 0 = Q )Tj/T1_2 7.97011 Tf6.5833 0 Td(1 ( B E R max ) as the SNR achie ving the B E R max to f acilitate us nding the boundary v alues of P t and From Eq.2.2and Eq.2.3, it is easy to get v u u t 2 P t G t G r h 2 t h 2 r d 4 1 L W N f cos = r 0 (2.4) and v u u t 2 P t G t G r h 2 t h 2 r d 4 2 L W N f sin = r 0 : (2.5) From Eq.2.4and Eq.2.5, we can get = arctan d 2 2 d 2 1 (2.6) and P tmin = r 0 N f ( d 4 1 + d 4 2 ) L W G t G r h 2 t h 2 r (2.7)

PAGE 47

35 Therefore, gi v en the distances between the transmitter and recei v ers, we can get the minimum transmission po wer P tmin and in the QPSK hierarchical modulation constellation, and use this combination to implement the proposed multiple-pack ets transmission and achie v e the BER requirements at both recei v ers. In practice, P-node A may rst use a predened transmission po wer co v ering T R pb and transmit the R TS frame using typical BPSK. After the R TS-CTS handshak e, A can determine the tw o parameters P tmin and based on the feedback from the tw o recei v ers. Then A pulls out from the w aiting queue one pack et to w ards C as the more important bitstream and one to w ards B as the less important bitstream. Further it can transmit symbols using the aforementioned QPSK hierarchical modulation. W e should note that, in order to a v oid the possible interference when a P-to-B period is shared by multiple P-nodes, the in v olv ed P-nodes, before enabling the multiple-pack ets transmission, should e xamine and mak e sure that P tmin is not greater than the po wer to co v er T R pb In f act, the minimum total transmission po wer to transmit the pack ets to each recei v er separately using BPSK can be written as follo ws: P ttoal = P t 1 + P t 2 = r 0 N f d 4 1 L W G t G r h 2 t h 2 r + r 0 N f d 4 2 L W G t G r h 2 t h 2 r : (2.8) Compared Eq.2.7to Eq.2.8, we can see that although P tmin alone is lar ger than either P t 1 or P t 2 P tmin is identical to P ttoal This v eries that multiple-pack ets transmission with QPSK hierarchical modulation does not require e xtra transmission po wer9compared to the case when multiple-pack ets transmission is not used. In f act, if multiple-pack ets transmission is implemented with uniform QPSK, in order to satisfy the same BER requirement at both recei v ers: p i B E R max where i = 1 ; 2 the transmission po wer w ould be p 2 P t 1 which is much lar ger than P tmin Thus, in our case, QPSK hierarchical modulation is a rather reasonable option to implement multiple-pack ets transmission. Further 9 Here we only consider the po wer used to transmit the information bits.

PAGE 48

36 if we consider the ener gy used for control frames, the proposed multiple-pack ets transmission requires fe wer control frames thus less ener gy consumption, compared to the case when pack ets are transmitted separately F or e xample, in Fig.2, transmission ener gy for three control frames (R TS/CTS/A CK) to w ards node B can be sa v ed when multiple-pack ets transmission is used. 2.6 Discussion 2.6.1 The Existence of Backw ard P aths In DELAR, the P-nodes are utilized in tw o w ays to conserv e ener gy: enabling Pnodes to directly communicate with other P-nodes within T R pp and enabling P-nodes to directly communicate with B-nodes within T R pb Unlik e the communications between Pnodes that need no special treatment, the communications between P-nodes and B-nodes are supported by the nodes on the backw ard paths as described in Section2.4.1, thus the backw ard path is v ery important for the proper functioning of DELAR. Then one may question the e xistence of such backw ard paths. W e belie v e that the e xistence of backw ard paths is related to se v eral f actors such as basic transmission range (BTR), node density and the location distrib ution of P/B-nodes. On the other hand, the backw ard path should not be too long in terms of hop count, otherwise A-MA C may not function well because a long backw ard path is more subject to breakage due to node mobility or other reasons. In practice, we should include this limitation into the aforementioned neighbor -selection criteria, for e xample, setting TTL v alue of a reply to the neighboring query to n (cf. Section2.4.1), such that only those B-nodes with backw ard paths of less than n hops can be considered as neighbors of a gi v en P-node. Ne xt we w ant to study the e xistence of such backw ard path in typical mobile ad hoc netw orks. W e assume n = 2 and T = 2 (cf.2.4.1) so that only those nodes with backw ard paths not more than 2 hops are considered as neighbors of a gi v en P-node. Let O denote the position of a P-node, and O 0 denote the position of a B-node in O s 2 B T R = 2 R range, as sho wn in Fig.2. W e use r to indicate the distance between O and O 0 It is

PAGE 49

37 ob vious that when r is not greater than R O 0 is denitely a neighbor of O So we shall only consider R < r 2 R when studying the e xistence of backw ard paths for those nodes that are r a w ay from O W e also assume that all the nodes are uniformly and independently distrib uted in a tw o-dimensional area with node density where the probability of ha ving k nodes in an area of size S follo ws Poission distrib ution: p ( k ; S ) = ( S ) k k e )Tj/T1_5 7.97011 Tf6.5833 0 Td( S : The mean of the number of nodes in the area of size S is S Then the e xistence of backw ard paths between O and O 0 or the problem whether O 0 is a neighbor of O or not can be reduced to a rather simple geometric problem: Is there an y other node located in the shado wed area S with both O and O 0 as neighbors? Let be the a v erage number of nodes e xisting in the shado wed area, and S 1 and S 2 be the areas of sector [ AB O and 4 AO B respecti v ely Then we ha v e = 2 ( S 1 )Tj/T1_1 11.95509 Tf11.4671 0 Td(S 2 ) where S 1 = R 2 2 2 arccos r = 2 R = R 2 arccos r = 2 R and S 2 = 1 2 R 2 sin 2 arccos r = 2 R : Therefore, we ha v e = 2 ( R 2 arccos r = 2 R )Tj/T1_4 11.95509 Tf13.149 8.08202 Td(1 2 R 2 sin 2 arccos r = 2 R ) : Thus, the probability of e xisting a backw ard path between O and O 0 which is equal to the probability that there is at least one node in the shado wed area, is 1 )Tj/T1_1 11.95509 Tf11.9261 0 Td(e )Tj/T1_5 7.97011 Tf6.5833 0 Td( according to the Poission distrib ution. Fig.2plots this probability with dif ferent v alues of r where the total number of nodes is 50 for all netw ork topologies. W e can see that, with a reasonable node density the probability of e xisting a backw ard path between a gi v en P-node and Bnode pair is pretty high, which justies the feasibility of our A-MA C protocol and in turn the o v erall DELAR scheme.

PAGE 50

38 R 2 R r O O A B S Figure 2: The e xistence of backw ard path. 2 0 0 2 5 0 3 0 0 3 5 0 4 0 0 4 5 0 5 0 0 0 0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 1 r ( m ) P r o b ( E x i s t e n c e o f a b a c k w a r d p a t h ) 1 5 0 0 3 0 0 m27 0 0 7 0 0 m28 0 0 8 0 0 m29 0 0 9 0 0 m21 0 0 0 1 0 0 0 m2 B T R = 2 0 0 B T R = 2 5 0 Figure 2: The a v erage number of nodes e xisting in the shaded area.

PAGE 51

39 2.6.2 DELAR and ZRP In f act, we can borro w some ideas from ZRP [56,11] to further impro v e the routing performance of DELAR as follo ws. A P-node maintains the routing information within a zone T R pb by using the procedures described in Section2.4.1or an y other routing protocol as the intra-zone routing protocol (IARP). It also maintains the information about its neighboring P-nodes in its T R pp In contrast, a B-node only needs to maintain the routing information in its T R bb = B T R where all the nodes are its one-hop neighbors. Once a node needs an ener gy-ef cient route to another node, it can disco v er the route on demand using the routing disco v ery procedure similar to that of A OD V b ut with what is dened in Eq.2.1instead of hop count as the cost metric. In this sense, DELAR can be vie wed as a special case of ZRP The dif ference lies in the f act that all nodes within a P-node' s T R pb zone, are one-hop a w ay from this P-node from the routing perspecti v e rather than multi-hop a w ay in the le g ac y ZRP Besides, the border -casting [56,11] technique used in ZRP is also a v ailable in DELAR in the sense that a P-node border -casts a route request (which we call simply a request) to all its peripheral nodes with corresponding backw ard paths embedded in the route request. In this w ay each peripheral node w ould learn the backw ard path used to return a route reply if needed. Moreo v er since a P-node can directly e xchange routing information with other P-nodes within its T R pp it also can border -cast the route request to its P-node neighbors, which in turn can look up their o wn neighbor tables to decide if the desired destination is in their o wn T R pb zones. Since T R pp is usually lar ger than T R pb apparently such inter -zone routing protocol (IERP) can further speed up the route disco v ery process. 2.6.3 The Choice of m and n Since a P-node can use higher transmission po wer to communicate with other P-nodes within its m B T R lar ger m w ould lead to less use of B-nodes in the communications, b ut also less spatial reuse. Similarly lar ger n may lead to more ener gy sa vings, b ut also imply possible longer backw ard paths and less spatial reuse, which may mak e A-MA C less

PAGE 52

40 ef cient as e xplained before. F or the similar reasons, usually n is less than m In order to well balance the ener gy sa vings and other system performance f actors, both m and n should be chosen cautiously While DELAR requires all the P-nodes ha v e the same v alue of m to a v oid producing unidirectional links between P-nodes in P-to-P periods, with A-MA C in place, the y can ha v e dif ferent v alues of n 2.6.4 Benets of the T ime Di vision Scheduling In f act, P-nodes can communicate with each other in one-hop or multi-hop manner during P-to-P periods to coordinate the use of the ne xt Superfr ame A P-node can also notify all the other nodes within its T R pb to adjust their transmission schedules. W ith these intelligent communications a v ailable, the slot allocation in each Superfr ame can be adjusted adapti v ely according to traf c conditions instead of being x ed as in the gi v en e xample. It is w orth pointing out that this time-di vision scheduling method can f acilitate the operations of PSM (cf. Section2.2) in that it can help nodes determine their sleep and w ak e schedules. F or e xample, B-nodes can turn of f the radios during the P-to-P periods to conserv e ener gy As we discussed before, time synchronization is of importance for the correct oper ation of DELAR. In literature, there e xist man y proposals for time synchronization and man y of them can be incorporated into DELAR. In f act, P-nodes can serv e as coordinators to f acilitate such synchronization by sending out some beacon information in some P-toP periods and P-to-B periods periodically Subsequently B-nodes can synchronize their clocks to these P-nodes. 2.7 Performance Ev aluation 2.7.1 Simulation Setup In order to e v aluate the performance of DELAR, we implemented our scheme including the routing layer and the A-MA C in the OPNET Modeler [57]. W e simulated a netw ork with 50 nodes randomly deplo yed in a 1500 300 m 2 area. The B T R w as 200 m and the transmission rate w as 2 M bps In our simulations, all the nodes were capable of mo ving

PAGE 53

41 in the netw ork according to the modied random W aypoint mobility model presented in [58]. The pause time w as set to be zero in our simulations, meaning nodes were al w ays mo ving. And each node mo v ed with a randomly chosen speed between [ V min V max ], where V min w as x ed to be 1 m/s and V max assumed dif ferent v alues to reect v arious netw ork mobility le v els. There were 20 constant bit rate (CBR) data sessions between randomly selected source and destination pairs, and each source generated data pack ets of 512 bytes in length at a rate of pack ets per second. In our simulation, B-nodes had the same initial ener gy reserv oir 6 k J and their transmission and reception po wer were 1560 mW and 930 mW respecti v ely [59]. The netw orks with 2 to 6 P-nodes were studied and these P-nodes were randomly deplo yed. Besides, we chose m = 4 n = 2 t pp = 0 : 02 s t pb = 0 : 02 s and t bb = 0 : 05 s10. By v arying the number of P-nodes, the maximum mo ving speed, and the CBR source rate, we were able to study the performance of DELAR under dif ferent congurations. Six runs were carried out to get an a v erage result for each simulation conguration and each run w as e x ecuted for 900 seconds of simulation time. Pre vious w ork [37,39] has sho wn that the ener gy ef cienc y of routing protocols can be much impro v ed by adopting such a routing metric as what we dened in Eq.2.1. Therefore, we shall only compare our DELAR and DELAR with multiple-pack ets transmission (denoted by DELAR+MPT) with the one referred to as EAR in this chapter which is dev eloped for comparison purpose and in f act a v ariant of DSD V with the routing cost metric dened in Eq.2.1. In EAR, all the P-nodes ha v e the same transmission range as B-nodes, b ut ha v e a zero cost to their neighbors because the y are assumed to ha v e almost unlimited ener gy reserv oir and other resources. And the follo wing metrics will be adopted in 10 W ith adv anced simulation tools such as OPNET NS-2, and Qualnet, it is possible to gure out a good conguration of these parameters to yield a good system performance, before putting DELAR into practice.

PAGE 54

42 comparison: a v erage ener gy consumption dened as the total ener gy consumption for all pack et transmissions and receptions normalized by the number of deli v ered pack ets; pack et deli v ery ratio dened as the ratio of deli v ered data pack ets to those generated by the sources; a v erage pack et end-to-end delay dened as the a v erage delay from when a pack et is generated and transmitted by the source till it is recei v ed by the destination. 2.7.2 Impact of The Number of P-nodes W e rst x ed the data rate to 4 pack ets/s and v aried the number of P-nodes to study its impact on the performance of DELAR. Here only the results for V max =4 m=s are presented in Fig.2, though DELAR has the similar performance with other v alues of V max Figure 2(a) comparesaverageenergyconsumptionofDELARandEARunder dif ferent numbers of P-nodes. Since EAR is also an ener gy-a w are routing protocol, it is of no surprise to see that its ener gy-sa ving performance impro v es with the number of P-nodes whose routing costs to their neighboring nodes are assumed to be zero. W ith DELAR in place, ho we v er the a v erage ener gy consumption can be much further reduced, for e xample, with a f actor of almost 50% if 6 P-nodes are a v ailable. The sho wn adv antage comes from the f act that DELAR mak es much better use of P-nodes than EAR through intelligent transmission scheduling and the allo w ance of P-nodes using dif ferent transmission po wer during v arious periods. And the more P-nodes, the more ener gy sa vings we can e xpect from DELAR. Compared to the basic DLEAR, DELAR with multiple-pack ets transmission can further reduce the ener gy consumption. It can be contrib uted to the reduced transmissions of control pack ets because multiple-pack ets transmission requires fe wer control frames than separate transmissions. From Fig.2(b), we can see that DELAR outperforms EAR in terms of pack et deli v ery ratio (PDR). The reason is that transmission scheduling in DELAR will lead to less congestion, and the lar ger transmission ranges of P-nodes during P-to-P periods and P-to-B periods can help reduce the number of hops a pack et may tra v el. Ag ain, the more P-nodes, the more PDR is impro v ed. Compared to the basic DLEAR, the multiple-pack ets

PAGE 55

432 3 4 5 6 1 2 1 3 1 4 1 5 1 6 1 7 1 8 1 9 2 0 2 1 2 2 2 3 2 4 2 5 2 6 2 7 2 8 2 9 A v e r a g e P a c k e t E n e e g y C o n s u m p t i o n ( m J )N u m b e r o f P n o d e s E A R D E L A R D E L A R + M P T(a) A v erage ener gy consumption2 3 4 5 6 0 9 2 8 0 9 3 0 0 9 3 2 0 9 3 4 0 9 3 6 0 9 3 8 0 9 4 0 0 9 4 2 0 9 4 4 0 9 4 6 0 9 4 8 0 9 5 0 0 9 5 2 0 9 5 4 0 9 5 6 0 9 5 8 0 9 6 0 0 9 6 2 P a c k e t D e l i v e r y R a t i oN u m b e r o f P n o d e s E A R D E L A R D E L A R + M P T(b) P ack et deli v ery ratio2 3 4 5 6 0 0 0 0 0 5 0 1 0 0 1 5 0 2 0 0 2 5 0 3 0 0 3 5 0 4 0 A v e r a g e P a c k e t E n d t o e n d D e l a y ( s )N u m b e r o f P n o d e s E A R D E L A R D E L A R + M P T(c) A v erage pack et end-to-end delay Figure 2: Simulation results with dif ferent number of P-nodes.

PAGE 56

44 transmission can further impro v e the pack et deli v ery ratio because transmitting multiple pack ets simultaneously can impro v e the channel utilization. In terms of end-to-end delay ho we v er EAR is better than DELAR as sho wn in Fig.2 11(c). This is because, in contrast to the contention-based transmissions in EAR, DELAR di vides the time into Super F r ames to schedule the transmission acti vities so that a pack et usually needs to be b uf fered at a node w aiting for the coming of a proper transmission period. In addition, we can observ e that the delay of EAR decreases with the increase of P-nodes because the e xistence of more P-nodes can achie v e better load balance, that is, the number of ener gy-ef cient paths may be increased. Ho we v er this is not al w ays the case with DELAR. Besides the better load balance, the increase of P-nodes also means for DELAR that the delay from the transmission scheduling becomes lar ger This contrib utes to the uctuation in the delay of DELAR depicted in Fig.2(c). Compared to the basic DLEAR, DELAR with multiple-pack ets transmission can reduce the pack et end-to-end delay 2.7.3 Impact of Node Mobility In this subsection, we study the impact of the node mobility on DELAR by v arying V max from 2 m=s to 16 m=s F or the reason of clarity only the results for 4 P-nodes and a data rate of 4 pack ets=s are presented. Fig.2(a)compares the a v erage ener gy consumption of DELAR and EAR under dif ferent mobility le v els. As we can see, DELAR al w ays has less ener gy consumption than EAR due to the reasons stated before. Generally the higher mobility leads to less ener gy consumption. After e xamining the a v erage number of hops a pack et may tra v el, we notice that higher mobility often results in shorter route, which statistically leads to less ener gy consumption because fe wer transmissions and receptions are in v olv ed. As sho wn in Fig.2(b), the pack et deli v ery ratio decreases with the increase of the mobility which is in accordance with pre vious studies. F or the similar reason we stated in pre vious subsection, DELAR al w ays has higher pack et deli v ery ratio than EAR in all kinds of mobility As

PAGE 57

450 2 4 6 8 1 0 1 2 1 4 1 6 1 8 1 4 1 6 1 8 2 0 2 2 2 4 2 6 2 8 3 0 A v e r a g e P a c k e t E n e e g y C o n s u m p t i o n ( m J )M a x i m u m N o d e S p e e d : V m a x ( m / s ) E A R D E L A R D E L A R + M T P(a) A v erage ener gy consumption0 2 4 6 8 1 0 1 2 1 4 1 6 1 8 0 8 2 0 8 4 0 8 6 0 8 8 0 9 0 0 9 2 0 9 4 0 9 6 0 9 8 P a c k e t D e l i v e r y R a t i oM a x i m u m N o d e S p e e d : V m a x ( m / s ) E A R D E L A R D E L A R + M T P(b) P ack et deli v ery ratio0 2 4 6 8 1 0 1 2 1 4 1 6 1 8 0 0 4 0 0 6 0 0 8 0 1 0 0 1 2 0 1 4 0 1 6 0 1 8 0 2 0 0 2 2 0 2 4 0 2 6 0 2 8 0 3 0 0 3 2 0 3 4 0 3 6 A v e r a g e P a c k e t E n d t o e n d D e l a y ( s )M a x i m u m N o d e S p e e d : V m a x ( m / s ) E A R D E L A R D E L A R + M T P(c) A v erage pack et end-to-end delay Figure 2: Simulation results with dif ferent maximum node speed.

PAGE 58

462 0 0 3 0 0 4 0 0 5 0 0 6 0 0 7 0 0 0 9 1 0 9 2 0 9 3 0 9 4 0 9 5 0 9 6 P a c k e t D e l i v e r y R a t i oT r a f f i c L o a d ( K b i t s / s ) E A R D E L A R D E A L R + M P T(a) P ack et deli v ery ratio2 0 0 3 0 0 4 0 0 5 0 0 6 0 0 7 0 0 0 0 0 0 0 5 0 1 0 0 1 5 0 2 0 0 2 5 0 3 0 0 3 5 0 4 0 A v e r a g e P a c k e t E n d t o e n d D e l a y ( s )T r a f f i c L o a d ( K b i t s / s ) E A R D E L A R D E L A R + M P T(b) A v erage pack et end-to-end delay Figure 2: Simulation results with dif ferent traf c load. the mobility increases, generally the delays of both EAR and DELAR get longer Ag ain, DELAR has longer delay than EAR due to DELAR' s time-di vision medium access control mechanism. One interesting observ ation is that the delays of both DELAR and EAR uctuate around V max = 2 m=s and V max = 2 m=s This can be attrib uted to the used routing cost metric which causes lots of pack ets sw arming to the P-nodes. This phenomenon results in longer w aiting time at P-nodes. Ho we v er nodal mo v ement helps alle viate such phenomenon by dispensing the traf c load. Compared to the basic DLEAR, DELAR with multiple-pack ets transmission can reduce the ener gy consumption, impro v e the pack et deli v ery ratio, and shorten the end-to-end delay in all kinds of mobility .

PAGE 59

47 2.7.4 Impact of T raf c Load In this subsection, we study the impact of the traf c load on DELAR by v arying the data generation rate from 3 pack ets=s to 8 pack ets=s Since the traf c load does not much impact a v erage ener gy consumption, we only depict the simulation results for the pack et deli v ery ratio and the end-to-end delay in Fig.2, where the number of P-nodes is four and V max is equal to 4 m=s Figures 2(a) and 2(b) demonstratethatthepacketdeliveryratiodecreases and the delay increases with the increase of the traf c load for both schemes, respecti v ely which are quite intuiti v e. Ag ain, our DELAR is better in terms of the pack et deli v ery ratio b ut w orse with re g ard to the end-to-end delay than EAR for the reasons stated pre viously Ho we v er Compared to the basic DLEAR, DELAR with multiple-pack ets transmission outperforms DELAR due to reason described before. In summary DELAR is more appropriate for delay-insensiti v e applications, such as le transfer and web access, which prefer higher ener gy-ef cienc y and pack et deli v ery ratio. W ith multiple-pack ets transmission, the performance can be further impro v ed. Thus it is a viable enhancement and can be ef fecti v ely used in DELAR frame w ork. W ith a better tuning of system parameters DELAR can strik e a good balance between ener gy ef cienc y and other system performance f actors. 2.8 Summary In this chapter we proposed a De vice-Ener gy-Load A w are Relaying frame w ork, namely DELAR, to achie v e ener gy conserv ation in mobile ad hoc netw orks. DELAR utilizes the de vice heterogeneity inherent in ad hoc netw orks and features the cross layer protocol design methodology T o tak e better adv antage of po werful nodes (P-nodes) while reducing their interference on other ongoing communications, a h ybrid transmission scheduling mechanism is used to schedule and coordinate the transmission acti vities among P-nodes and B-nodes (ordinary nodes). In addition, in order to support reliable transmissions in the

PAGE 60

48 presence of unidirectional links between P-nodes and B-nodes, we introduced the minirouting technique and the no v el Asymmetric MA C ( A-MA C ) protocol. W e demonstrated that A-MA C can ef fecti v ely enable the MA C layer ackno wledgements o v er unidirectional links. T o the best of our kno wledge, no pre vious ef fort has been made to address this issue at the MA C layer Moreo v er we proposed a multiple-pack ets transmission scheme which can be operated with hierarchical modulation scheme to further impro v e the performance of DELAR. Detailed simulations sho wed that DELAR can signicantly reduce the ener gy consumption and thus prolong the netw ork lifetime e v en with a fe w P-nodes e xisting in the netw ork. W ith this frame w ork, v arious ener gy conserv ation techniques such as po wer sa ving modes, transmission po wer control and po wer -a w are routing can be inte grated to jointly achie v e better ener gy conserv ation. More important, this frame w ork pro vides a platform to study other challenging issues such as QoS pro visioning and security support.

PAGE 61

CHAPTER 3 RESOURCE A W ARE MO VEMENT IN HETER OGENEOUS MOBILE AD HOC NETW ORKS 3.1 Introduction As discussed in Chapter2, there has been a rich literature addressing the ener gy conserv ation issue in MANETs, ranging from po wer -sa ving mode(PSM) [23,60,25,24,61], to transmission po wer control(TPC) [26,32,27,30,33,34], and to po wer -a w are routing(P AM) [37,40,41]. Though those ener gy-a w are MA C and routing protocols can in general help the whole system e xpend the ener gy resources more reasonably to some e xtent, there are still man y other aspects one can e xplore to further impro v e the system-wide ener gy ef cienc y As an e xample, similar to the traf c jam in daily life where a mass of v ehicles ocks to a single spot, unwise mo v ement may cause local netw ork traf c congestion, thus leading to unf a v orable ener gy w aste. This observ ation moti v ates us to address ener gy conserv ation from the perspecti v e of node mo v ement. In addition to node mo v ement, resource heterogeneity as another inherent character istic of MANETs, is often either o v erlook ed or underutilized in designing ener gy conserv ation schemes for MANETs. Though node heterogeneity can be interpreted in v arious w ays, we limit the scope of this chapter to heterogeneous netw orks in terms of ener gy supply In such a netw ork, most nodes (called B-nodes hereafter) are furnished with lightweight batteries ha ving limited po wer while a fe w others (called P-nodes hereafter) are po wered by almost unlimited ener gy supplies such as ener gy-sca v enging de vices (for e xample, solar cells) and dynamos when nodes are in some mobile v ehicles. In a relati v e sense, the ener gy consumption of P-nodes can be considered as small or e v en ne gligible. In this chapter we propose to address ener gy conserv ation by guiding nodes' s mo v ement and utilizing de vice heterogeneity The basic idea is that, instead of mo ving in the 49

PAGE 62

50 eld blind to the netw ork en vironment, for e xample, al w ays follo wing shortest-hop paths, nodes are instructed to tra v el much more intelligently by k eeping in mind the system-wide objecti v e of ener gy conserv ation and mo ving along resource-a w are paths in such a w ay that P-nodes can undertak e as much communication tasks as possible so that less po werful P-nodes can sa v e ener gy thus leading to the prolonging of the whole netw ork lifetime. The contrib utions in this chapter are mainly fourfold. First, we dene a general mobility model and formulate a general resource-a w are mo v ement problem, from which we deri v e a W aterhunter Mo v ement problem for B-nodes and a Firehunter Mo v ement problem for P-nodes. Second, we reduce the W atehunter Mo v ement problem to a NP-complete distance-constrained least-cost (DCLC) routing problem and propose an ef cient heuristic solution. Third, we propose a routing delay dif ferentiation mechanism to mak e full use of the benets pro vided by the resource-a w are mo v ement. Last, this resource-a w are mo v ement can be incorporate into other ener gy conserv ation schemes to further impro v e the ener gy ef cienc y The ef fecti v eness of the proposed schemes are justied and v alidated through e xtensi v e simulations. The rest of the chapter is or g anized as follo ws. W e start with Section3.2surv e ying the related w ork, then we formulate the resource-a w are mo v ement problem in Section3.3. In Section3.4, we focus on the W aterhunter Mo v ement problem and propose an ef cient heuristic solution. Section3.5e v aluates the performance of the proposed schemes. Finally this chapter is concluded in Section3.6. 3.2 Related W ork In this section, we brief some of related w ork that are closely related to this research. Grossglauser and Tse sho wed that node mobility can be used to impro v e netw ork throughput [62]. After predicting the destination' s location, a node forw ards a data pack et to a subset of its neighboring nodes in the direction of the destination to reduce the o v erall routing ener gy consumption [63]. Similarly mo v ement information w as also used in [64] to limit the ooding in a restricted area to reduce the ener gy consumption of route disco v ery .

PAGE 63

51 Chakraborty et al. proposed a strate gy to reduce the ener gy consumption by delaying communication until a mobile node mo v es close to its peer tar get, within an applicationimposed deadline [65]. Dif ferent from the abo v e w ork trying to utilize e xisting mobility information, there are some w ork discussing ho w to mo v e proacti v ely or control node mobility to impro v e the system performance. Li and Rus proposed an optimal algorithm to computer the trajectory of nodes for minimizing message transmission delay [66]. Goldenber g at al. proposed a distrib uted mobility-control scheme to guide the nodes to adjust their mo v ements with the purpose of attaining a potential ener gy-minimizing netw ork conguration [67]. In sparse ad hoc netw ork, proacti v e node mobility can as well be used to o v ercome netw ork partitions. More specically nodes should b uf fer and carry pack ets during netw ork partitions, and forw ard pack ets to other nodes when the y meet. Such a store-carry-forw ard paradigm w as proposed to help data deli v ery by making use of node mobility [68]. In addition, Zhao et al. [69] proposed a message ferrying (MF) approach to address the similar netw ork partition problem. In their approach, a set of special mobile nodes, called mobile ferries, pro vide data forw arding service to other nodes by either mo ving along the routes kno wn a priori to other nodes or proacti v ely mo ving to meet other nodes. Our resource-a w are mo v ement (RAM) strate gy proposed in this chapter is also a proacti v e mo v ement approach. It dif fers from all the pre vious w ork in its unique designing objecti v e, that is, nding the optimal mo v ement trails for each indi vidual node to minimize the total ener gy consumption of the whole system by taking into consideration the inherent resource heterogeneity of MANETs. 3.3 Problem F ormulation W e consider a MANET that consists of tens or e v en hundreds of mobile nodes, among which there are N r re gular battery-po wered nodes (called B-nodes hereafter) and N p po werful nodes (called P-nodes hereafter) ha ving almost unlimited ener gy supplies such as solar cells. Communication de vices installed on a mobile v ehicle and po wered by inside

PAGE 64

52 alternators are other e xamples of such P-nodes. Usually N p is much smaller than N r W e assume that all the nodes are able to generate traf c or forw ard pack ets for others no matter when the y are at rest or in motion. Intuiti v ely since P-nodes ha v e relati v ely innite ener gy reserv oir as opposed to battery-po wered B-nodes, the y should be utilized as much as possible to sa v e the scarce resources of B-nodes and thus prolong the whole netw ork lifetime. F or e xample, a pack et should be forw arded to a P-node whene v er possible if ener gy sa vings can be e xpected. On the other hand, we should reduce the use of B-nodes if we cannot completely a v oid using them. Ho w to realize this simple rationale, ho we v er is by no means a easy task. In this dissertation, we intend to address the issue of ener gy conserv ation from the vie wpoint of node mo v ement. In what follo ws, we rst present a general mobility model that is used to characterize nodes' mo v ement patterns. W e then introduce the resourcea w are mo v ement problem in its general form with the consumption of ener gy resources as the sole optimization objecti v e. 3.3.1 General Mobility Model (GMM) During an observ ation period T we assume that there are some designated locations that an y node i be it a P-node or B-node, should stop by at some designated time instances. F or e xample, a student carrying a mobile de vice may appear in the classroom during schooltime while in the cafeteria during lunchtime. Let J i denote the size of the ordered list of locations node i should visit during T which might be dif ferent for each indi vidual node. W e denote by pos i ( j ) ( 0 j < J i ) the j th location that node i should stop by and by t i ( j ) the required time instance. Then pos i (0) denotes the starting point of node i and pos i ( J i )Tj/T1_4 11.95509 Tf10.56709 0 Td(1) denotes the location of its last stop during T W e will also call as an epoc h [70] the time duration from one node lea ving the current stop until it reaches the ne xt stop henceforth. Whene v er arri ving at some designated location at the specied instance, each node is assumed to pause for a while according to concrete application requirements. Let pause i ( j ) indicate the time node i spends at pos i ( j ) .

PAGE 65

53 Based on the abo v e denitions, the order list f pos i ( j ) ; t i ( j ) ; pause i ( j ) ; 0 j < J i g can well characterize the itinerary of node i during the observ ation period T Let z }| { pos i ( j ) pos i ( j + 1) denote the path tra v elled by node i from its j th stop to the ( j + 1 )th stop. Pro vide that each node tra v els at a constant speed between tw o consecuti v e stops, the tra v elling speed of node i from pos ( i; j ) to pos ( i; j + 1) is l eng th ( z }| { pos i ( j ) pos i ( j + 1) ) t i ( j +1) )Tj/T1_3 7.97011 Tf6.5833 0 Td(t i ( j ) )Tj/T1_3 7.97011 Tf6.5833 0 Td(pause i ( j ) Notice that node i can follo w potentially man y dif ferent paths, for e xample, a straight one or a zigzag one or e v en a tortuous one, as long as the time constraint is satised, that is, it can reach pos i ( j + 1) at the time instance t i ( j ) Ho we v er once the path between tw o consecuti v e stops is determined, the v elocity of the node between these tw o stops is determined and x ed. Therefore, once the paths between all pairs of consecuti v e stops are determined, the mo v ement pattern of a node during T is also determined. The general mobility model (GMM) described abo v e bears both similarities and differences with the random w aypoint model (R WM), which is the most commonly-used mobility model in simulating MANET protocols [58,70]. Both models are characterized by a collection of locations of ne xt stops, tra v elling speeds, and tra v elling time. Dif ferent from GMM, R WM requires a node to rst choose the location of its ne xt stop and the tra v elling speed, which leads to the determination of the tra v elling time. Our GMM actually does the opposite by rst determining the ne xt visited location and the tra v elling time so as to determine the tra v elling speed. The biggest dif ference, ho we v er is that, in R WM nodes al w ays tra v el along the straight paths connecting tw o consecuti v e stops, while in GMM, nodes can tra v el along arbitrary paths as long as the y do not violate the time requirements, that is, the y should arri v e at the designated stops at the specied time instances. W e belie v e that our GMM outperforms R WM in reecting some practical scenarios, which can be seen from a simple e xample in daily life. Suppose a student carrying a mobile de vice should be in the classroom at x ed time e v eryday He/she might ha v e se v eral options of arri ving in time at the classroom from his/her apartment: by foot through the shortest path with the longest time, by bic ycle through the second shortest path with the second

PAGE 66

54 longest time, or by b us through the longest path b ut with the shortest time. Ob viously R WM f ails to model this case, b ut our GMM can. 3.3.2 Resource A w are Mo v ement As mentioned before, in the general mobility model there might be potentially man y dif ferent paths between an y tw o consecuti v e stops. Dene P AT H ( pos i (0) ; pos i ( J 1 )Tj/T1_2 11.95509 Tf12.1512 0 Td(1)) as node i s path set which is the concatenation of all the paths z }| { pos i ( j ) pos i ( j + 1) In this dissertation, we are interested in nding the optimal path sets for all the nodes such that the total ener gy consumption for communications by all the nodes during the observ ation period T is minimized (the objecti v e function), while all the ordered lists of visited locations and the corresponding time instances should not be violated (the constraints). T o help better understand the importance of this problem, we utilize the mo v ement of a single node between tw o consecuti v e stops as an e xample. As sho wn in Fig.3, suppose an B-node A should mo v e from the current location 1 to the ne xt location 2. It can choose the shortest straight path (the dashed one) as it does in the random w aypoint model. Ho we v er considering that node A may forw ard or generate pack ets destined for other nodes during the mo v ement process, the shortest straight path is not necessarily the best one for achie ving the system-wide ener gy ef cienc y Instead, the dotted and solid paths are much better candidates through which node A can tak e adv antage of more P-nodes by forw arding to the encountered P-nodes the pack ets destined for other nodes and letting them nish the rest of the task. Due to this reason, we call this problem the r esour ce-awar e mo vement problem in that nodes no w are mo ving with the system-wide resource (ener gy) consumption in mind instead of mo ving blindly as before. The general resource-a w are mo v ement problem itself is f ar too complicated to be solv able. T o render it tractable, we mak e some approximation and decouple it into tw o relati v ely simpler subproblems: the W aterhunter Mo v ement problem and the Firehunter Mo v ement problem. In the former we assume that only B-nodes are capable of mo ving

PAGE 67

55 D E F C A A B G T r a n s m i s s i o n R a n g e P n o d e R n o d e H I 2 1Figure 3: Multiple paths between tw o consecuti v e stops. and all the P-nodes are stationary whose locations are kno wn a priori to B-nodes. By contrast, in the latter we assume that all the B-nodes are stationary and only P-nodes are able to mo v e. W aterhunter Mo v ement : In the netw ork with N p stationary P-nodes and N r mobile B-nodes, gi v en all the order lists of f pos i ( j ) ; t i ( j ) ; pause i ( j ) ; 0 j < J i g the W aterhunter Mo v ement Problem1is to determine the optimal tra v elling path set for each B-node such that the total ener gy consumption of the whole netw ork during T is minimized. Fir ehunter Mo v ement : In the netw ork with N p mobile P-nodes and N r stationary B-nodes, gi v en all the order lists of f pos i ( j ) ; t i ( j ) ; pause i ( j ) ; 0 j < J i g the Firehunter Mo v ement Problem2is to determine the optimal tra v elling path set for each P-node such that the total ener gy consumption of the whole netw ork during T is minimized. 1 If we compare ener gy resources to w ater, the mo v ement of B-nodes is similar to the beha vior of w ater -hunters who are al w ays looking for fountains (P-nodes), hence the name. 2 If we compare ener gy resources to w ater, the mo v ement of P-nodes is similar to the beha vior of re-hunters who are al w ays looking for places on re or lack of w ater (B-nodes), hence the name.

PAGE 68

56 The W aterhunter Mo v ement problem is similar to the determination of people' s optimal tra v elling plans when lots of airline hubs and stops are a v ailable. On the contrary the Firehunter Mo v ement problem bears similarity with the planning of hub locations when airline companies b uild their global transportation netw orks with the purpose of pro viding more con v enience for passengers while reducing the o v erall system cost. Both problems are pretty interesting and w orth y of rigorous study Ho we v er due to the space limitation, we focus on nding a nearly optimal solution to the W aterhunter Mo v ement problem in this dissertation. Our in v estig ation on the Firehunter Mo v ement problem and the general Resource-A w are Mo v ement problem will be the future w ork. 3.4 W aterhunter Mo v ement In this section, we rst present a simplied v ersion of the W aterhunter Mo v ement problem, which is further reduced to a NP-complete distance-constrained least-cost (DCLC) problem. W e then present an ef cient heuristic solution. Finally we propose a routing delay dif ferentiation mechanism to utilize the benets resulting from the resource-a w are mo v ement. 3.4.1 Simplied W aterhunter Mo v ement Problem In the W aterhunter Mo v ement problem, we assume that all the N p nodes are stationary during the observ ation period T and are willing to forw ard pack ets for other less po werful B-nodes. F or simplicity we do not dwell on ho w to place P-nodes to attain the optimal system performance, which is belie v ed to be a challenging problem itself and is currently under in v estig ation. Instead, we assume that each B-node kno ws the locations of all the P-nodes and its o wn location at an y time, and can as well adjust its mo ving direction at will. F or the time being, we assume here that all the P-nodes and B-nodes ha v e the same transmission range T R W e will discuss the case that P-nodes ha v e the lar ger transmission range than B-nodes in Section3.4.4. It is w orth pointing out that the ndings in this dissertation can be easily e xtended to the case that each node has indi vidual transmission range.

PAGE 69

57 The original W aterhunter Mo v ement problem aims at minimizing the total ener gy consumption of the whole netw ork during the observ ation period T which is a global optimization problem and still too hard to be solv able. T o mak e it tractable, we ha v e to mak e some approximations to get a suboptimal solution. W e assume that a B-node mo v es at a constant speed in one epoch, that is, between tw o consecuti v e stops, and the maximum speed it can tak e is a system-wide v alue speed max Therefore, the longest path node i can tra v el in one epoch is l max = speed max ( t i ( j + 1) )Tj/T1_1 11.95509 Tf12.0791 0 Td(t i ( j ) )Tj/T1_1 11.95509 Tf12.0791 0 Td(pause i ( j )) T o achie v e the system-wide goal of ener gy conserv ation, instead of mo ving along the straight path connecting tw o consecuti v e stops pos i ( j ) and pos i ( j + 1) node i may tra v el along a resource-a w are path with the purpose of letting the encountered P-nodes forw ard on behalf of it as man y as possible pack ets destined for other nodes. F or simplicity we assume that when mo ving to w ards a P-node, node i al w ays goes along the straight path connecting the destined P-node and itself. Therefore, z }| { pos i ( j ) pos i ( j + 1) is a set of zigzag straight paths if there e xist multiple P -nodes. Notice that the simplied tar get no w is to nd an optimal path set for each indi vidual node to minimize its total ener gy consumption during the observ ation period T instead of that of the whole netw ork. W e intend to utilize the solutions to this local optimization problem to approximate the ones to the original global optimization problem, which is belie v ed to be too complicated to be tractable. Normally when mo ving between tw o consecuti v e stops pos i ( j ) and pos i ( j + 1) node i may ha v e se v eral potential P-nodes to utilize. It is, ho we v er usually unwise for node i to pass by each of them. The reason is that, the longer path node i tak es, the f aster speed it should mo v e at, as described in the aforementioned general mobility model. It is well-kno wn that a f aster mo v ement speed may cause some undesirable problems such as the instability of routing paths and the drop of pack ets. Therefore, some rules should be designed to guide each B-node in deciding which P-nodes and in what order it should pass through between tw o consecuti v e stops. A simple rule w ould be to only consider as candidates the P-nodes whose distance from the direct link between an y tw o consecuti v e

PAGE 70

58 B E d H s A r Figure 3: An e x emplary complete graph. steps are no more than a threshold F or e xample, can equal 1 : 5 T R where T R is the transmission range of each node. W ith node i as an e xample, we can put the simplied W aterhunter Mo v ement problem in another w ay: gi v en a source ( i s current stop), a destination ( i s ne xt stop), and some a v ailable intermediate P-nodes, and the path length constraint l max nd a path as ener gyef cient as possible from the source to the destination, which is might be either the direct link connecting the source and destination or a zigzag path through multiple P-nodes. Fig.3depicts such a topology where a rectangular area, called the -bounded r ectangular ar ea hereafter is formed such that only the P-nodes residing in this area are considered as v alid candidates. In addition, each link is of the forw ard direction from the source s to the destination d simply because tra v elling backw ard is ener gy inef cient. W e assign to each link tw o weights, of which one represents the ph ysical distance between tw o link ends and the other indicates the virtual ener gy cost (dened shortly) incurred by choosing this link. The simplied W aterhunter Mo v ement problem can be boiled do wn to a distanceconstrained least-cost (DCLC) [71] routing problem which is formally dened as follo ws. Consider a directed netw ork that can be modelled as a complete graph G = ( V ; E ) where V is the set of v ertices consisting of the source, the destination, and all the v alid candidate P-nodes, and E is the set of edges connecting each pair of nodes. V can be further di vided into tw o subsets, namely U including the source s and destination d and P containing all the P-nodes. In addition, each edge e 2 E represents the mo v ement from the

PAGE 71

59 D E F C A A B G T r a n s m i s s i o n R a n g e P n o d e R n o d e H IFigure 3: An e x emplary resource-a w are mo v ement. tail node to the head node. Let R + denote the set of non-ne g ati v e real numbers. Each edge e 2 E is associated with tw o non-ne g ati v e functions: a distance function dist ( e ) : E R + representing the ph ysical distance between the end nodes of e and an ener gy cost function cost ( e ) : E R + S f 0 g More specically for a gi v en edge e ( v i ; v j ) its ener gy cost is dened as cos t ( e ( v i ; v j )) = 8 > > > > < > > > > : f ( dist ( v i ; v j )) + g ( dist ( v i ; v j )) v i = s; v j = d f ( dist ( v i ; v j )) + g ( dist ( v i ; v j ) )Tj/T1_10 10.9091 Tf10.89729 0 Td(T R ) v i = s; v j 2 P or v j = d; v i 2 P f ( dist ( v i ; v j )) + g ( dist ( v i ; v j ) )Tj/T1_9 10.9091 Tf10.89729 0 Td(2 T R ) v i ; v j 2 P : Here f is the cost, such as g as, fuel or other types of resources, required for the mechanical mo v ement3; g is used to reect the cost for communications. g can be an y non-decreasing function that con v erts a gi v en distance v alue into a non-ne g ati v e cost, for e xample, g ( x ) = 8 > < > : x x > 0 0 x 0 : 3 In this chapter for simplicity we assume that people on foot carry the communities de vices and we do not tak e the cost for the mechanical mo v ement into account, that is, f ( ) = 0 .

PAGE 72

60 The moti v ation for the abo v e denition of the edge ener gy cost is as follo ws. Whene v er a B-node mo v es into the transmission range TR of a P-node, it is capable of forw arding to the P-node pack ets destined for other nodes so as to conserv e ener gy At one e xtreme, if a B-node mo v es along an edge not (partially) co v ered by an y P-node, all the pack ets from this B-node w ould be forw arded to other ener gy-constrained B-nodes, which is the most unf a v orable situation. At the other e xtreme, if a B-node mo v es along an edge completely co v ered by one or se v eral P-nodes, all the pack ets from this node could be forw arded to the P-node(s), which is the most desirable situation. Notice that the ener gy cost function gi v en abo v e can well capture this ef fect. Though there might e xist other meaningful metrics, we belie v e the chosen one is v ery simple and tractable. W e also dene the non-ne g ati v e delay and cost functions for an y path p as dist ( p ) = X e 2 p dist ( e ) and cost ( p ) = X e 2 p cost ( e ) : Gi v en the abo v e denitions, the DCLC routing problem is to nd a path p from s to d such that min f cost ( p ) ; p 2 P d g is achie v ed, where P d is the set of all feasible paths from s to d satisfying the distance constraint l max that is, dist ( p ) l max Moreo v er we dene P l d ( s; d ) as the path with the least distance from s to d and P l c ( s; d ) as the path with the least cost from s to d Apparently with the abo v e denition of dist ( e ) P l d ( s; d ) is the straight path directly connecting s and d It has been sho wn in [72] that the DCLC routing problem is NP-complete e v en for undirected netw orks. In the follo wing section we will propose an ef cient heuristic algorithm to pro vide a suboptimal solution to this DCLC problem and hence to the original W aterhunter Mo v ement problem.

PAGE 73

61 T able 3: RAM: resource a w are mo v ement 1. Determine the candidate P-node set C H S in the bounded r ectangular ar ea ; 2. Construct a complete graph G with virtual nodes s and d and all the nodes in C H S ; 3. Label each link e in G with cost ( e ) and dist ( e ) ; 4. Find the DCLC path P dcl c by calling RAM-DCLC( G s t l max ); 5. Determine the tra v elling speed along P dcl c as speed dcl c = dist ( P dclc ) t ( i;j +1) )Tj/T1_4 7.97011 Tf6.5833 0 Td(t ( i;j ) )Tj/T1_4 7.97011 Tf6.5833 0 Td(pause ( i;j ) ; 6. Mo v e along P dcl c at a speed of speed dcl c ; 3.4.2 RAM-DCLC Algorithm As mentioned before, we assume that a B-node, say i is a w are of its o wn itinerary f pos i ( j ) ; t i ( j ) ; pause i ( j ) ; 0 j < J i g and the locations of all the P-nodes during the observ ation period T The procedure of node i s resource-a w are mo v ement from the cur rent location pos i ( j ) to the ne xt location pos i ( j ) is summarized in T able3. Node i rst needs to determine the candidate P-nodes in the -bounded r ectangular ar ea and then constructs a complete graph lik e the one in Fig.3, consisting of the v ertices s (a virtual node at pos i ( j ) ), d (a virtual node at pos i ( j + 1) ), and all the found candidate P-nodes. It then proceeds to calculate the distance and the ener gy cost for each link and nally generates the weighted graph G The ne xt step is to call the process RAM-DCLC gi v en in T able3to get the DCLC path P dcl c whose length is bounded by l max = speed max ( t i ( j + 1) )Tj/T1_1 11.95509 Tf12.0611 0 Td(t i ( j ) )Tj/T1_1 11.95509 Tf12.05209 0 Td(pause i ( j )) It then mo v es to w ards pos i ( j + 1) at a constant speed of speed dcl c = dist ( P dclc ) t ( i;j +1) )Tj/T1_4 7.97011 Tf6.5833 0 Td(t ( i;j ) )Tj/T1_4 7.97011 Tf6.5833 0 Td(pause ( i;j ) along the found P dcl c Upon reaching pos i ( j + 1) node i pauses for a period pause i ( j + 1) F ollo wing the pre vious process, it can then mo v e to w ards the ne xt stop pause i ( j + 2) until all the required stops f pos i ( j ) ; 0 j < J i g during the observ ation period T are visited.

PAGE 74

62 T able 3: RAM-DCLC: a DCLC routing algorithm for the W aterhunter Mo v ement 1. F or each node v k in G nd the P l c ( v k ; d ) and P l d ( v k ; d ) and their respecti v e ne xt hops nid ( P l c ( v k ; d )) and nid ( P l d ( v k ; d )) ; 2. distS oF ar = 0 ; P dcl c = s ; T hisN ode = s ; 3. while ( T hisN ode 6= d ) do 4. if ( ( dist ( P l c ( T hisN ode; d )) + distS oF ar ) l max ) then 5. v = nid ( P l c ( T hisN ode; d )) ; 6. distS oF ar = distS oF ar + dist ( T hisN ode; v ) ; 7. P dcl c = P dcl c + f v g ; 8. T hisN ode = v ; 9. else 10. f or each neighboring node w = 2 P dcl c do 11. calculate w eig ht ( T hisN ode; w ) ; 12. end f or 13. v = extr act ( T hisN ode ) ; 14. distS oF ar = distS oF ar + dist ( T hisN ode; v ) ; 15. P dcl c = P dcl c + f v g ; 16. T hisN ode = v ; cp:ref35 17. end if 18. end while 19. Retur n P dcl c

PAGE 75

63 The proposed DCLC algorithm RAM-DCLC is summarized in T able3, in which the w eig ht () function is dened as follo ws: w eig ht ( v i ; v j ) = 8 > < > : cost ( v i ; v j ) + cost 0 ( v j ; d ) cond (1) + 1 o:w : ; and cost 0 ( v j ; d ) = 8 > < > : cost ( P l c ( v j ; d )) cond (2) cost ( P l d ( v j ; d )) o:w : ; where cond (1) = distS oF ar + dist ( v i ; v j ) + dist ( P l d ( v j ; d )) l max and cond (2) = distS oF ar + dist ( v i ; v j ) + dist ( P l c ( v j ; d )) l max : The function extr act () is used to choose the node, say w whose w eig ht ( v i ; w ) is the minimum one among all the neighboring nodes of T hisN ode If more than one node ha v e the same minimum v alue, it chooses the one with the smallest distS oF ar + dist ( v i ; v j ) + dist ( P l d ( v j ; d )) In the RAM-DCLC algorithm, for each node v k in G the Bellman-F ord or Dijkstra shortest-path algorithm can be used to nd the P l c ( v k ; d ) and P l d ( v k ; d ) and their respecti v e ne xt hops nid ( P l c ( v k ; d )) and nid ( P l d ( v k ; d )) Since the optimization objecti v e is the path cost, at each intermediate node v RAM-DCLC al w ays chooses the ne xt hop w with minimum cost ( v ; w ) + cost 0 ( w ; d ) while not violating the distance constraint l max RAM-DCLC is able to nd a feasible path satisfying the distance constant while k eeping the path cost as small as possible. In particular we ha v e the follo wing theorems for this algorithm4. Theor em 1 : RAM-DCLC can al w ays nd a feasible path from a source s to a destination d satisfying the gi v en distance constraint l max if such feasible paths e xist. 4 The correctness of these theorems can be justied follo wing the proof in [73].

PAGE 76

640 2 0 0 4 0 0 6 0 0 8 0 0 1 0 0 0 1 2 0 0 1 4 0 0 1 6 0 0 0 5 0 1 0 0 1 5 0 2 0 0 2 5 0 3 0 0 1 3 1 2 1 1 1 0 9 8 7 6 5 4 3 2 1 ( S t o p A r r i v a l t i m e ) ( S5, 8 5 8 4 ) ( S4, 6 2 5 8 ) ( S3, 3 8 7 4 ) ( S2, 2 4 7 8 ) ( S1, 0 ) R A M R W M P n o d e s S t o p s Y ( m )X ( m )Figure 3: Resource-a w are mo v ement (RAM) vs. random w aypoint mo v ement. A Bnode should consecuti v ely visit S 1 S 2 S 3 S 4 S 5 and the solid lines are labelled by the sequences the y were passed through. Theor em 2 : The path found by RAM-DCLC is loop-free. Theor em 3 : RAM-DCLC al w ays terminates in nite time. No w we utilize the e xample gi v en in Fig.3to illustrate the resource-a w are mo v ement process using the proposed R AM )Tj/T1_2 11.95509 Tf12.2141 0 Td(D C LC algorithm. Suppose node A intends to mo v e from its current location to the location where A resides according to its itinerary In this e xample is set to 1 : 5 T R so that there are four candidate P-nodes. Based on the output of R AM )Tj/T1_2 11.95509 Tf12.5741 0 Td(D C LC node A should mo v e along the DCLC path denoted by the solid line instead of the straight path denoted by the dash line. In this w ay ener gy sa vings can be e xpected by forw arding to the tw o P-nodes the pack ets it carries for other nodes and letting the P-nodes nish the rest transmissions (either single-hop or multi-hop) on behalf it. Fig.3compares a B-node' s resource-a w are mo v ement trail and its random w aypoint mo v ement trail with 8 P-nodes in a 1500 300 m 2 eld. The data used were generated using OPNET [57] and the pause time of the B-node w as 120s. In addition, the transmission range of each node w as set to 250m. It is clear that our proposed resourcea w are mo v ement strate gy enables the B-node to ha v e more opportunities of approaching and utilizing the P-nodes as compared to the random w aypoint mo v ement.

PAGE 77

65 3.4.3 A Routing Delay Dif ferentiation Mechanism The resource-a w are mo v ement strate gy only enables B-nodes to mo v e close to Pnodes by taking the ph ysical DCLC paths, b ut cannot guarantee a pack et will be forw arded from a B-node to a P-node e v en when the y are close to each other As a result, we need to incorporate such nodal information as de vice types and residual ener gy into the routing metric such that the P-nodes can be naturally selected onto the routing paths of the Bnodes. F or this purpose, we propose to map the residual ener gy information into the random jitter delay for which each node has to w ait before propag ating a route request in a re gular MANET routing protocol such as DSR [70]. The rational here is that the less residual ener gy a node has, the longer delay a routing request pack et should e xperience at this node. In this w ay the nodes with less residual ener gy w ould be less utilized, while the nodes with more residual ener gy such as P-nodes w ould be more utilized. Belo w we illustrate the delay dif ferentiation mechanism using DSR [70] as the under lying routing protocol, though our mechanism can as well be applied to other MANET routing protocols. Similar to the interframe spacing (IFS) dif ferentiation mechanism adopted in the IEEE 802.11e [74], we rst dene the Delay Dif fer entiation as E 0 =E t where is a delay control parameter and E 0 and E t are the residual ener gy at time instance 0 and t respecti v ely When recei ving a route request, each node should rst compute the delay it needs to w ait for before further propag ating the route request. Since the P-nodes ha v e relati v ely unlimited residual ener gy as compared to the B-nodes, that is, E 0 E t the y w ould forw ard route requests much more quickly than the B-nodes. Since the destination usually only replies to the rst arri v ed request, the P-nodes with po werful ener gy resources w ould ha v e higher chances of being chosen onto the routing paths and thus being in v olv ed in more data forw arding acti vities than the B-nodes with less ener gy W ith this delay differentiation mechanism, the benets resulting from the resource-a w are mo v ement can be better utilized. The nice feature of our mechanism is that it is v ery easy to implement.

PAGE 78

66 3.4.4 Incorporate RAM into DELAR In our pre vious discussion, we assume that the P-nodes ha v e the same transmission range T R as the B-nodes. Ho we v er in more practical heterogeneous MANETs, P-nodes may ha v e greater transmission capabilities than the B-nodes. T o mak e use of such po werful P-nodes especially when P-nodes can adjust their transmission po wer to co v er lar ger range than B-nodes, in our pre vious w ork DELAR, we proposed a ener gy ef cient relaying frame w ork as a joint design of scheduling, routing and po wer control to ef ciently utilize P-nodes to conserv e ener gy Since RAM and DELAR utilize po werful nodes to conserv e ener gy from totally orthogonal perspecti v es, RAM can be directly incorporated into the DELAR scheme. W ith the help of DELAR, RAM can further impro v es the ener gy ef cienc y when P-nodes and B-nodes ha v e dif ferent transmission capabilities. 3.5 Performance Ev aluation In this section, we use simulations to e v aluate the impact of the proposed schemes on the ener gy conserv ation and other system performance f actors. 3.5.1 Simulation Setup W e implemented the resource-a w are mobility model, the routing delay dif ferentiated mechanism, and DELAR (cf. Chapter2) in OPNET [57]. W e simulated a netw ork with N r B-nodes and N p P-nodes in a 1500 300 eld, where N r = 46 and N p = 4 All the B-nodes were capable of mo ving in the eld, while all the P-nodes were x ed. Though a careful deplo yment of P-nodes may impro v e the system performance [75], we simulated a w orse scenario that the P-nodes were randomly deplo yed in the eld. The transmission range of the B-nodes w as 250 m. F or the P-nodes, we simulated tw o cases in which the P-nodes had the transmission ranges 250 m and 500 m, respecti v ely The ener gy consumption for the B-nodes follo wed the linear ener gy model proposed in [76]: ener g y = m l eng th + b where m is an incremental cost of each operation, b is the

PAGE 79

67 T able 3: Ener gy consumption parameters. Symbol V alue Unit m send 1.89 uW -sec/byte b send 246 uW -sec m r ecv 0.49 uW -sec/byte b r ecv 56.1 uW -sec b sendctl 120 uW -sec b r ecv ctl 29 uW -sec x ed cost of each operation, and l eng th is the size of the frame sent/recei v ed5. The ( m; b ) v alues pro vided in [76] were summarized in T able3and used in all the calculations. W e intended to compare the proposed resource-a w are mobility model (denoted by RAM) with the modied random w aypoint model (denoted by R WM) presented in [58], which can guarantee the con v er gence of a v erage nodal speed throughout the simulation time. F or this purpose, we rst ran the simulations using R WM and recorded the stops, the starting/arri v al time instances, the mo ving directions, and the mo v ement speeds of all the mo v ement epochs. W e then used this mo v ement prole to generate the itinerary f pos i ( j ) ; t i ( j ) ; pause i ( j ) ; 0 j < J i g for each node such that in both models each node w ould stop by the same stops at the same time instances, b ut follo w totally dif ferent mo v ement trails and tak e dif ferent mo v ement speeds. Both models had the same maximum speed 20 m/s and we adjusted nodal pause time to v ary the netw ork mobility The traf c used were 20 CBR connections with randomly selected source-destination pairs. All the data pack ets were 64 bytes and were sent a speed of 4 pack ets/second. Each simulation w as e x ecuted for 15 simulated minutes and each data point represents an a v erage of ten runs with identical traf c models, b ut dif ferently generated mobility scenarios. 5 F or the A-MA C control frames P-R TS/P-CTS/P-A CK, the x ed costs b sendctl and b r ecv ctl were used because the y ha v e the similar size.

PAGE 80

68 3.5.2 Simulation Results W e compared RAM with R WM in terms of the commonly used metrics including pack et deli v ery ratio, a v erage pack et end-to-end delay a v erage pack et ener gy consumption, and a v erage routing o v erhead. Moti v ated by the small-w orld phenomenon [77], we used tw o additional metrics, a v erage path length and a v erage clustering coef cient, which are tw o dening characteristics of small-w orld netw orks. The former means the a v erage number of hops a pack et may tra v el through, while the latter indicates the connecti vity of an a v erage neighborhood in the netw ork, dened as the a v erage node de gree di vided by the netw ork size [77]. The simulation results are presented in Fig.3, where RAM-1 indicates the case that the P-nodes and B-nodes ha v e the same transmission range 250 m and RAM-2 denotes the case that the P-nodes ha v e a lar ger transmission range 500 m. Fig.3(a)and3(b)compare RAM and R WM with re g ard to a v erage path length and a v erage clustering coef cient, respecti v ely W e can see that RAM can shorten the a v erage path length and increase the clustering coef cient as compared to R WM. That is because in RAM the B-nodes are al w ays trying to mo v e to w ards some P-nodes between tw o consecuti v e stops. Such beha viors w ould ef fecti v ely bring more B-nodes to the vicinity of P-nodes, leading to shorter paths and lar ger clustering coef cients. In some sense, such resource-a w are mo v ement creates a small-w orld netw ork, which results in some performance g ains sho wn belo w In addition, since the P-nodes in RAM-2 ha v e a lar ger transmission range and thus ha v e more neighbors that those in RAM-1, we can observ e that RAM-2 further reduces the a v erage path length and increases the a v erage clustering coef cient. Fig.3(c)compares the a v erage pack et deli v ery ratios of RAM and R WM, which is dened as the ratio of deli v ered data pack ets to those generated by the sources. As we can see, the PDR of RAM-1 or RAM-2 is al w ays higher than that of R WM. This result is of no surprise since the shorter a v erage path length implies the netw ork-wide less traf c load and less pack et drops due to the MA C-layer contention and channel errors, all of which

PAGE 81

69 0 1 0 0 2 0 0 3 0 0 4 0 0 5 0 0 6 0 0 7 0 0 8 0 0 9 0 0 1 8 2 2 2 2 4 2 6 2 8 3 3 2 3 4 3 6 P a u s e t i m e ( s )A v e r a g e p a t h L e n g t h R W M R A M 1 R A M 2 (a) A v erage path length. 0 1 0 0 2 0 0 3 0 0 4 0 0 5 0 0 6 0 0 7 0 0 8 0 0 9 0 0 0 2 6 0 2 8 0 3 0 3 2 0 3 4 0 3 6 0 3 8 0 4 0 4 2 0 4 4 P a u s e t i m e ( s )C l u s t e r i n g c o e f f i c i e n t ( m / s ) R W M R A M 1 R A M 2 (b) A v erage clustering coef cient. 0 1 0 0 2 0 0 3 0 0 4 0 0 5 0 0 6 0 0 7 0 0 8 0 0 9 0 0 0 9 5 0 9 5 5 0 9 6 0 9 6 5 0 9 7 0 9 7 5 0 9 8 0 9 8 5 0 9 9 0 9 9 5 1 P a u s e t i m e ( s )P a c k e t d e l i v e r y r a t i o R W M R A M 1 R A M 2 (c) P ack et deli v ery ratio. 0 1 0 0 2 0 0 3 0 0 4 0 0 5 0 0 6 0 0 7 0 0 8 0 0 9 0 0 0 0 0 2 0 0 0 4 0 0 0 6 0 0 0 8 0 0 1 0 0 1 2 0 0 1 4 0 0 1 6 0 0 1 8 P a u s e t i m e ( s )P a c k e t e n d t o e n d d e l a y ( s ) R W M R A M 1 R A M 2 (d) A v erage pack et end-to-end delay 0 1 0 0 2 0 0 3 0 0 4 0 0 5 0 0 6 0 0 7 0 0 8 0 0 9 0 0 1 1 5 2 2 5 3 3 5 P a u s e t i m e ( s )A v e r a g e e n e r g y c o n s u m p t i o n ( m W ) R W M R A M 1 R A M 2 (e) A v erage ener gy consumption. 0 1 0 0 2 0 0 3 0 0 4 0 0 5 0 0 6 0 0 7 0 0 8 0 0 9 0 0 0 5 1 0 1 5 P a u s e t i m e ( s )A v e r a g e r o u t i n g o v e r h e a d R W M R A M 1 R A M 2 (f) A v erage routing o v erhead. Figure 3: Simulation results.

PAGE 82

70 w ould contrib ute to the increase of the PDR. Due to the same reason, RAM-2 demonstrates a higher PDR than RAM-1. Figure 3-5(d) depictsthecomparisonofaverageend-to-endpacketdelay,denedas the time duration from when a pack et is generated till it is recei v ed by the destination. The sho wn adv antage of RAM o v er R WM mainly results from the aforementioned shorter a v erage path length and higher clustering coef cient. Ag ain, RAM-2 outperforms RAM-1 in reducing a v erage pack et delay because of the shorter a v erage path length. Figure 3-5(e) showsaverageenergyconsumption,denedasthetotalenergyconsumption for transmitting and recei ving all data and routing pack ets di vided by the number of deli v ered pack ets. Apparently our RAM can conserv e a signicant amount of ener gy as compared to R WM because the B-nodes are al w ays mo ving along the paths through which the P-nodes can be utilized as much as possible. Since the P-nodes in RAM-2 ha v e a lar ger transmission range, statically less B-nodes are in v olv ed in pack et transmissions and thus RAM-2 can help the system conserv e more ener gy than RAM-1. Figure 3-5(f) demonstratesaverageroutingoverhead,denedastheaveragenumber of routing pack ets in v olv ed in deli v ering 100 data pack ets. As we can see, our RAM has smaller routing o v erhead than R WM. That is because in RAM pack ets can be forw arded to their destinations through shorter paths in shorter time, thus fe wer routing errors occur T o summarize, the proposed resource-a w are mo v ement strate gy has man y signicant and positi v e impacts on the system performance. It mak es the netw ork more lik e a smallw orld netw ork with shorter a v erage path length and higher clustering coef cient. This results in impro v ed pack et deli v ery ratio, shortened end-to-end delays, and most impor tantly much better ener gy conserv ation. Therefore, the combination of node mobility and heterogeneity is a v alid means to address the ener gy conserv ation issue in MANETs. The results also suggest that making the netw ork a small-w orld netw ork may ha v e a lot of positi v e ef fects on the system performance and it deserv es further in v estig ation.

PAGE 83

71 3.6 Summary In this chapter we studied the ener gy conserv ation problem from the ne w perspecti v e of node mobility that is, by analyzing the impact of node mo v ement on the system-wide ener gy conserv ation. W e proposed a no v el resource-a w are mo v ement strate gy to tak e full adv antage of some po werful nodes in heterogeneous mobile ad hoc netw orks. W e then for mulated the resource-a w are mo v ement as a NP-complete distance-constrained least-cost (DCLC) routing problem and proposed an ef cient heuristic solution. Moreo v er we proposed a simple yet ef fecti v e routing delay dif ferentiation mechanism to virtually utilize the benets from the resource-a w are mo v ement. In addition, the proposed resource a w are mo v ement strate gy can be incorporated into other ener gy conserv ation schemes, for e xample, DELAR, to further impro v e the ener gy ef cienc y W e used e xtensi v e simulations to sho w the ef fecti v eness of the proposed scheme.

PAGE 84

CHAPTER 4 SUPPOR T DIFFERENTIA TED SER VICES IN MOBILE AD HOC NETW ORKS 4.1 Introduction In the literature there are proposals addressing the issue of quality of service in wired netw orks, ho we v er these proposals may not be feasible in the wireless counterpart if we do not modify them for the wireless en vironments. The system dynamics [18] of multi-hop mobile ad hoc netw orks, such as time-v arying and error -prone wireless links, dynamic and limited bandwidth, time-v arying traf c pattern and user location, and ener gy constraints, pose ne w challenges that do not e xist in wired netw orks. T o conquer these challenges, in recent years, man y researchers adv ocate a cross-layer design philosoph y to de v elop protocols and applications for MANETs. This is a departure from the traditional layered design for the Internet. Though the cross-layer design philosoph y might not be an optimal solution, it does pro vide us ne w netw ork implementations that may better support the amalg amation of user services and QoS requirements [19]. T o ef ciently handle heterogeneous traf c o v er wireless links, we need to address tw o problems. The rst is to handle reliable mobile communications in MANETs. This problem has been e xtensi v ely studied in recent years, and man y proposed routing protocols such as DSD V [7], DSR [9], and A OD V [10], and medium access control mechanisms such as MA CA W [3], F AMA [4], and IEEE 802.11 [5], aim to achie v e ef cient reliable communications. The other problem is to pro vide QoS pro visioning for heterogeneous traf c with dif ferent quality-of-service (QoS) requirements in terms of BER, throughput, and delay Since the channel bandwidth in wireless en vironments is limited, one strate gy to support QoS is to set up some kind of priority scheme or service dif ferentiation mechanism [78][79], under which delay-sensiti v e traf c has higher priority to access the channel o v er less time-critical traf c. 72

PAGE 85

73 In the current literature, man y scheduling mechanisms for wireless netw orks are proposed for this purpose, though most of them are not directly designed for MANETs. In general, these scheduling mechanisms all attempt to combat the channel impairments and to support heterogeneous traf c with the follo wing goals: pro viding high wireless channel utilization, long-term f airness, bandwidth guarantees and delay bounds for o ws with error -free links or links with sporadic errors [80]. Ho we v er these algorithms may not be practical to be implemented in MANETs. Actually it is hard, if not impossible, to achie v e those goals simultaneously because of their conicting nature. F or e xample, there is a tradeof f between the throughput and f airness1or so-called inter -class ef fects [81] among traf c with dif ferent priorities. W ithout an y precautionary measures, this conict may lead to bandwidth starv ation for lo w-priority traf c when the high-priority traf c load is high. Meanwhile, most of these scheduling mechanisms are suitable for the reserv ation-based MA C protocols, especially for those designed for cell-structured wireless netw orks. In netw orks with contention-based MA C protocols such as IEEE 802.11 [5], the reserv ationbased scheduling mechanisms may not be applicable because it is not easy for a node to reserv e resource in a contention manner In this dissertation, we attempt to a v oid the con v entional scheduling approach, and propose a no v el scheme called Courtesy Piggybacking (CP) to alle viate the conict between throughput and f airness. The basic idea of CP is to let the high-priority traf c help the lo w-priority traf c by sharing unused residual bandwidth with courtesy Our scheme closely follo ws the cross-layer design principle and e xploits the system dynamics as much 1 W e only consider the f airness problem between dif ferent classes of traf c, for e xample, each class of service should be allocated some bandwidth rather than being completely starv ed; while the f airness problem between dif ferent nodes, for e xample, each node should ha v e f air opportunity to access the channel in the short or long term, is out of the scope of this dissertation.

PAGE 86

74 as possible; that is, we ef fecti v ely emplo y the dynamic channel conditions and the resulting dynamic bandwidth, and the dynamic characteristics of the heterogeneous traf c. Note that not only is our scheme suitable for multi-hop mobile ad hoc netw orks with underlying contention-based MA C protocols, b ut also it is applicable to those with reserv ation-based or h ybrid MA C protocols. Meanwhile, our scheme is sho wn to be easily implemented. The rest of the chapter is or g anized as follo ws. W e discuss some related w ork in Section4.2. In Section4.3, we sho w the moti v ation of our proposed scheme. In Section4.4, we discuss the relationship between the SNR and the optimal pack et length, and come up with a Finite State Mark o v Chain channel model based on the pack et length. Our Courtesy Piggybacking scheme is described in Section4.5. W e present some preliminary analytical results in Section4.6and e v aluate our scheme with e xtensi v e simulation in Section4.7. Finally we conclude the chapter in Section4.8. 4.2 Related W ork As we mentioned in Section4.1, scheduling is one promising w ay to support heterogonous traf c with dif ferent QoS requirements. F or scheduling mechanisms, throughput and f airness are tw o main objecti v es to be met through bandwidth allocation with admission control and congestion control. Man y scheduling algorithms such as f air queuing scheduling [82], and virtual clock [83] are capable of pro viding certain QoS guarantee for wireline netw orks, and man y scheduling algorithms such as IWFQ [84], CIF-Q [85], CSDPS [86], and CSDPS + CBQ [87] are proposed for the wireless netw orks, especially for wireless cellular netw orks. Ho we v er little progress has been made along this direction in wireless mobile ad hoc netw orks with underlying contention-based MA C protocols. CSDPS and its impro v ed v ersion CSDPS+ CBQ are tw o of scheduling mechanisms that may be applicable to the ad hoc netw orks with contention-based MA C protocols. In CSDPS, pack ets to be transmitted to the same recei v er are queued in the same queue and are serv ed in an FIFO f ashion. At a node, the dif ferent queues are serv ed according to some policies such as round robin, earliest timestamp rst, or longest queue rst. The basic idea

PAGE 87

75 of CSDPS is as follo ws: when the link to w ards a recei v er is bad, the node should defer the transmission of pack ets in the queue corresponding to that recei v er W ith CSDPS, it is easy to alle viate the head of line (HOL) problem when single FIFO queue is used. Since CSDPS mak es use of the channel state information, it can achie v e high data throughput and channel utilization. Ho we v er it does not address the f airness issue. T o impro v e the f airness in CSDPS, class-based queuing (CBQ) [88] is used together with the CSDPS. By using CBQ, a hierarchical channel-sharing mechanism, it can achie v e certain f airness, and ensure that dif ferent traf c classes can share the o v erall bandwidth, while maintaining the features of CSDPS to deal with the channel v ariations. Unfortunately this scheme is also complicated in k eeping track of the amount of service each class has been serv ed. Ef cient and less e xpensi v e mechanisms are v ery desirable to alle viate the conict of throughput and f airness in MANETs. More and comprehensi v e materials on scheduling can be found in [80]. Besides, some QoS adapti v e schemes such as SW AN [16] and Ha v ana [89] are also a v ailable in the literature. These schemes adapti v ely perform admission control and rate control according to the user QoS requirements and channel states. The main reason leading to the conict between throughput and f airness is the limited bandwidth of the wireless link. If the system can pro vide plenty of bandwidth, the conict problem w ould not be so signicant. Recently man y adapti v e transmission techniques are proposed to e xploit the channel dynamics to pro vide more bandwidth. These schemes can adapti v ely adjust the parameters such as modulation le v el and symbol rate to maintain an acceptable BER without w asting much bandwidth. K outsopoulous and T assiulas proposed to inte grate adapti v e transmission techniques, resource allocation and po wer control for TDMA/TDD system so that higher modulation le v els can be assigned to users in good channels to enhance the throughput, while po wer control can be used to reduce the interference and increase the system capacity [90]. In addition to these schemes proposed for wireless cellular netw orks, some rate-adapti v e schemes are also proposed to impro v e the system throughput in WLANs. Holland et al. proposed a rate adapti v e MA C protocol

PAGE 88

76 called RB AR, which uses the R TS/CTS to e xchange the channel state information and the optimal rate on a per -pack et basis [91]. Unfortunately this scheme needs to mak e some modications to the IEEE 802.11 MA C protocols. T o a v oid this modication, Che villat et al. proposed a scheme to select the optimal rate only with the local information at the transmitter [92]. This scheme is based on the history of attempted transmissions. It uses one successful transmission count and one f ailed transmission count to indicate the channel state and to determine the optimal rate the transmitter can use. F or IEEE 802.11 MA C protocols, adapti v e fragmentation schemes can also be designed with the rate adaptation to enhance the system throughput [93][94][95]. F or all the scheduling mechanisms and other channel-dependent schemes, including our Courtesy Piggybacking scheme, designed for wireless netw orks, the y all ha v e to monitor the channel quality based on the symbol error rate, bit error rate, and recei v er signal strength. The more accurate the channel information is, the more benets these schemes can bring to the system design. In general, the channel estimation can be performed by the sender or by the recei v er Since the channel information used in all channel-dependent schemes is the one seen by the recei v er the recei v er -based channel estimation is more attracti v e. Ho we v er the channel information needs to be sent back to the sender which is sometimes costly in terms of the resource used to transmit the channel information, thus certain performance tradeof f has to be made between estimation accurac y and o v erhead. More details about channel quality estimation can be found in [96]. 4.3 Moti v ation Consider the scenario depicted in Fig.4. In a mountain area, the only w ay from A nchorage to W hittier (the access to see the spectacular glacier) is to pass a tunnel near P ortage running through the C hug ach Mountain Range (that is, the longest tunnel in North America the Whittier T unnel in Alaska). The situation is the same from S e w ard to W hittier People ha v e se v eral choices to pass the tunnel: by train (high priority), by car by bic ycle or on foot (lo w priority). Only one direction traf c is allo wed during one period

PAGE 89

77 A P W S T u n n e l R a i l R o a d R o a d Figure 4: The Whittier T unnel scenario. of time. T o pass the tunnel, when the train approaches the tunnel, all other traf c stops and w aits until the train passes the tunnel. Often, there is a long traf c line w aiting to pass the tunnel, especially for the direction from W to P when traf c load is high, for e xample, during rush hour in the afternoon. In order to quickly pass the tunnel, a better approach for other transportation users w ould be to check if there is an y free space left in the train. If there is, these users could ask for permission to ride at a certain cost and according to some rules, for e xample, ho w man y free space in terms of basic units is left and what kind of traf c (priority) the train can accommodate. After passing through the tunnel, the piggyback ed traf c can get of f the train at P and continue on its o wn w ay Of course in the real situations, when passengers by car by bic ycle or on foot pass through the narro w and dark tunnel in a sequential manner the traf c usually mo v es v ery slo wly for the sak e of safety Thus it is advisable for cars that ha v e free space to piggyback those passengers by bic ycle or on foot according to some rules to benet all the traf c. W e can think of these rules as being concerned with the HO W MANY -WHO problem, that is, ho w much free space is a v ailable and who can enjo y such free space? If we only consider the free space F S in the train as a function of time, then we could consider the follo wing scenario as an e xample: one person w ould occup y 1 basic space unit, a bik e 2 units, and a car 6 units. If we ha v e some predened objecti v e to meet, then we can design dif ferent piggybacking rules to solv e the HO W MANY -WHO problem. F or e xample,

PAGE 90

78 suppose our objecti v e is to maximize the re v enue of the train. W ith dif ferent piggybacking costs, for a gi v en F S we can achie v e the optimal allocation scheme for the free space among dif ferent traf c: cars, bic ycles, and pedestrians. The abo v e scenario is v ery similar to multi-hop mobile ad hoc netw orks supporting dif ferentiated services. The piggybacking strate gy described abo v e moti v ates us to de v elop a more ef cient w ay to alle viate the conict between throughput and f airness for dif ferent prioritized services. First of all, we need to identify the free space in a MANET F ortunately we do ha v e tw o sources that can pro vide us with such free space. The rst one comes from the time-v arying channel conditions. In recent studies such as [90] [93], the MA C and PHY layers adapt to the channel state by using adapti v e transmission schemes to pro vide higher data rates when the channel is good. W ith a higher data rate, the transmission time for MA C protocol data unit (MPDU) can be shortened, leading to some potential idle time if the transmitting node does not ha v e further data to transmit. If the IEEE 802.11 MA C is used, the N A V (Netw ork Allocation V ector) setting may pre v ent other nodes from using the medium, e v en though it is idle (the rule of virtual collision a v oidance). This idle period will be the free space and should be more ef fecti v ely used. The second source comes from the traf c characteristics. When we look into the traf c patterns and the stochastic traf c beha vior sometimes the high priority traf c may not ha v e enough data during the reserv ed slots in a reserv ation-based system or their transmission period in a contention-based system (for e xample, a netw ork with IEEE 802.11) to fully utilize the channel capacity F or e xample, consider a netw ork with reserv ation-based MA C protocols. In addition to the free space pro vided by the channel dynamics, when the pack ets from one high-priority o w are not enough to ll the reserv ed slots, for e xample, during silent periods for v oice connections, some free space can be harv ested to piggyback some bits from the queue(s) with lo w priorities. When such free space is a v ailable, the ne xt problem w ould be ho w to mak e use of it to fulll certain objecti v es such as f air allocation of bandwidth. While one w ould think

PAGE 91

79 that it should be used to better support high-priority traf c in the rst place, we ar gue that it may not necessarily be the case. Rather the piggybacking rules should be properly designed in light of specic requirements of v arious applications. If some delay-sensiti v e applications lik e v oice or video-telephon y require that their pack ets get through the netw ork as quickly as possible, then the free space should be used to meet such needs. On the other hand, if high-priority traf c does not need more resource than needed, a piggybacking rule f a v oring lo w priority may be more reasonable. F or streaming multimedia applications, as an e xample, when the QoS requirement of one stream with high priority has already been met, there is no need to piggyback pack ets belonging to this stream ahead of the scheduled time; instead, piggybacking pack ets from other lo w-priority streams may be more benecial. In the follo wing sections, we will elaborate more on wh y free space e xists and ho w piggybacking can be used to achie v e our goal alle viating the conict between throughput and f airness for dif ferent prioritized services. 4.4 P ack et-length-based Channel Model In the current literature, the time-v arying channel is commonly modeled as the wellkno wn Gilbert-Elliott tw o-state Mark o v channel model (Fig.4). Each state in the tw ostate Mark o v chain model represents a binary symmetric channel (BSC). In Good state, the BSC has lo w crosso v er probability P g and int the Bad state, the BSC has high crosso v er probability P b The transition probability matrix can be gi v en as: < = 2 6 4 P GG P GB P B G P B B 3 7 5 : Gi v en the transition probability it is easy to determine that the steady state probabilities are = P B G P B G + P GB P GB P B G + P GB :

PAGE 92

80 G o o d B a d P G B P B G P B B P G G 1 P g 1 P g P g P g 0 1 0 1 1 P b 1 P b P b P b 0 1 0 1 G B Figure 4: The Gilbert-Elliott channel model. W e notice that if P g and P b are set to 0 and 1, respecti v ely that is, a pack et succeeds with probability 1 in the Good state and is lost with probability 1 in the Bad state, the tw o-state model is reduced to the simplied Gilbert model. When the channel quality v aries dramatically it is not accurate enough to model the channel as a tw o-state Gilbert-Elliott model. In this case, a nite-state Mark o v channel (FSMC) [97] can be used. By using the recei v ed signal-to-noise-ratio (SNR) as the only side information, the FSMC pro vides a mathematically tractable model for time-v arying channel. Let r denote the recei v ed SNR that is proportional to the square of the signal en v elop. Then, for a Rayleigh f ading channel, the probability density function of r can be written as f r = 1 )Tj/T1_2 11.95509 Tf-0.39601 -7.54201 Td(r e )Tj/T1_6 5.97758 Tf8.51399 3.69 Td(r )Tj/T1_6 5.97758 Tf0.53999 -4.06799 Td(r ; (4.1) where )Tj/T1_2 11.95509 Tf-0.39601 -7.54201 Td(r is the mean of r (actually it is an e xponential distrib ution with mean )Tj/T1_2 11.95509 Tf-0.39601 -7.54201 Td(r ). In order to b uild the nite state Mark o v chain model, we assume the recei v ed SNR remains at a certain le v el for the duration of a symbol, and we partition the range of the recei v ed SNR into a nite number of interv als. Let 0 = r 0 < r 1 < r 2 < :::::: < r K )Tj/T1_8 7.97011 Tf6.5833 0 Td(1 < r K = 1 be the thresholds. F or each interv al, we associate it with a state S k ; k = 0 ; 1 ; 2 ; 3 ; :::; K )Tj/T1_4 11.95509 Tf12.0791 0 Td(1 The channel is in the state S k if r is in the interv al [ r k ; r k +1 ] W e kno w that there is a crosso v er probability p for a gi v en SNR r When BPSK is used, this probability can be written as a function of r : p ( r ) = 1 )Tj/T1_4 11.95509 Tf11.94411 0 Td(b( p 2 r ) ; b( r ) = R r 1 p 2 e )Tj/T1_6 5.97758 Tf7.785 3.26701 Td(t 2 2 dt: (4.2)

PAGE 93

81 3 4 5 6 7 8 9 1 0 0 1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0 6 0 0 0 g ( d B )p a c k e t l e n g t h ( b i t )Figure 4: The optimal pack et length ( P L ) vs. SNR ( r ), h=128. P L 0 P L 1 P L 2 P L k 1 t 0 1 t 1 0 t 1 2 t 2 1 t 2 3 t 3 2 t k 2 k 1 t k 1 k 2 t 0 0 t 1 1 t 2 2 t k 1 k 1 Figure 4: P ack et-Length-Based nite-state Mark o v channel model. According to [98], for a gi v en crosso v er probability p the optimal pack et length, which is a function of p can be written as P L = )Tj/T1_3 11.95509 Tf9.28912 0 Td(h ln(1 )Tj/T1_3 11.95509 Tf11.95309 0 Td(p ) )Tj/T1_10 11.95509 Tf12.0327 14.472 Td(q )Tj/T1_8 11.95509 Tf9.28912 0 Td(4 h ln (1 )Tj/T1_3 11.95509 Tf11.94411 0 Td(p ) + h 2 ln 2 (1 )Tj/T1_3 11.95509 Tf11.94411 0 Td(p ) 2 ln (1 )Tj/T1_3 11.95509 Tf11.95309 0 Td(p ) ; (4.3) where h is the number of o v erhead bits per pack et. Fig.4sho ws the relationship between the recei v ed SNR and the optimal pack et length. F or a gi v en state S k the a v erage optimal pack et length P L k for this state can be deri v ed by using (4.1), (4.2), (4.3) to be R r k +1 r k 1 r e r r )Tj/T1_12 7.97011 Tf6.5833 0 Td(h ln(b( p 2 r )) )Tj/T1_18 10.9091 Tf6.60638 7.92899 Td(p )Tj/T1_11 7.97011 Tf6.5833 0 Td(4 h ln(b( p 2 r ))+ h 2 ln 2 (1 )Tj/T1_11 7.97011 Tf6.5833 0 Td((1 )Tj/T1_11 7.97011 Tf6.5833 0 Td(b( p 2 r ))) 2 ln(b( p 2 r )) dr R r k +1 r k 1 r e )Tj/T1_14 5.97758 Tf7.785 3.69 Td(r r dr : (4.4) Based on the abo v e analysis we present our pack et-length-based FSMC model in Fig.4. W e represent each state as the a v erage pack et length P L k which is the pack et size for a transmission in state k The transition probabilities between dif ferent states are denoted

PAGE 94

82 as t ij Further we can deri v e the state steady probability for each state as k = Z r k +1 r k 1 r e )Tj/T1_6 5.97758 Tf7.785 3.69902 Td(r r dr = e )Tj/T1_6 5.97758 Tf7.785 4.62604 Td(r k r )Tj/T1_3 10.9091 Tf10.89729 0 Td(e )Tj/T1_6 5.97758 Tf7.785 5.45403 Td(r k +1 r ; k = 0 ; 2 ; :::; K )Tj/T1_4 10.9091 Tf10.89729 0 Td(1 : (4.5) In practice, we may use dif ferent modulation schemes (not necessarily BPSK) in different channel states. Moreo v er by properly partitioning the range of the recei v ed SNR, we may obtain the multiplicati v e relationship between the a v erage optimal pack et lengths. 4.5 Courtesy Piggybacking In this section, we present our Courtesy Piggybacking scheme to alle viate the conict between throughput and f airness and to combat the starv ation problem for dif ferentiated services. 4.5.1 System Assumptions W e consider an ad hoc netw ork consisting of n mobile nodes uniformly distrib uted in some area. Nodes can communicate with each other directly if the y can hear each other or through other relay nodes in a single broadcast channel. The y emplo y some contentionbased MA C protocols, such as IEEE 802.11, to support their communications. Each node can generate services with N dif ferent priorities destined to other mobile node(s). A node' s mobility follo ws the random w aypoint model [58,70]. At rst, a node stays at a position for duration of pause time After that period, the node chooses a ne w random position and mo v es to w ards that position at a random speed uniformly distrib uted in the range from min speed to max speed After reaching the ne w position, the node will stay there for another pause time This process will continue for each node until the end of the simulation. W e assume some service dif ferentiation mechanism is emplo yed at the netw ork layer All the heterogeneous traf c is prioritized at its originating source node. When a pack et is handed do wn from the netw ork layer it will be k ept in the Tx queue corresponding to its priority and w ait for its turn to be transmitted at the MA C layer From the pre vious section, we kno w that the pack et length is related to the recei v ed SNR. The greater the SNR is, the greater the pack et length is. In the IEEE 802.11 MA C

PAGE 95

83 M S D U M A C h e a d e r M S D U C R C M A C h e a d e r M S D U C R C M A C h e a d e r M S D U C R C Figure 4: A fragmentation e xample. protocol [5], this pack et length may be called as frame length, which equals the fragmentation threshold plus the length of the MA C header and the length of CRC. In the IEEE 802.11 standard, the MA C layer tak es a MSDU from the Tx queue and adds MA C header and a CRC to each MSDU to generate a MPDU. In order to reduce the probability of transmission errors, the IEEE 802.11 limits the size of the body of a MPDU to be less than a x ed fragmentation threshold ( F T ), or it will break the long MSDU into multiple fragments, each of which will be no longer than the F T In Fig.4, we sho w a case where a long MSDU is partitioned into three small MSDUs in the IEEE 802.11. Since the length of the MA C o v erhead may be k ept unchanged, according to the analysis in the pre vious section, dif ferent channel states ha v e dif ferent frame lengths, we can say that dif ferent channel states ha v e dif ferent fragment thresholds ( F T s ). The greater the recei v ed SNR is, the greater the fragment threshold ( F T ) is. Hereafter we associate F T k with each state S k of the FSMC model as depicted in Fig.4. In order to impro v e the channel utilization, we assume that the MA C protocol can adapti v ely adjust the fragmentation threshold and the transmission rate according to the channel state. T o accurately determine the channel state when some pack ets need to be transmitted, we further assume that we ha v e some channel estimators or predictors, which can pro vide the accurate channel information for the proper MA C layer fragmentation. 4.5.2 The Courtesy Piggybacking Scheme In practice, the size of a pack et generated by an application may be x ed or may v ary from a minimum allo wed size to a maximum v alue P K max W e ar gue that the P K max should be properly chosen to reduce the o v erall o v erhead. Suppose we w ant to transmit c Mbits traf c. P ack ets are generated according to the P K max W e assume that each pack et

PAGE 96

84 Figure 4: The total o v erhead with P K max and F T can be correctly recei v ed without an y retransmission2. Then, the total o v erhead should be the sum of the o v erheads O ip at the IP layer (for e xample, 20 bytes for IPv4), O mac at the MA C layer (for e xample, 34 bytes for the IEEE 802.11) and O phy at the PHY layer (for e xample, 16 bytes). Thus, the total o v erhead to transmit the c Mbits traf c can be written as c P K max O ip + P K max F T ( O mac + O phy ) ; where d e is the function to round the element to the nearest inte ger greater than the element. W e sho w the relationship of o v erhead vs. P K max and F T when c = 1 in Fig.4. From Fig.4, we observ e that P K max should be reasonably chosen when multiple fragmentation thresholds are used. It cannot be too small, as it may cause too much o v er head; neither can it be too lar ge, as it may generate too man y fragments when the F T is small, which may further de grade the o v erall throughput. F or e xample, a lar ge pack et may 2 F or simplicity we only consider the case without an y retransmission and vie w the resulting o v erhead as a lo wer bound. Apparently the o v erhead with retransmissions is lar ger than this lo wer bound.

PAGE 97

85 be partitioned into man y small fragments. Each fragment is augmented with an indi vidual header sent as an independent transmission, and ackno wledged indi vidually Hence, a lar ge pack et will cause lots of D A T A/A CK e xchanges and result in suboptimal perfor mance. Thus, there must e xist an optimal v alue of P K max such that the o v erall o v erhead associated with the successful transmission of a message is minimized. Assume we obtain P K max which may not be equal to an y of F T k ( k = 0 ; 1 ; 2 ; :::; K )Tj/T1_5 11.95509 Tf11.98911 0 Td(1) ; it is thus advisable to approximate P K max with the closest fragmentation threshold corresponding to a certain channel state, say S m Therefore, we set P K max to F T m Ne xt, we w ant to sho w where the free space comes from. When a pack et with length strictly less than the P K max is transmitted in the channel state with F T less than F T m the pack et may be fragmented, and there is no free space a v ailable for all fragments possibly e xcept the last one. Ho we v er due to the time-v arying nature of the channel, when the pack et is transmitted in the channel state with F T greater than F T m one pack et does not ha v e enough bits to utilize the full capacity the channel pro vides in one transmission. W e ar gue that we could tak e adv antage of the free space to pack more bits as the channel allo ws. As a matter of f act, we ha v e sho wn, in our recent studies, that the fragmentation threshold can be up to 10K bits when the SNR is close to 20 dB and 64 QAM modulation scheme is used with tar geted FER 8% (Frame Error Rate) [93]. On the other hand, we observ e that in contention-based MA C protocols, it may tak e a long time for a node to seize the channel, and the node that has seized the channel should treasure e v ery transmission opportunity to transmit as man y bits as possible, especially when the channel condition is good. From no w on, we call state S k the free-space-ef fecti v e state when k is greater than m otherwise the non-free-space-ef fecti v e state, e v en though such a state may still has the possibility to pack more bits when the traf c dynamic is tak en into account. No w we describe ho w the Courtesy Piggybacking scheme mak es use of the free space. When a mobile node seizes the channel, it will rst check the channel state and determine if it is in a free-space-ef fecti v e state and if it is capable of piggybacking more pack ets in

PAGE 98

86 one transmission. If it is not in a free-space-ef fecti v e state, only one pack et (MSDU) with highest priority from the queues will be serv ed, as the current MA C protocol does. If the channel is in a free-space-ef fecti v e state, the node can transmit in one transmission as man y bits as channel allo ws and thus can piggyback more pack ets (MSDUs) from the queue(s), which may ha v e dif ferent priorities b ut the same ne xt hop in the routing table. Since the Courtesy Piggybacking scheme follo ws the cross-layer design principle so that the MA C layer has the access to the routing information, it is possible for the MA C layer to obtain such pack ets from the Tx queues. After identifying the e xistence of the free space, we no w discuss the piggybacking rules that guide the MA C layer to assemble enough and proper bits from the Tx queues (the HO W MANY -WHO problem) and piggyback them to the ne xt hop to alle viate the conict we intend to address. Since the channel state determines HO W MANY MSDUs the node can pack and transmit in one transmission, the fundamental issue of the rules should specify is who plays the role of train that of fers the piggybacking service to others, and who can enjo y such piggybacking service. W ithout an y scheduling mechanism, the role of train is al w ays tak en by the MSDU locating at the head of a non-empty queue with the highest priority currently Thus, the piggybacking rules should primarily address who has the pri vile ge to enjo y such free piggybacking service. As a guideline, the basic idea for such piggybacking rules is that under dif ferent channel states, the node assembles multiple MSDUs that may ha v e dif ferent priorities b ut share the same ne xt hop in the routing table, to form an MPDU whose length is channel dependent. In this w ay we can achie v e some e xtent of f airness between dif ferent prioritized services. When the channel is not in a free-space-ef fecti v e state, only the highest priority service in the Tx queues is supported, and the pack ets are fragmented if needed and are treated as usual. When the channel changes to a free-space-ef fecti v e state, according to the rules we dene, our Cour tesy Piggybacking scheme can pack other services, possible with lo wer priorities, to share the residual bandwidth with the high-priority traf c. One such rules is to gi v e preference

PAGE 99

87 q u e u e f o r p r i o r i t y l e v e l 0 q u e u e f o r p r i o r i t y l e v e l 1 q u e u e f o r p r i o r i t y l e v e l 2 q u e u e f o r p r i o r i t y l e v e l N 1 I n c o m i n g d a t a p a c k e t s f r o m r o u t i n g l a y e r w i t h a s s i g n e d p r i o i t y l e v e l d e q u e u e c o n t r o l l e r b M S D U b M S D U b M S D U A s s e m l e d M S D U M A C h e a d e r M S D U C R C M P D U B r e a k i n t o b M S D U i f n e c e s s a r yFigure 4: Illustration of the courtesy piggybacking scheme. to high-priority services. It al w ays, if possible, packs the high-priority services destined to the same ne xt hop in queue(s). Only when there are no more bits from the high-priority traf c tting into the free-space3will the bits from the lo wer -priority queue(s) be considered for piggybacking. Other rules may not prefer the high-priority service; for e xample, a high-priority service may trade-of f its o wn performance for f airer channel utilization by its courtesy piggybacking the lo w-priority service. One such rules is to al w ays piggyback the MSDUs from the longest Tx queue. W ith such piggybacking rules, the traf c dynamics (dif ferent pack et arri v al time and destinations) and channel dynamics are jointly utilized to strik e a good balance between throughput and f airness. Note that the piggybacking rules are not necessarily dened a priori ; the y could be designed to adapt to both channel and traf c uncertainty in the runtime. Intuiti v ely the Courtesy Piggybacking scheme can impro v e the performance of the lo w-priority traf c, since some lo w-priority pack ets may be pack ed with high-priority packets and be deli v ered to the ne xt hop for free; thus it can statistically reduce the time tak en to contend for accessing the channel for the lo w-priority services. This benet will be more 3 When we say no more bits in one queue tting into the free-space, it means either no pack et is left in the queue or pack ets in the queue do not share the same ne xt hop with the MSDU who of fers the piggybacking service.

PAGE 100

88 pronounced in mobile ad hoc netw orks using service dif ferentiation based MA C protocols [78][99] where the MA C protocols sacrice quality of the lo w-priority service to support high-priority service through either time spacing (dif ferentiation of Interframe Space (IFS)) or back of f parameters [5]. On the other hand, the reduction of contention from lo w-priority services can in turn benet the high-priority services: one node' s courtesy piggybacking of lo w-priority services may help its neighbors to transmit high-priority traf c, because less lo w-priority traf c will reduce the contention the high-priority traf c may encounter One may w onder wh y we do not simply release the channel so that other lo w-priority traf c can use the channel, that is, the so-called complete sharing scheme. The problem is that the time, for which the residual resource is a v ailable, is too short to be gi v en to other services due to the o v erhead associated with successfully seizing the channel. Besides, some MA C protocols such as the IEEE 802.11 f amily forbid others to use the channel during the time period specied by the Netw ork Allocation V ector (N A V). Ev en if the N A Vs are reset, the contention process may tak e too long to render the harv ested resource from the rate adaptation useless. Thus, the courtesy piggybacking by high-priority traf c o ws mak es more sense. In short, our Courtesy Piggybacking is able to achie v e better channel utilization and further impro v es the f airness between dif ferent prioritized traf c by the follo wing tw o means: lo wering the contention of the netw ork and decreasing the o v erhead required for a transmission. And as we sho w in simulation later our scheme signicantly impro v es the performance of lo w-priority traf c, while impro v es or at least k eeps unchanged the performance of high-priority traf c with appropriate piggybacking rules. T o illustrate the Courtesy Piggybacking scheme, we demonstrate the operation of the scheme in Fig.4. First, prioritized pack ets called MSDUs arri v e from the netw ork layer as b-MSDUs (basic MSDUs, the basic unit) whose lengths agree with the F T m W e assume the maximum pack et length P K max is strictly enforced at the upper layer; if not, an o v ersized MSDU will be further brok en do wn into se v eral b-MSDUs and the resulting b-MSDUs will inherit the IP header of the original MSDU. The b-MSDUs are k ept in the

PAGE 101

89 queues corresponding to their priorities. The dequeue controller operates according to the predened piggybacking rule, dequeues one or more b-MSDUs with the same ne xt hop and forms a MPDU satisfying the F T corresponding to the channel state. In order to reduce the o v erhead and the w ork to break a MSDU at the transmitter and to assembly the MSDU at the recei v er it is advisable to limit the pack et length at the netw ork layer to be no longer than F T m In order to a v oid further fragmentation of a b-MSDU to t the free space and assembly the b-MSDU, it is advisable to maintain the multiplicati v e relationship between the fragmentation threshold (FT) of the free-space-ef fecti v e state and F T m that is, the frame length for state k satises F T k = g k F T m where g k is a positi v e inte ger This can be achie v ed by properly partitioning the range of recei v ed SNR and adopting channeldependent modulation schemes. T o reduce the transmission time for a long frame, rate adapti v e transmission schemes may be used, so that the time for transmitting a frame does not v ary too much. T o a v oid making too man y modications to the MA C layer we prefer packing the b-MSDUs with the same ne xt hop in the routing table. T o f acilitate a recei v er in unpacking the bound pack ets, an unused bit [100] in the IP header of each b-MSDU is set to 1 at the transmitter to indicate that one bound b-MSDU is follo wed this b-MSDU, and the corresponding bit in the last b-MSDU is set to 0. At the recei v er the only thing it needs to do is to ackno wledge the recei v ed long frame, unpack the pack ed pack ets one by one according to the v alue of the unused bit. 4.5.3 Discussion Some properties of the CP. If we e xamine the destination of the bits in a single piggyback ed transmission, though these bits share the same ne xt hop, we nd out these bits may be destined to the ne xt hop or other nodes. Figure 4-8 showsthreescenariosfortheCourtesy Piggybacking. Consider that a mobile node A sends some bits (consisting of 2 b-MSDUs) to the ne xt hop B which ha v e three neighbors, including this mobile node A. Suppose that pack et 2 is piggyback ed by pack et 1, both pack ets should ha v e the same ne xt hop B. After the pack ets 1 and 2 arri v e at B, there are three cases at node B to process these tw o pack ets

PAGE 102

90 A B C D 1 2 1 2 A B C D 1 2 1 A B C D 1 2 C a s e 1 : E a c h h a s d i f f e r e n t n e x t h o p a t B C a s e 2 : B o t h d e s t i n e t o B C a s e 3 : O n e d e s t i n e s t o B t h e o t h e r n o t Figure 4: Three piggybacking cases. if we do not distinguish the dif ference between pack et 1 and pack et 2. Case 1 sho ws that pack et 1 and pack et 2 may be destined to dif ferent nodes and ha v e dif ferent ne xt hops at node B. Case 2 sho ws the case when both pack ets ha v e the same destination B. Case 3 sho ws that one pack et is destined to B while another one is destined to a node other than B. Since the probability that a pack et is piggyback ed lar gely depends on channel conditions and traf c pattern, the piggybacking may induce some delay jitter F or instance, due to our restriction that only pack ets sharing the same ne xt hop can be piggyback ed together one pack et that arri v es at one node later may lea v e the node earlier than some earlier arri ving pack ets. The signicance of piggybacking rules The piggybacking rules may play an impor tant role in allocating the bandwidth among the dif ferent prioritized traf c. Here, we w ant to discuss the design of a piggybacking rule based on a special case. When we ha v e plenty

PAGE 103

91 of dif ferent priority pack ets in the Tx queues w aiting for being serv ed, the design of piggybacking rule can be vie wed as an allocation problem. As discussed abo v e, by properly par titioning the range of recei v ed SNR, the fragmentation threshold of the free-space-ef fecti v e state k satises F T k = g k F T m where g k is a positi v e inte ger F or other non-free-spaceef fecti v e states, let g k = 1 Suppose we ha v e a total of K dif ferent channel states, and N dif ferent priority le v els. Let k j denote the number of b-MSDUs of j priority le v el to be pack ed when the channel is in state k W e should point out that in the non-free-spaceef fecti v e state, only the highest priority pack et is serv ed when there is plenty of traf c in the w aiting queue, thus k 0 = 1 and k j = 0 for 0 k m and 1 j N )Tj/T1_4 11.95509 Tf12.5201 0 Td(1 If we ne glect the MA C layer o v erhead, then the design problem can be reduced to choose a k j such that 8 > < > : 0 k j P j k j g k 0 k K )Tj/T1_4 11.95509 Tf11.95309 0 Td(1 ; 0 j N )Tj/T1_4 11.95509 Tf11.94411 0 Td(1 : (4.6) When the operation of packing pack ets with dif ferent priority le v els associates with certain objecti v e of f airness or prot, the design problem is a kind of knapsack problem which is considered to be NP-complete [72]. On the one hand, the resource (free space here) should ne fully utilized; On the other hand, the system attempts to maintain the best performance in terns of f airness among dif ferent classes of traf c. In f act, an upper bound for the e xpected v alue of throughput of priority le v el j at one node is can be written by P k R k p k k j ; 0 j N )Tj/T1_4 11.95509 Tf11.9621 0 Td(1 where p k is the probability that the channel is in state k and R k is the transmission rate in state k Some measures may impro v e CP. T o a v oid fragmentation of the b-MSDUs in a freespace-ef fecti v e state, in our piggyback scheme we should maintain the multiplicati v e relationship between the fragmentation threshold (FT) in the free-space-ef fecti v e state and the F T m Actually we can relax this requirement in the high traf c load case by allo wing fragmentation of the lo w-priority services at will to t into the free space the channel pro vides. Because at v ery hea vy traf c load, the piggybacking rule f a v oring high-priority

PAGE 104

92 services may still lead to bandwidth starv ation for lo w-priority services. By allo wing the fragmentation of the b-MSDUs from lo w-priority traf c, at least some lo w-priority traf c can be serv ed by piggybacking. Our Courtesy Piggybacking scheme does not preclude scheduling mechanisms; in f act, scheduling can still be emplo yed at higher layers to enhance the management of the heterogeneous traf c. F or e xample, we can use EDF (Earliest Deadline First) polic y to manage each priority queue, such that in a single queue, the pack et with early deadline may be put at the head of the queue and get transmitted earlier than those with later deadline. At the same time, the piggyback scheme can still tak e ef fect once the channel is in a freespace-ef fecti v e state. In our preliminary implementation of the piggybacking, the b-MSDUs are or g anized in the queues according to their priorities. When a transmitter w ants to pack more bits to the same recei v er e xhausti v e search is carried out to nd the proper bits in candidate queues according to the piggybacking rules. In addition, some processing time is also needed at the recei v er to unpack the pack ets. Thus, in the rst place we may e xpect some additional delay caused by courtesy piggybacking; ho we v er as can be seen from our performance e v aluation, the incurred delay is ne gligible and hence acceptable compared with the benet g ained from the Courtesy Piggybacking scheme. Furthermore, we ar gue that with more ef cient w ays to or g anize the b-MSDUs and to quickly acquire the proper b-MSDUs for packing, the benet from piggybacking scheme should be more visible. In our proposed Courtesy Piggybacking scheme, only the traf c sharing the same ne xt hop can be pack ed. One may w onder if it can be e xtended to the dif ferent ne xt hop scenarios. W e are cautions to mak e this mo v e. The main concern comes from the MA C layer The pre v alent MA C protocols including the IEEE 802.11 do not support multiple recei v ers at the same time. Undoubtedly it is a arduous task to coordinate the transmission and reception acti vities between multiple recei v ers. Especially when multiple recei v ers are in v olv ed in one transmission, the interference area should be much lar ger than that of one

PAGE 105

93 recei v er case and the coordination issue gets more complicated. Though a data ushing mechanism for multiple recei v er w as presented in [101], while without consideration of the enlar ged interference area, the benet may not outweigh the ne g ati v e impact of their approach. In our e x emplary piggybacking illustration (Fig.4), all the dequeued b-MSDUs are rst assembled as one MSDU and further encapsulated with only one MA C header and CRC. This e x emplary piggybacking method imposes no modication on current MA C protocol, e.g, IEEE 802.11, b ut it lacks e xibility F or e xample, when the recei v er detects something wrong with the recei v ed frame, it is not able to inform the transmitter which b-MSDU is damaged, and as a result, the transmitter has to retransmit the whole frame ag ain. In f act we can adopt an alternati v e method as the follo wing (Fig.4). Dif fer ent from the abo v e method, each b-MSDU will generate a separate MPDU with its o wn CRC and MA C header and the resulting MPDUs will nally be combined as an assembled MPDU and be sent out as one unit in one transmission. At the recei v er side, the recei v er can check the CRC for each b-MSDU and positi v ely or ne g ati v ely ackno wledge each bMSDU in a single short message, for e xample, A CK. W ith this alternati v e method, the transmitter can retransmit only those b-MSDUs f ailed CRC checking. In f act, these retransmitted b-MSDUs can be further bound with other fresh b-MSDUs from the queues to form another assembled MPDU. Compared the alterati v e one with our e x emplary one, the alterati v e method pro vides some de gree of e xibility b ut requires some modication on MA C protocol to f acilitate the abo v e communications. This modication can be possibly made similar to that in [101]. On the other hand, our e x emplary piggybacking method may incur less o v erhead of MA C header and CRC than the alternati v e one. In practice, some adapti v e piggyacking methods can be designed to accommodate the time-v arying channel condition, meanwhile, k eep the o v erhead reasonable.

PAGE 106

94 . A s s e m l e d M P D U M A C h e a d e r b M S D U C R C M A C h e a d e r b M S D U C R C M P D U Figure 4: An alternati v e piggybacking method. 4.6 Performance Analysis In this section, we present the performance analysis in order to theoretically sho w the ef fecti v eness of our proposed piggybacking scheme in alle viating the conict between throughput and f airness for dif ferent prioritized services. T o simplify the analysis, we consider the piggybacking at one node, and assume all the pack ets are destined for the same ne xt hop. Ag ain, we assume that there are a total of N dif ferent priority le v els, P 0 P 1 ..., P N )Tj/T1_5 7.97011 Tf6.5833 0 Td(1 whereby P i has higher priority than P j if i < j The arri v als of each priority P i service are P oisson processes with arri v al rate i The channel is modeled as FSMC as discussed in Section4.4: F T k = g k F T m where g k = 1 if k m4; otherwise g k is an inte ger greater than 1 and g i > g j if i > j m When the channel is in state j > m the channel adaptation scheme is used such that the time for transmitting F T j is almost the same as that for F T m Moreo v er we assume that the pack et length is P K max = F T m Under such assumptions, our Courtesy Piggybacking scheme can be modeled as a multiple-serv er queue system with non-preempti v e priority where the service rate is dependent on the channel state. More specically the service discipline is the follo wing if we limit the number of serv ers to 2. Serv er S R 0 operates in all channel states with a v erage service time X F or serv er S R 1 it operates when the channel is in the state S k ; k > m and does not w ork in an y state S k ; k m ; the service time for serv er S R 1 is X g k )Tj/T1_5 7.97011 Tf6.5833 0 Td(1 This means that when serv er S R 0 serv es one non-piggyback 4 Gi v en x ed transmitting po wer and modulation scheme, when the channel is in an y state j < m a fragmentation threshold that is smaller than F T m is required to maintain the same frame error rate. F or simplicity we assume F T m is also used in those states.

PAGE 107

95 pack et, serv er S R 1 can serv e g k )Tj/T1_5 11.95509 Tf12.5741 0 Td(1 piggyback ed pack ets. The non-w orking probability of serv er S R 1 is m P k =0 k where k is the steady-state probability of state S k Serv er S R 0 operates with non-preempti v e head-of-line priority service discipline, while serv er S R 1 operates according to the discipline dened in the piggybacking rules. F or e xample, it may rst serv e the traf c with the longest w aiting queue. W ithin a priority traf c, service is pro vided on a rst-come-rst-serv ed basis. Figure 4 showstheanalyticalmodel. F or the purpose of obtaining tractable analysis, we study the scheme in a simple scenario in which the traf c is of tw o priorities, and the channel has tw o states, with F T 1 = 2 F T 0 Accordingly the service time for each serv er is X As discussed abo v e, serv er S R 0 w orks all the time with non-preempti v e service discipline, whereas S R 1 w ork only when channel state is in S 1 and does not w ork otherwise. Then our piggybacking queue model can be reduced to an M/D/2 non-preempti v e priority queueing system, in which one serv er w orks all the time, while the other w orks with some probability This system seems to be simple, ho we v er it is hard to quantitati v ely analyze it. T o our kno wledge, there is no ready solution to this interesting queue system. Hence we seek to get some bounds on the performance metrics such as a v erage w aiting time (that is, the a v erage queueing delay) and queue size. W e only focus on the a v erage w aiting time hereafter as the a v erage w aiting time and the queue size lead to each other according to Little' s La w W e rst consider the upper bound of the a v erage w aiting time for the system. Ob viously if a transmitter does not ha v e an y kno wledge of the channel status and hence does not adopt an y rate adaptation based on the channel state, both traf c types, that is, the high priority traf c and the lo w priority traf c, will suf fer from longer delay than the y will when piggybacking is used. In this case, only one serv er that is, serv er S R 0 w orks. The system thus is an M/D/1 priority queue system [102], and it is easy to obtain the a v erage w aiting time for each type of traf c: 8 > < > : W 0 = ( 0 + 1 ) X 2 2(1 )Tj/T1_3 7.97011 Tf6.5833 0 Td( 0 X ) W 1 = ( 0 + 1 ) X 2 2(1 )Tj/T1_3 7.97011 Tf6.5833 0 Td( 0 X )(1 )Tj/T1_3 7.97011 Tf6.5833 0 Td( 0 X )Tj/T1_3 7.97011 Tf6.5833 0 Td( 1 X ) : (4.7)

PAGE 108

96 . . S R0 S R1 0 l 1 l 1 N l S 0 S 1 S k 1 t 0 1 t 1 0 t 1 2 t 2 1 t k 2 k 1 t k 1 k 2 t 0 0 t 1 1 t k 1 k 1 0 1 m 1 1 m P r i o r i t y Q u e u e s S e r v e r s S m t m 1 m t m m 1 t m m + 1 t m + 1 m t m mFigure 4: The queue model for piggybacking. Recall that serv er S R 0 w orks all the time and serv er S R 1 w orks only when channel state is in S 1 If the piggybacking rule is set such that when S R 1 operates, it al w ays serv es the high-priority traf c rst, the follo wing case will pro vide the lo wer bound for the a v erage w aiting time. In this case, the transmitter is a w are of the channel state and adopts the channel dependent transmission rate. Moreo v er the channel is al w ays in good state. Therefore, the system becomes a tw o-serv er queue system with non-preempti v e priority where the tw o serv ers, serv er S R 0 and S R 1 w ork all the time. According to the results in [103], we can obtain the a v erage w aiting time for each type of traf c: 8 > < > : W 0 = 4( 0 + 1 ) 2 X 3 3(2+( 0 + 1 ) X )(2 )Tj/T1_12 7.97011 Tf6.5833 0 Td( 0 X ) W 1 = 4( 0 + 1 ) 2 X 3 3(4 )Tj/T1_9 7.97011 Tf6.5833 0 Td(( 0 + 1 ) 2 X 2 )(2 )Tj/T1_12 7.97011 Tf6.5833 0 Td( 0 X ) : (4.8) T o v alidate the upper and lo wer bounds, we implement the abo v e queue model in OPNET [57] and present the analytical results and simulation results in Fig.4, where the service time X = 0.01s. It can be observ ed that the a v erage w aiting time for the piggybacking scheme in three dif ferent channel models as specied in T able4is completely

PAGE 109

97 6 5 7 0 7 5 8 0 8 5 9 0 9 5 1 0 0 1 03 1 02 1 01 1 00 A r r i v a l r a t e ( p a c k e t / s )A v e r a g e w a i t i n g t i m e f o r p r i o r i t y 1 ( s ) p0= 0 7 5 p1= 0 2 5 : s i m u l a t i o n p0= 0 5 0 p1= 0 5 0 : s i m u l a t i o n p0= 0 2 5 p1= 0 7 5 : s i m u l a t i o n L o w e r b o u n d : a n a l y s i s U p p e r b o u n d : a n a l y s i s (a) A v erage w aiting time of priority 1 service. 6 5 7 0 7 5 8 0 8 5 9 0 9 5 1 0 0 1 04 1 03 1 02 A r r i v a l r a t e ( p a c k e t / s )A v e r a g e w a i t i n g t i m e f o r p r i o r i t y 0 ( s ) p0= 0 7 5 p1= 0 2 5 : s i m u l a t i o n p0= 0 5 0 p1= 0 5 0 : s i m u l a t i o n p0= 0 2 5 p1= 0 7 5 : s i m u l a t i o n L o w e r b o u n d : a n a l y s i s U p p e r b o u n d : a n a l y s i s (b) A v erage w aiting time of priority 0 service. Figure 4: A v erage w aiting time. T able 4: Channel model statistics Setting State i P r ( i ) t 1 ;i t 0 ;i 1 0 0.75 0.0075 0.9975 1 0.25 0.9925 0.0025 2 0 0.5 0.002 0.998 1 0.5 0.998 0.002 3 0 0.25 0.0025 0.9925 1 0.75 0.9975 0.0075 bounded by the obtained bounds. Therefore, we kno w the g ain that can be accrued by adopting the piggybacking scheme as opposed to the case without piggybacking. 4.7 Performance Ev aluation In this section, we implement our proposed piggybacking scheme in OPNET with DSD V and the IEEE 802.11 as underlying routing and MA C protocols, and conduct e xtensi v e simulations to e v aluate the performance of the piggybacking scheme. The simulation results v alidate the ef fecti v eness and ef cienc y in impro ving the channel utilization and f airness thus alle viating the conict between throughput and f airness. 4.7.1 Simulation Setup In our simulation study we assume the ph ysical channel is a slo w f ading channel with only tw o states satisfying F T 1 = 2 F T 0 Three dif ferent channel settings are adopted

PAGE 110

98 in our studies. The channel statistics are listed in T able4. While setting 2 represents a neutral channel in the sense that the steady state probabilities of both states are identical, channel setting 1 and 3 respecti v ely represent a relati v ely bad channel and a relati v ely good channel. W ithout loss of generality we limit the number of priorities to 2. W e simulate an ad hoc netw ork consisting of 50 mobile nodes5, whose mobility follo ws the random w aypoint mobility model in a 1500 300 m 2 area. The transmission range of each node is 250 m Each node generates traf c according to a Poisson process with parameter and the destination for each generated pack et is randomly chosen among all other nodes. W e assume that the pack et length is 1024 bits that agrees with F T 0 and each pack et is a b-MSDU. The generated traf c is further assigned with 1 (lo w priority) or 0 (high priority) with probability 0.5. All pack ets are b uf fered in the queues according to their priorities. In our simulations, when the channel is in the state corresponding to F T 0 we use basic transmission rate 1Mbps, while for state corresponding to F T 1 we use 2Mbps, so that the transmission time for one fragment with channel-dependent length in tw o states does not change too much. Each simulation runs for 300 seconds. W e dene tw o piggybacking rules for comparison. Rule 1 f a v ors the high priority traf c. When the channel state is in F T 0 a priority-1 pack et can be serv ed only when no priority-0 pack et e xists in the queue. When the state is in F T 1 a priority-1 pack et is piggyback ed by priority-0 pack ets only when no priority-0 pack et is left in the queue (of course, the pack ets should share the same ne xt hop in the routing table). In contrast, rule 2 f a v ors the lo w priority traf c. In rule 2, when the channel is in F T 0 the stations act as in the rule 1. When the state is in F T 1 a priority-1 pack et is piggyback ed no matter ho w man y priority-0 pack ets are left in the queue. W e study the performance of the netw ork in four 5 T o minimize the possible unf airness between dif ferent nodes due to unequal channel access, we use such conguration and treat all the nodes equally for e xample, using the same traf c and mobility pattern.

PAGE 111

99 scenarios: case 1. the netw ork is una w are of channel states; case 2. the netw ork is a w are of channel states and thus adopts dynamic transmission rate; case 3. the netw ork emplo ys the Courtesy Piggybacking with rule 1; and case 4. the netw ork emplo ys the Courtesy Piggybacking with rule 2. W e choose tw o metrics to analyze and compare the performance of piggybacking: End-to-end delay and P ack et deli v ery ratio End-to-end delay measures the a v erage onew ay latenc y observ ed between the time instant that the pack et is generated at the source and the time instant the pack et is recei v ed at the destination. This metric should count in all the delays, such as propag ation delay queueing delay and transmission delay which the pack et has e xperienced during the whole process. P ack et deli v ery ratio measures the ratio of the total number of pack et successfully recei v ed by the destinations to the total number of pack ets generated at the sources. This metric reects the o v erall throughput and f airness of each prioritized traf c. T o observ e the ef fect of the channel characteristics, we rst disable node mobility and compare the performance of piggybacking under dif ferent channel settings and under dif ferent traf c loads. Then we x the channel setting as setting 2, and enable node mobility according to the random w aypoint mobility model described in the Section4.5with min speed = 1 m=s and max speed = 19 m=s in order to study the ef fect of traf c load and mobility on the performance of piggybacking. 4.7.2 Impact of Channel Characteristics Figure4 illustratesperformanceofthenetworkinthefourdifferentcases,when three channel settings are specied as in T able4. Since in case 1, the netw ork has no kno wledge of the channel, the performance is the same no matter what channel settings are used. In all the other three cases, the performance changes with the channel setting. More specically as the channel is getting better that is, the channel setting changes from 1, to 2, and to 3, for each traf c priority the end-to-end delay decreases and the pack et deli v ery ratio increases. This sho ws that the dynamic channel conditions are v aluable resources that

PAGE 112

100 1 2 3 0 1 2 3 4 5 e n d t o e n d d e l a y ( s )c h a n n a l s e t t i n g ( a ) p r i o r i t y 1 c a s e 1 c a s e 2 c a s e 3 c a s e 4 1 2 3 0 0 1 0 2 0 3 0 4 0 5 e n d t o e n d d e l a y ( s )c h a n n a l s e t t i n g ( b ) p r i o r i t y 0 1 2 3 0 0 2 0 4 0 6 0 8 1 p a c k e t d e l i v e r y r a t i oc h a n n a l s e t t i n g ( c ) p r i o r i t y 1 1 2 3 0 0 2 0 4 0 6 0 8 1 p a c k e t d e l i v e r y r a t i oc h a n n a l s e t t i n g ( d ) p r i o r i t y 0Figure 4: Simulation results with dif ferent channel settings. should be e xplored. More impro v ements in terms of both metrics can be observ ed when piggybacking is adopted, for e xample, case 3 and case 4, compared to case 2. Further our piggybacking scheme can achie v e a better f airness between these tw o prioritized traf c while still maintaining a higher pack et deli v ery ratio, thus higher throughput. This is because, in addition to taking adv antage of channel states, the piggybacking scheme reduces the time to contend for the channel, thereby further impro ving channel utilization. 4.7.3 Impact of T raf c Load The performance of the netw ork is studied under dif ferent traf c loads as well, as sho wn in Fig.4, in which the a v erage pack et inter -arri v al time of 0.3s represents the relati v ely light traf c load and the a v erage pack et inter -arri v al time of 0.25s represents relati v ely hea vy traf c load. From Fig.4, we can see the inter -class ef fects in the dif ferentiated services system, which becomes more pronounced in the case where the highpriority traf c load is high. The delay for priority 1 (7.89 second) is f ar greater than that for priority 0 (0.30 second), and the pack et deli v ery ratio for priority 1 is much smaller than that for priority 0. It also can be observ ed that case 3 and 4, where our piggybacking scheme is emplo yed, ha v e better performance than case 1 and 2. Under both hea vy traf c load and

PAGE 113

101 0 2 5 0 3 0 2 4 6 8 e n d t o e n d d e l a y ( s )p a c k e t a r r i v a l i n t e r v a l ( s ) ( a ) p r i o r i t y 1 0 2 5 0 3 0 0 1 0 2 0 3 0 4 e n d t o e n d d e l a y ( s )p a c k e t a r r i v a l i n t e r v a l ( s ) ( b ) p r i o r i t y 0 c a s e 1 c a s e 2 c a s e 3 c a s e 4 0 2 5 0 3 0 5 0 6 0 7 0 8 0 9 1 p a c k e t d e l i v e r y r a t i op a c k e t a r r i v a l i n t e r v a l ( s ) ( c ) p r i o r i t y 1 0 2 5 0 3 0 5 0 6 0 7 0 8 0 9 1 p a c k e t d e l i v e r y r a t i op a c k e t a r r i v a l i n t e r v a l ( s ) ( d ) p r i o r i t y 0Figure 4: Simulation results with dif ferent pack et arri v al rates. light traf c load, the piggybacking scheme can greatly reduce the end-to-end delay and impro v e the pack et deli v ery ratio for both priorities. It is note w orth y that although channel a w are mechanisms (case 2) can impro v e both metrics compared to case 1, our piggybacking scheme pro vides more benets that the channel a w are mechanism alone cannot achie v e for the reasons described abo v e. 4.7.4 Impact of Node Mobility Ne xt, we study the impact of mobility on the performance of the proposed Courtesy Piggybacking scheme when the a v erage pack et inter -arri v al time is 0.25s. In Fig.4, the lar ger the pause time the lo wer the mobility W e observ e that in all the cases the performance is sensiti v e to mobility and de grades as mobility increases. The reason for this is that high mobility may frequently cause route breakages and pack et losses, and hence increase delay and decrease pack et deli v ery ratio. In addition, we can clearly observ e that case 2, case 3, and case 4 ha v e better performance than case 1, which is consistent with our observ ations in pre vious simulations. In general, these three cases ha v e shorter end-to-end delay and higher pack et deli v ery ratio than case 1 for both priorities, especially for the lo w priority traf c, the disadv antageous traf c in the dif ferentiated services system. Since all

PAGE 114

102 0 5 0 1 0 0 1 5 0 2 0 0 2 5 0 3 0 0 0 5 1 0 1 5 p a u s e t i m e ( s )e n d t o e n d d e l a y ( s ) c a s e 1 c a s e 2 c a s e 3 c a s e 4 (a) A v erage end-to-end delay of priority-1 service. 0 5 0 1 0 0 1 5 0 2 0 0 2 5 0 3 0 0 0 0 5 1 1 5 2 2 5 3 3 5 p a u s e t i m e ( s )e n d t o e n d d e l a y ( s ) c a s e 1 c a s e 2 c a s e 3 c a s e 4 (b) A v erage end-to-end delay of priority-0 service. 0 5 0 1 0 0 1 5 0 2 0 0 2 5 0 3 0 0 0 4 0 5 0 6 0 7 0 8 0 9 1 p a u s e t i m e ( s )p a c k e t d e l i v e r y r a t i o c a s e 1 c a s e 2 c a s e 3 c a s e 4 (c) P ack et deli v ery ratio of Priority-1 service. 0 5 0 1 0 0 1 5 0 2 0 0 2 5 0 3 0 0 0 6 0 6 5 0 7 0 7 5 0 8 0 8 5 0 9 0 9 5 1 p a u s e t i m e ( s )p a c k e t d e l i v e r y r a t i o c a s e 1 c a s e 2 c a s e 3 c a s e 4 (d) P ack et deli v ery ratio of Priority-0 service. Figure 4: Simulation results with dif ferent pause time.

PAGE 115

103 the cases e xcept case 1 mak e use of the channel conditions and rate adaptation, ag ain, we v alidate that the dynamic channel conditions can be used to impro v e the channel utilization. W e further compare case 3 and case 4 as a group with case 2 to sho w the ef fecti v eness of our Courtesy Piggybacking scheme. From Fig.4, we can clearly see that our scheme can further shorten the end-to-end delay and impro v e the pack et deli v ery ratio for both types of traf c. Moreo v er the Courtesy Piggybacking scheme not only impro v es the performance of the priority-0 traf c, b ut also impro v es signicantly the performance of the priority-1 traf c. This v eries that our Courtesy Piggybacking scheme is capable of alle viating the conict between the dif ferent prioritized traf c. As discussed in Section4.5, all these g ains come from the Courtesy Piggybacking scheme. In case 2, the channel state information is e xploited only to some e xtent, b ut not fully harv ested in the sense that the free space cannot completely be utilized. Ho we v er our piggybacking scheme can mak e use of these system dynamics, not only the channel dynamics b ut also the traf c dynamics, so that the free space can be best e xploited without an y w aste. 4.7.5 Impact of Piggybacking Rules Finally we focus on case 3 and 4 and study the impact of the piggybacking rules. The piggybacking rule in case 3 f a v ors the high-priority traf c in the system, priority 0, while the rule in case 4 f a v ors the lo w-priority traf c, priority 1. Thus, there is no surprise that in Fig.4(a)and Fig.4(c), the end-to-end delay for the priority 1 in case 4 is generally shorter than that in case 3, and the pack et deli v ery ratio is generally greater than that in case 3. F or the priority-0 traf c, all the measured metrics generally ha v e better performance in case 3 than those in case 4. Compared with the performance in case 3, the Courtesy Piggybacking in case 4 sacrices the priority-0 traf c a little bit to piggyback the priority-1 traf c. In Fig.4(d), we also observ e some oscillations in the pack et deli v ery ratio when the mobility is high, for e xample, when the pause time is less than 60. The pack et deli v ery ratio of priority 0 in case 3 seems v ery sensiti v e to the high mobility and has w orse performance than that in the case 4, the one with piggybacking rule preferring

PAGE 116

104 the lo w priority This can be e xplained as follo ws. When the mobility is high, the pack et loss may primarily result from the mobility of nodes in v olv ed in the communications, not necessarily from the channel impairments due to other f actors. On the other hand, the high mobility prolongs pack et deli v ery and brings do wn the pack et deli v ery ratio, which further results in man y w aiting pack ets of both types in the queues. In case 3, since the piggybacking rule prefers the traf c of priority 0, quite often we may ha v e tw o priority-0 pack ets pack ed together for transmission to the ne xt hop when the channel is in state 1. If the recei v er does not recei v e them successfully due to high mobility in this case, then more pack ets of priority 0 will be dropped, leading to lo wer pack et deli v ery ratio, thus the pack et loss due to high mobility under piggybacking rule in case 3 may be amplied and accordingly de grades the performance further than in case 4 for high-priority traf c. On the contrary in case 4, instead of packing tw o priority-0 pack ets when possible, a sender packs one pack et of priority 1 with one pack et of priority 0. When the pack ed pack ets cannot be successfully recei v ed due to high mobility only one pack et of each priority is in v olv ed, hence the impact on the high-priority traf c is less se v ere. Thus the Courtesy Piggybacking with properly designed piggybacking rules may compensate for the ne g ati v e ef fect of high mobility 4.8 Summary In this chapter we propose a no v el Courtesy Piggybacking scheme to alle viate the conict between throughput and f airness for dif ferent prioritized services in mobile ad hoc netw orks. By making use of system dynamics, such as the v ariable channel quality and changing traf c conditions, it can harness the a v ailable residual bandwidth that w ould otherwise be w asted. Thereby it signicantly impro v es the end-to-end delay and pack et deli v ery ratio. More precisely when the traf c load is light, it can shorten the end-toend delay; when the traf c load is high, it can not only shorten the end-to-end delay b ut also impro v e the pack et deli v ery ratio for all prioritized services. W ith properly dened piggybacking rules, the piggybacking scheme can e xibly allocate the bandwidth among

PAGE 117

105 dif ferent types of traf c, thus achie v e good f airness between dif ferent priority services without using con v entional costly scheduling mechanisms. By deri ving the delay bounds, we e xplicitly specify the performance g ain the piggybacking scheme can achie v e in some simplied scenarios. Further e xtensi v e simulations v erify the performance of our proposed piggybacking scheme. Our scheme is also sho wn to be easily implemented in a distrib uted f ashion, and thus could be incorporated into man y scheduling schemes to pro vide better support of the dif ferentiated and heterogeneous services in both mobile ad hoc netw orks and traditional wireless netw orks.

PAGE 118

CHAPTER 5 R OB UST AND ENERGY -EFFICIENT D A T A DISSEMIN A TION IN WIRELESS SENSOR NETW ORKS 5.1 Introduction Recent adv ances in Micro-Electro-Mechanical Systems (MEMS) technology and wireless communications ha v e resulted in the emer gence of small, lo w-cost sensors with more and more po werful processing and netw orking capabilities, which mak es wireless sensor netw orks (WSNs) be identied as one of the most important emer ging technologies [104]. Especially wireless sensor netw orks can furnish us with ne-granular observ ation about the ph ysical w orld we are li ving in. Potential applications include the remote sensing in nuclear plants, mines, and other hazardous industrial v enues, real-time traf c monitoring, realtime weather monitoring, wild animal monitoring and tracking, disaster rescue, ener gy management, medical monitoring, logistics and in v entory management, and military reconnaissance. While much research has been focused on making sensor netw orks feasible and useful [105] [106], some important problems resulting from the error -prone and resource-constrained nature of WSNs ha v e not been well addressed yet. Notable are the issues associated with scalability reliability and ener gy ef cienc y F or instance, since a WSN may consist of hundreds or thousands or e v en millions of sensor nodes, an ef cient data dissemination technique should w ork well not only in small-scale sensor netw orks b ut also in lar ge-scale ones. In addition, it should be rob ust ag ainst the harsh en vironmental ef fects and temporal or permanent f ailures of sensors and wireless links in between them, so that the functionality of the WSN can be sustained without an y interruption. Moreo v er it should ha v e good ener gy ef cienc y in terms of both lo w a v erage ener gy consumption per observ ation report and balanced ener gy usage instead of o v erb urdening a small set of nodes in the netw ork. 106

PAGE 119

107 The research presented in this chapter tar gets at real-time and continuous monitor ing applications such as battleeld monitoring netw orks and v olcano monitoring netw orks, where sensors are deplo yed in an ad hoc manner and the aforementioned nice properties are desirable. Those sensors collaborati v ely accomplish the sensing task and forw ard the sensing data to the closest data processing centers or sink nodes through wireless links. T raditional routing protocols proposed for ad hoc netw orks may not be suitable for our tar get applications due to the substantial dif ferences between ad hoc netw orks and sensor netw orks [105]. In contrast, ooding, as a reacti v e technique with inbred reliability seems to be a good candidate for sensor netw orks because it does not in v olv e costly topology maintenance and comple x route disco v ery algorithms. Ho we v er the main problems with ooding are that it typically causes unproducti v e and often harmful bandwidth congestion, as well as inef cient use of node resources such as ener gy which is scarce in resource-constrained sensor netw orks. Though se v eral data dissemination schemes ha v e been proposed specically for sensor netw orks [105] [106], research on nding a scheme that can strik e a good balance among reliability scalability and ener gy ef cienc y is still lacking. In this chapter we propose a h ybrid data dissemination frame w ork for WSNs that features lo w o v erhead, high reliability good scalability and e xibility and preferable ener gy ef cienc y Our contrib utions are mainly three-fold. First of all, to the best of our kno wledge, this is the rst ef fort to study a wireless sensor netw ork from the point of vie w of supply chain management. W e introduce the notion of supply c hain into the design of sensor netw orks and conceptually partition a sensor eld into se v eral functional re gions according to the supply chain management methodology Secondly we apply dif ferent routing techniques to dif ferent re gions in order to pro vide better performance in terms of reliability and ener gy consumption. Lastly we propose a no v el zone ooding scheme which is a combination of con v entional ooding and geometric routing. Our rationale here is to of fer the desired reliability and routing simplicity with ooding and to mitig ate the decienc y of blind ooding with geometric routing.

PAGE 120

108 The remainder of this chapter is or g anized as follo ws. W e start with Section5.2describing the related w ork and discussing some basics of supply chains and the resemblance between supply chains and wireless sensor netw orks in Section5.3. Then, we detail our h ybrid data dissemination frame w ork in Section5.4. In Section5.5, simulation studies are carried out to e v aluate the performance of the proposed scheme. The concluding remarks are gi v en in Section5.6. 5.2 Related W ork Ho w to ef ciently deli v er the information in densely deplo yed sensor netw orks is a v ery challenging task. Directed dif fusion [107], and SPIN [108] are tw o e x emplary data dissemination paradigms for sensor netw orks. SPIN adopts meta-data ne gotiation to eliminate the redundant data transmission, and it is suitable for the scenarios where an individual sensor disseminates its observ ations to all the sensors in a netw ork. As a datacentric approach, directed dif fusion emplo ys lo w rate ooding to establish gradients and uses gradual reinforcement of better paths to accommodate certain le v els of netw ork and sink dynamics. Recently Heidemann et al. proposed tw o v ariations of directed dif fusion and the y adv ocated the importance of matching dissemination algorithms to application performance requirements [109]. Besides the abo v e schemes, the cluster -based LEA CH [110] and hierarchical-based TTDD [111] are also a v ailable in the literature. In LEA CH, nodes are or g anized into clusters, each of which has a randomly selected cluster header Sensor nodes send their observ ations to af liated cluster headers and request cluster headers to forw ard their reported data to a base station. As the authors mentioned, their scheme w as designed for applications where a sensor node is able to transmit data to a f ar a w ay base station with higher transmission po wer In TTDD, each data source rst b uilds a grid structure, based on which the sink' s query and sensing data are forw arded through this grid structure and further ooded in a local grid cell. TTDD can accommodate the e xistence of mobile sinks. W e notice that the abo v e approaches need lots of o v erhead to be w orkable, for e xample, the y need e xchange a lot of information with neighbors to b uild routing tables

PAGE 121

109 or to form and propag ate the cluster/grid structures. In addition, reliability in the presence of the error -prone wireless links is another constraint of the abo v e approaches because the y usually maintain only a single route for each source-sink pair Though the multipath e xtension of [107] w as reported [112], it is quite dif ferent from our approach. P articularly the alternati v e paths are only used when some node f ailures occur along the primary path. Moreo v er the multipath e xtension of directed dif fusion is still not able to deal with other transmission f ailures due to the wireless links. In contrast, our RRP inherits the inbred reliability of ooding and is rob ust ag ainst transmission f ailures coming from both node f ailures and wireless link impairment. As a reacti v e technique with inbred reliability ooding seems to be a good candidate for sensor netw orks because it does not in v olv e costly topology maintenance and comple x route disco v ery algorithms. In vie w of this, quite a fe w research ef forts ha v e been done to optimize the use of ooding in terms of reducing the unproducti v e and often harmful bandwidth congestion, as well as inef cient use of nodes resources caused by ooding. Gossiping [113], probabilistic-based ooding [114] [115], and other controlled-ooding approaches [116] belong to this cate gory In contrast, our zone ooding scheme utilizes the node location information instead of a probabilistic method to control the ooding range. Furthermore, based on the partition strate gy and applying dif ferent routing strate gies to different functional re gions, our data dissemination paradigm can well balance the reliability and the ener gy ef cienc y and achie v e good scalability and notable e xibility Geographic routing is another potential candidate for sensor netw orks. It uses nodes' locations as their addresses based on which pack ets are forw arded to w ards the destination in a greedy manner Se v eral approaches ha v e been proposed to impro v e the ef cienc y and accurac y of the geographic routing. In GPSR [117], a detour algorithm w as proposed to o v ercome the possible dead ends with the greedy method. In addition, LAR [64] uses restricted area ooding to reduce the cost of disco v ery Moreo v er in DREAM [63], each

PAGE 122

110 node obtains its geographic location through e xternal de vices such as GPS, and periodically transmits its location coordinates to other nodes in the netw ork. A source sends a data pack et to a subset of its neighbors in the direction of the destination. Unlik e the abo v e approaches that require a mass of location information e xchange among neighboring nodes or require a node to be a w are of the locations of all the other nodes, our zone ooding scheme only requires a node to learn the locations of its o wn and sink nodes, and to mak e a forw arding decision based on its o wn location and the ooding zone information carried in recei v ed pack ets. Therefore, our scheme demonstrates good scalability Besides, our scheme pro vides the option of multi-zone ooding and hence is more resilient to node or route f ailures. T rajectory based routing (TBF) [118] is another notable w ork in geographic routing. TBF embeds a trajectory in each pack et and intermediate nodes around the trajectory collaborati v ely forw ard a gi v en pack et. Dif ferent from TBF our zone ooding scheme uses tw o curv es to specify a ooding zone and all the nodes in the specied zone will par ticipate in forw arding a gi v en pack et in an autonomous w ay Hence our approach is more reliable than TBF More important, our approach is completely loop free, while trajectory based forw arding still risks possible routing loops in some scenarios lik e other geographic routing. Our data dissemination frame w ork demands the location information of nodes, which is indispensable for man y applications, such as tar get tracking and en vironment monitoring [119], and is fundamental for man y geographic routing schemes ([64,63,117]). There is a rich literature in ad hoc netw orks on ho w to retrie v e the location information as accurate as possible. F or e xample, se v eral approaches based on time dif ference of arri v al (TDO A) [120], angle of arri v al (A O A) [121], or signal strength [122] ha v e been proposed for the scenarios without GPS de vices. More recently tw o no v el approaches ha v e been proposed specically for sensor netw orks [123] [124]. W e belie v e that the location information should be ef fecti v ely utilized to pro vide better performance in wireless sensor netw orks, the more accurate the location information is, the more benet we can get.

PAGE 123

111 A B B C C D E F 2 1 3 p a r t s s h i p m e n t p r o d u c t s s u p p l i e r s p a r t s w a r e h o u s e s m a n u f a c t u r e r s p r o d u c t w a r e h o u s e r e t a i l e r s c u s t o m e r s v o l u m e s h i p m e n t A B D A B C D E B C F Figure 5: An e x emplary supply chain. 5.3 Modelling Sensor Netw orks as Supply Chains In this section, we rst introduce some basics of supply chains and then discuss wh y and ho w we can model a WSN as a supply chain. 5.3.1 Introduction to Supply Chains In the b usiness w orld, a supply c hain (SC) is a series of links and shared processes e xisting between suppliers and consumers, which in v olv e all acti vities from the acquisition of ra w materials to the deli v ery of nished goods to the end consumers [125]. Generally speaking, a supply chain consists of se v eral components: ra w material manuf acturers, inter mediate product manuf acturers, end product manuf acturers, wholesalers and distrib utors, and retailers. These components are connected by transportation and storage acti vities, and are inte grated through information, planning, and inte gration acti vities. Fig.5sho ws an e x emplary SC, where ra w materials or parts enter into a manuf acturing or g anization via a supply system and are transformed into nished goods. The nished goods are then supplied to consumers through a distrib ution system. Usually supply chains are operated with certain strate gy to coordinate all the components' functions and to smooth the material and information o ws in the supply chains. In the past 40 years, SC strate gies ha v e e v olv ed from push strate gies to pull strate gies and nally to push-pull strate gies [125]. In a push-type SC, production and distrib ution decisions are based on long-term forecasts. A manuf acturer uses orders recei v ed from retailer w arehouses to forecast demands. The decisions of consumers' orders are based on in v entory

PAGE 124

112 P u s h S t r a t e g y : M a k e t o s t o c k m o d e l P u l l S t r a t e g y : A s s e m b l y t o o r d e r m o d e l P u s hP u s h--P u l l B o u n d a r yP u l l B o u n d a r yP U S H S T R A T E G Y P U L L S T R A T E G YP u s h P u l l S t r a t e g y : H y b r i d m o d e ls u p p l i e r s A s s e m b l y C o n f i g u r a t i o ns u p p l i e r s A s s e m b l y C o n f i g u r a t i o n Figure 5: Supply chain strate gies. rather than consumers' demands. Bull whip ef fects such as e xcessi v e in v entory e xcessi v e production v ariability and poor service le v els, are v ery common in the push-type SC. In contrast, in a pull-type SC, production and distrib ution are demand-dri v en and are based on actual customer demands rather than forecasts. The rm no longer needs to hold an y in v entory and only responds to specic orders. Ho we v er in a pull-type SC, it is hard to le v erage economies of scale. These adv antages and disadv antages of push strate gies and pull strategies ha v e resulted in hybrid strate gies: push-pull strate gies. In push-pull strate gies, push strate gies are applied in the initial stages such as parts in v entory where production and distrib ution decisions are based on long-term demand forecasts by manuf acturers on the basis of orders recei v ed from retailers; and pull strate gies tak e ef fect in the nal stages such as product assembly where production and distrib ution are purely demand dri v en and rely on actual customer demands rather than forecasts. Usually a b uf fer in v entory such as a w arehouse, is located at the push-pull boundary Dell Computer a giant computer manuf acturer and retailer is an e xcellent e xample of the push-pull-type SC. Dell k eeps an in v entory of components and assembles only when there is an actual order Fig.5sho ws those three dif ferent strate gies. On the other hand, supply c hain mana g ement (SCM) is the act of optimizing all acti vities throughout the supply chain, so that products and services are supplied in the right quantity to the right location, at the right time, and at the optimal cost. One of the fundamental concepts in SCM is that all the autonomous entities in the SC may ha v e their o wn inner operations and management strate gies, b ut the y w ork in a cooperati v e f ashion

PAGE 125

113 to achie v e the management goal of the whole SC: satisfying the customer service requirements while minimizing the o v erall system cost and obtaining as much re v enue as possible from the cooperations. No w adays, SCM has been widely accepted in the b usiness w orld as a vital f actor for competiti v e adv antage and sustainable b usiness impro v ement. 5.3.2 Ho w Could Supply Chains Help Us Similar to supply chains, wireless sensor netw orks are designed to cooperati v ely transmit meaningful sensing data (products) to the sinks (retailers) with certain QoS requirements. F or a designed sensing task, the sensors (ra w material manuf acturers) sense the ra w data (ra w materials) as suppliers and send them to some nodes (product manuf actur ers) for further processing. These processed data (products or sometimes intermediate products) are deli v ered hop-by-hop to sinks with the help from intermediate sensor nodes (transporters). In f act, supply chains and wireless sensor netw orks ha v e man y k e y components or functions in common. F or e xample, the parts w arehouse in Fig.5is designed to consolidate ra w parts from dif ferent suppliers and it serv es as a multiple x er to decouple the need from the a v ailability Its counterpart in sensor netw orks is the desirable functionality of data aggre g ation, which is used to combine the data gradually at inter mediate nodes enroute from dif ferent sensor nodes to the sink node. The objecti v es of the data aggre g ation are eliminating redundanc y minimizing the number of transmissions, and thus sa ving ener gy [126]. Further the product w arehouse close to retailers, ensuring lo w in v entory reduced transportation costs, and quick replenishment capability acts a similar role as that of the mechanisms for reducing information implosion [105]. V ital to cost management of supply chains, transportation planning is equi v alent in functionality to routing in sensor netw orks, whose major objecti v e is to transfer sensing data from sensors to sink nodes as ef ciently as possible. Moreo v er both require all the entities in the system to w ork cooperati v ely for goals common in nature: pro viding good quality of service (QoS) and k eeping the o v erall system cost as lo w as possible. In T able5we list the analogous components between supply chains and sensor netw orks. It is a rather soft matching in

PAGE 126

114 T able 5: Analogue between supply chains and wireless sensor netw orks Items in supply chains Counter parts in sensor netw orks Ra w materials or parts Phenomena of interest, for e xample, sound, images, and mo v ements of observ ed objects Suppliers or manuf acturers Sensor nodes generating sensing data T ransportation netw ork Intermediate sensor nodes Distrib utors or retailers Sink nodes Finished products Data processed by sink nodes Consumers End-users of the data of fered by sink nodes that not e v ery sensor node can be absolutely or perfectly t into a matching component in a supply chain architecture. A sensor may act as multiple roles under dif ferent scenarios. F or e xample, one sensor may be a supplier of one communication, while be a tr ansporter of the other communication in the meantime. Intermediate nodes sometimes are also called semi-manuf acturers or parts w arehouses when performing data aggre g ation to reduce the data redundanc y What we emphasize here is the conceptual and technical moti v ations and the approaches leading to the viable solutions behind the tw o seemingly dif ferent systems. Interestingly we notice that tw o notable routing protocols for sensor netw orks, namely directed dif fusion [107] and SPIN [108], conform well with the SC methodology In particular directed dif fusion can be vie wed as a pull-type SC, in which the sink node propag ates its interests throughout the sensor netw ork and sensors possessing the data of interest respond with sensing data via intermediate sensors. Thus, directed dif fusion is a pull-type SC in that pr oduction is demand (inter est) driven In contrast, SPIN with AD V -REQ-D A T A handshaking is more lik e a push-type SC. It is designed for disseminating information to all nodes in a sensor netw ork, where the nodes generating the data can be re g arded as suppliers or manuf acturers in a SC. Thus, SPIN is a push-type SC where or der decisions (REQ) ar e based on in ventory (AD V) The abo v e resemblance between supply chains and wireless sensor netw orks moti v ates us to model a WSN as a SC, thus we can apply the sophisticated kno wledge of SCM in the b usiness w orld to impro v e the performance of the sensor netw orks. More specically we

PAGE 127

115 C o n s u m e r 1 C o n s u m e r 2 C o n s u m e r 3 C o n s u m e r 4 I a m i n t e r e s t e d i n t h e b e a k w h a t c o l o r ? h o w l o n g ? S e r v i c e a r e a W a r e h o u s e a r e a T r a n s p o r t a t i o n a r e a M a n u f a c t u r e a r e a C a n y o u s e n d m e t h e c h i r m o f t h e b i r d ? I c a n t e l l y o u t h e n a m e o f t h e b i r d b y i t s c h i r m L e t m e s e e t h e w h o l e p i c t u r e o f t h e b i r d I c o l l e c t p h o t o s o f f e e t o f a l l k i n d s o f b i r d s T h e y a r e b e a u t i f u l A B C D E F G H I J K L M N O P Q R S T w 4 w 5 w 6 w 7 w 8 w 9 R 3 R 2 w 1 w 2 w 3 Z o n e 1 Z o n e 2 R 1 R 4P u l l P u s h B o u n d a r yP U S H S T R A T E G Y P U L L S T R A T E G Y Figure 5: A system architecture for habitat monitoring kno w that it yields better performance and lo wer cost in SCM by partitioning the supply chain into se v eral dif ferent components ( partition str ate gy ), applying dif ferent management mechanisms to dif ferent components ( hybrid str ate gy ), and designing cooperations among dif ferent components ( cooper ation str ate gy ). Similar ideas can be introduced into sensor netw orks for solving reliability and ef cienc y problems. 5.4 A Hybrid Data Dissemination Frame w ork for W ireless Sensor Netw orks Inspired by SCM methodology we design a no v el h ybrid data dissemination framew ork for wireless sensor netw orks. In this section, we rst describe the system model and then detail the management strate gies or routing techniques applied to dif ferent functional components. 5.4.1 System Model In what follo ws, we use a WSN for habitat monitoring as an e xample to illustrate our scheme. As sho wn in Fig.5, for one specic sensing task, the whole sensor eld is conceptually partitioned into se v eral functional areas according to the aforementioned push-pull strate gies of supply chains (see Section5.3.1). In the manufactur e ar ea some nodes such as those from A to J are in v olv ed in generating the ra w data about the objects of interest, that is, birds in this case, while other nodes such as K L and P are responsible

PAGE 128

116 for data aggre g ation, that is, consolidating the ra w data and reducing possible information o v erlapping. The ltered data is fed into the tr ansportation ar ea to be relayed by inter mediate sensors to sink nodes. In addition, we introduce the war ehouse ar ea as a b uf fer area between the transportation area and the service area to reduce the possible traf c congestion and information implosion [105] at the sink nodes. The service area consists of sink nodes which can directly communicate with each other through f ast and reliable links, either wired or wireless. The sink nodes perform collaborati v e reception of sensing e v ents and of fer dif ferent data items to end-users or consumers with dif ferent interests. W e assume that each node has the kno wledge of its o wn position1and the positions of the sink nodes, which is a reasonable assumption for man y monitoring applications [105,106]. W e further assume that all the nodes including common sensor nodes and sink nodes are identied by their geographic information. W e should emphasize here that the sensor eld partition is only conceptual and application-dependent rather than a x ed one. Basically v arious sensing tasks may ha v e quite dif ferent partition instances. In particular for each specic sensing task, those nodes sensing the e v ents of interest can form a manuf acture area together with their neighboring nodes just for that task. Accordingly the transportation area lies in the forw ard direction from the manuf acture area to w ards the sink nodes. In this dissertation, we assume that the sinks are f ar enough a w ay from the manuf acture area, and thus the transportation area does e xist2. Moreo v er we dene the w arehouse area to be the area within the sink nodes' n -hop range, where n is a tunable design parameter T o form its w arehouse area, one sink node needs to simply broadcast a special request with the TTL v alue set to n An y node recei ving this request becomes a w arehouse member 1 Note that a node can obtain its location information at lo w cost from GPS or some localization system [120,122,121,123,124]. 2 F or those sensing tasks close to the sinks, a modied partition method can be used where transportation area may not be necessary .

PAGE 129

117 (w arehouse node) of the w arehouse area of the requesting sink node. In f act, the v alue of n can be application-dependent, and v arious sensing tasks may ha v e dif ferent v alues of n to reect the di v erse QoS requirements. In contrast to the abo v e three areas whose locations and sizes are closely related to sensing tasks, the service area is determinate because of the in v ariable locations of sink nodes. One of the no v el features of our h ybrid data dissemination paradigm is that we apply dif ferent data forw arding mechanisms in dif ferent functional areas. More specically local broadcasting in the manuf acture area, a unicasting-based routing in the w arehouse area, and a specially designed zone ooding, which is essentially a combination of con v entional ooding and geometric routing, in the transportation area, are used respecti v ely W ith this h ybrid data dissemination paradigm in place, a better balance between ener gy ef cienc y and reliability can be e xpected. In what follo ws, the h ybrid data dissemination paradigm is elaborated in more detail. 5.4.2 Manuf acture Area W e postulate that the nodes in the manuf acture area are a w are of their o wn missions3. Each mission might represent a sensing task of the sensor netw ork. In this e xample, the mission may be collecting the information of birds, such as the beak color the feet length, or e v en the bird chirm. Due to the limitation of sensors' capabilities, each sensor may only sense part of the interested e v ent so that the y might need to locally e xchange some sensing e v ents and select among themselv es one node as an aggre g ation center to fulll the data fusion task. F or e xample, nodes K L and P in Fig.5are selected as aggre g ation centers. Since aggre g ation centers, in most cases, are only se v eral hops a w ay from the sensing nodes, the simplest w ay to forw ard the sensed ra w data to aggre g ation centers is 3 Here we assume that certain mission allocation scheme is used during the deplo yment and initialization phases of sensor netw orks.

PAGE 130

118 to broadcast pack ets with limited TTL v alues. F or lack of space, we do not detail ho w to manage the sensing tasks and accomplish data aggre g ation in this disseratation. Besides data fusion, each aggre g ation center assumes a special role in our data dissemination frame w ork. It is also responsible for determining the transportation method for the ltered data by itself, that is, using single zone ooding or multi-zone ooding, and the proper transportation zone(s) through which the data will tra v el in the tr ansportation ar ea F or e xample, after nishing the aggre g ation of the ra w data from nodes E F and G node P mak es choice of using tw o ooding zones and then chops the ltered data into tw o parts, both of which are labelled with their respecti v e designated ooding zone. In f act, the operations in the manuf acture area e xhibit lots of e xibility during the selection of the transportation methods and the ooding zone(s). W ith proper selection, our scheme can strik e a good balance between reliability and ener gy ef cienc y F or e xample, if the en vironment is good, an aggre g ation center can choose a single ooding zone instead of multiple ooding zones which are usually used in the f ace of harsh en vironment to forw ard the data. Moreo v er if no w arehouse area is allocated and the ooding zone is the whole sensor eld, our zone ooding e xpands to blind ooding; on the other hand, if we can squash the ooding zone into an area containing only a single path, our zone ooding reduces to single path routing. F or a b ursty and b ulk y e v ent report, an aggre g ation center can choose multiple ooding zones and split the report into se v eral portions which are simultaneously deli v ered to multiple sinks through dif ferent ooding zones. Since sink nodes can communicate with each other through f ast and reliable links, the y can e xchange the recei v ed portions and easily reconstruct the original e v ent report. In this w ay a reduction in the end-to-end delay can be e xpected. Such multipath approaches are also well kno wn for their ef fecti v eness in increasing the reliability and security of data disseminations [127]. Moreo v er to combat the node or wireless link f ailures especially in a harsh en vironment, an aggre g ation center may introduce some redundanc y by sending duplicate reports to the sinks through multiple ooding zones. W e should note that, aggre g ation centers can v ary

PAGE 131

119 the sizes and the locations of ooding zones with time in order to distrib ute the traf c load more e v enly among sensors and thus a v oid o v erb urdening a small set of nodes. Besides, if proper scheduling is a v ailable, nodes can enter into w ak e or sleep modes zone by zone to sa v e ener gy In what follo ws, we shall discuss ho w an aggre g ation center chooses appropriate ooding zones, what data structure a data pack et manufactur ed by an aggre g ation center may use, and ho w a node in the w arehouse area processes a zone-ooded pack et. 5.4.3 T ransportation Area Sensor nodes in the transportation area undertak e the task of relaying data to possible multiple sink nodes. T o a v oid costly topology maintenance and comple x route disco v ery algorithms, we propose a no v el zone ooding scheme, which is a combination of con v entional ooding and geometric routing techniques. The basic idea is as follo ws: a zone containing the source and the destination is specied by some geometric means; then, instead of netw ork-wide omnidirectional ooding, the zone ooding is guided along the direction from the source to the destination and is restricted in the designated zone. Once a node recei v es a pack et carrying parameters that identify a ooding zone, it rst checks whether or not it is in the indicated zone through some tri vial calculations based on its o wn location information and the recei v ed zone parameters. Only when located in the ooding zone w ould it rebroadcast the pack et. Figure 5-3 showsanexampleofthezoneoodingscheme,inwhichellipses 4 areused to specify the ooding zones. Suppose one of the aggre g ation centers, say node P has the coordinates ( x 1 ; y 1 ) in the cartesian plane of Fig.5, and the intended sink node R 3 has the coordinates ( x 2 ; y 2 ). W e should mention that all the required information for zone ooding are embedded in the transmitted data pack ets rather than the control pack ets. 4 As we will see, ellipse can be represented with lo w o v erhead. More important, it is easy to manipulate.

PAGE 132

120 P ( x1, y1) R3 ( x2, y2) U ( x3, y3) X Y f ( x y ) g ( x y ) F1 fF2 fF1 gF2 gO O Y X 1 2 3 4 I I I I I I U U U U C a s e 1 : b 1 b 2 ( b 1 < b 2 ) C r i t e r i a : y 3 = ( y 3 ( y 1 + y 2 ) / 2 ) > 0 a n d L ( b 1 ) < 2 a < L ( b 2 ) C a s e 2 : b 1 b 2 ( b 1 < b 2 ) C r i t e r i a : y 3 = ( y 3 ( y 1 + y 2 ) / 2 ) < 0 a n d L ( b 1 ) < 2 a < L ( b 2 ) C a s e 3 : b 1 b 2 ( b 1 < b 2 ) C r i t e r i a : y 3 = ( y 3 ( y 1 + y 2 ) / 2 ) > = 0 a n d L ( b 2 ) < 2 a o r y 3 = ( y 3 ( y 1 + y 2 ) / 2 ) < 0 a n d L ( b 1 ) < 2 a C a s e 4 : b 1 b 2 ( b 1 < b 2 ) C r i t e r i a : y 3 = ( y 3 ( y 1 + y 2 ) / 2 ) > = 0 a n d L ( b 1 ) < 2 a o r y 3 = ( y 3 ( y 1 + y 2 ) / 2 ) < 0 a n d L ( b 2 ) < 2 a .P R3P R3P R3P R3Y Y Y Y Y X X X X O O O O Figure 5: The forw arding-decision-making process of nodes in the transportation area. A C l o c a t i o n S i n k l o c a t i o n I n n e r S e m i m i n o r A x i s P a y l o a d I n t e r e s t ( e v e n t ) D e s c r i p t i o n ( l o c a t i o n t i m e t y p e ) O t h e r C o n t r o l F i e l d s ( m u l t i p l e z o n e f l o o d i n g r e d u n d a n c y . ) W a r e h o u s e F l a g O u t e r S e m i m i n o r A x i sFigure 5: An e x emplary pack et structure.

PAGE 133

121 Figure 5-5 showsthetypicalstructureofadatapacketmanufacturedbyanaggregation node. Besides the ltered data, each data pack et sent from node P will carry four e xtra zone parameters: A C location indicating the coordinates of the aggre g ation node P Sink location indicating the coordinates of the sink node R 3 and Inner -SemiminorAxis and Outer -SemiminorAxis indicating the semiminors of the inner and outer ellipses5of the desired ooding zone respecti v ely The eld W arehouse Flag is used to inform intermediate nodes whether this pack et has e v er been processed by a w arehouse node. The usage of this eld will be discussed in Section5.4.4. In addition, based on the elds Interest (e v ent) Description and Other Control Fields, intermediate nodes and sinks can determine whether one recei v ed pack et has already been processed, or it needs to assemble those partitioned pack ets belonging to the same interest, or to remo v e the possible redundanc y added. The concrete use of these elds depends on specic applications, which is be yond the scope of this dissertation. From the ellipse geometry we kno w that when the tw o endpoints of the major axis are x ed (that is, A C location of P and Sink location of R 3 ), a v alue of the semiminor axis can uniquely determine an ellipse. And tw o dif ferent semiminor v alues (that is, Inner SemiminorAxis b 1 and outer -SemiminorAxis b 2 ) will determine tw o ellipses as sho wn in Fig.5, f ( x; y ) and g ( x; y ) Such ellipses can be used to specify multiple ooding zones. F or e xample, the tw o ellipses with the same endpoints of the major axis can jointly deter mine three non-o v erlapping ooding zones Zone I between curv e 1 and curv e 2, Zone II between curv e 2 and curv e 3, and Zone III between curv e 3 and curv e 4. In f act, an y tw o of the four curv es can specify a ooding zone, for e xample, a bigger zone determined by curv e 1 and curv e 3. W e already kno w that, with the tw o endpoints of the major axis x ed, the semiminor axis uniquely determines an ellipse. But we need a means to dif ferentiate 5 Here the inner ellipse means the ellipse with smaller semiminor axis, while the outer ellipse means the ellipse with bigger semiminor axis.

PAGE 134

122 the tw o curv es constituting the ellipse the upper half (abo v e the X 0 axis, for e xample, curv e 1 and 2) and the lo wer half (belo w the X 0 axis, for e xample, curv e 3 and 4). In our design, we adopt a simple rule we use the positi v e v alue, for e xample, b 1 or b 2 to denote the upper half of an ellipse while using the ne g ati v e v alue to denote the lo wer half curv e of the ellipse in the shifted coordinate plane. F or e xample, b 1 together with b 2 determines zone I, )Tj/T1_1 11.95509 Tf9.28912 0 Td(b 1 and )Tj/T1_1 11.95509 Tf9.28912 0 Td(b 2 jointly specify zone III, and )Tj/T1_1 11.95509 Tf9.28912 0 Td(b 1 and b 2 select zone I+II. In particular Inner -SemiminorAxis b 1 can be 0 to denote the X 0 axis as a special elliptic curv e. Thus, a f (+ = )Tj/T1_6 11.95509 Tf9.28912 0 Td() b 1 (+ = )Tj/T1_6 11.95509 Tf9.28912 0 Td() b 2 g pair of real numbers can uniquely determine a ooding zone. Moreo v er by v arying the v alues of semiminor axis, we can easily get ph ysically separated or interlea v ed multiple paths (multiple ooding zones) without incurring an y signicant additional costs. As discussed in Section5.4.2, the multipath routing is a po werful technique to impro v e the reliability and security among other system performance f actors. When one node, say U with coordinates ( x 3 ; y 3 ), recei v es a data pack et containing the abo v e-mentioned zone parameters f ( x 1 ; y 1 ) ; ( x 2 ; y 2 ) ; (+ = )Tj/T1_6 11.95509 Tf9.28912 0 Td() b 1 ; (+ = )Tj/T1_6 11.95509 Tf9.28912 0 Td() b 2 g the question whether or not it should rebroadcast the pack et is reduced to a simple geometric problem: whether or not point U lies between the tw o elliptic curv es specied by the embedded parameters? Suppose the semiminor axis of an ellipse with tw o major -axis endpoints P and R 3 is b Then the sum of the distance from point U to tw o x ed points F 1 and F 2 (the foci) can be e xpressed as L ( b ) = D 1 + D 2 [128], where D 1 = r ( q ( x 1 )Tj/T1_10 7.97011 Tf6.5833 0 Td(x 2 ) 2 +( y 1 )Tj/T1_10 7.97011 Tf6.5833 0 Td(y 2 ) 2 4 )Tj/T1_7 10.9091 Tf10.89729 0 Td(b 2 + ( x 3 )Tj/T1_10 7.97011 Tf12.105 4.49101 Td(x 1 + x 2 2 )) 2 + ( y 3 )Tj/T1_10 7.97011 Tf12.105 4.93201 Td(y 1 + y 2 2 ) 2 and D 2 = r ( q ( x 1 )Tj/T1_10 7.97011 Tf6.5833 0 Td(x 2 ) 2 +( y 1 )Tj/T1_10 7.97011 Tf6.5833 0 Td(y 2 ) 2 4 )Tj/T1_7 10.9091 Tf10.9063 0 Td(b 2 )Tj/T1_8 10.9091 Tf10.89729 0 Td(( x 3 )Tj/T1_10 7.97011 Tf12.105 4.49098 Td(x 1 + x 2 2 )) 2 + ( y 3 )Tj/T1_10 7.97011 Tf12.105 4.93198 Td(y 1 + y 2 2 ) 2 : T o determine whether a node itself is in the specied ooding zone, it needs to compare the distance to the foci of each ellipse with 2 a where a = p ( x 1 )Tj/T1_1 11.95509 Tf11.95309 0 Td(x 2 ) 2 + ( y 1 )Tj/T1_1 11.95509 Tf11.94411 0 Td(y 2 ) 2 = 2 is the semimajor axis of the ellipse with major -axis endpoints P and R 3 In Fig.5we

PAGE 135

123 gi v e four possible cases and the corresponding decision criteria that a node may lie in the specied ooding zone. Therefore, for Case 1 where b 1 < b 2 6= 0 in our e xample (Fig.5), node U needs to rebroadcast the pack et if ( y 3 )Tj/T1_6 7.97011 Tf12.85201 5.25604 Td(y 1 + y 2 2 ) > 0 and L ( b 1 ) < 2 a < L ( b 2 ) Otherwise, it will simply ignore the pack et because it is not in the designated ooding zone for that pack et. F ollo wing the abo v e procedures, sensor nodes ( tr ansporter s ) in the transportation area can nally relay a data pack et to the w arehouse area through multi-hop wireless links. In our e xample, since we use tw o elliptic curv es to specify one ooding zone, the Flooding Zone P arameters eld only needs to include tw o v alues for the inner and the outer semiminor ax es, respecti v ely Although an y tw o noncrossing curv es sharing the same tw o ends could be used to specify a ooding zone, we should cautiously choose those curv es that not only can be represented with as fe w bytes as possible to reduce the communication o v erhead, b ut also can simplify the forw arding-decision-making processes of intermediate nodes. In this sense, arcs and ellipses are tw o promising candidates. Moreo v er a ooding zone specied by tw o curv es should be wide enough to ha v e suf cient nodes to forw ard the pack ets while maintaining high ener gy ef cienc y in the meantime. T o accomplish this, the aggre g ation center in our e xample, say node P should properly choose the v alues of the semiminors of the tw o ellipses. Besides, to balance the nodal usage in the transportation area, aggre g ation centers should v ary the ooding zones by using dif ferent and alternating ne g ati v e and positi v e v alues for semiminor ax es. By doing so, our scheme could achie v e e v en load balance and f air ener gy usage without incurring an y signicant additional costs. 5.4.4 W arehouse Area and Service Area In the e x emplary SC sho wn in Fig.5, the w arehouse near the retailers creates a break point in the mo v ement of the products and acts as a b uf fer to reduce the cost of stock at the retailers, hence impro ving the e xibility of the retailers. The w arehouse frequently updates its in ventory list to the retailers, and the retailers can quickly get the products out

PAGE 136

124 W a r e h o u s e A r e a S e r v i c e a r e a w 4 w 5 w 6 w 7 w 8 w 9 R 3 R 2 w 1 w 2 w 3 A D V R E Q D A T A R 1 R 4 Figure 5: The routing process in the w arehouse area. of stock in the stores replenished from the w arehouse. In addition, the w arehouse may consolidate small shipments into a lar ger shipment to the same retailer to sa v e transportation costs. W e notice that such a w arehouse component is also needed in sensor netw orks for realtime and continuous monitoring applications. In these applications, b ursty and b ulk y traf c may be simultaneously transmitted to sink nodes, as a result of which notorious traf c congestion may happen frequently in the vicinity of sink nodes and thus cause the unf a v orable loss of information and the w aste of scarce netw ork resources. Moreo v er the redundant pack ets ooded to w ards sinks may result in the information implosion problem as well. Thus, the introduction of the w arehouse area as a b uf fer area can decouple need ( inter ests ) from a v ailability ( r edundant e vent r eports ) and hence help mitig ate the abo v e information implosion problem and possible traf c congestion. F or the w arehouse area, we use a modied SPIN [108] instead of zone ooding as the underlying routing protocol. Dif ferent from the AD V -REQ-D A T A e xchange in SPIN that happens between neighbors, the AD V -REQ-D A T A e xchange in our scheme is between a w arehouse node and a sink node se v eral hops a w ay and is realized by end-to-end unicast. The unicasting path could be established using the Destination-Sequenced Distance-V ector Routing protocol (DSD V) [7]. W e note that only those w arehouse nodes may participate in

PAGE 137

125 the routing maintenance acti vities. W ith a limited w arehouse size, small o v erhead of routing maintenance can be e xpected. Of course, other routing schemes are applicable to this area as well. Fig.5illustrates our routing strate gy used in the war ehouse ar ea Once recei ving data pack ets from outside the w arehouse area (by e xamining the eld W arehouse Field), a w arehouse member of the desired sink of the pack ets (cf. Section5.4.1), say W 7 lying on the boundary of the w arehouse area, rst sets the eld W arehouse Flag and temporarily stores the pack ets. Afterw ards, W 7 the data holder will unicast an AD V message, essentially an in ventory containing the descriptors of stored pack ets, to the tar geted sink node R 3 either on a per -pack et basis or periodically or when the number of stored pack ets e xceeds a threshold. F ollo wing the reception of the AD V message, R 3 will send a REQ message requesting the interested data. Upon recei ving the REQ message, W 7 can unicast the requested data to R 3 via a D A T A message. In case that R 3 does not recei v e the requested data in time after sending out a REQ, it can resend a REQ to the same data holder W 7 or another data holder who also sent to it an AD V containing the descriptors of the same data. Note that, just as what the w arehouse does in the b usiness SC, if a w arehouse node has enough storage space, it can periodically consolidate se v eral stored pack ets and send them in a v olume shipment to the same sink, in which w ay ener gy sa vings can be e xpected. Here we w ant to e xplain ho w the w arehouse area can help reduce the information implosion. Suppose pack ets describing the same e v ent for sink R 3 arri v e at both W 4 and W 7 which are located in the same ooding zone and the w arehouse area of R 3 When no AD V -REQ-D A T A e xchange is used, depending on the routing protocol, both W 4 and W 7 will broadcast or unicast the pack ets for the same e v ent to the sink R 3 which will lead to information implosion and ener gy w aste o wing to redundant pack ets, especially when numerous nodes send the same e v ent report to the sink. Ho we v er using the proposed AD V REQ-D A T A procedure, both W 4 and W 7 will send short AD V messages rather than relati v e long data pack ets to R 3 It is up to R 3 to mak e a decision on which data holder it should

PAGE 138

126 pull the data from based on certain criteria such as hop count or delay Suppose W 7 is chosen, R 3 will send a REQ to W 7 and accordingly W 7 can unicast the requested data in a D A T A message to R 3 After a certain period, W 4 may delete the stored stale data. From this e xample, we can see that redundant pack ets can be successfully eliminated by the means of AD V -REQ-D A T A e xchange. Moreo v er sink nodes in the service area perform collaborati v e reception in the sense that the y could communicate with each other through f ast and reliable means, for e xample, wired links or separate wireless channels. F or e xample, if sink R 2 recei v es an AD V message from node W 1 it can contact other sink nodes f ar from W 2 to see if the y need the pro vided data, though R 2 itself may not need it in some cases. Suppose R 4 needs the data, R 2 can help obtain the data from W 1 and send it to R 4 Such collaboration also helps a w arehouse node deal with cases when the w arehouse node has no unicasting route to the desired sink of the report due to node f ailures or other reasons. In such a case, the w arehouse node may choose to send the AD V message to a nearest sink. F or instance, in Fig.5, W 1 is a common w arehouse member of both sinks R 2 and R 3 F or some reason, W 1 temporarily has no unicasting route to the reports' destination R 3 and thus it unicasts an AD V to R 2 instead. After recei ving the AD V from R 2 R 3 may request R 2 to help collect the other portions of the information which are destined to R 3 After collecting al the portions, R 3 can reconstruct the original information manuf actured by node P and serv e consumers later 5.4.5 Discussion Our h ybrid data dissemination frame w ork is an open frame w ork and allo ws dif ferent routing techniques to coe xist in the same netw ork. In f act, depending on the applications, a v ariety of routing protocols, not limited to the ones presented in this chapter can nd their possible applications in this frame w ork. In addition, similar to its support of scalability in database engineering [129], the partitioning strate gy enables our scheme to be more scalable and e xible, and reduces the dif culty of designing a feasible o v erall routing scheme.

PAGE 139

127 The decision on ho w to partition the sensor eld and what routing technique should be used for each functional area depends on the application requirements and sensor eld features. F or e xample, multipath routing could be used in the w arehouse area if it is in a v ery harsh en vironment. Further with more information about the sensor eld, the ooding zones can be specied in an ef cient and adapti v e w ay F or e xample, some irre gular curv es can be used to a v oid the ph ysical obstacles in the eld or to a v oid the dead zone with sparse sensor nodes or empty of sensor nodes. As an another e xample, a ooding zone can be specied in a smart w ay such that it co v ers a relati v e small area where communications are of good quality while co v ers a relati v e lar ge area where communications are of bad quality Moreo v er our zone ooding scheme does not e xclude other ooding enhancement mechanisms. Note that our zone ooding scheme attempts to impro v e the ener gy ef cienc y of ooding by restricting the ooding range in the spatial domain. W e can further impro v e the ener gy ef cienc y in the temporal domain. F or e xample, if wireless links are reliable enough, a redundanc y elimination technique can be enabled to further optimize the pack et ooding, which w orks as follo ws: sensor nodes k eep track of redundant pack ets recei v ed o v er a short time interv al, termed Random Assessment Delay (RAD), randomly chosen from a uniform distrib ution between 0 and T max seconds, where T max is the longest possible delay; each node needs to rebroadcast one gi v en pack et if not recei ving redundant ones during the RAD. This RAD method is designed to reduce the collisions among neighboring nodes and to eliminate unnecessary transmissions for one pack et. Besides, ener gy information can be incorporated into the routing decision. F or e xample, an ener gy-ef cient cost metric can be used in the w arehouse area to set up ener gyef cient paths. In the abo v e RAD enhancement, residual ener gy information can also be mapped into delays such that nodes with more ener gy can assume smaller delays than those

PAGE 140

128 nodes short of ener gy Actually ener gy ef cienc y can be further impro v ed in other dimensions. F or instance, if the ooding zone is properly chosen, we can mak e use of Gossipingbased ooding [113], probabilistic-based ooding [114] [115], and other controlled-ooding approaches to further impro v e the ener gy ef cienc y Considering the f act that some sensor nodes may ha v e much more po werful capabilities in terms of ener gy and other important resources than common sensor nodes, more enhancement schemes can be proposed to mak e use of such heterogeneity to operate and manage the whole sensor netw ork more ef fecti v ely and ef ciently In some cases, such po werful nodes may perform data fusion and other management acti vities. The y may e v en act as freight agencies as what UPS or Fede x does in the b usiness supply chains. The aggre g ation centers can send their data to these freight agencies and further these agencies multiple x the data and ship the data in a manner of mass transportation through some high speed data link. In some circumstances, some mobile sensor nodes may be deplo yed in the eld. In such cases, the mobile sensor nodes can perform as mobile data collectors, patrolling around some area and collecting data and carrying them to freight agencies or data processing agencies. These mobile nodes can greatly f acilitate the data forw arding process. Moreo v er in practice, it is possible to replace the depleted nodes near the sinks, or deplo y some nodes with solar po wer supply in the eld. 5.5 Performance Ev aluation In this section, we e v aluate the performance of our h ybrid data dissemination framew ork through simulations. W e rst describe the simulation congurations, the performance metrics, and our simulation methodology W e then study the impact of some en vironment and control parameters on the performance of our scheme. W e compare our scheme with ooding and directed dif fusion [107] in terms of ener gy ef cienc y and reliability 5.5.1 Methodology and Metrics T o v alidate the ef fecti v eness and ef cienc y of our proposed scheme (denoted by RRP), we ha v e de v eloped an e v aluation en vironment within GlomoSim [130] and implemented

PAGE 141

129 s r c 1 s r c 2 s r c 3 s i n k 1 s i n k 2 s i n k 3 1 2 3 4 5 Figure 5: The simulated sensor eld. our h ybrid data dissemination paradigm, including the zone ooding scheme designed for the transportation area and the unicast-based AD V -REQ-D A T A e xchange for the w arehouse area. W e simulate a sensor eld consisting of 606 sensor nodes. W e ha v e 3 independent equally-spaced sources at the left boundary of the sensor eld and 3 independent equallyspaced sinks at the right boundary of the sensor eld. The other 600 sensors are uniformly deplo yed in the sensor eld. The simulated netw ork is sho wn in Fig.5, where the sensor eld is composed of the transportation area and the w arehouse area only The manuf acture area and the service area are on the boundaries of the eld and omitted for simplicity Besides, each of the three sources generates a 128-byte data pack et (e v ent) destined to a randomly selected sink e v ery 1.5 seconds, 1.0 second, and 2.0 seconds, respecti v ely Since our purpose is to study the routing performance of the h ybrid data dissemination framew ork, with an emphasis on the zone ooding scheme, there is no data fusion and collaboration among sink nodes implemented in our simulation. Therefore, a source acts as both an e v ent observ er and a data aggre g ation center T able5lists the conguration parameters of our simulation, where the transmission/reception po wer consumption of sensors are in line with those of Motes [131]. In our simulation, we v ary the size of the w arehouse area, dened as the area within a sink node' n -hop range, to see its impact on RRP' s performance. Besides, each sourcesink pair uses equally-spaced elliptic curv es to partition the entire sensor eld into d (another control parameter) non-o v erlapping ooding zones. F or e xample, Fig.5sho ws the curv es used by source sr c 2 and sink 2 where d = 4 W e e v aluate the impact of the ooding zone size on RRP' s performance by v arying the v alue of d Since the eld size is

PAGE 142

130 T able 5: Simulation conguration Simulation Area 500 m 300 m Number of Nodes 606 T ransmission Range 40 m Initial Ener gy 60 J T ransmit Po wer 81 mW Recei v e/Idle Po wer 30 mW MA C Protocol IEEE 802.11 Radio Bandwidth 2 M bps Data P ack et 128 B y tes Directed Dif fusion Interests 36 B y tes AD V/REQ 12 B y tes Simulation time 15 min x ed, the bigger the v alue of d the narro wer the ooding zone is. Furthermore, we introduce an en vironment parameter namely pack et error rate ( PER ), to reect the error -prone natures of wireless links in WSNs. F or simplicity we assume the error properties of all radio transmissions are independent b ut ha v e the identical PERs during one simulation run. Therefore, by v arying the control parameters n and d and the en vironment parameter PER we can study the performance of our proposed RRP under dif ferent settings. Moreo v er we compare our RRP with ooding and directed dif fusion [107] in terms of ener gy ef cienc y and reliability In our implementation, directed dif fusion uses delay as the criterion to select preferred neighbors during reinforcement. F or our RRP we use the AD V -REQ-D A T A e xchanges (cf. Section5.4.4) and our proposed zone ooding (cf. Section5.4.3) as underlying routing techniques for w arehouse area and transportation area, respecti v ely There are four performance metrics of interest to us. The e v ent deli v ery ratio (EDR) reecting the reliability is dened as the ratio of the total number of e v ent reports (data pack ets) that are successfully deli v ered from the sources to the intended sinks o v er the total number of pack ets generated at the sources. The normalized ener gy consumption reecting the ener gy ef cienc y is dened as the ener gy consumption per pack et per node normalized by the ener gy consumption for one single pack et reception. The a v erage e v ent

PAGE 143

131 end-to-end delay is dened as the a v erage delay from when a pack et (e v ent) is generated and transmitted by the source till it is recei v ed by the sink. And the a v erage r outing o v erhead is dened as the a v erage number of routing pack ets generated per data pack et. 5.5.2 Simulation Results F or the rst set of gures (Fig.5Fig.5), we x the number of ooding zones to be 4, that is, d = 4 and study the performance of RRP under dif ferent en vironment conditions, that is, dif ferent v alues of pack et error rate ( PER ), and with dif ferent w arehouse size n 0 0 0 0 1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0 6 0 0 0 7 0 0 0 8 0 0 0 9 0 0 1 0 8 6 0 8 8 0 9 0 9 2 0 9 4 0 9 6 0 9 8 1 1 0 2 P a c k e t e r r o r r a t e ( P E R )E v e n t d e l i v e r y r a t i o ( E D R ) d i r e c t e d d i f f u s i o n f l o o d i n g R R P ( n = 1 ) R R P ( n = 2 ) R R P ( n = 3 ) R R P ( n = 4 ) R R P ( n = 5 ) Figure 5: Ev ent deli v ery ratio vs. pack et error rate. 0 0 0 0 1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0 6 0 0 0 7 0 0 0 8 0 0 0 9 0 0 1 0 2 4 6 8 1 0 1 2 1 4 1 6 1 8 P a c k e t e r r o r r a t e ( P E R )N o r m a l i z e d e n e r g y c o n s u m p t i o n d i r e c t e d d i f f u s i o n f l o o d i n g R R P ( n = 1 ) R R P ( n = 2 ) R R P ( n = 3 ) R R P ( n = 4 ) R R P ( n = 5 ) Figure 5: Normalized ener gy consumption vs. pack et error rate. Figure 5-8 compareseventdeliveryratioofRRP,pureooding,anddirecteddiffusion under dif ferent PERs. As we can see, since directed dif fusion maintains single path

PAGE 144

132 0 0 0 0 1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0 6 0 0 0 7 0 0 0 8 0 0 0 9 0 0 1 0 1 0 1 5 0 2 0 2 5 0 3 0 3 5 0 4 0 4 5 0 5 P a c k e t e r r o r r a t e ( P E R )E v e n t e n d t o e n d d e l a y ( s ) d i r e c t e d d i f f u s i o n f l o o d i n g R R P ( n = 1 ) R R P ( n = 2 ) R R P ( n = 3 ) R R P ( n = 4 ) R R P ( n = 5 ) Figure 5: A v erage e v ent end-to-end delay vs. pack et error rate. 0 0 0 0 1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0 6 0 0 0 7 0 0 0 8 0 0 0 9 0 0 1 0 5 1 0 1 5 2 0 2 5 3 0 3 5 P a c k e t e r r o r r a t e ( P E R )A v e r a g e r o u t i n g o v e r h e a d d i r e c t e d d i f f u s i o n f l o o d i n g R R P ( n = 1 ) R R P ( n = 2 ) R R P ( n = 3 ) R R P ( n = 4 ) R R P ( n = 5 ) Figure 5: A v erage routing o v erhead vs. pack et error rate. for each source-sink pair its EDR is v ery sensiti v e to the change of PER, dropping almost linearly from 99 % to 86 % with the increase of PERs. In contrast, the EDRs of ooding al w ays stabilize around 100 % This result is not surprising because of the inbred reliability of ooding techniques. Our RRP demonstrates a stable EDR greater than 99 % under all v e dif ferent w arehouse sizes. This result indicates that our RRP using zone ooding in the transportation area and unicasting in the w arehouse area, has achie v ed the reliability comparable to that of ooding, b ut superior to that of directed dif fusion. W e can also observ e the small de gradation of EDR with the increase of n in RRP which can be e xplained as follo ws: since our RRP uses unicasting in the w arehouse area, data pack ets tra v elling in the w arehouse area may suf fer from pack et dropping as in direct dif fusion. Intuiti v ely ,

PAGE 145

133 the greater n is, the more hops a pack et may tra v el in the w arehouse area, hence increasing the dropping probability of pack ets. F ortunately our AD V -REQ-D A T A e xchanges can compensate for such possible pack et droppings by allo wing sink nodes to recei v e AD Vs from multiple data holders and retransmit REQs to a data holder if the e xpected D A T A message is not recei v ed in time. Therefore, our RRP can still maintain a pretty high EDR comparable to that of ooding in the f ace of error -prone wireless links. Figure 5-9 showsnormalizedenergyconsumptionofourRRP,oodinganddirected dif fusion. Compared with the other tw o schemes, directed dif fusion demonstrates the minimum ener gy consumption because it uses lo w rate ooding for interest propag ation and unicasting for data pack ets. Our RRP outperforms ooding because of the use of zone ooding instead of netw ork-wide ooding. W e observ e that the greater n is, the less ener gy our RRP consumes. The reason is that zone ooding and unicasting are respecti v ely used in the transportation area and the w arehouse area. W ith the increase of the w arehouse size n more unicasting and less zone ooding will be in v olv ed and the former is kno wn to be more ener gy ef cient than the latter Figure 5-10 givesaverageeventend-to-enddelayofeachschemeunderdifferent PERs Since directed dif fusion adopts minimum-delay paths, it has the shortest delay among the three schemes. F or pure ooding, the netw ork-wide pure ooding of data packets may result in much more collisions than our zone ooding and pack ets belonging to the same source-sink pair may follo w dif ferent, quite unpredictable, and possibly v ery long routes. Therefore, pure ooding e xperiences longer a v erage e v ent delay than that of our RRP F or our RRP since an e v ent report rst tra v els through the transportation area and then the w arehouse area, there are tw o sources contrib uting to the delay One is the zone ooding in the transportation area, and the other is the AD V -REQ-D A T A e xchanges in the w arehouse area. The a v erage delay coming from zone ooding decreases with the increase of n while the a v erage delay coming from AD V -REQ-D A T A e xchanges increases with the

PAGE 146

134 0 0 0 0 1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0 6 0 0 0 7 0 0 0 8 0 0 0 9 0 0 1 0 8 6 0 8 8 0 9 0 9 2 0 9 4 0 9 6 0 9 8 1 1 0 2 P a c k e t e r r o r r a t e ( P E R )E v e n t d e l i v e r y r a t i o ( E D R ) d i r e c t e d d i f f u s i o n f l o o d i n g R R P ( d = 2 ) R R P ( d = 3 ) R R P ( d = 4 ) R R P ( d = 5 ) Figure 5: Ev ent deli v ery ratio vs. pack et error rate. increase of n An interesting observ ation here is that the delay performance is not monotone with re g ard to n Our RRP achie v es smallest delay when n = 1 follo wed by n = 5 and n = 2 the other tw o are v ery close and not easy to tell them apart. This observ ation suggests a tradeof f between zone ooding and w arehouse routing should be made to achie v e a desired latenc y Figure 5 presentsaverageroutingoverheadofeachscheme.Amongthethree schemes, directed dif fusion requires sinks to periodically ood inter ests to maintain the gradients, as a result of which it encounters the lar gest routing o v erhead. F or our RRP the routing o v erhead comes from the routing maintenance in the small w arehouse area. Therefore, its routing o v erhead is much smaller than that of directed dif fusion b ut lar ger than that of pure ooding which is supposed to ha v e zero routing o v erhead. Besides, since the o v erhead of RRP mainly comes from the routing maintenance in the w arehouse area, it is of no surprise that we observ e the o v erhead of RRP increases with the increase of n that is, the w arehouse size. In the second set of gures (Fig.5Fig.5), we present our results on the RRP performance with a x ed w arehouse size n = 3 b ut dif ferent PER s. W e also v ary the control parameter d to study the impact of dif ferent ooding zone sizes.

PAGE 147

135 0 5 1 0 1 5 2 0 d i r e c t e d d i f f u s i o n f l o o d i n g R R P ( d = 2 ) R R P ( d = 3 ) R R P ( d = 4 ) R R P ( d = 5 ) (a) Normalized ener gy consumption vs. d (PER=0.0005). 0 0 2 0 4 0 6 0 8 d i r e c t e d d i f f u s i o n f l o o d i n g R R P ( d = 2 ) R R P ( d = 3 ) R R P ( d = 4 ) R R P ( d = 5 ) (b) A v erage e v ent end-to-end delay vs. d (PER=0.0005). Figure 5: Ener gy consumption and e v ent end-to-end delay of RRP with dif ferent d

PAGE 148

136 Fig.5sho ws the e v ent deli v ery ratio with dif ferent d W e observ e that the size of ooding zone has signicant impact on the performance of RRP When d = 2 the EDR of RRP is nearly 98% b ut when d increases to 3 or 4, the EDR of RRP becomes better and is comparable to that of pure ooding. Ho we v er the lar ger d does not al w ays imply better EDR. As we can see, when d increases to 5, the EDR drops to 96% This observ ation can be interpreted as follo ws. When d is small, meaning a lar ge ooding zone, the collisions in the zone may occur quite often and result in relati v ely lo wer EDR. Though pure ooding suf fers from collisions as well, ooding has higher probability to recei v e a cop y of the original data pack et because it is ooded in the whole eld rather than in a zone. W ith the increase of d and thus the decrease of the ooding zone size, pack et collisions may decrease, leading to a better EDR. Ho we v er the ooding zone size should be reasonably wide to ha v e enough nodes to relay pack ets. Ho we v er our RRP outperforms directed dif fusion in most cases e v en when the ooding zone is small. Figure 5(a) showsnormalizedenergyconsumptionofourRRPwithregardto d where P E R = 0 : 0005 The results of pure ooding and directed dif fusion are sho wn for reference only W e observ e that the lar ger d is, the less ener gy each pack et consumes. This result is reasonable because the lar ger d is, the narro wer the zone is, the fe wer nodes are in v olv ed in the ooding, and hence the less the ener gy is consumed. Considering the impact of d on both EDR and ener gy consumption, it clearly indicates that the ooding zone size d is an important parameter in RRP and should be well chosen to strik e a good balance between reliability and ener gy ef cienc y Figure 5(b) showsaverageeventend-to-enddelayofourRRPwithregardto d with P E R equal to 0.0005. W e can see that the e v ent latenc y decreases with the increase of d When d is small, meaning a lar ge ooding zone, pack ets recei v ed by a sink may tra v el along unpredictable paths, leading to a longer a v erage delay In contrast, when d is lar ge, meaning a small ooding zone and fe wer nodes in v olv ed in the pack et forw arding, pack ets recei v ed by a sink may tra v el along more predictable paths limited by the small zone, which

PAGE 149

137 results in a shorter a v erage delay W e can further e xpect a small latenc y comparable to that of directed dif fusion if we carefully choose a small enough ooding zone. T o sum up, our simulation results sho w that ooding and our zone ooding techniques are less sensiti v e to the link f ailures thereby pro vide more rob ust pack et deli v ery service. Ho we v er pure ooding suf fers from inef cient ener gy usage and long end-to-end pack et latenc y By choosing appropriate w arehouse size and ooding zone size, we can adjust the ener gy consumption, the e v ent end-to-end latenc y and the routing o v erhead of RRP to competiti v e le v els while maintaining nearly perfect reliability Therefore, our RRP provides a good solution to balance the reliability and ener gy ef cienc y requirements in sensor netw orks. 5.6 Summary In this chapter we introduced the concept of supply chain into wireless sensor netw orks and propose a h ybrid data dissemination frame w ork based on the supply chain management methodology F or each sensing task, a whole sensor eld is conceptually partition into se v eral functional re gions, and v arious routing schemes are applied to dif ferent re gions in order to pro vide better performance in terms of reliability and ef cient ener gy usage. F or this purpose, we also proposed a no v el zone ooding technique which is a combination of geometric routing and ooding. On top of our scheme, ph ysically separated or interlaced multipath routing can be easily implemented without incurring an y signicant additional costs. Our h ybrid data dissemination frame w ork features lo w o v erhead, high reliability good scalability and notable e xibility The ef fecti v eness and ef cienc y of our scheme were v alidated through simulation studies. W e demonstrated that our scheme can strik e a good balance between reliability and ener gy ef cienc y with proper sizes of the w arehouse area and ooding zones.

PAGE 150

CHAPTER 6 CONCLUSIONS AND FUTURE W ORK 6.1 Conclusions In this dissertation, we study the issues of ener gy conserv ation and QoS pro visioning in resource-constrained wireless ad hoc netw orks. W e particularly identify the inherent heterogeneity in such netw orks, and classify it into tw o groups: de vice heterogeneity and en vironment heterogeneity W e propose to adopt the cross-layer design philosoph y to guide the protocol design. F or the issue of ener gy conserv ation. W e rst propose a de vice-ener gy-load a w are relaying frame w ork (DELAR) to utilize heterogeneity in terms of po wer supply to conserv e ener gy This frame w ork is a joint design of po wer control, scheduling and routing. The proposed A-MA C protocol also addresses the issue of reliable communications o v er unidirectional links. Further based on the DELAR frame w ork, a multiple-pack ets transmission scheme is proposed to impro v e the o v erall performance of DELAR. Through simulation studies, DELAR is sho wn to be ener gy-ef cient. W e also study ho w to utilize mobility to conserv e ener gy W e present a general mo v ement model and formulate the general mo v ement problem. W e particularly study the W aterhunter problem and reduce it to a path-constrained path-optimization routing problem. W e present a heuristic algorithm for this NP-complete routing problem. Through simulation, it sho ws that the resource-a w are mo v ement scheme can better utilize the P-nodes in the netw orks and lead to more ener gy sa vings. F or the issue of Quality of Service, we study ho w to use system dynamics to better support dif ferentiated services in wireless ad hoc netw orks. W e v alidate that there are unused resources, that is, bandwidth due to system dynamics. W e propose a cross-layer designed scheme called courtesy piggybacking to mak e use of such unused resources. This 138

PAGE 151

139 scheme is sho wn to be v ery ef fecti v e in alle viating the conict between throughput and f airness and the starv ation problem in a dif ferentiated services system. W e study the issues of reliability and ener gy conserv ation in wireless sensor netw orks. W e propose a supply-chain based wireless sensor netw ork model and propose a h ybrid data dissemination frame w ork. This frame w ork is scalable, reliable and ener gy-ef cient. 6.2 Future W ork F or the future research on the issue of ener gy conserv ation, there are se v eral aspects can be impro v ed and in v estig ated. Since the h ybrid scheduling has hea vy impacts on the system performance of DELAR. W e w ould lik e to seek a more dynamic transmission scheduling scheme to strik e a better balance between ener gy conserv ation and other system performance f actors. W e also w ant to in v estig ate the impact on ener gy conserv ation of node placement of P-nodes [75,132]. W e belie v e that the controlled and intelligent mo v ement and its inte gration with resource heterogeneity can bring lots of benets to mobile ad hoc netw orks. From the resource-a w are mo v ement scheme, it suggests that netw orks e xhibiting features of small-w orld tend to ha v e better o v erall performance, so we w ould lik e to rst study the Firehunter Mo v ement problem (cf. Section3.3.2) and stri v e to propose a unied solution for the general resource-a w are mo v ement problem. Further we w ould lik e to study ho w to b uild up such a small w orld in wireless ad hoc netw orks. As to the research on the issue of QoS pro visioning, we w ould lik e to apply this piggybacking scheme to more concrete applications to e v aluate its performance, and we are also interested in the inte gration of this scheme into the DELAR frame w ork. F or the wireless sensor netw orks, the supply chain model can be further impro v ed to t dif ferent application scenarios. W e w ould lik e to gi v e a formal formulation of this model and to carry out more strict analysis about the partition boundary between dif ferent

PAGE 152

140 functional re gions. W e are particularly interested in b uilding some real test beds to e v aluate this model and this frame w ork. F or wireless ad hoc netw orks as a whole, there are still lots of open problems such as security issues, and the interoperateability with other wireless and mobile netw orks. In the future, we will de v ote ourselv es to these interesting topics.

PAGE 153

REFERENCES [1] E. Biglieri and P T orino, Digital transmission in the 21st century: Conating modulation and coding, IEEE Communications Ma gazine v ol. 40, no. 5, pp. 128, May 2002. [2] S. K umar V Ragha v an, and J. Deng, Medium access control protocols for ad-hoc wireless netw orks: A surv e y Else vier Ad Hoc Networks J ournal 2005, to appear [3] V Bhar gha v an, A. Demers, S. Shenk er and L. Zhang, Maca w: A media access protocol for wireless lans, in Pr oc. of A CM SIGCOMM'94 1994. [4] C. Fullmer and J. Garcia-Luna-Ace v es, Floor acquisition multiple acess (f ama) for pack et radio netw orks, in Pr oc. of A CM SIGCOMM'95 1995. [5] P art 11: W ireless lan medium access control (mac) and ph ysical layer (ph y) specications, Standard, IEEE, IEEE 802.11WG, August 1999. [6] D. Lang, A comprehensi v e o v ervie w about selected ad hoc netw orking routing protocols, T echnische Uni v ersity Munchen, Department of Computer Science, T ech. Rep., March 2003. [Online]. A v ailable:http://wwwbib .informatik. tumuenchen.de/infberichte/2003/TUMI0311.pdf[7] C. Perkins and P Bhagw at, Highly dynamic destination-sequenced distance-v ector routing (dsdv) for mobile computers, in Pr oc. of A CM SIGCOMM'94 Sept. 1994, pp. 234. [8] Optimized link state routing protocol (olsr), IETF T ech. Rep., October 2003. [9] D. Johnson, D. Maltz, and Y .-C. Hu, The dynamic source routing protocol for mobile ad hoc netw orks, IETF Internet Draft, April 2003, draft-ietf-manet-dsr -09.txt. [10] C. Perkins, E. Belding-Ro yer and S. Das, Ad hoc on demand distance v ector (aodv) routing, RFC 3561, July 2003. [11] Z. Haas, M. Pearlman, and P Samar The zone routing protocol (zrp) for ad hoc netw orks, IETF Internet Draft, draft-ietf-manet-zone-zrp-04.txt, July 2002. [12] M. Gerla, X. Hong, L. Ma, and G. Pei, Landmark routing protocol (lanmar) for lar ge scale ad hoc netw orks, IETF Internet Draft, MANET W orking Group, draftietf-manet-lanmar -05.txt, No v ember 2002. 141

PAGE 154

142 [13] C. Jones, K. S. P Agra w al, and J.-C. Chen, A surv e y of ener gy ef cient netw ork protocols for wireless netw orks, A CM/Kluwer J ournal on W ir eless Networks (WINET) v ol. 7, no. 4, August 2001. [14] P Mohapatra, J. Li, and C. Gui, Qos in mobile ad hoc netw orks, IEEE W ir eless Communications Ma gazine v ol. 10, no. 3, pp. 44, June 2003. [15] X. Hannan, C. Chaing, and S. W inston, Quality of service models for ad hoc wireless netw orks, in The handbook of ad hoc wir eless networks Boca Raton, FL, USA: CRC Press, Inc., 2003, pp. 467. [16] G.-S. Ahn, A. Campbell, A. V eres, and L.-H. Sun, Supporting service dif ferentiation for real-time and best ef fort traf c in stateless wireless ad hoc netw orks (sw an), IEEE T r ans. on Mobile Computing v ol. 1, no. 3, pp. 192, July-September 2002. [17] D. Black, Dif ferentiated services and tunnels, RFC2983, 2000. [18] Dening cross-layer design for wirelss netw orking, in Pr oc. of IEEE ICC'03 2003, panel. [Online]. A v ailable:http://www .eas.asu.edu/ junshan/ICC03panel. html[19] S. Shakk ottai, T Rappaport, and P Karlsson, Cross-layer design for wireless netw orks, IEEE Commun. Ma g October 2003. [20] I. Ak yildiz, X. W ang, and W W ang, W ireless mesh netw orks: a surv e y Computer Networks 2005. [21] I. Chak eres and E. Belding-Ro yer Resource biased path selection in heterogeneous mobile netw orks, in UCSB T ec hnical Report 2003-18 July 2003. [22] M. Krunz, A. Muqattash, and S.-J. Lee, T ransmission po wer control in wireless ad hoc netw orks: challenges, solutions and open issues, IEEE Network v ol. 18, no. 5, pp. 8, 2004. [23] S. Singh and C. S. Ragha v endra, P amas-po wer a w are multi-access protocol with signaling for ad hoc netw orks, Computer Communication Re vie w v ol. 28, no. 3, pp. 5, 1998. [24] J. H. W Y e and D. Estrin, An ener gy-ef fcient mac protocol for wireless sensor netw orks, in Pr oc. of INFOCOM'02 2002. [25] Y .-C. Tseng, C.-S. Hsu, and T .-Y Hsieh, Po wer -sa ving protocols for ieee802.11based multi-hop ad hoc netw orks, in Pr oc. of INFOCOM'02 2002. [26] V Roodoplu and T Meng, Minimum ener gy mobile wireless netw orks, IEEE J Select. Ar eas Commun. v ol. 17, no. 8, pp. 1333, August 1999.

PAGE 155

143 [27] B. Chen, K. Jamieson, H. Balakrishnan, and R. Morris, Span: an ener gy-ef cient coordination algorithm for topology maintainence in ad hoc wireless netw orks, A CM/Kluwer J ournal on W ir eless Networks (WINET) v ol. 8, no. 5, Sept. 2002. [28] J. W u, F Dai, M. Gao, and I. Stojmoeno vic, On calculating po wer -a w are connected dominating sets for ef cient routing in ad hoc wireless netw orks, J ournal of Communications and Networks v ol. 4, no. 1, March 2002. [29] Y Xu, J. Heidemann, and D. Estrin, Geograph y-informed ener gy conserv ation for ad hoc routing, in Pr oc. of MobiCom'01 2001. [30] D. Blough, M. Leoncini, G. Resta, and P Santi, The k-neigh protocol for symmetric topology control in ad hoc netw orks, in Pr oc. of MobiHoc'03 2003. [31] L. Feene y A qos a w are po wer sa v e protocol for wireless ad hoc netw orks, in Pr oc. of INFOCOM'02 2002. [32] R. Ramanathan and R. Rosales-hain, T opology control of multihop wireless netw orks using transmit po wer adjustment, in Pr oc. of INFOCOM'00 2000. [33] J. Gomez and A. T Campbell, A case for v ariable-range transmission po wer control in wireless ad hoc netw orks, in Pr oc. of IEEE INFOCOM'04 Hongk ong, China, March 2004. [34] W .-Z. Song, Y W ang, X.-Y Li, and O. Frieder Localized algorithms for ener gy ef cient topology in wireless ad hoc netw orks, in Pr oc. of MobiHoc'04 2004. [35] A. Muqattash and M. Krunz, Po wmac: A single-channel po wer -control protocol for throughput enhancement in wireless ad hoc netw orks, IEEE J Select. Ar eas Commun. 2005, special Issue on Adv ances in Military W ireless Communications (to appear). [36] E.-S. Jung and N. V aidya, A po wer control mac protocol for ad hoc netw orks, in Pr oc. of MobiCom'02 2002. [37] S. Singh, M. W oo, and C. Ragha v endra, Po wer -a w are routing in mobile ad hoc netw orks, in Pr oc. of MobiCom'98 1998. [38] J.-H. Chang and L. T assiulas, Ener gy conserv ation routing in wireless ad-hoc netw orks, in Pr oc. of INFOCOM'00 2000. [39] C. T oh, Maximum battery life routing to support ubiquitous mobile computing in wireless ad hoc netw orks, IEEE Commun. Ma g pp. 138, June 2001. [40] S. Banerjee and A. Misra, Minimum ener gy paths for reliable communication in multi-hop wireless netw orks, in Pr oc. of MobiHoc'02 2002.

PAGE 156

144 [41] J. Gomez, A. Campbell, M. Naghshineh, and C. Bisdikian, P aro: Supporting dynamic po wer controlled routing in wireless ad hoc netw orks, A CM/Kluwer J ournal on W ir eless Networks (WINET) v ol. 9, no. 5, September 2003. [42] J.-H. Ryu and D. Cho, A ne w routing scheme concerning po wer -sa ving in mobile ad hoc netw orks, in Pr oc. of ICC'00 2000. [43] A. A vudainayag am, W Lou, and Y F ang, Dear: A de vice and ener gy a w are routing protocol for heterogeneous ad hoc netw orks, J ournal of P ar allel and Distrib uted Computing Feb 2003. [44] U. K ozat, I. K outsopoulos, and L. T assiulas, A frame w ork for cross-layer design of ener gy-ef cient communication with qos pro visioning in multi-hop wireless netw orks, in Pr oc. of IEEE INFOCOM'04 Hongk ong, China, March 2004. [45] R. Bhatia and M. K odialam, On po wer ef cient communication o v er multi-hop wireless netw orks: Joint routing, scheduling and po wer control, in Pr oc. of IEEE INFOCOM'04 Hongk ong, China, March 2004. [46] V Ka w adia and P R. K umar Principles and protocols for po wer control in ad hoc netw orks, IEEE J Select. Ar eas Commun. v ol. 23, no. 5, pp. 76, January 2005, special Issue on W ireless Ad Hoc Netw orks. [47] A. Leon-Garcia and I. W idjaja, Computer Networks: Fundmental Concepts and K e y Ar c hitectur es McGra w-Hill, 1999. [48] L. Bao and J. Garcia-Luna-Ace v es, Link-state routing in netw orks with unidirectional links, in Pr oc. of ICCCN'99 1999. [49] M. Marina and S. Das, Routing performance in the presence of unidirectional links in mulithop wireless netw orks, in Pr oc. of MobiHoc'02 Lausanne, Switzerland, June 2002. [50] R. Prakash, A routing algorithm for wireless ad hoc netw orks with unidirectional links, A CM/Kluwer J ournal on W ir eless Networks (WINET) v ol. 7, pp. 617, 2001. [51] S. Nesar gi and R. Prakash, A tunneling approach to routing with unidirectional links in mobile ad-hoc netw orks, in Pr oc. of ICCCN'00 2000. [52] V Ramasubramanian, R. Chandra, and D. Mosse, Pro viding a bidirectional abstraction for unidirectional ad hoc netw orks, in Pr oc. of INFOCOM'03 2003. [53] M.-S. Alouini, X. T ang, and A. Goldsmith, An adapti v e modulation scheme for simultaneous v oice and data transmission o v er f ading channels, IEEE J Select. Ar eas Commun. v ol. 17, no. 3, pp. 837, may 1999.

PAGE 157

145 [54] M. Pursle y and J. Shea, Nonuniform phase-shift-k e y modulation for multimedia multicast transmission in mobile wireless netw orks, IEEE J Select. Ar eas Commun. v ol. 17, no. 9, pp. 774, May 1999. [55] P V itthalade vuni and M.-S. Alouini, Exact ber computation of generalized hier archical psk constellations, IEEE T r ans. on W ir eless Comm. v ol. 51, no. 12, pp. 2030, December 2003. [56] M. Pearlman and Z. Haas, Determining the optimal conguration for the zone routing protocol, IEEE J Select. Ar eas Commun. v ol. 17, no. 8, pp. 1395, August 1999. [57] OPNET Making netw orks and applications perform, http://www .opnet.com. [58] J. Y oon, M. Liu, and B. Noble, Sound mobility models, in Pr oc. of A CM MobiCom'03 San Die go, CA, September 2003. [59] Proxim, Orinoco 11a/b/g combocard specications. [Online]. A v ailable:www proxim.com/learn/library/datasheet/11abgcombocard.pdf[60] C. Chiasserini and R. Rao, Ener gy ef cient battery management, in Pr oc. of INFOCOM'00 v ol. 2, 2000, pp. 396. [61] S. Jayashree, B. Manoj, and C. S. R. Murth y On using battery state for medium access control in ad hoc wireless netw orks, in Pr oc. of MobiCom'04 Sept. 2004. [62] M. Grossglauser and D. Tse, Mobility increase the capacity of ad-hoc wirelss netw orks, in Pr oceedings of IEEE INFOOCM'00 T el A vi v Israel, March 2000, pp. 1360. [63] S. Basagni, I. Chlamtac, and V Syrotiuk, A distance routing ef fect algorithm for mobility (dream), in Pr oc. of the F ifth Anuual A CM/IEEE International Confer ence on Mobile Computing and Networking (MobiCom'98) Dallas, TX, 1998. [64] Y .-B. K ou and N. V aidya, Location-aided routing (lar) in mobile ad hoc netw orks, Pr oc. of the F ifth Anuual A CM/IEEE International Confer ence on Mobile Computing and Networking (MobiCom'98) October 1998. [65] S. Chakraborty D. Y au, and J. Lui, On the ef fecti v eness of mo v ement prediction to reduce ener gy consumption in wireless sommunication (e xtended abstract), in Pr oc. of A CM SIGMETRICS'03 San Die go, CA, June 2003. [66] Q. Li and D. Rus, Sending messages to mobile users in disconnected ad-hoc wireless netw orks, in Pr oc. of A CM MobiCom'00 August 2000. [67] D. Glodenber g, J. Lin, A. Morse, B. Rosen, and Y Y ang, T o w ards mobility as a netw ork control primiti v e, in Pr oceedings of A CM MobiHoc'04 Roppongi, Japan, May 2004, pp. 163.

PAGE 158

146 [68] R. Shah, S. Ro y S. Jain, and W Brunette, Data mules: Modeling a three-tier architecture for sparse sensor netw orks, in Pr oc. of IEEE SNP A W orkshop 2003. [69] W Zhao, M. Ammar and E. Ae gura, A message ferry approach for data deli v ery in sparse mobile ad hoc netw orks, in Pr oc. of A CM MobiCom'04 26 Sep. 1 Oct. 2004, pp. 187. [70] D. Johnson and D. Maltz, Dynamic source routing in ad hoc wireless netw orks, in Mobile Computing Imielinski and K orth, Eds. Kluwer Academic Publishers, 1996, v ol. 353. [71] D. Ree v es and H. Salama, A distrib uted algorithm for delay-constrained unicast routing, IEEE/A CM T r ansaction on Networking v ol. 8, no. 2, pp. 230, 2000. [72] M. Gare y and D. Johnson, Computer and Intr actability: A Guide to the Theory of NP-completeness San Francisco, CA: Freeman, 1979. [73] W Liu, W Lou, and Y F ang, An ef cient qos routing algorithm for delay sensiti v e applications, Computer Networks v ol. 47, no. 1, pp. 87, 2005. [74] Draft supplement to part 11: W ireless lan medium access control (mac) and ph ysical layer (ph y) specications: Medium access control (mac) enhancements for quality of service (qos),ieee standard 802.11e/d4.3, Standard, IEEE, IEEE 802.11WG, May 2003. [75] Z. Y e, S. Krishnamurth y and S. K. T ripathi, A frame w ork for reliable routing in mobile ad hoc netw orks, in Pr oc. of INFOCOM'03 2003. [76] L. Feene y and M. Nilsson, In v estig ating the ener gy consumption of wireless netw ork interf ace in an ad hoc netw orking en vironment, in Pr oc. of IEEE INFOCOM'01 2001, pp. 1548. [77] R. Albert and A.-L. Barabsi, Statisctical mechanics of somple x netw orks, Re vie w of Modern Physics pp. 47, Jan. 2002. [78] M. Barry A. T Campbell, and A. V eres, Distrib uted control algorithm for service dif ferentiation in wireless pack et netw orks, in Pr oc. of IEEE INFOCOM'01 2001. [79] X. P allot and L. Miller Implementing message priority policies o v er an 802.11 based mobile ad hoc netw ork, in Pr oc. of IEEE MILCOM'01 2001. [80] Y Cao and V Li, Scheduling algorithms in broad-band wirelss netw orks, in Pr ocedings of the IEEE v ol. 89, no. 1, 2001, pp. 76. [81] J. W ang and K. Nahrstedt, Hop-by-hop routing algorithms for premium-class traf c in dif fserv netw orks, in Pr oc. of IEEE INFOCOM'02 2002. [82] A. Demers, S. K esha v and S. Shenk er Analysis and simulation of a f air queueing algorithm, in Pr oc. of A CM SIGCOMM'89 1989, pp. 3.

PAGE 159

147 [83] L. Zhang, V irtual clock: A ne w traf c control algorithm for pack et switching neto wrks, in Pr oc. of A CM SIGCOMM'90 1990, pp. 19. [84] S. Lu and V Bhar gha v an, F air scheduling in wireless pack et netw orks, IEEE/A CM T r ans. Networking v ol. 7, no. 4, pp. 473, 1999. [85] T Ng, I. Stoica, and H. Zhang, P ack et f air queueing algorithms for wireless netw orks with locaiton-dependent errors, in Pr oc. of IEEE INFOCOM'98 Mar 1998, pp. 1103. [86] P Bhagw at, A. Krishna, and S. T ripathi, Enhance throughput o v er wireless lans using channel state depentdent pack et scheduling, in Pr oc. of IEEE INFOCOM'96 Mar 1996, pp. 1133. [87] C. Fragouli, V Si v araman, and M. Sri v asta v a, Controlled multimedia wirelss link sharing via enhanced class-based queuing with channel-state dependent pack et scheduling, in Pr oc. of IEEE INFOCOM'98 1998, pp. 572. [88] S. Flo yd and V Jacobson, Link-sharing and resource management models for pack et netwroks, IEEE/A CM T r ans. Networking v ol. 3, pp. 365, August 1995. [89] J. Gomez and A. T Campbell, Ha v ana: Supporting application and channel dependent qos in wireless netw orks, A CM/Kluwer J ournal on W ir eless Networks (WINET) v ol. 9, no. 1, pp. 21, Jan. 2003. [90] I. K outsopoulous and L. T assiulas, Channel state-adpti v e techiniques for throughput enhancements in wireless broadband netw orks, in Pr oc. of IEEEINFOCOM'01 Apr 2001, pp. 757. [91] G. Holland, N. V aidya, and P Bahl, A rate-adpati v e mac protocol for multi-hop wireless netw orks, in Pr oc. of A CM MobiCom'01 2001. [92] P Che villat, J. Jelitto, A. Barreto, and H. T ruong, A dynamic link adaptation algorithm for ieee 802.11a wireless lans, in Pr oc. of IEEE ICC'03 2003. [93] B. Kim, Y F ang, T F W ong, and Y Kw on, Dynamic fragmentation scheme for rate-adapti v e wireless lan, in Pr oc. of IEEE PIMRC'03 Sept. 2003. [94] J. T ourrilhes, Dwell adapti v e fragmentation: Ho w to cope with short dwells required by multimedia wirelss lans, in Pr oc. of IEEE GLOBECOM'00 2000. [95] D. Qiao and S. Choi, Goodput enhancement of ieee 802.11a wireless lan via link adaptation, in Pr oc. of IEEE ICC'01 2001. [96] K. B. S. R. Kadaba and S. Nanda, Channel quality estimation and rate adaption for cellular mobile radio, IEEE J Select. Ar eas Commun. v ol. 17, no. 7, pp. 1244 1256, July 1999.

PAGE 160

148 [97] H. W ang and N. Moayeri, Finite-state mark o v channel-a useful model for radio communication channels, IEEE T r ans. V eh. T ec hnol. v ol. 44, no. 1, 1995. [98] E. Modiano, An adapti v e algorithm for optimizing the pack et size used in wireless arq protocols, A CM/Kluwer J ournal on W ir eless Networks (WINET) v ol. 5, no. 5, pp. 279, 1999. [99] I. Ada and C. Castelluccia, Dif ferentiation mechanisms for ieee 802.11, in Pr oc. of IEEE INFOCOM'01 2001. [100] A. S. T anenbaum, Computer Networks 3rd ed. Prentice Hall, 1996. [101] S.-T Sheu, T Chen, J. Chen, and F Y e, An impro v ed data ushing mac protocol for ieee 802.11 wireless ad hoc netw ork, in Pr oc. of IEEE VTC 2002 F all 2002. [102] D. Bertsekas and R. Gallager Data Networks 2nd ed. Prentice Hall, 1992. [103] K. Altink emer I. Bose, and R. P al, A v erage w aiting time of customers in an m/d/k queue with nonpreempti v e priorities, Computer s & Oper ations Resear c h v ol. 25, no. 4, pp. 317, 1998. [104] emer ging technologies that will change the w orld, T ec hnolo gy Re vie w v ol. 106, no. 1, pp. 33, Feb 2003. [105] I. Ak yildiz, W Su, Y Sankarasubramaniam, and E. Cayirci, A surv e y on sensor netw orks, IEEE Communication Ma gazine August 2002. [106] C.-Y Chong and S. P K umar Sensor netw orks: e v olution, opportunities, and challenges, Pr oceedings of the IEEE v ol. 91, no. 8, August 2003. [107] C. Intanagonwiw at, R. Go vindan, and D. Estrin, Directed dif fusion: a scalable and rob ust communication paradigm for sensor netw orks, in Pr oc. of the Sixth Anuual A CM/IEEE International Confer ence on Mobile Computing and Networking (MobiCom'00) Boston, MA, August 2000. [108] J. K ulik, W R. Heinzelman, and H. Balakrishnan, Ne gotiation-based protocols for disseminating information in wireless sensor netw orks, A CM W ir eless Networks v ol. 8, no. 2-3, pp. 169, 2002. [109] J. Heidemann, F Silv a, and D. Estrin, Matching data dissemination algorithms to application requirements, in Pr oc. of the F ir st International Confer ence on Embedded Network ed Sensor Systems (MobiSys'03) Los Angeles, CA, 2003, pp. 218. [110] W Heinzelman, A. Chandrakasan, and H. Balakrishnan, Ener gy-ef fcient rotuing protocols for wireless microsensor netw orks, in r oc. of the 33r d Hawaii Intenational Confer ence on System Sciences (HICSS'00) Ha w aii, Jan. 2000. [111] F Y e, H. Luo, J. Cheng, S. Lu, and L. Zhang, A tw o-tier data dissemination model for lar ge-scale wireless sensor netw orks, in Pr oc. of the Ninth Anuual A CM/IEEE

PAGE 161

149 International Confer ence on Mobile Computing and Networking (MobiCom'02) Atlanta, Geor gia, Spet. 2002. [112] D. Ganessan, R. Go vindan, S. Shenk er and D. Estrin, Highly-resilient,ener gyef cient multipath routing in wireless sensor netw orks, A CM MC 2 R v ol. 5, no. 4, October 2001. [113] J. H. Z. Hass and L. Li, Gossip-based ad hoc routing, in Pr oc. of IEEE INFOCOM'02 Ne w Y ork, June 2002. [114] Y Sasson, D. Ca vin, and A. Schiper Probabilistic broadcast for ooding in wireless mobile ad hoc netw orks, in Pr oc. of IEEE W ir eless Communications and Networking Confer ence (WCNC'03) Ne w Orleans, Louisiana, March 2003. [115] C. Barrett and S. E. L. Kroc, P arametric probabilistic sensor netw ork routing, in Pr oc. of the 2nd International W orkshop on W ir eless Sensor Networks and Applications (WSN A 2003) San Die go, CA, Spet. 2003. [116] S. T ilak, A. Murph y and W Heinzelman, Non-uniform information dissemination for sensor netw orks, in Pr oc. of the 11th International Confer ence on Network Pr otocols (ICNP'03) Atlanta, Geor gia, No v 2003. [117] B. Karp and H. K ung, Greedy perimeter stateless routing, in Pr oc. of the Se venth Anuual A CM/IEEE International Confer ence on Mobile Computing and Networking (MobiCom'00) Boston, MA, August 2000. [118] D. Niculescu and B. Nath, rajectory based forw arding and its applications, in Pr oc. of the T enth Anuual A CM/IEEE International Confer ence on Mobile Computing and Networking (MobiCom'03) San Die go, CA, Sept. 2003. [119] A. Mainw aring, J. Polastre, R. Sze wczyk, D. Culler and J. Anderson, W ireless sensor netw orks for habitat monitoring, in Pr oc. of the 1st International W orkshop on W ir eless Sensor Networks and Applications (WSN A 2002) Atlanta,Geogia,USA, 2002. [120] N. B. Priyantha, A. Chakraborty and H. Balakrishnan, The crick et location-support system, in Pr oc. of the Se venth Anuual A CM/IEEE International Confer ence on Mobile Computing and Networking (MobiCom'00) Boston, MA, August 2000. [121] D. Niculescu and B. Nath, Ad hoc position system (aps) using aoa, in Pr oc. of IEEE INFOCOM'03 San Fransciso, CA, April 2003. [122] P Bahl and V P admanabhan, Radar: an inb uiding rf-based user location and tracking system, in Pr oc. of IEEE INFOCOM'00 T el A vi v Israel, March 2000. [123] A. Rao, S. Ratnasamy C. papadimitriou, S. Shenk er and I. Stoica, Geograhpic routing without location information, in Pr oc. of the T enth Anuual A CM/IEEE International Confer ence on Mobile Computing and Networking (MobiCom'03) San Die go, CA, Sept. 2003.

PAGE 162

150 [124] T He, C. Huang, B. Blum, J. Stank o vic, and T Abdelzaher Range-free location schemes for lar ge scale sensor netw orks, in Pr oc. of the T enth Anuual A CM/IEEE International Confer ence on Mobile Computing and Networking (MobiCom'03) San Die go, CA, Spet. 2003. [125] L. D. Fredenhall and E. Hill, Basics of supply c hain mana g ement St. Lucie Press, 2001. [126] B. Krishnamachari, D. Estrin, and S. W ick er Impact of data aggre g ation in wireless sensor netw orks, in Pr oc. of International W orkshop on Distrib uted Event-Based Systems V ienna, Austria, July 2002. [127] W Lou, W Liu, and Y F ang, Spread: enhancing data condentiality in mobile ad hoc netw orks, in Pr oc. of IEEE INFOCOM'04 HongK ong, China, March 2004. [128] E. W eisstein, Ellipse: From mathw orlda w olfram web resource, http://mathw orld.w olfram.com/Ellipse.html. [129] T Ozsu and P V alduriez, Principles of distrib uted database mana g ement systems 2nd ed. Prentice Hall, 1999. [130] L. Bajaj, M. T akai, R. Ahuja, K. T ang, R. Bagrodia, and M. Gerla, Glomosim: A scalable netw ork simulation en vironment, UCLA Computer Science Department, T ech. Rep. 990027, May 1999. [131] Crossbo w Crossbo w technology inertial & gyro systems, wireless sensor netw orks, smart dust, adv anced sensors, http://www .xbo w .com. [132] M. Y arvis, N. K ushalnag ar H. Singh, A. Rang arajan, Y Liu, and S. Singh, Exploiting heterogeneity in sensor netw orks, in Pr oc. of IEEE INFOCOM'05 2005.

PAGE 163

BIOGRAPHICAL SKETCH W ei Liu recei v ed the B.E. and M.E. de grees in electrical engineeringfromHuazhong Uni v ersity of Science and T echnology W uhan, China, in 1998 and 2001, respecti v ely He is currently pursuing the PhD de gree in the Department of Electrical and Computer Engineering,Uni v ersity of Florida, Gainesville, where he is a research assistant in the W ireless Netw orks Laboratory (WINET). His research interest includes QoS, Secure, and Po wer Ef cient Routing, and MA C protocols in Mobile Ad Hoc Netw orks and W ireless Sensor Netw orks. 151

PAGE 164

I certify that I ha v e read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality as a disser tation for the de gree of Doctor of Philosoph y Y uguang F ang, Chair Associate Professor of Electrical and Computer Engineering I certify that I ha v e read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality as a disser tation for the de gree of Doctor of Philosoph y Shig ang Chen Assistant Professor of Computer and Information Science and Engineering I certify that I ha v e read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality as a disser tation for the de gree of Doctor of Philosoph y Sartaj Sahni Professor of Conputer and Information Science Engineering I certify that I ha v e read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality as a disser tation for the de gree of Doctor of Philosoph y John M. Shea Assistant Professor of Electrical and Computer Engineering I certify that I ha v e read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality as a disser tation for the de gree of Doctor of Philosoph y Dapeng W u Assistant Professor of Electrical and Computer Engineering

PAGE 165

This dissertation w as submitted to the Graduate F aculty of the Colle ge of Engineering and to the Graduate School and w as accepted as partial fulllment of the requirements for the de gree of Doctor of Philosoph y. August 2005 Pramod P Khar gonekar Dean, Colle ge of Engineering K enneth Gerhardt Interim Dean, Graduate School

PAGE 166

CR OSS-LA YER DESIGN OF RESOURCE A W ARE PR O T OCOLS FOR HETER OGENEOUS WIRELESS AD HOC NETW ORKS W ei Liu (352) 392–2746 Department of Electrical and Computer Engineering Chair: Y uguang F ang De gree: Doctor of Philosoph y Graduation Date: August 2005 W ireless ad hoc netw orks are v ery attracti v e in military and ci vil applications for which x ed infrastructure are una v ailable or unreliable. In this dissertation we in v estig ate the protocol design in resource-constrained wireless ad hoc netw orks, for the purpose of supporting communications with high ener gy-ef cienc y and desired quality-of-service (QoS). W e propose se v eral schemes that can utilize inherent heterogeneity in the netw orks to better use the precious resources. The nds in this dissertation are no v el. The research on ener gy conserv ation is the rst w ork that comprehensi v ely addresses this challenging issue. The research on QoS can be applied to man y other wireless and mobile netw orks and systems. The research on wireless sensor netw orks pro vides a ne w w ay to ef ciently manage wireless sensor netw orks. All of the proposed schemes are v ery ef fecti v e and easy to implement, and can be readily applied to real applications.


Permanent Link: http://ufdc.ufl.edu/UFE0011611/00001

Material Information

Title: Cross-Layer Design of Resource Aware Protocols for Heterogeneous Wireless Ad Hoc Networks
Physical Description: Mixed Material
Copyright Date: 2008

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0011611:00001

Permanent Link: http://ufdc.ufl.edu/UFE0011611/00001

Material Information

Title: Cross-Layer Design of Resource Aware Protocols for Heterogeneous Wireless Ad Hoc Networks
Physical Description: Mixed Material
Copyright Date: 2008

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0011611:00001


This item has the following downloads:


Full Text











CROSS-LAYER DESIGN OF RESOURCE AWARE PROTOCOLS FOR HETEROGENEOUS
WIRELESS AD HOC NETWORKS
















By
Wei Liu
















A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY


UNIVERSITY OF FLORIDA


2005
































Copyright 2005

by
Wei Liu

















To my wife, Shu, and my family members in China.















ACKNOWLEDGMENTS

First and foremost, I would like to express my sincere gratitude to my advisor, Prof.

Yuguang Fang, who leaded me to this exciting research area, for his invaluable advice,

encouragement and motivation during these years of my Ph.D. studies. I also thank him for

his philosophical advice on both my academic and nonacademic life. As a mentor, he has

helped me become more mature, scholastically and personally.

I also want to thank my committee members, Prof. Shigang Chen, Prof. Sartaj K.

Sahni, Prof. John. M. Shea, and Prof. Dapeng Wu, for their attention and advices.

I would like to extend my thanks to all my colleagues in Wireless Networks Laboratory

(WINET) at University of Florida for providing me such a warm family-like environment.

The constructive discussions with them either individually or together deserve special ac-

knowledgments.

Finally, on a more personal level, I am particularly indebted to my wife Shu, for her

enduring so many late nights, and for countless sacrifices made to me. Her love, support

and encouragement have made it possible for me to carry out this work. I am also very

grateful to all my family members in China for their endless encouragement and continued

support in one way or the other. This work is dedicated to all of them.

This material is based upon work supported in part by the U.S. Office of Naval Re-

search and National Science Foundation. Any opinions, findings, and conclusions or rec-

ommendations expressed in this material are those of the authors and do not necessarily

reflect the views of the U.S. Office of Naval Research and National Science Foundation.
















TABLE OF CONTENTS


page


ACKNOWLEDGMENTS.


LIST OF TABLES ............. ............ viii


LIST OF FIGURES.


ABSTRACT. .......... .........


CHAPTERS


1 INTRODUCTION


Wireless Ad Hoc Networks.
Current Research
1.2.1 Reliable Communications
1.2.2 Energy Conservation
1.2.3 Quality of Service ...
1.2.4 Cross-Layer Design ..
1.2.5 Heterogeneity in Wireless
Our Research
Outline. ... ....


Ad Hoc Networks.


2 A DEVICE-ENERGY-LOAD AWARE RELAYING FRAMEWORK
IN HETEROGENEOUS WIRELESS AD HOC NETWORKS.


2.1 Introduction.
2.2 Related Work.
2.3 Overview of DELAR.
2.3.1 Problem Statement
2.3.2 Our Solution: DELAR. ......
2.4 Design of DELAR. .. .......
2.4.1 P-nodes' Neighbor-Selection Criteria ...
2.4.2 Routing Component of DELAR ....
2.4.3 Hybrid Transmission Scheduling ....
2.4.4 Asymmetric Media Access Control Protocol
2.5 Multiple-Packets Transmission in DELAR ...
2.5.1 Multiple-packets Transmission .. ..
2.5.2 Hierarchical Modulation .. ....
2.6 Discussion.
2.6.1 The Existence of Backward Paths ....


(A-MAC)











2.6.2 DELAR and ZRP . .... .. 39
2.6.3 The Choice of m and n .... . .. .. 39
2.6.4 Benefits of the Time Division Scheduling .. .. .. .. 40
2.7 Performance Evaluation . ... .. .. 40
2.7.1 Simulation Setup . ... .. .. 40
2.7.2 Impact of The Number of P-nodes .. .. .. .. 42
2.7.3 Impact of Node Mobility .... ... .. 44
2.7.4 Impact of Traffic Load ..... .. . .47
2.8 Summary . .... ..... .. .47

3 RESOURCE AWARE MOVEMENT IN HETEROGENEOUS
MOBILE AD HOC NETWORKS ..... .... .. 49

3.1 Introduction ......... .. ... .. 49
3.2 Related Work ........ .. .. 50
3.3 Problem Formulation . ... ... .. 51
3.3.1 General Mobility Model (GMM) ... ... .. .. 52
3.3.2 Resource Aware Movement .... ... . .. 54
3.4 Waterhunter Movement . . . .. .. 56
3.4.1 Simplified Waterhunter Movement Problem .. . 56
3.4.2 RAM-DCLC Algorithm .... . .. .. 61
3.4.3 A Routing Delay Differentiation Mechanism .. . 65
3.4.4 Incorporate RAM into DELAR ... .. .. .. .. 66
3.5 Performance Evaluation . .... .. .. 66
3.5.1 Simulation Setup . .... .. 66
3.5.2 Simulation Results ...... .. .. 68
3.6 Summary ............. ... ...... 7

4 SUPPORT DIFFERENTIATED SERVICES IN MOBILE AD HOC NETWORKS 72

4.1 Introduction ......... .. ... .. 72
4.2 Related Work ........ .... .. 74
4.3 Motivation .......... ...... ...... 76
4.4 Packet-length-based Channel Model .... ..... .. 79
4.5 Courtesy Piggybacking . ..... .. .. 82
4.5.1 System Assumptions ...... .... .. 82
4.5.2 The Courtesy Piggybacking Scheme ... .. .. .. .. 83
4.5.3 Discussion ........ .. .. 89
4.6 Performance Analysis . ...... ... .. 94
4.7 Performance Evaluation . .... .. .. 97
4.7.1 Simulation Setup . ..... .. .. 97
4.7.2 Impact of Channel Characteristics .. .. .. .. .. 99
4.7.3 Impact of Traffic Load .............10
4.7.4 Impact of Node Mobility .............11
4.7.5 Impact of Piggybacking Rules .............13
4.8 Summary .............14












5 ROBUST AND ENERGY-EFFICIENT DATA DIS SEMINATION
IN WIRELESS SENSOR NETWORKS ... ... .. 106

5.1 Introduction . .. ..... .06
5.2 Related Work .. .. .. . ... . .08
5.3 Modelling Sensor Networks as Supply Chains .. . . 111
5.3.1 Introduction to Supply Chains ... .. .. .. 111
5.3.2 How Could Supply Chains Help Us .. .. .. .. .. 113
5.4 A Hybrid Data Dissemination Framework
for Wireless Sensor Networks .... .... .. 115
5.4.1 System Model . .... .. .. .115
5.4.2 Manufacture Area . ... .. .. .117
5.4.3 Transportation Area .... .. .. .119
5.4.4 Warehouse Area and Service Area ... .. .. .. .. 123
5.4.5 Discussion . .... .. .. .. .. 126
5.5 Performance Evaluation . . ... .. .128
5.5.1 Methodology and Metrics ... ... .. 128
5.5.2 Simulation Results ...... .... .. .131
5.6 Summary ......... .. .. .. .13

6 CONCLUSIONS AND FUTURE WORK ... ... .. 138

6.1 Conclusions . .. ..... .. .38
6.2 FutureWork ............. ..........139

REFERENCES ......... . .. .. .. 141

BIOGRAPHICAL SKETCH . ..... ... . 151

















LIST OF TABLES
Table pg

3-1 RAM: resource aware movement .... .. .. 61

3-2 RAM-DCLC: a DCLC routing algorithm for the Waterhunter Movement 62

3-3 Energy consumption parameters. ..... .. . 67

4-1 Channel model statistics ..... ... .. 97

5-1 Analogue between supply chains and wireless sensor networks .. .. .. 114

5-2 Simulation configuration . ...... ... .. .. 130

















LIST OF FIGURES
Figure page

2-1 The structure of a Super Frame. . .. .. .. 18

2-2 An example of the neighbor determination process (m=4, n=2, T=3.).. .. 22

2-3 The topology in homogeneous and heterogeneous cases.. .. .. .. .. 23

2-4 A unidirectional link between A and B. .. .. .. 27

2-5 The A-MAC operation procedure. ..... .... .. 28

2-6 The multiple-packets transmission enhancement to DELAR. .. .. .. 30

2-7 General hierarchical 2/4-PSK constellation. ... .. . .. 32

2-8 Implement multiple-packets transmission with hierarchical modulation .. 33

2-9 The existence of backward path. ...... .... .. 38

2-10 The average number of nodes existing in the shaded area. .. .. .. .. 38

2-11 Simulation results with different number of P-nodes. .. .. .. .. .. .. 43

2-12 Simulation results with different maximum node speed. .. . 45

2-13 Simulation results with different traffic load. ... .. . .. 46

3-1 Multiple paths between two consecutive stops. ... .. . .. 55

3-2 An exemplary complete graph. ...... .... .. 58

3-3 An exemplary resource-aware movement. ... . . .. 59

3-4 Resource-aware movement (RAM) vs. random waypoint movement. A B-
node should consecutively visit S1-S2- 3- 4- 5 and the solid lines are
labelled by the sequences they were passed through. .. . 64

3-5 Simulation results. ......... .. .. 69

4-1 The Whittier Tunnel scenario. . .. .. .. 77

4-2 The Gilbert-E11iott channel model. ..... .. .. 80

4-3 The optimal packet length (PL) vs. SNR (y), h=128... .. .. .. .. 81

4-4 Packet-Length-Based finite-state Markov channel model. .. .. .. .. 81











4-5 A fragmentation example. . ..... .. .. 83

4-6 The total overhead with PKmax and FT. ... .... .. 84

4-7 Illustration of the courtesy piggybacking scheme. .. .. .. .. 87

4-8 Three piggybacking cases. . ..... .. .. 90

4-9 An alternative piggybacking method. .... ... .. 94

4-10 The queue model for piggybacking. .... .. .. 96

4-11 Average waiting time. ....... .. ... .. 97

4-12 Simulation results with different channel settings. .. .. . 100

4-13 Simulation results with different packet arrival rates. .. .. .. .. .. .. 101

4-14 Simulation results with different pause time. .. .. .. .. 102

5-1 An exemplary supply chain. . .... .. .. .. 111

5-2 Supply chain strategies. ....... ... .. .. 112

5-3 A system architecture for habitat monitoring .. .. .. .. .. 115

5-4 The forwarding-decision-making process of nodes in the transportation area. 120

5-5 An exemplary packet structure. ...... .... .. .. 120

5-6 The routing process in the warehouse area. .. .. .. .. .. 124

5-7 The simulated sensor field. . ..... .. .. .. 129

5-8 Event delivery ratio vs. packet error rate. ... .. .. .. .. 131

5-9 Normalized energy consumption vs. packet error rate. .. .. .. .. 131

5-10 Average event end-to-end delay vs. packet error rate. .. .. .. .. 132

5-11 Average routing overhead vs. packet error rate. .. .. .. .. .. 132

5-12 Event delivery ratio vs. packet error rate. ... .. .. .. .. 134

5-13 Energy consumption and event end-to-end delay of RRP with different d 135















Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy

CROSS-LAYER DESIGN OF RESOURCE AWARE PROTOCOLS FOR
HETEROGENEOUS WIRELESS AD HOC NETWORKS

By

Wei Liu

August 2005

Chair: Yuguang Fang
Major Department: Electrical and Computer Engineering

Their features of rapid establishment and self-organization have rendered wireless ad

hoc networks to be identified as an indispensable component to support ubiquitous com-

munications. However, the viable deployment of such networks faces many challenges

resulting from some innate unfavorable features such as the constrained resources, for ex-

ample, finite energy supply and limited bandwidth. In this dissertation we investigate the

cross-layer design of resource-aware protocols that conserve energy and support quality-

of-service (QoS) in resource-constrained wireless ad hoc networks.

We first study the issue of energy conservation in heterogenous wireless ad hoc net-

works, where most nodes are powered by batteries with small capacity, while some others

are more powerful, for example, powered by batteries with large capacity. We propose

a cross-layer designed device-energy-load aware relaying framework, called DEIAR, to

capitalize such powerful nodes to conserve energy. Different from the previous work that

addresses the energy conservation issue at one layer, DELAR is a joint design of schedul-

ing, power control and routing that can conserve energy thus prolong the network lifetime.

We further propose a resource-aware movement mechanism to utilize such powerful nodes










for data relaying. Such a resource-aware movement mechanism provides a new way to

conserve energy in wireless ad hoc networks.

We then study the issue of how to support QoS in wireless ad hoc networks. We

propose a scheme called Courtesy Piggybacking to utilize system dynamics including time-

varying channel condition and stochastic traffic characteristics, to support differentiated

services in such networks. The basic idea of Courtesy Piggybacking is to let high priority

traffic help the low priority traffic by sharing the unused residual bandwidth. As a result,

the scheme improves the overall performance for different prioritized services.

Finally, we propose an energy efficient, reliable and scalable data dissemination scheme

for wireless sensor networks. Based on supply chain management methodology, for each

sensing task, the sensor field is conceptually partitioned into several functional areas, and

then different routing schemes are applied in different areas. This scheme addresses energy-

efficiency, reliability, and scalability at the same time.















CHAPTER 1
INTRODUCTION

1.1 Wireless Ad Hoc Networks

A wireless ad hoc network is a collection of autonomous communication devices,

sometimes called nodes, which can communicate with each other by forming a multi-hop

radio network and maintaining connectivity in a decentralized manner. There are two major

types of wireless ad hoc networks that are of particular interest to academia, industry and

government: mobile ad hoc networks (MANETs) and wireless sensor networks (WSNs).

Both types of wireless ad hoc networks share several typical features in common. Wireless

ad hoc networks can be deployed quickly and on-demand. They are self-organized and

independent of infrastructure. Moreover, each node in a wireless ad hoc network functions

as both a host and a router, helping relay packets for other nodes. These favorable features

make wireless ad hoc networks very attractive in military and civil applications for which

fixed infrastructures are unavailable or unreliable, yet fast network establishment and con-

stant reconfiguration are required. As an instance of mobile ad hoc networks, after an

earthquake, a rescue team sets up a temporary wireless networks and shares information to

coordinate the rescuing work. As an example of wireless sensor networks, a large number

of small sensor nodes are deployed in a field to monitor the habitat behavior of animals.

However, there are some unfavorable features impeding wireless ad hoc networks

from widespread deployment. Due to the time-varying and error-prone wireless channel,

nodes have to contend with the adverse effects of radio communications, such as noise, fad-

ing and interference. Since nodes may depart, join, or move in the field, or even perish due

to energy depletion, the network topology is in general dynamic and may change rapidly

and unpredictable over time. In such hostile, dynamic environment, reliable communica-

tions are of great importance. Additionally, the wireless channel has limited bandwidth.










Therefore, it is a challenging task to support heterogeneous traffic with various quality-

of-service requirements. Moreover, most nodes, especially those small sensors deployed

in wireless sensor networks, are battery-powered thus with limited lifetime. These nodes

would become useless once depleting the batteries. Some adverse consequences of such

node diminution include degradation of network performance and unfavorable network

partition. Thus, energy conservation, that is, how to expend the energy resources more

frugally and more evenly so as to prolong the network lifetime, becomes a crucial issue

for MANETs. The broadcast nature of wireless communications also makes it easy for

enemies to snoop the ongoing transmissions, thus communications in wireless ad hoc net-

works have security hazard. Depending on the applications running in such networks, the

networks range from small-scale networks with a small number of nodes to large-scale net-

works with thousands even millions of nodes, for example, large-scale wireless sensor net-

works. The above unfavorable factors, especially variable wireless links, topology changes,

limited bandwidth, and constrained energy supply, make the protocol design a tremendous

challenge. Particular, when heterogenous applications are finding their niches in wireless

ad hoc networks, specially designed protocols that can overcome the above adverse factors

are desired. Such protocols, scalable themselves, should provide reliable, QoS-supporting,

energy-e~fficient, and secure communications for those eye-catching applications. In the fol-

lowing we give a nutshell of the current research for the issues of reliable communications,

energy conservation and QoS provisioning.

1.2 Current Research

1.2.1 Reliable Communications

The primary goal of protocol design for wireless ad hoc networks is to support reli-

able communications which is the basic requirement for any applications, and all the other

solutions for the issues of energy conservation and quality of service should be based on re-

liable communications. In such hostile and dynamic settings, packet losses may be caused










for various reasons including transmission errors, collisions, or route changes due to mo-

bility. Thus, there are many works addressing reliable communications at different layers.

For example, at the physical layer, modulation and error coding schemes are proposed to

combat the error-prone feature of wireless communications [1]. In order to reduce the in-

terference and transmission collisions, at the MAC layer there are many MAC protocols

that are devised to to efficiently control the channel access [2]. These protocols can be

classified into three categories: contention-based (for example, MACAW [3], FAMA [4]

and IEEE 802.11 [5]), contention-free (for example, TDMA, FDMA and CDMA) and hy-

brid. The most famous one is IEEE 802.11, and it uses CSMA/CA and four way handshake

(RTS/CTS/DATA/ACK) to coordinate the channel access and provide reliable communica-

tions [5]. At the network layer many routing protocols have been proposed to deal with

the dynamic topology and support reliable communications [6]. They can be proactive

in that they determine routes independent of traffic patterns, for example, DSDV [7] and

OLSR [8]; they can be reactive in that they maintain routes on demand, such as DSR [9]

and AODV [10]; they even can be hybrid of the above two, for example, ZRP [l l] and

LANMAR [12]. Though these schemes function at different layers, they all work for the

same goal: supporting reliable communications in wireless ad hoc networks.

1.2.2 Energy Conservation

How to lengthen the lifetime of communication devices is a crucial issue for the wire-

less ad hoc networks and many power conservation techniques are applied when hardware

and software including the protocol stack are specifically designed. Many efforts have been

made at the physical layer due to the factor that the system hardware is the primary con-

sumption object in a mobile device. Two different perspectives are commonly adopted to

approach the problem. Though no breakthrough has been experienced in the past 30 years,

the direct way is to increase the battery capacity, while keeping the weight of the battery

tolerative. Another way is, hard yet attainable, to apply low-power techniques to decrease

the energy dissipation when wireless terminals are designed. Considerable energy savings










are resulted from the hardware design; however, it is pertinent to explore other venues to

further improve the energy efficiency, that is, to design the protocol stack with energy effi-

ciency in mind [13]. For example, at the data-link layer transmission power control can be

used to reduce the interference. Power-saving mode also can be used to conserve energy

by putting the network interface into the sleep mode when no communication is needed.

At the network layer, power aware routing is used to find energy-efficient paths and route

packets over the energy-efficient paths to save energy. At the application layer, energy-

efficient OS/middleware and applications (for example, power-aware video processing) are

proposed to save energy. Because the key to energy conservation in wireless ad hoc net-

works lies within the higher levels of the wireless protocol stack [13], it would be beneficial

if these multiple layers can jointly function to conserve energy.

1.2.3 Quality of Service

It is hard to agree on a common definition of QoS, but a QoS enabled network shall

ensure that its applications and/or their users have their QoS parameters fulfilled, while

at the same time ensuring an efficient resource usage, for example the bandwidth, and

also ensure that the most important traffic still has its QoS parameters fulfilled during net-

work overload. The most important QoS parameters may include: throughput, availability,

delay, jitter and packet loss. The network can serve the applications with "hard-QoS",

"soft-QoS" or "adaptive QoS" [2, 14]. Several prominent service models or overall archi-

tectural frameworks, within which certain types of services can be provided in the network,

are suggested for ad hoc networks, such as flexible quality of service model for MANETs

(FQMM) [15] and stateless model for wireless ad-hoc networks (SWAN) [16]. Since the

resource is limited and different traffic has very different quality of service requirements,

usually Differentiated Services (DiffServ) architecture [17] is used in which traffic is clas-

sified into different classes or priority levels, and the wireless networks serve different

classes of traffic with different processing rules or policies. There are several mechanisms

proposed for supporting QoS [14], such as QoS routing, scheduling, admission control,










resource reservation, and signaling techniques. These mechanisms can coexist in a net-

work. For example, QoS routing is used to find a path satisfying certain QoS requirements,

then signalling techniques and resource reservation are used to reserve resources along the

found path. We believe that in order to efficiently utilize the resources and support QoS,

various mechanisms implemented at different layers should work together.

1.2.4 Cross-Layer Design

As a matter of fact, the issues of energy conservation and QoS provisioning are direct

results of constrained resources: energy supply and bandwidth. If nodes and networks are

endowed with plenty of bandwidth and energy supply, for example, unbounded frequency

band and heavy duty power supply with light wight, these two issues may never come to

the fore. Since the tough reality of constrained resources is there and the chance to envision

breakthrough for these two resources is rather slim, the natural response to these two issues

is to check: Do all the current protocols proposed for wireless ad hoc networks make full

use of such precious resources? The answer to the question is negative. There do have lots

of resources thar are wasted without any yield, owing to low efficiency or their overlook of

these protocols. For example, in networks with contention-based medium access control

mechanisms, lots of energy and bandwidth are wasted due to collisions.

One of the reasons for resource underutilization is that vast majority of the protocols

are based on traditional layered-design principle, which lacks the interaction and informa-

tion sharing among multiple layers. The lack of interaction causes low efficiency of those

protocols. On one hand, many resources at one layer are wasted due to undesirable de-

cisions made at another layer. For example, poor routing decisions cause congestion and

collisions at the physical layer thus waste precious resources, for example, bandwidth and

energy. On the other hand, many available resources are neglected by those protocols. For

example, time-varying channel condition leads to time-varying channel capacity, however,

due to the lack of interaction between the MAC layer and the physical layer, many MAC

protocols cannot recognize such a resource but neglect it. Hence, in recent years a new










cross-laver design philosophy emerges [18] [19]. In contrast to the layered-design phi-

losophy, cross-layer design philosophy stresses on the information sharing and interaction

among multiple layers. Cross-layer design can be performed in two ways [20]. In the

first way, parameters or information in other protocol layers are considered to improve the

performance of a protocol layer. The first way of practice does not totally abandon the

transparency between protocol layers. In contrast, the second way blurs this transparency

by merging several layers into one component to achieve much better performance.

The cross-layer design philosophy attempts to make use of the inter-layer coupling to

use the resources more efficiently. With this philosophy, the decision for resource utiliza-

tion can be made more advisably based on the information from multiple layers. Though

the cross-layer design may make the design issue more complicated than the layered design,

the appealing benefits make cross-layer design a strong candidate as the design principle in

resource-constrained wireless ad hoc networks.

1.2.5 Heterogeneity in Wireless Ad Hoc Networks

Most of the previous research assumes that networks are homogeneous or nodes are

in fully symmetrical environment. For example, all nodes have identical capabilities and

responsibilities. In reality, however, the heterogeneity of mobile devices seems to be inher-

ent and has been commonly observed in MANETs [21]. Strictly speaking, heterogeneity in

a network usually means the device heterogeneity regrading the physical hardware or soft-

ware nodes use. Nodes in the same network may differ in their communication, processing

capabilities, which can be reflected by CPU speed, memory size, battery life, transmission

range and radio, network interface, security level, operating system or protocol stack, and

so on. Nodes in the same network may differ in their responsibilities as well. They may

undertake different roles, for example, routers, servers, cluster heads, and authentication

centers, thus operate with different policies. Owing to such differences in capabilities and

responsibilities, along with the various applications running on each individual node, nodes

may also have different traffic characteristics and mobility characteristics. For example,










nodes may have different bit rate, timeliness constraints, reliability requirements, speed,

direction, and predictable/unpredictable movement patterns for each individual applica-

tion. Broadly speaking, heterogeneity in a network also means that nodes may experience

heterogeneous environment characterized by time-varying channel and network conditions.

Such environment heterogeneity may have different properties in different settings. For ex-

ample, different nodes at different locations and at different time may perceive different

channel conditions.

The heterogeneity, either device heterogeneity or environment heterogeneity, coupling

with the aforementioned unfavorable factors, makes the protocol design much more com-

plicated. However, such heterogeneity also introduces opportunities to design more ef-

ficient protocols. In fact, some types of heterogeneity are resource-related when they

themselves are resources or can be mapped into certain type(s) of resources. Moreover,

device heterogeneity and environment heterogeneity can jointly affect the usage of con-

strained resources. For example, nodes with less available battery reserve would refrain

from transmissions in poor channel condition. Thus, to recognize the inherent hetero-

geneity and further to efficiently utilize the resource-related heterogeneity to overcome the

impact stemming from those unfavorable features, for example, limited energy supply and

constrained bandwidth, is a rather challenging task. In this dissertation, we make our ef-

forts along this line to address the issues of energy conservation and QoS provisioning in

resource-constrained wireless ad hoc networks.

1.3 Our Research

As mentioned above, there is a big volume of work studying how to support reliable

communications, from the physical layer to the data link layer and to the routing layer.

Nevertheless, the research on energy conservation and QoS provisioning is still thin on

the ground. Either of the two issues is quite interesting, however, on top of the previous

research on reliability, it is pretty hard, if not impossible, to design a one-size-fit-all pro-

tocol that can simultaneously resolve these two issues. Thus, in this dissertation, based on










the previous work for reliable communications, we attempt to study these two issues and

resolve them separately, but bear in mind the possibility of future integration of these two

1SSUeS.

The cross-layer design philosophy serves as our design principle for its potential in

better utilizing the available resources and providing desired energy efficiency and QoS

provisioning to applications. More important, the cross-layer design philosophy will help

us recognize the inherent heterogeneity in wireless ad hoc networks, and further help us

utilize it to address the issues we are interested in. More specifically, we utilize the hetero-

geneity in terms of power supply or battery life to address the issues of energy conservation.

Such heterogeneity is used to design more efficient routing, power control and transmission

scheduling. We also utilize the time-varying channel condition and stochastic traffic char-

acteristics to address issues associated with QoS provisioning in wireless ad hoc networks.

For wireless sensor networks, we particularly recognize the different roles undertaken by

each individual sensor node, and utilize such heterogeneity to design more efficient com-

munication protocols.

1.4 Outline

This dissertation is organized as follows.

Chapter 2 concerns the issue of energy conservation in heterogeneous wireless ad

hoc networks in terms of power supply. In such networks, most nodes called B-nodes are

battery-powered, and some nodes called P-nodes have relatively unlimited energy supply

such as solar cell or dynamos. We propose a device-energy-load aware relaying framework,

called DELAR, to capitalize such P-nodes to conserve energy. The basic idea of DELAR

is to divide time into Super Frames, each of which is composed by several periods, and

to enable P-nodes to transmit with different power in different periods. In this framework,

device heterogeneity, nodal residual energy, and local load status are mapped into a routing

cost metric and incorporated into routing protocols to find energy-efficient paths. We also










propose a hybrid medium access control mechanism to schedule the transmission activi-

ties of P-node and B-nodes. Moreover, we introduce the concept of "mini-routing" into

the MAC layer and propose an Asymmetric MAC (A-MAC) to support reliable communi-

cations over unidirectional links resulted from the asymmetric power capabilities between

P-node and B-nodes. Further, we propose a multiple-packets transmission scheme as an

enhancement to improve the overall performance of DELAR.

Chapter 3 discusses a resource-aware movement scheme that can be incorporated

into DELAR. In this scheme, a node is enabled to determine its movement characteris-

tic (speed, direction, and so on) based on the network environment and its own residual

resources. Specifically, such kinds of movement can be mapped into a path-constrained

path-optimization problem. A heuristic algorithm is proposed to solve this NP-complete

problem. With this resource-aware movement scheme in place, B-nodes have more oppor-

tunities to utilize P-nodes and thus conserve energy and prolong the network lifetime.

Chapter 4 focuses on the issue of QoS provisioning. We consider heterogeneous mo-

bile as hoc networks on how to utilize system dynamics, a combination of environment and

device heterogeneity including time-varying channel condition and stochastic traffic char-

acteristics. A scheme called Courtesy Piggybacking is proposed to support differentiated

services in mobile ad hoc networks. The basic idea is to let the high priority traffic help

the low priority traffic by sharing the unused residual bandwidth with courtesy. Courtesy

Piggybacking is able to utilize such system dynamics to improve the overall system per-

formance for different prioritized services: improved packet delivery ratio and shortened

end-to-end delay. More important, the piggybacking scheme can be readily incorporated

into the DELAR framework.

In Charter 5, we particularly address the issues of energy efficiency and reliability in

wireless sensor networks, a special type of wireless ad hoc networks. We introduce the

concept of supply chain into wireless sensor networks and model wireless sensor networks

as supply chains. Based on supply chain management strategies, we propose an energy










efficient, reliable and scalable data dissemination scheme. Basically, for each sensing task,

the sensor field is conceptually partitioned into several functional areas, and different rout-

ing schemes are applied in different areas. Such partition strategy, hybrid strategy, and

cooperation strategy help us address the issues of energy conservation and reliable commu-

nications at the same time.

Chapter 6 concludes this dissertation and outlines the future research.















CHAPTER 2
A DEVICE-ENERGY-LOAD AWARE RELAYING FRAMEWORK
IN HETEROGENEOUS WIRELESS AD HOC NETWORKS

2.1 Introduction

In this chapter, we focus on heterogeneous wireless ad hoc networks, where most

nodes, denoted as B-nodes, are equipped with limited power sources like batteries, while

some other nodes, denoted as P-nodes, have relatively unlimited power supplies, for ex-

ample, power scavenging units such as solar cells, or dynamos when they are installed in

mobile vehicles, and so on. Our goal is to develop more energy conscious protocols by tak-

ing advantage of the heterogeneity of mobile devices, that is, being generous in using the

P-nodes while conservative in using the B-nodes. More specifically, the contributions are

mainly fourfold. First, following the cross-layer protocol design philosophy, we propose a

Device-Energy-Load Aware Relaying framework, named DELAR, to achieve energy con-

servation by utilizing the inherent heterogeneity of nodal power capabilities. Second, we

design a hybrid transmission scheduling scheme, which is a combination of reservation-

based and contention-based medium access control schemes, to coordinate the transmis-

sion activities among P-nodes and B-nodes. Such scheduling can make full use of powerful

nodes while reducing the interference and collisions. Third, we develop the "mini-routing"

and Asymmetric MAC (A-IMAC) protocols to support the MAC layer acknowledgements

over unidirectional links resulting from the asymmetric transmission power levels at P-

nodes and B-nodes. To the best of our knowledge, this is the first effort to address this

issue at the MAC layer. Last, we present a multiple-packets transmission technique to fur-

ther improve the delay performance. Detailed simulation studies are carried out to justify

the effectiveness and efficiency of the proposed framework.










The proposed DELAR can serve as a general framework where various energy con-

servation techniques such as power saving modes, transmission power control and power-

aware routing can be integrated to jointly achieve better energy conservation. In addition,

it also offers a platform to study other challenging issues, for example, quality of service

(QoS) provisioning and security support. For instance, P-nodes can act as distributed ad-

mission controllers to coordinate the access to limited network resources such as available

bandwidth. As another example, in security-sensitive MANET applications, P-nodes can

help B-nodes perform resource-hungry public-key operations. As far as we know, no simi-

lar framework has appeared elsewhere in the literature.

In the rest of this chapter, we start with the review of related work. We then introduce

the system model and the overall framework of DELAR in Section 2.3. In Section 2.4,

we elaborate the network layer components of DELAR and a hybrid transmission schedul-

ing scheme, and present a novel Asymmetric MAC protocol called A-IMAC followed by

a multiple-packet transmission scheme to further improve the performance of DELAR. In

Section 2.7, we evaluate the performance of DELAR through simulations. Finally, we

summarize this chapter in Section 2.8.

2.2 Related Work

Recent years have seen a growing body of research concerned with energy conserva-

tion in wireless ad hoc networks, among which many efforts have been made at the physical

layer to improve the hardware design of mobile devices [13]. Though important, it is still

pertinent to explore other venues to further ameliorate the energy efficiency of mobile de-

vices. For the lack of space, here we only review works performed at the MAC or network

layer of ad hoc networks, which are closely related to our study in this chapter. Based

on the mechanisms used, the solutions seen in the literature can be roughly classified into

three categories: Power-Saving Modes (PSM), Transmission Power Control (TPC) [22],

and Power-Aware Routing (PAR).










PSM is usually implemented at the MAC layer and the basic idea is to put the network

interface into the sleep mode when no communication is needed. One fundamental issue

in PSM is when to enter the sleep mode and for how long to stay in this mode. Some work

along this line includes PAMAS [23] and S-MAC [24]. In addition, Tseng et al. proposed

three asynchronous protocols, namely, Dominating-Awake-Interval, Quorum-based, and

Periodical-Fully-Awake-Interval protocols [25]. These proposals strive to efficiently and

intelligently control nodes' sleep and wake schedules and at the same time deal with such

factors as clock synchronization, neighbor discovery, and network partitions which are

inherent in multi-hop ad hoc networks [25].

TPC adapts the transmission power to the propagation and interference characteristics

experienced by the link [26, 27, 28, 29, 30, 31, 32, 33, 34, 35]. TPC sometimes is called

topology control when it attempts to control the transmission power or even turn off the

radio so that a desirable topology or connectivity can be maintained for saving energy.

Many proposals in this category are concerned with maintaining a dominant set of nodes

or forming some virtual backbone with certain clustering mechanisms. In addition, PCM

[36] was proposed to use different transmission power levels for RTS/CTS and DATA/ACK

frames on a per-packet basis.

In contrast to TPC protocols aiming at making each link as energy-efficient as possible,

a PAR protocol determines which of these links to be used for end-to-end paths so that

additional energy savings can be obtained by routing packets over energy-efficient paths.

Singh et al. proposed five power-aware routing metrics that could be incorporated into

routing protocols to reduce the per-packet energy consumption, to prolong the per-node

lifetime, or to prolong the overall network lifetime [37]. Besides, Chang and Tassiulas [38]

proposed a routing cost metric that is a combination of the transmission power level and

the residual energy. Toh [39] studied both the minimum energy and the network lifetime

issues and proposed a conditional min-max battery capacity routing protocol which tries to

strike a balance between these two competing objectives. While the link quality has been










suggested as a routing metric to reduce queuing delays and data loss rates, Banerjee and

Misra proposed a routing cost formulation that considers not only the residual power but

also the link characteristics, for example, the cost for potential retransmissions, to capture

the effects of both the link distance and the link error rate [40]. As an addition, PARO [41]

is another notable approach designed for scenarios where nodes can dynamically adjust

their transmission power. In PARO, a candidate intermediate node monitors an ongoing

direct communication between two nodes and inserts itself in the forwarding path if its

action can lead to some energy savings.

Most previous proposals do not take into account device heterogeneity inherent in

MANETs to achieve better energy conservation. How to take full advantage of P-nodes

to prolong the network lifetime as much as possible has been previously addressed in [21,

42, 43]. However, all of them focus on the network layer and many challenging issues

relating to the MAC layer are either left untouched or overlooked. For example, none of

them consider how to support the MAC-layer acknowledgements over unidirectional links

caused by different transmission power at P-nodes and B-nodes (cf. Section 2.4.4). By

contrast, our DELAR addresses the energy conservation from multiple facets, including

routing, transmission scheduling and power control.

In brief, although all the aforementioned proposals can achieve certain level of energy

conservation, how to design a comprehensive, practical framework that not only integrates

PSM, TPC, PAR, and transmission scheduling, but also makes full use of inherent device

heterogeneity remains an open, challenging problem. Our DELAR framework is proposed

to address this crucial issue.

2.3 Overview of DELAR

In this section, we first make a brief introduction to the problems we intend to address.

We then provide a high-level overview of our solution, DELAR, which stands for Device-

Energy-Load Aware Relaying framework for heterogeneous ad hoc networks.










2.3.1 Problem Statement

In this chapter, we focus on heterogeneous ad hoc networks comprised of mobile nodes

with different energy supplies, though heterogeneity may have other meanings in different

settings. We assume that besides a majority of battery-powered B-nodes, there exist some

powerful P-nodes having relatively unlimited energy supplies (cf. Section 2.1). Our objec-

tive is to develop energy conservation protocols by utilizing such heterogeneity in energy

resources. Intuitively speaking, since P-nodes are of "relatively" infinite energy reservoir

as opposed to the B-nodes with usually irreplaceable batteries, data communications should

try to utilize these P-nodes as much as possible in order to prolong the network lifetime.

Therefore, on the one hand, a packet should be forwarded to a P-node whenever an energy

saving can be expected. On the other hand, communications in the networks should avoid

using B-nodes if possible. For these purposes, it is desirable to allow P-nodes to have higher

transmission power so as to cover a larger transmission area, which can statistically reduce

the number of B-nodes involved in packet forwarding. However, this straightforward pro-

posal may pose lots of challenges to the protocol design How can a B-node be aware of

the existence of P-nodes in its vicinity? If there exist multiple paths through P-nodes to the

destination, which path should be chosen? Should the transmission range of a P-node be as

large as possible, or be kept within certain "optimal" ranges? How can the protocol support

reliable communications over the unidirectional links caused by asymmetric transmission

power at P-nodes and B-nodes along with error-prone and time-varying wireless channels?

In addition, higher transmission power often implies more reachable neighbors, decreased

spatial reuse, and increased local contention for the shared wireless medium, then how can

the protocol schedule the transmission activities so that a good balance can be struck be-

tween energy savings and other system performance factors such as packet delivery ratio

and end-to-end delay? These are all non-trivial questions and need to be answered before

we can indeed make full use of the aforementioned heterogeneity in ad hoc networks.










After a careful investigation on these interwoven issues, we believe that they are

closely related to routing, transmission scheduling, and power control. For example, with

the adjustment of the transmission power of P-nodes and B-nodes, the topology and neigh-

borhood, and thus the routing information, would change accordingly. So would the sched-

ule of transmission activities. Moreover, there exists a strong interaction between routing

and MAC layers. Apparently, our original design objective can be boiled down to designing

a joint routing, scheduling, and power control scheme, which should be addressed across

the whole protocol stack, especially at the routing and MAC layers [44][45]. To achieve

this, a cross-layer designed framework is demanded. In this framework, power control

should be implemented to optimize the transmission power of each node (both P-nodes and

B-nodes) to achieve optimal energy utilization and maintain a reasonable network topol-

agy; routing should be designed to inform all the nodes of the existence of P-nodes and find

the optimal energy-efficient routes; and transmission scheduling should be able to adjust

the transmission activities so that the energy spent on channel contentions and collisions

can be minimized. In addition, an appropriate scheduling scheme should be capable of

striking a good balance between energy efficiency and other system performance factors

such as end-to-end delay and packet delivery ratio.

2.3.2 Our Solution: DELAR

We consider a mobile ad hoc network consisting of N,, P-nodes and Nb B-nodes,

where N,, and Nb are system design parameters. We assume a single wireless broadcast

channel shared by all the nodes, though our DELAR can be easily extended to multi-

channel cases. We also adopt a simple power control scheme as follows. Each B-node

transmits omni-directionally and can maintain a circular transmission range BTR (basic

transmission range) before using up its battery, which can be properly set to achieve a good

tradeoff between energy efficiency and network connectivity [46]. In addition, we pos-

tulate that P-nodes are able to adjust their transmission power so as to cover larger areas

than B-nodes if needed. Moreover, all the P-nodes are assumed to have identical maximum










transmission range of PTRmos = M~x BTR, where M~ is a positive integer greater than

1. As revealed in [46], using common transmission power between the same type of nodes

can ensure bidirectional links and thus the correct operations of existing routing and MAC

protocols. With this simple yet efficient power control scheme, a unidirectional link only

exists between a P-node and a B-node when they use different transmission power, instead

of between any two B-nodes or P-nodesl According to [35], such simple power control

is believed to be more practical than other expensive transmission power control schemes,

either making unrealistic assumptions or having extra hardware requirements.

As mentioned before, DELAR arises from the following intuition: the P-nodes should

be utilized as much as possible. In other words, we should attempt to reduce the use of B-

nodes if we cannot avoid using them at all. Thus it is advantageous to enable a P-node to di-

rectly communicate with other P-nodes nearby or in distance by using higher transmission

power so that the number of B-nodes involved in the data forwarding can be much reduced.

However, higher transmission power or larger transmission coverage usually implies more

neighbors and increased local contention for the shared wireless channel. Therefore, in-

stead of granting these P-nodes unlimited privileges of reaching any other node at any time

at will, it makes more sense to constrain P-nodes' transmission power under certain bound

and to limit their transmission activities within some pre-planned periods in order to reduce

the collisions with other ongoing transmissions and thus maintain good channel utilization.

In order to better schedule the transmissions of P-nodes and B-nodes, we adopt a

time-division multiplexing method. We divide time into equal length time slots called

Supufmou'~lr l \. In each of the superframes, some time periods are exclusively designated

to P-nodes, while the rest are shared by all P-nodes and B-nodes in the network. More




SIn this chapter, we only consider asymmetric transmission power as the primary cause
for unidirectional links and omit others such as various collision/noise/interference levels
at different nodes.










P-to-P P-to-B P-to-B mini-slot P-to-B B-to-B

Pnd P-node 1 ... P-node k
Time Frame Duration

Figure 2-1: The structure of a Super Frame.


specifically, during one cycle of the Sep cT inlric (see Fig. 2-1), there is a P-to-P period

with length tp,, in which only P-nodes are allowed to communicate with each other by

using transmission range TR,, = m x BTR (1 < m < M~), while all B-nodes just keep

silent, as if the network were merely formed by these "mobile core" P-nodes. Addition-

ally, in one Sequ cT hana each P-node has its own exclusive period called P-to-B period

with equal length tpb, in which it can boost its transmission power to cover a range of

TRpb = a x BTR (1 < n < M~)2 The rest of one Sep cT homelc is called B-to-B period

with length tbb in which all the nodes in the network can contend for the channel and initi-

ate transmissions towards nodes in their TRbb = BTR. Obviously, all the P-nodes should

act as common B-nodes in the B-to-B period by adjusting their transmission range back to

TRbb. Notice that during one P-to-B period, since the P-node owning this period and the

B-nodes it intends to communicate with have different transmission power, unidirectional

links between them may be formed. Therefore, in contrast to the P-to-P and B-to-B periods

where some common contention-based MAC protocols such as the IEEE 802.11 can be

used, the P-to-B periods) demands an enhanced MAC protocol to support reliable com-

munications over unidirectional links. Our Asymmetric MAC protocol A-IMAC is exactly

developed for this purpose. Such rendezvous of reservation-based and contention-based

MAC schemes is able to schedule the packet transmission more efficiently, which we will

see shortly.


2 To provide reliable communications during P-to-B periods, usually n is less than m.










In DELAR, the heterogeneity of mobile nodes are also incorporated into the construc-

tion of routing tables. Routes are discovered based on routing metrics which take consid-

eration of the residual energy and the load status of mobile nodes. Once generating a data

packet, a node looks up its routing table and sends the data packet to the next hop as it does

in common ad hoc routing protocols. If residing in the forwarding path and having received

a forwarding request, a node will forward the data in an appropriate time period to the next

hop according to its own routing table. More specifically, for a B-node, when the next hop

is in its TRbb range, it can only forward the data packet during the B-to-B period. While

for a P-node, if the next hop is another P-node located in this P-node's TR,, the P-node

can forward the packet to the next hop in the P-to-P period, if the next hop is a B-node

located inside its TRyb, the P-node can forward the packet to the next hop in its exclusive

P-to-B period. In summary, with such time-division scheduling and a device-energy-load

aware routing metric, we intend to utilize P-nodes as much as possible in an efficient and

cautious way while reducing collisions and interference to an acceptable level, so that we

can achieve the expected energy conservation without harming other system performance

factors.

Several research challenges remain in supporting the seemingly simple operations of

DELAR as described above. Given a B-node (P-node) X located in a P-node P's transmis-

sion range TR,b (TR,,), for instance, what criteria should P adopt to determine if X is a

neighbor (in one hop range) or not, that is, forwarding a packet to this node X in a one-hop

manner or a multiple-hop manner? What kind of routing metric should be adopted to reflect

the heterogeneity in device types, nodal residual energy, and local load status when setting

up routing paths? How can we divide time into Supo,:,'~ll Wau. and how can one P-node

register a P-to-B period without conflicting with others' P-to-B periods? How does node

X send MAC layer acknowledgements back to P in the presence of unidirectional links

resulting from the asymmetry in transmission power? The remainder of this chapter will

address these questions one by one in more detail.










2.4 Design of DELAR

In this section, we will first discuss the neighbor-selection criteria of P-nodes. We then

illustrate the routing component of DELAR. Next, we introduce in more detail the hybrid

transmission scheduling of DELAR. Then, we present the Asymmetric Media Access Con-

trol Protocol (A-MAC) and the multiple-packet transmission scheme. Last we give some

further discussions.

2.4.1 P-nodes' Neighbor-Selection Criteria

In the literature, two nodes are usually considered as neighboring nodes of each other

when they are one hop away and they can directly communicate with each other. However,

in heterogeneous networks, we have to change the criteria to cope with the existence of

P-nodes whose transmission ranges are much larger than those of B-nodes. In this case,

any node in a P-node's TRpb COuld be a neighbor candidate of the P-node3 Nevertheless,

in order to support the MAC layer acknowledgements, not all the candidates can be finally

chosen as neighbors or be next hops in the routing table. Before presenting the rules that

guide P-nodes to make selection decisions, we first introduce the notions of Forward Path

and Backward Path. For any node pair s and t, a Forward Path indicates the path derived

from normal routing tables. For example, the Forward Path(s, t) can be represented as

s NI ~... Nk, t, where (Ni) 1 < i < k) denote the k intermediate nodes be-

tween s and t. For a given P-node P and any B-node X located in P's transmission range

TRpb, the Backward Path(P, X) is defined as the minimum-hop Forward Path(X, P) when

all the nodes have a transmission range of BTR. It is worth noting that the minimum-hop

Forward Path(X, P) is not necessarily the same as the Forward Path(X, P). Although For-

ward Pathyr\ are defined for any node pairs in the network, Backward Pathyr\ are only valid

between a P-node and the B-nodes located within the P-node's TRpb range. Furthermore,




SHowever, due to the asymmetric transmission power, the discussed P-node may not be
a neighbor of an individual node of those neighbor candidates.









for any neighbor candidate X of a given P-node P, this B-node X can be considered as

P's neighbor only when the Backward Path(P, X) satisfies the following criteria: All the

intermediate nodes along the Backward Path(P, X) should be in P's TRI>> range. In other

words, a neighbor candidate X can be considered as a P-node P's neighbor if and only if

all the intermediate nodes along the Backward Path(P, X) are P's neighbors as well.

With the above definitions, the remaining issue is how to set up these Backward rathir.

A simple way is to let a P-node broadcast a query message with a certain transmission

power, that is, covering all the B-nodes in its TRI,b = n x BTR range. Once seeing

such a query, each node broadcasts a special reply with the TTL value set to T 4 Each

node appends its own ID in the reply when relaying such a special reply. The querying

P-node will wait some time until collecting enough replies. The initiator of a reply would

be considered as a neighbor if and only if the querying P-node also receives replies initiated

from all the relaying nodes of that reply. We will see later, in order to facilitate the operation

of A-MAC, the path length of a Backward Path should be limited. We need to point out that,

even when a P-node, say P1, receives a query message initiated by another P-node, say P2,

P1 should reply like common B-nodes with a transmission range of BTR. Since our scheme

is targeted for networks with low or moderate mobility, P-nodes can execute this process

infrequently or in their respective P-to-B periods when topology changes are detected by

the MAC protocol. Therefore, the resulting overhead is expected to be affordable.

P-nodes also need to discover the neighboring relationship among themselves. To

achieve this, during the P-to-P period a P-node may send a query with appropriate trans-

mission power that is set to cover a range of TRI>; = m x BTR. P-nodes receiving this

query may directly send replies back to the requesting P-node.

Fig. 2-2 gives an example of the neighbor selection process. Suppose A is a P-node

with TRI,b = 2 x BTR and TRI~, = 4 x BTR, and the Backward Paths for neighbor


4 T can assume an integer value slightly larger than n to allow more replies.







22


/TR" O B-nod















Figure 2-2: An example of the neighbor determination process (m=4, n=2, T=3.).

candidates C, F, G and I are C B A, F E A, G F E A, and

I H C B A, respectively. Since node H does not initiate an reply to A, only

C, D, and G are considered as A's neighbors. Of course, B, D, and E are A's neighbors

as well. In this example, another P-node J is also a neighbor of node A because J is in A's

TR,, and they can directly communicate with each other.

2.4.2 Routing Component of DELAR

In homogeneous ad hoc networks, a node can only communicate with other nodes in

its BTR range, while in heterogeneous ad hoc networks, a P-node is able to reach any

other node within a larger transmission range, for example, TRpb and TR,,. Therefore, the

resulting topology and routing strategy may be quite different from that in homogeneous

networks. As an example, a network topology without P-nodes is depicted in Fig. 2-

3.a, where all the links are bi-directional and labelled with equal or unequal costs on both

directions. For instance, al/bi indicates that the link cost from A to B is al while bl from

B to A. In contrast, if one node, say A, is identified as a P-node who can reach much

further in the network, more unidirectional links may be added as shown in Fig. 2-3.b. We

label unidirectional links from P-node A to its neighbors with cost 0 to represent node A's

"unlimited" power supply.










B D




C E
a Topology in Homogeneous case



A FD




b Topology in Heterogeneous case

Figure 2-3: The topology in homogeneous and heterogeneous cases.


To cope with such heterogeneous networks, each P-node needs to maintain an internal

neighbor table recording its chosen neighbors within TRpb and the corresponding Back-

ward Pathr\ of those neighbors. In addition, each node in the network, either a P-node or

a B-node, needs to maintain a forwarding routing table similar to that in a normal table-

driven routing protocol such as DSDV [7]. For each node i, we define


P(i) = residual-en~ ,.ml(i) pL x queue_1en(i),

where residual-e 1, ml(i) indicates current remaining energy level at node i, queue _en(i)

represents the current load status at node i, and pL is a parameter representing the energy

consumption per unit data transmission Then the device-energy-load aware routing cost

metric we adopt is given in Eq. 2.1, though other cost metrics are applicable in DELAR as

well.


cost(i) = 1/P, S > (2.1)






5 In our simulation, for example, pL is equal to the average energy consumption per
packet transmission.










In the above cost metric, cost(i) is the cost of using node i as a relay, it could be

used as the cost of all directional links (arcs) starting from node i and directed to any of

its neighbors; y is a parameter used to adjust the weight of the awareness of load and

energy in the overall cost metric; constant a assumes a relatively large value to avoid us-

ing the nodes short of energy. Different types of devices may assume different values of

these parameters: pL, y and a~. For example, in this chapter, to represent a P-node's unique

power capability or device type, a P-node assumes a zero link cost6 fOr all the links to-

ward its B-node neighbors within TRpb Of P-DOde neighbors within TR,,. Ideally, in order

to find energy-efficient paths, each node should be informed about the routing costs of

other nodes as accurate and prompt as possible which may lead to excessive overhead. In

practice, however, the employed routing protocol should strike a good tradeoff between

energy efficiency and overhead. Proactive routing protocols are known for their capability

of propagating network conditions through the whole network in due course so that ap-

propriate QoS decisions, for example, admission control and route selection, can be made

intelligently. Thus, we adopt a proactive routing protocol, for example, DSDV, as the un-

derlying routing protocol to periodically exchange the routing information. We note that

other types of routing protocols can also be used in this framework. After gathering enough

routing information, a node can employ a shortest path algorithm to decide the minimum

cost paths and the related costs to all the other nodes in the network. Here the path cost is

actually the sum of the cost defined in Eq. 2.1 of all the B-nodes along a forwarding path.

Similar to those energy-aware cost metrics proposed in the literature, by choosing the

proper values of a~, pL and y, the cost function defined in Eq. 2.1 can help prolong the

network lifetime by distributing the traffic more evenly throughout the network, avoiding

the overuse of a small set of nodes, and consuming nodal energy resources in a more bal-

anced manner [37, 39]. Moreover, DELAR spontaneously incorporates P-nodes' unique


6 In practice, a very small value can be used to avoid possible routing loops.










power capabilities or device types, residual energy information, and local load status into

the routing protocol without using redirect tables in DEAR [43] any more.

2.4.3 Hybrid Transmission Scheduling

In order to reduce the interference a P-node's communication may impose on other

ongoing transmissions, it is reasonable to only allow a P-node to boost its transmission

power during some exclusively reserved periods. For this purpose, time is divided into

equal length time periods called Sequ' ,:nnlrti \ in which each P-node is assigned an exclu-

sive time interval, called a P-to-B period, to communicate with its B-node neighbors. In

addition, a P-to-P period is allocated to allow P-nodes to communicate with each other if

they are within each other's TR,, range. The rest of a Sty>
shared by all the nodes. Therefore, as shown in Fig. 2-1, a P-node has three different trans-

mission ranges in each Sty>
and TRbb in the rest of the Sep'cT; inrit In contrast, a B-node only has one transmission

range TRbb = BTR and it can only initiate transmissions during the B-to-B period.

Fig. 2-1 gives an instance of such a Srlet ,ntlric structure including multiple reserved

P-to-B periods, one for each P-node. The one-minislot-length paddings between consec-

utive P-to-B periods are used to further reduce the possible interference. The period allo-

cation of the Sty>
nodes use high transmission power to communicate and negotiate with each other, deciding

the lengths of the P-to-P period, the P-to-B period and the B-to-B period, also associating

each P-node with a P-to-B period. After finishing the negotiation, the P-nodes broadcast

such allocation information to all the B-nodes in their own TRpb. In this way, ultimately










all the nodes are informed the Supuf:T,~~ rame allocation, and can synchronize to such alloca-

tion7 In our current design, one P-to-B period is not allowed to share by multiple P-nodes

for simplicity only. However, a P-to-B period can be shared among P-nodes far away from

each other, if such sharing can ensure the transmissions conflict-free.

Since a P-node can communicate with other P-nodes in the P-to-P periods and com-

municate with the B-nodes within its BTR range in the B-to-B periods, it is natural that

in its own P-to-B period, a P-node should give priority to packets intended to its B-node

neighbors which are outside its BTR range but inside its TRI,, range. Thus, packet schedul-

ing is needed at a P-node to determine the appropriate transmission schedule for the packets

to be relayed or initiated by itself.

One may notice that the aforementioned transmission scheduling may need time syn-

chronization among the nodes, we argue that nodes equipped with GPS devices can easily

fulfill this task. This requirement is quite realistic today since such devices are inexpensive

and can provide reasonable precision. In this chapter we assume that nodes have perfect

time synchronization and leave the synchronization problem in the networks without GPS

devices as our future work.

In fact, the time-division scheduling, essentially a reservation-based access control

mechanism, and the MAC protocols employed in the three types of periods in each Super-

frame, either contention-based or reservation-based, form a hybrid transmission scheduling

for DELAR. Moreover, each type of periods may have different MAC protocol. For exam-

ple, the conventional contention-based MAC protocols, such as the IEEE 802.11 MAC

protocol, can be used during P-to-P periods and B-to-B periods. Since unidirectional links




SFor simplicity, we assume a fixed allocation scheme is used in this chapter, however,
more adaptive allocation is possible when P-nodes periodically exchange local load infor-
mation and negotiate a new allocation scheme during the P-to-P periods.







27

-L----- ~ Transmission







Figure 2-4: A unidirectional link between A and B.


are basically unavoidable in P-to-B periods, special measures are needed to deal with them.

In what follows, we will delineate the A-MAC protocol developed for P-to-B periods.

2.4.4 Asymmetric Media Access Control Protocol (A-MAC)

The presence of unidirectional links is pretty common in heterogeneous networks es-

pecially when employed power control schemes cause unequal transmission powers among

nodes. As an example, node A in Fig. 2-4 has a larger transmission range than node B.

Thus, node B can hear node A's transmission, however, node A cannot detect node B's trans-

mission. Obviously, a unidirectional link exists between node A and node B. The dilemma

is that the stop-and-wait ARQ (Automatic Repeat Request) scheme [47] employed in cur-

rent contention-based MAC protocols only works well with bidirectional links. In the face

of unidirectional links, the receiver B (Fig. 2-4) has no way to directly and successfully

send the acknowledgements back to the transmitter A, which means that the transmitter

would continuously transmit the same frame before timeout no matter whether the receiver

has received it or not. Moreover, the unidirectional links may severely affect the functional-

ities of ad hoc networks at various layers [48, 49, 50]. For example, many routing protocols

such as DSR and AODV rely on hop-wise acknowledgments for discovering route errors.

Therefore, how to support the MAC layer acknowledgements over the unidirectional links

is very important [51, 52] and has not yet been well addressed. Fortunately, we can make

use of the aforementioned Backward Pathr\ and the following "mini-routing method to

tackle this problem in an elegant way.









A (P-node)











Figure 2-5: The A-MAC operation procedure.


In current contention-based MAC protocols such as the IEEE 802.11, a receiver can

only transmit an acknowledgement frame to its one-hop-away transmitter. With the cross-

layer design methodology, we introduce a new concept of "mini-routing intoTrrack your

eBay activities the MAC layer, which requests intermediate nodes to relay the receiver's ac-

knowledgement frames, that is, CTS/ACK frames, along the established Backward Path(transmitter

receiver) in a multi-hop fashion to the transmitter (a P-node) at the MAC layer. Here the

routing information is no longer exclusively used by the network layer but shared by the

MAC and network layers.

Based on the IEEE 802.11 [5], we introduce into A-IMAC four special frames: P-

RTS, P-CTS, P-DATA, and P-ACK, all of which can only be transmitted in P-to-B pe-

riods. When a P-to-B period of a P-node comes and the P-node happens to have some

packets to transmit, it first boosts its transmission power to cover the range of TRpb =

n x BTR. With the scheduling described in Section 2.4.3, all the other nodes should re-

frain from initiating a transmission and temporarily stop transmitting usual frames, that

is, RTS/CTS/DATA/ACK. The P-node associated with this P-to-B period can send packets

to any neighboring B-node in the range of TRpb through P-RTS/P-CTS/P-DATA/P-ACK

exchanges.

Next, we illustrate the A-IMAC by using Fig. 2-5, where we assume n = 2 and P-node

A intends to send a packet to C, one of its B-node neighbors. The location relationship

among A, B, and C is also depicted in Fig. 2-5. First, A sends the P-RTS with TRpb



























































SSIFS stands for Short Inter-frame Space, and DIFS stands for DCF Inter-frame Space.


2 x BTR containing the Backward Path(A, C) (C B A). Then according to the length

of the Backward Path(A,C) in this example which is 2, A sets its waiting timer for the P-

CTS to be 2(SIFS+4T-crs+-tTprop) Where TP-CTs and Tprop are the transmission time

and air propagation time for one P-CTS respectively. Upon receiving the P-RTS destined to

it, and waiting for a SIFS period, node C will send node B a P-CTS including the addresses

of A and C. For an intermediate node residing on the backward path, it needs to set its

waiting timer according to its order in the Backward Path. For example, the ith node (the

intended receiver like C is assumed to be the Oth node on the path) on the path should set

its timer to be i x (SIFS +t TP-Crs +t Tprop). In this example, node B, when overhearing

the above P-RTS from A, it starts a timer equal to SIFS +t TP-ors + Tprop Since it is

the 1st intermediate node on the backward path. Once receiving a P-CTS from node C

before timeout, B simply appends its address and relays the modified P-CTS to the next

hop which is the P-node A in this example. Otherwise, node B sends a P-CTS containing

its own address to A after the timer expires, and the reason for doing this will be explained

later. If node A does not receive any P-CTS before timeout, it can retransmit the P-RTS

until reaching an admissible number of retries. If the same situation happens, A temporarily

saves this packet for future transmission and switches to another packet with a different next

hop. When A successfully receives a P-CTS from B containing both B 's and C's addresses,

the P-RTS/P-CTS exchange finishes. After a SIFS, A can send a P-DATA frame to node C

and set the timer to 2(SIFS+- TP-ACK -tTprop). Then the similar procedures apply. After

receiving the P-ACK from node C relayed by the intermediate node B, the P-node A can

start transmitting a new packet after a DIFS in the same manner. When its P-to-B period

expires, A lowers its transmission power and acts as a B-node in other P-nodes' P-to-B

periods and in the B-to-B period.




















Figure 2-6: The multiple-packets transmission enhancement to DELAR.


Besides the purpose of resolving the well-known hidden/exposed terminal problems,

the P-RTS/P-CTS exchange is also used to eliminate possible errors resulting from stale

routes or nodes' mobility. For example, in the above example, if node C moves out of

P-node A's 2 x BTR range while B is still in A 's BTR range, node C will not hear the P-

RTS from node A and hence A could only receive from node B a P-CTS including only B 's

address. In this case, A will think that node C is currently unreachable and may temporarily

save the packets to C for future transmissions. Another situation may happen that node C

is still in A's 2 x BTR range while node B moves out of A's BTR range, in which the

P-node A will delete node C from its neighbor table. Moreover, when node C moves into

A's BTR range, A will hear the P-CTS from node C directly. Hence A can optimize the

future transmissions to C without the help of node B any more.

2.5 Multiple-Packets Transmission in DELAR

In the basic DELAR design, time-division transmission scheduling is used to coordi-

nate the transmission activities of P-nodes and B-nodes. One undesirable consequence is

the excessive delay one packet may experience because it may be buffered at intermediate

nodes to wait for appropriate transmission periods. On the other hand, since DELAR is

energy aware and it costs P-nodes almost "nothing" to transmit a packet, many data pack-

ets may swarm to P-nodes. This may make P-nodes the "bottlenecks" of the network and

further increase the delay that packets accumulated at P-nodes would experience. In what

follows, we seek a way to alleviate this phenomenon.










2.5.1 Multiple-packets Transmission

In the basic design (see Fig. 2-5), for example, A can only transmit packets to either B

or C each time. When P-node A transmits a packet to B-node C, node B is only involved in

forwarding the control frames (for example, P-CTS/P-ACK). However, since the channel is

reserved for P-node A during its P-to-B periods and node B is also in P-node A's transmis-

sion range, node B has the capability to receive and demodulate any signal transmitted by

node A if it is allowed. Therefor, if node A can recognize B's capability and utilize it, that

is, transmitting packets to node B and C at the same time, the system performance could

be improved. This multiple-packets transmission mechanism can be illustrated in Fig. 2-6.

Suppose P-node A has some packets to B and some packets to C. Node A would put one

packet for C and another for B together, and send them in a single transmission from which

nodes B and C can acquire their own part, respectively. In this way, we expect to see the

improvement of the end-to-end delay performance of DELAR.

To do this, node A first makes sure that both B and C are within its effective range

TRpa after the P-RTS/P-CTS exchange as before. Then A can pull out from the waiting

queue one packet towards C and another packet towards B, and send them in one P-DATA

frame as depicted in Fig. 2-6. When seeing such a frame, nodes B and C can extract their

own parts and dump the rest. The same procedures as specified in A-MAC are executed

with the exception that node B also needs to indicate in the P-ACK that it has successfully

received the packet to itself. The similar procedure can be applied to the cases with more

than one interlineate nodes on the backward pathir.

With the above multiple-packets transmission mechanism, one may expect that the

total number of packets that can be packed and transmitted at one time is bounded by

the length of the Backward Path. Considering the possibility of adopting high-data rate

modulation schemes with a higher power level so as not to degrade the received signal

strength, more packets towards different receivers on the Backward Path, can be assembled

together and transmitted at the same time.
















C/ D




Figure 2-7: General hierarchical 2/4-PSK constellation.

2.5.2 Hierarchical Modulation

To support the multiple-packets transmission, hierarchical modulation (or nonuniform

modulation) schemes can be use to ensure all the receivers have enough received signal

strength to demodulate the useful information. In hierarchical modulation schemes, the

constellations consist of nonuniformly spaced symbols and allow for unequal error protec-

tion, that is, different degrees of protection for transmitted bits within a symbol are allowed

according to the importance of the information. For example, suppose that there are two

streams of data, each of which has a priority (a target BER), and QPSK hierarchical mod-

ulation (see Fig. 2-7) is used. One bit from each stream is taken to form a symbol of two

bits. The bit from bitstream with high priority (lower allowed BER) is assigned to the most

significant position, for example, the first bit in the constellation in Fig. 2-7. The bit from

the bitstream with low priority is assigned to the less significant position. By adjusting

the angle of 8, the constellation can be determined to achieve the unequal error protection

for each bitstream. Such unequal error protection has make hierarchical modulation very

attractive in multimedia services [53, 54, 55].

We notice that in the previous research, the hierarchical modulation is used in the cases

that the transmission power is fixed and the constellation is chosen in a way to allocate the

power among multiple data streams to achieve unequal error protection. In addition, in

the previous research, bitstreams usually are destined to the same receiver. However, in our









~A (P-node)







b, o




Figure 2-8: Implement multiple-packets transmission with hierarchical modulation


DELAR scheme, the transmission power of P-nodes is not fixed. Moreover, when multiple-

packets transmission is used, the bitstreams are destined to different receivers. As shown

in Fig. 2-6, multiple receivers involved in the multiple-packets transmission have different

distances to the transmitter. Therefore, in order to use hierarchical modulation to support

multiple-packets transmission at a P-node, more power ought to be allocated to far away

receivers and less power for close receivers. More specific, we need to find out the constel-

lation and the minimum overall transmission power for a P-node so that all the receivers

can successfully get their own packets. In the following we use QPSK hierarchical modula-

tion to implement the multiple-packets transmission when two receivers are involved (Fig.

2-8).

Suppose BERmay is the BER requirement for both receivers B (d2 aWay ffom trans-

mitter A) and C (dl away from transmitter A) to demodulate the received signal. Moreover,

we assume two-ray ground model with a~ = 4 is used to model the radio propagation:

PtcrG,h,2h,
P=
r 4L

where L is the system loss factor, Pt is the transmission power, Pr is the received power,

Gr(G,) is the antenna gain at the transmitter(receiver), and ht(h,) is the height of antenna

at the transmitter(receiver) For simplicity, we assume all the nodes have the same antenna

gain and height. We assume that 8 is less than 450 so that the inphase signal towards far










away receiver C is assigned with more power and quadrature signal towards close receiver

B with less power. It is easy to get the BER at C as

PtceG,h~h h
pl Q cos 8) = (1 2 cos 8) (2.2)
Nf \Nf

where N is the noise signal power, f is the transmission bit rate, W is the channel band-

width (in Hz), and Q() is Gaussian Q-function [54, 55]. Similarly, we can get the BER at

the receiver B as

PtGeG,h~h h
p2 = Q( L Sill 8) = \ 2 Sin 8). (2.3)


Thus, in order to correctly demodulate the received signal, the BER should satisfy the fol-

lowing conditions: pi < BERmax where i = 1, 2. Since we are interested in the minimum

transmission power of node A when equality is satisfied: pi = BERmax where i = 1, 2,

we define

To = Q- (BERmax)

as the SNR achieving the BERmax to facilitate us finding the boundary values of Pt and

0. From Eq. 2.2 and Eq. 2.3, it is easy to get

PtGeGThh~ h
S2 cos 8 = To (2.4)
\Nf

and
PtGeG,h h?
S2 sin 8 = To. (2.5)
\Nf
From Eq. 2.4 and Eq. 2.5, we can get


0 = arctan 2~j (2.6)


and

Pti (2.7)
mm-WGtG, ht2h









Therefore, given the distances between the transmitter and receivers, we can get the

minimum transmission power Ptain and 8 in the QPSK hierarchical modulation constel-

lation, and use this combination to implement the proposed multiple-packets transmission

and achieve the BER requirements at both receivers. In practice, P-node A may first use

a predefined transmission power covering TRpb, and transmit the RTS frame using typ-

ical BPSK. After the RTS-CTS handshake, A can determine the two parameters Ptmin

and 8 based on the feedback from the two receivers. Then A pulls out from the waiting

queue one packet towards C as the more important bitstream and one towards B as the

less important bitstream. Further it can transmit symbols using the aforementioned QPSK

hierarchical modulation. We should note that, in order to avoid the possible interference

when a P-to-B period is shared by multiple P-nodes, the involved P-nodes, before enabling

the multiple-packets transmission, should examine and make sure that Ptain is not greater

than the power to cover TRpb.

In fact, the minimum total transmission power to transmit the packets to each receiver

separately using BPSK can be written as follows:

ToN fdiL ToN fd L
ptot,, = p,1 + Pt2 = .(2.8)
WGtG,h~h~ WGtG,h~h~

Compared Eq. 2.7 to Eq. 2.8, we can see that although P,,in alone is larger than ei-

ther Ptl or Pt2, Ptmin is identical to Peroat. This verifies that multiple-packets transmission

with QPSK hierarchical modulation does not require extra transmission power 9 compared

to the case when multiple-packets transmission is not used. In fact, if multiple-packets

transmission is implemented with uniform QPSK, in order to satisfy the same BER re-

quirement at both receivers: pi < BERmax where i = 1, 2, the transmission power would

be JZPri which is much larger than Ptain. Thus, in our case, QPSK hierarchical modu-

lation is a rather reasonable option to implement multiple-packets transmission. Further,


9 Here we only consider the power used to transmit the information bits.










if we consider the energy used for control frames, the proposed multiple-packets trans-

mission requires fewer control frames thus less energy consumption, compared to the case

when packets are transmitted separately. For example, in Fig. 2-6, transmission energy for

three control frames (RTS/CTS/ACK) towards node B can be saved when multiple-packets

transmission is used.

2.6 Discussion

2.6.1 The Existence of Backward Paths

In DELAR, the P-nodes are utilized in two ways to conserve energy: enabling P-

nodes to directly communicate with other P-nodes within TR,,, and enabling P-nodes to

directly communicate with B-nodes within TRpb. Unlike the communications between P-

nodes that need no special treatment, the communications between P-nodes and B-nodes

are supported by the nodes on the backward paths as described in Section 2.4.1, thus the

backward path is very important for the proper functioning of DELAR. Then one may

question the existence of such backward paths. We believe that the existence of backward

paths is related to several factors such as basic transmission range (BTR), node density, and

the location distribution of P/B-nodes. On the other hand, the backward path should not be

too long in terms of hop count, otherwise A-MAC may not function well because a long

backward path is more subject to breakage due to node mobility or other reasons. In prac-

tice, we should include this limitation into the aforementioned neighbor-selection criteria,

for example, setting TTL value of a reply to the neighboring query to n (cf. Section 2.4.1),

such that only those B-nodes with backward paths of less than n hops can be considered as

neighbors of a given P-node.

Next we want to study the existence of such backward path in typical mobile ad hoc

networks. We assume n = 2 and T = 2 (cf. 2.4.1) so that only those nodes with backward

paths not more than 2 hops are considered as neighbors of a given P-node. Let O denote

the position of a P-node, and O' denote the position of a B-node in O's 2 x BTR = 2R

range, as shown in Fig. 2-9. We use r to indicate the distance between O and O'. It is









obvious that when r is not greater than R, O' is definitely a neighbor of O. So we shall

only consider R < r < 2R when studying the existence of backward paths for those nodes

that are r away from O. We also assume that all the nodes are uniformly and independently

distributed in a two-dimensional area with node density rl, where the probability of having

k nodes in an area of size S follows Poission distribution:


p(k, S) = e-e


The mean of the number of nodes in the area of size S is rlS. Then the existence of

backward paths between O and O' or the problem whether O' is a neighbor of O or not

can be reduced to a rather simple geometric problem: Is there any other node located in the

shadowed area S with both O and O' as neighbors?

Let -r be the average number of nodes existing in the shadowed area, and S1 and S2 be

the areas of sector A1BO and AAOB, respectively. Then we have -r = 2rl(S1 S2), where


S1 = x 2 arccos =I21/ R2 RECCOSI
2x R R

and
1 r/2
S2 -R2 Sill 2 arCCOS ~
2 R

Therefore, we have

r/2 1 r/2
-r = 2rl(R2 arCCOS 8 -2 Sill 2 arCCOS )
R2 R

Thus, the probability of existing a backward path between O and O', which is equal to the

probability that there is at least one node in the shadowed area, is 1 e-' according to the

Poission distribution. Fig. 2-10 plots this probability with different values of r, where the

total number of nodes is 50 for all network topologies. We can see that, with a reasonable

node density, the probability of existing a backward path between a given P-node and B-

node pair is pretty high, which justifies the feasibility of our A-MAC protocol and in turn

the overall DELAR scheme.





















































I


Figure 2-9: The existence of backward path.


1
I
1
BTR=~50




i


BTR=20


S1500*300 m2
h-7070 m
900*900 m2
-9-1000*1000 m2


200 250 300 350
r (m)


400 450 500


Figure 2-10: The average number of nodes existing in the shaded area.










2.6.2 DELAR and ZRP

In fact, we can borrow some ideas from ZRP [56, 11] to further improve the routing

performance of DELAR as follows. A P-node maintains the routing information within a

zone TRI~a by using the procedures described in Section 2.4.1 or any other routing proto-

col as the intra-zone routing protocol (IARP). It also maintains the information about its

neighboring P-nodes in its TRI~,. In contrast, a B-node only needs to maintain the routing

information in its TRbb = BTR, where all the nodes are its one-hop neighbors. Once a

node needs an energy-efficient route to another node, it can discover the route on demand

using the routing discovery procedure similar to that of AODV but with what is defined

in Eq. 2.1 instead of hop count as the cost metric. In this sense, DELAR can be viewed

as a special case of ZRP. The difference lies in the fact that all nodes within a P-node's

TRI,b zone, are "one-hop" away from this P-node from the routing perspective rather than

multi-hop away in the legacy ZRP. Besides, the border-casting [56, 11] technique used in

ZRP is also available in DELAR in the sense that a P-node border-casts a route request

(which we call simply a request) to all its peripheral nodes with corresponding backward

paths embedded in the route request. In this way, each peripheral node would learn the

backward path used to return a route reply if needed. Moreover, since a P-node can directly

exchange routing information with other P-nodes within its TRI>,, it also can border-cast

the route request to its P-node neighbors, which in turn can look up their own neighbor

tables to decide if the desired destination is in their own TRI>> zones. Since TRI>; is usually

larger than TRI,b, apparently, such "inter-zone routing protocol (IERP)" can further speed

up the route discovery process.

2.6.3 The Choice of m and n

Since a P-node can use higher transmission power to communicate with other P-nodes

within its m x BTR, larger m would lead to less use of B-nodes in the communications,

but also less spatial reuse. Similarly, larger n may lead to more energy savings, but also

imply possible longer backward paths and less spatial reuse, which may make A-MAC less










efficient as explained before. For the similar reasons, usually n is less than m. In order to

well balance the energy savings and other system performance factors, both m and n should

be chosen cautiously. While DELAR requires all the P-nodes have the same value of m to

avoid producing unidirectional links between P-nodes in P-to-P periods, with A-MAC in

place, they can have different values of n.

2.6.4 Benefits of the Time Division Scheduling

In fact, P-nodes can communicate with each other in one-hop or multi-hop manner

during P-to-P periods to coordinate the use of the next Sep cT f~rame. A P-node can also

notify all the other nodes within its TRpb to adjust their transmission schedules. With

these intelligent communications available, the slot allocation in each Sep cT f~~rame can be

adjusted adaptively according to traffic conditions instead of being fixed as in the given

example. It is worth pointing out that this time-division scheduling method can facilitate

the operations of PSM (cf. Section 2.2) in that it can help nodes determine their sleep and

wake schedules. For example, B-nodes can tumn off the radios during the P-to-P periods to

conserve energy.

As we discussed before, time synchronization is of importance for the correct oper-

ation of DELAR. In literature, there exist many proposals for time synchronization and

many of them can be incorporated into DELAR. In fact, P-nodes can serve as coordinators

to facilitate such synchronization by sending out some beacon information in some P-to-

P periods and P-to-B periods periodically. Subsequently, B-nodes can synchronize their

clocks to these P-nodes.

2.7 Performance Evaluation

2.7.1 Simulation Setup

In order to evaluate the performance of DELAR, we implemented our scheme includ-

ing the routing layer and the A-IMAC in the OPNET Modeler [57]. We simulated a network

with 50 nodes randomly deployed in a 1500 x 300m2 area. The BTR was 200m and the

transmission rate was 2M~bps. In our simulations, all the nodes were capable of moving










in the network according to the modified random Waypoint mobility model presented in

[58]. The pause time was set to be zero in our simulations, meaning nodes were always

moving. And each node moved with a randomly chosen speed between [ niEnc,Vm]z, where

Vni,z was fixed to be 1 m/s and V,,,,. assumed different values to reflect various network

mobility levels.

There were 20 constant bit rate (CBR) data sessions between randomly selected source

and destination pairs, and each source generated data packets of 512 bytes in length at a rate

of A packets per second. In our simulation, B-nodes had the same initial energy reservoir

6k J and their transmission and reception power were 1560m W and 930m W, respectively

[59]. The networks with 2 to 6 P-nodes were studied and these P-nodes were randomly

deployed. Besides, we chose m = 4, n = 2, tz>; = 0.02s, 1;>> = 0.02s, and tbb = 0.05S10

By varying the number of P-nodes, the maximum moving speed, and the CBR source rate,

we were able to study the performance of DELAR under different configurations. Six runs

were carried out to get an average result for each simulation configuration and each run

was executed for 900 seconds of simulation time.

Previous work [37, 39] has shown that the energy efficiency of routing protocols can

be much improved by adopting such a routing metric as what we defined in Eq. 2.1. There-

fore, we shall only compare our DELAR and DELAR with multiple-packets transmission

(denoted by DELAR+MPT) with the one referred to as EAR in this chapter, which is de-

veloped for comparison purpose and in fact a variant of DSDV with the routing cost metric

defined in Eq. 2.1. In EAR, all the P-nodes have the same transmission range as B-nodes,

but have a zero cost to their neighbors because they are assumed to have almost unlim-

ited energy reservoir and other resources. And the following metrics will be adopted in




to With advanced simulation tools such as OPNET, NS-2, and Qualnet, it is possible to
figure out a good configuration of these parameters to yield a good system performance,
before putting DELAR into practice.










comparison: average energy consumption defined as the total energy consumption for

all packet transmissions and receptions normalized by the number of delivered packets;

packet delivery ratio defined as the ratio of delivered data packets to those generated by

the sources; average packet end-to-end delay defined as the average delay from when a

packet is generated and transmitted by the source till it is received by the destination.

2.7.2 Impact of The Number of P-nodes

We first fixed the data rate to 4 packets/s and varied the number of P-nodes to study its

impact on the performance of DELAR. Here only the results for neu.=4 m/s are presented

in Fig. 2-11, though DELAR has the similar performance with other values of Exca--

Figure 2-11(a) compares average energy consumption of DELAR and EAR under

different numbers of P-nodes. Since EAR is also an energy-aware routing protocol, it is of

no surprise to see that its energy-saving performance improves with the number of P-nodes

whose routing costs to their neighboring nodes are assumed to be zero. With DELAR in

place, however, the average energy consumption can be much further reduced, for exam-

ple, with a factor of almost 50% if 6 P-nodes are available. The shown advantage comes

from the fact that DELAR makes much better use of P-nodes than EAR through intelligent

transmission scheduling and the allowance of P-nodes using different transmission power

during various periods. And the more P-nodes, the more energy savings we can expect from

DELAR. Compared to the basic DLEAR, DELAR with multiple-packets transmission can

further reduce the energy consumption. It can be contributed to the reduced transmissions

of control packets because multiple-packets transmission requires fewer control frames

than separate transmissions.

From Fig. 2-11(b), we can see that DELAR outperforms EAR in terms of packet

delivery ratio (PDR). The reason is that transmission scheduling in DELAR will lead to

less congestion, and the larger transmission ranges of P-nodes during P-to-P periods and

P-to-B periods can help reduce the number of hops a packet may travel. Again, the more

P-nodes, the more PDR is improved. Compared to the basic DLEAR, the multiple-packets









































12. I


0.00 .


SEAR
SDELAR
SDELAR+MPT


23 4
Number of P-nodes


5 6


(a) Average energy consumption


0.962-
0.960-
05-
0.958-
05-

0.952
-
~i0.950
CC: 0.948
S0.946
2 0.944
"0.942-
0.940 -
S0.938 -
,0.936 -
0.934 -
0.932 -
0.930
0.928


SEAR
SDELAR
tDELAR+MPT


234~
Number of P-nodes


5 ~6


(b) Packet delivery ratio


0.40

0.35-

0.30-

S0.25

S0.20

0i .15-




a,0.05 -


SEAR
SDELAR
SDELAR+MPT


Number of P-nodes


5 6


(c) Average packet end-to-end delay


Figure 2-11: Simulation results with different number of P-nodes.










transmission can further improve the packet delivery ratio because transmitting multiple

packets simultaneously can improve the channel utilization.

In terms of end-to-end delay, however, EAR is better than DELAR as shown in Fig. 2-

11(c). This is because, in contrast to the contention-based transmissions in EAR, DELAR

divides the time into Super Frames to schedule the transmission activities so that a packet

usually needs to be buffered at a node waiting for the coming of a proper transmission

period. In addition, we can observe that the delay of EAR decreases with the increase of

P-nodes because the existence of more P-nodes can achieve better load balance, that is,

the number of energy-efficient paths may be increased. However, this is not always the

case with DELAR. Besides the better load balance, the increase of P-nodes also means for

DELAR that the delay from the transmission scheduling becomes larger. This contributes

to the fluctuation in the delay of DELAR depicted in Fig. 2-11(c). Compared to the basic

DLEAR, DELAR with multiple-packets transmission can reduce the packet end-to-end

delay.

2.7.3 Impact of Node Mobility

In this subsection, we study the impact of the node mobility on DELAR by varying

Vm,, from 2 m/s to 16 m/s. For the reason of clarity, only the results for 4 P-nodes and a

data rate of 4 packets/s are presented.

Fig. 2-12(a) compares the average energy consumption of DELAR and EAR under

different mobility levels. As we can see, DELAR always has less energy consumption than

EAR due to the reasons stated before. Generally the higher mobility leads to less energy

consumption. After examining the average number of hops a packet may travel, we notice

that higher mobility often results in shorter route, which statistically leads to less energy

consumption because fewer transmissions and receptions are involved. As shown in Fig.

2-12(b), the packet delivery ratio decreases with the increase of the mobility, which is in

accordance with previous studies. For the similar reason we stated in previous subsection,

DELAR always has higher packet delivery ratio than EAR in all kinds of mobility. As





































































































I I I II I I I I I I I I I
0 2 4 6 8 10 12 14 16 18
Maximum Node Speed: Vmax (m/s)




(c) Average packet end-to-end delay


EAR
SDELAR
SDELAR+MTP


-


Maximum Node Speed: Vmax (m/s)




(a) Average energy consumption


SEAR
SDELAR
SDELAR+MTP


096-

0 96 -


094
; 0 92 -

S0 90 -



$ 0 66-

0 4 -

0 2 -


4


-


2 4 6 6 10 12 14 16 16
Max mum Node Speed: Vmax (m/s)


(b) Packet delivery ratio


036-
034.
032-

S028-
o 026-
,0 24 -
,0 22 -
S0 20 -
~018-
016-
S014-
S0.O12-
S010-


0 04 -


-tEAR
SDELAR
SDELAR+MTP


--~


Figure 2-12: Simulation results with different maximum node speed.














0 96-

0 95-
EAR
cc DELAR
~094-1 DEALR+MPT

0 93

092-

0 01
200 300 400 500 600 700
Traffic Load (Kbits/s)



(a) Packet delivery ratio


0 40-
0 35-
S0 30 -


S0 20-
015-
S010-
EAR
~005- *DELAR
~ ,,,~ ~DELAR+MPT

200 300 400 500 600 700
Traffic Load (Kbits/s)



(b) Average packet end-to-end delay


Figure 2-13: Simulation results with different traffic load.



the mobility increases, generally the delays of both EAR and DELAR get longer. Again,


DELAR has longer delay than EAR due to DELAR's time-division medium access con-


trol mechanism. One interesting observation is that the delays of both DELAR and EAR


fluctuate around Encer = 2m/s and Encer = 2m/s. This can be attributed to the used rout-


ing cost metric which causes lots of packets swarming to the P-nodes. This phenomenon


results in longer waiting time at P-nodes. However, nodal movement helps alleviate such


phenomenon by dispensing the traffic load. Compared to the basic DLEAR, DELAR with


multiple-packets transmission can reduce the energy consumption, improve the packet de-


livery ratio, and shorten the end-to-end delay in all kinds of mobility.










2.7.4 Impact of Traffic Load

In this subsection, we study the impact of the traffic load on DELAR by varying the

data generation rate from 3 packets/s to 8 packets/s. Since the traffic load does not much

impact average energy consumption, we only depict the simulation results for the packet

delivery ratio and the end-to-end delay in Fig. 2-12, where the number of P-nodes is four

and V,0, is equal to 4 m/s.

Figures 2-13(a) and 2-13(b) demonstrate that the packet delivery ratio decreases

and the delay increases with the increase of the traffic load for both schemes, respectively,

which are quite intuitive. Again, our DELAR is better in terms of the packet delivery ratio

but worse with regard to the end-to-end delay than EAR for the reasons stated previously.

However, Compared to the basic DLEAR, DELAR with multiple-packets transmission out-

performs DELAR due to reason described before.

In summary, DELAR is more appropriate for delay-insensitive applications, such as

file transfer and web access, which prefer higher energy-efficiency and packet delivery

ratio. With multiple-packets transmission, the performance can be further improved. Thus

it is a viable enhancement and can be effectively used in DELAR framework. With a better

tuning of system parameters DELAR can strike a good balance between energy efficiency

and other system performance factors.

2.8 Summary

In this chapter, we proposed a Device-Energy-Load Aware Relaying framework, namely

DELAR, to achieve energy conservation in mobile ad hoc networks. DELAR utilizes the

device heterogeneity inherent in ad hoc networks and features the cross layer protocol

design methodology. To take better advantage of powerful nodes (P-nodes) while reduc-

ing their interference on other ongoing communications, a hybrid transmission scheduling

mechanism is used to schedule and coordinate the transmission activities among P-nodes

and B-nodes (ordinary nodes). In addition, in order to support reliable transmissions in the










presence of unidirectional links between P-nodes and B-nodes, we introduced the "mini-

routing" technique and the novel Asymmetric MAC (A-IMAC) protocol. We demonstrated

that A-IMAC can effectively enable the MAC layer acknowledgements over unidirectional

links. To the best of our knowledge, no previous effort has been made to address this issue

at the MAC layer. Moreover, we proposed a multiple-packets transmission scheme which

can be operated with hierarchical modulation scheme to further improve the performance

of DELAR. Detailed simulations showed that DELAR can significantly reduce the energy

consumption and thus prolong the network lifetime even with a few P-nodes existing in the

network. With this framework, various energy conservation techniques such as power say-

ing modes, transmission power control and power-aware routing can be integrated to jointly

achieve better energy conservation. More important, this framework provides a platform to

study other challenging issues such as QoS provisioning and security support.















CHAPTER 3
RESOURCE AWARE MOVEMENT IN HETEROGENEOUS
MOBILE AD HOC NETWORKS

3.1 Introduction

As discussed in Chapter 2, there has been a rich literature addressing the energy con-

servation issue in MANETs, ranging from power-saving mode(PSM) [23, 60, 25, 24, 61],

to transmission power control(TPC) [26, 32, 27, 30, 33, 34], and to power-aware rout-

ing(PAM) [37, 40, 41]. Though those energy-aware MAC and routing protocols can in

general help the whole system expend the energy resources more reasonably to some ex-

tent, there are still many other aspects one can explore to further improve the system-wide

energy efficiency. As an example, similar to the traffic jam in daily life where a mass of

vehicles flocks to a single spot, unwise movement may cause local network traffic conges-

tion, thus leading to unfavorable energy waste. This observation motivates us to address

energy conservation from the perspective of node movement.

In addition to node movement, resource heterogeneity, as another inherent character-

istic of MANETs, is often either overlooked or underutilized in designing energy conserva-

tion schemes for MANETs. Though node heterogeneity can be interpreted in various ways,

we limit the scope of this chapter to heterogeneous networks in terms of energy supply. In

such a network, most nodes (called B-nodes hereafter) are furnished with lightweight bat-

teries having limited power, while a few others (called P-nodes hereafter) are powered by

almost unlimited energy supplies such as energy-scavenging devices (for example, solar

cells) and dynamos when nodes are in some mobile vehicles. In a relative sense, the energy

consumption of P-nodes can be considered as small or even negligible.

In this chapter, we propose to address energy conservation by guiding nodes's move-

ment and utilizing device heterogeneity. The basic idea is that, instead of moving in the










field blind to the network environment, for example, always following shortest-hop paths,

nodes are instructed to travel much more intelligently by keeping in mind the system-wide

objective of energy conservation and moving along "resource-aware" paths in such a way

that P-nodes can undertake as much communication tasks as possible so that less powerful

P-nodes can save energy, thus leading to the prolonging of the whole network lifetime.

The contributions in this chapter are mainly fourfold. First, we define a general mo-

bility model and formulate a general resource-aware movement problem, from which we

derive a Waterhunter Movement problem for B-nodes and a Firehunter Movement prob-

lem for P-nodes. Second, we reduce the Watehunter Movement problem to a NP-complete

distance-constrained least-cost (DCLC) routing problem and propose an efficient heuristic

solution. Third, we propose a routing delay differentiation mechanism to make full use of

the benefits provided by the resource-aware movement. Last, this resource-aware move-

ment can be incorporate into other energy conservation schemes to further improve the

energy efficiency. The effectiveness of the proposed schemes are justified and validated

through extensive simulations.

The rest of the chapter is organized as follows. We start with Section 3.2 surveying

the related work, then we formulate the resource-aware movement problem in Section 3.3.

In Section 3.4, we focus on the Waterhunter Movement problem and propose an efficient

heuristic solution. Section 3.5 evaluates the performance of the proposed schemes. Finally,

this chapter is concluded in Section 3.6.

3.2 Related Work

In this section, we brief some of related work that are closely related to this research.

Grossglauser and Tse showed that node mobility can be used to improve network

throughput [62]. After predicting the destination's location, a node forwards a data packet

to a subset of its neighboring nodes in the direction of the destination to reduce the overall

routing energy consumption [63]. Similarly, movement information was also used in [64] to

limit the flooding in a restricted area to reduce the energy consumption of route discovery.










Chakraborty et al. proposed a strategy to reduce the energy consumption by delaying

communication until a mobile node moves close to its peer target, within an application-

imposed deadline [65].

Different from the above work trying to utilize existing mobility information, there

are some work discussing how to move proactively or control node mobility to improve the

system performance. Li and Rus proposed an optimal algorithm to computer the trajectory

of nodes for minimizing message transmission delay [66]. Goldenberg at al. proposed a

distributed mobility-control scheme to guide the nodes to adjust their movements with the

purpose of attaining a potential energy-minimizing network configuration [67]. In sparse ad

hoc network, proactive node mobility can as well be used to overcome network partitions.

More specifically, nodes should buffer and carry packets during network partitions, and

forward packets to other nodes when they meet. Such a store-carry-forward paradigm was

proposed to help data delivery by making use of node mobility [68]. In addition, Zhao et al.

[69] proposed a message ferrying (MF) approach to address the similar network partition

problem. In their approach, a set of special mobile nodes, called mobile ferries, provide

data forwarding service to other nodes by either moving along the routes known a priori to

other nodes or proactively moving to meet other nodes.

Our resource-aware movement (RAM) strategy proposed in this chapter is also a

proactive movement approach. It differs from all the previous work in its unique designing

objective, that is, finding the optimal movement trails for each individual node to minimize

the total energy consumption of the whole system by taking into consideration the inherent

resource heterogeneity of MANETs.

3.3 Problem Formulation

We consider a MANET that consists of tens or even hundreds of mobile nodes, among

which there are N, regular battery-powered nodes (called B-nodes hereafter) and N, pow-

erful nodes (called P-nodes hereafter) having almost unlimited energy supplies such as

solar cells. Communication devices installed on a mobile vehicle and powered by inside










alternators are other examples of such P-nodes. Usually, N, is much smaller than Nr. We

assume that all the nodes are able to generate traffic or forward packets for others no matter

when they are at rest or in motion. Intuitively, since P-nodes have relatively infinite en-

ergy reservoir as opposed to battery-powered B-nodes, they should be utilized as much as

possible to save the scarce resources of B-nodes and thus prolong the whole network life-

time. For example, a packet should be forwarded to a P-node whenever possible if energy

savings can be expected. On the other hand, we should reduce the use of B-nodes if we

cannot completely avoid using them. How to realize this simple rationale, however, is by

no means a easy task.

In this dissertation, we intend to address the issue of energy conservation from the

viewpoint of node movement. In what follows, we first present a general mobility model

that is used to characterize nodes' movement patterns. We then introduce the resource-

aware movement problem in its general form with the consumption of energy resources as

the sole optimization objective.

3.3.1 General Mobility Model (GMM)

During an observation period T, we assume that there are some designated locations

that any node i, be it a P-node or B-node, should stop by at some designated time in-

stances. For example, a student carrying a mobile device may appear in the classroom

during schooltime while in the cafeteria during lunchtime. Let Ji denote the size of the

ordered list of locations node i should visit during T, which might be different for each

individual node. We denote by 1., -- (j) (0 < j < Ji) the jth location that node i should

stop by and by ti(j) the required time instance. Then 1., -- (0) denotes the starting point of

node i, and posi(Ji 1 ) denotes the location of its last stop during T. We will also call as an

epoch [70] the time duration from one node leaving the current stop until it reaches the next

stop henceforth. Whenever arriving at some designated location at the specified instance,

each node is assumed to pause for a while according to concrete application requirements.

Let pausei(j) indicate the time node i spends at 1., -- (j).









Based on the above definitions, the order list {ye- (j), ti(j), pausei(j),0 < j <

Ji} can well characterize the itinerary of node i during the observation period T. Let

posi(j)posi(j +t 1) denote the path travelled by node i from its jth stop to the (j 1t )th

stop. Provide that each node travels at a constant speed between two consecutive stops,

the travelling speed of node i from pos(i, j) to pos(i, j +t 1) is 1..1. tij)pouse(j ))

Notice that node i can follow potentially many different paths, for example, a straight one

or a zigzag one or even a tortuous one, as long as the time constraint is satisfied, that is,

it can reach 1., -- (j +t 1) at the time instance ti(j). However, once the path between two

consecutive stops is determined, the velocity of the node between these two stops is de-

termined and fixed. Therefore, once the paths between all pairs of consecutive stops are

determined, the movement pattern of a node during T is also determined.

The general mobility model (GMM) described above bears both similarities and dif-

ferences with the random waypoint model (RWM), which is the most commonly-used mo-

bility model in simulating MANET protocols [58, 70]. Both models are characterized by a

collection of locations of next stops, travelling speeds, and travelling time. Different from

GMM, RWM requires a node to first choose the location of its next stop and the travelling

speed, which leads to the determination of the travelling time. Our GMM actually does

the opposite by first determining the next visited location and the travelling time so as to

determine the travelling speed. The biggest difference, however, is that, in RWM nodes

always travel along the straight paths connecting two consecutive stops, while in GMM,

nodes can travel along arbitrary paths as long as they do not violate the time requirements,

that is, they should arrive at the designated stops at the specified time instances.

We believe that our GMM outperforms RWM in reflecting some practical scenarios,

which can be seen from a simple example in daily life. Suppose a student carrying a mobile

device should be in the classroom at fixed time everyday. He/she might have several options

of arriving in time at the classroom from his/her apartment: by foot through the shortest

path with the longest time, by bicycle through the second shortest path with the second










longest time, or by bus through the longest path but with the shortest time. Obviously,

RWM fails to model this case, but our GMM can.

3.3.2 Resource Aware Movement

As mentioned before, in the general mobility model there might be potentially many

different paths between any two consecutive stops. Define PATH(I., -- (0),1~ 4. (J1 1))

as node i's path set which is the concatenation of all the paths 1., -- (j J., -- (j +t 1). In this

dissertation, we are interested in finding the optimal path sets for all the nodes such that

the total energy consumption for communications by all the nodes during the observation

period T is minimized (the objective function), while all the ordered lists of visited loca-

tions and the corresponding time instances should not be violated (the constraints). To help

better understand the importance of this problem, we utilize the movement of a single node

between two consecutive stops as an example. As shown in Fig. 3-1, suppose an B-node

A should move from the current location 1 to the next location 2. It can choose the shortest

straight path (the dashed one) as it does in the random waypoint model. However, con-

sidering that node A may forward or generate packets destined for other nodes during the

movement process, the shortest straight path is not necessarily the best one for achieving

the system-wide energy efficiency. Instead, the dotted and solid paths are much better can-

didates through which node A can take advantage of more P-nodes by forwarding to the

encountered P-nodes the packets destined for other nodes and letting them finish the rest of

the task. Due to this reason, we call this problem the resource-aware movement problem in

that nodes now are moving with the system-wide resource (energy) consumption in mind

instead of moving blindly as before.

The general resource-aware movement problem itself is far too complicated to be

solvable. To render it tractable, we make some approximation and decouple it into two

relatively simpler subproblems: the Waterhunter Movement problem and the Firehunter

Movement problem. In the former, we assume that only B-nodes are capable of moving










Transmission
Range

I





eP-node () R-node

Figure 3-1: Multiple paths between two consecutive stops.


and all the P-nodes are stationary whose locations are known a priori to B-nodes. By con-

trast, in the latter, we assume that all the B-nodes are stationary and only P-nodes are able

to move.

Waterhunter Movement: In the network with N, stationary P-nodes and N, mobile

B-nodes, given all the order lists of {posi(j), ti(j), pausei(j), O < j < Ji}, the

Waterhunter Movement Probleml is to determine the optimal travelling path set for

each B-node such that the total energy consumption of the whole network during T

is minimized.

Firehunter Movement: In the network with N, mobile P-nodes and N, stationary

B-nodes, given all the order lists of {posi(j), ti(j), pausei(j), O < j < Ji}, the

Firehunter Movement Problem2 is to determine the optimal travelling path set for

each P-node such that the total energy consumption of the whole network during T

is minimized.




SIf we compare energy resources to "water", the movement of B-nodes is similar to
the behavior of water-hunters who are always looking for "fountains" (P-nodes), hence the
name.

2 If We COmpare energy resources to "water", the movement of P-nodes is similar to
the behavior of fire-hunters who are always looking for places "on fire" or "lack of water"
(B-nodes), hence the name.










The Waterhunter Movement problem is similar to the determination of people's opti-

mal travelling plans when lots of airline hubs and stops are available. On the contrary, the

Firehunter Movement problem bears similarity with the planning of hub locations when

airline companies build their global transportation networks with the purpose of providing

more convenience for passengers while reducing the overall system cost. Both problems

are pretty interesting and worthy of rigorous study. However, due to the space limitation,

we focus on finding a nearly optimal solution to the Waterhunter Movement problem in

this dissertation. Our investigation on the Firehunter Movement problem and the general

Resource-Aware Movement problem will be the future work.

3.4 Waterhunter Movement

In this section, we first present a simplified version of the Waterhunter Movement

problem, which is further reduced to a NP-complete distance-constrained least-cost (DCLC)

problem. We then present an efficient heuristic solution. Finally, we propose a routing

delay differentiation mechanism to utilize the benefits resulting from the resource-aware

movement.

3.4.1 Simplified Waterhunter Movement Problem

In the Waterhunter Movement problem, we assume that all the N, nodes are stationary

during the observation period T and are willing to forward packets for other less powerful

B-nodes. For simplicity, we do not dwell on how to place P-nodes to attain the optimal

system performance, which is believed to be a challenging problem itself and is currently

under investigation. Instead, we assume that each B-node knows the locations of all the

P-nodes and its own location at any time, and can as well adjust its moving direction at

will. For the time being, we assume here that all the P-nodes and B-nodes have the same

transmission range TR. We will discuss the case that P-nodes have the larger transmis-

sion range than B-nodes in Section 3.4.4. It is worth pointing out that the findings in this

dissertation can be easily extended to the case that each node has individual transmission

range.










The original Waterhunter Movement problem aims at minimizing the total energy con-

sumption of the whole network during the observation period T, which is a global optimiza-

tion problem and still too hard to be solvable. To make it tractable, we have to make some

approximations to get a suboptimal solution. We assume that a B-node moves at a constant

speed in one epoch, that is, between two consecutive stops, and the maximum speed it can

take is a system-wide value speedmax. Therefore, the longest path node i can travel in one

epoch is Im=, = speedmax x (ti(j +t 1) 14(j) pausei(j)). To achieve the system-wide

goal of energy conservation, instead of moving along the straight path connecting two con-

secutive stops 1.. -- (j) and posi(j +t 1), node i may travel along a resource-aware path with

the purpose of letting the encountered P-nodes forward on behalf of it as many as possible

packets destined for other nodes. For simplicity, we assume that when moving towards

a P-node, node i always goes along the straight path connecting the destined P-node and

itself. Therefore, 1., -- (j J., -- (j +t 1) is a set of zigzag straight paths if there exist multi-

ple P-nodes. Notice that the simplified target now is to find an optimal path set for each

individual node to minimize its total energy consumption during the observation period T

instead of that of the whole network. We intend to utilize the solutions to this local op-

timization problem to approximate the ones to the original global optimization problem,

which is believed to be too complicated to be tractable.

Normally, when moving between two consecutive stops posse (j) and posse (j +t 1), node

i may have several potential P-nodes to utilize. It is, however, usually unwise for node

i to pass by each of them. The reason is that, the longer path node i takes, the faster

speed it should move at, as described in the aforementioned general mobility model. It

is well-known that a faster movement speed may cause some undesirable problems such

as the instability of routing paths and the drop of packets. Therefore, some rules should

be designed to guide each B-node in deciding which P-nodes and in what order it should

pass through between two consecutive stops. A simple rule would be to only consider as

candidates the P-nodes whose distance from the direct link between any two consecutive















S~


Figure 3-2: An exemplary complete graph.


steps are no more than a threshold p. For example, p can equal 1.5 x TR, where TR is the

transmission range of each node.

With node i as an example, we can put the simplified Waterhunter Movement problem

in another way: given a source (i's current stop), a destination (i's next stop), and some

available intermediate P-nodes, and the path length constraint Imax, find a path as energy-

efficient as possible from the source to the destination, which is might be either the direct

link connecting the source and destination or a zigzag path through multiple P-nodes. Fig.

3-2 depicts such a topology, where a rectangular area, called the p-bounded rectangular

area hereafter, is formed such that only the P-nodes residing in this area are considered

as valid candidates. In addition, each link is of the forward direction from the source s

to the destination d simply because travelling backward is energy inefficient. We assign

to each link two weights, of which one represents the physical distance between two link

ends and the other indicates the virtual energy cost (defined shortly) incurred by choosing

this link. The simplified Waterhunter Movement problem can be boiled down to a distance-

constrained least-cost (DCLC) [71] routing problem which is formally defined as follows.



Consider a directed network that can be modelled as a complete graph G = (V, E),

where V is the set of vertices consisting of the source, the destination, and all the valid

candidate P-nodes, and E is the set of edges connecting each pair of nodes. V can be

further divided into two subsets, namely, ZA including the source s and destination d, and P

containing all the P-nodes. In addition, each edge e EE represents the movement from the









Transmission
Range




CI





SP-node O R-node

Figure 3-3: An exemplary resource-aware movement.


tail node to the head node. Let R*+ denote the set of non-negative real numbers. Each edge

e EE is associated with two non-negative functions: a distance function dist(e) : E R W

representing the physical distance between the end nodes of e and an energy cost function

cost(e) : E R +U {0}. More specifically, for a given edge e(vei, vy), its energy cost is
defined as

cos t(e(vi, vj))

f (dist(vi, vj)) g(dist(vi, vj)) vi = S, vj = d

f (dist(vi, vj)) g(dist(vi, vj) TR) vi = s, vj E p or vj = d, vi E p

f (dist(vi, vj)) g(dist(vi, vj) 2 x TR) vi, vj E p.

Here f is the cost, such as gas, fuel or other types of resources, required for the mechanical

movement ; g is used to reflect the cost for communications. g can be any non-decreasing

function that converts a given distance value into a non-negative cost, for example,



9(1~0 x < 0.





SIn this chapter, for simplicity, we assume that people on foot carry the communities
devices and we do not take the cost for the mechanical movement into account, that is,
f () = 0.










The motivation for the above definition of the edge energy cost is as follows. Whenever a

B-node moves into the transmission range TR of a P-node, it is capable of forwarding to

the P-node packets destined for other nodes so as to conserve energy. At one extreme, if

a B-node moves along an edge not (partially) covered by any P-node, all the packets from

this B-node would be forwarded to other energy-constrained B-nodes, which is the most

unfavorable situation. At the other extreme, if a B-node moves along an edge completely

covered by one or several P-nodes, all the packets from this node could be forwarded to the

P-node(s), which is the most desirable situation. Notice that the energy cost function given

above can well capture this effect. Though there might exist other meaningful metrics, we

believe the chosen one is very simple and tractable.

We also define the non-negative delay and cost functions for any path p as


dist(p) = dist(e)
etp

and

cost(p) = costle).
etp
Given the above definitions, the DCLC routing problem is to find a path p from s to d

such that min~cost(p), p E Pd} is achieved, where Pd is the set of all feasible paths from

s to d satisfying the distance constraint Imax, that is, dist(p) < Imax. Moreover, we define

Pra(s, d) as the path with the least distance from s to d, and Ple(s, d) as the path with the

least cost from s to d. Apparently, with the above definition of dist(e), Pla(S, d) is the

straight path directly connecting s and d.

It has been shown in [72] that the DCLC routing problem is NP-complete even for

undirected networks. In the following section we will propose an efficient heuristic algo-

rithm to provide a suboptimal solution to this DCLC problem and hence to the original

Waterhunter Movement problem.









Table 3-1: RAM: resource aware movement

1. Determine the candidate P-node set CHS
in the p-bounded rectangular area;
2. Construct a complete graph G with virtual
nodes s and d, and all the nodes in CHS;
3. Label each link e in G with cost(e) and
dist(e);
4. Find the DCLC path Pacle by calling
RAM-DCLC(G, s, t, Ims,);
5. Determine the travelling speed along Pacle
as seeddle = dist(Pacic)
as seeduic =t(i,j+1) -t(i~j) -pause(i,j) '
6. Move along Pacle at a speed of speeddcic;


3.4.2 RAM-DCLC Algorithm

As mentioned before, we assume that a B-node, say i, is aware of its own itinerary

(1** (j), ti(j), pausei(j), O < j < Ji} and the locations of all the P-nodes during the

observation period T. The procedure of node i's resource-aware movement from the cur-

rent location 1y. -- (j) to the next location 1y...- (j) is summarized in Table 3-1. Node i

first needs to determine the candidate P-nodes in the p-bounded rectangular area and

then constructs a complete graph like the one in Fig. 3-2, consisting of the vertices s

(a virtual node at 1.. -- (j)), d (a virtual node at 1., -- (j +t 1)), and all the found candi-

date P-nodes. It then proceeds to calculate the distance and the energy cost for each

link and finally generates the weighted graph G. The next step is to call the process

RAIM-DCLC given in Table 3-2 to get the DCLC path Pacle whose length is bounded by

Imaz = speedmax x (ti(j +t 1) ti(j) pausei(j)). It then moves towards 1., -- (j +t 1) at

a constant speed of speetddcic = ti"+1)(t, d- sei') along the found Pace,. Upon reach-

ing 1y. .- (j +t 1), node i pauses for a period pausei(j +t 1). Following the previous pro-

cess, it can then move towards the next stop pausei(j +t 2) until all the required stops

{posi (j), O < j < Ji} during the observation period T are visited.




















1. For each node I1 in G, find the Plc(I, ,d)
and Ple('r -, d) and
their respective next hops nid(Pic( -.d))
and nid(P~d(r I, d));
2. distSoFar = 0; Pacle = S; ThisNode = s;
3. while (ThisNode / d) do
4. if ((dist(Ple(ThisNode, d))+
distSoFar) < Imom) then
5. vu = nid(Ple(ThisNode, d));
6. distSoFar = distSoFar+
dist (This Node, u) ;
7. Pacle = Pacle + {v);
8. ThisNode = vI;
9. else
10. for each neighboring node w ( Pacle
do
11. calculate weight(ThisNode, w);
12. end for
13. v1 = extract(ThisNode);
14. distSoFar = distSoFar+
dist (This Node, u) ;
15. Pacle = Pacle + {v);
16. ThisNode = vu;
cp:ref35 17. end if
18. end while
19. Return Pacle


Table 3-2: RAM-DCLC: a DCLC routing algorithm for the Waterhunter Movement









The proposed DCLC algorithm RAIM-DCLC is summarized in Table 3-2, in which the

weight() function is defined as follows:


Cost (vi, vj) cost' (vy d) cond (1)

and

cost'(vj, d) = cotPev,))...1
cost(Pza(v ,d)) 0.w.
where

cond(1) = distSoFar dist(vi, vj) dist(P~a(vj, d)) < lmax

and

... c!ol(1) = distSoFar dist(vi, vj) dist(Ple(vj, d)) < lma,

The function extract() is used to choose the node, say w, whose weight(vei, w) is the

minimum one among all the neighboring nodes of ThisNode. If more than one node have

the same minimum value, it chooses the one with the smallest distSoFar +t dist(ve, vyj)+

dist(Pza(vj, d)).
In the RAIM-DCLC algorithm, for each node vk~ in G', the Bellman-Ford or Dijkstra

shortest-path algorithm can be used to find the PIe(r -, d) and Pra(r-, d) and their respective

next hops nid(Ple d)) and nid(Pza(r-, d)). Since the optimization objective is the path

cost, at each intermediate node vI, RAIM-DCLC always chooses the next hop w with mini-

mum cost(vl, w) + cost'(w, d) while not violating the distance constraint Imax. RAIM-DCLC

is able to find a feasible path satisfying the distance constant while keeping the path cost as

small as possible. In particular, we have the following theorems for this algorithm 4

Theorem 1: RAIM-DCLC can always find a feasible path from a source s to a destina-

tion d satisfying the given distance constraint Imax if such feasible paths exist.


4 The correctness of these theorems can be justified following the proof in [73].








64




Pnds (S 0) .(S 858.4)
SStops
250 -RAM1 3
--RWM 1
200 (Stop, Arrival time)
E (S2, 247.8)"
150- 11


( S 4 6 2 5 8 )- ,8~


.S,387. 1)
U200 400 600 800 1000 1200 1400 1600
x (m)



Figure 3-4: Resource-aware movement (RAM) vs. random waypoint movement. A B-
node should consecutively visit S1-S2-S3-S4-S5 and the solid lines are labelled by the se-
quences they were passed through.


Theorem 2: The path found by RAIM-DCLC is loop-free.

Theorem 3: RAIM-DCLC always terminates in finite time. Now we utilize the example

given in Fig. 3-3 to illustrate the resource-aware movement process using the proposed

R 4M DCLC algorithm. Suppose node A intends to move from its current location to

the location where A' resides according to its itinerary. In this example p is set to 1.5 x TR

so that there are four candidate P-nodes. Based on the output of RAM~ DCLC, node

A should move along the DCLC path denoted by the solid line instead of the straight path

denoted by the dash line. In this way, energy savings can be expected by forwarding to

the two P-nodes the packets it carries for other nodes and letting the P-nodes finish the rest

transmissions (either single-hop or multi-hop) on behalf it.

Fig. 3-4 compares a B-node's resource-aware movement trail and its random way-

point movement trail with 8 P-nodes in a 1500 x 300m2 field. The data used were gen-

erated using OPNET [57] and the pause time of the B-node was 120s. In addition, the

transmission range of each node was set to 250m. It is clear that our proposed resource-

aware movement strategy enables the B-node to have more opportunities of approaching

and utilizing the P-nodes as compared to the random waypoint movement.










3.4.3 A Routing Delay Differentiation Mechanism

The resource-aware movement strategy only enables B-nodes to move close to P-

nodes by taking the physical DCLC paths, but cannot guarantee a packet will be forwarded

from a B-node to a P-node even when they are close to each other. As a result, we need

to incorporate such nodal information as device types and residual energy into the routing

metric such that the P-nodes can be naturally selected onto the routing paths of the B-

nodes. For this purpose, we propose to map the residual energy information into the random

jitter delay for which each node has to wait before propagating a route request in a regular

MANET routing protocol such as DSR [70]. The rational here is that the less residual

energy a node has, the longer delay a routing request packet should experience at this node.

In this way, the nodes with less residual energy would be less utilized, while the nodes with

more residual energy such as P-nodes would be more utilized.

Below we illustrate the delay differentiation mechanism using DSR [70] as the under-

lying routing protocol, though our mechanism can as well be applied to other MANET rout-

ing protocols. Similar to the interframe spacing (IFS) differentiation mechanism adopted

in the IEEE 802.11e [74], we first define the Delay Diffi I~i~illaio as 6 x E,3/Et, where 6

is a delay control parameter, and Ec, and Et are the residual energy at time instance 0 and

t, respectively. When receiving a route request, each node should first compute the delay

it needs to wait for before further propagating the route request. Since the P-nodes have

relatively unlimited residual energy as compared to the B-nodes, that is, Ec, a Et, they

would forward route requests much more quickly than the B-nodes. Since the destination

usually only replies to the first arrived request, the P-nodes with powerful energy resources

would have higher chances of being chosen onto the routing paths and thus being involved

in more data forwarding activities than the B-nodes with less energy. With this delay dif-

ferentiation mechanism, the benefits resulting from the resource-aware movement can be

better utilized. The nice feature of our mechanism is that it is very easy to implement.










3.4.4 Incorporate RAM into DELAR

In our previous discussion, we assume that the P-nodes have the same transmission

range TR as the B-nodes. However, in more practical heterogeneous MANETs, P-nodes

may have greater transmission capabilities than the B-nodes. To make use of such pow-

erful P-nodes especially when P-nodes can adjust their transmission power to cover larger

range than B-nodes, in our previous work DELAR, we proposed a energy efficient relaying

framework as a joint design of scheduling, routing and power control to efficiently utilize

P-nodes to conserve energy. Since RAM and DELAR utilize powerful nodes to conserve

energy from totally orthogonal perspectives, RAM can be directly incorporated into the

DELAR scheme. With the help of DELAR, RAM can further improves the energy effi-

ciency when P-nodes and B-nodes have different transmission capabilities.

3.5 Performance Evaluation

In this section, we use simulations to evaluate the impact of the proposed schemes on

the energy conservation and other system performance factors.

3.5.1 Simulation Setup

We implemented the resource-aware mobility model, the routing delay differentiated

mechanism, and DELAR (cf. Chapter 2) in OPNET [57].

We simulated a network with N, B-nodes and Nz, P-nodes in a 1500 x 300 field,

where N, = 46 and N,, = 4. All the B-nodes were capable of moving in the field, while

all the P-nodes were fixed. Though a careful deployment of P-nodes may improve the

system performance [75], we simulated a worse scenario that the P-nodes were randomly

deployed in the field. The transmission range of the B-nodes was 250m. For the P-nodes,

we simulated two cases in which the P-nodes had the transmission ranges 250m and 500m,

respectively.

The energy consumption for the B-nodes followed the linear energy model proposed

in [76]: ener yy = m x len ythtb, where m is an incremental cost of each operation, b is the







67

Table 3-3: Energy consumption parameters.

Symbol Value Unit
mend 1.89 uW-sec/byte
bsend 246 uW-sec
meec, 0.49 uW-sec/byte
bree, 56.1 uW-sec
bsenacet 120 uW-sec
breevenl 29 uW-sec


fixed cost of each operation, and length is the size of the frame sent/received5 The (m, b)

values provided in [76] were summarized in Table 3-3 and used in all the calculations.

We intended to compare the proposed resource-aware mobility model (denoted by

RAM) with the modified random waypoint model (denoted by RWM) presented in [58],

which can guarantee the convergence of average nodal speed throughout the simulation

time. For this purpose, we first ran the simulations using RWM and recorded the stops,

the starting/arrival time instances, the moving directions, and the movement speeds of

all the movement epochs. We then used this movement profile to generate the itinerary

{posi (j), ti(j), pausei(j), O < j < Ji} for each node such that in both models each node

would stop by the same stops at the same time instances, but follow totally different move-

ment trails and take different movement speeds. Both models had the same maximum

speed 20 m/s and we adjusted nodal pause time to vary the network mobility. The traffic

used were 20 CBR connections with randomly selected source-destination pairs. All the

data packets were 64 bytes and were sent a speed of 4 packets/second. Each simulation was

executed for 15 simulated minutes and each data point represents an average of ten runs

with identical traffic models, but differently generated mobility scenarios.




5 For the A-MAC control frames P-RTS/P-CTS/P-ACK, the fixed costs bsenactl and
breevenl were used because they have the similar size.










3.5.2 Simulation Results

We compared RAM with RWM in terms of the commonly used metrics including

packet delivery ratio, average packet end-to-end delay, average packet energy consump-

tion, and average routing overhead. Motivated by the small-world phenomenon [77], we

used two additional metrics, average path length and average clustering coefficient, which

are two defining characteristics of small-world networks. The former means the average

number of hops a packet may travel through, while the latter indicates the connectivity of

an average neighborhood in the network, defined as the average node degree divided by

the network size [77]. The simulation results are presented in Fig. 3-5, where RAM-1

indicates the case that the P-nodes and B-nodes have the same transmission range 250m

and RAM-2 denotes the case that the P-nodes have a larger transmission range 500m.

Fig. 3-5(a) and 3-5(b) compare RAM and RWM with regard to average path length

and average clustering coefficient, respectively. We can see that RAM can shorten the

average path length and increase the clustering coefficient as compared to RWM. That is

because in RAM the B-nodes are always trying to move towards some P-nodes between

two consecutive stops. Such behaviors would effectively bring more B-nodes to the vicin-

ity of P-nodes, leading to shorter paths and larger clustering coefficients. In some sense,

such resource-aware movement creates a "small-world" network, which results in some

performance gains shown below. In addition, since the P-nodes in RAM-2 have a larger

transmission range and thus have more neighbors that those in RAM-1, we can observe

that RAM-2 further reduces the average path length and increases the average clustering

coefficient.

Fig. 3-5(c) compares the average packet delivery ratios of RAM and RWM, which is

defined as the ratio of delivered data packets to those generated by the sources. As we can

see, the PDR of RAM-1 or RAM-2 is always higher than that of RWM. This result is of

no surprise since the shorter average path length implies the network-wide less traffic load

and less packet drops due to the MAC-layer contention and channel errors, all of which





























































0 100 200 300 400 500 600 700 800 900
Pause t me (s)


(d) Average packet end-to-end delay.



-4- RAM-1
-8- RAM-2



-


0c-_~


5i


X-


100 200 300 400 500 600 700 800 9000
Pause tme (s)


(a) Average path length.





-a RAM-1
-8 RAM-2


(b) Average clustering coefficient.









h-RA-


~0014
0 012
0 01


S0- 006


t


0 100 200 300 400 500 600 700 800 900
Pause tme (s)


(c) Packet delivery ratio.






RAM-1
X-* I,- RAM-2


100 200 300 400 500 600 700 800 900 0 100 200 300 400 500 600 700 800 900
Pause t me (s) Pause t me (s)


(e) Average energy consumption. (f) Average routing overhead.


Figure 3-5: Simulation results.










would contribute to the increase of the PDR. Due to the same reason, RAM-2 demonstrates

a higher PDR than RAM-1.

Figure 3-5(d) depicts the comparison of average end-to-end packet delay, defined as

the time duration from when a packet is generated till it is received by the destination.

The shown advantage of RAM over RWM mainly results from the aforementioned shorter

average path length and higher clustering coefficient. Again, RAM-2 outperforms RAM-1

in reducing average packet delay because of the shorter average path length.

Figure 3-5(e) shows average energy consumption, defined as the total energy consump-

tion for transmitting and receiving all data and routing packets divided by the number of

delivered packets. Apparently, our RAM can conserve a significant amount of energy as

compared to RWM because the B-nodes are always moving along the paths through which

the P-nodes can be utilized as much as possible. Since the P-nodes in RAM-2 have a larger

transmission range, statically less B-nodes are involved in packet transmissions and thus

RAM-2 can help the system conserve more energy than RAM-1.

Figure 3-5(f) demonstrates average routing overhead, defined as the average number

of routing packets involved in delivering 100 data packets. As we can see, our RAM has

smaller routing overhead than RWM. That is because in RAM packets can be forwarded to

their destinations through shorter paths in shorter time, thus fewer routing errors occur.

To summarize, the proposed resource-aware movement strategy has many significant

and positive impacts on the system performance. It makes the network more like a "small-

world" network with shorter average path length and higher clustering coefficient. This

results in improved packet delivery ratio, shortened end-to-end delays, and most impor-

tantly, much better energy conservation. Therefore, the combination of node mobility and

heterogeneity is a valid means to address the energy conservation issue in MANETs. The

results also suggest that making the network a small-world network may have a lot of pos-

itive effects on the system performance and it deserves further investigation.










3.6 Summary

In this chapter, we studied the energy conservation problem from the new perspective

of node mobility, that is, by analyzing the impact of node movement on the system-wide

energy conservation. We proposed a novel resource-aware movement strategy to take full

advantage of some powerful nodes in heterogeneous mobile ad hoc networks. We then for-

mulated the resource-aware movement as a NP-complete distance-constrained least-cost

(DCLC) routing problem and proposed an efficient heuristic solution. Moreover, we pro-

posed a simple yet effective routing delay differentiation mechanism to virtually utilize

the benefits from the resource-aware movement. In addition, the proposed resource aware

movement strategy can be incorporated into other energy conservation schemes, for exam-

ple, DELAR, to further improve the energy efficiency. We used extensive simulations to

show the effectiveness of the proposed scheme.















CHAPTER 4
SUPPORT DIFFERENTIATED SERVICES IN MOBILE AD HOC NETWORKS

4.1 Introduction

In the literature there are proposals addressing the issue of quality of service in wired

networks, however, these proposals may not be feasible in the wireless counterpart if we

do not modify them for the wireless environments. The system dynamics [18] of multi-hop

mobile ad hoc networks, such as time-varying and error-prone wireless links, dynamic and

limited bandwidth, time-varying traffic pattern and user location, and energy constraints,

pose new challenges that do not exist in wired networks. To conquer these challenges, in re-

cent years, many researchers advocate a cross-layer design philosophy to develop protocols

and applications for MANETs. This is a departure from the traditional layered design for

the Intemet. Though the cross-layer design philosophy might not be an optimal solution,

it does provide us new network implementations that may better support the amalgamation

of user services and QoS requirements [19].

To efficiently handle heterogeneous traffic over wireless links, we need to address

two problems. The first is to handle reliable mobile communications in MANETs. This

problem has been extensively studied in recent years, and many proposed routing protocols

such as DSDV [7], DSR [9], and AODV [10], and medium access control mechanisms

such as MACAW [3], FAMA [4], and IEEE 802.11 [5], aim to achieve efficient reliable

communications. The other problem is to provide QoS provisioning for heterogeneous

traffic with different quality-of-service (QoS) requirements in terms of BER, throughput,

and delay. Since the channel bandwidth in wireless environments is limited, one strategy to

support QoS is to set up some kind of priority scheme or service differentiation mechanism

[78][79], under which delay-sensitive traffic has higher priority to access the channel over

less time-critical traffic.










In the current literature, many scheduling mechanisms for wireless networks are pro-

posed for this purpose, though most of them are not directly designed for MANETs. In

general, these scheduling mechanisms all attempt to combat the channel impairments and

to support heterogeneous traffic with the following goals: providing high wireless chan-

nel utilization, long-term fairness, bandwidth guarantees and delay bounds for flows with

error-free links or links with sporadic errors [80]. However, these algorithms may not be

practical to be implemented in MANETs. Actually, it is hard, if not impossible, to achieve

those goals simultaneously because of their conflicting nature. For example, there is a

tradeoff between the throughput and fairness or so-called inter-class effects [81] among

traffic with different priorities. Without any precautionary measures, this conflict may lead

to bandwidth starvation for low-priority traffic when the high-priority traffic load is high.

Meanwhile, most of these scheduling mechanisms are suitable for the reservation-based

MAC protocols, especially for those designed for cell-structured wireless networks. In

networks with contention-based MAC protocols such as IEEE 802.11 [5], the reservation-

based scheduling mechanisms may not be applicable because it is not easy for a node to

reserve resource mna contention manner.

In this dissertation, we attempt to avoid the conventional scheduling approach, and

propose a novel scheme called Courtesy Piggybacking (CP) to alleviate the conflict be-

tween throughput and fairness. The basic idea of CP is to let the high-priority traffic help

the low-priority traffic by sharing unused residual bandwidth with courtesy. Our scheme

closely follows the cross-layer design principle and exploits the system dynamics as much




SWe only consider the fairness problem between different classes of traffic, for example,
each class of service should be allocated some bandwidth rather than being completely
starved; while the fairness problem between different nodes, for example, each node should
have fair opportunity to access the channel in the short or long term, is out of the scope of
this dissertation.










as possible; that is, we effectively employ the dynamic channel conditions and the result-

ing dynamic bandwidth, and the dynamic characteristics of the heterogeneous traffic. Note

that not only is our scheme suitable for multi-hop mobile ad hoc networks with underlying

contention-based MAC protocols, but also it is applicable to those with reservation-based

or hybrid MAC protocols. Meanwhile, our scheme is shown to be easily implemented.

The rest of the chapter is organized as follows. We discuss some related work in

Section 4.2. In Section 4.3, we show the motivation of our proposed scheme. In Section 4.4,

we discuss the relationship between the SNR and the optimal packet length, and come up

with a Finite State Markov Chain channel model based on the packet length. Our Courtesy

Piggybacking scheme is described in Section 4.5. We present some preliminary analytical

results in Section 4.6 and evaluate our scheme with extensive simulation in Section 4.7.

Finally, we conclude the chapter in Section 4.8.

4.2 Related Work

As we mentioned in Section 4.1, scheduling is one promising way to support hetero-

gonous traffic with different QoS requirements. For scheduling mechanisms, throughput

and fairness are two main objectives to be met through bandwidth allocation with admis-

sion control and congestion control. Many scheduling algorithms such as fair queuing

scheduling [82], and virtual clock [83] are capable of providing certain QoS guarantee for

wireline networks, and many scheduling algorithms such as IWFQ [84], CIF-Q [85], CS-

DPS [86], and CSDPS + CBQ [87] are proposed for the wireless networks, especially for

wireless cellular networks. However, little progress has been made along this direction in

wireless mobile ad hoc networks with underlying contention-based MAC protocols. CS-

DPS and its improved version CSDPS+ CBQ are two of scheduling mechanisms that may

be applicable to the ad hoc networks with contention-based MAC protocols. In CSDPS,

packets to be transmitted to the same receiver are queued in the same queue and are served

in an FIFO fashion. At a node, the different queues are served according to some poli-

cies such as round robin, earliest timestamp first, or longest queue first. The basic idea










of CSDPS is as follows: when the link towards a receiver is bad, the node should defer

the transmission of packets in the queue corresponding to that receiver. With CSDPS, it is

easy to alleviate the head of line (HOL) problem when single FIFO queue is used. Since

CSDPS makes use of the channel state information, it can achieve high data throughput

and channel utilization. However, it does not address the fairness issue. To improve the

fairness in CSDPS, class-based queuing (CBQ) [88] is used together with the CSDPS. By

using CBQ, a hierarchical channel-sharing mechanism, it can achieve certain fairness, and

ensure that different traffic classes can share the overall bandwidth, while maintaining the

features of CSDPS to deal with the channel variations. Unfortunately, this scheme is also

complicated in keeping track of the amount of service each class has been served. Efficient

and less expensive mechanisms are very desirable to alleviate the conflict of throughput

and fairness in MANETs. More and comprehensive materials on scheduling can be found

in [80]. Besides, some QoS adaptive schemes such as SWAN [16] and Havana [89] are also

available in the literature. These schemes adaptively perform admission control and rate

control according to the user QoS requirements and channel states.

The main reason leading to the conflict between throughput and fairness is the lim-

ited bandwidth of the wireless link. If the system can provide plenty of bandwidth, the

conflict problem would not be so significant. Recently, many adaptive transmission tech-

niques are proposed to exploit the channel dynamics to provide more bandwidth. These

schemes can adaptively adjust the parameters such as modulation level and symbol rate to

maintain an acceptable BER without wasting much bandwidth. Koutsopoulous and Tassi-

ulas proposed to integrate adaptive transmission techniques, resource allocation and power

control for TDMA/TDD system so that higher modulation levels can be assigned to users

in good channels to enhance the throughput, while power control can be used to reduce the

interference and increase the system capacity [90]. In addition to these schemes proposed

for wireless cellular networks, some rate-adaptive schemes are also proposed to improve

the system throughput in WLANs. Holland et al. proposed a rate adaptive MAC protocol










called RBAR, which uses the RTS/CTS to exchange the channel state information and the

optimal rate on a per-packet basis [91]. Unfortunately, this scheme needs to make some

modifications to the IEEE 802.11 MAC protocols. To avoid this modification, Chevillat

et al. proposed a scheme to select the optimal rate only with the local information at the

transmitter [92]. This scheme is based on the history of attempted transmissions. It uses

one successful transmission count and one failed transmission count to indicate the chan-

nel state and to determine the optimal rate the transmitter can use. For IEEE 802.11 MAC

protocols, adaptive fragmentation schemes can also be designed with the rate adaptation to

enhance the system throughput [93][94][95].

For all the scheduling mechanisms and other channel-dependent schemes, including

our Courtesy Piggybacking scheme, designed for wireless networks, they all have to mon-

itor the channel quality based on the symbol error rate, bit error rate, and receiver signal

strength. The more accurate the channel information is, the more benefits these schemes

can bring to the system design. In general, the channel estimation can be performed by

the sender or by the receiver. Since the channel information used in all channel-dependent

schemes is the one seen by the receiver, the receiver-based channel estimation is more at-

tractive. However, the channel information needs to be sent back to the sender, which is

sometimes costly in terms of the resource used to transmit the channel information, thus

certain performance tradeoff has to be made between estimation accuracy and overhead.

More details about channel quality estimation can be found in [96].

4.3 Motivation

Consider the scenario depicted in Fig. 4-1. In a mountain area, the only way from

Anchorage to Whittier (the access to see the spectacular glacier) is to pass a tunnel near

Portage running through the Chugach Mountain Range (that is, the longest tunnel in North

America the Whittier Timnel in Alaska). The situation is the same from Seward to

Whittier. People have several choices to pass the tunnel: by train (high priority), by car,

by bicycle or on foot (low priority). Only one direction traffic is allowed during one period


















// Tunnel

( S -Road


Figure 4-1: The Whittier Tunnel scenario.


of time. To pass the tunnel, when the train approaches the tunnel, all other traffic stops

and waits until the train passes the tunnel. Often, there is a long traffic line waiting to pass

the tunnel, especially for the direction from W to P when traffic load is high, for example,

during rush hour in the afternoon. In order to quickly pass the tunnel, a better approach

for other transportation users would be to check if there is any free space left in the train.

If there is, these users could ask for permission to ride at a certain cost and according to

some rules, for example, how many free space in terms of basic units is left and what kind

of traffic (priority) the train can accommodate. After passing through the tunnel, the piggy-

backed traffic can get off the train at P and continue on its own way. Of course in the real

situations, when passengers by car, by bicycle or on foot pass through the narrow and dark

tunnel in a sequential manner, the traffic usually moves very slowly for the sake of safety.

Thus it is advisable for cars that have free space to piggyback those passengers by bicycle

or on foot according to some rules to benefit all the traffic.

We can think of these rules as being concerned with the HOW MANY-WHO problem,

that is, how much free space is available and who can enjoy such free space? If we only

consider the free space FS in the train as a function of time, then we could consider the

following scenario as an example: one person would occupy 1 basic space unit, a bike 2

units, and a car 6 units. If we have some predefined objective to meet, then we can de-

sign different piggybacking rules to solve the HOW MANY-WHO problem. For example,










suppose our objective is to maximize the revenue of the train. With different piggyback-

ing costs, for a given FS we can achieve the optimal allocation scheme for the free space

among different traffic: cars, bicycles, and pedestrians.

The above scenario is very similar to multi-hop mobile ad hoc networks supporting

differentiated services. The piggybacking strategy described above motivates us to develop

a more efficient way to alleviate the conflict between throughput and fairness for different

prioritized services. First of all, we need to identify the "free space" in a MANET. Fortu-

nately, we do have two sources that can provide us with such free space. The first one comes

from the time-varying channel conditions. In recent studies such as [90] [93], the MAC and

PHY layers adapt to the channel state by using adaptive transmission schemes to provide

higher data rates when the channel is good. With a higher data rate, the transmission time

for MAC protocol data unit (MPDU) can be shortened, leading to some potential idle time

if the transmitting node does not have further data to transmit. If the IEEE 802.11 MAC is

used, the NAV (Network Allocation Vector) setting may prevent other nodes from using the

medium, even though it is idle (the rule of virtual collision avoidance). This idle period will

be the "free space" and should be more effectively used. The second source comes from

the traffic characteristics. When we look into the traffic patterns and the stochastic traffic

behavior, sometimes the high priority traffic may not have enough data during the reserved

slots in a reservation-based system or their transmission period in a contention-based sys-

tem (for example, a network with IEEE 802.11) to fully utilize the channel capacity. For

example, consider a network with reservation-based MAC protocols. In addition to the

"free space" provided by the channel dynamics, when the packets from one high-priority

flow are not enough to fill the reserved slots, for example, during silent periods for voice

connections, some "free space" can be harvested to piggyback some bits from the queue(s)

with low priorities.

When such free space is available, the next problem would be how to make use of it

to fulfill certain objectives such as fair allocation of bandwidth. While one would think









that it should be used to better support high-priority traffic in the first place, we argue

that it may not necessarily be the case. Rather, the piggybacking rules should be properly

designed in light of specific requirements of various applications. If some delay-sensitive

applications like voice or video-telephony require that their packets get through the network

as quickly as possible, then the free space should be used to meet such needs. On the other

hand, if high-priority traffic does not need more resource than needed, a piggybacking rule

favoring low priority may be more reasonable. For streaming multimedia applications,

as an example, when the QoS requirement of one stream with high priority has already

been met, there is no need to piggyback packets belonging to this stream ahead of the

scheduled time; instead, piggybacking packets from other low-priority streams may be

more beneficial.

In the following sections, we will elaborate more on why free space exists and how

piggybacking can be used to achieve our goal alleviating the conflict between throughput

and fairness for different prioritized services.

4.4 Packet-length-based Channel Model

In the current literature, the time-varying channel is commonly modeled as the well-

known Gilbert-Elliott two-state Markov channel model (Fig. 4-2). Each state in the two-

state Markov chain model represents a binary symmetric channel (BSC). In "Good" state,

the BSC has low crossover probability, P,, and int the "Bad" state, the BSC has high

crossover probability, Pb. The transition probability matrix can be given as:



SPGGPB PGBPB


Given the transition probability, it is easy to determine that the steady state probabilities

are













PG ( Good ; ( Bad ) Pss

0 0G
0P9~ ----- lPb
G B
199 Pbl


Figure 4-2: The Gilbert-E11iott channel model.


We notice that if P, and Pb are set to 0 and 1, respectively, that is, a packet succeeds

with probability 1 in the "Good" state and is lost with probability 1 in the "Bad" state, the

two-state model is reduced to the simplified Gilbert model.

When the channel quality varies dramatically, it is not accurate enough to model the

channel as a two-state Gilbert-E11iott model. In this case, a finite-state Markov channel

(FSMC) [97] can be used. By using the received signal-to-noise-ratio (SNR) as the only

side information, the FSMC provides a mathematically tractable model for time-varying

channel. Let g* denote the received SNR that is proportional to the square of the signal

envelop. Then, for a Rayleigh fading channel, the probability density function of ]* can be

written as




where 1' is the mean of ]* (actually, it is an exponential distribution with mean 1'). In order

to build the finite state Markov chain model, we assume the received SNR remains at a

certain level for the duration of a symbol, and we partition the range of the received SNR

into a finite number of intervals. Let 0 = yo < yl
thresholds. For each interval, we associate it with a state Sp, kc = 0, 1, 2, 3, ..., K 1. The

channel is in the state SI, if g* is in the interval [ys, y*, ]. We know that there is a crossover

probability p for a given SNR y. When BPSK is used, this probability can be written as a

function of ]*:













5000-

S4000-

$3000

S2000-

1000-



y (dB)


Figure 4-3: The optimal packet length (PL) vs. SNR (y ), h=128.




0, 10 1,1 t 2 ,2 k-1,k-1


Figure 4-4: Packet-Length-Based finite-state Markov channel model.


According to [98], for a given crossover probability p, the optimal packet length,

which is a function of p, can be written as


-h ln(1 p)- -4 n1-p+h22
PL =(4.3)
2 In(1 p)

where h is the number of overhead bits per packet. Fig. 4-3 shows the relationship between

the received SNR and the optimal packet length. For a given state Sk, the average optimal

packet length PLk for this state can be derived by using (4.1), (4.2), (4.3) to be






Based on the above analysis we present our packet-length-based FSMC model in Fig.

4-4. We represent each state as the average packet length PLk, which is the packet size for

a transmission in state k. The transition probabilities between different states are denoted










as tis. Further, we can derive the state steady probability for each state as




In practice, we may use different modulation schemes (not necessarily BPSK) in dif-

ferent channel states. Moreover, by properly partitioning the range of the received SNR,

we may obtain the multiplicative relationship between the average optimal packet lengths.

4.5 Courtesy Piggybacking

In this section, we present our Courtesy Piggybacking scheme to alleviate the conflict

between throughput and fairness and to combat the starvation problem for differentiated

services.

4.5.1 System Assumptions

We consider an ad hoc network consisting of n mobile nodes uniformly distributed in

some area. Nodes can communicate with each other directly if they can hear each other

or through other relay nodes in a single broadcast channel. They employ some contention-

based MAC protocols, such as IEEE 802.11, to support their communications. Each node

can generate services with N different priorities destined to other mobile node(s). A node's

mobility follows the random waypoint model [58, 70]. At first, a node stays at a position

for duration of pause_time. After that period, the node chooses a new random position

and moves towards that position at a random speed uniformly distributed in the range from

min~speed to maxspeed. After reaching the new position, the node will stay there for an-

other pause_time. This process will continue for each node until the end of the simulation.

We assume some service differentiation mechanism is employed at the network layer.

All the heterogeneous traffic is prioritized at its originating source node. When a packet is

handed down from the network layer, it will be kept in the Tx queue corresponding to its

priority and wait for its turn to be transmitted at the MAC layer.

From the previous section, we know that the packet length is related to the received

SNR. The greater the SNR is, the greater the packet length is. In the IEEE 802.11 MAC












hMad~ r MSDU CRC hIadC SDU CRC h~ajdC MSDU CR

Figure 4-5: A fragmentation example.


protocol [5], this packet length may be called as frame length, which equals the fragmen-

tation threshold plus the length of the MAC header and the length of CRC. In the IEEE

802.11 standard, the MAC layer takes a MSDU from the Tx queue and adds MAC header

and a CRC to each MSDU to generate a MPDU. In order to reduce the probability of

transmission errors, the IEEE 802.11 limits the size of the body of a MPDU to be less

than a fixed fragmentation threshold (FT), or it will break the long MSDU into multiple

fragments, each of which will be no longer than the FT. In Fig. 4-5, we show a case

where a long MSDU is partitioned into three small MSDUs in the IEEE 802.11. Since the

length of the MAC overhead may be kept unchanged, according to the analysis in the previ-

ous section, different channel states have different frame lengths, we can say that different

channel states have different fragment thresholds (FTs). The greater the received SNR is,

the greater the fragment threshold (FT) is. Hereafter, we associate FTI, with each state SI,

of the FSMC model as depicted in Fig. 4-4. In order to improve the channel utilization, we

assume that the MAC protocol can adaptively adjust the fragmentation threshold and the

transmission rate according to the channel state. To accurately determine the channel state

when some packets need to be transmitted, we further assume that we have some channel

estimators or predictors, which can provide the accurate channel information for the proper

MAC layer fragmentation.

4.5.2 The Courtesy Piggybacking Scheme

In practice, the size of a packet generated by an application may be fixed or may vary

from a minimum allowed size to a maximum value PK,nct i We argue that the PK,,c

should be properly chosen to reduce the overall overhead. Suppose we want to transmit c

Mbits traffic. Packets are generated according to the PK,nct i We assume that each packet











4



S10 -i ~:







4000 1001200(
Fragmentation Threshold (blt) PKrnax (bit)


Figure 4-6: The total overhead with PKmax and FT.


can be correctly received without any retransmission2 Then, the total overhead should be

the sum of the overheads 04, at the IP layer (for example, 20 bytes for IPv4), Omac at the

MAC layer (for example, 34 bytes for the IEEE 802.11) and Ophy at the PHY layer (for

example, 16 bytes). Thus, the total overhead to transmit the c Mbits traffic can be written

as




where [*] is the function to round the element to the nearest integer greater than the ele-

ment. We show the relationship of overhead vs. PKmax and FT when c = 1 in Fig. 4-6.



From Fig. 4-6, we observe that PKmax should be reasonably chosen when multiple

fragmentation thresholds are used. It cannot be too small, as it may cause too much over-

head; neither can it be too large, as it may generate too many fragments when the FT is

small, which may further degrade the overall throughput. For example, a large packet may




2 For simplicity, we only consider the case without any retransmission and view the
resulting overhead as a lower bound. Apparently the overhead with retransmissions is
larger than this lower bound.










be partitioned into many small fragments. Each fragment is augmented with an individ-

ual header, sent as an independent transmission, and acknowledged individually. Hence,

a large packet will cause lots of DATA/ACK exchanges and result in suboptimal perfor-

mance. Thus, there must exist an optimal value of PKlax such that the overall overhead

associated with the successful transmission of a message is minimized. Assume we obtain

P~lax, which may not be equal to any of FTe(k = 0, 1, 2, ..., K 1); it is thus advisable

to approximate P~lax with the closest fragmentation threshold corresponding to a certain

channel state, say So, Therefore, we set PK,>e,. to FTen.

Next, we want to show where the "free space" comes from. When a packet with length

strictly less than the PK,>e,. is transmitted in the channel state with FT less than FTen, the

packet may be fragmented, and there is no "free space" available for all fragments possibly

except the last one. However, due to the time-varying nature of the channel, when the

packet is transmitted in the channel state with FT greater than FTen, one packet does not

have enough bits to utilize the full capacity the channel provides in one transmission. We

argue that we could take advantage of the "free space" to pack more bits as the channel

allows. As a matter of fact, we have shown, in our recent studies, that the fragmentation

threshold can be up to 10K bits when the SNR is close to 20 dB and 64 QAM modulation

scheme is used with targeted FER 8% (Frame Error Rate) [93]. On the other hand, we

observe that in contention-based MAC protocols, it may take a long time for a node to seize

the channel, and the node that has seized the channel should treasure every transmission

opportunity to transmit as many bits as possible, especially when the channel condition is

good. From now on, we call state SI, the free-space-effective state when k is greater than

m, otherwise the non-free-space-effective state, even though such a state may still has the

possibility to pack more bits when the traffic dynamic is taken into account.

Now we describe how the Courtesy Piggybacking scheme makes use of the free space.

When a mobile node seizes the channel, it will first check the channel state and determine

if it is in a free-space-effective state and if it is capable of piggybacking more packets in










one transmission. If it is not in a free-space-effective state, only one packet (MSDU) with

highest priority from the queues will be served, as the current MAC protocol does. If the

channel is in a free-space-effective state, the node can transmit in one transmission as many

bits as channel allows and thus can piggyback more packets (MSDUs) from the queue(s),

which may have different priorities but the same next hop in the routing table. Since the

Courtesy Piggybacking scheme follows the cross-layer design principle so that the MAC

layer has the access to the routing information, it is possible for the MAC layer to obtain

such packets from the Tx queues.

After identifying the existence of the free space, we now discuss the piggybacking

rules that guide the MAC layer to assemble enough and proper bits from the Tx queues

(the HOW MANY-WHO problem) and piggyback them to the next hop to alleviate the

conflict we intend to address. Since the channel state determines "HOW MANY" MSDUs

the node can pack and transmit in one transmission, the fundamental issue of the rules

should specify is "who" plays the role of "train" that offers the piggybacking service to

others, and "who" can enjoy such piggybacking service. Without any scheduling mecha-

nism, the role of "train" is always taken by the MSDU locating at the head of a non-empty

queue with the highest priority currently. Thus, the piggybacking rules should primarily

address "who" has the privilege to enjoy such "free" piggybacking service. As a guideline,

the basic idea for such piggybacking rules is that under different channel states, the node

assembles multiple MSDUs that may have different priorities but share the same next hop

in the routing table, to form an MPDU whose length is channel dependent. In this way we

can achieve some extent of fairness between different prioritized services. When the chan-

nel is not in a free-space-effective state, only the highest priority service in the Tx queues

is supported, and the packets are fragmented if needed and are treated as usual. When the

channel changes to a free-space-effective state, according to the rules we define, our Cour-

tesy Piggybacking scheme can pack other services, possible with lower priorities, to share

the residual bandwidth with the high-priority traffic. One such rules is to give preference
















o"queue fvo priority _+|b-MSDU |b-MSDU |. |b-MSDU|
c. 2 Asseml d MSDU
h drMSDU CRC
3 I MPDU
queue for priority level


dequeue
controller

Figure 4-7: Illustration of the courtesy piggybacking scheme.


to high-priority services. It always, if possible, packs the high-priority services destined to

the same next hop in queue(s). Only when there are no more bits from the high-priority

traffic fitting into the free-space3 will the bits from the lower-priority queue(s) be consid-

ered for piggybacking. Other rules may not prefer the high-priority service; for example,

a high-priority service may trade-off its own performance for fairer channel utilization by

its courtesy-piggybacking the low-priority service. One such rules is to always piggyback

the MSDUs from the longest Tx queue. With such piggybacking rules, the traffic dynamics

(different packet arrival time and destinations) and channel dynamics are jointly utilized to

strike a good balance between throughput and fairness. Note that the piggybacking rules

are not necessarily defined a priori; they could be designed to adapt to both channel and

traffic uncertainty in the runtime.

Intuitively, the Courtesy Piggybacking scheme can improve the performance of the

low-priority traffic, since some low-priority packets may be packed with high-priority pack-

ets and be delivered to the next hop for free; thus it can statistically reduce the time taken to

contend for accessing the channel for the low-priority services. This benefit will be more




3 When we say "no more bits in one queue fitting into the free-space", it means either
no packet is left in the queue or packets in the queue do not share the same next hop with
the MSDU who offers the piggybacking service.










pronounced in mobile ad hoc networks using service differentiation based MAC protocols

[78][99] where the MAC protocols sacrifice quality of the low-priority service to support

high-priority service through either time spacing (differentiation of Interframe Space (IFS))

or backoff parameters [5]. On the other hand, the reduction of contention from low-priority

services can in tumn benefit the high-priority services: one node's courtesy piggybacking of

low-priority services may help its neighbors to transmit high-priority traffic, because less

low-priority traffic will reduce the contention the high-priority traffic may encounter. One

may wonder why we do not simply release the channel so that other low-priority traffic can

use the channel, that is, the so-called complete sharing scheme. The problem is that the

time, for which the residual resource is available, is too short to be given to other services

due to the overhead associated with successfully seizing the channel. Besides, some MAC

protocols such as the IEEE 802.11 family forbid others to use the channel during the time

period specified by the Network Allocation Vector (NAV). Even if the NAVs are reset, the

contention process may take too long to render the harvested resource from the rate adap-

tation useless. Thus, the courtesy piggybacking by high-priority traffic flows makes more

sense. In short, our Courtesy Piggybacking is able to achieve better channel utilization

and further improves the fairness between different prioritized traffic by the following two

means: lowering the contention of the network and decreasing the overhead required for

a transmission. And as we show in simulation later, our scheme significantly improves

the performance of low-priority traffic, while improves or at least keeps unchanged the

performance of high-priority traffic with appropriate piggybacking rules.

To illustrate the Courtesy Piggybacking scheme, we demonstrate the operation of the

scheme in Fig. 4-7. First, prioritized packets called MSDUs arrive from the network

layer as b-MSDUs (basic MSDUs, the basic unit) whose lengths agree with the FTm. We

assume the maximum packet length PKmax is strictly enforced at the upper layer; if not,

an oversized MSDU will be further broken down into several b-MSDUs and the resulting

b-MSDUs will inherit the IP header of the original MSDU. The b-MSDUs are kept in the