<%BANNER%>

Quality of Service Design Issues in Multimedia Communications over Power Line 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 E20101119_AAAACE INGEST_TIME 2010-11-19T14:12:32Z PACKAGE UFE0013881_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES
FILE SIZE 21843 DFID F20101119_AABKIA ORIGIN DEPOSITOR PATH katar_s_Page_108.QC.jpg GLOBAL false PRESERVATION BIT MESSAGE_DIGEST ALGORITHM MD5
57112da3c4a9b90deb0711ef91bc3f6f
SHA-1
d07701fe4e68a8f10dcfe55a920cbea14f838ac9
13411 F20101119_AABKHL katar_s_Page_041.QC.jpg
da74ae9e66092cc959d8e386e82eef53
5287bafddc88d759f319e48f3c3461426e5ce351
19960 F20101119_AABKGX katar_s_Page_028.QC.jpg
9f79ec0256fad861be609a6f895bea1b
f3e3488cd885806276d4529d8894a04d6e332253
99812 F20101119_AABIZD katar_s_Page_042.jpg
f41b10798e54b3153e1c58e68137ac28
631593479a0888806e55570808bd548d3f7516a5
101208 F20101119_AABJDV katar_s_Page_010.jpg
eaad8ed1247f3bac90b3b97d66195f30
21d160f5b214af1a1686a0ace9b879dc54caa168
72131 F20101119_AABJEJ katar_s_Page_024.jpg
a4478cd7309ce6f027c87a6f636858ea
f83f85a0f627f5a1d759c751f6751155a6125423
22208 F20101119_AABKIB katar_s_Page_137.QC.jpg
79bcdaae4aef57eac2185a287df51e8a
8092569ca15b07e5f0fb696ae09e1735fefd7646
4762 F20101119_AABKHM katar_s_Page_135thm.jpg
f5d6e0ba67645fb1b85ee7b6d3de8cd8
973f33f29b041bbcd9ba31456b304a3c5c8bdbe9
5599 F20101119_AABKGY katar_s_Page_137thm.jpg
97bebafc407a4f8695992e0ffb35bf8e
0733a090e0f0c3c06e03d35e33ff64f896097eda
72119 F20101119_AABIZE katar_s_Page_005.pro
2b9836f9e706c69f5442fcf6ff2d082b
f4911dc3f8573e9f0d135347359188724a9debd9
112323 F20101119_AABJDW katar_s_Page_011.jpg
24c9114fcecba6605aa02f1076fd96a0
ba77133a1f7e55e1d2ad86b0a0acad7c3e8629ec
78572 F20101119_AABJEK katar_s_Page_025.jpg
b0eef702566ebe598645ef631d39a4a2
2fe476724cbbe8259a075f0989f8d69e6f2de503
15732 F20101119_AABKIC katar_s_Page_096.QC.jpg
a33e00a0d558fef298a616c388496fd8
6e8cec45f346a86a736f246e4c44f0f52acb7a1c
5518 F20101119_AABKHN katar_s_Page_128thm.jpg
1cb82f261740f92f8987f66007355378
1f28e02cf85de6e99745f15988f733dca8f6ec88
2013 F20101119_AABKGZ katar_s_Page_069thm.jpg
3baa73e261c264c3f79b1d0cdb4f1319
18dc1ffa248f6fac01ecc56c01ed448fbf11b8e6
3189 F20101119_AABIZF katar_s_Page_042.txt
555930ee85ff60712dfc4be2a85543a3
3fdfa070263a33725dc6e6496ccab988dd2fbd42
75921 F20101119_AABJDX katar_s_Page_012.jpg
c3f1b0503c245d9e653fe3785bacbc45
280e3250d97bf775ce7d6751ed893b56f145f714
83876 F20101119_AABJFA katar_s_Page_044.jpg
776dbaaf9a39685dd53c732eb5af4682
94215b6036635f7b05f4a934f9b46a127ef3a407
81796 F20101119_AABJEL katar_s_Page_026.jpg
9f381690350c51b07c1f9497a8f7aa50
822d2e47949bf97b6f9e75e8f7e4e140a33dd849
3205 F20101119_AABKID katar_s_Page_134thm.jpg
33f1280433e327e3e96c306b6ac85df9
48782c9c2e47209d59b8df14f0536b718c5017b3
6349 F20101119_AABKHO katar_s_Page_085thm.jpg
e559ca08400f3b08f552c1ec6a0d5f16
052bf3fc2f68f9aebf54f4ce670493ddb5596c11
1053954 F20101119_AABIZG katar_s_Page_120.tif
e7fd25aba83a72d6baa66d0d47dfc612
be5968a9e4faa4c351216480658c19474f946f63
64321 F20101119_AABJDY katar_s_Page_013.jpg
21e23fbed88690a8c4a9237c26f4a6fe
29da1971f66e61587939dc545807fa46499b9783
87992 F20101119_AABJFB katar_s_Page_045.jpg
0a1dea91747398243dc3bc89a2a7a9e3
5abeb6594d8645221656e2939811fac79213ff84
79188 F20101119_AABJEM katar_s_Page_027.jpg
299ea901d780a46d0102d319d794f7fa
4e74822be8bad44f1c6f9ba09a897dcf00a9f167
6434 F20101119_AABKIE katar_s_Page_117thm.jpg
ceed78c77ba56eaa9eef5f270646ecfa
a2276a04d244f4b7fd3b38a5b35d2176d3c1c7fc
6373 F20101119_AABKHP katar_s_Page_086thm.jpg
c8ee416fb2f455b79131a13f474b80e1
467ee1a0c561252d47ec9e987e7543eebe97e70a
110109 F20101119_AABIZH katar_s_Page_040.jp2
15ef90efeae864b149ca6771796460da
30cbb50bd89d01a77204ec089957cd2d4c3c6979
74261 F20101119_AABJDZ katar_s_Page_014.jpg
7c408ade4bbea59b317b57c5953d4665
b2776244eda89e35ff8df6e286af3c18e03db6a1
100336 F20101119_AABJFC katar_s_Page_046.jpg
7ed8418bc8806430123a07a79b51d106
848b3bd23f58eac31b8f41212ac0cda10cad0831
62817 F20101119_AABJEN katar_s_Page_028.jpg
0da96ee130bba8625db73c24c795d35a
2f56ddfd36749cf70771588bc1cef980b174433b
17222 F20101119_AABKIF katar_s_Page_007.QC.jpg
7aedc2a21c6c458094e85bbc8afac128
431860390ba85c5e7aec1dedf28564a46311fb77
6628 F20101119_AABKHQ katar_s_Page_001.QC.jpg
4aef349b83123ce331c342c529f9e16f
7b6ddfc29f03c57442be49e25305e3b7cd8e93e9
20949 F20101119_AABIZI katar_s_Page_091.QC.jpg
db0dbbe17dfacb0884326ac111c912f9
c63b9d3aa72f85797c47265db5c7c2babc51d6d1
77599 F20101119_AABJFD katar_s_Page_048.jpg
b84ba1420d74bdb7689082b3ef3f9553
e49cef622c7d87e7fd20f7c670f530fe42a3c15b
59385 F20101119_AABJEO katar_s_Page_029.jpg
9cc01020585c8f7203c3a8467e82fd03
ce617fca65a52e94d60a6e003acb1c41f7095c33
4289 F20101119_AABKIG katar_s_Page_151thm.jpg
945013ab3c57209f1d5567fb778f1638
6812c299024c50072ba7c2673770c43e332714d7
21620 F20101119_AABKHR katar_s_Page_124.QC.jpg
21b7faf0b511c5d8b7f5ba119a83e050
bd518f11543c11179f6a768a6f9d16f67d85d3d3
4083 F20101119_AABIZJ katar_s_Page_149thm.jpg
b600e8a81ccbb586b72511978e4caa2b
6fe2a31466b9ba856e6d271da3543538fa7586f5
61800 F20101119_AABJFE katar_s_Page_049.jpg
1e822a1f6a83a1570ead82a82fce508f
8b7cee9af146ba3a677fc80f3d3d058f1a554823
55814 F20101119_AABJEP katar_s_Page_031.jpg
597b5ab77ac4c0d375051da69216ba36
af08ab56d52ac29b1905c7cfb77f8f3c09af9854
1743 F20101119_AABKIH katar_s_Page_003.QC.jpg
d38c858aa5912474db1e6dbe34ae0842
71d113e57d7660a52d6e58654ef40ca202f82a4c
4466 F20101119_AABKHS katar_s_Page_083thm.jpg
05a6fe406771f2f30d3cae93b62f9cb8
07d277f80fe762e66a02b64ab641311cffaba799
32028 F20101119_AABIZK katar_s_Page_131.pro
042ce67d5451a8676f3ad622cb877fb0
c7f0ca8c630b0336420eab61d7117428c99b87df
47559 F20101119_AABJFF katar_s_Page_050.jpg
58d897969501ab9d6362e390e81f2474
fef6f9fbde97206f45da8193744d34e3971b00ab
62944 F20101119_AABJEQ katar_s_Page_032.jpg
2cbad0e3e2e3324de79cb2bd5fc87615
d1baab185b5d4685fc08e28178cbe76a98933503
22075 F20101119_AABKII katar_s_Page_082.QC.jpg
a0161a5004ba889fab02db15b52abe91
537c49e14e9a74e93d6a3b44d64c53ef24e38915
24554 F20101119_AABKHT katar_s_Page_118.QC.jpg
5c2b79162778a5030cb68e8392e86c87
8f3c49074395aa3f01d220cdb36f4591eaaba902
1496 F20101119_AABIZL katar_s_Page_041.txt
f39a6f6d68fe43c2b95be72d22f21730
7866d04bbcc9617e6c617de99f226d6855b6eb2d
52492 F20101119_AABJFG katar_s_Page_051.jpg
2423ea9427a9de3a65f23cc49188c41a
8acd098ebd5621f00b52fff5fbac8b8a57897ad7
77492 F20101119_AABJER katar_s_Page_033.jpg
20144e70b0147ee679623ddafc1fe7e4
44e31cdb2db3ea94ff045bb869f3187f16e3dee3
21243 F20101119_AABKIJ katar_s_Page_035.QC.jpg
d82412f33094007225112c47eb9e520b
9031c2e0f9cc483af57305178b83738e34acf82d
18448 F20101119_AABKHU katar_s_Page_092.QC.jpg
45ed41071ec6376dbec05187dc0c62d8
a69f0a9a358aefc3f0d7fa94bad97505d5b9fe1c
23112 F20101119_AABIZM katar_s_Page_140.QC.jpg
491de41b3f2acc87277847c58a5cf6c1
34061859e14603eee960cbbf870e9904c20dd733
83418 F20101119_AABJFH katar_s_Page_052.jpg
0a420b6c717fad3e258800f1a011382e
5d56bfbb24de5c8f0d82b9ee743b6b358568111c
77947 F20101119_AABJES katar_s_Page_034.jpg
3cf7e7d5aa0ed6d71539b6248be5d4e8
d33b70ff54e44190c608a7acfcc643f6de170b2d
17940 F20101119_AABKIK katar_s_Page_122.QC.jpg
0b5c2aba6d1bd99039fc91a4717ab4a0
322bf1af815f32ab0c577b2ba182478d2b05b253
26968 F20101119_AABKHV katar_s_Page_085.QC.jpg
8a4920a65338b5dc689cd12bc77448c6
e51843a2a5dd0e538f1700b9477c8168c4b29a60
24791 F20101119_AABIZN katar_s_Page_097.QC.jpg
b583745e7ee88178e2f628c7e666ab1c
fc58b0cd8dc9fc86fda550a0803dc335683ef003
83651 F20101119_AABJFI katar_s_Page_053.jpg
06d1b5bf25414b40b75fc4aede7ad903
5c65a1dc79aad404a5798701c66e19ddb6bad10d
68772 F20101119_AABJET katar_s_Page_035.jpg
35232b4f1f3d3397048f6ae474306b01
cde062b43e59aaba4b2d337e8608d6694001e6ab
19864 F20101119_AABKHW katar_s_Page_120.QC.jpg
d057ff44a0c4d430a76b192963acbf77
dfb033f78286b7b65a2b1d952185eb295c6874ba
19518 F20101119_AABIZO katar_s_Page_090.QC.jpg
d027c945006f9ff0d2393da7240a1344
84702d1d055bae272c23dd34ccd479be77582577
73956 F20101119_AABJEU katar_s_Page_036.jpg
6590c2b5ef567d41005e006a3b134589
c61d5a5b8429cd80ff2fb32849b848a3b3e02737
22267 F20101119_AABKJA katar_s_Page_128.QC.jpg
1a2958edbdc1ab733542f0e159e36600
197b5317ea12538d394fbb8688fc78f8caeadcd3
27919 F20101119_AABKIL katar_s_Page_117.QC.jpg
a53036d6a43fdc56f414625df035ca0e
b568904d487b77c3a5895b1325c894a5f20a46a3
4004 F20101119_AABKHX katar_s_Page_081thm.jpg
14f24b8ba23eff49e3ea09318ee1b416
43bf871ce349128d10363f156236006569a12a67
2043 F20101119_AABIZP katar_s_Page_114.txt
d57882b3d115a40bc1d19ea57eff65e8
b50803096aa674d7bb687b052cc70c6c9a8ba13f
67202 F20101119_AABJFJ katar_s_Page_054.jpg
7ed0da45b103afec28778d1d3bc3a4b9
6a61659652a504e7b20137a341884a76777669c9
84559 F20101119_AABJEV katar_s_Page_037.jpg
9fe3f53e30ed7be339bf74a32d98900c
2b7862f5ad8508ccdf4e8827cf32883cde4caae4
6040 F20101119_AABKJB katar_s_Page_111thm.jpg
0bba5840586b5e2fc3290ffdfdbf75f8
99d2dcbe369a8c4057a943b82bb704ff4f58e821
5288 F20101119_AABKIM katar_s_Page_058thm.jpg
bacc57becaf3f68bdf2333bd18dacf61
3b024690cb99ea092061fd792b30908baa1522bc
6832 F20101119_AABKHY katar_s_Page_010thm.jpg
a1d2f33769e302033eb7fa6fd5777f5d
092d268a4dbb6c70b57366a7ee557a3077a56c56
73077 F20101119_AABIZQ katar_s_Page_007.jpg
9b783439d8be1fcb349fc0ddeb94219d
b750b63ca7f4af095c013fb8e512a2ec200ae146
85685 F20101119_AABJFK katar_s_Page_055.jpg
8285b2308d5a8d4c05d70f6496bcffc6
005bbcd0d6e9a8735d72823feb1b9664b66dcef3
81077 F20101119_AABJEW katar_s_Page_039.jpg
ac5934eaed1ff9b10b3fd1ad101a6a15
e2ba55efaa7343beeb5f350ca6e6ebbe04471d4b
17513 F20101119_AABKJC katar_s_Page_103.QC.jpg
8a5c10eecabffb44bd2b5223fd48009b
52d367664105383ae9d16b81acdb80da4a4af7eb
4202 F20101119_AABKIN katar_s_Page_132thm.jpg
bf40a611729f80c0560707d8c9be8c2d
8809cb6e493ee0628779947b5180cf179a2dfe8e
5903 F20101119_AABKHZ katar_s_Page_129thm.jpg
cf0abca5f3142b9dd7d3527271cc4d1c
a1a1f5320674e9b360e937954f3a40a536ec9a4c
6081 F20101119_AABIZR katar_s_Page_026thm.jpg
7dd7e0f06ea3524bebb5674b5202aaa0
63733e02de150c7d518c6fe3724ea150d1d0b561
57750 F20101119_AABJGA katar_s_Page_072.jpg
b93bc0be367feb8f34d89f1e60191a5d
5f2d32fb9f335b8e4ee3d3de403016fba492ba77
85670 F20101119_AABJFL katar_s_Page_056.jpg
ac7ae642ca3ddd4b7bdf12cff8ba13bc
e94ac0eef423bcec853ee5c571325ec842d28250
85805 F20101119_AABJEX katar_s_Page_040.jpg
26dfb8eaf807aa0f66038ef82c9ec857
926e5a873f187462e6f324854fcad2c0a263dd80
25109 F20101119_AABKJD katar_s_Page_110.QC.jpg
2b63d84f5584abfb55934fb25a48e499
b172c479bd288b1eefa5e8f23289c40f77def0e6
17328 F20101119_AABKIO katar_s_Page_102.QC.jpg
2c1946b88054b453b39405547a08f5eb
64f20867c912f7e2392fb45aadeda4fbea73e3db
25271604 F20101119_AABIZS katar_s_Page_094.tif
fbcf1cbe06157b5ae323dcd3be55ee10
315e7f4cdcaf8569a04394ad027a79ecfaf7b409
75407 F20101119_AABJGB katar_s_Page_073.jpg
80410cd14a5878c2f8bf94648a3a1a20
41183284b19c634f06a711f3fa38643379abafbe
46830 F20101119_AABJFM katar_s_Page_057.jpg
472954517df9e843713709b71618d9bd
4cda4af8b5f8ce47fac2655f636fbec83718943a
41571 F20101119_AABJEY katar_s_Page_041.jpg
eb8d2f5d30ed66433f76b63e1c5c42a9
2d50c986a84057b796b5a243b20a282b4c127c4e
18423 F20101119_AABKJE katar_s_Page_009.QC.jpg
a0ec7c4cb6f14f1f2987204fd5c90111
378d0d1df7433c5d338c1d88f406a9923f8b1259
19978 F20101119_AABKIP katar_s_Page_049.QC.jpg
1ceb6b8bbc1fb58794f76cfb4eacc2cd
2aaa84043794edabf7f26ac272cf90407a8e7f91
6694 F20101119_AABIZT katar_s_Page_042thm.jpg
bc50ff3f48e0abd1130df786e03d21c3
1f232151fbd246a9ee094ff9b55dd30d31ad74ea
66806 F20101119_AABJGC katar_s_Page_074.jpg
afd8b578aea0d5dc217a4005158bf401
b8a42dee928f7b88d3dbe956d498a2cfd9c8328c
70636 F20101119_AABJFN katar_s_Page_058.jpg
cbe8962be161d03842c26dd1654e4df8
15727f5d7a35b2634e2600f7387cbb15ca9fc335
81521 F20101119_AABJEZ katar_s_Page_043.jpg
6cea004039d2bd2663e4d4bb123e8c90
fd246ed8437f1a0b479f80eb6acdca4c8e191417
5479 F20101119_AABKJF katar_s_Page_125thm.jpg
bad2b9013cddc2bbe0f6248980b908fd
65a5c5477249af15c9ce54850ed34bbf4b8c0177
21636 F20101119_AABKIQ katar_s_Page_133.QC.jpg
e793ac718ed45b53877f1e19fed7dedd
b6c13859de05d501578eda9bb29036b93b51fcdf
26083 F20101119_AABIZU katar_s_Page_055.QC.jpg
68e19b4ca4baf823a8cee0f2490b2de3
3fa248500ae9d484c1518a2f200cc7dd08cfd63a
64481 F20101119_AABJGD katar_s_Page_075.jpg
a1703bb2421df9aec4a453e1d21f3c47
618cfb58960fdf8130b56890356725bff3812cba
84429 F20101119_AABJFO katar_s_Page_059.jpg
3aa06d63db491810512fade60906c733
2305c05221ec7052e0c86878db4a97ac31f0dc13
5667 F20101119_AABKJG katar_s_Page_078thm.jpg
25b9eeeb8424fda29250c67c2f88c706
d5c73eec8eae37ebd93b622bb10f332d2f416ae4
11270 F20101119_AABKIR katar_s_Page_134.QC.jpg
3ea14bd5776647957d5d0d779afca855
4da6ad1a7cbab19f9a62dc99b662a7f9fe410e19
33043 F20101119_AABIZV katar_s_Page_028.pro
1dbdcfa0fd8fa50aed9aa646da446aaa
155569c726545469b2bc66d92b0984c0603b5eff
45370 F20101119_AABJGE katar_s_Page_076.jpg
89153f889f16dcd71d3165da30c65139
7720554583ac98a5866ce6cbe34322568619ecec
86184 F20101119_AABJFP katar_s_Page_060.jpg
2d232a2acdf6b55b0160dd2f2d334c19
b55a5e2426507a7ceb1b3f18e34b2517fd10fc64
25453 F20101119_AABKJH katar_s_Page_039.QC.jpg
190a21aec2e71d011b24241ecc94b1a9
6d1d3e02c9f0fd2ec351a69481aa28ce2a39cbff
5998 F20101119_AABKIS katar_s_Page_048thm.jpg
07a55c314722d06bac2ec6d6b095210a
855b5a9975f992dc31422f6f61e119bdb0af104f
F20101119_AABIZW katar_s_Page_016.tif
5b035fab8b23eb5a3c6843cc06181a2d
798ff5ded65da957c75b69216fd0915e5e0360fc
76250 F20101119_AABJGF katar_s_Page_077.jpg
e8335fb9ca3c28de31b6f14de0a93c3e
b32dc2cd46ecb5d3c1db225a9992700f2679d88d
56180 F20101119_AABJFQ katar_s_Page_061.jpg
85271f73c557e5def5a123dd6f30ffed
1f9ed9679aec4d302952dde5048b7f9bd168bc5d
6502 F20101119_AABKJI katar_s_Page_158thm.jpg
b4dfe0e6b35925b6277b0706613db701
c92a6179d5efd1f5683933977c34b63dc4802f5f
4788 F20101119_AABKIT katar_s_Page_122thm.jpg
75daea53c76a5ca164167819cbc59203
759c39cae7bb0ac85835c8bf3b49f034b7823d69
876691 F20101119_AABIZX katar_s_Page_035.jp2
852af123986fc15c42d992f814c1a69f
323d9759b6fe5aa76216fc5ff682def7c1f212ea
70538 F20101119_AABJGG katar_s_Page_078.jpg
50733720cbac0fe54e7681d2192b875a
232e2e27439bb44aecc615e8e79d2a783ca28c79
85269 F20101119_AABJFR katar_s_Page_062.jpg
88420869b92d63c0b3e7e9fe2bdc7121
36b04a2f2cde9b2d45f9115127c559e3ccd0d47c
23264 F20101119_AABKJJ katar_s_Page_116.QC.jpg
b7481e6cf2a69b8cc7037a3e96b32c62
f714671297fa43628dab2c2245c8f0950408fdbd
3928 F20101119_AABKIU katar_s_Page_007thm.jpg
321c1c94e837bb64f405d6531bae0fb7
14eb17b8712239ccbc9b123129d4bcddf40e35f0
48327 F20101119_AABIZY katar_s_Page_123.pro
33eba4bbfbd55b2d66e8dba9e72ec2fc
bca5b766ade07e05f58002fbe47e40cb999f9a4e
66675 F20101119_AABJGH katar_s_Page_080.jpg
de2fbe092ab88a393ea783e78e08f93c
fa353d108d1cd87c354188c6ec350895e2818d3f
81237 F20101119_AABJFS katar_s_Page_063.jpg
95a9172792577ebb336591bb5034c43a
d3db0673620bceb63e005d960b13d4f41209ff04
6193 F20101119_AABKJK katar_s_Page_044thm.jpg
559002ac282b83159c281482d69ba7f6
6d235fb6fe4a6c083b6cff333579181bed8c8ccd
28184 F20101119_AABKIV katar_s_Page_042.QC.jpg
cd5f7d870efde332a033f70313ffe480
8977ce9a92afeed70cc16627e62d2a8d4dec6b1c
14223 F20101119_AABIZZ katar_s_Page_030.QC.jpg
1833829849bfc34b799f0f24b89e0c1d
027d57b457e08d2216046549aabbe0d376838830
59462 F20101119_AABJGI katar_s_Page_081.jpg
149102b2d748c5152e99dc49b23ac171
7f47b0a184d6f9b1cdca0ddc135379dee2475849
58654 F20101119_AABJFT katar_s_Page_064.jpg
faa3b1e513d1aafd6ef0ecbcff98ff80
746b7230bd9a1c9371cc40c984d32d246682dc72
19483 F20101119_AABKJL katar_s_Page_032.QC.jpg
46a341bef0df95a84443c652c6511091
a15f3a6c3b7461ea0c381a17264e73313251e9df
19663 F20101119_AABKIW katar_s_Page_068.QC.jpg
88b25c7d08ea3347dc18473ade399713
386f38a7c3ea8e48b7ec7a33f62c1a8ed5c1f9d4
70585 F20101119_AABJGJ katar_s_Page_082.jpg
bfb3a8070c24e5ca4b8d40f518bfce59
b1faa0c8cf49081836efb657a1be6637accaca65
82007 F20101119_AABJFU katar_s_Page_065.jpg
e5a16b259fb41e8f7d7d0d836bc9bf83
8df3d7dd25ad9eecd32a8fb323974cfb5cc0f6c9
27399 F20101119_AABKKA katar_s_Page_160.QC.jpg
d0e45f6c2146a018d19292544506ecbf
ea26b7047ecb91ddfc6d9d4c98f2e64a751d5611
5511 F20101119_AABKIX katar_s_Page_015thm.jpg
d4dc42eb8e29312a1f5aed77ec20cb26
b44924f1b28fea556846b0863c13cf05ace8e20f
81063 F20101119_AABJFV katar_s_Page_066.jpg
5e58c51977c3665a5103df65cbd054a4
d1fdaa5d13865458fbe8037c1b581ae796331c8a
17704 F20101119_AABKKB katar_s_Page_083.QC.jpg
31d3cab377f548a81e5e5411b1cd0acd
67260bf9ca6f4f23a0486099d55cde1dd76670ba
1332 F20101119_AABKJM katar_s_Page_002.QC.jpg
c9f9d67d30c6492bd84b11860f33436f
276410fde287a584606f73a39d308da0f07d4777
24929 F20101119_AABKIY katar_s_Page_063.QC.jpg
4d12dfe6c47e51909310d45ab752f5c6
692d66e64e6c11f66d205c4f4190b467355cd578
60662 F20101119_AABJGK katar_s_Page_083.jpg
d525ece8f27529f7b874fcd517dd7ed9
6d798dc2e7f8490a4016c0f7bba1ec17a116503c
83208 F20101119_AABJFW katar_s_Page_067.jpg
264b4677725f1e8f315b24b18bb7e599
8659933463606ac29d612238b6e608416b553da8
17267 F20101119_AABKKC katar_s_Page_156.QC.jpg
5f89bb584e95a6c38f49347fd311febb
fb3d1a659821d15f38aee4c576f6c2d7d2281e5c
26732 F20101119_AABKJN katar_s_Page_071.QC.jpg
b444e73cb10bf30e353eab78745b96e8
bc044f4f0d316119dae814b986f0e6f085529492
6105 F20101119_AABKIZ katar_s_Page_155thm.jpg
b851b5ff07529ad941a48c380bf5084d
bbaa879f854e56e12946cf1a112cc80af2d2059f
74431 F20101119_AABJGL katar_s_Page_084.jpg
5b9599f7bcec6389c89cbf3667237aae
07c03c65125447b437ff6897cdcf2b3ac19f3b35
71155 F20101119_AABJFX katar_s_Page_068.jpg
1b595823e8dd89eb61c8cd70de7ac6f4
f8927b52e53bea98033c72a48ffa1b7ef587768e
67921 F20101119_AABJHA katar_s_Page_100.jpg
78a94e4d746cc4926c7ffe750a5b4f41
7e796b454753905bc99966554dc7979bd5f4dea1
29038 F20101119_AABKKD katar_s_Page_158.QC.jpg
75532ff518b126089dd78dae191f7352
fa3a07e4fbe50f7d9c309db40200d7af5cb144d7
24710 F20101119_AABKJO katar_s_Page_111.QC.jpg
6f9379d3db6cd58108f2703aabbd3e80
e65fb142a0ad0896d558a03e6a8174bc15016021
85621 F20101119_AABJGM katar_s_Page_085.jpg
68143cc1e149775974eaa16c165236f5
7726c24bb257398bd7bb83371efe5aa93d1651c9
28002 F20101119_AABJFY katar_s_Page_069.jpg
2913bc8b4b9a8d6eed215a03c712942d
2c2372f95af427b09341c8b049502947263a5175
79959 F20101119_AABJHB katar_s_Page_102.jpg
1228e7502f163411daf3b14cb3028caf
1fc3e87711fae579f133a434c82310e33bd4a11c
18925 F20101119_AABKKE katar_s_Page_100.QC.jpg
a479e0404243deef4df548d41dc499a5
423d30416f7f2f90347cb7f188004f79afec4ed2
4349 F20101119_AABKJP katar_s_Page_076thm.jpg
273f6730b85057344d30729eb6e8514c
1dcf4c2975fb0bba117b530a768080181fd5920d
83719 F20101119_AABJGN katar_s_Page_086.jpg
9ec68c48555e1fbf36d12b25e44f70e7
076f797509e48d05f564728a78b5c2cb3e57e5e4
76738 F20101119_AABJFZ katar_s_Page_070.jpg
7b5e63ef2c7d4acf0e1aba0d3e3b6d9c
7ae1a9848cd9de01a752216d6cd7ba9610bc118e
60897 F20101119_AABJHC katar_s_Page_103.jpg
c6054ab4ce1b5830c0f455f51777c598
24b46e933148406864b0074cb69b389d3ee545cb
1637 F20101119_AABKKF katar_s_Page_107thm.jpg
1e78a9dbc9e784dfecc365f7548f8809
8b7aff05a70c322f37c0b4cd8201607880a00078
4958 F20101119_AABKJQ katar_s_Page_013thm.jpg
4a09ddf916a57472254acb8f7477ed04
4232027f601a568d4abeb815eedfa63572b7f125
82769 F20101119_AABJGO katar_s_Page_087.jpg
de4d502bffa6085ae6da0aedad0d7861
516167408d684c351257c2b4ce4d5c372c880379
91904 F20101119_AABJHD katar_s_Page_104.jpg
5b0cf6c791d01797751936815c10545b
8bbae80d17837bef2ed69d7810c187c7943b51f3
6190 F20101119_AABKKG katar_s_Page_055thm.jpg
0df7534d5f4b8da1a154f6b37f50066e
969e94d05e32c10cde8d6f3ec5c199fc8029028f
2304 F20101119_AABKJR katar_s_Page_130thm.jpg
8f623e1d672f31172f7e4dfca56ca4e2
ddd7d6c94cd101dd4c0182da5f9a6d8300292b4b
69896 F20101119_AABJGP katar_s_Page_088.jpg
2544ffff62896526a2b75ec94638c2d8
f60843fabba7aadbd5896789434e75394c353010
75582 F20101119_AABJHE katar_s_Page_105.jpg
e39201af2bb6ed55e3f8c0993bd542b4
6a2b69d376ad6af84dc9f357938de9a896ffa8ae
6419 F20101119_AABKKH katar_s_Page_045thm.jpg
82d00483c9d2331396e8dcfc82e6d0e6
47e9a7723e0d684c92c247871d92cdf38452898d
6322 F20101119_AABKJS katar_s_Page_059thm.jpg
8618f88157cc2d19392d775141adef09
1eaf32d0297373f654bc071eb66fc7d5bad226bd
99916 F20101119_AABJGQ katar_s_Page_089.jpg
73e478621ca6c66c5da15399f664252e
98567a27958c84ebace8b8453864fdd54109bc0c
76933 F20101119_AABJHF katar_s_Page_106.jpg
a9fd758322277c2a3ccde2a44e3aa524
e1bed0e55292bc4fcb8f42ececc0fe7800aacc44
25116 F20101119_AABKKI katar_s_Page_065.QC.jpg
5fdb43a5613ba774de418d07cc4bfc97
dc6e7142c7f3f52d3b5604f1b08068d345911c8a
6071 F20101119_AABKJT katar_s_Page_047thm.jpg
04e4a88816692202a334be1bb3cdb330
b89d0db42182edf7163c932c386f44adb7244455
61865 F20101119_AABJGR katar_s_Page_090.jpg
6acf03ace1a9f444005a3bde0cba574f
057e0186ac6ba79bffecb23dd68b06a7483b8e04
19784 F20101119_AABJHG katar_s_Page_107.jpg
be5d7f829e9b83e17ece6a11b6754756
604d5f88658e680fc2fc39e3926bda8dc09ec181
4897 F20101119_AABKKJ katar_s_Page_098thm.jpg
97683506891527cd2c9bb7ea05b37025
808e1c39825ce0732987641041b26d8ca9a86089
19625 F20101119_AABKJU katar_s_Page_150.QC.jpg
4abe969d9a4d15febe8d233b180041f4
c7fd946195b5e7ab718b4a9a5d8886888bc010a6
68878 F20101119_AABJGS katar_s_Page_091.jpg
efb996a25c2d5e82af905bf88c7900f8
eb332e594e54eb196318a19995d835cc0aa68c3c
72230 F20101119_AABJHH katar_s_Page_108.jpg
c642623d0359a707c0c0ea7133898c78
23b07add942af4b41f72a1ee3fcf79f01963af44
5495 F20101119_AABKKK katar_s_Page_008.QC.jpg
cad4dd24aa7a0e73ea74630b613d43b3
6f137d4eb6ca97d658fa2e9bc5b2a2e575e22c2c
6454 F20101119_AABKJV katar_s_Page_056thm.jpg
73677628300f9c6e410df93f74aa83f9
11a884247f2a3a3c1956bf3905945f1aa29f3eb4
58038 F20101119_AABJGT katar_s_Page_092.jpg
a8923955ee83348f5ea0f4781cdacee8
99a5a76e47170ff97d7b22edadc4f6505807bdfc
79466 F20101119_AABJHI katar_s_Page_109.jpg
2328ad4bdaf3ecca5090ed3f6070e30b
9a4176d74f9ed4cc6269a89001b06107266db2be
18545 F20101119_AABKKL katar_s_Page_064.QC.jpg
42638a5320b23388b975af15ba0af00a
513329ae040a0010c8950ca7f3d89b5361ce1e4a
27537 F20101119_AABKJW katar_s_Page_154.QC.jpg
0ba4e7d46424e0e6101f64825ffbf220
266fc601cbac4ecc980ad68510987752a945c413
71867 F20101119_AABJGU katar_s_Page_094.jpg
b1fbf51549893eca3387c8d89c8baf7b
3b58fa82fc13a8668ba09a81e5ab4e922d1deef6
81820 F20101119_AABJHJ katar_s_Page_111.jpg
00b45b2c6d61a649992e8168b85c1e2a
bce2e9e7175a46887a22898383ea9d2b9aa4e589
3896 F20101119_AABKLA katar_s_Page_057thm.jpg
69563a3262f62c61c54f2d66b2930fe7
8a23aaba36a2851cf1a988fbb483fca8e49895ce
5200 F20101119_AABKKM katar_s_Page_082thm.jpg
9e25fe00da14c9485733c872ee528158
52e4478f85149618ce916636fea81a2d062a9971
4701 F20101119_AABKJX katar_s_Page_121thm.jpg
de9ea621b4bd94079a40e33abf550200
6d9ac588e3b1d1cd5c04e1a43e491c5e20f5c284
74993 F20101119_AABJGV katar_s_Page_095.jpg
9b283d72480b6069ab5d6dbe38ef9c1e
85813e20005f249d2ea9f1692f4b20631ef79356
82253 F20101119_AABJHK katar_s_Page_112.jpg
5f90a1d33d5760f9c4eb05c15f2d1283
df34f38c22c18170b2e91a7c4d9327d80d2a7c38
5022 F20101119_AABKLB katar_s_Page_072thm.jpg
5c41472f87ad36f421d74d4914b262c0
b8fd34b540699672509586c644c0b2b736c2032f
30303 F20101119_AABKJY katar_s_Page_138.QC.jpg
53de080f05a0d437674de65068347bea
ea9bb9ff6ea6135ff0a03374ee894853e7e31626
57477 F20101119_AABJGW katar_s_Page_096.jpg
78d2a4f08b6c0c4e3298837ae013339f
acced64d2f6dd2def4d5939ce632a4bc8e426d3e
6232 F20101119_AABKLC katar_s_Page_063thm.jpg
aaba180d679157401632778f7254e56c
a84532ad6bbb87e7d968d3c682f9062f05487867
5166 F20101119_AABKKN katar_s_Page_054thm.jpg
899e173177f1fbb7432e616471c8c248
21a0539c8f039020d894f5e2cb81312dc3944202
6478 F20101119_AABKJZ katar_s_Page_153thm.jpg
c0e107da43460250618f54c1e4a273d3
46378ad8556e78716de0698c4ea355b9a1f663fc
77544 F20101119_AABJGX katar_s_Page_097.jpg
2337967aa4d82a888302da7ad420386c
a9a7159a3f3ffe9d33893dc2774c251d044e6556
49526 F20101119_AABJIA katar_s_Page_132.jpg
191b11a932d3c0796544b074ee1cd97a
b00a81e0f9015f77872d26fa4ced102d37930a02
62105 F20101119_AABJHL katar_s_Page_113.jpg
798b9c16450446cf93d82898fb6c0803
4b2fcbaf88aacf3e57fb693af8d53033c13adcec
23124 F20101119_AABKLD katar_s_Page_014.QC.jpg
f61d2705da94bcb498d30abd1e5a91fe
1bee376815f5b2e38638ac62d5e8313a5cfd3eb0
6219 F20101119_AABKKO katar_s_Page_043thm.jpg
6eec7686f382ab2320acb04a45375b9a
197f14e48b82ce88a7a1bb9f466bf3180f7f8710
F20101119_AABJGY katar_s_Page_098.jpg
cfd50718109e5816e193e5794d1cbf63
c2b0762272adb257bdd9e4c7700d1a2c6c4b7474
71188 F20101119_AABJIB katar_s_Page_133.jpg
1f86d28a3ebd8ca4f1b9359c77593aa0
7270bf386bec175cf98ab0392fdaea0153e5ee2e
54298 F20101119_AABJHM katar_s_Page_115.jpg
f43595022f1466fc06a0198afd69f347
5daedf20093bfeb8659a2e6214b2e9fc383a7603
6121 F20101119_AABKLE katar_s_Page_067thm.jpg
9ab9f78ff08e8a4e257836daa51e736f
e5627373c5d6b2939006c7f60c60ac328ffee218
21658 F20101119_AABKKP katar_s_Page_058.QC.jpg
b2798bed014b9bcea90e1522515f050d
f9b2f10e1bb41044c0d5a616d5866593190ec5fa
53366 F20101119_AABJGZ katar_s_Page_099.jpg
e3d205c04a4f8c6eb20b18e2c3f64ebf
629b2e5362dcaf23101d5167a63fd02c79df581a
35328 F20101119_AABJIC katar_s_Page_134.jpg
98b22c2c0e968be49cd3d1751f424602
0273241d7bbb5fd858a1f66954c17ea975244e92
92743 F20101119_AABJHN katar_s_Page_116.jpg
13d0080ce09c6dc7329f369cf548cd32
695a902d7313dee58df06551ae71cc9906937d20
5779 F20101119_AABKLF katar_s_Page_116thm.jpg
966696c9946e1f52befde0eca9e02688
acdf109111db6d6839d160ceda5ecdaadff0f1e9
4333 F20101119_AABKKQ katar_s_Page_080thm.jpg
0229ca6ed051eb33311ba7ddebb89ae2
3aecec8bcf44069580825f0814d72445c1038d8a
65164 F20101119_AABJID katar_s_Page_135.jpg
dbc72535cd3e6e6cd56cad897001efc2
31a2aa170d535c15abf092f4a04af26e1842d227
105202 F20101119_AABJHO katar_s_Page_117.jpg
45cb59cad41ea8c34eb65e3dda0b67dc
8914acb0bf671979d35db476247d280251fbf641
4200 F20101119_AABKLG katar_s_Page_127thm.jpg
c1df89f5a3f8ae0a681bf3e143535831
e9f8893cad113e02c8aad852f02396d0b391b0d0
27993 F20101119_AABKKR katar_s_Page_153.QC.jpg
9afca9f71b552d02e91e3b7d22ca7d70
7aad0344b6acc76dcc449849cc9fe066d5f20b94
37376 F20101119_AABJIE katar_s_Page_136.jpg
58dd98ec2f5deed7f2f1730c3b00a3fe
f96f5a38ae105df06123be5174db91b4258005b6
83077 F20101119_AABJHP katar_s_Page_118.jpg
91f8808783e1e5e9a055fe7ca509b613
f4f59d443f93d9286010dfc9737f6776d87289d6
240768 F20101119_AABKLH UFE0013881_00001.xml FULL
0bb0031f9a1640ba646d2ff64c014509
c8ad2e0ed413e92f6df40e067aaea8410db45b19
4579 F20101119_AABKKS katar_s_Page_148thm.jpg
f48bf67034e8d5fca94300be24f72897
8c59a123aef7465920310b05e03ed85dca7f9225
113270 F20101119_AABJIF katar_s_Page_138.jpg
2cbee7116a382c1859d9be65ce5268a9
a8f51ec336857ac518f6dd6f9e5e59f4d3eacdff
62017 F20101119_AABJHQ katar_s_Page_120.jpg
e0abc34024502e567d89b7c110e217a5
255386e3811b22f63e806d26865065013ba4dc36
607 F20101119_AABKLI katar_s_Page_003thm.jpg
f573eb29465dd65b4667d3f2e43313f2
d5654ed57e1d4e9efe5a1e30c3dadd9e3c37b2fe
5765 F20101119_AABKKT katar_s_Page_097thm.jpg
094754532332f235a400597dd974243e
d64296d18cf58ad3ae9deb397ee3fb53dfc29a83
92024 F20101119_AABJIG katar_s_Page_139.jpg
931965e73a9c70384f70addfca8f62d4
dae9eb4168e613c362580df64ac78b34a61faae8
56155 F20101119_AABJHR katar_s_Page_121.jpg
1a6a3a4375d6fd15bf86880eaf753774
7bdedb4094a3afcc93ba556a63744d3219b33905
3298 F20101119_AABKLJ katar_s_Page_004thm.jpg
a83dba6a03c697bc72a5ca38480b33d4
94ce0dbb5f6d2e0f0e3bc7c4ded651dcfed40337
21515 F20101119_AABKKU katar_s_Page_074.QC.jpg
1c2b8648604de6d3172a61f0b48835d5
1f5c1bae1b71a2a7cf18013121515b23de5b9104
67187 F20101119_AABJIH katar_s_Page_141.jpg
5c33e55d3d4efc18e48919d617efc64d
5ab20463b8b498bc518984e299451d7042f1b12f
58986 F20101119_AABJHS katar_s_Page_122.jpg
8e0b4ba3dc3abea2ea80e2016f0730f8
e11a9de7a335e2411701bfc509bdc6ccecd58e0f
12652 F20101119_AABKLK katar_s_Page_004.QC.jpg
8c2c9b73e3b01d85db172c2f1c7f8f6e
6f8a693802d476a28d8286e96b422f48ee0e719b
5056 F20101119_AABKKV katar_s_Page_120thm.jpg
327411c68b1627cb6ef6691c1c394553
b31d390018975a0407ff5ce9713e8d5389b7a37a
63640 F20101119_AABJII katar_s_Page_142.jpg
08b790ab32d58a87c5bc7b081661f4c6
6e69bb0a90c73bbbaf37b3f7e760f0e6044e4537
77878 F20101119_AABJHT katar_s_Page_123.jpg
e22879f22519ae6af174d340840a305d
7a72c2dd292764b67a7657212dabf2de6a6ff240
3947 F20101119_AABKLL katar_s_Page_005thm.jpg
c1ce153360f114044bf071d4842b6507
89f532c8b3282e7b025f3b12cff2b8c50a885275
4682 F20101119_AABKKW katar_s_Page_030thm.jpg
b1a28eb2e0f5763061561f6402e7d8d5
4897ad4cbccd77fd2f5f73d63c872bda09ce4dd0
59776 F20101119_AABJIJ katar_s_Page_143.jpg
f7c9012083e0acacc0c7b43834242fab
bd3a85d4e288e89deaf40cac54396ae63300e554
63210 F20101119_AABJHU katar_s_Page_125.jpg
4468975676921a219f6dc57925931d76
c6f360451f6ea0007f715bd27197cbfe8a80b8f2
25246 F20101119_AABKMA katar_s_Page_019.QC.jpg
39c79f6b11715e6b809d9e2bc97e0eda
8709cee960b1b163c2371449034b24acccd75aac
16206 F20101119_AABKLM katar_s_Page_005.QC.jpg
d7b32dd55111a477f9a3585e8786ce41
db8ac07cc214abcd05956b15f237a3b514549388
6560 F20101119_AABKKX katar_s_Page_138thm.jpg
7118952228632fc051f0f398af015bb4
34b11863d0532f6c6cea20bcddc482e0c011bc92
58154 F20101119_AABJIK katar_s_Page_144.jpg
7032eaeae22648820102fb2f35d02814
7b254f9dae966b5368b36f9765aba668390cf785
47389 F20101119_AABJHV katar_s_Page_127.jpg
9f02c852872316e029b3146472b5c3d7
f13191c2f425a7a0e41c0be0c2a422a07d2b73cc
8323 F20101119_AABKMB katar_s_Page_020.QC.jpg
d6f62c661af986db304cb74dc24aa08c
f3d0ae8d42d90b74429aa24a16e45a9a9582f49f
5846 F20101119_AABKLN katar_s_Page_006thm.jpg
0db9e00926ed22ab2627e5b2ac275329
83e9cae8c4f8c72025ed6ec5e34538c261b6edf3
1137 F20101119_AABKKY katar_s_Page_161thm.jpg
2f887cda6454c51dc01744180cc77438
bfe0ff16d324b4f0cb7e44c26c8247475bc60bad
56832 F20101119_AABJIL katar_s_Page_145.jpg
c54ea569345a5ab3e4bfedf350292799
06fbc8947b91e2216c4c886a0261cb658bad2f1a
69341 F20101119_AABJHW katar_s_Page_128.jpg
9efb4ffc67a1584eef813e2f09f8f8d8
604957681d9be02e89f99def7a27cec5856cfdf8
5154 F20101119_AABKMC katar_s_Page_021thm.jpg
c4a1da5822e49a5ab9700de7ea679311
daae21fef188d9b4920c8ba1c13b7faaf9647356
20228 F20101119_AABKKZ katar_s_Page_054.QC.jpg
d5fc85fdb72c88c53c7e9218fa310243
32887d4c5c9a2ef3ca5b9493defa06ecda921480
28864 F20101119_AABJJA katar_s_Page_162.jpg
1910f48d01bc06112b28172a2f9af6a2
693030a2b3e93526e539e9e5cd573a1bb2f80690
72696 F20101119_AABJHX katar_s_Page_129.jpg
fb2800047744d4470c9e3f99b528f840
4c57f7569581120f728eb4898c80bff601345c39
21610 F20101119_AABKMD katar_s_Page_021.QC.jpg
e126c99b8c78d12eba473ba4518ba266
88fe134e6e3dbd8d945dd4644e6a7d94d92dedca
1604 F20101119_AABKLO katar_s_Page_008thm.jpg
15202974b625cfb46cd3f9c8a17de2e5
5c63d8bdd9b61bf1715b3e62ebafe58bcb347d21
24792 F20101119_AABJJB katar_s_Page_001.jp2
aae10c5a7476fa22dbf907f8de0ecab5
6484aaa9812e817935a6c52f10b138a706686222
64112 F20101119_AABJIM katar_s_Page_146.jpg
271cf21cb2d256528dc63f0b8906007d
b46369656a738c6e4bb60cfa9feaa232ec0842e0
23914 F20101119_AABJHY katar_s_Page_130.jpg
f0e6dd9cdda41f5dffedf8ff068c5aa8
9f1bf9139747f853da7ae8c04a5d9eb2067558b0
6381 F20101119_AABKME katar_s_Page_022thm.jpg
535120f451bb3758408a83db0a372b76
4a21433399c8378fdccb666b8321980c6551f7fe
4957 F20101119_AABKLP katar_s_Page_009thm.jpg
6bbfec172f6e077e81fecb065b678982
cc11e36ff969eb0ac7882a9095da45fde465bc08
5530 F20101119_AABJJC katar_s_Page_002.jp2
40178f9426ced7fbec37003b2e2221f0
48d7e7348e0e0a40ab6708cb404a1ce43f4cffb4
77390 F20101119_AABJIN katar_s_Page_147.jpg
49e16ed48364bf00582660d5120fc2e9
0025832010ca38881af37cb55c17b2d27e1e3f09
50937 F20101119_AABJHZ katar_s_Page_131.jpg
2e981e53a5ef4b51ecfa2548c54e8c40
ac811ab6ec4c283c115085296b4e4ea800725352
27820 F20101119_AABKMF katar_s_Page_022.QC.jpg
0884dba46f1a3cf168794974c89fa3c3
f1a2e1ae0241bf3ee0eea1ce7869fc28c77aa4f2
7004 F20101119_AABKLQ katar_s_Page_011thm.jpg
639e1232cddc6dcb79eb4c13976d2491
97f0bef2c22f5e67b4e3e850183f11a2b997d9c7
69675 F20101119_AABJIO katar_s_Page_148.jpg
7de17d8d1230cd9cc788b4a82d58088a
ff8889987e7b765e455270385aaac2a56c53c332
7922 F20101119_AABJJD katar_s_Page_003.jp2
2633ad2cb403be6f88d65f22685bf58b
912ee03409df8d167b9df2bbca74754fa48f2e00
25156 F20101119_AABKMG katar_s_Page_023.QC.jpg
0220d37df20de618da0dfeb5c7de366d
023fe8538c2837eae912db32cf770bc00bfa884a
30334 F20101119_AABKLR katar_s_Page_011.QC.jpg
49689712e1fb76e1d8cadec403f8b073
1071893103eb14fef50c1af3d16586d5feae94d0
69456 F20101119_AABJIP katar_s_Page_150.jpg
0093c21d824341443389be27025b642c
e83406fb4beb95b925e80618c375ffbb88ea32f6
50203 F20101119_AABJJE katar_s_Page_004.jp2
e9e597f05a983cef02d9281e0c85c6a6
ddb5939d02ef8794398682a4acaeeaa91be2a276
5112 F20101119_AABKMH katar_s_Page_024thm.jpg
e7fde6ce9aa21f35bcbf09f9e55ce24e
02ad7dec57d75222b52af98f217c32d173aa6f34
F20101119_AABKLS katar_s_Page_012thm.jpg
39229aacf58f4e627211518462dd4134
3553b121e313a0356e896b9f19bdbda00c4d0a48
54614 F20101119_AABJIQ katar_s_Page_151.jpg
9e898282fbe6614b826eaf7854b0b95e
2ed4fd3eef40f3ecf3554a534788f26e789b1107
1051971 F20101119_AABJJF katar_s_Page_005.jp2
0ba1c2fd55a6f1ba9aa908813747553f
549449d7b79db8e74dd50a09a402bc2826d3ab33
21167 F20101119_AABKMI katar_s_Page_024.QC.jpg
6672eb0b27a5bbae02be769ee45ba76a
8a9eb819f67ec7dfaa1693f35c551cc7524a536e
20390 F20101119_AABKLT katar_s_Page_012.QC.jpg
68714522dde244c9c5aa2eec201d2fbf
b88f88531f513876308c1a7c33829d1952f538c4
69830 F20101119_AABJIR katar_s_Page_152.jpg
287c4a30b91d79d19a195996108ce341
96a6dd3b39ed8ecbdf66882d80bfd675bcef6cc8
1051967 F20101119_AABJJG katar_s_Page_006.jp2
da1d6920d95a294a0ecedf8da14d1e00
fec0ebb3993a9d497762a66f1b244895de6ce67e
5560 F20101119_AABKMJ katar_s_Page_025thm.jpg
0c4c453746b05757ea072ece01087340
4b03c2a8f4c430b1d1909c3fbff6fce1264ef42d
19563 F20101119_AABKLU katar_s_Page_013.QC.jpg
c16e2a54aafb0a30b6fcff278a0a34c5
09d799c941156cc70829d9dc88893e343913af28
96764 F20101119_AABJIS katar_s_Page_153.jpg
b3abd1ed34d5a3d26258f3ada74b1ba4
f20b64a8963b1b9e971b5c2137f551744aa88f91
1051986 F20101119_AABJJH katar_s_Page_007.jp2
5066826e4ae79a24bda55ba21f0a9174
7befa7f7d3d2d3ba33340a68abe061e62485044d
23887 F20101119_AABKMK katar_s_Page_025.QC.jpg
44a1722f292d4550661e4d432d5f19e8
b9418821c449be4199df57fc6b09b703029b3cc7
5639 F20101119_AABKLV katar_s_Page_014thm.jpg
889b1f38686527ed6c54ca140d7b3ae2
4fac00dfb3410223f96b2dce8a8b73ff5faf60fa
90565 F20101119_AABJIT katar_s_Page_154.jpg
a4437f76c923e4b9bde6a0dbd619c8cb
7129f1d1a25baa2535e5a68b6c3167a1fdc3f3eb
366533 F20101119_AABJJI katar_s_Page_008.jp2
34175a04ae7451d1c5b27a6925bed624
d022cf5841e2af3e07b4c22226811b0f7cc9e197
25601 F20101119_AABKML katar_s_Page_026.QC.jpg
7d2358d5c88256f2a49e4ad0b599193b
22575fffd76bb9b302cbed88d839b621a115d615
23364 F20101119_AABKLW katar_s_Page_015.QC.jpg
1dd1bc9919344c6f75346d5498e7e0b3
2e00130112b849f06753e4ec995b2aab20801ad1
82937 F20101119_AABJIU katar_s_Page_155.jpg
bd231323f87232dae2177e319d7191f1
3ce8a761a18f90a5cd99f147e504fa5730d6c04b
1051985 F20101119_AABJJJ katar_s_Page_009.jp2
1cdda1ac4ea52abae5ec143b2a15e41b
2a1008dabe10eb19da34c2c05b42f188dbc46522
26138 F20101119_AABKNA katar_s_Page_044.QC.jpg
c1ceaced112bb8130a3e1a9c23924aa2
80feb887296d21be9e27a7a99cd4f487273a9012
6236 F20101119_AABKMM katar_s_Page_027thm.jpg
9aa1e0663df53b87abd516ca987b8093
d094d4689a52729dea67f67eaf63a69cda97eb53
26179 F20101119_AABKLX katar_s_Page_016.QC.jpg
6299c0c084d14380f1481b45f59bf5b9
e9a23f8c45528a245a4aaa81a6af5a5250438e5f
63029 F20101119_AABJIV katar_s_Page_156.jpg
611fed1bb77681e127baa44375cc94cc
6631d0e53e2a7ab67e2fe52866c8d5f57d3afca5
1051978 F20101119_AABJJK katar_s_Page_010.jp2
c3e00a3c4c30a956e71752ea11ebe76d
9fd664947d98a5cbbc709ad825085aaa1d53fa83
7031 F20101119_AABKNB katar_s_Page_046thm.jpg
24a7dfa7fd7fb4dda482b5f04d909644
2f074bb5e08eead1b398e44dff5debe4c5d545cf
5033 F20101119_AABKMN katar_s_Page_028thm.jpg
7148e7f07f971cca9e12a1dfc082a7ab
1eef90499bb46d50f4183dfeddd93453b20fa4b7
6272 F20101119_AABKLY katar_s_Page_017thm.jpg
c4caa5064ce1dad2124fce76ca40439c
a1dc3251e302ade82c2c8c4af4b3cfd718b05b30
94005 F20101119_AABJIW katar_s_Page_157.jpg
75e93e4156623774b60996c6d9c9187f
d52aeaf57844fe6d78d96bd0d4ca1fa1129b142a
F20101119_AABJJL katar_s_Page_011.jp2
cc5e3977ead084e9724665052c0984eb
7b1de68e690e9f39ba77288230ec95443bc1e65e
29669 F20101119_AABKNC katar_s_Page_046.QC.jpg
21bb744864fba1e51283516e9d747e59
31ab6b35a83124d84618e96d2cc25189eeccf2f0
18354 F20101119_AABKMO katar_s_Page_029.QC.jpg
1fce220b50740a4982423f77b478871d
641803a3a8b00681d7ea5ae6f6b2f03d4dcbcacf
6173 F20101119_AABKLZ katar_s_Page_018thm.jpg
af1b03a88855f2e7687554b61e4900da
f951047b1077df6c16df0dc688fb35f5bdc3d203
108898 F20101119_AABJIX katar_s_Page_158.jpg
0f346f0547cf24bbf8c8738a62323349
a718a725cf81615e05ca0161ebed5d2ff6b7b794
1038939 F20101119_AABJKA katar_s_Page_027.jp2
65c051775377b5e5df83a30a41b2cf83
80e4dab0c0ed0ce18350a4501af0f46bd5cf8cab
1051982 F20101119_AABJJM katar_s_Page_012.jp2
461db1fb796504f0f4ba978f05a02a4b
a9bac4fc272b57f0f25e13c14ca4f542b06d551b
24342 F20101119_AABKND katar_s_Page_047.QC.jpg
edd4413600ec8d7e360f7d7feba13c56
fc71cbb7835cfb99a89808e537319c3f0ab3fb51
101812 F20101119_AABJIY katar_s_Page_159.jpg
e01d61ad43035e8b6836ab4309fd6e6d
3bd15c9df84356f509c1a1c5b051c6f37e4a0276
942349 F20101119_AABJKB katar_s_Page_028.jp2
055b377ae495d158fec9926e25972bfd
c0885aa6aef4b1a5f9e124f772bdbc15034daed3
3638 F20101119_AABKNE katar_s_Page_050thm.jpg
8df4c8a03fc2f032f56ebfec0fb3b267
cc1481888b6a54608336060c7ca94eccdc33395e
5278 F20101119_AABKMP katar_s_Page_031thm.jpg
f211d67619a3aa937f5f91e13e7859cd
e43297e69aa229303d194a9db82d99998d90d742
104442 F20101119_AABJIZ katar_s_Page_160.jpg
c721c86ed6e60ac7fe849edeca38212e
f1d3c0a24a9334ab4ccdffd73d235ab9014375ea
918680 F20101119_AABJKC katar_s_Page_029.jp2
f2993053ca1bd6cf605b7d78d92f5d7d
7966d46b710d7c891a06c4204df2b2465cb7c369
83792 F20101119_AABJJN katar_s_Page_013.jp2
eaa101fd153b84fed78b15116ee48ab5
91e2e5d1751d23c2aa603292cf8fc140669a7152
13701 F20101119_AABKNF katar_s_Page_050.QC.jpg
aec664a366354d7b425ab7f9a66a5a92
92a21b1ad4205948fbd375c5569bedb487493218
17122 F20101119_AABKMQ katar_s_Page_031.QC.jpg
78233f4d4bcce21ec9190d9e3390f971
839df1659f9d43f764c1314e466abb4ef26c2987
1051984 F20101119_AABJKD katar_s_Page_030.jp2
d1b3aa75cac55f9de7d5c93c48a7dbfc
982ed66d1a1669b6951cd0ab3cc254f24d84cf72
96738 F20101119_AABJJO katar_s_Page_014.jp2
c95cef54b7a6fd714badb1dc4b2d3846
5362546c163ed47eba1135ab825d2acca5be6fa5
4228 F20101119_AABKNG katar_s_Page_051thm.jpg
af59637e222bfcab0311277a00e9b765
fffda6266030ce0843e207043aa56642f5ef20cf
5861 F20101119_AABKMR katar_s_Page_033thm.jpg
1b7779bed86e184c6fdacde94fe83507
24aaea42e91caab07ba262f4904109844546cd46
1051918 F20101119_AABJKE katar_s_Page_031.jp2
df3403eef1423785122b7ab6eac78953
6d4b83cdd9207519e7f659261798fd9d42fa4556
97570 F20101119_AABJJP katar_s_Page_015.jp2
5a4f343eb21ecf6b75ed407750b50cf6
f0ba06654937373ae08882c3c38164800c9367fe
16136 F20101119_AABKNH katar_s_Page_051.QC.jpg
f3fba5d439085c90e23628f861689ef6
3c28dd5f31a85a2503bcfff5f9a0055f026558be
23889 F20101119_AABKMS katar_s_Page_033.QC.jpg
be9b5d265baeaac12bb1951140c555d8
982e48307ab32d2c1cc3b59d268065e3fba20fd3
81683 F20101119_AABJKF katar_s_Page_032.jp2
fc73e300cf9ee578247be744b0cbe400
1ac290ebad920bc36d2a9bbe9d59074e2d3ead11
1051977 F20101119_AABJJQ katar_s_Page_016.jp2
9607711b584b9cf45e77f862e7ef189e
9b593875947218a35c452a3143824f045b0fc753
6452 F20101119_AABKNI katar_s_Page_052thm.jpg
e373f5bcd979626e0f6013fee684bce0
6cba59ca086727503e1e2e80d318e17ca2446bf5
5945 F20101119_AABKMT katar_s_Page_036thm.jpg
fc28028e71e04b1f4889679a4cd51ca1
b95b2c69bd57f88d40c48d9954168a573b972bd1
1008054 F20101119_AABJKG katar_s_Page_033.jp2
9839b3aa95deaaf6a248fcdda2e5b1d0
8e661cb088535eb1dd6a9bfa48b0a74552711cdc
108082 F20101119_AABJJR katar_s_Page_017.jp2
aa9e6e813efbacb9b337ddc17a860bf3
ffddc82b7477cead9fba060ffc3ead643cdd9004
25722 F20101119_AABKNJ katar_s_Page_052.QC.jpg
2dcc61ce75bb85bdba34761d83a0fb31
a181553d84d07ca038266187e7f7450f64ba0610
6302 F20101119_AABKMU katar_s_Page_037thm.jpg
2be69885e57ab154731a7a4f25e364f2
782a4059b330d0854e03ce14d1117f87471c3342
1003534 F20101119_AABJKH katar_s_Page_034.jp2
c49aeb31e1cb043f8c6bfa5c97887dee
ea38d39e02594e2927f6c67ee840bbf4561ccf6c
108134 F20101119_AABJJS katar_s_Page_019.jp2
302958d572c452431903f2383e314e41
bd9d0482d77293fa526200e3a522c47a864b9e15
6260 F20101119_AABKNK katar_s_Page_053thm.jpg
1603bd33f279e3411844c2c51edc444e
4de7543257f889d96d184db11b30b1e0f4981a44
5960 F20101119_AABKMV katar_s_Page_038thm.jpg
ec85cc4cfe3e336ed37710ba59ae4a4d
2bb12b980e78de857d67391f09f6f85826fdfe4d
972388 F20101119_AABJKI katar_s_Page_036.jp2
217ff425736dc6a2267fe5fb9ca4ffe0
fb499dc0a18e81cf60c1ed7783d7952067b6e8b7
33187 F20101119_AABJJT katar_s_Page_020.jp2
3b5a114327e576e868dc93e4b20bd008
e84c399f1427ae7acd038753eec157fdbfa03448
26002 F20101119_AABKNL katar_s_Page_053.QC.jpg
7c3481f29636456a9e325e02649a99f6
8ee4600f69ccf0c405fa877f3498b49b46c2f371
6206 F20101119_AABKMW katar_s_Page_039thm.jpg
6109b448da62f122fac8395ead06f2d1
e212ba48b5e2d1e66d4acf7deb9a0e81f0f00d9c
101073 F20101119_AABJKJ katar_s_Page_038.jp2
e7a79ea974d2eb0d5b699130fdef40c6
768cc00738969514f25ca3bbd5424d9a1c294505
96078 F20101119_AABJJU katar_s_Page_021.jp2
5f6b2cd4a8917212d81c28bf06429e75
c5a452f610ef295b306f56a90822022da0d3941b
6026 F20101119_AABKOA katar_s_Page_077thm.jpg
9b45f52c30d473e7b199bf5dc5aa0eb9
8a553fec63d0285d77c62e7664fafcc664cd84fd
14267 F20101119_AABKNM katar_s_Page_057.QC.jpg
0795ac14019f7ac3f56d6239cc3db689
9473f19316790257286d5382aebc596610aa7593
6552 F20101119_AABKMX katar_s_Page_040thm.jpg
19ab44b5a22549490e0154e54f49b3b6
be69e55889097965ff2dc9a615a607b2b297e73e
105569 F20101119_AABJKK katar_s_Page_039.jp2
e974d19bda429a71f59c253a950426c1
c16980e93f8ed86c1d910dfdd4150f898406cfda
F20101119_AABJJV katar_s_Page_022.jp2
097ce69d60741e54ef62d65613e2a99d
c4ff3c9173883ab790b470e28b90ea29d0b88e38
23892 F20101119_AABKOB katar_s_Page_077.QC.jpg
5aa3bdb41d39b296d5b4fab7196cd276
9e40256e0fd067b954f0f6202c10499b14070e0c
27550 F20101119_AABKNN katar_s_Page_060.QC.jpg
e9187f91e4dfa907d59b148d86cd84b2
031c7667a4dc676dde483b228849ca476cd54a65
27012 F20101119_AABKMY katar_s_Page_040.QC.jpg
5c2f37d5c5c9ba6464aabb2970a9c9ff
46e52ef6759524c5da328a36dacfc4a4965b64ec
529270 F20101119_AABJKL katar_s_Page_041.jp2
dd82c044d33f0a523602078e25306c64
b67e8e8c33b5c1a89dd863ac94d4055fd0390df3
1033268 F20101119_AABJJW katar_s_Page_023.jp2
80c76158e56b77772c710d6aa2d288be
aa4a67adeeda0d352f88d65c02cf996309cae8fa
21721 F20101119_AABKOC katar_s_Page_078.QC.jpg
124b0ba5dd69940a7771d5ba1b67dc9e
2d58e33fff36c53f20e9121e71e1052f7ca8eb16
6503 F20101119_AABKNO katar_s_Page_062thm.jpg
ccb5a248d555784fce631742414b06e4
6afc9169f4941ec36b85eb8195a1019e19aa6a20
3740 F20101119_AABKMZ katar_s_Page_041thm.jpg
1049fb8cf2d7a860a3311efbaae615c2
01f337d62f7b7406adf59918db6fe52f5fe4c611
1051960 F20101119_AABJLA katar_s_Page_056.jp2
2ac1e162a016f67f18be38d8f9e8a15a
2da4a49c1a4adf7878864f695bf898a0cf120690
1051941 F20101119_AABJKM katar_s_Page_042.jp2
47cd54a1996f22c7022d9672fd2819bc
8eb321b014b180ef8bb2ea75ffd25be9f0d889d8
913388 F20101119_AABJJX katar_s_Page_024.jp2
61b28cb3303a0fe5acd155195599c18a
deafe5d726da9dd8342ddab71e8b29b9fbad5fc3
17732 F20101119_AABKOD katar_s_Page_079.QC.jpg
7b2cf61bc867414d10873c03cc515d50
96d66537b51bc33273794a7ddaf08efac8aeaa90
26564 F20101119_AABKNP katar_s_Page_062.QC.jpg
4484f79dce54d1d2182f6bdd11aa7bdb
5f8b0bd94ff7508dd7ae60838972a21f4ebfa55c
566249 F20101119_AABJLB katar_s_Page_057.jp2
f48a5dda0363b86ce31708d50aefeb0c
95faa41de615acf43a68965eac40faaaa8d178fe
1051959 F20101119_AABJKN katar_s_Page_043.jp2
f3ff4041c105ba74f402387af35aef37
3bf5de07d48e201b931d17b53c964c1093b08114
100243 F20101119_AABJJY katar_s_Page_025.jp2
3bf99e6c1b041610033803257675ff9a
f1ce5de6f9ddbebaa693f895b7f8d7943227215e
18684 F20101119_AABKOE katar_s_Page_080.QC.jpg
022d594350850599337932ed49baa1f4
1952cc22a18af52c72ee416944a0b48c5931af22
920381 F20101119_AABJLC katar_s_Page_058.jp2
aeb2ea12ea8c480b116dc66ddee23321
a0dc5b80f06c5990010a6ae380b13a2eb71866bf
105794 F20101119_AABJJZ katar_s_Page_026.jp2
58e3753a847633c445d3f524f55c2ee6
27a13db7367ce6754faeb7ffe8b18221595be689
16728 F20101119_AABKOF katar_s_Page_081.QC.jpg
bc6f1eb048b7331185c012d53fbe180b
86e08c20427f0fad9dfbcfb3116efdeb27be1f7a
4861 F20101119_AABKNQ katar_s_Page_064thm.jpg
9a22749cc78d68c427d9f75826d3f934
3d403ce11f27c8df04c69e1045a0da93996943db
109449 F20101119_AABJLD katar_s_Page_059.jp2
f26a1f3155bd1e447b0a417a35b76f54
28d41466fa74a83ef5a462451741b880a44fe549
106777 F20101119_AABJKO katar_s_Page_044.jp2
652d011f9754f51e3c5d58e8b0c743a4
2d865adef9be78f874ec4660b6195106642f845f
5470 F20101119_AABKOG katar_s_Page_084thm.jpg
59c2814550a28ad5498a31dc91dbefca
2dc7b7b7f7f252b30bd949062a9217476a8e245a
6237 F20101119_AABKNR katar_s_Page_065thm.jpg
09f0a3532b8fe7ba1b4f5f909f3bf5c0
ed792d9cb694aeda047641a5cf6689324b8b5f83
1051934 F20101119_AABJLE katar_s_Page_060.jp2
b33783e94e4dc2168edbca5ce36d0dba
914ca9a969f4e65ce7763855d3db05ea9acdc062
110617 F20101119_AABJKP katar_s_Page_045.jp2
21a419f16114760f554b0b4e0e17a01b
699871d60dafa24e7c1ba8132da6bc09fba8dc02
23132 F20101119_AABKOH katar_s_Page_084.QC.jpg
7743c125cface54c3ecc6198e205a04b
17c726000c4b1d1b66e38240c16b02e62f78e4e9
6028 F20101119_AABKNS katar_s_Page_066thm.jpg
65e45a178b3c2d9c202970dad6497717
fed360f80b81dab013dc3cb7016263d13fdc79e1
723127 F20101119_AABJLF katar_s_Page_061.jp2
66ba171f3874fdb91575ec2bd1599e55
c2f790e9ed45fca881f46f28ee80c0cb054e65ae
1051957 F20101119_AABJKQ katar_s_Page_046.jp2
3b48c79d3f648f972a8a1e02864ee1ba
ec617b3b1ab147dfd781238bcc2b26906be43326
25967 F20101119_AABKOI katar_s_Page_086.QC.jpg
8dc20b2d8d88c54f93b27320cbabec8a
2ea1caac7572a3fa8100dba2316676a58bad1abc
24222 F20101119_AABKNT katar_s_Page_066.QC.jpg
e5c18ce83e3f1fec13ca3ece2e8bcbd0
74ace7f4b7653dd62e459ffea7b24e7b24e5b1ad
1051973 F20101119_AABJLG katar_s_Page_062.jp2
766ed73956214780724a5e6ba5eafaff
4d8ce746b5fe761c1f08ea0d5e91f9b8e8fcbbfc
98637 F20101119_AABJKR katar_s_Page_047.jp2
f2a9f62af6ab7587e0f46d4b16267f0b
686a95de5c26980c9904f5d517b6f1a2139223f9
6082 F20101119_AABKOJ katar_s_Page_087thm.jpg
4d139e2406fb4150b80853d6082236b2
ebda3c19e7e953819d3506d5ce4fbdb31fce1b18
25321 F20101119_AABKNU katar_s_Page_067.QC.jpg
4b8486899c409243e06711abfc593d1f
0ed69631939672460a60a7b4b0b878d769af6c0e
F20101119_AABJLH katar_s_Page_063.jp2
db07450b2ba5ff28b059d706756c50ba
b4ec34c8387cde08cb6948ea22d5df0ebacfd6a9
102561 F20101119_AABJKS katar_s_Page_048.jp2
66b719bc352f21367eb10686da0c72eb
c8ffd8ff8c2a405489598423f45550f81fd45e33
5471 F20101119_AABKOK katar_s_Page_088thm.jpg
73b617aa4420fdc2d785079f0883d69b
3e979f2f6f0f269122f26c2453ea95302103d808
4703 F20101119_AABKNV katar_s_Page_068thm.jpg
264681ac2792550b3d1eecdeab496b31
8a2533b779ce42fc8130c278ebfafb597408cd6d
76618 F20101119_AABJLI katar_s_Page_064.jp2
588ef1e0e81ff2709965f19ceabfbdfb
9a04d9b743cd56e06719710f77c611eb52fa096a
791544 F20101119_AABJKT katar_s_Page_049.jp2
7346c190c222ace60c2b9335b347c7ae
3ab7275e42e1441aff154066267e9a62b3496292
21665 F20101119_AABKOL katar_s_Page_088.QC.jpg
5110fced534e787ec8805eeba0c4d723
35c95accf28b9f90278b71e94ff8fa71ce6e0112
5760 F20101119_AABKNW katar_s_Page_070thm.jpg
b8f10ddf65b27c11b6d2735e9d2722c9
57e7958fdfa48991322716eacd78991b62d452ee
106963 F20101119_AABJLJ katar_s_Page_065.jp2
e1d63f16cd6b29c332c73d06889b9d03
01ff575136638cc47377441d592bfd5df1001eb5
535501 F20101119_AABJKU katar_s_Page_050.jp2
e8efad274cc8a017677810c4665634c6
ddc18ff7941aa05ac709fe22a782ca077a51161d
6403 F20101119_AABKPA katar_s_Page_104thm.jpg
a164886d186f6071f4730721ec6cf95f
6bfe4b91725ca9b73e5b4fcfaa4b67db1b5d8459
6277 F20101119_AABKOM katar_s_Page_089thm.jpg
497b1f335b385f97cef49d482871b8e0
1b545214e3a45475591cc894563c0b7390dfc2af
23507 F20101119_AABKNX katar_s_Page_070.QC.jpg
a95c90de1232e48df9d3760c43078b53
67613e5b9acddcf7f567d5f1e8f707ebadd78384
104226 F20101119_AABJLK katar_s_Page_066.jp2
986f884507c33bce9e070a45fcb260ca
3b8a89515d0666badc78e31816016ad254675895
746748 F20101119_AABJKV katar_s_Page_051.jp2
0f380f6dfdda7686f2737cacc14cc127
9b440fead43e82a992f975d2d368a30f755f0909
27319 F20101119_AABKPB katar_s_Page_104.QC.jpg
6f240eb03921d5f6c23bef65e1047ab1
637c22e6203b70c546fb53cf836f0f35668e9505
27191 F20101119_AABKON katar_s_Page_089.QC.jpg
16492d4e938c76303cde6f64cdb2156a
a6058832a8da03c370232124a4928814e4c726e7
6518 F20101119_AABKNY katar_s_Page_071thm.jpg
5f25b72f53ec92b2c1f23fc385206c19
f2d259ef3d7ec3b83217c73c682b79a19f1076ba
109812 F20101119_AABJLL katar_s_Page_067.jp2
4fbd056d7b9ccbef2b99c35ea642d0c8
2e818016e8bce87ce93cf347d85d1c22231875d9
108956 F20101119_AABJKW katar_s_Page_052.jp2
48227e4e3e6aa99a24afd80e4ed47645
1037ad5782fa8d543f918a37fc7f0a3d4ec975dd
5210 F20101119_AABKPC katar_s_Page_105thm.jpg
ce5cc0ebae4c9e2b631db10e1c3e9fa9
b90d99e7bdc5ec256c5ff1a8f3d5860cfc4b6314
5574 F20101119_AABKOO katar_s_Page_091thm.jpg
7c474da96a6ad7d963c5c3eb5acf1d0b
2ceee839b10bc6b6d97e0c499b05989d621b974d
24091 F20101119_AABKNZ katar_s_Page_073.QC.jpg
bf0e09eab6867931d93fc3fe07ca839b
60aa46942a5c307d5a2e5fa9f3c90d1cf7759d83
1051919 F20101119_AABJLM katar_s_Page_068.jp2
f44f4b6b8d369b93ecf41c670127442e
ab9d811b0472c5b5fa14085edaaf7d8f0038ef54
107562 F20101119_AABJKX katar_s_Page_053.jp2
0640291b1bc1b765632667989fbbe406
dfde8370c3222bfbe9202f14dd8c6e2e23f93f91
755068 F20101119_AABJMA katar_s_Page_083.jp2
0be69da558afa55db7063c464ac3882d
087a428a7cc46e18873976a1fc72a3a6da199760
5153 F20101119_AABKPD katar_s_Page_106thm.jpg
c9c963e62df37bedc2f06a277fb6e5d6
bf6fa34fc15414b417d7122feb897d151f302e35
5164 F20101119_AABKOP katar_s_Page_092thm.jpg
3d166e468d8cae52ef596d1acaf07f6b
7f12463472deaa381a0acc11e384c03be49ced08
297282 F20101119_AABJLN katar_s_Page_069.jp2
15dbad649a21cfb71b6598bd9ea877f9
0df4474046c9554f78e8b9be8d38316d013d4abd
84294 F20101119_AABJKY katar_s_Page_054.jp2
e495b9362eb38617991d6c2fbe84ab62
c55def7a29113dc17898a87375026a2011d583b3
94885 F20101119_AABJMB katar_s_Page_084.jp2
80db7ed2a02cb40069af275e121b52f2
ac9530cc430f4c7ca924af54f68528ee879dfcc3
5688 F20101119_AABKPE katar_s_Page_108thm.jpg
8275a2762c0987f69198cf6024ccd321
39e55ab2c44ef62451275e3bc4973e9d3a4c49c6
4727 F20101119_AABKOQ katar_s_Page_093thm.jpg
e0ee07257074c6f6f66c6d6354e446be
25fdfdead7667904a01c2ff04d0f1f93e1764df8
1000958 F20101119_AABJLO katar_s_Page_070.jp2
7bda8824c711fd8f227c520d656dac17
f0d058fa1ed97d1588119c0a5473a5b4a81bf6ba
1051949 F20101119_AABJKZ katar_s_Page_055.jp2
3f1752cbe52e36af0ecce8a2b4a88942
45d8b6a4de274e0104593c43de0f4d7f04641d61
1051935 F20101119_AABJMC katar_s_Page_085.jp2
2601f837d0b756afa389d44a62dbb0a9
40fecb2bb6fc844ebaf6cca99be9de9d9c305669
5840 F20101119_AABKPF katar_s_Page_109thm.jpg
c8d977de17c3fdd4e1f8b5e066d47982
564f6479f4608874161a432249b984736f48e0c1
1051974 F20101119_AABJMD katar_s_Page_086.jp2
7e09cc61bb5ceaa3b0be669a4decc45c
7cf3b6cfdce46898c1a70fdda8a15022f47bdef0
25138 F20101119_AABKPG katar_s_Page_109.QC.jpg
0a94b2585c7d304982d979a87922fa51
c33415191cbe3cfb4d8c5a1ea9f70c43054b4b01
19248 F20101119_AABKOR katar_s_Page_093.QC.jpg
306e6076cc4673f1cf6afc5438a5f46b
3ff954a25434b1f91b843f446c29085157f12337
110444 F20101119_AABJLP katar_s_Page_071.jp2
ac58ee73a5aa1908e22f29bfd96761ca
3e11a31f7feb066ce681b6542ba02285cf0382a3
108504 F20101119_AABJME katar_s_Page_087.jp2
8a2dcf30c89275079f684fc95d4c5cfd
fdc98e9bc51678d8625a7659ff14e6e4e2c62cd1
5925 F20101119_AABKPH katar_s_Page_110thm.jpg
176c4060f55325b4a97cc53b09b1c2d0
509e1d466fd57e4efa7780e5d39e4bedba2bfa49
5649 F20101119_AABKOS katar_s_Page_094thm.jpg
7443e6ccd87b544d136a96993b531163
630b0290e18526726c12c3c5d26a19b496d4f648
703910 F20101119_AABJLQ katar_s_Page_072.jp2
6de15f2716a8e89160f5c542593d8ca6
a122cdca9f1bdb85a3226ee1b790b0b7af465818
890310 F20101119_AABJMF katar_s_Page_088.jp2
95680e4cdf4eecf13e8a8500eedb3400
e177929c6fc57cc6ae862bc95e107203a2430550
6375 F20101119_AABKPI katar_s_Page_112thm.jpg
7891a1642194a2d63461d2806ffd4844
8a2ec667b7ebf5122ac54fa44b7a12e068a86c29
21663 F20101119_AABKOT katar_s_Page_094.QC.jpg
1eed4b39b7601d6be59369df4eccf243
ebfdfe7fbb195920594545f01a843174953d489d
97183 F20101119_AABJLR katar_s_Page_073.jp2
2213e14d80faa0007af331bfccfc34fb
3c824a9ace4bf5e2e939a75ce1c528f9ce6950c5
125126 F20101119_AABJMG katar_s_Page_089.jp2
a031d02f818284588b95f607fe8b7bfb
59980547c9df9e377bb71e797510abdc218c4501
25409 F20101119_AABKPJ katar_s_Page_112.QC.jpg
5eae625bb9c7fa0169e10411a93a1366
3c7b9b17b7b115dd6837bb0ad5cf5b8641c49021
5746 F20101119_AABKOU katar_s_Page_095thm.jpg
25c71de4cf0ef123fa894b97e754d2ee
3fa1a9fe4745f051ff147abaa01b56c913ba622a
850297 F20101119_AABJLS katar_s_Page_074.jp2
69e5fa6625e59af0a4535c526912eace
d7a0f4aa8e7fdd4d59b3d87b65744f59dceff596
81745 F20101119_AABJMH katar_s_Page_090.jp2
7ee3d12d8e39a7e42936471d793b378a
c670c8895a810e1a2a0b2b37c32ebf2a9dfc8d7a
5521 F20101119_AABKPK katar_s_Page_113thm.jpg
f1bb8eba002a1f411fb5f41bbfa0606c
b098175f94b06d7936017fae740bb396146e1e81
24085 F20101119_AABKOV katar_s_Page_095.QC.jpg
054dc4fd056474c74b65139116e00850
35121a924395a4e1b8dc69cfa702db2d9d83cb68
805480 F20101119_AABJLT katar_s_Page_075.jp2
aa700df1bf5a10c3768bb35c06071078
338d9316c8c7ceca62283fceacab5d6712f317e8
875555 F20101119_AABJMI katar_s_Page_091.jp2
3c9b5b9562d5be901370eab6891acd3d
88e1dabcfce188d9138c0a7fb56bdecd919d0d02
20520 F20101119_AABKPL katar_s_Page_113.QC.jpg
0af03fd04779a74fed04c18705f3d4c9
89676e3993f134b80981e2b2146e39a604b50880
4233 F20101119_AABKOW katar_s_Page_096thm.jpg
9a4b467b9c55d8593e5ebb9f7862d4ca
95c92a280c60f0297e867e2d5e5719124c2e9f66
508163 F20101119_AABJLU katar_s_Page_076.jp2
09f16724a5a836fe10e4dc23106cb0f9
7bdc5867e4f858416129e77104dfc2a938ea3d72
782486 F20101119_AABJMJ katar_s_Page_092.jp2
a0b36c255c353e3e0577d1360cc96383
c78c34365960acefa697a185202aeea4a41e25c4
3319 F20101119_AABKQA katar_s_Page_136thm.jpg
2b3c1e74051b922dbd627cfefd9e6803
53f9df6ee4ef0b05438593cff353b22319644aa2
21226 F20101119_AABKPM katar_s_Page_114.QC.jpg
4d60e1d7b838fad88404d42a9595bfe6
fce220e738c9caa05c3ef1eb500ec0a1677fa482
3447 F20101119_AABKOX katar_s_Page_099thm.jpg
147e23791fb90746b28c4e7aca655dba
8b5d361b636915389d249767c4e12890b4886f90
976283 F20101119_AABJLV katar_s_Page_077.jp2
df3a739838a6e587c87e0f9027ef6eb3
5ef8bf0e035fed5f3eaeb084e6d423fa02956129
993099 F20101119_AABJMK katar_s_Page_094.jp2
38071b9a0f469fbb9018bb8bba88097c
cb98fd824c4c0304557d6e1d3547bcbfda3a5a4a
11811 F20101119_AABKQB katar_s_Page_136.QC.jpg
264293398ace05fd2ac4343bacf6eccf
270a2c162f746a362a26bf14c13381a03c062b95
16954 F20101119_AABKPN katar_s_Page_115.QC.jpg
1ba1cb39923bffe08471867758e1a15f
fea3a9066371ae09e7f1f60edd9b0fb378c456fe
4591 F20101119_AABKOY katar_s_Page_101thm.jpg
86710986c7be1d5ecb31a25813d690be
0351e463fa142c78f2bbef05fa7eeb812d7c0179
90748 F20101119_AABJLW katar_s_Page_078.jp2
63450d8adf8cec69c43d93a2a07d6ffb
733c1b1bd93b77a45077f46e89fddf3fabe30ab0
98302 F20101119_AABJML katar_s_Page_095.jp2
889e9da042d58d9880c27d29063fa93a
5c966ce3f82f495a1d562a3dfe736036f0efc0ac
5618 F20101119_AABKQC katar_s_Page_140thm.jpg
16e767fc258b66dfeff83e9b81f7a886
edc6295a69a8626111470854b4a921edae9396ae
6143 F20101119_AABKPO katar_s_Page_118thm.jpg
742135f5004e7392b1c936ca67fad368
ae2760f6160b7e1d86487bc18c4a89b8ab4ebb0b
4430 F20101119_AABKOZ katar_s_Page_102thm.jpg
2863f75ad704da2fd3de9ca9d7f710af
ad60a2d6c14c411c2e5f1f52cac84c13d49dd74e
764050 F20101119_AABJLX katar_s_Page_079.jp2
3b05b93a6018677b0e57b095fbbb9267
1963cc142469f1c76b20847d316ef1cd5a9a63c3
107818 F20101119_AABJNA katar_s_Page_112.jp2
8ce1ffd5309063f97b2eea34f49e6bf4
d25194035ee84aa668d51227598b84666856dda8
743799 F20101119_AABJMM katar_s_Page_096.jp2
b7a21a28863284645bda3d2c5bbe320e
6802a74ed75ce8eae2a2192e944f489c5c048168
4529 F20101119_AABKQD katar_s_Page_141thm.jpg
d37a7110e5e6e4b65972e10ecc911232
44a66dc8269fb2b9d1ea460554e689444e4fa5fd
5740 F20101119_AABKPP katar_s_Page_119thm.jpg
242d86d7f20177f0626c610e32e016c1
b2bdaf74db7e8622d4c56f9a991f092764a9515f
845610 F20101119_AABJLY katar_s_Page_080.jp2
5ac07e0360f647861b70fbeb2b2232e1
479610ed909ab0ebf7269a5b2143ee2d39d8ac70
74480 F20101119_AABJNB katar_s_Page_113.jp2
4a5eb29e3cff3bd033bb440eb4085892
18d57615644a8f2df7fb86e6f9abd350e569825c
F20101119_AABJMN katar_s_Page_098.jp2
a988b3aa15e17cf82d276d2aa28cd6d6
11f5b4b35c3c07aa68e48052e366fb4af7ff7138
18223 F20101119_AABKQE katar_s_Page_141.QC.jpg
3fc59d780ab679ce79956567c1211983
98a23a84fbcd2de4c1c7b92df9f57d599163a6ac
18293 F20101119_AABKPQ katar_s_Page_121.QC.jpg
2e0c89a1a632fdd8e05dea3e218f2f40
ea68af536410f8f7d1128830b22c741535e46ec2
672407 F20101119_AABJLZ katar_s_Page_081.jp2
cdb409a4b25dddf544f8b20812141f0f
2edae082abd2dec2df7479a83601c101f9459a1a
79488 F20101119_AABJNC katar_s_Page_114.jp2
4e034fa4d3d2b6219ac4b7b0d36ff729
ae6b5d9186aa323111c82645720145fb37ba4a1b
661325 F20101119_AABJMO katar_s_Page_099.jp2
51d261654b3cc3a010d10204c5677d0b
7ea55157c4220b6db0192a697bd8104292d592fa
4225 F20101119_AABKQF katar_s_Page_142thm.jpg
b9dbfe20ecc082a3f32f7e1d254a65a9
926276ff5680a0e689cedf2749873495dfb34d51
5301 F20101119_AABKPR katar_s_Page_124thm.jpg
cb08adc44001ebe78dc34e6e8b5a4e2a
f22ef35a5a84b1726eaa5feee95f390ed4b91102
1051962 F20101119_AABJND katar_s_Page_117.jp2
221100af6ce8b10f4915ce63cc8ab73f
da3fb1525e1a663681b4696bc38824c0f994e8d9
868010 F20101119_AABJMP katar_s_Page_100.jp2
21c502aa469b140e2e848e2851291cf6
0492a9ffe0a869404b8e1187e787a4fdf7be6b18
16742 F20101119_AABKQG katar_s_Page_142.QC.jpg
1eb60c953052463796da43ed0c1206f0
a8d19448f5891795adc49a1a26fe94958cacf8ac
107283 F20101119_AABJNE katar_s_Page_118.jp2
bc0fc33752bdc6a2d8b27fec951f4902
fdc08204b7bfce5839ce5900a42364b9f0ff13f2
3909 F20101119_AABKQH katar_s_Page_144thm.jpg
48d508f84485e1bf94ec23d6c85a6ae7
09e1c81848b4a2c88ec7644102d62256f5cc108d
19910 F20101119_AABKPS katar_s_Page_125.QC.jpg
5e89f530887834997166321b2e2d5b02
d0718262c65c7ebf733cebd8e9aec6d9565ee2a2
98017 F20101119_AABJNF katar_s_Page_119.jp2
f8b4ab8d50a2bdfb341914a8e3e89df2
e14ab7463334104d1c1078a078a83c65ea41dfef
72237 F20101119_AABJMQ katar_s_Page_101.jp2
a381d93283ef65eadf75d20ca4804404
7739a552b446a56645580eeefec2b28cab2d2cf4
15255 F20101119_AABKQI katar_s_Page_144.QC.jpg
ae3301b7627f3e5957faabedfd1f0a27
e265ef3211345b744f108141c621739c6d27001b
16245 F20101119_AABKPT katar_s_Page_127.QC.jpg
eb44a2166da970fa8c717d986e77608a
ae2bdeb61f5bce238150ab33d8da4f232b4295c6
79731 F20101119_AABJNG katar_s_Page_120.jp2
13bd8b4d8b006cea2112ba02eea06fff
729d28c0d0e6edae405561cf55e07d9ca5b57d7f
F20101119_AABJMR katar_s_Page_102.jp2
03ae935006c9c7f8f2e1df12425c7c7a
775f54275545f49f535b0c44945b7819f3fd3952
4262 F20101119_AABKQJ katar_s_Page_146thm.jpg
6cc9ade163e2f3ce678243fff25cbc47
c9ab67166d3dfa1e19b4a64c5b04528b55fcce62
21879 F20101119_AABKPU katar_s_Page_129.QC.jpg
99ee23a1158643c9e50a50e309e57f34
fb5b9a78c1949b9d66de227ce8ca017a2a138d8b
71228 F20101119_AABJNH katar_s_Page_121.jp2
54f8a8a8f72b2991a55b71381c83c401
32adc69caa5941df3fac40bda0ff6a253b07017f
699982 F20101119_AABJMS katar_s_Page_103.jp2
af9d305a2f04fcdbf7e94c581636e83c
79c3573b4021c8e22970bc2f1b2c3610dda9245e
16798 F20101119_AABKQK katar_s_Page_146.QC.jpg
7075811b9e52c3970e1c0e99d92f43ae
127b2fac53a0209325e593b4a83591eca16a7fa6
7667 F20101119_AABKPV katar_s_Page_130.QC.jpg
34639eeaba1870570474108f25e9cd03
23a33e9c95f9f3b953906bec46c4addf214b6236
75509 F20101119_AABJNI katar_s_Page_122.jp2
ec09a86bc97ab7cc8d007cb6e385f50f
079a43e76462bb614d8859b843a971942618f5d8
119188 F20101119_AABJMT katar_s_Page_104.jp2
d240ecd4e1ed17f327921b24e0624a0e
edd1616f31677554a178b86b022fad4e0e7903a9
20328 F20101119_AABKQL katar_s_Page_147.QC.jpg
134cdc9e78efef08207866582257fb5b
50e6c1e72d92574a34df5caae0f71b5deff84563
4154 F20101119_AABKPW katar_s_Page_131thm.jpg
d0f6cb8fa044b30ad03ee956de027f85
c3d48f738ac725573b7f0c2aafad7bd86b0ab3d2
100963 F20101119_AABJNJ katar_s_Page_123.jp2
54a2747d99f0f25152705019faf279e7
7b607617657f2979adfa3d2df18ff313555001f3
F20101119_AABJMU katar_s_Page_105.jp2
ef5ba42d58fbfa94e78420cc2286a9e6
d9503ed8c39a7d5ed60d6712d5a7fbed6ea9886a
18037 F20101119_AABKQM katar_s_Page_148.QC.jpg
956680e1ec06e20ac20b6988a02cdc03
839f7a0a24b3e43bcae10ea971dedd163591dccb
15486 F20101119_AABKPX katar_s_Page_131.QC.jpg
d172398cf443b35ff4edf711849aa3f2
c627b2be891df61bbe9723d20a7d2d04e0c2b46e
84789 F20101119_AABJNK katar_s_Page_124.jp2
e8d548291aa0fd7328d19ac8d3748011
36434e10165e8d65b4fd131f645594f82e97c93b
1051981 F20101119_AABJMV katar_s_Page_106.jp2
fe920507cd6f135b1232b7226b0f9e02
945abb63af60242413f9edcb9c2fcc2d36ef241a
16401 F20101119_AABKQN katar_s_Page_149.QC.jpg
016c1e2b191ae68b48ff6155c720a9d0
371e7f1ddbbbb8a13283dcd133efc0a2745e65ee
16251 F20101119_AABKPY katar_s_Page_132.QC.jpg
27c3c5c9eea7a4d484a6a1d19eba158a
951353825646de2b5667b1acf7a8f84d2e5681a4
81655 F20101119_AABJNL katar_s_Page_125.jp2
423ad57df1f7b1aa0d403cf78e56196f
29af7824cd8c51a18059afbf1d2ddd864a00ebdb
25314 F20101119_AABJMW katar_s_Page_107.jp2
051d7f32634693bbc83d684390c4fd85
d8579641fced18e422b32c2d243e5915bce57976
5024 F20101119_AABKQO katar_s_Page_150thm.jpg
138eb04a9869a4a6ad4b628c7de4e4b0
60eda8b2fa45fb9ccacd3c55c40cecb935af3aac
5122 F20101119_AABKPZ katar_s_Page_133thm.jpg
ca5804f804a16eefc17ddf0fc755825d
f3e88f5b8c854bf3cf406f77cf3c069ff517c04b
94760 F20101119_AABJNM katar_s_Page_126.jp2
4fdb2bcc37ba83667a57f43266a20474
751e9c115345c4f79139e6b648f61d589a9576bf
940981 F20101119_AABJMX katar_s_Page_108.jp2
8440650d6631636b2532528c327d7d38
a3946e5ac35c369e09e375608b603ba4d6425e6c
957334 F20101119_AABJOA katar_s_Page_141.jp2
a9cab46e82e9ea739352a026545c6fec
2ebf077c95f2e3dc017ea2fdc1b4348e72a15040
17331 F20101119_AABKQP katar_s_Page_151.QC.jpg
4c5d6b152b9c0eba0304c795bbf88477
0507e7026509dbd1c84027e664a68426da068f56
579499 F20101119_AABJNN katar_s_Page_127.jp2
8928aa9103ac104367fdb4edb4ed5b0a
a8b4de7c65805e0e9e121104dd9c8dc364ad657c
103499 F20101119_AABJMY katar_s_Page_109.jp2
8fe79256e00b41df868b50ffdaab41b7
1102c6ecfb93d4250f0f9b407a16fb5ba9dc2156
891462 F20101119_AABJOB katar_s_Page_142.jp2
3b0e67b3fb57291d88893bfb02889fef
d62135fce6c2e6aa65671369c2b19e9b07f8c4b7
5246 F20101119_AABKQQ katar_s_Page_152thm.jpg
668459360b2f4fffa64b5aa239181c7d
cf1197f8bd6fcecab79d74e8e4d5d665843a924d
92572 F20101119_AABJNO katar_s_Page_128.jp2
cec2f11f589b560a1806513cc7c6a5fb
95510d9634986e7d95d84decca9c33d507be3356
105582 F20101119_AABJMZ katar_s_Page_110.jp2
722150b4a50f6feba1e3cec802ff4808
ef26de3c54927ca43aaaa7bb6d71fcf2c83762e9
825468 F20101119_AABJOC katar_s_Page_143.jp2
cb5f2a690b96c4c1deb2fe4751fa6ec6
72e577f9fb6809a25d9c0d9e765720d4ee62c505
21654 F20101119_AABKQR katar_s_Page_152.QC.jpg
0afa8612660773c672cc078c4adcd7c0
39e1e3026d9a773d9380e6b4ac01c70bfbddafe9
892523 F20101119_AABJNP katar_s_Page_129.jp2
55400add1dc529f12203e1d19cf8590e
74379ce611593b3d55c83ed71551611bb3a35471
771468 F20101119_AABJOD katar_s_Page_144.jp2
4ebe578c1a79a9bd73353770a72f5835
8438a016ef33466f197ac957cd3dce0428d10e25
6365 F20101119_AABKQS katar_s_Page_154thm.jpg
6f55b27bda94876378d6cd3ab2d02d7c
a33957e52619487e15f9750901299a653e51f1de
480474 F20101119_AABJNQ katar_s_Page_130.jp2
ed613e4b70777a9eaafaf2e310d05feb
d6f0fb61c2e4669ccfbc0496c90139d74fabeb39
700668 F20101119_AABJOE katar_s_Page_145.jp2
148945644bb77784086a4fba33c47543
f4868e1f98bfa940fa840d21ee956b945d5372b5
820506 F20101119_AABJOF katar_s_Page_146.jp2
7c1ab3b49d35db313b8279cb773addb0
eb640833798a930347eb5cd76796802421b3294f
25361 F20101119_AABKQT katar_s_Page_155.QC.jpg
ef268934d47eeeb56bd2120288c6d460
73e83f31839fc1771b294f40fdad48161be4b461
64226 F20101119_AABJNR katar_s_Page_131.jp2
99cae5f2cc5c31c2234d55ef25948b2f
9783100702aad8b6e714a560971f401bcb9b7ddd
1051980 F20101119_AABJOG katar_s_Page_147.jp2
89f2f77e15d2fc4df6b7080a418ee830
5449393a273d2136c5adaeec82c4114fbed91461
6641 F20101119_AABKQU katar_s_Page_159thm.jpg
78df29162bac238a8eb5996726a66a52
785dfbfdcf4b4066813cf077ac63a54ce053d1c9
62938 F20101119_AABJNS katar_s_Page_132.jp2
91afa18431e968f308db749be745ec1c
3d3dcd18d55d537f83829728e846f28e07214d3e
994479 F20101119_AABJOH katar_s_Page_148.jp2
2a8220e90b69038aad7a43973506fbcb
ad9225d3da34beb719b571224f18593adac1fed3
6313 F20101119_AABKQV katar_s_Page_160thm.jpg
738a02165f4daf28741b78a787312d6f
8ddb140f0907fc6a39bfa38a2ad0bac4910988c8
679905 F20101119_AABJNT katar_s_Page_134.jp2
5e86d577c5be79bda9115f922d063614
de8c000dcc0c85dd21d7fe82edcb4a816362a6b1
854041 F20101119_AABJOI katar_s_Page_149.jp2
7171a2396e4b026d58e926542065a31e
ce7328c94cbb568d01ad2a67cbbedfcf179686d6
4803 F20101119_AABKQW katar_s_Page_161.QC.jpg
85698c4e785b07d5d1616c4854716b57
a5ceb9e107ce856bd0cc46d5b544da4adbb79b33
76866 F20101119_AABJNU katar_s_Page_135.jp2
e86b11d9e4834b42401c30b688c5a284
d222299e8d4c2df573720f349e9053045fa9f76c
890357 F20101119_AABJOJ katar_s_Page_150.jp2
26db535721c5f3f7b7c0c79b9938bca0
aae547618663c701ab30b93d5483385a96fd5166
9039 F20101119_AABKQX katar_s_Page_162.QC.jpg
20d498fb0137c15bc100875ec4538297
f247cd7b85b96c03d202356095a931a0f2fa00ee
46218 F20101119_AABJNV katar_s_Page_136.jp2
57eaa527ecdd32578f32c3ccbb3722f8
7eefd4b190935ef26e8615ea80b704e00a81e865
72928 F20101119_AABJOK katar_s_Page_151.jp2
3d93b521ded955750d093ea5caaf46d4
46b2f3c133da175363a5b42dd3c1d587bef5c039
92019 F20101119_AABJNW katar_s_Page_137.jp2
8197149a95f28c45ebd449cd3e08b974
db65ee0846e2bc7f7b4df611d3ca51e702663e64
90075 F20101119_AABJOL katar_s_Page_152.jp2
1dd278b2870d89e255da2698a1a8a6ac
377a0bef45cc7d627e077fda8614c79f8db228eb
146075 F20101119_AABJNX katar_s_Page_138.jp2
00838c4a885af9f1c1d4363efb5b8136
8416c2d819881fd19d3d45601d5405b645fc88f9
F20101119_AABJPA katar_s_Page_006.tif
9d4ccc04d837400c7ee41da40d3149d8
a3149f6a29c98b25c2cb6f7187d0187299155075
122889 F20101119_AABJOM katar_s_Page_153.jp2
f8717929dce3e3886bbd1d13ab6228e5
7695c09aed3e9a6fa8b4e4c6eb2b9e73245a1cb2
121331 F20101119_AABJNY katar_s_Page_139.jp2
219a29cfe4eb5a41f67a04a6e44bebe3
fb09b5996a7086d38d91c45b6a6c4d3c0730a345
F20101119_AABJPB katar_s_Page_007.tif
0db35755fba92c2caefcb03cc43e111d
38b70c420587c26964b46401e7c0f991adaffa57
116105 F20101119_AABJON katar_s_Page_154.jp2
a90ac8f606f308fdd692b1de5c396c68
80d5a6dd1f41bf30743d07baad5ddc5429351934
F20101119_AABJNZ katar_s_Page_140.jp2
13c433953464a6dcdb67d4eb1176239a
2e48f676d8948f7146d244807f1c3d78ec0da37d
F20101119_AABJPC katar_s_Page_008.tif
2595e370a53d7ddd4d31724f9ce61ae2
7cad165909286112e6889cf4fa3fd80102231d25
108267 F20101119_AABJOO katar_s_Page_155.jp2
b035bb3e4e82ea7b31499f6ff4634378
ff3297834d0818a7068219a048e1a57f58090237
F20101119_AABJPD katar_s_Page_009.tif
a31c650df426f9dc5708f5561245531b
60e6ae47c4676e8f8d66ea3dc4e3e2b9e00a7ffe
80378 F20101119_AABJOP katar_s_Page_156.jp2
32a39dad376642e0143d74693b44fce3
097b68b97358f3295d0ddc91859159344f7b43d6
F20101119_AABJPE katar_s_Page_010.tif
ffd941c2f8c92eda2920f4d0aa5418d7
d2cf876bda43aba6639445b29ee9e9170c68859d
117134 F20101119_AABJOQ katar_s_Page_157.jp2
ae1c8a4d82d6103d273933e61b6c23a6
c8f6b8171f29f7109b3fd40d98ab4331e5987f9a
F20101119_AABJPF katar_s_Page_011.tif
2fed21e833f36f94c7c6b00fa849633f
15216f9469a9fd855c016bfe684cf9aa472c2df3
1051888 F20101119_AABJOR katar_s_Page_158.jp2
16077e66de17b21dab3e4d25a41f0e19
5ad46dd9be75f8ff73deecf3019911c0911654da
F20101119_AABJPG katar_s_Page_012.tif
a864061fe9e92cb71b562328f166ddef
001e424d2a60e6cb8dbc252999c75cfd42257ef1
F20101119_AABJPH katar_s_Page_013.tif
f8f150df8e2e2ff94688f00109d3d912
ab97b9a1e22ec2a06ac742cc3e57e59a0b17dd07
F20101119_AABJOS katar_s_Page_159.jp2
addf27f91e2ffb1534a7a97b2c25ceb4
21403b15c00de03b3f36d7a025d142a8e473e99b
F20101119_AABJPI katar_s_Page_014.tif
be3a513bfc0f5992f2ac2e41dccbc2ca
bb880af0c2aa2f5c1a2924b34acc3994c29eb4bb
132307 F20101119_AABJOT katar_s_Page_160.jp2
83be2cd10ed75aed51d56e658d6f466b
884d492980d5686c16834e6d1e40cf1dbee4e71e
F20101119_AABJPJ katar_s_Page_015.tif
0db5a9d132f9f9c531235384073ce4fd
5e8f63ebd7531786f991cbb37f5f5d1ffa67e312
21792 F20101119_AABJOU katar_s_Page_161.jp2
698dce458ba3d9915a177cc9e5cf78d5
5c93ba7321f209a783501f38be4e6e851ce2cce6
F20101119_AABJPK katar_s_Page_017.tif
885d40e8ee421c071d8400407e27dd8d
e0b6373ecfc39c8c2db5da6c0cd2917e8384c4d1
F20101119_AABJOV katar_s_Page_001.tif
5258db986e16dac9d3af4e5e9c3efbba
18ebf2500bdae000691e8a957e5ca95ef0f1c47f
F20101119_AABJPL katar_s_Page_018.tif
ff21eb436a35d220a6f0c46577cae896
5c475932088a41f08d4c9d3ddf0ecb82d986f5cd
F20101119_AABJOW katar_s_Page_002.tif
c1f0a7547da4204521632ebb206dd37c
ab09a8683fed645994f95b17a6beafef9606616e
F20101119_AABJQA katar_s_Page_035.tif
3448db4b45fa4c987a6d53e8776e2748
3c7dd9c257c0ddaa89ae0cd47b68de22c7cea46a
F20101119_AABJPM katar_s_Page_019.tif
b8e2490bd2fba5c103e4008cd76c7a96
33e1dddf4539b414dbf36ff78b6ef700be912119
F20101119_AABJOX katar_s_Page_003.tif
1c4aae9cba1756deb71847998a1f1ce1
bc97cfb44b4281d39384c220b3c1d3c1dccc0fb0
F20101119_AABJQB katar_s_Page_036.tif
302487b6818c82f50f92a4d5f8ab9d68
0e1d51d6ffd3a07779c4dd482dd01c472db57985
F20101119_AABJPN katar_s_Page_020.tif
ff7ea563ebe05deb15a39da4198e61e3
9265a7a2d92e9df96dda069ad09069e41648793e
F20101119_AABJOY katar_s_Page_004.tif
82051ff8e41f2131c46395ec906053e7
447bcf3f75e3d6681b7f08cbc254c5c7691db572
F20101119_AABJQC katar_s_Page_037.tif
5fefe51fbf17ca854b77eceaa83b780a
dea2bc0f3471b2571d74ea8795bf2c6a093dff37
F20101119_AABJPO katar_s_Page_021.tif
7cbbef7ed64336d56f857af322115322
cca5235c2b79d80967171b363395a0d9384d9784
F20101119_AABJOZ katar_s_Page_005.tif
86c14e469df9eb0615c657ff0d1f6303
eb3d4e64a9f97b3715d5c6f03e4849fdc3903ad6
F20101119_AABJQD katar_s_Page_038.tif
ad1fa78f6577277eb30f649706d89d4e
8e17ec3adb3aad538f4ec05bf15d7eb896657858
F20101119_AABJPP katar_s_Page_022.tif
a83aa4d74935a6718fe865982585b4f3
f982761ffae299a6acd4d18714a360b929857875
F20101119_AABJQE katar_s_Page_039.tif
6b59a593cbad333e0f1eec8e9a759f66
da1980ada33d84b9654b0d8266dd04e5be7393ca
F20101119_AABJPQ katar_s_Page_025.tif
1bb914f223b337f4057415506e254b64
a1a222cfbd16164f88b31563e9ecbb0a362e927e
F20101119_AABJQF katar_s_Page_040.tif
d3f2f9ccb16bb71a366a70e2eb1026ce
d5d0005a6b5dfb6890070d624a561dd70f8de9a3
F20101119_AABJPR katar_s_Page_026.tif
50d37f42c438e691cd8e7c59ea55ae04
14a30d327821a159403bbd20b8e5c986f73b57a9
8423998 F20101119_AABJQG katar_s_Page_041.tif
12569ad24ddb6c2f33a89054e85cf739
6b9689e9b9a077e0504ebc68293b7976f11d0ae9
F20101119_AABJPS katar_s_Page_027.tif
f7313687603b7ef0c34d716bc1918f5e
cf83df9530fae95ed8bcf76160737aaa7a7c9b62
F20101119_AABJQH katar_s_Page_042.tif
9c4fdb60116802c30f359d0f02958b76
e100bcb1497372725185f450d6d2b5a4d67b497d
F20101119_AABJQI katar_s_Page_043.tif
d71b250be4cbb4d493964c74e20dd0bf
ad89e2fbeee7557ef5b093c866b544ddcc30923e
F20101119_AABJPT katar_s_Page_028.tif
396c75c1e927b59d16aeb7f49ab33956
3132108b2695deff16d9d66dcb58a6581030b1c6
F20101119_AABJQJ katar_s_Page_044.tif
c9226697f0a58aeca189af121fcf9513
cd8acebf5263c5460fa405ec34e6c98002f66b75
F20101119_AABJPU katar_s_Page_029.tif
a7b96408309d83e49bb2861696f09784
fc40147dc6ace5ce635916de64afb9327d017c89
F20101119_AABJQK katar_s_Page_045.tif
9cfd7c495e1f39db94df63461fbff54e
17bc6d6ddfa9127cbd341af4b1b49bace92d2022
F20101119_AABJPV katar_s_Page_030.tif
c591f79264e5576ef5274fd08dddba70
83ee8d79e688706ed7115820417d5bb921f3ce0e
F20101119_AABJQL katar_s_Page_046.tif
9a28c8b5b2cbb4e1f2405d17bd65b115
31df510554cfc00ca13b52a3fd059aefe0ca1737
F20101119_AABJPW katar_s_Page_031.tif
5b5a4cd5e9b4cc155c283ca4fca69dc9
788d9c137ace0e6f97c74487c26918a83450d1e9
F20101119_AABJQM katar_s_Page_048.tif
ff017448e34cf10a2ed50983faec574e
36a8525d8d75b7411c349802bbf4a09b6fc77f25
F20101119_AABJPX katar_s_Page_032.tif
c2196ddc3d9786a08c82c214abb147b2
e1d449c04e2266073efd1495857ce147c44fcdcf
F20101119_AABJRA katar_s_Page_064.tif
09e230e3540ff79b8c4c7a0b6a6396c2
0a0415ba02451ab3c207074ee14f588c06a1d1e0
F20101119_AABJQN katar_s_Page_049.tif
9a5683b2efb97791ecfd22449c379469
eb3ce543d9620eaa17fccde5e9f4418cea3b3011
F20101119_AABJPY katar_s_Page_033.tif
7179dbf83850446f00298785cc0e375b
403c7afe33612a59853f87686c5d76aef446ac29
F20101119_AABJRB katar_s_Page_065.tif
47b4a02e63e404bcbe853573de29e4bc
b6f6e69ad323bd60bb1e055bfc709d6e67dead82
F20101119_AABJQO katar_s_Page_050.tif
371adfb55ed468b30f43e70b63704e07
35c513b07d95dc720ebb7602944d9d8938b486b6
F20101119_AABJPZ katar_s_Page_034.tif
9331a1ec6bfb70dd7f113900c6b3ad0f
12e1388ae4b5a660a287e519c16af1262cdf025c
F20101119_AABJRC katar_s_Page_066.tif
16788a67aed10f5137a730c99c0e2cac
476a12ffc5728a5fe0bc082f0d1021d912b75297
F20101119_AABJQP katar_s_Page_051.tif
0677c8c824cb0ef14f0426dce11100b3
5efa38859e3f6c4596d13301d5de7f1dc7aae538
F20101119_AABJRD katar_s_Page_067.tif
010fb3f14fdf7884ce60be8aaaf4c9f5
d6b5ff21195efe9e43bbe46160762ceaa36ffc39
F20101119_AABJQQ katar_s_Page_052.tif
f70124cd576b6dc76c898e538b3d8cd5
21f24a568f494a6e2beadfab2d23c3b968c720cb
F20101119_AABJRE katar_s_Page_068.tif
7b7903c3d7929bc7f3185b1793494ea5
900547e2476dac9fc7852de63171ae0181da1189
F20101119_AABJQR katar_s_Page_053.tif
720cf4023ca75cf8a65bbfbc01c79c2a
fb43ae3716c0789cd3c10d8fc6266d4f45829b29
F20101119_AABJRF katar_s_Page_069.tif
1321d9729f25b4fccce9614fc0679677
5a9ad485e07c15823aa79b5d445a3dd3e4ecdcc9
F20101119_AABJQS katar_s_Page_054.tif
2a26b19b602fee45b14ca749820d2ee6
e996c7cb31d67195376db06be4a57dd188cf405f
F20101119_AABJRG katar_s_Page_070.tif
d14fe48f32fe63f90292214c094559bf
0ed42645398424199855af0fb6560e0e02233b3b
F20101119_AABJQT katar_s_Page_056.tif
5b12d8cd4795908c1059715cd87814df
e9b6719d9d53478b4b8bba97a0616a63ec80e3eb
F20101119_AABJRH katar_s_Page_071.tif
8b68ab7af36cee21d0fd43d0489507f1
f1fc9bd7ffe2ff8cff74ad061f965180a2ff57ed
F20101119_AABJRI katar_s_Page_072.tif
af733f0d57fccb08489c4a5f8a83dc2b
17f09a1e4dd3c86abe5468eb1b7b3ece71eefe1b
F20101119_AABJQU katar_s_Page_057.tif
26ab7b47d86a8f7c7f734e4e8bed5f71
ac7ac11837e13846a3405e54550447430307d9c4
F20101119_AABJRJ katar_s_Page_073.tif
0fdf257f0acc25532092e031bc4f42a3
8b9915ce1c560d498998fbd6f81367a67c3e90a6
F20101119_AABJQV katar_s_Page_058.tif
4cd9db8063ff03e5e16156b045d485a9
e277b31724551d52b1dd2c25f31ad05d10c33812
F20101119_AABJRK katar_s_Page_074.tif
061dcc7a68b5fd9ca07f4c72bd9630cd
ce243960a97b2994be6d5e850087aaaaefe1499e
F20101119_AABJQW katar_s_Page_060.tif
27437e04378332c6b2056ea6499a5501
6cc6f4bf7ad7ae1cc27754b233b349cd3d5ae9a3
F20101119_AABJRL katar_s_Page_075.tif
bd46ec8628cc3d0deaa857a797652f24
1248d00a2dcbe87687361f5c3397bb4bdd576a70
F20101119_AABJQX katar_s_Page_061.tif
80e8a540c6ed2c125a88a79257e19e05
fa9adf5b7eba16beee30a7fdd9411ec06c8b2f77
F20101119_AABJSA katar_s_Page_093.tif
70bb69ffe5f447e6e88ac2f3d6fc5441
250ceefcab2ce2f666558213fbd58936511c4f61
F20101119_AABJRM katar_s_Page_076.tif
6a9af8430dc21b4f0bc36796c12b25b7
c8a506cba6baffe61efdbacdebdf85c495190786
F20101119_AABJSB katar_s_Page_095.tif
26ef366a33d87a708f8a4e69430cd4f2
dd8c85ed2652e52652e0e5e2fd0e6b5f73941113
F20101119_AABJRN katar_s_Page_077.tif
dc48ed428df23ca6001e097039ddaffd
f818a49f35acca8a11adf8bf911749eef5c38ba3
F20101119_AABJQY katar_s_Page_062.tif
de02ae5bf1158a6b2839802410f96282
02982c075497003d94637c5c9d3df5fa414db7c0
F20101119_AABJSC katar_s_Page_096.tif
aac99320312a3bfec0827b8f25be527e
71423473047065a67c6bc0cb320df89ec6187724
F20101119_AABJRO katar_s_Page_079.tif
8a7021a368e4573ce4bbfc29474175ba
be4b39167b5e50408508a3d00c86d92c2a02d3f7
F20101119_AABJQZ katar_s_Page_063.tif
ae2747d9019fec80ab7039f2878d69dc
2b0b8b0f2964e60d16eb420b2b314b07d74f741d
F20101119_AABJSD katar_s_Page_097.tif
338ffc6bd9d5d4ac67808e4458fb9f4d
744318713a3f6b3d577d209cf27325708b0dd6a2
F20101119_AABJRP katar_s_Page_080.tif
1b3f2762c79b024ccb5a76646c0e4ff2
589882e23c50e75bafe2384b02185f9f1fb0d772
F20101119_AABJSE katar_s_Page_098.tif
0f2d75ee062324d6ff84dc62a9a543ba
b28f1808c3d0177e5b166600ef9c49b9253c0a43
F20101119_AABJRQ katar_s_Page_081.tif
e187b8ea79887cf9d753b3f0e0b07a78
6f9935e5ada655574951d7f4663f4ec0c5919cc4
F20101119_AABJSF katar_s_Page_099.tif
66a141f64a5515e279209c57e734bb3f
e717d71352fb7bfaf3de10bf1e9469f4c888fdb8
F20101119_AABJRR katar_s_Page_082.tif
e0729abd50cdebf38ce6d88040cb9b40
470bdcec2e1b016763fcae21fea96b75164dc00b
F20101119_AABJSG katar_s_Page_101.tif
f5325551127163e691f16fa011b8a153
a8a77b200f596bb30b68e9ff897e753e2b6b22e6
F20101119_AABJRS katar_s_Page_083.tif
c0ffa561c2d129375721eee1d89cfa81
02fbbbd77cf6951fffe7885c802eaeda91a848ce
F20101119_AABJSH katar_s_Page_102.tif
d60bb26f4662fe5b5750d3d6278f36bb
f8990337a2600bc5edfa7a1447ae9004f1bb8d6d
F20101119_AABJRT katar_s_Page_084.tif
017c6d5cd93063dd81d55d461cec2111
8d0320c6f6ee5c48efefa843606d2ab5751560d4
F20101119_AABJSI katar_s_Page_103.tif
328be6b0ebaf378ae8a765b418e60bb5
56ae79585dda920e5193f7b4aa4018437237989d
F20101119_AABJRU katar_s_Page_085.tif
cea0a0296a1053c9fff39df288cce87b
fba3310ffc12f6633913389a8c70feafb087cd0e
F20101119_AABJSJ katar_s_Page_104.tif
b6360f53004afebc1d1d57c40e0fb3be
4aba377aaf53c6c5d6c01b107983884653e20281
F20101119_AABJSK katar_s_Page_105.tif
5d41c622c084840956b4cb8707bd6b89
46e7a78a9151f8bb9a0bbd7b2f6665883d12ee7b
F20101119_AABJRV katar_s_Page_087.tif
365f5418a17327e708bd5fd99588e3d1
eafe15fa7471ca3e500e477e527c0bde8b6e4d1e
F20101119_AABJSL katar_s_Page_107.tif
d125babb3c26f21e4cbce18596900fbe
3661e0259c3eb502a65ed19a5db0b151d9d90f28
F20101119_AABJRW katar_s_Page_089.tif
23772c600196e123a821b7fb70764b6a
e202f5bd36819f99941d79676b837cebe2027958
F20101119_AABJSM katar_s_Page_108.tif
aef0ce82138c4ed31c131ea713d8bcae
c76d1b448ffce7bed9c4e1a28da504f94108533b
F20101119_AABJRX katar_s_Page_090.tif
60789f12a2a37ee9c4811379b70f4a3e
fce53fd7d8da4991857e42422a08f89858d8ecf4
F20101119_AABJTA katar_s_Page_123.tif
3fb498b07374bbe5b9546db9ef5b75af
442b1759115e92a8b7c283e2e5226e43de9a46ef
F20101119_AABJSN katar_s_Page_109.tif
786815630bf93f4e2190e8484bc5dd4d
56cb713b42cd082db6386f4cfaf8d309cb4faf11
F20101119_AABJRY katar_s_Page_091.tif
52f7663d885dc64e93e499e37e4c3890
ecd8242727722bda6dc3a3d35be41c93998550a8
F20101119_AABJTB katar_s_Page_124.tif
b125da6a2030339647b4b603f18abbb6
31992cef2930e704cc9ddc9c8b73dd47f044210e
F20101119_AABJSO katar_s_Page_110.tif
9bf1374d150c3c287802265053cec21c
7a63f90a1ca548df760da9f7d47ae538b4a03214
F20101119_AABJRZ katar_s_Page_092.tif
9c8344c7234e4fe3072bf659a52852f4
9dfde65b17d4c29988e717756907f12bfd6372b3
F20101119_AABJTC katar_s_Page_125.tif
b13733e55da8361cda5d477047c13992
42515e735645f37f08ce80f75137bf7c04f6c325
F20101119_AABJSP katar_s_Page_111.tif
f617fbb8c478619526ae2b0cf6d49151
442b56890de3dd58bdebcb342cfef42b77f84893
F20101119_AABJTD katar_s_Page_126.tif
9f30d12949b83107296cf98173415a01
896225fcce36ef68008c262ccbfc5f1de3bece9f
F20101119_AABJSQ katar_s_Page_112.tif
1b993f51b65319551a66716109ce7373
0230730fef646556751a2b60c261ee628e17daea
F20101119_AABJTE katar_s_Page_127.tif
a3cc6eb812a13c57680b6f32cfb599d8
df266c8ec7cf6f29d3a219415ce5724906cb1af8
F20101119_AABJSR katar_s_Page_113.tif
470b583a3e6f690c252ac8808b2123bb
5043e594cc7ab11f02046a348da479e96fdeb7d7
F20101119_AABJTF katar_s_Page_128.tif
65f91dec59fc8206ebf4e0568c7ff938
530a671a1bf0c364b46939c69ac70620eedd8108
F20101119_AABJSS katar_s_Page_114.tif
09b4e45948847f12db2ac2bbd887aa0d
653ff0c741f7f31e6906899e8b29a8630b857ea5
F20101119_AABJTG katar_s_Page_129.tif
ba55f32e2d55876c24fa2c7b50441a15
a91489eb53a4745789b8875cf7f753b684176b96
F20101119_AABJST katar_s_Page_115.tif
a2e28674fdb6e0e4d09116f1c6a51046
905c46769155f14043ddd3ba416e36867e69a957
25265604 F20101119_AABJTH katar_s_Page_130.tif
671e297ca59d5ab895f192e916ceb688
004120facc1a513249c41edb62208c7e34354ead
F20101119_AABJSU katar_s_Page_116.tif
bafe44c6b192664583ca96b2cb74bb25
adf9fb16205e60791e5d939bea84d310f3db0047
F20101119_AABJTI katar_s_Page_131.tif
3268710034a1b4c0a839a393454a8a78
aeef0e8f88b139ac723e5ed193ff171e37652be6
F20101119_AABJSV katar_s_Page_117.tif
d08ac2164e1a75228b9d7e58c6b4b822
ae8d3395b2cb0950fbbc0b15a710923d6a61c405
F20101119_AABJTJ katar_s_Page_132.tif
4095832a4c80e9e6c2eec90d24fd0b87
66ce6dae03c2250e0e8fdbd4d494f0ad2fc23139
F20101119_AABJTK katar_s_Page_133.tif
ed99ac5ec8d81590c096029f6e571ede
119287db865a18b37dfff6790f7be30d7561c5ff
F20101119_AABJSW katar_s_Page_118.tif
80d6bad86dde8a3022f58089af435284
69eb8018dd1fa7e72936e517019f60c135a3916a
F20101119_AABJTL katar_s_Page_134.tif
3482edbf768c86201843691b58897c4f
78699c624663156ebac79496b41261b38470aca3
F20101119_AABJSX katar_s_Page_119.tif
04b1e129549c973f2625ee4ea8ab1a1d
ebb40b7acc4bb2d899b5991e5de731cd52e175f7
F20101119_AABJUA katar_s_Page_151.tif
4b0c43ac8af71aa2616159440ae75e69
f490d4ce2b07b898d6d3ed1f6519f2ef2b7dc2db
F20101119_AABJTM katar_s_Page_136.tif
46ddd9a271b3f1d980fe4d26c0179884
d43983fb47f1d916e05d0f3a06af06631cff705b
F20101119_AABJSY katar_s_Page_121.tif
62c74e71bb1ba5433d32130be6790b01
dcd005e11f5240e2e12bc116360263e532669632
F20101119_AABJUB katar_s_Page_152.tif
9cfebef0c08c84038648787e218c8c9e
ed0674130fefb72eeecb7348e8184bb20b72ba97
F20101119_AABJTN katar_s_Page_137.tif
8eab37fbf591bc81b0f8dbec22dccde3
6314d6bbc8cea39b39e22f193497e22b5cdbf3c9
F20101119_AABJSZ katar_s_Page_122.tif
639ab2b008046c70fb3de43d4484e08d
ceac66ac99c4b9a836abe37b42d5850765a9affd
F20101119_AABJUC katar_s_Page_153.tif
a39c63d0921fd3de61477fe94c0e0d24
cbd60baea454369975da243b8f82ba2d7dfef4bc
F20101119_AABJTO katar_s_Page_138.tif
0111c7884b050ed383e0782f5998195e
082f0ed6ab15864cecac33c70cb56f3fed5d46f8
F20101119_AABJUD katar_s_Page_154.tif
6b4ff3e0e7777054002bb0a512f173ea
18cfcd71bc750ec187b6129b38f33a72a4cf7655
F20101119_AABJTP katar_s_Page_139.tif
444a5c91dc1d3559e794f37aaf909d54
7f46c4a1f53caf8de55f5e7726e8c52f490bd7b1
F20101119_AABJUE katar_s_Page_155.tif
43cf841ce0907738d4637bcbba41afa6
b1b96f40907301aac0371d0e49062cbacab0d183
F20101119_AABJTQ katar_s_Page_140.tif
6f22e294516958cdf278ef53817a4fb2
05a4e1454b3f039d024dabd3fb62938c1cc73acd
F20101119_AABJUF katar_s_Page_156.tif
9eaacd75d698e39b253c405179108879
a908ef5c33c6ef18a72cf4230528ee5887fa228e
F20101119_AABJTR katar_s_Page_141.tif
139e6fd1282c445a9d6cf9d871969ac1
438543088ada7460dc1bdd10360e380cd4a0baca
63699 F20101119_AABKAA katar_s_Page_158.pro
c380b6972015d1c83515ebe97a7f2173
9b0e422ea810ab206cdd8286e3a32ba2f96bd2e5
F20101119_AABJUG katar_s_Page_157.tif
37359622bd1d0498df2921adb1eeda14
598ab72289a3f0c12569bff1d6fea5eacb370bcd
F20101119_AABJTS katar_s_Page_142.tif
dbf00450212e62a8b72fc012c4c59a13
6f429f089c7f9e1e485eb58f44485c58962b5dc6
56745 F20101119_AABKAB katar_s_Page_159.pro
a4a8be8ba5f21e7e8bf72e1c6b8db9df
457ccd8b236f710d904ec15107561d0a8923f03e
F20101119_AABJUH katar_s_Page_158.tif
139633acc5be8e7c5be16c778c4bd9cb
f2d7ef2bf64987734fcea33b362ec306876210b4
F20101119_AABJTT katar_s_Page_143.tif
c8d7cfcc7eabe976ce817edddf9fd551
2b41e1bd3a8c18775f04bcac4cac47d22eee530e
62202 F20101119_AABKAC katar_s_Page_160.pro
b270339b593ae316a4a79c927e25564a
8bb4e8e6fe59fc414efbc6e9d9bac18c74c7b91a
F20101119_AABJUI katar_s_Page_159.tif
ee1775ca1b4f10cc912a191cc48d43d0
9736fbb3f0acdb8df87a36ac163085fc47b7411a
F20101119_AABJTU katar_s_Page_144.tif
0bb3c6d9cd49012d8905206ada452a6c
0208d714f4972f3cae76cb4f7e320d4be7bb16a5
F20101119_AABJUJ katar_s_Page_160.tif
d445e203cc4b42206f05a38e28d09527
426d31261b3fe93f80bdd7bbc4d7408f577b0b62
F20101119_AABJTV katar_s_Page_145.tif
6aa0100df78e31021118e9211d7e5d10
97a9d698abb24b96051f4f5d2d5cc86709b7c306
8372 F20101119_AABKAD katar_s_Page_161.pro
a827124b0446a4001f279270cf1007c3
9b965288c990d4b6116c614b3f76a5e2579d2b8f
F20101119_AABJUK katar_s_Page_161.tif
dd9b1dd47f3fc78579ba130daa6f4ec0
846ed977c959c9697ac2cbda30c51ee84088f374
F20101119_AABJTW katar_s_Page_146.tif
d87118709be0787bf487705e178295e5
10b56dbf7a14ded92bddcccd60ea26e1b853993c
14898 F20101119_AABKAE katar_s_Page_162.pro
7ba958111c69b980476e358ffbe4ad00
4437e95d037691b6213089b6a28772f94b84d94a
F20101119_AABJUL katar_s_Page_162.tif
f2af33f44822a902051255bf27ccefb8
212659d6b9963d8eaef3002b9ccdc984924b0d0d
469 F20101119_AABKAF katar_s_Page_001.txt
8deb46b3b9455bdf834daffb19c50740
f1627d42ffac6dc75db0c3dd20102f4f3738b8b2
49323 F20101119_AABJVA katar_s_Page_017.pro
be30bb4b584676de6e702d5fa3955a0b
50479fbb833840dd8768e3f2555ad66e5251502e
8224 F20101119_AABJUM katar_s_Page_001.pro
80dbd777a5b763835f0d37c76d55098e
00466b1c997195ea90a7e5008c71a147094606cc
F20101119_AABJTX katar_s_Page_147.tif
f09251678f9b1fea5e1c9dee6f40f775
255e2f41b2a945f729ddd113edd28895f6290fbe
110 F20101119_AABKAG katar_s_Page_002.txt
2cb4a4118c8ca58a61be71653d2ef04f
650829015c938c589389d3e46ab23fa28327f146
49658 F20101119_AABJVB katar_s_Page_018.pro
16a93d12a69f64bfe8e2d81c24f3b07b
da96286980c40d272a4f879f903c8ca86baa7123
1147 F20101119_AABJUN katar_s_Page_002.pro
d9460280684fd606e5a71139bc3cf464
03f620e0eedcaca713f342c82d0141c59f06f919
F20101119_AABJTY katar_s_Page_148.tif
95bdaca509734a85d678690bd8066800
888def617ddf8ad84595ccf9253ce640b7ef6a83
152 F20101119_AABKAH katar_s_Page_003.txt
68c8758d7d22950adf06b86399839a7a
e0d855428a5270ea6da49dd7c6b90731d3e9efeb
49653 F20101119_AABJVC katar_s_Page_019.pro
d346f367d63f321348dc214698835bbe
46e90d39c8f47aacf3a73ea0ab5fe6bd535c5518
2354 F20101119_AABJUO katar_s_Page_003.pro
e01c6cbbe6fc27428aa95bdd54ed161f
8aa7f4542adf3483e29c05d96ecc6321b2be13e5
F20101119_AABJTZ katar_s_Page_150.tif
4adae9acf9f1a6be8b80d273e581fa4a
cb643120177db9464cccca5851fd6c675bede387
913 F20101119_AABKAI katar_s_Page_004.txt
ed1f97707f02ad3f20385d823ef05fbc
405327c078c7f6d8a15295982576a00307f98cc9
F20101119_AABJVD katar_s_Page_020.pro
82e69a3d4debe118df0f509138cbbad2
da5fec533cd7bc42010fdba5fddf56632eabfe09
21459 F20101119_AABJUP katar_s_Page_004.pro
f402673206843a5bef4d90bea947af64
68ccd3399595d7b2f3f5684db159d5efe8fe612c
3192 F20101119_AABKAJ katar_s_Page_005.txt
93514e0ebe2ede0c7966f13341c34cbb
d3674759f9959f767943876a90b9c67e10139725
45502 F20101119_AABJVE katar_s_Page_021.pro
e3fab077c5bdeee0610b5f1b732ec962
b8aabff324c870e83a8a5e203519c14c7da2b4c9
95319 F20101119_AABJUQ katar_s_Page_006.pro
aa4ba21f69981aa12bbb6e7e5565ef3a
d81ee677ac44311049fbac517aa5fc24cb7c864e
4005 F20101119_AABKAK katar_s_Page_006.txt
dca4c25a3e762793d12b7cd4e1a48e7e
8c2a543205eebe098bbc789568afe316b430fb64
52295 F20101119_AABJVF katar_s_Page_022.pro
d6f6907a4035b1ff741596f8819f8723
0fd1a1393ab6a0fd2aa393a8199d134b17a82602
13282 F20101119_AABJUR katar_s_Page_008.pro
de9e23b57b8c700ec503b547c67e580b
b0c69718caab037328baaecbd8d42e6c82d0f69f
1855 F20101119_AABKBA katar_s_Page_023.txt
62a3f6b3bcb039b4d31afc7faf915d5a
e045806735a66808beba2af4e72955f7d3239ac1
2731 F20101119_AABKAL katar_s_Page_007.txt
f504c262252ca265a88ddca8fca54fbf
ef03bced553064c10c6b86f97aa5e349a1f27b5f
45843 F20101119_AABJVG katar_s_Page_023.pro
f1502dd71bd35ff320662d1e5b43302b
8133ad4c472f0ba5f558a0e45302dda315ef3fa5
62038 F20101119_AABJUS katar_s_Page_009.pro
4f09ef41b652eebca94c20f7a10b8661
646f161011deab8b36542d54b8a8fb3041bd9d80
2190 F20101119_AABKBB katar_s_Page_024.txt
585422c447dade7f4fa946f729400998
f4f3a450e66b2b65d09ca6945ff06a22f892bf92
604 F20101119_AABKAM katar_s_Page_008.txt
5107dbee71a0c4a1fda1c7202173c64a
4e7889d03a6e8cb706b2ff4b779f53e2070833b6
46109 F20101119_AABJVH katar_s_Page_024.pro
b6192e49bb7758e8216abde8c2b6b6b2
277ef9ae4a6b53d54c5d3cc74426c0bc94aade68
70872 F20101119_AABJUT katar_s_Page_010.pro
660bd75f807bdd3902df72d0fcae4812
d9b1b7da202e515751eb01f102a4be34b896b82e
1839 F20101119_AABKBC katar_s_Page_025.txt
8b44e147ba2b8669f3a7674720ea4725
b817f4500ac02432c567bea1a2471d0d54bb861c
2477 F20101119_AABKAN katar_s_Page_009.txt
9f1c85189153217660b5d092938066e4
8f641d05395fdcc2dca74d8e023a52be409905cf
48091 F20101119_AABJVI katar_s_Page_026.pro
77494619bcbbaf30994b90f902222fa2
1220fb52039ccf6c92c14c0f208de647316766d5
73382 F20101119_AABJUU katar_s_Page_011.pro
0f265b01533d7d117f605c8ab4c3b12c
11168ac1ca2bec929700e14f4483e6e07c37c215
1920 F20101119_AABKBD katar_s_Page_026.txt
d3dffb1ac1c202e3052eda89eb5b7fb0
88ffced765f8eff73c318e370a3a0c1e7268ddae
2811 F20101119_AABKAO katar_s_Page_010.txt
7b7017c234760103a9ec42389917058f
aa29adbda924d6e12041f932fc9633fe194ccdc1
46084 F20101119_AABJVJ katar_s_Page_027.pro
d488b1dcd885643df46e42915d9c2598
1e9cf10716128dc64297e31960e11bdebb132070
46299 F20101119_AABJUV katar_s_Page_012.pro
e213d391654eebc8f43aa7d1f6fbe62b
67a3f7ee39d5ba318bd9e1ba3a6ecd7aa73f9276
2915 F20101119_AABKAP katar_s_Page_011.txt
e22841451459dfcbea5c0e338b567ce2
cf3dbac490e4a40297533defc73321407691dc4b
30036 F20101119_AABJVK katar_s_Page_029.pro
46a1469f45ff6b5ade60c8841742909b
b218ebf8ccd482f03c81e19baad8d6db79adf9c0
36782 F20101119_AABJUW katar_s_Page_013.pro
fc720958922b2b84d8ed78fc737bab49
7f67c5ecb2e9a3e8b235dcf01b585864b49ac224
1820 F20101119_AABKBE katar_s_Page_027.txt
00b294e523b5e50f96820027ec9b4635
e12fe9a2622c11ac68f0f287f7548b84682be958
1873 F20101119_AABKAQ katar_s_Page_012.txt
093143d9aa103562da47e1999510eff7
92db02f06c44558d209a42cfe3373afd7da4ef16
12622 F20101119_AABJVL katar_s_Page_030.pro
1b7d62fb1f569f95cee45b0aa7891c23
355abd4168d7c2ae01c8fbae67e94dc651b7da9d
43156 F20101119_AABJUX katar_s_Page_014.pro
ee615f1dca8ae9ee78ad0f22b4f92092
103e7e96b65ca2b7c6d0b43fa154b9e9ca17ea63
1456 F20101119_AABKBF katar_s_Page_028.txt
5fbebf63ccd2e4f0a933beacba3e566f
8a78ecab4c1e6a7635f3eec9345a24460f685d06
1723 F20101119_AABKAR katar_s_Page_014.txt
8e0c979b51608b5713451fee4a7cd3f4
5c4508b0216ef4a6d430062c75263f8e5d31d2ae
7299 F20101119_AABJVM katar_s_Page_031.pro
d5f32a4a087ea2dc97f84a58cf7721d2
3b1fe2f045bcc7617aeea542487c7b3471182f1c
1397 F20101119_AABKBG katar_s_Page_029.txt
7c8fe640e985b32d2bd1856d1a826f2a
a6fd3bf7608a3eea87c73a3f83ae306831e5007d
52249 F20101119_AABJWA katar_s_Page_045.pro
167f5da15f9fdd767de2d87febe61d72
e456367abe96f6f9723504fd66f035b5fdc17ab6
1851 F20101119_AABKAS katar_s_Page_015.txt
3704b66e966e1ef754806a4730c255a1
406d003ec3fbdb4fa15e7e7e5cb3f7eff9f7fa9b
35448 F20101119_AABJVN katar_s_Page_032.pro
662670252c3278b220737664fde29d25
50c324b57c10f84f12ff9ffdeb27d3973a196343
45065 F20101119_AABJUY katar_s_Page_015.pro
19ed3acc1b9742da607df11a4bd84e8c
2b3b81c4c5c3029a9c2621d5fdcb8bd37c606c89
668 F20101119_AABKBH katar_s_Page_030.txt
2d081ed4f679dcdec6501cab2ee8442e
d7162ad19c82847913b5b3836dd6134558a95bf3
52717 F20101119_AABJWB katar_s_Page_046.pro
f3d9e48dd01e5f18327987d2babd7a50
7b8941c76ab79c7a6f50a75ad3c59bdcf311735b
1948 F20101119_AABKAT katar_s_Page_016.txt
0e79eaf07fae319a40bc5e2d0fd572a8
5f4cd7a7847dc0f28d7ed272a7c2f8f6bf09fc60
44058 F20101119_AABJVO katar_s_Page_033.pro
af9b173c3acaeac8b1061796f4b9de49
ab983eac92a2a77cbfdadd2739683fb56ef150ea
49419 F20101119_AABJUZ katar_s_Page_016.pro
92ae5b6b36014b9f513412e0d2f6af3c
2adf1f6b83914357ddfb5515aa0ac5bda8c6d20b
510 F20101119_AABKBI katar_s_Page_031.txt
8ebe93f24dd6cc384b0bb45fc938324d
fa881e80473821b01ce13b7f38ed1c898e1da235
44971 F20101119_AABJWC katar_s_Page_047.pro
df9b129dbc512af4196a71a22f4e107e
7ae48a3b1ca90978be377882d4d3fa9c578997d2
1949 F20101119_AABKAU katar_s_Page_017.txt
547922a9d2b284864928a1c914dd5e21
6a02150a1d6c6d46c1c639ff1beb13f662264aff
44916 F20101119_AABJVP katar_s_Page_034.pro
8c5ed4d8c44774b9ddde9bc4864c791f
cbf78fad7aa5781042298a920bf1384ac23100f2
1424 F20101119_AABKBJ katar_s_Page_032.txt
3e31f2f28d8aa373739bed4c09d27929
92207730bad6b9fd4c17861d53c5be672e00c09e
47005 F20101119_AABJWD katar_s_Page_048.pro
21414c2d2759220344840333a009002e
41d82d565eb86e0fd0ce1cda0abf061ead10be18
1955 F20101119_AABKAV katar_s_Page_018.txt
613f632a4412f1e506553fe8e4cf995f
1154b5e2705b095ab721387aae2cdd91cb57fdf6
37543 F20101119_AABJVQ katar_s_Page_035.pro
e3d4f784cb1ada8d4ce1e6cc771c1706
a037537156ba7c4b3a02abb996345f823c04925f
1814 F20101119_AABKBK katar_s_Page_033.txt
1c762ef5a60e0be4dee1e8b65ea92fd8
70ab740dac4b9112bf99e53bfc5a8ba6c87f1fed
37465 F20101119_AABJWE katar_s_Page_049.pro
e1368988e3b08dd0a5b44f3dce6d0184
e4bbc8b3f8d7af0f91c5f52b21ea174310d58731
F20101119_AABKAW katar_s_Page_019.txt
e784f05595d197e3f62a9e00f90107db
d6b06215c9350cd97addcd15155826323ed43206
40763 F20101119_AABJVR katar_s_Page_036.pro
a28ea05101b46252231e72759baca4a7
f6509a0b36883e962fffb89d084f00e13ce2496e
1983 F20101119_AABKCA katar_s_Page_053.txt
e4575ea1160af0f0104bdd40565f48e6
9efb663f745ea3e4f20410748a95ba10ab839cba
1802 F20101119_AABKBL katar_s_Page_034.txt
7f13783fc3efeb2fc5ce0e961e3f307d
ca5de64f51d75fae6353ab746f3b4474fdc27dfc
26236 F20101119_AABJWF katar_s_Page_050.pro
05dff915fba99fa01a782be9014a831d
3e5763c233c72b2279e399636d93654384e2c734
580 F20101119_AABKAX katar_s_Page_020.txt
bcc27aee8cbdb2c1fcf72e00d65295a1
5354e4ec3873fa497c45b0b5db7e8ce50589f6de
48201 F20101119_AABJVS katar_s_Page_037.pro
8211980f0b83f1bba98d3891c12c238d
5206f5666837a07e8da3ddff6037b6ad4b753026
1576 F20101119_AABKCB katar_s_Page_054.txt
f903bff46e4da45ae8bb765a41073736
14459d1fc18c1a269f253eafb281f8e214578f46
1506 F20101119_AABKBM katar_s_Page_035.txt
dd850d170778b11d81c49d02100e0d3d
1e6195b2fd40406d4bdbf3f8d3ae63e798b58732
25025 F20101119_AABJWG katar_s_Page_051.pro
337ab36bf67486f0d2c7365fcad7570a
c27eb2481519733f85b7cd8454fbc665cdb08b60
1884 F20101119_AABKAY katar_s_Page_021.txt
307e763adcd45aee2ee7eb1444e22ed2
ca8538177ee75d563f927b8a037159f9fd0c000e
46200 F20101119_AABJVT katar_s_Page_038.pro
e9a140c85db07ea2e0a4ca36b0f9dcf2
63f4342d57ee187f012daee750942d733a6491f5
2411 F20101119_AABKCC katar_s_Page_055.txt
4d9c8d1e480d99a5d44e24de29c25907
2e5033e0de3654402b6d07e8665d7528e1557b04
1829 F20101119_AABKBN katar_s_Page_038.txt
fa98ae90a554315f9e8408376036ac52
3e35c5d7f20400c05717350c4a21bf340ae4e82d
50476 F20101119_AABJWH katar_s_Page_052.pro
0183cc64f56014e348ad9687fdf984a9
298b75a22f02fb01a7ad09acdd378f01d6617397
2077 F20101119_AABKAZ katar_s_Page_022.txt
0bb43c810dad3a6330bd64871b634f6a
36258264ffe80b08f56a091f26779d7b7d8579fb
47427 F20101119_AABJVU katar_s_Page_039.pro
c68c1cbda725c2212669b3b4b16468bc
79ac2af5c040758f2da4a90ac87227f8bf8dbea8
1963 F20101119_AABKCD katar_s_Page_056.txt
da9564bc9d3e76557bb222bdbf26b64f
580bb9fa7ca90ef7fa859e32a8704ab75ce1f023
1883 F20101119_AABKBO katar_s_Page_039.txt
b8a89c23ba8a61417109f84e731b7db8
8d229e0e2120734666137d49ee019e880b56cfec
50400 F20101119_AABJWI katar_s_Page_053.pro
55c6d0594fde7c8c1930de16af580ad5
5f5f2a9b01f5a8a34b254cfefb85c57c0d14ca05
51526 F20101119_AABJVV katar_s_Page_040.pro
46d1685d8844d75b062a255bb3392be6
539ab3e146f3e4691793e9ef56cb4b363944ffdd
990 F20101119_AABKCE katar_s_Page_057.txt
bc30c5e86357481552f1210a43dd4c21
8243bf26c05009f38d977f07f379d117b1dbde12
2023 F20101119_AABKBP katar_s_Page_040.txt
351888ebd12b774b9a74e4e7bf37e73f
e1601d4fdbf8f3b5943e53c26c084cc3575b168d
38700 F20101119_AABJWJ katar_s_Page_054.pro
53a66a0fcfecfec5ac0fc5bf86235f79
4aa400e85298778278be763e8ea8ed6cb567ba83
2112 F20101119_AABKBQ katar_s_Page_043.txt
0078bce5dcbcfce19f9ea6282d93e383
4f2e47c1a4cd0514fb1c91e1be4cf7708f3ee254
48093 F20101119_AABJWK katar_s_Page_055.pro
c1c26e38af2dea19d96168f12ed50838
a6aa509c7fd534bc9ba8376303ab4d2bdd09befd
33383 F20101119_AABJVW katar_s_Page_041.pro
dd58150f756f49e9dce85dbac53316aa
72edf0826a7df41cd31aec97b8a376141118925c
1653 F20101119_AABKCF katar_s_Page_058.txt
063364e11dafae8490320893d32246c2
a9cde10c5d945ed40960da7cbeec22cfed81249a
1980 F20101119_AABKBR katar_s_Page_044.txt
87469f9f42581da784d16070588d285e
8c3935bd7240ee543e25b9d64a2ebf4fe7faf759
49234 F20101119_AABJWL katar_s_Page_056.pro
a33aea1cf78d049c8adb214afcf8d662
e42336723864333f408009d563fc88e01ce30d6f
58689 F20101119_AABJVX katar_s_Page_042.pro
831725eace8416d1fcd453a102034297
2aac8fc18dd731b7e4618c5213ab3d01d97a42f9
2029 F20101119_AABKCG katar_s_Page_059.txt
42f4a0c4dd51329ae70427603277ec2b
1f3d2f4763100775487810db1836f47a5d2bd3c8
49765 F20101119_AABJXA katar_s_Page_071.pro
ab6c73a3aa7e2d9dd4dc4b1615b84406
da464c46099a0473e91ff944563928dbeb069013
2075 F20101119_AABKBS katar_s_Page_045.txt
76b24ca84e2c69d07df8d3784d3782c8
baa789b9ea368f26418732bd005ca1a62a468c6e
24302 F20101119_AABJWM katar_s_Page_057.pro
df86dcc727a850af5ee5c94d3ed33fb5
dea21ca9661a8bb0b998e820094ca9a132aabdaa
50616 F20101119_AABJVY katar_s_Page_043.pro
9a31a5a95c5de68f530b62dab733af02
5ebfd99ea91a18405a2973ff29ba3e09a1751311
2052 F20101119_AABKCH katar_s_Page_060.txt
9688770fec1e2e764d6206a96d7d6569
ae644306b119ad6250ea2b0b91a9a7aa6727701e
30997 F20101119_AABJXB katar_s_Page_072.pro
637b2df39df44e8675528445b68f92c1
69401b2ad09b435ce3091747149f37f305530249
2473 F20101119_AABKBT katar_s_Page_046.txt
a5e3c6962fbed048e9f04decdac99644
fc7015d7c7e22edab39c4fe2b6180e05dc59e1d0
39562 F20101119_AABJWN katar_s_Page_058.pro
4780256d07d94f43938625ce6bf6fc3b
5ac6f1b41eebbfe8c168a9241844518e98d56d68
1127 F20101119_AABKCI katar_s_Page_061.txt
fd1e17f4cbe3510e53e2d1d325135810
208367a0753fb5f794f0bf65d64f0565137fcd99
43569 F20101119_AABJXC katar_s_Page_073.pro
5907bf4eb4409df1969ff7de9089afa4
ac3d7ae13c97a1daeff371ffdf852074eca96052
1804 F20101119_AABKBU katar_s_Page_047.txt
8986c286828f271b8d02fd39f36f9b77
b4da5f75476d0a588427e9bd2a245989bfdd2d3d
50507 F20101119_AABJWO katar_s_Page_059.pro
f1d6acd3adffd3af5cbbba804d7d84e7
cc0a002fbd2d891b4de816a2051883cbe7a70264
50269 F20101119_AABJVZ katar_s_Page_044.pro
13832afddbc5b11ffe8a5697f7c25680
70ab6df77535f62a670b87a5d0645103a6cd607d
1966 F20101119_AABKCJ katar_s_Page_062.txt
6d831241398997d7935502b4c70cf383
2bd5966e0cf001a2cf5ad25e0c5ff98c72883e86
39290 F20101119_AABJXD katar_s_Page_074.pro
974ab8b826b85c0e222ff5dc289d1072
c57781f7a346978548681d382a7f0dce58473bee
1859 F20101119_AABKBV katar_s_Page_048.txt
3dceb4e04422a025a71e067707c785ff
042e95382595b945ff149e3bcef403df4938a583
51272 F20101119_AABJWP katar_s_Page_060.pro
3473c1d70bbf840fa694d552bd3934f9
5a0a3b299c179ac2c4f17e5630a0896fdee2a7a6
1905 F20101119_AABKCK katar_s_Page_063.txt
0cb8654c97b266cc99dae30b25a5870d
c38d7f9aab0cb7c12cb6432be4802f24f98cfa70
35353 F20101119_AABJXE katar_s_Page_075.pro
da262a955f3a309ba4cc5992d108c9c0
46196b645a383447988d99a0f0a43fe68d641b6a
1675 F20101119_AABKBW katar_s_Page_049.txt
1ffbb26b28316ed2742a9ac52bb44f46
556060de1af657c7c2991d82d97f27a58cd13395
28260 F20101119_AABJWQ katar_s_Page_061.pro
b4aca7fc2c3d41732e3d8e41cbff7d5c
8dfd0bba53f0903e0a366bf8ef522e926b4564b1
1782 F20101119_AABKCL katar_s_Page_064.txt
16ea094cbcce434944ff1acba38deae7
56dd354a77661e46c9bdbd12c448665a48fcd903
23536 F20101119_AABJXF katar_s_Page_076.pro
46066bb8abe2d45f1d759732538192ae
a84a8aacb83d6d7393c91ecd88356390950c7781
1379 F20101119_AABKBX katar_s_Page_050.txt
caf421b99337d8e4c6146fa5d056e6d4
a96e962c49729d8f6982ac508b3f2b7ff55a4154
49119 F20101119_AABJWR katar_s_Page_062.pro
e3c8a21c9c1f9f4ee798f37d820f4f60
bdabc38634a8023845cb8b5872dfd17516b1a2a5
2626 F20101119_AABKDA katar_s_Page_080.txt
22c1cf80ad825d0538999d85ad068164
4f57bff8775a2932c1fa03a896d6d7aada443a48
2016 F20101119_AABKCM katar_s_Page_065.txt
bbb7aad06c18697f100c6e7feed52549
2ca8d409d53ff29c5e98bdffc643473740d2ab2b
43980 F20101119_AABJXG katar_s_Page_077.pro
89bffeeae95e7c610172da201883e462
ce0245bd4da2672eeb3f1e3fcb861cbd2443ee85
1294 F20101119_AABKBY katar_s_Page_051.txt
d7a4c2b60c71f11808a86033abdc4592
3fc94c251ad8921833facf98c8830e9ef9c53929
47359 F20101119_AABJWS katar_s_Page_063.pro
4d4545981a23a7860e5e0edd70613ae7
317e82b96098386fcf1d14d51f4f37e298578144
1621 F20101119_AABKDB katar_s_Page_082.txt
0ba429c25bcffb66df4a1b3d4e191c26
5c1e500e5569aeae9431cdcab7c3a5709e86839e
2056 F20101119_AABKCN katar_s_Page_066.txt
2684eb1ab012e7f063a68d6d5ad63720
c4d379b7544e68f79cfba20a8f3a9059a43b73d8
41435 F20101119_AABJXH katar_s_Page_078.pro
ad37a111f81b6b9f073afa3a4da27b05
351ddb50b35ba6f60c94fe56ba88beb0c3652989
1982 F20101119_AABKBZ katar_s_Page_052.txt
5e013a18adb09b6272f3b9e35303c394
b76bb388cffc666ddb2dd23973da87a95527e83c
33836 F20101119_AABJWT katar_s_Page_064.pro
23305c47fb7bd56d439cd1cd9fb6b0a0
e26d2d9c51b2bb1dcd824f93b85129f73185189d
17625 F20101119_AABJAA katar_s_Page_101.QC.jpg
c67d973b4906cc763665552f19f345cd
61052e8f40a10fe8805ef1a8a13dfd43a650619c
1297 F20101119_AABKDC katar_s_Page_083.txt
b2428e5221aa8ab5237d0de281ca8fbd
d7926a8f63b212c683834486d08e39f356f4119d
F20101119_AABKCO katar_s_Page_067.txt
37fc34bbda9aef5e1f1d4352a5c34092
6aabae6dd8ce05afb7f237fc031b951b94ee1249
48969 F20101119_AABJXI katar_s_Page_079.pro
f1692663f6d58a04253500091a558136
ab6f4bf2d55413d1157a5de7c294c863365232b9
50010 F20101119_AABJWU katar_s_Page_065.pro
9e4fb9ff4de0fa8dd0f86422d5abfbc9
5612b818d8ac43ec6e317c2450fee28d8bfc9972
18398 F20101119_AABJAB katar_s_Page_072.QC.jpg
96054197e7431b80fff4d895ed3112d8
a17ed14b6ea4403ae8ca5f5d16c9ae9686cf477e
1783 F20101119_AABKDD katar_s_Page_084.txt
b2962e734211b82f5da6f1b72a634924
1f4de9c1709b761679dc90e6885381ad2cc2baeb
2417 F20101119_AABKCP katar_s_Page_068.txt
377ede1f0bfa68addd5f3e3ef6475775
0aecaa9225008a11f2748a075d6c79d8cc63539e
49357 F20101119_AABJXJ katar_s_Page_080.pro
cd9163c961226645b24a4e1afe15bbdb
14dcd55f4d30f79f0aba2b38bbaf3a2f62cb6a91
48478 F20101119_AABJWV katar_s_Page_066.pro
0d81b49d97242b6aaf792ba0b2739601
34b8abc46e716a5021baf11967bbda0b88a5db76
F20101119_AABJAC katar_s_Page_106.tif
deef4bb371cc59ab5ccd4a9133060bea
770286295d149d83d98c23fab0dd1a90cdadeb4a
1959 F20101119_AABKDE katar_s_Page_085.txt
9e3f998294796d13f90d9d3f2ee1f32c
4d8b7210cf6a1f8ae20fce06b54e49506d406da4
886 F20101119_AABKCQ katar_s_Page_069.txt
0fd2fea5f3e240e6cdc7d3b72f5ea3c3
eac56329e46d4676e108422bed6e9ff560fa9ad1
34821 F20101119_AABJXK katar_s_Page_081.pro
ad1655b3ea978c5f0a9e80dfe029654a
6ee0f535748d48e0934a15e83a6deb392adc04b1
50255 F20101119_AABJWW katar_s_Page_067.pro
a7ac78ef9d3986c0e9420d63e34105e4
21b2207065ceaa26e125fdcea8749ae76a95db70
F20101119_AABJAD katar_s_Page_149.tif
d88b7f7ac1815147d497c2184120370b
b5b5f016c625056fab07296c9df9810312c2e607
1890 F20101119_AABKDF katar_s_Page_086.txt
5eadccb8fae263e29c7bf7d3822e2954
d9260d4c85268feb4968c122c30e57dfd50042a0
1986 F20101119_AABKCR katar_s_Page_071.txt
a01467ffe76c610c9220c4c9ad37cb99
8911674da049a543967079965a9110f819c4243f
40213 F20101119_AABJXL katar_s_Page_082.pro
c0dcc291b042817336160ee3051c2bf7
1490c8e20eb674e744a3132030c6f0fab572c5f1
45494 F20101119_AABJWX katar_s_Page_068.pro
90093c2d7d3ba0d7a21a62f2e0d537fb
cea4e873f63a4b1bdb45c28b3321d8322e71f207
1481 F20101119_AABKCS katar_s_Page_072.txt
824192fe7f54964fe587ab740e388a93
e9583abefee7415317a39950a8780488f0eca887
30251 F20101119_AABJXM katar_s_Page_083.pro
96a003d1b42d36c7d45cd2880e9dd1b0
269a6bcf4d9520c055b514be827fa0476e7f9a90
17707 F20101119_AABJWY katar_s_Page_069.pro
9c9fc037d9f491154373de6263465e1e
15e07cfb8683f09c60220d4c9cc34de5f55b6ff1
69424 F20101119_AABJAE katar_s_Page_115.jp2
0d75562d7220f2f21d254a6dac2d5781
1c24fcc6e5bc4f45cbd1aca894d9d4047eea9911
1942 F20101119_AABKDG katar_s_Page_087.txt
db8d60dd9bc6f93ada15a3877ee4b32c
b30d15732c8ca5cfa5c020aeb712a74915cc37d1
43304 F20101119_AABJYA katar_s_Page_099.pro
3d65867edcb2f1df64208b91a04e82bd
70890422735c4465c39c2020994b176c0e2e444a
43132 F20101119_AABJXN katar_s_Page_084.pro
a5199b7ac45f3589d60a17d97ba6bceb
d2ee66c3e4d02e3cd15fca17fa21a73285d8d6ad
43746 F20101119_AABJWZ katar_s_Page_070.pro
b18658032f22f0763a5d2d4b76f92da3
63b33e5f77a5f1b16346022e11cdfaada05ac9f9
4821 F20101119_AABJAF katar_s_Page_049thm.jpg
8969db98db07759d8ac96447822eba0f
27bc8ea2faa8fa3ae993425260548f72a4962285
2335 F20101119_AABKDH katar_s_Page_088.txt
f09d7473b0de4b2655172b7955efe10a
3fe7c87f163754d9ff4371168ac35e6b84904a1a
40678 F20101119_AABJYB katar_s_Page_100.pro
ae928feab7bf887f0ac00806373bf96d
c820872db54d4b3d1b02d68de56ed5519bbdc89f
1731 F20101119_AABKCT katar_s_Page_073.txt
b14ce86fa792aa190193c279ea172054
68512b03baaff3da6c8b83a9e21ec9fbba5e390b
49029 F20101119_AABJXO katar_s_Page_085.pro
ae0a6392ed817df7e20ffe18ca915826
d19ac45a4414a6304306ea8873ca105b11dec818
F20101119_AABJAG katar_s_Page_135.tif
89c4f9cefe337234fc8268ec8001efda
fae16e48415fb926de0136a38b1dd625028ac154
2495 F20101119_AABKDI katar_s_Page_089.txt
a8d6a0949386fd59e3b81f744ff21264
22bedc42761de2dca68496016f8da10785cfe45a
58664 F20101119_AABJYC katar_s_Page_102.pro
1bae24fef42265f7c3f0a3d2d31d1aed
68e41febb6aa0f392e0badfc79ae7b239d2670e8
1688 F20101119_AABKCU katar_s_Page_074.txt
3f2d480e2039e14c9228332a1335cdce
56abe1babca2738f9b823bd76d7ec91a2c74b811
1659 F20101119_AABKDJ katar_s_Page_090.txt
e644cacd00d56c96647bbcd37aa82421
5d876ca58c9c8423a13e8e0a75d38f10e3ee8d19
47951 F20101119_AABJXP katar_s_Page_086.pro
469664eba5c2959ca22a4208ac85f103
94707513ce4458785bb732881f6cd6b22a030297
71093 F20101119_AABJAH katar_s_Page_005.jpg
176553c501b84fe625aa817d9e180b89
defa53fac24d4a70f791399b740219eea40b117e
37410 F20101119_AABJYD katar_s_Page_103.pro
a3f24883556b488129a7f657120440ea
4d54d0743a70b145cabd6ba610bc74b6f07de878
1527 F20101119_AABKCV katar_s_Page_075.txt
e4bccd6fec95857db13e9d0e0e1f4d7c
c6d345c70ca625ae22623ad24c882cfb5ad1bae8
1737 F20101119_AABKDK katar_s_Page_091.txt
b3063909b4cf95403b6af915002e10ec
7c2cfd76b1084d07520f4790962c2071faacf537
49382 F20101119_AABJXQ katar_s_Page_087.pro
2449d5a7990baa778b624d1107f83eb2
6132162179db221b04bb636c93c86c103657380f
65001 F20101119_AABJAI katar_s_Page_124.jpg
319732c9b59247c09984e07a18efb380
e45129ea7f222703b76800fa8be849ad27a2c3c3
56473 F20101119_AABJYE katar_s_Page_104.pro
1a91aa95d21f1395cbb34f8c7884f06a
5ebb526b351dcf0ec73867b3902ada0a32bca51a
1237 F20101119_AABKCW katar_s_Page_076.txt
144b14e026e762a6c7f597d775a8a13b
43d6cff0223175a0079374ff5fa1bcea20057d5d
1690 F20101119_AABKEA katar_s_Page_108.txt
72f4c516353a144ecd47f0e889eaaa12
54d07b8bba247f4c4c32841ee7e64a3febb5c448
924 F20101119_AABKDL katar_s_Page_092.txt
e7346086cc17be88d0f134b241146e31
bece9946cec9340f604ba6cc1619d6b1a4eedd37
61142 F20101119_AABJXR katar_s_Page_089.pro
79a4cccef5d16849101571297252db33
6fc3012850576f08d5e5634171d8bc25bff22dcd
4692 F20101119_AABJAJ katar_s_Page_061thm.jpg
5b417bfc5eb7bd7c1c43efb32aa3bc4c
e0b6aa60b52d9856dc8cd59502851142f4f84249
29174 F20101119_AABJYF katar_s_Page_105.pro
0e2839b6a791e452a01b721b0470098b
cc811d4c0fd7a5d9964582cb6c3e60b87f285037
1916 F20101119_AABKCX katar_s_Page_077.txt
23f8c83884cac702816fcf373e2e0c26
2b78e11220ccd5714268407501c7bbe17789ee2a
1846 F20101119_AABKEB katar_s_Page_109.txt
b43e4e060a6ef20c91c083b80310fc1a
cbf95a0e19164f32ef2a96d4e885effa0d0cb505
1817 F20101119_AABKDM katar_s_Page_093.txt
354cee2ff653567f43c2de0c38b85ce9
0053c0abccd5dfbf7594084fee7da76a7ed0fd95
36104 F20101119_AABJXS katar_s_Page_090.pro
eb90f6d977485ea455b12a0a9cf8f0c2
859b99fe5e49d61f1b6a1f503f63102ab513850a
17885 F20101119_AABJAK katar_s_Page_161.jpg
721f7baecb86f023f614becfb8bd2dc1
6a7dca55b96b7d4b724ae99db9396f7b76357fbe
54213 F20101119_AABJYG katar_s_Page_106.pro
332bad133925bc1c5d64991fda7ed344
8473d3e55c3b9a7118fc92fe4ce5bcd93168f382
1757 F20101119_AABKCY katar_s_Page_078.txt
9622ac3e869add05a692696aec855e2f
f147deb997981634af911a277f72467a228e1a0c
1898 F20101119_AABKEC katar_s_Page_110.txt
12de6732bffea9124f3ef75096c43b45
fdf78d39747887ca3f0a61ef457e626abfe83708
1825 F20101119_AABKDN katar_s_Page_094.txt
76a50b91d16caa593bb62ac9e4648902
0b143e83de4e247a6d4743dd98ad7982937d7275
38248 F20101119_AABJXT katar_s_Page_091.pro
3409dc762b1bde6528c4d07a704e8ea4
66736f1c527f37d724c9e6341a6f201fe0b6d3de
507 F20101119_AABJBA katar_s_Page_002thm.jpg
b45b86e32b67d159b8930529a8a87134
a946abe25c56d2949d0cb99fc053a158084f6dcf
2114 F20101119_AABJAL katar_s_Page_099.txt
fd7e4b987132f6bad4e2f08eed9d755c
27e10599d16cdb3b7eecd743572cddf1c20e50ec
10285 F20101119_AABJYH katar_s_Page_107.pro
1187112aba19d5c51bf367211d4d3c89
811b9a38af812070c7ffddf66a3c9870e95acf80
2459 F20101119_AABKCZ katar_s_Page_079.txt
0ffd027d49c8dff37f57ac851bfc8135
a45e53ffbb05cb6bca44baa91a7b461e1b114b9c
F20101119_AABKED katar_s_Page_111.txt
8c77081c056eeb092e0a091a683b12fd
aff684fc782809a9aca226addf0613bf40346c29
1828 F20101119_AABKDO katar_s_Page_095.txt
60092d86ff908a1628faf2e42dd83b47
a3de1d6d2c62f798b97690e1edf97110a09e7eb5
21643 F20101119_AABJXU katar_s_Page_092.pro
1ff77ae77efd99faacea67278b236fa3
d81be69834d8195b19e0afa74925da0497766060
15000 F20101119_AABJBB katar_s_Page_145.QC.jpg
93a71618faa8f16372dc3adebd258fad
d5fd7456c1aa21a341ef220d70273621d7a581b1
26731 F20101119_AABJAM katar_s_Page_096.pro
cf3bc0b6d9deea89b59184c1e40db6e8
72a3c6afee3f1869684d4cd02f19a6d1f9faae1b
40411 F20101119_AABJYI katar_s_Page_108.pro
9b23bbce5250aa68b9c2e3d7b5a9b1ac
2ff36e06a023e9b0830febfc0106c0acd86c7d10
1962 F20101119_AABKEE katar_s_Page_112.txt
6a9055fe1cf577a1bebbf8c5c951aece
49e93bd756ac8d4f2893c93e6cb5dccd95226789
1370 F20101119_AABKDP katar_s_Page_096.txt
42f4383df7fb170b2e85a659a1b06701
04f10eda0526cbf8c301a280055d8d950b6831fc
37014 F20101119_AABJXV katar_s_Page_093.pro
8be137dda6a086ba14a43607623012ba
befde82c283429b2dd8ef364a50f910e1f26d204
1645 F20101119_AABJBC katar_s_Page_036.txt
5ec0664a89db930fc3f9cc29d20ae802
f011c8017b635ebcb775554bf1978b54f9066cf6
F20101119_AABJAN katar_s_Page_100.tif
4d0135315495f9bb8944bf2982d44350
8fa26a4018aefbfaee56bac6777921e45a0bdacb
46731 F20101119_AABJYJ katar_s_Page_109.pro
aa48a4e74390a4561ae8a10ea26203f9
feb95cbb5e5f2037f658c650ab045d5491e5d316
1768 F20101119_AABKEF katar_s_Page_113.txt
5cda41761f8698626959baad3defb60c
17cc71c7d692ce512d6d84fc1b61c40bd500c240
1901 F20101119_AABKDQ katar_s_Page_097.txt
0c8696ea205c089a2bc36a62d3bc72a7
17a45164190517bb6f3efa233120eb212019d56f
37841 F20101119_AABJXW katar_s_Page_094.pro
5452b7b12fd2547739ee6fb47e740d50
c99a5e3f69d3c5dfc9d940d1ea549c677b90673d
76050 F20101119_AABJBD katar_s_Page_119.jpg
6e594331bc186cab65f06b3d4fa98b75
af9c6774154291802bf9ece64f0947aa0c3415d9
1646 F20101119_AABJAO katar_s_Page_013.txt
4a2e2097d7a352def9a8ffd75148d261
080f5eb60fb6fe83098a29bf76a2d069b4d247e3
47890 F20101119_AABJYK katar_s_Page_110.pro
9343345d67c48a585562d9155877a531
f4977b697a422755c34c0def7def237ce53c9d05
1595 F20101119_AABKEG katar_s_Page_115.txt
e5a648b8ada59c53f6f4f1a293f3219f
7af8ab30592159cc0e3bff965642c940644656ad
2577 F20101119_AABKDR katar_s_Page_098.txt
c9e61aed921ea027a1b2139882e28c8c
d73acea25e648e3d2c62731c4e7efdc639dd85db
45182 F20101119_AABJXX katar_s_Page_095.pro
9712ed60e59fcf2ca46b5ca204306ee0
c27adeff1f60d7942d261b5f9c599f0ed3be306b
80283 F20101119_AABJBE katar_s_Page_140.jpg
24bedd04ffa91f4230203fc374243576
1e27352286011618fae590d03b3e26f928776b39
27636 F20101119_AABJAP katar_s_Page_045.QC.jpg
fd4112a16de0428c1c09ba0195a1d5d2
ac9af7126c46b39c9f1bc4b87d3e1df0bc6e8019
48944 F20101119_AABJYL katar_s_Page_111.pro
b598025636906a058e74e03398b8bfaa
103fbbd571308c265b7c873312bebacd06a12917
1889 F20101119_AABKDS katar_s_Page_100.txt
1c1591d362203db8db5a96b98cbe9557
72a513e9bbdb1c756e71e72acc66375af481e79e
47244 F20101119_AABJXY katar_s_Page_097.pro
0488377a177b9e4242695a6c23d8d6e5
85bb0f0a3c89345e707b5cc1e3b8f4b7b32324ae
24321 F20101119_AABJAQ katar_s_Page_006.QC.jpg
506c004a38b1b010d16fd2a177b290e3
3286318a1ccca1e84755e7813338c28fce2ce592
45824 F20101119_AABJZA katar_s_Page_129.pro
2254796c42af2952abfde1cd0cb2b02d
5d90af31f4f6a788c13f37278a61891ccdc49f09
36394 F20101119_AABJYM katar_s_Page_113.pro
da21eb9eb765faede311a34b911dd392
b1088d7d2d755e2b8840b2d13725b615f300325d
2592 F20101119_AABKEH katar_s_Page_116.txt
62fc533abe05df851a493026561b0395
202c036f4ac91b8767ec64716730ee1c1084ba2c
1612 F20101119_AABKDT katar_s_Page_101.txt
6ad1260f032f0fc26a6f30db3daff888
f1600441a0770798f6e17ce9f811ace0afe9d925
56002 F20101119_AABJXZ katar_s_Page_098.pro
9c75ce5198f7eb9cdd33b31b3eb4d61a
4e859a3359049c2b2c3974a09230f1a2a7714258
F20101119_AABJBF katar_s_Page_086.tif
53381fbac0ce5e079e66d35139cf6e35
9aa0bbb7e2d4e87d4ea829cea628a78a8985a048
46074 F20101119_AABJAR katar_s_Page_025.pro
9e656a571596fa0e632286c27032afe2
0431f13b14a0d561d52cf1898a2e7a1c18f8e2cc
2893 F20101119_AABJZB katar_s_Page_130.pro
9bc2c4ad331772e2e010dfe912197fb8
9e0442b91c11c091889876e0d20b2e6278dd3a24
40156 F20101119_AABJYN katar_s_Page_114.pro
becc027ec291099e7fcbb788c9914b2f
b27e2050070d6448153a434ab9e28fcfe8d91329
2647 F20101119_AABKEI katar_s_Page_117.txt
884faa5e8a47059ecccfd89b90866f85
b68662657598fb0dedc4556376ba7199673d4a50
3299 F20101119_AABKDU katar_s_Page_102.txt
0dc838ff997ec7fcb0cff4f6e8b68bf9
1c66c6cbce38e4ced8add3cb5e028b9ff90dc02a
1881 F20101119_AABJBG katar_s_Page_081.txt
22eddf27ee8ba699b8370eace4e649eb
b68fb7baf3885a52c719ddd0218cf4919894920a
88438 F20101119_AABJAS katar_s_Page_133.jp2
61d4f6bb94b67eed7fd2b0dc3d8f14d1
823b1429c2236da820504f752c855a2a0cf5a125
31733 F20101119_AABJZC katar_s_Page_132.pro
f9c542a406663c5dd6f0bf850edf9a8b
d070ea893bc6b09a2b68799a3494d4b2174c716d
32734 F20101119_AABJYO katar_s_Page_115.pro
6c13943c2f90e5d0b383c398f33ceba2
6506687b31b76f34caa7ac7748c44c112d5ddded
2074 F20101119_AABKEJ katar_s_Page_118.txt
bb154807f36098f0657aa894e7ad430a
d656ad8bcc4f59be753b59478bf7415298499372
1924 F20101119_AABKDV katar_s_Page_103.txt
98495ccf93a0a6715bdd543cdd35c413
9c1272b96f8dd2d29cd2168b7246ac8694c32af8
6607 F20101119_AABJBH katar_s_Page_139thm.jpg
667263232871353afd8d781d526cd3c5
8acc099c752a33195056988be688aa6564297ae3
72595 F20101119_AABJAT katar_s_Page_126.jpg
5eac31828a15de4e8ab3de8005b32f55
1e7c643bb3f157cb081ccae59960819f7ec1ac81
43081 F20101119_AABJZD katar_s_Page_133.pro
51d6db4848d97c2ac62f20c156ddf7b6
956176cf1988a51242e1407404c0dd6c421c4ffd
59236 F20101119_AABJYP katar_s_Page_116.pro
94ccb97495c67fbda522d0509b4fd549
0256ea134c0c3856a33f40956996db3c8c80d45d
F20101119_AABKEK katar_s_Page_119.txt
ec66bc23e8430ede81b09769d7117f01
1772d19fa6eac2745ec866579f79247aa682eac1
2263 F20101119_AABKDW katar_s_Page_104.txt
5b327c00c100bfe4658b035028cc4cf0
39d762ac4b83f8044660ef49bfacbab9bd6e37db
F20101119_AABJBI katar_s_Page_088.tif
1c33c6ec0e746f552be3fb913fa401aa
c99bfa4e6146ca4f4daa6863c8bb0ee97fb4c820
35870 F20101119_AABJAU katar_s_Page_162.jp2
1cf602eb74f366b39f9efdfa44489553
4a0f36886f972bf5f188e3bb14ba1fdd17ac14e9
3342 F20101119_AABJZE katar_s_Page_134.pro
746a5a8f3683efc890b9d423c4a61d44
1c1b7422b4515efbd362f58aa4c11ae058018a1d
50670 F20101119_AABJYQ katar_s_Page_118.pro
a0220acfa9987b93c32e8892ce30f567
66dba26e8217da98190844a5c10a82be7468f4f7
1115 F20101119_AABKFA katar_s_Page_136.txt
64ef16aaf6666b16fc117111534ccf14
5823eda1bdd15a3eddb0dcb2a29fb028a6bed6e5
1767 F20101119_AABKEL katar_s_Page_120.txt
cf2c37937872a8c425a63c356be9fac8
cb75829543a9234a99933686f539b6ed11c45769
1469 F20101119_AABKDX katar_s_Page_105.txt
b307ad379b3520701b6eb33ecee1da35
213e3035cc6086ff7bfc9b58c04e6a6c76a77092
68793 F20101119_AABJBJ katar_s_Page_093.jpg
3d9b05f881986bcfb88101992f1e0a67
62bbc17f236305ccff0d0e2a93acdcd5c2a3db14
20608 F20101119_AABJAV katar_s_Page_075.QC.jpg
51691c869665f4993090bba22049ce2c
70de25ad539f4972fa4633f5029fcdeec73a0da1
39809 F20101119_AABJZF katar_s_Page_135.pro
6a5c331c46c6a0ce2fd2a1e590eff01c
6f45992a0e1f810be7dc27e0ef17d0906f83c2d0
45021 F20101119_AABJYR katar_s_Page_119.pro
5894dda4370c3c3dee4df9588737314a
35d4c6c2e41988286cef164b91f8d671343203ed
1939 F20101119_AABKFB katar_s_Page_137.txt
388184ac36940b78b8bd37fbea18bd65
56a9bf8be1b2643a17326c442d983144fba1030e
1707 F20101119_AABKEM katar_s_Page_121.txt
9d5830bfa2337254c5836cad20901c51
48e73d678daec9ac4b1a0fefd8de6d9538c442f7
2575 F20101119_AABKDY katar_s_Page_106.txt
33475ff11a75b9f1b644e106e9f05405
a65c48e71116f04d6ea2e1d70138941918564f55
6055 F20101119_AABJBK katar_s_Page_107.QC.jpg
52a942797b2b0b07e255a5324985d6a2
fc53ce9c1e3137782588e0eff8ff65d2fc04fc21
71922 F20101119_AABJAW katar_s_Page_137.jpg
c57f27207e4ab4fdbf9c47fd8eb55068
d65a2e6d9309511ad1189aee82cf1527274dcdae
20677 F20101119_AABJZG katar_s_Page_136.pro
ce54ac39bcb011fe99d9e6fdc702fd12
b64727462835b13d2409176fedaa12abef738bae
36867 F20101119_AABJYS katar_s_Page_120.pro
c04158814c7e9364f7029ea3aef5c968
1f0663c562faff98ce229336bdcf8d7d466bf4d5
2994 F20101119_AABKFC katar_s_Page_138.txt
e6f2e2c270916100055f540744e860fd
6c5a3057a06e96548bbb7022b2689b070df9a335
1716 F20101119_AABKEN katar_s_Page_122.txt
540d161a0542de583d7ae52ec7db5715
66a3c1d35b6526a106a84aacef3028ea04c17308
452 F20101119_AABKDZ katar_s_Page_107.txt
bd012d2266e941aca19c07afb24cd7e3
eb2962eaa0473f80ae2951100252d19fd15e2ab1
5601 F20101119_AABJBL katar_s_Page_003.jpg
0353d8919886bf38702d561c04bde7b8
092121a2f3cf1a7245e68f809d8ce55185f3f7bd
49617 F20101119_AABJAX katar_s_Page_112.pro
60830e6cd77adff11236ef7fa4e5fb35
307c3ee1cc8addfce77ad72f651ef4710c7b3148
42083 F20101119_AABJZH katar_s_Page_137.pro
cfdf3e6167834f089c3b86451190a15f
cfb68787c0f1d3e066b4de1eca43ecb3611a0cf8
35104 F20101119_AABJYT katar_s_Page_121.pro
cae5dcfb94185868048323ae2440efbe
fc825974d2e5e62966518743f86b61e3cbf0d67f
4571 F20101119_AABJCA katar_s_Page_032thm.jpg
a455d0ebf461b8b3690f3818f9820af4
06167ad604c5b128017e216e66889a772e3794d3
2339 F20101119_AABKFD katar_s_Page_139.txt
4c8a1a4588dcedf7615f41601ce3a658
b1e421bbecc1ca4771386dc09752da9d87bf9297
2258 F20101119_AABKEO katar_s_Page_123.txt
37501ce5eb3ca03e1447c07f4aac799b
8138c703e5bcd8ddbb70e400c342323ee8cb2314
47392 F20101119_AABJBM katar_s_Page_088.pro
d14bea48bdcb8feec7f04294533c7040
c42e744d83147cb9847fc6cbf660dabf8f9b5f98
62516 F20101119_AABJAY katar_s_Page_149.jpg
c17bca03fc31dc9aeccbae84c8cb744b
0112afcd381eebe922c15c12e5621e582393f5b8
73114 F20101119_AABJZI katar_s_Page_138.pro
800aba48398d23883c6fe63c8550e28d
5480f659122fab1182be9d3a5a53c4e488f9ffe0
33979 F20101119_AABJYU katar_s_Page_122.pro
359988ab643ae173da081fcb5020df8c
2e65ec7569c5409c2c02d802513f0bb8e95894cd
25905 F20101119_AABJCB katar_s_Page_043.QC.jpg
509793601338913e3c486e8e6270593a
4acb4105f19b7e20d63ffad6eb0a5d1737eac04b
2219 F20101119_AABKFE katar_s_Page_141.txt
c9f5d16d09a086f6dd33cc247fb77793
914459b80b0cf899fd85d05e30812f17e53dfe2c
1657 F20101119_AABKEP katar_s_Page_124.txt
d8ae6da2786f9b09e0c1eb3e364b0e48
cd6023fef05a12150c563acdb1b9cd814752f41a
F20101119_AABJBN katar_s_Page_059.tif
7a7ead7ba4b492a5d701bad7b3778bfa
0736741bb42e49eafdfb40561d863078412f3ccb
27053 F20101119_AABJAZ katar_s_Page_037.QC.jpg
d953db99c28457b82c887ef09ac2f719
51cefc13f93ca226f22d73017b4259de7857d55b
58520 F20101119_AABJZJ katar_s_Page_139.pro
9676783f92257d36480559f994d9a252
eda14b6c578cd84dc5776e69ee9d949b0774d2be
37990 F20101119_AABJYV katar_s_Page_124.pro
666a58ebdbdfa979bbc21b826578c3c9
376922e2ca373ae0e9ba12f30f9c3db1629fc274
F20101119_AABJCC katar_s_Page_024.tif
9201137b4ee5455bad9edf272a263f88
47a7c3a2172f6c23ffb437bb8d6945213359e28f
F20101119_AABKFF katar_s_Page_142.txt
028606461f11be91599c470d045836ab
40afcbfeeec4ebaaaeefdfd204dd5b23830d784e
1765 F20101119_AABKEQ katar_s_Page_125.txt
5be05c5258770dc7071e6926248bd3e3
c391a75ed3afd393757fa34b64cc51071a097de5
47915 F20101119_AABJBO katar_s_Page_030.jpg
62a98f1a17afc17b9dab1aabff7e4578
845463ffe4e7903b0e4bd246a3bcc7adc51a89ec
50161 F20101119_AABJZK katar_s_Page_140.pro
f8b54e025325fe507dfd3e0a2cf41599
329d73ddf5ddadc9006ed6d6520539dd6287a84d
32845 F20101119_AABJYW katar_s_Page_125.pro
8f87564462625d7bd5b436add329a2c0
27793adc6fef75fa777094b26be9b72ced4c1081
28723 F20101119_AABJCD katar_s_Page_159.QC.jpg
8e8677e9d16e15ddec416ac5947a1c1f
748f931d878d6e0cc72c515de91df122035ae08b
1365 F20101119_AABKFG katar_s_Page_143.txt
82f8bd4d33f26d8c14df63526a313d70
55784ee1d4becdc8c9a8a68af0352e38cea3395c
1791 F20101119_AABKER katar_s_Page_126.txt
6167a4f1c991b6e16f303ffee0cc21a5
a1ddab9291961294fae994cdacddf262103e4295
1051933 F20101119_AABJBP katar_s_Page_037.jp2
5ea5f11e4f91f01f0a1e91cc094e8f80
a30b52a1d86d1401cbd7b8fc83335bf691e2c4b1
44099 F20101119_AABJZL katar_s_Page_141.pro
55d1b65f7a5b453171c26f5f20fefeba
4b3a4198127a0121f38d551194b9857ca89b7b24
43068 F20101119_AABJYX katar_s_Page_126.pro
33739fbd4734261a829c4fb7d0c9e484
01a4d07cc0c6431d8a56c27de6558ab14e235ce4
85622 F20101119_AABJCE katar_s_Page_071.jpg
05bdfa15d4ba4b2661ba99b707160a7e
b455093e3538ab520bcb2acea8cbd6a6e9ad4a4b
2168 F20101119_AABKFH katar_s_Page_144.txt
d989dd3c729a4daf569a3c00e64fba5c
4b86560f7c52321bbe160d9835f163f9daccb2d5
1098 F20101119_AABKES katar_s_Page_127.txt
6c84ae91fe63475dcf204eeec5d5bb2d
ce84386cfa8e09ddc8112c52e17759a599d8899c
25846 F20101119_AABJBQ katar_s_Page_027.QC.jpg
732561154dcf1aac73adc6b0eceb895c
166d05fa0b6192d05e32f36d23acd99464cfd670
35307 F20101119_AABJZM katar_s_Page_142.pro
e695d98f341455ddd2b1afb120bed46a
9e8c7d28b75e8ebbf0c4cfc3f130cbb41920c394
18394 F20101119_AABJYY katar_s_Page_127.pro
47a2e657c7a64e47d36166e7b8ec3c99
8d34bb6abbb741063e98623df76a2ca989cca831
15313 F20101119_AABJCF katar_s_Page_076.QC.jpg
0685ac3ecae8e1cbccf8306290d185b5
c9d8e39ab74bd8b3482698259418dd56c1014178
1904 F20101119_AABKET katar_s_Page_128.txt
fc5613519ec4496cb8189a4b819b1651
c9ae339064bd7086ee268d1d3de79bd4eab7d6fd
78066 F20101119_AABJBR katar_s_Page_047.jpg
2c31ee20ad60c57fe96f31d4f4a385a0
7f773edfb7926a1d326717d24b5d3855e7147b34
24808 F20101119_AABJZN katar_s_Page_143.pro
ab0c0af5e134b2d9cd2659440c9b3db7
c1d2c734a9cf16087191729896d91c106bea7a4e
43176 F20101119_AABJYZ katar_s_Page_128.pro
ac75cce88c672434238ac2aad0f2f85f
088c9200888434417fb2fa6fa5e6a73a086aebaf
1490 F20101119_AABKFI katar_s_Page_145.txt
362c6d366c60847693ee8b163bdd3420
1ec5fa2cceff8176ec0687792b3800b0f790aaa1
2025 F20101119_AABKEU katar_s_Page_129.txt
3355d2fb7b49f9ae6035c973b064b7a6
29355849d9fabe4bceee57127a50ec220d56a8a4
1800 F20101119_AABJBS katar_s_Page_070.txt
3501e2f14c10133747787f8682c67843
fee3e86b5a3b4aef3ffc1896847802ad20444ff5
36287 F20101119_AABJZO katar_s_Page_144.pro
e9edba753012e93ca5f3ba2e8299f434
51d0321a76488d23003b18926ef660f254508143
5589 F20101119_AABJCG katar_s_Page_073thm.jpg
92a71aa1c9611015995cfaa2c1caea91
1cfc5ad479bf881676a82ad451c65af6c425fc71
1888 F20101119_AABKFJ katar_s_Page_146.txt
9caaa8c7955a96eff8fc04944ec6d8e5
0649e5ba1afc554b092eb0f338025160994b831f
219 F20101119_AABKEV katar_s_Page_130.txt
ee27f4a3a2f294ec77e8489200c5b4f3
3d1bbaf8d100c090dc213c99afc958321ce163a9
5334 F20101119_AABJBT katar_s_Page_075thm.jpg
db3928bd16e74f61118819d4f20c7b5e
54ffc0890db5b8255544ef8e25636098d9498c0a
F20101119_AABJZP katar_s_Page_145.pro
351d780b871d3ec669d906b69b9abe05
3a5a9d1ee605dc44e9aca8e2ab9baca666bcfce2
19476 F20101119_AABJCH katar_s_Page_105.QC.jpg
ded12c425fd583a57cdfb21a10cae6e2
d40b7ad3a66edb336ea4ebfe4f56132fa8af2f0e
2270 F20101119_AABKFK katar_s_Page_147.txt
4c8151920ddcc86ba23ac1d4ca196226
976ff1da1a704563a2ec9f39fde932fedbea77cd
1627 F20101119_AABKEW katar_s_Page_131.txt
5ed6f01ed246c87957ce3e44c99f91c2
5ded7ff12b9568f6b7cad09fc345efdec2ef380d
5018 F20101119_AABJBU katar_s_Page_147thm.jpg
0f475af3c1f34dd929dae40ee39bc57a
89c6d607bd8a0c644f7a1d9a3f63521d8139fee9
34832 F20101119_AABJZQ katar_s_Page_146.pro
795f4ccc5649a8840aec1dcdab4bebea
35a3cc85cd4006398410dc2cfa496b7cd2dca1b6
63373 F20101119_AABJCI katar_s_Page_114.jpg
541d7313fc1e1438e4093e57761f9d7a
4f774626e682b524bd13ec0bccc4f5856b640e71
2668719 F20101119_AABKGA katar_s.pdf
f70020f8cf45011a94fc31f8cede7540
69814dc74501417dcbaba8504b365f6262db2282
2150 F20101119_AABKFL katar_s_Page_148.txt
c27333322b36c4a98a39743155ae501a
d34ed07020c77dfc9658cb8bdc6e6f71f62cf802
1922 F20101119_AABKEX katar_s_Page_132.txt
25aaf7a86e9e889dd65774e0ea9fa582
42f52db8fa22c800c4437f385d2b2dd2b17e33d6
2097 F20101119_AABJBV katar_s_Page_020thm.jpg
6246c8957eab6258d41efa07ad60a180
15df8bb29fe6c88665bae6144bc32360b365584e
F20101119_AABJZR katar_s_Page_147.pro
1313b0fc3238a2fe2abae0ae23d61d2c
3a66983bdc450a420c60e9a380c4cec6db6acfd4
64626 F20101119_AABJCJ katar_s_Page_117.pro
4c11914c34f0863436db25b8258aade9
722a6d2ee048c70402ac063f76e76b69d10717c4
27942 F20101119_AABKGB katar_s_Page_010.QC.jpg
09cc3f959fc6fe6b0232635d3d85d79a
a943434b3286467f61baad0bd678e3bc6dd71189
1261 F20101119_AABKFM katar_s_Page_149.txt
5461c1601ad61a9365b59c4c5356d131
86c103ec6d649bf0caa327253a648a2f99ccbf65
1810 F20101119_AABKEY katar_s_Page_133.txt
a394924eedb2eca14cbd83b7b039e4ea
464c2723f9028ac505195626d76c5319b4a85a47
108452 F20101119_AABJBW katar_s_Page_018.jp2
4d8983f3fb333ee78c1854304cf81192
5c68c8ff92b963fc93693d4f59b296515797aaf0
41022 F20101119_AABJZS katar_s_Page_148.pro
095ee13e1852cd54f57117f839078615
33a4d2430aa18a1df44c3ca2f0e373b3bd9e664f
919472 F20101119_AABJCK katar_s_Page_093.jp2
52192646ca2dd67894a0c26fa617bfeb
66c5d577741734d2d9b30ff1d44f6debdb38a65c
6615 F20101119_AABKGC katar_s_Page_060thm.jpg
dd8e9ed428864e35663390d189a707ca
f35cf78a5d47816a2e559645c11d4bff4e442a0e
1565 F20101119_AABKFN katar_s_Page_150.txt
733e812e869a362e75ea037a2abf1f31
2f1e2f45eb754ea634d782dac57a85d46d305b19
159 F20101119_AABKEZ katar_s_Page_134.txt
5240d0bae665eae8d6a12d7e3adce361
4c0644d5b1624aeb6a5c78420381d6849021bcd6
2314 F20101119_AABJBX katar_s_Page_162thm.jpg
621ed6e1d69f38325a185ce89b9f5a8f
8e339bf7da8488cd63cf77af5d0bc365695c2292
25653 F20101119_AABJZT katar_s_Page_149.pro
d1cbb2aaaef4f311d158fdc345a3f669
e6e65a03ddc71e015f6606f2f669242bbbde200f
25358 F20101119_AABJDA katar_s_Page_018.QC.jpg
b61279198cc8cf15a1031141f0c7ab2c
069072fb69f12b5edf15f7a54cc3887f3ad13ec1
F20101119_AABJCL katar_s_Page_116.jp2
bf9509c837555319e473f88b9dca3522
aca39c86c1c72d4bb970f5766326307f53ec7813
5037 F20101119_AABKGD katar_s_Page_100thm.jpg
11e1cb0f0dca24ac1030df4199493f5c
329e3e181d0d2c9f245bb1f8af2c37e76bfceadf
1310 F20101119_AABKFO katar_s_Page_151.txt
f71f62693f67f3927db5112073d0ffbd
c27627c5bbfbc4f5274d3a9042dca6d6acd87f79
5293 F20101119_AABJBY katar_s_Page_074thm.jpg
fb6a397a87f6b0d52ee910e9301c5ad9
6a498bb48605a3eac8a64411a8727539dc8ba758
32192 F20101119_AABJZU katar_s_Page_150.pro
0af903f1877a1e6a9aa30c8a9c44ffc9
23b2eb087bd716f586ce9664b55ad5735e350140
24829 F20101119_AABJDB katar_s_Page_034.QC.jpg
f85e08e683487a791faa9003192c335b
8ff3168a8304ce86cbd05a8ab0347cd0274d06bf
F20101119_AABJCM katar_s_Page_078.tif
910dc9ac8fd4b4c7dc27684b03910f7a
4d878edfe5a9f56766ba6ce4211d210a2d4edc23
4530 F20101119_AABKGE katar_s_Page_029thm.jpg
f7d06bf5698eda1117337890a433776e
e5b410967b1a340fd90584c40d4ea6590051ec0b
1689 F20101119_AABKFP katar_s_Page_152.txt
898b3f28786ddeae15c7e7a1784836b9
d166a2f3ce5289d2450bd966c28a270a01c694e9
63895 F20101119_AABJBZ katar_s_Page_007.pro
c5635bace3bed61034ff27b70b32c784
35259a3a800ca4488c4efe383eaa1e8761179d96
32490 F20101119_AABJZV katar_s_Page_151.pro
e548f041abfa19dd43176186d6798386
38a3bbb6bb7a30f245db47372b887c31114b040a
89452 F20101119_AABJDC katar_s_Page_082.jp2
aefa7f3fba51cb9f9cc898ac090e5152
44c0cf87049772a2fd90fdf7175a8f6d59e548da
F20101119_AABJCN katar_s_Page_023.tif
ce178d70381f9f72dcfc3d635f2091a1
dcd29eaf3073f5920bf205a395306f2e1dcb3dc7
20433 F20101119_AABKGF katar_s_Page_135.QC.jpg
de6090e43649556aa9ef3b8cb8eebd42
11c24339dd866e410c2687adc77cd6a93a277590
2330 F20101119_AABKFQ katar_s_Page_153.txt
1a4e2075a28cb5450216bc22aa6329fc
389fdf75d8c34f84b78891e150f18034dc4e275c
41000 F20101119_AABJZW katar_s_Page_152.pro
99f809b2efcc015adfdca7eda55965a6
dc1619b3924ceb63c413de65eeb5f2b5e78eaf2b
1833 F20101119_AABJDD katar_s_Page_001thm.jpg
13ccaf1ca08f7281ac59a28c96fa24d4
404e4e8d16b0cd19c7bf3c79af60e8a5f8a789aa
6505 F20101119_AABJCO katar_s_Page_016thm.jpg
d0cbeb569bebe7e3b2a6b620443c6fdb
ad51507a01013ff8ff30ef0e21847abe8708a65c
22129 F20101119_AABKGG katar_s_Page_123.QC.jpg
6e08680334a5e92d3d2003565f37006b
4431260dd0d7cf6c5f5a3764f394231a757dd2e2
2140 F20101119_AABKFR katar_s_Page_154.txt
e04b3be059f3e2e4fee9f063eb18401c
680f934f9f5f09ee9c4c43817350b92201701f34
54107 F20101119_AABJZX katar_s_Page_154.pro
35419cc19a80de1532ed2bd4ca2f984d
32ae771a075b3d978b4ce20aea6c2803755f53ac
3948 F20101119_AABJDE katar_s_Page_145thm.jpg
3f0c846ac3e0399c70d8024605b1bf55
a89bc87fb3f7a80045f765228170e4a0c6b46508
21028 F20101119_AABJCP katar_s_Page_106.QC.jpg
18366856010c366a83332434de1b6b3c
619dbcdd7e80efec2b14dcd48b68a6a9c1cc9cb8
25526 F20101119_AABKGH katar_s_Page_087.QC.jpg
b5776a63cb25ef1c35d6958e8a6b901d
99189e144b329b29cf4f20fec85a139b101840e7
1945 F20101119_AABKFS katar_s_Page_155.txt
7357501608f14cd32000519dce64b114
ba6cd6fceaef74cb6e606f75903b0930ebb790f2
49395 F20101119_AABJZY katar_s_Page_155.pro
dd49fe4224018ee7256a46e3596cd4a4
317204058db1677dac47a6501d67af25b84dc186
99185 F20101119_AABJDF katar_s_Page_097.jp2
6849bf76c3b27ca8a2cff6089d65f5f0
42d5e8611171b0a46e7a88b5e2f49617a1c9c0e0
7131 F20101119_AABJCQ katar_s_Page_069.QC.jpg
26c18f2455f55684a2aedb00985ca5b0
dd70b0faedb6ac2ec0d80f38dc76aad19a137c05
4016 F20101119_AABKGI katar_s_Page_143thm.jpg
dc122fd4147897210dd1b578c412f840
cd1476cd42b55ddac2c7f62c5660ded4b5f958d4
1570 F20101119_AABKFT katar_s_Page_156.txt
978d44369ec99ed184523d5cf43966d3
0da4ebcfd89d6539e2b17b832a4598deb440a3a6
55866 F20101119_AABJZZ katar_s_Page_157.pro
647e17eeb00a6bac8ea00fd7f8627bfa
456669afde832af4d6dbe100a9926e7dcf69a5d1
1798 F20101119_AABJDG katar_s_Page_135.txt
9cbfeb99d24b139f9961a9f918d1765a
9b338c20f299b04448e2262fcc0713de859e07b3
2274 F20101119_AABJCR katar_s_Page_140.txt
52092760a5a3aee275d5418c00054f99
6f763ffc1d5f9cac2350af08f90aaf3269533ed7
2300 F20101119_AABKFU katar_s_Page_157.txt
aa48c1aa6aae17c901bf7756e35844f5
a02758bef7d25b6a6cf9f2e0d226f60ddc25049f
80895 F20101119_AABJCS katar_s_Page_110.jpg
b1d481440ab3dce70d1c1a3297ee62bc
4393cb51ce481bdc6e7adb74475b93fb61d95485
26151 F20101119_AABKGJ katar_s_Page_059.QC.jpg
4dce11acc14fd90a0cf32012c3fe0039
8ba4afcb6c4410b8bd1c4db2ae37dddca5cd7877
2636 F20101119_AABKFV katar_s_Page_158.txt
ef1f751c95b54decd3ad93506fb6178c
5f39ffe2aba9053390241b5c66e92d824e3e7f33
22557 F20101119_AABJDH katar_s_Page_126.QC.jpg
a632460179d7fa1cd4da7eda2d7dbe8b
55d8f032ec73ce1a1cf44d4eba3a07ca4c4fe4e9
13407 F20101119_AABJCT katar_s_Page_099.QC.jpg
c5d8b610b877afeb86c041acc1780fff
99a87072d3e6c5990e4c055340bb9cc068ce1f62
20107 F20101119_AABKGK katar_s_Page_098.QC.jpg
69ba984c89c6c128c2ab7f7608973e72
37c2fb4d16469c7aa969a190903bd65391e37210
2328 F20101119_AABKFW katar_s_Page_159.txt
52de18c12da5889f783c0bdfe7275075
5f85a2b5f1f26bd9c90cdc0d5fd0a7328c7f2f49
F20101119_AABJDI katar_s_Page_047.tif
f619ae9384af27c5b3f18fc4491a26c6
12515350560787e79c3a58efcb64357e23ec74db
62354 F20101119_AABJCU katar_s_Page_079.jpg
c78df1f8f8e9238b064090a3ed7b4ce2
0c7fd431d73c8738c97cada1071eb319bf762829
5389 F20101119_AABKHA katar_s_Page_123thm.jpg
f81afb58212e31e8216ec29587aeb2f9
11bdb99882223081a177d378c032ece1b68c5667
25702 F20101119_AABKGL katar_s_Page_017.QC.jpg
8a2601c4b49ec034d79e774e813e2288
fa487803e57a6d0240f5397e3ed88fb4d701158f
2591 F20101119_AABKFX katar_s_Page_160.txt
1adb17c8505d35018bfa877d1bfd5758
6795042e9cef4d4781d7e04611d60a4b6684e7d0
37997 F20101119_AABJDJ katar_s_Page_156.pro
4c9dbcaa9c9d8999bad27e2d86c79c89
18e5b6462aefff4d66f4ca1d9811f17cf19e0968
31950 F20101119_AABJCV katar_s_Page_101.pro
7c1b1aea68d6a1dea28804ebcedbaa92
7fd88665380958a3c6c7bd86fcf28a40f19fdd57
4525 F20101119_AABKHB katar_s_Page_103thm.jpg
b39198d15b7c0b8d4e07b53df5a07975
0d731729ea4b32dacf9ba8ba27fee215950ff6cf
25797 F20101119_AABKGM katar_s_Page_056.QC.jpg
26544955e22055bd0a25939cd734ca89
67bb0df0cdfffb9977609860ae15461a99f872e4
385 F20101119_AABKFY katar_s_Page_161.txt
b09623a7573789627a12c185309df373
bae6dcb7d9499c0e9ff340dbe9c35881ef4df730
F20101119_AABJDK katar_s_Page_055.tif
600e041a5283f53e027b77f1faee5843
b0a97487b2c745b39f7d1f1bb2ea3ef4a6fa3fc7
56297 F20101119_AABJCW katar_s_Page_101.jpg
c55af9050347448d4b45cf397cc8a01a
5810f5f60dfad27d58055742673c369c94480e26
4268 F20101119_AABKHC katar_s_Page_079thm.jpg
58717d5d15b008ca6ed258c02067a294
35c8e259d2fa7f279b9800df7e814a12862eb8bc
17842 F20101119_AABKGN katar_s_Page_061.QC.jpg
dfa3e20c22feded8a72d60e8264e1b1b
729050b9fd9c69c642b96c972f82dd697fdb147f
641 F20101119_AABKFZ katar_s_Page_162.txt
cfe4c673a1dc63ba17145aaec2caa02e
2e03a3556ba391d9b358f6d90a393e9aeffb150d
75004 F20101119_AABJEA katar_s_Page_015.jpg
903e358da668154e7bb45951340c47be
d280fc0927e2d4a7c72eeb0a70862f0a8bf67e12
1937 F20101119_AABJDL katar_s_Page_037.txt
cb9b3100ee6441c3c7473263a4e7e9d5
32d0214550094a1f8fe72a3b5c5c637f084ac72a
78805 F20101119_AABJCX katar_s_Page_038.jpg
59525fbaee8e1d5335bb2bbd3d866353
152ce47d77e56f78cb9743096d82f9289930d0fb
24459 F20101119_AABKHD katar_s_Page_038.QC.jpg
e98941f1e1ca632e41b7f2d869cb18e0
a86f6a071b9b0adc008da2d02d74a7ada5d5a4e3
24971 F20101119_AABKGO katar_s_Page_157.QC.jpg
45c03136febcb2a013a16ee561813455
f293e3d19006420a6922d4baf3c94f0123c62a50
85221 F20101119_AABJEB katar_s_Page_016.jpg
cbad42beffbf38c4c4c2d1b8f1c356f6
d0c75cca3b8c1a75ee7aed6e8d794e9b579b0677
185976 F20101119_AABJDM UFE0013881_00001.mets
a4b9403591fd18fa9791664d3450d6a9
30125371a6ec828094de3a22e2f3adceb26ecdf1
58713 F20101119_AABJCY katar_s_Page_153.pro
fd8eaa8e02e8634190999059efdf8e08
24e2653c1647c7cafcb6c9e91eb3aa3e935191d4
24075 F20101119_AABKHE katar_s_Page_119.QC.jpg
10e6dc9e8f2ec429815016a9962d837e
014d78ea83ad7c2cffc7a36c3d20b176250fa02b
5643 F20101119_AABKGP katar_s_Page_126thm.jpg
88c0d94bc30841e984cbda16dd71da15
0fef017fe4c8ecc23d09c4af63b4bf0b717d745b
82847 F20101119_AABJEC katar_s_Page_017.jpg
f847ddfe4842f95cab88df5944ee4bc9
ce75bbbbc93c6b54e68bbfc5a3e8cbd7a263f742
105364 F20101119_AABJCZ katar_s_Page_111.jp2
0b4d22dba0605bc7ac22e4d4bce2d2eb
260ef53eb553b3b4b8e79a55bd9197ab39aea937
3917 F20101119_AABKHF katar_s_Page_156thm.jpg
e52f2687e3fb19e84298e5f489e9bd2a
ff7edd60722444d09664c89e815c9ac07b38443f
15547 F20101119_AABKGQ katar_s_Page_143.QC.jpg
bc7cc8f0094d35f3f871b15bc2023124
1f8b37fe1b8c487642b3b3708f3fffb97b53be4c
83020 F20101119_AABJED katar_s_Page_018.jpg
d20eb14a7e9b6c1dfa99fcd27942466d
7eb2eb0a4fa669a9656a50fba802a62c9921df40
5191 F20101119_AABKHG katar_s_Page_035thm.jpg
7047cf29c0f15843e08094e7f45f7659
eb0a10ee2b32907345fdbbcaf21a4e309518a5f4
5069 F20101119_AABKGR katar_s_Page_090thm.jpg
e24a34d43a4a8abb7a786ea5b839d3ca
1612933380997a13d0dfd65cfd261e6349ae8321
82409 F20101119_AABJEE katar_s_Page_019.jpg
3c7530aa72fd7d15a76a25b740cd9b67
190cb7c4c13be766d33c02e4b3d2dc7237f23c91
21920 F20101119_AABJDP katar_s_Page_001.jpg
cbc68c94b9c1f4bdde7098f80a5eb035
7e3c5f7ba8e4a13ebc1ec333644f51cafcdb6794
28133 F20101119_AABKHH katar_s_Page_139.QC.jpg
16d4163e583683391b2de60b9bf76ead
28fba232472941b53cbff614d23c282b3c7e97ac
5952 F20101119_AABKGS katar_s_Page_034thm.jpg
447c19b17ec4f320435c7e9c37c8a1d4
3b40d1897a7d14baca267644f7cb12d833320acc
25537 F20101119_AABJEF katar_s_Page_020.jpg
7304e63d77760db14ac2c7268cddd6b0
e2b4fc77f9a4fda52521af75d0abfebfa85f7a68
4110 F20101119_AABJDQ katar_s_Page_002.jpg
b657fa28af600d16c1ff6e15da617990
55c5e51009014e129cb6a24d3c2f5bbc7ac4b7ab
5579 F20101119_AABKHI katar_s_Page_114thm.jpg
a15ae4dd52f6643107621a4a11454f42
060c937ed09d093fdf0de4d3716bc78c937507bf
23380 F20101119_AABKGT katar_s_Page_036.QC.jpg
ed97785133e64c2873c81ff58c79adfa
c3b646781033b38883372164ee69eda0c32bdcf5
40260 F20101119_AABJDR katar_s_Page_004.jpg
479d4bdcd942e42be959d20f3c9663f9
5bddf7d56d8db19263671fa41912421dd8ec283d
73599 F20101119_AABJEG katar_s_Page_021.jpg
56f83e3a524239c234930155029763e2
4a0f1cfcec3a3b5196a882bfaf179a987d4d92ac
5768 F20101119_AABKHJ katar_s_Page_157thm.jpg
a437041388c636620bb66869a24fd1d2
388198b2b13d9a292be0d61c9ff249f04c3fa6c7
F20101119_AABKGU katar_s_Page_019thm.jpg
fedb33b37e2b53618340a2ebd4dd7bc8
0ba5084698d812b486b57e04a480745280d39ad2
106986 F20101119_AABJDS katar_s_Page_006.jpg
3a7b37d17906b18db4a42c343d6ae75c
d7ed89bfbee35e6e29e3fc28ab328b4218265513
90537 F20101119_AABJEH katar_s_Page_022.jpg
3f10fcab0cd66830e35d6107bae31b27
2edbf83d8c2297c3c2da8d7b311353aa698f5aa1
25377 F20101119_AABKGV katar_s_Page_048.QC.jpg
16ef8199e5a3b2ac783f527f9024fe9a
4f070eff8f032fea13cd715e940df091863506cb
19134 F20101119_AABJDT katar_s_Page_008.jpg
fa8d63fc214da538e6283ebdf2f5d311
953bd68dd709186a1bd266a6a5c078b8a9848709
4373 F20101119_AABKHK katar_s_Page_115thm.jpg
60a9227b7335c1e958a20b95ea782c9b
8b23aff1761f5733e82795c3ed35af6a4fea1368
6155 F20101119_AABKGW katar_s_Page_023thm.jpg
22bda835d57f664c019f69cf412996fc
348b81cc8471b0d007bb9bf75647f3f6a3b25835
70938 F20101119_AABJDU katar_s_Page_009.jpg
6853908b915a25c71fafa86fe5b456d8
f7b6eb77244e1e3f9d0a2ad0e36ed689d6b7d2e2
79031 F20101119_AABJEI katar_s_Page_023.jpg
a33e41109b55a9bc8317dcb025a9e43a
ea2237a7f240159613452fbd7d2bf31f32d0b8c8



PAGE 1

QUALITY OF SERVICE DESIGN ISSUES IN MULTIMEDIA COMMUNICATIONS OVER POWER LINE NETWORKS By SRINIVAS KATAR A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2006

PAGE 2

Copyright 2006 by Srinivas Katar

PAGE 3

I dedicate this work to my parents, Katari Lakshmipathi Raju and Krishna Kumari

PAGE 4

ACKNOWLEDGMENTS I would like to thank my parents for their support and encouragement over all these years. None of this work would have possible without them. I would also like to express my heartfelt gratitude to my wife Keerti for her support and understanding throughout this endeavor. I would especially like to thank Larry Yonge and Intellon Corporation for allowing me to work part time on my PhD. I would like to thank my colleagues, Brent Mashburn, Kaywan Afkhamie and Manjunath Krishnam for their valuable feedback. Working with them has always been a pleasure. Finally, I would like to thank my advisors, Dr Haniph Latchman and Dr Richard Newman, for their valuable guidance throughout my studies. I would also thank Dr Oscar Crisalle and Dr Tan Wong for being part of my PhD supervisory committee. iv

PAGE 5

TABLE OF CONTENTS page ACKNOWLEDGMENTS .................................................................................................iv LIST OF TABLES ...........................................................................................................viii LIST OF FIGURES ...........................................................................................................ix ABSTRACT .....................................................................................................................xiii CHAPTER 1 INTRODUCTION...................................................................................................1 2 BACKGROUND.....................................................................................................7 2.1 Low Bandwidth PLC...................................................................................9 2.1.1 Regulatory constraints.....................................................................9 2.1.2 Technologies..................................................................................11 2.1.2.1 X-10...................................................................................11 2.1.2.2 CEBus................................................................................11 2.1.2.3 LonWorks..........................................................................12 2.2 High Bandwidth PLC.................................................................................12 2.2.1 Regulatory constraints...................................................................13 2.2.2 Channel characteristics..................................................................13 2.2.2.1 Attenuation characteristics.................................................13 2.2.2.2 Noise characteristics..........................................................15 2.2.3 Technologies..................................................................................18 3 HOMEPLUG 1.0 TECHNOLOGY DESCRIPTION AND PERFORMANCE EVALUATION......................................................................................................19 3.1 Home Networking Solutions......................................................................20 3.1.1 New wire solution..........................................................................20 3.1.2 Wireless solution............................................................................21 3.1.3 No new wire solutions...................................................................22 3.2 HomePlug 1.0 Technology........................................................................23 3.2.1 HomePlug Physical layer...............................................................23 3.2.2 HomePlug 1.0 MAC......................................................................25 3.2.2.1 Frame formats....................................................................25 v

PAGE 6

3.2.2.2 Channel access mechanism................................................26 3.2.2.3 Segmentation and reassembly mechanism.........................30 3.3 HomePlug 1.0 Performance based on Simulations....................................31 3.4 HomePlug 1.0 Field Test Results and Comparison with IEEE 802.11b/a.32 3.4.1 Experiment method........................................................................33 3.4.2 Results 34 3.4.2.1 IEEE 802.11a indoor performance....................................34 3.4.2.2 IEEE 802.11b and HomePlug 1.0......................................37 4 POWER LINE MULTIMEDIA NETWORKING.................................................41 5 CHANNEL ADAPTATION BASED ON CYCLO-STATIONARY NOISE CHARACTERISTICS...........................................................................................44 5.1 Introduction................................................................................................45 5.2 Channel Measurements..............................................................................46 5.3 Channel Adaptation...................................................................................48 5.4 Computation of Channel Capacity.............................................................49 5.5 Discussion of Results.................................................................................51 5.6 Conclusions................................................................................................53 6 MAC FRAMING...................................................................................................56 6.1 MAC Framing and Framing Efficiency.....................................................57 6.2 MAC Framing Strategies...........................................................................58 6.2.1 Simple concatenation.....................................................................59 6.2.2 Concatenation with explicit demarcation.......................................60 6.2.3 2-Level framing.............................................................................61 6.3 Efficiency Analysis....................................................................................63 6.3.1 Simple concatenation.....................................................................63 6.3.2 Concatenation with explicit demarcation.......................................64 6.3.3 2-level framing...............................................................................64 6.4 Comparison of Performance between various Framing Schemes.............68 6.5 Conclusions................................................................................................69 7 ALLOCATION REQUIREMENTS FOR SUPPORTING LATENCY BOUND TRAFFIC...............................................................................................................70 7.1 Introduction................................................................................................71 7.2 Allocation Requirements without Transmission Overheads......................76 7.2.1 Fixed Allocation, L = 1..................................................................76 7.2.2 Fixed Allocation, L > 1..................................................................77 7.3 Allocation Requirements with Transmission Overheads...........................80 7.3.1 Fixed Allocation.............................................................................81 7.3.2 Average Allocation........................................................................86 7.4 Effect of Transmission overheads on Allocation Requirements................89 7.5 Conclusions................................................................................................92 vi

PAGE 7

8 BEACON SCHEDULE PERSISTENCE TO MITIGATE BEACON LOSS.......94 8.1 Introduction................................................................................................94 8.2 Persistent Beacon Schedules......................................................................97 8.3 Efficiency Analysis..................................................................................100 8.3.1 Non-persistent schedule...............................................................100 8.3.2 Persistent current schedules.........................................................100 8.3.3 Persistent preview schedules........................................................101 8.3.4 Persistent current and preview schedules....................................101 8.3.5 Efficiency comparison.................................................................102 8.4 QoS Considerations.................................................................................103 8.5 Lower Bound on PLEP............................................................................104 8.5.1 Non-persistent schedules.............................................................105 8.5.2 Persistent current schedules.........................................................105 8.5.3 Persistent preview schedules........................................................106 8.5.4 Persistent current and preview schedules....................................108 8.5.5 Lower Bound comparison............................................................109 8.6 Markov Chain Analyses...........................................................................109 8.6.1 Non-persistent beacon schedules.................................................110 8.6.2 Persistent current schedules.........................................................111 8.6.3 Persistent preview schedules........................................................114 8.6.4 Persistent current and preview schedules....................................118 8.7 Comparison of Persistent Scheduling Mechanism..................................123 8.8 Conclusions..............................................................................................136 9 CONCLUSIONS AND FUTURE WORK..........................................................138 LIST OF REFERENCES.................................................................................................143 BIOGRAPHICAL SKETCH...........................................................................................148 vii

PAGE 8

LIST OF TABLES Table page 3-1 Frame Control information fields.............................................................................28 3-2 HomePlug 1.0 simulation results.............................................................................32 4-1 Bandwidth and QoS requirements for multimedia applications..............................41 6-1 MAC framing overheads..........................................................................................62 viii

PAGE 9

LIST OF FIGURES Figure page 1-1 FCC and CENELEC frequency band allocation......................................................10 2-1 Power line channel impulse response.......................................................................14 2-2 Channel frequency response.....................................................................................15 2-3 Noise generated by light dimmer.............................................................................16 2-4 Noise generated by yard light...................................................................................16 2-5 Noise generated by halogen lamp............................................................................17 2-6 Noise generated by hair dryer..................................................................................17 3-1 Long Frame format...................................................................................................27 3-2 Short Frame format..................................................................................................27 3-3 Priority resolution and backoff scheme....................................................................29 3-4 IEEE 802.11a indoor performance...........................................................................35 3-5 HomePlug 1.0 and IEEE 802.11b indoor performance............................................36 5-1 Channel measurement setup.....................................................................................47 5-2 Performance gain in PHY data rates as a function of macroslot size......................54 5-3 Performance gain in MAC data rates as a function of macroslot size.....................54 5-4 Average gain in MAC and PHY data rates..............................................................55 6-1 MAC framing Process..............................................................................................58 6-2 Simple concatenation...............................................................................................60 6-3 Concatenation with explicit demarcation.................................................................61 6-4 2-Level MAC framing..............................................................................................62 6-5 Simple concatenationefficiency with varying FEC block rrror rate and N = 25...65 ix

PAGE 10

6-6 Simple concatenation efficiency with varying number of FEC blocks and p = 0.1.65 6-7 Explicit demarcation based concatenation efficiency with varying FEC block error rate and N = 25................................................................................................66 6-8 Explicit demarcation based concatenation efficiency with varying number of FEC blocks and p = 0.1............................................................................................66 6-9 2-Level framing efficiency with varying FEC block error rate and N = 25..........67 6-10 2-Level framing efficiency with varying number of FEC blocks and p = 0.1.......67 6-11 Performance comparison of various MAC framing strategies with varying FEC block error rate, N= 25, L msdu = 1518.......................................................................69 7-1 Transmission overheads in HomePlug AV..............................................................74 7-2 Illustration of delayed retransmission strategy.........................................................74 7-3 State transition diagram for no MPDU overhead case.............................................78 7-4 Excess allocation as a function of FEC block error probability when no transmission overhead present.................................................................................79 7-5 Average packet loss per packet loss event when no transmission overheads are present......................................................................................................................80 7-6 Transmission opportunities required to achieve PLEP = 10 -5 at various FEC block error probabilities...........................................................................................82 7-7 Fixed Allocation estimation error for N = 100 and L = 1, 2, 3, 4, 6, and 8.............84 7-8 Probability distribution function of the Fixed Allocation estimation error..............85 7-9 Average packet loss per packet loss event when transmission overheads are present......................................................................................................................86 7-10 Average Allocation estimation error for N = 100 and L = 1, 2, 3, 4, 6 and 8..........88 7-11 Probability distribution function of the Average Allocation estimation error.........89 7-12 Percentage difference in Fixed Allocation and Optimal Allocation for N = [50, 200], p= [0.01, 0.10].................................................................................................91 7-13 Percentage difference in Average Allocation and Optimal Allocation for N = [50, 200], p= [0.01, 0.10].........................................................................................91 7-14 Effective transmission overhead for Average Allocations with N = 50 and 200.....92 x

PAGE 11

8-1 Example of 2-persistent current schedules...............................................................99 8-2 Example of 2-persistence preview schedules...........................................................99 8-3 Example of 2-persistent current and preview schedules........................................100 8-4 Efficiency loss comparison for various scheduling strategies................................102 8-5 Illustration of relationship between latency and persistent preview schedule.......107 8-6 State transition diagram for non-persistence beacon schedules.............................111 8-7 State transition diagram for 2-persistence current schedules.................................113 8-8 Illustration of state changes in Markov chain for 2-persistent preview schedule..115 8-9 State transition diagram for 2-persistence preview schedules...............................116 8-10 State transition diagram for 2-persistence current and preview schedules............120 8-11 Minimum latency required to achieve PLEP of {10 -5 10 -7 } at various beacon loss probabilities in a system with non-persistent schedules.................................126 8-12 Minimum latency required to achieve PLEP of {10 -5 10 -7 } at various beacon loss probabilities in a system with r-persistent current schedules..........................127 8-13 Minimum latency required to achieve PLEP of {10 -5 10 -7 } at various beacon loss probabilities in a system with r-persistent preview schedules........................127 8-14 Minimum latency required to achieve PLEP of {10 -5 10 -7 } at various beacon loss probabilities in a system with r-persistent current and preview schedules.....128 8-15 Comparison of minimum latency supported by various scheduling strategies at various beacon loss probabilities for PLEP = 10 -5 .................................................128 8-16 Comparison of minimum latency supported by various scheduling strategies at various beacon loss probabilities for PLEP = 10 -7 .................................................129 8-17 Minimum schedule persistence required to support L = 2 for various scheduling strategies at PLEP = 10 -5 ........................................................................................129 8-18 Minimum schedule persistence required to support L = 2 for various scheduling strategies at PLEP = 10 -7 ........................................................................................130 8-19 Over allocation required to support various latencies in a system with non-persistent beacon schedules, N = 100, PLEP = 10 -7 ...............................................130 8-20 Over allocation required to support various latencies in a system with 2-persistent current schedules, N = 100, r = 8,PLEP = 10 -7 ......................................131 xi

PAGE 12

8-21 Over allocation required to support various latencies in a system with 2-persistent preview schedules, N = 100, r = 4,PLEP = 10 -7 .....................................131 8-22 Over allocation required to support various latencies in a system with 2-persistent current and preview schedules, N = 100, r = 4,PLEP = 10 -7 ..................132 8-23 Latencies required to support a stream with optimal over allocation for a system with non-persistent schedules for PLEP = {10 -5 10 -7 }..........................................132 8-24 Latencies required to support a stream with optimal over allocation for a system with r-persistent current Schedules for PLEP = {10 -5 10 -7 }..................................133 8-25 Latencies required to support a stream with optimal over allocation for a system with r-persistent preview schedules for PLEP = {10 -5 10 -7 }.................................133 8-26 Latencies required to support a stream with optimal over allocation for a system with r-persistent preview schedules for PLEP = {10 -5 10 -7 }.................................134 8-27 Comparison of optimal latency supported by various scheduling strategies at various beacon loss probabilities for N = 100, PLEP = 10 -5 ..................................134 8-28 Comparison of optimal latency supported by various scheduling strategies at various beacon loss probabilities for N = 100, PLEP = 10 -7 ..................................135 8-29 Minimum schedule persistence required to support L = 2 with X = 1 for various scheduling strategies at PLEP = 10 -5 ......................................................................135 8-30 Minimum schedule persistence required to support L = 2 with X = 1 for various scheduling strategies at PLEP = 10 -7 ......................................................................136 xii

PAGE 13

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 QUALITY OF SERVICE DESIGN ISSUES IN MULTIMEDIA COMMUNICATIONS OVER POWER LINE NETWORKS By Srinivas Katar May 2006 Chair: Haniph A. Latchman Cochair: Richard E. Newman Major Department: Electrical and Computer Engineering Home Networking is one of the major hurdles in enabling ubiquitous data and multimedia distribution. Most houses are not equipped with specialized wiring for networking purposes and retrofitting them with new wiring is prohibitively expensive. Hence, the use of existing in-home power line infrastructure for networking purposes can go a long way towards solving the in-home connectivity problem. In this dissertation we first investigate the viability of power line communications (PLC) by evaluating the performance of HomePlug 1.0. Detailed event based simulation models were used to show that HomePlug 1.0 can provide a maximum UDP and TCP data rates of 7.86Mbps and 5.90Mbps. Extensive field testing in 20 houses showed that HomePlug 1.0 provides an average of 2.3Mbps higher throughput compared to IEEE 802.11b. While HomePlug 1.0 is good enough for data networking, it will not be able to support the newly emerging multimedia applications like audio and video streaming. We xiii

PAGE 14

study three issues that the new generation PLC systems need to overcome to enable multimedia communications: (a) AC line cycle variation in noise, (b) Impulse noise, and (c) Beacon Loss. We use channel characterization measurements on 72 channels to investigate the performance enhancements that can be achieved by using a channel adaptation mechanism that is synchronized to the AC line cycle. The results show that a 30% improvement in physical layer (PHY) data rates can be obtained by continuously adapting to the AC line cycle. A time slot based adaptation that is more suitable for practical systems provided an average of 10% improvement in MAC data rates. To effectively overcome impulse noise, we propose a novel 2-level MAC framing mechanism that enables efficient retransmissions. Our simulations and analysis show that this mechanism provides linear degradation in performance even under high FEC Block errors. Further investigation of the effect of transmission overheads on efficiency showed that low data rate, low latency QoS applications can incur significant loss in efficiency under high FEC block errors. Hence, PLC channel adaptation should take into account the application QoS requirements to optimize the overall system capacity. Due to the unreliable nature of the power line channels, beacon reception cannot always be guaranteed. We investigate various persistent scheduling approaches to mitigate beacon loss. Our analysis shows that using persistent current and preview schedules, or persistent preview schedules, up to 10% beacon loss can be tolerated with minimal impact on MAC efficiency and QoS. xiv

PAGE 15

CHAPTER 1 INTRODUCTION The explosive growth of the Internet is driving the need for ubiquitous data and multimedia communications in the twenty-first century. Major hurdles in realizing this objective are the last mile (access) and last hundred feet (in-home) connectivity problems. Approximately 50% of all network investments are spent on providing the last mile infrastructure. The high expense incurred in laying new wiring for broadband access to individual customers has forced service providers to use the existing cable and telephone lines. As much of the world still awaits broadband access, there is a need for alternative viable approaches to solving the last mile problem. In addition in-home distribution of multimedia content is still a challenge particularly for homes that are not equipped with proper wiring to support high-speed data and multimedia communications. Retrofitting the house with new wiring is prohibitively expensive and hence the need arises for new LAN technologies that enable affordable connectivity within the home. With more than a century of ongoing deployments throughout the world, electrical power lines represent by far the most pervasive network of wiring in the world. The use of power lines as a communications medium can come a long way toward solving the last mile and last hundred feet problems. Several attempts were made starting from the early 1920s to develop a reliable power line communication (PLC) technology. However, the lack of sophisticated digital signal processing technologies hampered earlier PLC technologies from gaining widespread acceptance. For several decades, their applicability was limited to operation 1

PAGE 16

2 and maintenance of the power supply grid. The explosive growth of the Internet in the recent decades has refueled the interest in developing high bandwidth (i.e., operating in 1-30 MHz frequency band) PLC technologies for in-home networking as well as for broadband access. However, the challenges to enabling high-speed communication over the power line are many. Power lines were originally devised for transmission of power at 50 Hz and at most 400 Hz. At high frequencies the power line is very hostile for signal propagation [1, 2]. Power line networks operate on standard in-building electrical wiring and as such consist of a variety of conductor types and cross-sections joined almost at random. Therefore a wide variety of characteristic impedances will be encountered in the network. This impedance mismatch causes a multi-path effect resulting in deep notches at certain frequencies. While a typical channel may present an average attenuation of approximately 40 dB, it is not uncommon for portions of the bands to experience greater than 60 dB of attenuation. Power line networks are also affected by interference. Electric appliances with brush motors, switching power supplies and halogen lamps produce impulse noise that can reduce the reliability of communication signals. Due to high attenuation over the power line, the noise is also location dependent. Power line channels also show strong dependence on the underlying AC line cycle. All these factors make the design of PLC technologies extremely challenging. In this dissertation we take a practical approach to understanding the issues in enabling high bandwidth communication over the power lines and how the PLC technologies based on HomePlug 1.0 and HomePlug AV standards overcome these challenges. A combination of analysis, simulations and field tests are used to determine

PAGE 17

3 the effectiveness of the various approaches used by these standards. The organization of this dissertation is as follows. Chapter 2 builds the background on power line communications. This chapter introduces the two major categories of PLC devices, low-bandwidth PLC and high-bandwidth PLC, and describes the various regulations constraints under which PLC devices should operate. Technology choices currently available for low and high bandwidth PLC systems are presented. This chapter also includes a detailed description of the power line channel characteristics at high frequencies (i.e., 1-30 MHz). The first generation of high-bandwidth PLC technology, HomePlug 1.0, is investigated in Chapter 3. A detailed description of HomePlug 1.0 standard was followed by simulation based evaluation of its performance. Simulation results showed that HomePlug 1.0 provides a maximum throughput of 7.86 Mbps and 5.90 Mbps for UDP and TCP respectively. The MAC saturation throughput showed graceful degradation with increase in number of station in the network. Simulations with Voice over IP traffic showed that the priority contention mechanism in HomePlug 1.0 could guarantee low latencies for delivery delay sensitive traffic. Extensive field tests were also conducted to evaluate the performance of HomePlug 1.0 with competing home networking solutions, IEEE 802.11b and IEEE 802.11a. These results show that the first generation PLC technologies provide a ubiquitous and reliable home networking solution. The HomePlug 1.0 standard was designed to transfer data with a maximum MAC throughput of 8 Mbps. While this is good enough for file sharing and Internet access, it will not be able to support newly emerging multimedia applications like audio and video streaming. Multimedia applications require not only significantly larger bandwidths, but

PAGE 18

4 also require guarantees on QoS parameters like latency, jitter and packet loss probability. To address this new breed of applications, a new generation of HomePlug Standard, HomePlug AV, was released by the HomePlug Powerline Alliance. In Chapter 4 we provide a brief overview of the QoS requirements for various applications multimedia applications and the salient features of the HomePlug AV standard that enables it to support these applications. To achieve data rates necessary to support multimedia applications, HomePlug AV uses a channel adaptation mechanism that is synchronized to the underlying AC line cycle. In chapter 5, we use channel characterization measurements and analysis to investigate the performance enhancements that can be received by using AC line cycle based channel adaptation. These results show that 30% improvement in the physical layer data rates can be obtained by continuously adapting to the underlying AC line cycle. Since continuous adaptation with respect to time is unduly complex for practical systems, we propose dividing the AC line cycle into multiple regions and adapting independently in each region. Results show that using a 1-2 milliseconds region size will provide optimal MAC performance on most paths tested. Impulse noise sources (like dimmers, halogen lamps, etc.,) are very common in power line networks. Impulse noise will produce temporal effects in channel capacity. Between two impulse noise events, the channel capacity can be very high, while during the impulse noise the capacity is very low. To obtain low FEC block error rates, the channel has to be adapted close to the lower end of its capacity (i.e., capacity near impulse noise event) and hence is undesirable. Physical layer adapting to the higher end of the channel capacity hinges on the ability to efficiently retransmit corrupted portions

PAGE 19

5 of the data at the MAC layer. Chapter 6 presents a novel 2-level MAC framing mechanism that enables efficient re-transmission of corrupt data. Both analytical and simulation results are used to compare the performance of the 2-level MAC framing with other viable approaches. These results show that 2-level MAC framing provides linear degradation in performance even under high error conditions and it outperforms other approaches. Apart from the MAC framing overheads, each transmission will also incur overheads due to physical layer headers, acknowledgments and interframe spacings. The impact of these transmission overheads on the TDMA allocation requirements and efficiency of multimedia streams is investigated in Chapter 7. Results show that the allocation has to be increased by one or more transmission overhead to support low latency applications. Further the increase the increase in allocation requirements is not highly dependent on the application data rate. Hence, low data rate applications can incur significant loss in efficiency. These results show that channel adaptation should take into consideration the application data rate to improve the overall capacity of the system. HomePlug AV uses a beacon based Time Division Multiple Access (TDMA) technique for providing guaranteed bandwidth to multimedia streams. Beacons in HomePlug AV carry TDMA allocation (or scheduling) information for various active multimedia streams. Since a station that missed a beacon will not know the schedule for that beacon period, it will not be able to transmit. Due to the large attenuation and impulse noise, beacon detection cannot always be guaranteed. To overcome occasional beacon loss, HomePlug AV uses persistent beacon schedules. Persistent schedules are valid for multiple beacon periods, thus enabling a station to transmit even if it misses one

PAGE 20

6 or more beacons. In Chapter 8, we use a Markov chain based analysis to investigate the impact of beacon loss on systems with and without persistent beacon schedule. Results show that by using persistent current and preview schedule or persistent preview schedules, HomePlug AV systems can tolerate up to 10% beacon loss with minimum impact of MAC efficiency and QoS guarantees. A brief summary of the major contributions of this dissertation and the scope for future work are presented in Chapter 9.

PAGE 21

CHAPTER 2 BACKGROUND Power line communications stands for the use of power supply grid for communication purpose. Power line network has very extensive infrastructure in nearly each building. Because of that fact the use of this network for transmission of data in addition to power supply has gained a lot of attention. Since power lines are devised for transmission of power at 50-60 Hz and at most 400 Hz, the use this medium for data transmission, at high frequencies, presents some technically challenging problems. Besides large attenuation, power line is one of the most electrically contaminated environments, which makes communication extremely difficult. Furthermore, the restrictions imposed on the use of various frequency bands in the power line spectrum limit the achievable data rates. Power lines connect the power generation station to a variety of customers dispersed over a wide region. Power transmission is done using varying voltage levels and power line cables. Power line cable characteristics and the number of crossovers play an important role in determining the kind of communication technology that needs to be used. Based on the voltage levels that are used, power lines can be categorized as follows: 1. High-tension lines: These connect electricity generation stations to distribution stations. The voltage levels on these lines are typically in the order of hundreds of kilovolts and they run over distances of the order of tens of kilometers. 2. Medium-tension lines: These connect the distribution stations to pole mounted transformers. The voltage levels are of the order of a few kilo volts and they run over distances of the order of a few kilometers. 7

PAGE 22

8 3. Low-tension lines: These connect pole-mounted transformers to individual households. The voltage levels on these lines are of the order of a few hundred volts and these run over distances of the order of a few hundred meters. High-tension lines represent excellent carriers for RF energy as we only find open wire equipment with very few crossovers. A transmission power of about 10 watts is often sufficient to overcome distances of more than 500 kilometers. Around the year 1922 the first carrier frequency system (CFS) began to operate on high-tension lines in the frequency range of 15-1500 KHz [3, 4, 5]. During the past and even nowadays the main purpose of CFS was to maintain the operability of the power supply. While in former times speech transmission was dominant, today we have more and more digital data communications due to the rapid progress of overall automation. Through the application of modern digital modulation and coding schemes, a significant enhancement of bandwidth efficiency could be achieved for CFS. Mediumand low-tension lines are characterized by a large number of cross connections and different conductor types (e.g., open wire and cable). Long distance RF signal propagation is extremely bad in this environment because of high attenuation and impedance matching problems. Around the year 1930 ripple carrier signaling (RCS) began to operate on these lines. These used a frequency range below 3 KHz down to 125 Hz with amplitude shift keying (ASK) modulation technique. The data rate achieved by RCS is in the order of a few bits per second. Load management and automatic reconfiguration of power distribution networks were among the most important tasks performed by RCS. We see that historically the use of power line communications was mainly for use by the utility corporations (UCs) in maintaining the seamless power supply. The UCs generally regarded the power distribution wiring as a natural medium for their

PAGE 23

9 communication needs, as all important stations are connected. The lack of sophisticated signal processing techniques at affordable cost points limited the applicability of PLC technologies in the past. 2.1 Low Bandwidth PLC In the mid 90s, data communications over low-tension lines regained a lot of attention for enabling smart houses. A smart house can be defined as a building equipped with numerous sensors and actuators, where for example heating, air-conditioning and illumination can be automatically and remotely controlled and supervised. Furthermore safety systems such as burglar or fire alarms may be included [4, 6, 7]. New generations of low bandwidth PLC systems were designed for addressing this market. 2.1.1 Regulatory constraints Frequency bands used by these devices are restricted by the limitations imposed by the regulatory agencies [8]. These regulations are developed to ensure harmonious coexistence of various electromagnetic devices in the same environment. The frequency restrictions imposed in two of the main markets, North America and Europe, are shown in Figure 1-1. The Federal Communications Commission (FCC) and the European Committee for Electro technical Standardization (CENELEC) govern regulator rules in North America and Europe respectively. In North America frequency band from 0 to 500 KHz can be used for power line communications. However the regulatory rules in Europe are more stringent. The spectrum is divided into five bands based on the regulations. They are 4. Frequency Band from 3 9 KHz: The use of this frequency band is limited to energy provides; however, with their approval it may also be used by other parties inside consumers premises.

PAGE 24

10 5. Frequency Band from 9 95 KHz: The use of this frequency band is limited to the energy providers and their concession-holders. This frequency band is often referred as the A-Band 6. Frequency Band from 95 125 KHz: The use of this frequency band is limited to the energy providers costumers; no access protocol is defined for this frequency band. This frequency band is often referred as the B-Band 7. Frequency Band from 125 140 KHz: The use of this frequency band is limited to the energy providers customers; in order to make simultaneous operation of several systems within this frequency band possible, a carrier sense multiple access protocol using center frequency of 132.5 KHz was defined. This frequency band is often referred to as the C-Band 8. Frequency Band from 140 148.5 KHz: The use of this frequency band is limited to the energy providers customers; no access-protocol is defined for this frequency band. This frequency band is often referred to as the D-Band. FCC General UseFrequency Band FCC ProhibitedFrequency Band0540 CENELECProhibited FrequencyBand A-Band D-Band C-Band B-Band39125140148.595(a) FCC Frequency Band Allocation for North AmericaKHzKHz(b) CENELEC Frequency Band Allocation for Europe Figure 1-1: FCC and CENELEC frequency band allocation Thus in Europe power line communications are restricted to operating in the frequency range from 95 148.5 KHz. Apart from band allocation, regulatory bodies also impose limits on the radiations that are emitted by these devices. These reflect as restrictions on the transmitted power in each of these frequency bands.

PAGE 25

11 2.1.2 Technologies Various protocols have been developed for use by low bandwidth digital devices for communication on power line. Each of these protocols different in the modulation technique, channel access mechanism and the frequency band they use. Various products based on these protocols are available in the market and are mainly used for home automation purposes. A brief overview of these protocols is presented here. 2.1.2.1 X-10 The X-10 technology is one of the oldest power line communication protocol. It uses a form of amplitude shift keying (ASK) technique for transmission of information. Although it was originally unidirectional (from controller to controlled modules) recently some bi-directional products are being implemented. X-10 controllers send their signals over the power line to simple receivers that are used mainly to control lightning and other appliances. Some controllers available today implement gateways between the power line and other medium such as RF and infrared. A 120 KHz amplitude modulated carrier, 0.5 watt signal, is superimposed into AC power line at zero crossings to minimize the noise interference. Information is coded by way of bursts of high frequency signals. To increase the reliability each bit of information is transmitted separately, which limits the transmission rate to 60 bits per second. This represents poor bandwidth utilization while the reliability of transmission is severely compromised in a noisy environment. These are the main reasons why this technology has limited applications. 2.1.2.2 CEBus The CEBus protocol uses peer-to-peer communication model. To avoid collisions a carrier sensed multiple access with collision resolution and collision detection

PAGE 26

12 (CSMA/CRCD) is used. The power line physical layer of the CEBus communication protocol is based on spread spectrum technology patented by Intellon Corporation. Unlike traditional spread spectrum techniques (that used in frequency hopping or time hopping or direct sequence), the CEBus power line carrier sweeps through a range of frequencies as it is transmitted. A single sweep covers the frequency band from 100-400 KHz. This frequency sweep is called a chirp. Chirps are used for synchronization, collision resolution and data transmission. Using this chirp technology data rate of about 10 KHz can be obtained. The frequency used by this technology restricts its use in only North American market. 2.1.2.3 LonWorks LonWorks is a technology developed by Echelon Corporation and provides a peer-to-peer communication protocol, implemented using Carrier Sensed Multiple Access (CSMA) technique. Unlike CEBus, LonWorks is a narrowband spread spectrum modulation technique using the frequency band from 125 KHz to140 KHz. It uses multi-bit correlator intended to preserve data in the presence of noise with a patented impulse noise cancellation. An advantage of the narrowband signaling is that it can be used in both North American and European markets. 2.2 High Bandwidth PLC High-speed communication over low-tension power lines has recently gained lot of attention. This is fueled by the unparalleled growth of the Internet, which has created accelerating demand for digital telecommunications. High bandwidth PLC devices are designed to exploit this market. More specifically, these devices use the existing power line infrastructure within the apartment, office or school building for providing a local area network (LAN) to interconnect various digital devices. Some of the applications

PAGE 27

13 include high-speed Internet access, multimedia, smart appliances/remote control, home automation and security; data back up, telecommunications, entertainment and IP-telephony. 2.2.1 Regulatory constraints High bandwidth PLC devices for communication on power line use the frequency band between 1 MHz and 30 MHz. In contrast to low bandwidth digital devices, the regulatory standards for this region of the spectrum are still being developed. Currently, United State is the only country where the regulations for High Bandwidth PLC systems are well defined. In United States, PLC systems operate under FCC Part 15 rules using the frequency band between 1.8-30 MHz. Several sub-bands within this range have to be notched to prevent interference with licensed services. 2.2.2 Channel characteristics Power lines were originally devised for transmission of power at 50-60 Hz and at most 400 Hz. At high frequencies power line is very hostile for signal propagation [1, 2]. In the section that follows gives a brief overview of power line channel characteristics in the frequency band between 1 MHz and 30 MHz. 2.2.2.1 Attenuation characteristics High frequency signals can be injected on to the power line by using an appropriately designed high pass filter. Received signal power will be maximum when the impedance of the transmitter, power line and the receiver are matched. Dedicated communication channels like Ethernet have known impedance, thus impedance matching is not a problem. However, power line networks are usually made of a variety of conductor types and cross sections joined almost at random. Therefore a wide variety of characteristic impedances will be encountered in the network. Further, the network

PAGE 28

14 terminal impedance will tend to vary both at communication signal frequencies and with time as the consumer premises load pattern varies. The multipath effect resulting from impedance mismatch and the attenuation caused by the electric wiring makes power line channels highly frequency selective [9]. While a typical channel may present an average attenuation of approximately 40 dB, it is not uncommon for portions of the bands to experience greater than 60 dB of attenuation. Similarly, while most power line channels have a significant delay spread of 1 to 2 s, it is not uncommon for some channels to exhibit a delay spread larger than 5 s. Figure 2-1 and Figure 2-2 show a sample power line channel impulse and frequency response. One of the unique characteristics of power line channels is the dependence of channel characteristics on the AC line cycle. Electric appliances may turn on and off, and/or draw electric power as a function of the AC line cycle. While this may change both the channels frequency response and noise profile, it is more common to see changes in only the noise profile. Figure 2-1: Power line channel impulse response

PAGE 29

15 Figure 2-2: Channel frequency response 2.2.2.2 Noise characteristics The major sources of noise on the power line are from electrical appliances, which generate noise components that extend well into the high frequency spectrum. Worse, there are often impulse noise sources. Impulse noise can be categorized as periodic or continuous based on when it occurs relative to the underlying AC line cycle. Figure 2-3, 2-4 and 2-5 respectively show periodic impulse noise from light dimmer, yard light and halogen lamp in addition to the data signal (rectangular shape). Devices that contain brush motors (Figure 2-6) also produce impulse noise. Induced radio frequency signals from broadcast, commercial, military, citizen band and amateur stations also impair certain frequency bands. Because much of the noise experienced by each node may be highly localized due to attenuation, the noise profile seen by each PLC device may be significantly different. Therefore, power line channels are not typically symmetric.

PAGE 30

16 Figure 2-3: Noise generated by light dimmer Figure 2-4: Noise generated by yard light

PAGE 31

17 Figure 2-5: Noise generated by halogen lamp Figure 2-6: Noise generated by hair dryer

PAGE 32

18 2.2.3 Technologies The first generation of high bandwidth PLC systems were developed to enable high-speed data communications in a Small Office Home Office (SOHO) environments. These technologies represented a gain leap in exploiting the power lines as a true communication medium. The most widely deployed first generation PLC systems are based on HomePlug 1.0 Standard. HomePlug 1.0 was developed by HomePlug Powerline Alliance; a non-profit industry association that was formed in March of 2000 by a group of industry leading companies to enable standards based power line networking products. In the next chapter, we take a detailed look at the technology behind HomePlug 1.0 and how it enabled it to overcome the harsh power line environment. The performance of this technology against other popular SOHO networking options was also presented. As home networks become more and more common, there is increasing demand for enabling multimedia communications within the home. Multimedia applications not only require high bandwidth but also need guarantees on various QoS elements like delay, jitter and packet loss probability. A second generation of HomePlug Standards, HomePlug AV was developed to address this need. In Chapters 5, 6, 7 and 8, we investigate some of the design choices made in HomePlug AV and evaluate their performance.

PAGE 33

CHAPTER 3 HOMEPLUG 1.0 TECHNOLOGY DESCRIPTION AND PERFORMANCE EVALUATION Affordable broadband Internet communication to residential customers is now available via Cable Modems and various flavors of Digital Subscriber Lines (DSL). In turn there is a growing need for in-home networks to share this single full-time Internet access link, along with network connectivity for various computers and printers in the home. A good home networking solution should be easy to setup, zero maintenance, cost effective and should also be able to provide reliable performance throughout the house With multiple outlets in every room, residential power lines are already the most pervasive networks in the home. Using this existing infrastructure to provide high speed networking capabilities provides several benefits. First of all, there is no need for expensive rewiring of the house. Secondly, almost all devices that need to be networked are already connected to the AC wiring. Thus, home networking becomes as simple as plugging the device in the AC outlet. HomePlug 1.0 Standard was developed to enable ubiquitous, easy to use and affordable home networking. In this chapter we present detailed description of HomePlug 1.0 technology and evaluate its performance based on simulations and field tests. A more thorough treatment of the material can be found in the following papers we published [10, 11, 12]. The organization of this section is as follows. In Section 3.1 we provide a brief overview of various options for home networking and their pros and cons. Details of HomePlug 1.0 technology and simulation based evaluation of its performance is 19

PAGE 34

20 presented in Section 3.2 and Section 3.3, respectively. Field test based performance comparison of HomePlug 1.0 and IEEE 802.11b/a are presented in Section 3.3. 3.1 Home Networking Solutions A variety of technological alternatives are available to home networking. These technologies can be grouped into three different categories based on the medium they use for transfer of data. They are: 1. New Wire 2. Wireless 3. No New Wires 3.1.1 New wire solution New Wired solutions use specialized cable to connect various devices. There are the most matured technologies and have found wide spread acceptance in enterprise, campus and industrial networks. Some examples include Ethernet, Token Bus and Token ring. Ethernet is one of the most widely used networking solution inside the home these days. The major drawback is that it would require retrofitting the house with these specialized cables. This is quite an expensive proposition as it could involve drilling holes in the walls. Further, most of the electricians have little or no knowledge on how to install these cables. For example, a simple mistake of improper termination of Ethernet wire will make the network useless. Another drawback of Ethernet is the lack of proper QoS provisions that are very critical for any networking technology to be successful in the home networking market. The recently developed IEEE 1394b [13] specification also targets the home networking market. This is an extension of the popular IEEE 1394 [14] (also known as FireWire or i.Link) that supports cord lengths up to 100 meters. Using Cat 5 UTP cords, it can support data rate of 100 Mbps over 100 meters distance. Using 50m

PAGE 35

21 fiber, data rates of up to 3200 Mbps can be supported for cord lengths of up to 100 meters. 3.1.2 Wireless solution Wireless technologies use high frequency radio waves to transfer data. The most predominant technologies include, 1. IEEE 802.11b/a/g, 2. BlueTooth, 3. HomeRF, and 4. Ultrawide band. The main benefit of using wireless networks is the freedom to move around the house while maintaining network connection. IEEE 802.11x [15, 16, 17, 18] are the most popular wireless home networking technology. IEEE 802.11b uses Direct Sequence Spread Spectrum (DSSS) in the 2.4GHz Industrial, Scientific and Medical (ISM) band and provides PHY rates of up to 11 Mbps. The extensions of this technology include IEEE 802.11g and IEEE 802.11a. IEEE 802.11g uses the same frequency bands as IEEE 802.11b, but with OFDM modulation and provides PHY rate of up to 54 Mbps. IEEE 802.11a uses the 5 GHz Unlicensed National Information Infrastructure (U-NII) bands along with OFDM modulation to provide up to 54 Mbps data rates. The main draw back of this technology has been coverage and quality of service. BlueTooth technology [19, 20] uses the 2.4GHz ISM band with frequency hopped spread spectrum technology. This technology should be treated more like a Personal Area Networking solution rather than a Home Networking solution due to lower transmit power required by the kind of applications they serve.

PAGE 36

22 HomeRF also uses the 2.4GHz ISM band with frequency hopped spread spectrum technology. This technology had limited acceptance in the market. The only advantage of this when compared to IEEE 802.11b is a better support for VoIP traffic. The use of Ultrawide Band (UWB) technology [21, 22] was recently approved by the FCC in February 2002. UWB is defined as any radio technology having a spectrum that occupies a bandwidth greater than 20 percent of the center frequency, or a bandwidth of at least 500 MHz. There are two competing UWB standards. 1. Multi-Band OFDM (MBOA) UWB standard developed by WiMedia Alliance, and 2. Direct Sequence Code Division Multiple Access (DS-CDMA) UWB standard developed UWB Forum UWB technologies can enable data rates of up to 480Mbps. As with Bluetooth, Ultrawide band signals have limited range and are more suitable for Personal Area Networks. 3.1.3 No new wire solutions No New wire solutions use existing wiring inside the house to provide high-speed communication capabilities. These technologies include 5. HomePlug 1.0, HomePlug AV 6. HomePNA, and 7. MoCA HomePlug 1.0 and HomePlug AV technologies uses the existing power line wiring inside the house and provide maximum PHY data rates of 14Mbps and 150Mbps, respectively. The ubiquity of power outlets is the greatest advantage when compared to other wired technologies. HomePNA uses existing phone line as an infrastructure to provide home networking. The HomePNA 3.0 standard [23] released in 2003 enables data rates up to

PAGE 37

23 128 Mbps with optional extensions to 240 Mbps. It also has deterministic QoS, but suffers from a limited number of available outlets in the house. Multimedia over Coax (MoCA) [24] uses existing coax wiring to provide home networking. To avoid interference with video channels, MoCA uses 850MHz to 1500MHz frequency band. Similar to using existing telephone wiring, cable outlets are typically limited to 3 or 4 in the average home and are certainly not present in all rooms. Use of splitters may also limit bi-directional transmission. As we have seen in this section, there are several technologies that have targeted the home networking technologies. Most of the technologies have gained limited acceptance due to coverage or ubiquity problems. HomePlug and IEEE 802.11x are the most important contenders at present due to their ubiquity. 3.2 HomePlug 1.0 Technology HomePlug 1.0 technology [11, 12, 25, 26] overcome the harsh channel conditions on the power lines by using an adaptive approach that uses robust transmission technique combined with sophisticated forward error correction (FEC) [27, 28], error detection, data interleaving, and automatic repeat request (ARQ) [29]. 3.2.1 HomePlug Physical layer Orthogonal Frequency Division Multiplexing (OFDM) is the basic transmission technique used by the HomePlug. OFDM is well known in the literature and in industry [30]. It is currently used in DSL technology [31], terrestrial wireless distribution of television signals, and has also been adapted for IEEE's high rate wireless LAN Standards (802.11a and 802.11g). The basic idea of OFDM is to divide the available spectrum into several narrowband, low data rate sub-carriers. To obtain high spectral efficiency the frequency response of the sub-carriers are overlapping and orthogonal,

PAGE 38

24 hence the name OFDM. Each narrowband sub-carrier can be modulated using various modulation formats. By choosing the sub-carrier spacing to be small the channel transfer function reduces to a simple constant within the bandwidth of each sub-carrier. In this way, a frequency selective channel is divided into many flat-fading sub-channels, which eliminates the need for sophisticated equalizers. The OFDM used by HomePlug is specially tailored for power line environments. It uses 84 equally spaced sub-carriers in the frequency band between 4.5MHz and 21MHz. Cyclic prefix and differential modulation techniques are used to completely eliminate the need for any equalization. Impulsive noise events are overcome by means of forward error correction and data interleaving. HomePlug payload uses a concatenation of Viterbi and Reed-Solomon FEC. Sensitive frame control data is encoded using turbo product codes. The power line channel between any two links has a different amplitude and phase response. Furthermore, noise on the power line is local to the receiver. HomePlug technology optimizes the data rate on each link by using channel adaptation. Channel adaptation is HomePlug 1.0 involves Tone Allocation, modulation selection and FEC selection. Tone allocation is the process by which certain heavily impaired carriers are turned off. This significantly reduces the bit error rates and helps in targeting the power of FEC and Modulation choices on the good carriers. HomePlug allows for choosing from differential binary phase shift keying (DBPSK), differential quadrature phase shift keying (DQPSK) on all the carriers, with either or FEC code rates. The end result of this adaptation is a highly optimized link throughput.

PAGE 39

25 Certain types of information, such as broadcast packets, cannot make use of channel adaptation techniques. HomePlug uses an innovative modulation called ROBO, so that information is reliably transmitted. ROBO modulation uses a DBPSK with heavy error correction with bit repetition in time and frequency to enable highly reliable communication. ROBO frames are also used for channel adaptation. 3.2.2 HomePlug 1.0 MAC The choice of Medium Access Control (MAC) protocol provides a different set of challenges. Home networks should be able to support a diverse set of applications ranging from simple file transfer to QoS demanding applications such as Voice-over-IP (VoIP) and low data rate streaming media. The HomePlug MAC is built to seamlessly integrate with the physical layer and addresses these needs. HomePlug MAC is modeled to work with IEEE 802.3 frame formats. This choice simplifies the integration with the widely deployed Ethernet. HomePlug MAC appends the Ethernet frames with encryption and other management before transmitting it over the power line. A segmentation and reassembly mechanism is used to in cases where the complete packet cannot be fit in a single frame. 3.2.2.1 Frame formats HomePlug technology uses two basic frame formats (refer Figure 5-1). A Long Frame consists of a Start of Frame (SOF) delimiter, Payload and End of Frame delimiter (EOF). A Short Frame consists of a Response Delimiter and is used as part of the Stop-and Wait automatic repeat request (ARQ) process. ARQ mechanism causes retransmission of corrupt packets, thus reducing the packet error rate. All the delimiters share a common structure. A delimiter consists of a Preamble and Frame Control information field. The Preamble is a form of spread spectrum signal that is

PAGE 40

26 used to determine the start of a delimiter. This is followed by Frame Control information, which is encoded using a robust Turbo Product Code and can be detected reliably even at several dB below the noise floor. Among other things, delimiters convey timing information that is used by MAC to determine the availability of the medium. The robust design of the delimiter helps the nodes to obtain a very high level of synchronization, thus reducing unintended collisions. The details of the various field contained in the Frame Control are given in Table 3-1. The Payload of the Long Frame delimiter is encoded based on the channel adaptation. The first 17 bytes of the payload contain the Frame Header. This field contains the source address, destination address and segmentation information. HomePlug technology limits the maximum length of the payload field in the Long Frame to 160 OFDM Symbols (~1.3 msec). This manifests as better guarantees of QoS as the delay incurred by higher priority traffic due to on going lower priority transmission is reduced. If the packet cannot be fitted into a Long Frame, a segmentation and reassembly mechanism is used to send it in multiple Long Frames. The frame header contains information that is used by the receiver to properly reconstruct the segmented packet. The Payload is protected by a Frame Check Sequence (FCS) to detect uncorrected errors. 3.2.2.2 Channel access mechanism The channel access mechanism used by the HomePlug MAC is a variant of the well-known CSMA/CA protocol. A typical CSMA/CA protocol would require nodes to sense the medium (this is the carrier sensing part) for other traffic. If the medium is busy, nodes will defer from transmitting until the medium becomes idle. When the medium becomes idle, nodes will wait for a randomly chosen duration (this is the collision avoidance part). A node will transmit only if it detects no other traffic on the medium

PAGE 41

27 during this randomly chosen duration. HomePlug channel access scheme builds upon this mechanism by providing prioritized access along with high network utilization. The overall protocol includes a carrier sensing mechanism, a priority resolution mechanism and a backoff algorithm. Preamble Frame Control Frame Header Frame Body FCS Preamble Frame Control 25 bitsvariable bytecount2 bytes25 bits4 symbolsvariablesymbol count4 symbols17 bytes Start of FrameDelimiter Payload20-160 OFDM Symbols End of FrameDelimiterEFG B-PAD variablebytecount Figure 3-1: Long Frame format Preamble Frame Control 25 bits4 symbols Response DelimiterA Figure 3-2: Short Frame format The carrier sense mechanism helps HomePlug nodes to synchronize with each other. At the heart of this mechanism are the delimiters. HomePlug technology uses a combination of Physical Carrier Sense (PCS) and Virtual Carrier Sense (VCS) to determine the state of the medium (i.e., if the medium is idle or busy and for how long).

PAGE 42

28 PCS is provided by the HomePlug PHY and basically indicates whether a preamble signal is detected on the medium. VCS is maintained by the HomePlug MAC layer and is updated based on the information contained in the delimiter (refer Table 1). Delimiters contain information not only on the duration of current transmission but also on which priority traffic can contend for the medium after this transmission. PCS and VCS information is maintained by the MAC to determine the exact state of the medium. Table 3-1: Frame Control information fields Delimiter Type Fields Meaning Type This can be SOF with response expected or an SOF with no response expected depending on whether a Short Frame delimiter is expected at the end of this Long Frame Contention Control When set to 1, this prevents all HomePlug nodes with packets of priority level equal to or less than the current Long Frame's priority from accessing the channel. However, higher priority nodes can still interrupt this transmission Frame Length This indicates the length of the payload in multiples of OFDM symbol blocks Start of Frame (SOF) Tone Map Index This is an index to the channel adaptation information stored at the receiver. Note that the variable length Payload is encoded using the maximum transfer rates that can be achieved by the link. Type This can be EOF with response expected or an EOF with no response expected depending on whether a Short Frame delimiter is expected at the end of this Long Frame. Contention Control The information conveyed is same as that conveyed by this field in SOF delimiter. This redundancy helps in better synchronization. End of Frame (EOF) Channel Access Priority (CAP) This field indicates the priority of the current Long Frame. Type This can be ACK (positive acknowledgment), NACK, (negative acknowledgment indicating faulty reception) or FAIL (negative acknowledgment indicating lack of resources) Response (Resp) Channel Access Priority (CAP) This field indicates the priority of the preceding Long Frame

PAGE 43

29 The Priority resolution mechanism provides prioritized access of the medium in a highly distributed manner. Due to the distributed nature of this mechanism, there is no need for a central node to coordinate access over the medium as in the case of some networking technologies. HomePlug allow up to four different priority levels. At the heart of this priority resolution mechanism are the priority resolution slots (PRS) and the priority resolution signals. Priority resolution signals use a form of spread spectrum signal that has a high tolerance to delay spread and is also very robust. High tolerance to delay spread prevents destructive interference when multiple nodes assert in the same priority resolution slot. After the end of every transmission, two slots are allocated for priority resolution. Figure 3-3 shows the occurrence of priority resolution signals with respect to the end of previous transmission and the subsequent contention period. After priority resolution slots, contention will be only between nodes that have the highest available priority in the network. For example, when all four priorities are present in the network, Priority 3 and Priority 2 nodes will transmit priority resolution signal in PRS0. Priority 1 and Priority 0 nodes will detect this signal, causing them to defer to the higher priority traffic. Priority 3 nodes will transmit a priority resolution signal in PRS1, which will be detected by the Priority 2 nodes, causing them to defer. Thus only Priority 3 nodes will contend in the Contention Period. Long Frame PRS0 PRS1 Contention PeriodPriority ResolutionSlots Backoff Slots ShortFrame Previous Transmission Figure 3-3: Priority resolution and backoff scheme The backoff algorithm used by HomePlug MAC is designed to provide high network utilization (which manifests itself as high network throughput) even under

PAGE 44

30 heavily loaded conditions. It is also tailored to seamlessly integrate with the priority level construction and the applications that each of these priorities is expected to support. Note that only nodes that have the highest priority available in the network will contend during the Contention Period and all other nodes will defer. From a HomePlug view of the power line network, all traffic with priority a less than the maximum available traffic priority in the network is pretty much ignored. Lower priority traffic will have to wait till all the higher priority traffic is transmitted. As with other CSMA/CA algorithms, the backoff slot is chosen to be a random integer between 0 and the Contention Window Size. The growth of the contention window under HomePlug is controlled by the estimation of traffic on the network as well as on the priority of the traffic. This helps HomePlug nodes achieve higher network utilization and also control the latency for higher priority traffic. 3.2.2.3 Segmentation and reassembly mechanism HomePlug uses an adaptive approach to maximize the throughput on each link. Furthermore, the maximum size of the long frame is limited to 160 OFDM symbols for the payload. For cases in which the complete Ethernet packet cannot be fitted in a single long frame, segmentation and reassembly mechanism is used. The necessary information for proper reassembly at the receiver is included in the segment control field of the frame header. Multiple segments can be transmitted in a single burst to obtain high throughputs. However, each segment is required to go through priority resolution mechanism. This ensures that a segment burst can be interrupted by higher priority traffic, thus reducing the latency for higher priority traffic. HomePlug allows extension of this segment bursting across multiple Ethernet packets to provide contention free access.

PAGE 45

31 3.3 HomePlug 1.0 Performance based on Simulations In this section we report the measurements observed using an event-based C program to simulate a HomePlug 1.0 power line network. All scenarios assume QPSK and a 3/4 coding rate on various links. In this simulation, we use UDP, TCP, and VOIP traffic. Both TCP and UDP traffic sources are configured to generate 1460 byte packet at priority 0. TCP and UDP traffic is generated with exponential inter-arrival time with 100 microsecond average. This will ensure that the network will remain saturated. VOIP traffic is modeled as isochronous with a 20 msec interarrival time. The packet size of VOIP is 160 bytes and is assigned the highest priority, 3. In Table 3-2, we provide the of simulation results of a power line network. The UDP traffic simulation scenario 1 shows the best throughput in our simulations since there is no contention at all. Table 3-2 also shows channel contention with 2 and 3 UDP nodes causes a modest reduction in channel throughput. In the TCP traffic simulation, though scenario 1 has only one traffic source, the bandwidth must be shared with data and response frames (e.g., TCP ACK packets) thus it provides lower performance than the UDP traffic simulation. The MAC throughput represents the total number of transmitted bytes of Ethernet payload transmitted by the MAC divided by the simulation time. The UDP throughput represents the total number of transmitted bytes of UDP payload divided by the simulation time. Similarly, TCP throughout represents the total number of transmitted bytes of TCP payload divided by the simulation time. The third metric we provide in Table 3-2 is the PLC simulation results of one VOIP and multiple UDP connections. The high priority VOIP always wins the contention and the UDP nodes can send packets only when there is no VOIP traffic. In this simulation, the VoIP Delay indicates the average network delay (i.e., queuing delay and channel

PAGE 46

32 access delay) incurred by the VoIP packets. These results show that the priority contention mechanism in HomePlug 1.0 can guarantee very low latency for delay sensitive traffic. The Table 3-2 also shows the simulation results of one VOIP and multiple TCP connections. The throughput of VOIP is small; the total throughput is dominated by the TCP component. The slight increase in VoIP delay is due to the increase in channel access time due to higher number of contending source (i.e., contention between TCP Data and TCP ACK). Table 3-2: HomePlug 1.0 simulation results Simulation Results Throughput of multiple UDP Traffic Streams Scenario 1 (1 UDP) Scenario 2 (2 UDP) Scenario 3 (3 UDP) MAC Throughput 8.08 Mbps 7.46 Mbps 7.46 Mbps UDP Throughput 7.86 Mbps 7.26 Mbps 7.26 Mbps Throughput of multiple TCP Traffic Streams Scenario 1 (1 TCP) Scenario 2 (2 TCP) Scenario 3 (3 TCP) MAC Throughput 6.16 Mbps 6.15 Mbps 6.12 Mbps TCP Throughput 5.90 Mbps 5.89 Mbps 5.87 Mbps Throughput of one VoIP and multiple UDP traffic streams Scenario 1 1 VoIP, 1 UDP Scenario 2 1 VoIP, 2 UDP Scenario 3 1 VoIP, 3 UDP MAC Throughput 7.89 Mbps 7.33 Mbps 7.29 Mbps UDP Throughput 7.53 Mbps 6.99 Mbps 6.96 Mbps VoIP Delay 2.75 msec 3.00 msec 3.00 msec Throughput of one VoIP and multiple TCP traffic streams Scenario 1 1 VoIP, 1 TCP Scenario 2 1 VoIP, 2 TCP Scenario 3 1 VoIP, 3 TCP MAC Throughput 6.04 Mbps 5.85 Mbps 5.77 Mbps TCP Throughput 5.65 Mbps 5.47 Mbps 5.39 Mbps VoIP Delay 3.25 msec 3.25 msec 3.25 msec 3.4 HomePlug 1.0 Field Test Results and Comparison with IEEE 802.11b/a To understand the real-world performance of HomePlug 1.0 and to compare it with a popular home networking alternative, IEEE 802.11a/b, we conducted field tests in 20 houses located in the Gainesville, Ocala, Orlando, and Belleview areas of Florida. The

PAGE 47

33 choices of the houses used in the tests were in the medium to large range (1500 ft 2 ), since larger houses provide a better range on the performance parameters of interest. The equipment used in this test included the following. AP server: A Sony notebook with a 700 MHz Pentium III processor and 128 Mbytes RAM running Windows2000 Mobile Station: An HP notebook with a 500MHz Pentium III processor and 128 Mbytes RAM running Windows2000 Linksys HomePlug 1.0-based Powerline-to-Ethernet bridges Netgear IEEE 802.11b Access Point and PCMCIA Card D-Link DWL-5000AP IEEE 802.11a Access Point and D-Link DWL-A650 PCMCIA card For PLC testing the two laptops were connected through the power line via Powerline-to-Ethernet bridges. For wireless testing, a modified infrastructure mode (MIM) was used. The AP server was connected to an access point using an Ethernet crossover cable to the built-in Ethernet socket. A PCMCIA slot in the mobile station was used to connect the wireless card. Note that typical wireless networks use an infrastructure mode (IM). In this mode, all wireless nodes communicate with each other through the access point, and must share the bandwidth over two hops. Furthermore, since these tests had the AP server connected to the access point via Ethernet, there was no other contention possible, so the test results should represent the best case scenarios in this regard. 3.4.1 Experiment method The TCP throughput and distances were measured for various locations of AP and mobile stations inside the house. The AP server was located close to a phone or cable outlet, the most probable locations for the home network to be connected to the

PAGE 48

34 broadband access network. The mobile station was located at various places where it would be likely to find other networked devices in the home. The AP Server antenna and Mobile Station antenna were placed randomly to minimize the effect of directional antenna gain. We argue that this is the typical antenna placement since ordinary users probably dont know how to set antenna directions to maximize throughput. Besides, not all locations are susceptible to antenna direction adjustment due to the surrounding environment. WSTTCP, a popular TTCP implementation ported to Windows sockets, was used. The TCP buffer size was chosen to be 11,680 bytes (1460 8). The number of TCP buffers transmitted was chosen such that each test ran for approximately 60 seconds. A single run of WSTTCP involved starting the WSTTCP in receive mode at the receiver on a selected port. WSTTCP was then started at the transmitter with a specific TCP buffer size, number of TCP buffers to be transferred, receiver IP address, and receiver port number. At the end of transmission, WSTTCP (at both the transmitter and receiver) provided the throughput observed on the link. For several of these tests, real-time packet capture was also obtained to observe TCP stability. All the procedures were automated and required minimal human operation. 3.4.2 Results The amount of data collected is too large to be presented, so only a summary of the most interesting findings is presented. 3.4.2.1 IEEE 802.11a indoor performance The performance and coverage results of IEEE 802.11a are depicted in Fig. 3-4. Figure 3-4a shows the connectivity (i.e., percentage of good links) as a function of house area. As expected, the connectivity decreased as the house area increased. Results show

PAGE 49

35 that connectivity is poor even in moderate size (2500 ft 2 ) houses. For larger houses (>4000 ft 2 ) the connectivity decreased to 20 percent. Figure 3-4b shows a scatter plot of throughput as a function of distance. It is interesting to note that IEEE 802.11a connectivity is almost zero when the distance is larger than 50 ft. Figure 3-4c shows the percentage of links that exceed the throughput values indicated on the x-axis. Note that the maximum IEEE 802.11a throughput obtained from the product being tested was larger than those expected from theory. This could be because of manufacturer-specific proprietary enhancements like the use of higher-level modulations. In summary, the statistics show that: IEEE 802.11a failed on at least one link in 19 of the 20 houses tested. 802.11a failed to connect in 45 percent of the links that were tested. 802.11a showed close to zero connectivity at distances larger than 50 ft. For shorter distances, 802.11a provided excellent throughput in most cases. (A) Figure 3-4: IEEE 802.11a indoor performance A) Connectivity versus house area B) Throughput versus distance C) Percentage of links versus throughput

PAGE 50

36 (B) (C) Figure 3-4: Continued (A) Figure 3-5: HomePlug 1.0 and IEEE 802.11b indoor performance A) Connectivity versus house area B) Throughput versus distance C) Percentage of links versus throughput (MIM) D) Percentage of links versus throughput (MIM)

PAGE 51

37 (B) (C) (D) Figure 3-5: Continued 3.4.2.2 IEEE 802.11b and HomePlug 1.0 To facilitate comparison, the performance and coverage results of IEEE 802.11b and HomePlug 1.0 are shown together in Fig. 3-5. Figure 3-5a shows connectivity as a

PAGE 52

38 function of house area. Both technologies show high connectivity for houses of less than 4000 ft 2 For houses larger than 4000 ft 2 the connectivity for IEEE 802.11b dropped dramatically to 50 percent in both of the houses tested, while HomePlug 1.0 continued to show high connectivity. Figure 3-5b shows a scatter plot of throughput as a function of distance. IEEE 802.11b typically provides close to maximum throughput at distances of less than 50 ft; for distance larger than 50 ft, the performance exhibited large variations. On the other hand, the HomePlug 1.0 system performance is not correlated with the line of sight distances measured in this experiment. This is because HomePlug 1.0 signals have to pass through the convoluted power line cable runs to reach the mobile station. Figure 3-5c shows the percentage of links that exceed the throughput value depicted on the x-axis. Our experiments showed that the overall coverage of 802.11b was 92 percent. The maximum throughput observed in field-testing was 5.13 Mb/s. Figure 3-5c shows that around 70 percent of the connections operated at more than 4 Mb/s and 10 percent above 5 Mb/s. For HomePlug 1.0, the overall coverage is 98 percent. The maximum throughput observed in testing was 5.98 Mb/s. For HomePlug 1.0, 58 percent of the connections operated above 4 Mb/s and 38 percent had throughput above 5 Mb/s. The interesting crossover phenomena displayed in the graph reflect three aspects of the systems. First, the paucity of data rates supported by 802.11b hurts its performance when channel conditions are sub optimal. Second, HomePlug 1.0s ability to adapt to the channel conditions with a nearly continuous selection of data rates allows it to perform better under mediocre channel conditions. Finally, the higher maximum data rate of HomePlug 1.0 allows it to outperform 802.11b when channel conditions are favorable.

PAGE 53

39 Note that, as mentioned earlier, in this experiment, throughput was measured between the access point and the mobile station in a MIM. However, this may not always be the way IEEE 802.11b stations communicate with each other. IEEE 802.11b networks can be configured in either ad hoc mode or IM. In ad hoc mode, wireless stations communicate with each other directly. However, typical home networks use an IM in which each wireless station communicates with the access point, which in turn forwards the data to the designated receiver. Some of the reasons for using IM include ease of setup, better coverage, and security. Furthermore, most wireless equipment is configured in IM out of the box, and must be reconfigured to ad hoc mode by the customer. Thus, in a typical IEEE 802.11b home network, all station-to-station transmissions, other than those designated to the access point itself (i.e., the access point is the final destination of the transmission) or those that originate from the access point, will be retransmitted by the access point. This reduces the effective throughput experienced between such stations. We use a simple method to extrapolate the IM throughput from the MIM link throughput data collected in the field tests. A random sample was chosen from the set of collected data and used as the throughput (R1) from a TCP source to the access point. Another random sample was chosen from the sample and was used as the throughput (R2) from the access point to the TCP destination. The aggregate IM throughput then can be obtained by assuming that a fixed packet size of x bits is transmitting through two links with speeds R1 and R2. The total time to transmit this packet will require x/R1 + x/R2 s. Thus, throughput can be calculated by R1 R2/(R1 + R2). Multiple iterations were used to obtain the distribution of the IM throughput. Although this method is not

PAGE 54

40 fully accurate, it is reasonable to expect the actual performance in IM to be close to the values obtained. Figure 3-5d shows the percentage of links that exceed the calculated IM throughput value depicted on the x-axis. These results show that HomePlug 1.0 stations provide superior coverage and throughput to IEEE 802.11b stations in IM. From the statistics, we make the following key observations: HomePlug 1.0 had a larger maximum throughput than 802.11b (about 1 Mb/s larger). On 60 percent of the links HomePlug 1.0 performed better than 802.11b links in MIM. On an average basis, HomePlug 1.0 gave approximately 0.2 Mb/s higher TCP throughput than 802.11b in MIM. In six of the 20 houses tested, IEEE 802.11b failed on at least one link. In two of the 20 houses tested, HomePlug 1.0 failed on at least one link. On an average basis, HomePlug 1.0 gave approximately 2.3 Mb/s higher TCP throughput than 802.11b would be expected to give in IM. In summary, HomePlug 1.0 was found to provide better coverage and slightly better average TCP throughput than IEEE 802.11b in MIM, which would be typical for Internet access. In IM, HomePlug 1.0 was estimated to have throughput about 2.3 Mb/s greater than 802.11b. For shorter line-of-sight distances, 802.11x performed better than HomePlug 1.0, but for longer distances, the nearly continuous adaptation capability of HomePlug 1.0 allowed it to make better use of mediocre channels.

PAGE 55

CHAPTER 4 POWER LINE MULTIMEDIA NETWORKING Current generation PLC-based home networking technologies are mainly data centric. While these are good enough for applications like web browsing and file transfer, they cannot support multimedia applications such as audio/video streaming and multi-player gaming as these applications require much higher bandwidths and QoS than those provided by current technologies [32]. Table 4-1 shows the bandwidth and QoS requirements for typical multimedia applications. For example, High Definition MPEG2 streams (HDTV streams) will require quasi-error free performance at up to 24 Mbps of application throughput, and about 300 msec of delay with less than a few microseconds of jitter. Furthermore, typical home network usage scenarios include simultaneous transfer of multiple multimedia streams. Hence, home networks should be capable of providing 50-60 Mbps application level throughput within the network. Table 4-1: Bandwidth and QoS requirements for multimedia applications Application Bandwidth (Mbps) Latency (msec) Jitter Packet Loss Probability High Definition (HD) Video Streaming 11 24 100 300 0.5 s to several msec Quasi Error Free Standard Definition (SD) Video Streaming 2 6 100 300 0.5 s to several msec Quasi Error Free DVD quality Video 6 8 100 300 0.5 s to several msec Quasi Error Free Internet Video Conferencing 0.1 2 75 100 several msec 10 -3 Home Theater Audio (Multiple streams of audio) 4 6 100 300 0.5 s to several msec Quasi Error Free Voice over IP (VoIP) < 0.064 10 30 10 30 msec 10 -2 Network Gaming < 0.1 10 30 10 30 msec Quasi Error Free 41

PAGE 56

42 1) msec => milliseconds 2) Quasi Error Free => Less than 1 packet dropped in 2 hours To support the data rates and QoS required by the new generation of multimedia applications HomePlug Powerline alliance has released a new generation of HomePlug Standards, HomePlug AV. The HomePlug AV physical layer is designed to operate as close as possible to the channel capacity of the Power line. Due to the regulatory constraints, a spectrally efficient PHY is essential to ensure that multiple multimedia streams can be supported simultaneously and delivered to the whole house. HomePlug AV PHY incorporates the following salient features. 1. Bit-loading and higher order modulations (including 8, 16, 64, 256, and 1024-QAM) to allow each carrier to take full advantage of the SNR it experiences. 2. Coherent modulation, for increased the SNR given a fixed transmit power. 3. Long symbol time to minimize guard interval overhead. 4. Forward error correction based on parallel concatenated convolutional codes, to decrease the probability of path no-connects and ensure high throughput. 5. Transmission synchronized with respect to the AC line cycle, for stable and improved throughput. 6. MAC-PHY cross-layer design for handling impulse noise. These enhancements enable HomePlug AV to achieve a maximum physical layer data rate of 150 Mbps. The exact details and rationale for the choice of various PHY parameters is beyond the scope of this work. Further details of HomePlug AV physical layer can be found in [33, 34]. The HomePlug AV Medium Access Control (MAC) layer is based on a hybrid of Time Division Multiple Access (TDMA) and Carrier Sense Multiple Access (CSMA). Each network has a designated device, the Central Coordinator (CCo), that coordinates the access to medium. The CCo periodically generates a beacon with information on

PAGE 57

43 TDMA and CSMA allocation. TDMA is used by applications requiring QoS, while CSMA is used for transport of regular data. Some of the salient features of the HomePlug AV MAC are, 1. Beacon Period Synchronized to the AC line cycle to enable AC line cycle based channel adaptation by the PHY, 2. 2-Level MAC framing to effectively deal with impulse noise, 3. Beacon Schedule Persistence to mitigate beacon loss, 4. Dynamic TDMA allocations to deal with changing channel conditions, 5. Jitter Control based on the use of Arrival Time Stamps and Clock tracking, 6. Proxy Coordination function to deal with hidden stations, 7. Neighbor network mode of operation to deal with overlapping networks, 8. Coexistence with HomePlug 1.0 devices, For further details of HomePlug AV MAC, refer to [33, 34]. In this dissertation, we investigate the first three features of the HomePlug AV MAC and evaluate their performance.

PAGE 58

CHAPTER 5 CHANNEL ADAPTATION BASED ON CYCLO-STATIONARY NOISE CHARACTERISTICS As power line communication (PLC) systems become more pervasive, there is increasing demand for new generations of PLC systems that enable bandwidth intensive applications like audio/video streaming and broadband access. To achieve the data rates necessary to support such applications within the current regulatory constraints, PLC systems should exploit the inherent characteristics of the power line channel that enable a more aggressive and optimized channel adaptation. One such characteristic is the cyclic variation of the noise power with the phase of the underlying AC line cycle. In this section, we use extensive channel characterization measurements and analysis to investigate the performance enhancements that can be achieved by using a channel adaptation mechanism that exploits the cyclic variation in power line noise characteristics. The results show that a 30% improvement in physical layer (PHY) data rates can be obtained by continuously adapting to, and synchronizing communications with, the AC line cycle. Since continuous adaptation with respect to time is unduly complex for practical systems, we propose dividing the AC line cycle into multiple regions and adapting independently in each region. Results show that using a region size of 1-2 milliseconds provides optimal MAC throughput on most paths tested. A subset of the results presented in this chapter is also published in [35]. 44

PAGE 59

45 5.1 Introduction The future evolution of PLC systems is driven primarily by an increasing demand for higher throughput. In LAN scenarios, the requirement to support multimedia applications like audio and video streaming is driving the need for higher throughput. In broadband access, the need to provide higher access speeds to a larger customer base (sharing the same medium) is causing the demand for higher throughputs. The design of PLC systems that enable this new generation of high data rate applications should exploit the unique channel characteristics observed on the power line channels to provide higher throughput within the current regulatory constraints. A typical power line network consists of a variety of conductor types joined almost at random and terminating into a variety of electrical appliances. The conductors were primarily designed for transmission of power at 50-60 Hz and at most 400 Hz. At higher frequencies they provide characteristic impedances that not only vary based on the conductor type but also as a function of frequency. Impedance mismatches in power line networks result in a multipath effect that often cause deep nulls in certain frequency bands. Electrical appliances connected to the power line also significantly impact the PLC channel characteristics. Several common electric appliances turn on/off via switching power supplies and/or draw a varying amount of electric power as function of the underlying AC line cycle voltage. During these processes noise is generated that extends well into the higher frequency bands where PLC systems operate. One of the unique characteristics of this man-made noise is its cyclic dependency on the phase of the underlying AC line cycle. The cyclic nature of noise in power line networks has been well known for several years. One of the first PLC based home automation technology, X10, exploited the low

PAGE 60

46 noise characteristics at the zero crossing of the AC line cycle to increase the system robustness. O. Ohno and M. Katayama [36, 37] were among the first to characterize the noise in (0-500Kz) frequency band as cyclo-stationary. In [38] Y. Hirayama et al., have investigated and confirmed the cyclo-stationary noise characteristics in the band from DC to 25MHz. Apart from noise, the channel response (or attenuation characteristics) can also change with the AC line cycle. F.J. Canete, J.A. Cortes, et al., [39, 40] has recently proposed a Linear Periodically Time Varying (LPTV) model for the PLC channel response. While in general both channel response and noise characteristics can vary as a function of the AC line cycle, it is more common to see significant variations in noise profile compared to that of the channel impulse response. In this section, we study the effect of the cyclic variation of power line noise on the performance of PLC systems. The organization of the remaining sections is as follows. Section 5.2 describes the test setup used to measure channel attenuation and noise characteristics. Mechanisms for channel adaptation over cyclic channels and the various tradeoffs involved are presented in Section 5.3. Processing of the field test data to compute channel capacities is presented in Section 5.4. The results obtained are discussed in Section 5.5 and conclusions are presented in Section 5.6. 5.2 Channel Measurements A total of 72 channels in two locations were characterized for use in this study. The characterization of each path involved computation of channel impulse responses and time domain noise captures spanning a single 60 Hz AC line cycle. The paths characterized were chosen at random within each location. The setup for measuring the channel impulse response is shown in Figure 5-1. An arbitrary waveform generator, functioning as a transmitter with a digital-to-analog

PAGE 61

47 converter (DAC), was loaded with a sounding waveform sampled at 75MS/s. The signal out of the DAC was amplified and coupled on to the power line. At the receiver side, the signal is coupled off of the power line medium, passed through a 2 to 30 MHz band pass filter, digitized at 75 MS/s using an analog-to-digital converter (ADC) card, and passed to Matlab for subsequent processing. Similarly, noise captures were performed by turning off the transmitter and simply digitizing a long portion of the ambient noise on the power line. The transmit waveform consisted of a repeating OFDM symbol, specifically a frequency sweeping chirp occupying approximately 2 to 30 MHz. After digitizing the waveform at the receiver, ten consecutive chirps were time domain averaged to reduce noise. Frequency domain division of the received signal by the transmitted signal (along with calibration factors to remove the gain and/or frequency response of the amplifiers, couplers, etc.) was used to estimate the channel frequency response. Figure 5-1: Channel measurement setup

PAGE 62

48 5.3 Channel Adaptation Channel adaptation in communication systems enables fine-tuning of the underlying PHY parameters to optimize the system performance. For periodically time-varying PLC channels, channel adaptation should also determine the dependency of the PHY parameters on the underlying phase of the AC line cycle. One simple means of accomplishing this is by dividing the AC line cycle into multiple regions and determining the PHY parameters independently for each such region. Note that there is an important trade-off involved in the choice of number of regions and the system performance/complexity. By increasing the number of regions, a finer adaptation to temporal periodic channel characteristics can be obtained. However, this also increases the complexity of the system. A larger number of regions requires the transmitter and receiver to determine and maintain a larger set of PHY parameters and to apply them with higher accuracy within the AC line cycle. The bandwidth and memory requirements for generation, storing and exchanging the PHY parameters can be prohibitively expensive. Apart from this, inefficiencies are also incurred at the MAC Layer due to the need to limit the transmission of each physical layer protocol unit (PPDU) to a single region. Since most practical implementations will apply a single set of PHY parameters on a single PPDU, any transmission that spans across two or more regions will have to be divided into multiple transmissions (or PPDUs) with each PPDU occupying a single region. The extra overhead incurred in each PPDU transmission, due to frame synchronization and control overheads at the beginning every PPDU, will adversely affect the overall efficiency. For the analysis in this section, we use 130ms of incurred overhead per PPDU, based on the HomePlug AV standard [33, 34].

PAGE 63

49 By decreasing the number of regions, the complexity of the channel adaptation mechanism can be reduced but the performance of the system suffers. This is due to the need to choose worst case PHY parameters within each region so that reliable communications can be guaranteed throughout the region. In the following section we investigate the effect of region size on the PHY and MAC layer data rates. 5.4 Computation of Channel Capacity To investigate the effect of cyclic variation of noise on the performance of a PLC system, we first divide the 60Hz AC line cycle into 160 microslots. Each microslot (@104s) forms the basic atomic for estimating the variation of signal-to-noise ratio (SNR) and Shannon Capacity [41] with respect to the AC line cycle. When computing SNR and Capacity, the bandwidth from DC to Nyquist (37.5 MHz) was divided into 192 subbands. This value was chosen as it provides an intermediate level of frequency domain resolution while at the same time enabling accurate noise power estimation as twenty consecutive 384-FFT outputs (5.12s per FFT) may be averaged in each microslot. This entire process is described in the following steps. First, a fixed subband transmit power, S TX was used for all subbands. This fixed power was picked to be comparable to the peak power spectral density allowed by HomePlug standards and hence compliant with FCC regulations. (Since this is merely a measurement of channel response, and not a data transmission, adherence to FCC regulations is not strictly required, but resulting SNR measurements are realistic for a practical system). Next, the received carrier signal power, S RX is measured across the frequency band of interest, and used to derive the channel attenuation, Preliminary observations show

PAGE 64

50 that the channel attenuation coefficients typically exhibit little variation throughout the AC line cycle compared to the variation of the channel noise. Therefore only one set of coefficients is used in our analysis 153 ..., ,21 ,11 ,10 ][][cforcSScRXTX Noise power in the j th microslot, N RX was then computed by averaging over twenty consecutive 384-FFTs, F, of the noise waveform as shown below, 19,...,2,1,0)383384:384(5.7812],[ iwhereiijNoiseWfmfftjiF 190201|],[|],[iRXjiFjiN Finally, the Shannon capacity was used to estimate throughput in the j th microslot, D[j], 15310],[][21925.37)1(log110153][cjcNcSRXRXjD For a PLC system that can adapt independently in each microslot, the average channel capacity can be obtained by averaging the microslot capacities over a full AC line cycle. In practice, supporting 160 channel adaptation regions is unduly complex. We study the effect of a smaller number of regions on performance by grouping a varying number of microslots to form a macroslot. For example, by grouping 20 microslots into a macroslot, only 8 (160/20) independent regions need to be supported. The average capacity of a macroslot that contains N microslots is obtained as follows, 1...0,maxNiforNiNiDDmacroslot

PAGE 65

51 Where, {D[0], D[1], D[N-1]} are the capacities of the constituent microslots arranged in ascending order. This choice can be justified as follows. In each macroslot, the PHY parameters could be chosen such that reliable trans-mission is guaranteed in all its microslots. In this case, the minimum of the capacity for all constituent microslots (i.e., D[0]) will be the capacity of a macroslot. However, for PLC system that can tolerate errors at the physical layer (for example, by using MAC layer re-transmissions), it is possible to achieve a larger aggregate throughput (in certain channel conditions) by choosing more aggressive PHY parameters. In general, PHY parameters based on, D[i] (where i>0), can be chosen. However, in such cases, transmission in i microslots will not be successful as transmission rate will exceed capacity. Hence the average capacity will be NiNiD In our analysis we chose the PHY parameters that provide the optimal PHY data rate. However, it has to be noted that maximization of the PHY data rates may not be the best choice under all conditions. For example, in PLC systems that support Quality of Service (QoS) sensitive multimedia streams, the probability of physical layer errors should also be bounded. Performance analysis for such cases is beyond the scope of this study. 5.5 Discussion of Results For each of the measured paths, we computed the average PHY capacity for various choices of macroslot sizes. Since we are interested in the performance gain obtained by AC line cycle based adaptation, we measured the gain in data rate in comparison with the case when the macroslot spans a complete AC line cycle (i.e., no AC line cycle based adaptation). Figure 5-2 shows the inverse cumulative distribution (i.e., 1-cdf) of the percentage gain in PHY data rates ob-served over the measured paths for various choices

PAGE 66

52 of macroslot size. The average gain in PHY data rates over all measured paths as a function of macroslot size is shown in Figure 5-4 (solid line). These results allow the following observations: Using a 1 millisecond long macroslot, more that 35% improvement in PHY data rates can be obtained on 30 % of the paths, Using a 1 millisecond long macroslot, more that 55% improvement in PHY data rates can be obtained on at least 15 % of the paths, Using continuous adaptation to cyclic noise (i.e., microslot based adaptation), a 30% increase in PHY data rates can be obtained (MAC overhead renders this adaptation method impractical how-ever, as seen in Figure 54), MAC layer data rates can be computed from the PHY data rates by taking into account the additional overheads incurred for transmissions that span multiple regions (or macroslots). We compute the MAC level throughput as follows, sinSizeMacroSlotssinSizeMacroSlotThroughputPHYThroughputMAC 130 Note that a 130s incurred overhead per PPDU was used for this analysis, based on the HomePlug AV standard as mentioned earlier. Figure 5-3 shows the inverse CDF of the percentage gain in MAC layer data rates observed over the measured paths for various choices of macroslot size. The average gain in MAC layer data rates over all measured paths as a function of macroslot size is shown in Figure 5-4. It is notable that MAC performance suffers when the macroslot size is chosen to be either too large or too small. In the former case, the drop in PHY data rates is due to lack of proper AC line cycle based adaptation. In the latter case, excessive PPDU overhead incurred due to the large number of regions negatively impacts the MAC performance. For the paths tested, these results can be summarized as follows,

PAGE 67

53 Using 1 millisecond long macroslots, more than 30% improvement in MAC layer data rates can be obtained on 20% of the paths, Using a macroslot that spans 1-2 milliseconds pro-vides the highest MAC performance, AC line cycle based adaptation can provide an average of 10% improvement in MAC data rates, In 40% of the paths, less than 5% improvement in MAC performance was observed. This suggests that AC line cycle based adaptation does not al-ways improve performance. The choice of using AC line cycle based adaptation should be made selectively based on the observed channel characteristics. 5.6 Conclusions In this section, we investigated the performance enhancements that can be achieved by using a channel adaptation mechanism that is synchronized to the AC line cycle. Analysis of channel characteristics on 72 channels has shown that an average of 10% improvement in MAC data rate can be achieved by using AC line based cycle adaptation. Furthermore, on 20% of the paths, we have seen more than 30% improvement in MAC layer data rate. The analysis also showed that using 1-2msec regions provides the largest gain in MAC performance for most of the channels. Apart from an increase in throughput, AC line cycle based channel adaptation also enhances the stability of channel adaptation and allows for better control of physical layer errors. Both of these features are invaluable for supporting QoS applications over power lines. The results presented in this section can be considered preliminary as they are based on 72 paths from two buildings in North America. Further research incorporating a larger number of channels, as well as taking into account both the frequency response and noise power variation with the AC line cycle, would help to fully understand the potential performance gains offered by the use of line cycle based adaptation in PLC systems.

PAGE 68

54 Figure 5-2: Performance gain in PHY data rates as a function of macroslot size Figure 5-3: Performance gain in MAC data rates as a function of macroslot size

PAGE 69

55 Figure 5-4: Average gain in MAC and PHY data rates

PAGE 70

CHAPTER 6 MAC FRAMING In power line communications (PLC) more so than in many other media, there is a strong interaction between the Medium Access Control (MAC) layer and the Physical (PHY) layer. This chapter investigates one of the manifestations of this interaction, that of MAC framing and error recovery with PHY transmission characteristics. MAC framing for high-speed PLC systems should be capable of concatenating multiple MAC Service Data Units (MSDUs, e.g., Ethernet or IP packets) in order to amortize the high and fixed overheads associated with each MAC Protocol Data Unit (MPDU). By itself, this is easy. However, transmission errors may damage portions of an MPDU, which can further rob a system of efficiency. This chapter presents a 2-level MAC framing mechanism that enables efficient re-transmission of corrupted data. Both analytical and simulation results are used to compare the performance of the proposed MAC framing scheme with other existing viable approaches. The results show that the proposed scheme is significantly superior to existing approaches. A subset of the results presented in this chapter is also published in [42]. The organization of the remaining sections is as follows. Section 6.1 gives an overview of MAC framing and provides an intuitive definition of Framing Efficiency. In Section 6.2, various approaches for MAC framing are presented. Simulation and analytical results on the performance of each of these approaches are presented in Section 6.3. Performance comparison between various MAC framing approaches and conclusions are presented in Section 6.4 and Section 6.5 respectively. 56

PAGE 71

57 6.1 MAC Framing and Framing Efficiency MAC framing deals with the process of converting Service Data Units received by the MAC from higher layer (i.e., MAC Service Data Unit, MSDU) into a Protocol Data Unit that the MAC handovers to the physical layer (i.e., MAC Protocol Data Unit, MPDU). MAC framing process not only depends on the PHY, but also on MAC features like channel access mechanism, network architecture, Layer-2 encryption etc. For example if one of the MAC features is to provide privacy, the MAC framing process will include fields for passing encryption relevant information. For a token passing MAC, the MAC framing process will contain fields to indicate busy token etc. When the underlying PHY is unreliable, MAC framing will also have to add fields for error detection and retransmissions. Thus the MAC framing process depends heavily on system details and the approach used at the MAC layer to address them. Once the MAC framing process generates an MPDU, it is handed over to the physical layer. The physical layer adds the PHY headers and transmits it to the destination (refer Figure 6-1). Apart from the physical layer overheads, MAC protocols also incur overheads due to interframe spacing, transmission of control information, collisions etc. All these overheads along with the MAC framing overheads will determine the MAC Efficiency. In this chapter, we are interested in studying the performance of the MAC framing process with respect to recovery of errors. To decouple analysis from other system level details, we assume that the only overheads incurred are those required for error recovery at the MAC layer. We define the MAC framing Efficiency to be the average number of MSDU bits delivered per MPDU divided by the total bits per MPDU. MAC framing Efficiency depends on PHY errors, MAC framing approach, MSDU size, and MPDU size.

PAGE 72

58 Figure 6-1: MAC framing Process 6.2 MAC Framing Strategies Concatenation is essential for high-speed PLC systems due to the large and fixed overheads per MPDU. These occur due to PHY overheads and interframe spacings. In Home-Plug 1.0 [11], each MPDU transmission will incur a minimum overhead of 351.02 microseconds. If a system is ported to 100 Mbps without any concatenation, the maximum efficiency (assuming 1518 byte Ethernet packet as the MSDU) that can be obtained per transmission is ~25%, even in the absence of errors. Thus, concatenation is essential. Three different packet concatenation mechanisms are investigated here. All three assume that some of the contents of an MPDU may be deliverable even if the MPDU is partially damaged. Without this stipulation, the efficiencies are so low in the presence of errors that these approaches are not considered further. In each case, we assume that the MPDU body is sent in one or more fixed size FEC blocks of a PPDU.

PAGE 73

59 Given that multiple MSDUs may be concatenated to form an MPDU, we consider several basic stages. First, the MPDU may be acknowledged as a whole either all of the constituent MSDUs are received correctly or none are, thereby requiring retransmission of the entire MPDU contents. This approach has a very low overhead, but is too inefficient when there are errors. The second main stage is to allow each MSDU to occupy its own, individually acknowledged, MAC frame. In this case, only the MAC frames that could not be recovered from a damaged MPDU must be retransmitted. We consider two variations of this approach; Simple Concatenation and Explicit Demarcation based Concatenation. Finally we consider two-level concatenation. Here, the individual FEC blocks hold their own header information, which allows retransmission of uncorrectable FEC blocks as well as resynchronization of MAC frames. 6.2.1 Simple concatenation This is a very straightforward extension of the case without concatenation. Each MSDU is prepended with a Sequence Number (SN) and a Length Field (Len). A CRC based error detection field in attached at the end of the MSDU. Sequence numbers are needed to ensure in-order delivery and the Length field enables demarcation of each packet at the receiver. Thus, multiple MSDUs can be concatenated at the transmitter and de-concatenated at the receiver. Figure 6-2 shows the MAC Frame format for simple concatenation. This approach has two main disadvantages. Firstly, each MPDU has to be padded to fit into the PPDU. Secondly, a single FEC Block error will result in loss of information contained in all subsequent FEC Blocks.

PAGE 74

60 Figure 6-2: Simple concatenation 6.2.2 Concatenation with explicit demarcation One of the main disadvantages of simple concatenation is that all information following the location of the first FEC block error is lost. This loss is mainly due to the loss of MAC Frame demarcation. So, it is quite likely that several MAC Frames might have been received without errors at the receiver but the receiver has to discard them due to loss in demarcation information. One simple approach is to add a Header Check Sequence (HCS) for detecting errors on the header fields in the MAC Frame. Thus, whenever the receiver detects a MAC Frame with errors, it can scan through the received bits to check for the next header. Once a proper header is confirmed by means of the HCS, it has the necessary demarcation information for receiving the remaining MAC Frames within the MPDU. MAC Frame header should also include an Identifier (ID) field. ID can be set to zero for the first MAC Frame in an MPDU and incremented for each subsequent MAC Frame. This enables the receiver to use a simple bit map for indicating the reception status of each MAC Frame within a MPDU. Figure 6-3 shows the MAC Frame format for Concatenation with Explicit demarcation.

PAGE 75

61 Figure 6-3: Concatenation with explicit demarcation 6.2.3 2-Level framing PHY errors occur on an FEC block basis. Hence, an optimal framing strategy is one in which only the corrupted FEC blocks need retransmission. 2-Level framing is based on this principle. Here, framing is done in two steps. In the first step each MSDU is prepended with a Length (Len) field. This entity is referred as a MAC Frame. Multiple MAC Frames that are awaiting delivery are concatenated together to form a MAC Frame Stream. MAC Frame Stream thus generated is treated as a stream of bytes and is then divided into segments of fixed size that match the FEC Block size. Each segment is prepended with a Sequence Number (SN) and a MAC Frame Boundary Offset (MFBO) field. The SN provides the relative location of the Segment within the stream while the MFBO field indicates the location of the first MAC Frame boundary (if any) within the Segment. A CRC based error detection field in attached at the end. The resulting entity is referred to as a PHY Block. Each PHY Block is mapped on to FEC Block at the PHY layer (refer Figure 6-4). It might be necessary to

PAGE 76

62 add a pad to the MAC frame stream when there are not enough bytes to transmit a Segment. Note that a pad is not necessary if there are enough pending MAC Frames. All the results presented in this section assume that the padding overhead for 2-Level Concatenation is negligible. Figure 6-4: 2-Level MAC framing Table 6-1: MAC framing overheads SN Len HCS ID CRC MFBO Simple Concatenation 2 2 4 Explicit Demarcation 2 2 2 1 4 2-Level Concatenation 2 2 4 2

PAGE 77

63 6.3 Efficiency Analysis We have derived equations for the efficiency of the Simple Concatenation and the 2-Level Framing schemes. For Explicit Demarcation based concatenation, simulation results were used to obtain the efficiency. For the results presented in this section, the overheads shown in Table 6-1 are used. We also assume that all MSDUs are of the same size. Results are presented for the case of 520-byte FEC Blocks [33, 34]. The FEC block errors in an MPDU are assumed to be independent of each other. 6.3.1 Simple concatenation In the following analysis of Simple Concatenation, we assumed that a single FEC block error would result in loss of all MAC Frames that are contained (either partially or fully) within it. Due to the loss in demarcation information of MAC Frames, all information subsequent to the first FEC Block error is also lost. Thus, if k th FEC block is the first FEC block that was corrupted the number of MAC Frames that will be successfully delivered is given by mffecLLk where L fec is the length of the FEC Block and L mf is the length of the MAC Frame. L mf is the sum of the size of the MSDU, L msdu and the overhead added per MSDU, i.e., L OH,SC If p is the probability of an unrecoverable FEC Block error and N is the number of FEC Blocks, the efficiency for Simple Concatenations SC is given by, fecmsdumffeckmffecNkkSCLNLLLkpLLkpp1111 Figure 6-5 shows the efficiency of Simple Concatenation as a function of FEC Block error rate for N = 25 and various MSDU sizes. Figure 6-6 shows the efficiency as a function of N for p = 0.1 and various MSDU sizes.

PAGE 78

64 6.3.2 Concatenation with explicit demarcation For Explicit Demarcation based concatenation, we assume that a single FEC block error will result in loss of all MAC Frames that are contained (either partially or fully) within it. The number of MAC Frames successfully delivered will not only depend on the number of FEC Block errors, but also on how and where they occur. For example, when two side-by-side FEC blocks have errors, then they tend to produce fewer MAC Frame errors than if they are far apart. Hence, it is not possible to derive a simple analytical result for this case. Simulation results were used to study the performance of this MAC framing Scheme. Figure 6-7 shows the efficiency of Concatenation with Explicit Demarcation as a function of FEC Block error rate for N = 25 and various MSDU sizes. Figure 6-8 shows the efficiency as a function of N for p = 0.1 and various MSDU sizes. 6.3.3 2-level framing With 2-Level Framing, only FEC Blocks that were corrupted need to be retransmitted. If p is the probability of FEC Block error, then the average number of FEC Blocks delivered per MPDU is (1-p) times the number of FEC Blocks, The efficiency with 2-level concatenation, L2 is given by sbmsdufecLsbOHfecLLLLLLp2,,21 where L OH, sb, 2L is the overhead per FEC block. Figure 6-9 shows the efficiency of 2-Level Concatenation and Segmentation as a function of FEC Block error rate for N = 25 and various MSDU sizes. Figure 6-10 shows the efficiency as a function of N for p = 0.1 and various MSDU sizes.

PAGE 79

65 Figure 6-5: Simple concatenationefficiency with varying FEC block rrror rate and N = 25 Figure 6-6: Simple concatenation efficiency with varying number of FEC blocks and p = 0.1

PAGE 80

66 Figure 6-7: Explicit demarcation based concatenation efficiency with varying FEC block error rate and N = 25 Figure 6-8: Explicit demarcation based concatenation efficiency with varying number of FEC blocks and p = 0.1

PAGE 81

67 Figure 6-9: 2-Level framing efficiency with varying FEC block error rate and N = 25 Figure 6-10: 2-Level framing efficiency with varying number of FEC blocks and p = 0.1

PAGE 82

68 6.4 Comparison of Performance between various Framing Schemes Using Simple Concatenation, a single FEC block error results in loss of all MAC Frames that it intersects and all subsequent MAC Frames in the MPDU. This approach is very sensitive to FEC block error rate, p, and number of FEC blocks per MPDU, N. For example at 10% FEC block error rate, the efficiency drops to 30%. The effect of MSDU size on efficiency is small compared to that of the FEC block loss probability. Larger MSDUs will also incur higher overheads due to padding (since a number of whole MSDUs are required to fit in each MPDU), thus resulting in lower performance. Using Explicit Demarcation based Concatenation, a single FEC block error results in loss of all MAC frames that it intersects. Simulation results show that the efficiency loss is linear with p and is strongly dependent on the MSDU size. At low p, smaller MSDU sizes show poor performance due to framing overhead. However, as p increases, larger MSDUs sizes suffer higher loss in efficiency since each single FEC block error results in larger number of MAC frame bits lost. With 2-Level Concatenation, only FEC blocks that were corrupted need to be retransmitted. Results show that the efficiency is linear with respect to p, and independent of N. Figure 6-11 shows the relative performance of the three framing approaches for 1518 byte MSDU size with varying FEC block error rate. The results show that the performance of 2-level concatenations is superior to that of other approaches.

PAGE 83

69 Figure 6-11: Performance comparison of various MAC framing strategies with varying FEC block error rate, N= 25, L msdu = 1518 6.5 Conclusions This chapter has highlighted the necessity of concatenation when framing at the MAC level in high-speed PLC communication systems. Five approaches were considered, with three analyzed and simulated in some detail. Approaches varied according to their ability to retransmit at the MAC frame or the FEC block level, and in their ability to resynchronize at the MAC frame level. The results show that 2-Level Framing was far superior to the others in its efficiency. It was also seen to be independent of the number of FEC blocks per MPDU, unlike the other two approaches.

PAGE 84

CHAPTER 7 ALLOCATION REQUIREMENTS FOR SUPPORTING LATENCY BOUND TRAFFIC Impulse noise sources such as dimmers, halogen lamps etc., are quite common in power line networks. Hence, high-speed power line communications systems should be designed to deal effectively with them. HomePlug AV uses aggressive channel adaptation at the physical layer and efficient retransmissions at the MAC layer to deal with impulse noise. This approach can lead to very high probability (up to 20%) of Forward Error Correction (FEC) block errors under some channel conditions. HomePlug AV also has large transmission overhead to ensure reliable communication over the power line. Transmission overhead can negatively impact the MAC efficiency on channels with high FEC Block Error Rate due the need for a large number of retransmissions to successfully deliver packets. This section studies the complex interaction between FEC block errors, transmission overhead, Quality of Service (QoS) requirements of a multimedia stream and its TDMA allocation requirements. Our results show that, to support low latency applications at high BLERs, the average allocation size has to be increased by one or more transmission overheads. Further, the increase in average allocation is not a strong function of application data rate. Hence, low data rate applications can suffer significant loss in efficiency at high FEC Block errors. Our results also show that using a fixed size TDMA allocation can lead to poor utilization when the latency requirements of the application are less than four beacon periods. These results indicate that channel 70

PAGE 85

71 adaptation in HPAV should take into consideration the application QoS requirements to optimize the overall system capacity. 7.1 Introduction The HomePlug AV (HPAV) Medium Access Control (MAC) layer is based on a hybrid of Time Division Multiple Access (TDMA) and Carrier Sense Multiple Access (CSMA). Each HPAV network has a designated device, the Central Coordinator (CCo), which coordinates access to the medium. The CCo periodically (with a period of either 33.33msec or 40msec [33, 34]) generates a beacon with information about the temporal locations of TDMA and CSMA allocations. Multimedia applications use TDMA allocations to provide guarantees on QoS while CSMA allocations are used by regular data traffic. Since power line channels can change with time, each station continuously provides information on the physical layer (PHY) data rates and the size of its queue (i.e., backlog) to the CCo. The CCo uses this information to dynamically adjust the size of TDMA allocations. HPAV stations use channel adaptation to choose the best set of physical layer parameters (carrier modulation, FEC code rate etc.,) for communication with other stations. Channel adaptation in HPAV is also tailored to effectively deal with impulse noise [43, 44]. In most communication systems, the channel adaptation mechanism is used to reduce the PHY layer errors to a range where they can be effectively corrected by the MAC layer. For example, the channel adaptation can be chosen such that the FEC block errors are less than 1/100 and the MAC retransmissions can be used to recover the uncorrectable errors. Such a technique is not viable under power line channel environments due to the presence of impulse noise. Impulse noise channels will display temporal effects in channel capacity. In between two impulse noise events, the channel

PAGE 86

72 capacity can be very high, while at the impulse noise the capacity is very low. To obtain less than 1/100 FEC block error rate, the channel has to be adapted close to the lower end of its capacity (i.e., capacity near impulse noise events) and hence is undesirable when high throughput is required. A better approach is to adapt aggressively and use a MAC ARQ mechanism to correct errors. In some environments, to achieve optimal performance an HPAV station may need to operate at FEC block error rate as high as 20%. To effectively recover from large number of FEC Block errors, HPAV uses a novel 2-level MAC framing mechanism and selective acknowledgments (SACK). The 2-level MAC framing essentially concatenates and then segments higher layer packets (or MAC Service Data Units) into fixed size block (referred to as Segments) that fit into a single 520-octet FEC block. Since errors at the physical layer occur on an FEC block basis, only Segments that get corrupted needs to be retransmitted. Further details on 2-level MAC framing can be found in Chapter 6 [42]. Apart from the MAC framing overheads, each transmission will also incur overheads due to physical layer headers, selective acknowledgments and interframe spacings. Figure 7-1 illustrates the overheads incurred in HPAV during TDMA allocations. Segments are transmitted in Long Physical Layer Protocol Data Units (PPDUs). Each long PPDU consists of a Start of Frame (SOF) delimiter followed by one or more FEC blocks that carry Segments. A Delimiter in HPAV consists of a preamble sequence followed by 128-bit frame control information. Preambles are used to demarcate the start of each PPDU. A SOF frame control contains broadcast information that is necessary for proper demodulation of the PPDU payload. The frame control of a

PAGE 87

73 SACK delimiter contains the reception status of the FEC Blocks in the Long PPDU. Since proper reception of the delimiter is critical for reliable exchange of data, delimiters in HPAV are designed to be highly robust and hence are quite long (110.48sec). Typically, a Response Interframe Spacing (RIFS_AV) of 100sec is used between the transmission of long PPDU and the corresponding SACK. A SACK and the subsequent Long PPDU are typically separated by 100sec (note that HPAV allows variable interframe spacing to enable flexibility for a variety of implementations). Thus we see that each PPDU transmission will incur an overhead of around 420sec. Due to the fixed transmission overhead in HPAV, the retransmission strategies can greatly influence the MAC efficiency. For example, Figure 8-1 also shows a scenario where Segment contained in FEC Block#2 gets corrupted during the initial transmission and is retransmitted immediately. Since a 420sec overhead is incurred for retransmitting a single Segment, this reduces efficiency. In this section, we refer to this retransmission strategy as Immediate Retransmission strategy. A strategy that can overcome the inefficiencies of immediate retransmissions is shown in Figure 8-2. In this approach, the transmitter waits until sufficient numbers of new Segments are available before retransmitting the corrupt Segment(s). We refer to this retransmission strategy as Delayed Retransmission strategy. In essence, the delayed retransmission strategy reduces the average overhead incurred for transmission of each segment, thus improving the efficiency. The drawback of this strategy is that, additional delays are incurred while the transmitter aggregates new Segments. While such additional delays are not an issue for data traffic, they may not be acceptable for QoS sensitive multimedia streams.

PAGE 88

74 Multimedia streams require guarantees on both latency and packet loss probability. Since multimedia streams are provided with one TDMA allocation per beacon period, each Segment will have a fixed number, L, of TDMA allocations within which it need to be delivered. In contrast, to achieve a given packet loss probability, Segments belonging to the applications will need to be guaranteed a certain number of transmission opportunities, TXOPS. In situations where L TXOPS, delayed retransmissions can be used. Otherwise, a hybrid of immediate and delayed retransmission strategies has to be used to meet the packet loss requirement. Inefficiencies will be incurred in the later case, due to additional transmission overheads. Figure 7-1: Transmission overheads in HomePlug AV Figure 7-2: Illustration of delayed retransmission strategy In this section we analyze the effect of FEC block errors and transmission overheads on the allocation requirements for the multimedia streams. The characteristics

PAGE 89

75 of the streams that are of interest are its data rate, latency and packet loss tolerance. We are interested in two different types of allocation requirements, 1. Fixed Allocation This is the duration of the fixed size TDMA allocation that is required to satisfy the QoS requirements of the multimedia stream. It is important to note that due to the statistical nature of the FEC errors, it is possible that the Fixed Allocation may not always be fully utilized. 2. Average Allocation This is the average duration of a variable size TDMA allocation that will be required to satisfy the QoS requirements of the multimedia stream. To simplify our analysis and presentation of the results we make the following assumptions, We assume that the application generates constant bit rate traffic. Application data rate is measured in terms of the number of Segments (N), that it generates per beacon period, We measure the application latency L in multiples of beacon period. Each Segment will have transmission opportunities in L number of TDMA allocations before it is considered stale. We measured packet loss probability in terms of the packet loss event probability (PLEP) and average packet lost (APL) per packet loss event, We assume that FEC Block errors are uncorrelated and occur with a probability (p) Since the efficiency of a TDMA allocation is primarily dependent on the ratio of the transmission overhead to transmission duration of a FEC block, we do our analysis in units of FEC Block transmission duration. For example, at physical layer data rate of 150Mbps, the transmission duration of an FEC block is 27.7msec. Hence the ratio of transmission overhead to FEC block transmission time (represented by H) is 15.16 (i.e., 420/27.7). Since the maximum physical layer data rate of HomePlug AV is 150Mbps, we are interested in H (0, 15.16]. The organization of the rest of the section is as follows. Analysis of Fixed Allocation and Average Allocation requirements without transmission overheads is presented in Section 7.2. An approximate analysis of Fixed Allocation and Average Allocation requirements in the presence of transmission overheads is presented in Section 7.3. This section also includes simulation based validation of the analysis. The effect of

PAGE 90

76 transmission overhead on allocation requirements is presented in Section 7.4 and conclusions are presented in Section 7.5. 7.2 Allocation Requirements without Transmission Overheads In this section we investigate the allocation requirements when no transmission overheads are present. In practice, it is not possible to design systems that do not such overheads. Hence, the results presented are of theoretical importance as they provide optimal bounds with which to compare actual achieved efficiency. To successfully transmit each Segment, an average of 11p number of transmission opportunities will be needed. Hence the Average Allocation, requirement for transmitting N Segments is given by, avgT pNpNTavg1, The computation of the Fixed Allocation that can enable the required PLEP is more involved and depends upon whether L = 1 or L > 1. 7.2.1 Fixed Allocation, L = 1 Fixed Allocation for L=1 can be computed by nothing that the probability that k out of N Segments remain when an allocation of size A (A>N) is granted on a channel with FEC Block error probability p, is given by 0;110;1,,Pr1NkkNkNAmNmNAmppkANmppmAApNremainSegmentsm From this, the packet loss event probability (PLEP) and Average packet loss (APL) per packet loss event can be obtained for an allocation A, as follows,

PAGE 91

77 mppmAAPLppmAPLEPNmmNmNANmmNmNA1111 Based on the above equation, the Fixed Allocation can be obtained by determining the minimum value of A that will achieve the required PLEP. fixedpNT, 7.2.2 Fixed Allocation, L > 1 Analytical solution for the Fixed Allocation cannot be obtained for L > 1. Hence we resort to a Markov chain [45] based analysis. This analysis assumes a retransmission strategy in which the oldest segment in the queue will be transmitted before transmitting any other Segments. Since no transmission overhead is incurred, this strategy is optimal in the sense that it provides the retransmission opportunities to the most urgent Segment. First, consider the buffer backlog of the stream at the end of each beacon period. It can be readily seen that the variation of the buffer backlog is only dependent on the current backlog and the size of the Fixed Allocation for a given FEC block error probability. Furthermore, the maximum buffer backlog is bounded by N(L-1). Note that this upper bound is a manifestation of latency bound associated with the Segments. A discrete-time, finite state Markov chain model can be used to determine the steady state probabilities of buffer backlog from which the PLEP can be computed. Figure 7-3 shows the state transition diagram of buffer backlog when L > 2. The state transition probabilities, are given by jip, otherwiseiNjAiNNNLiApiNjpNPji00,max,0,,Pr,

PAGE 92

78 Where, ApiNj,,Pr is given by, 0;110;1,,Pr1iNkkiNkiNAjiNjiNAjppkANjppjAApiNj The steady state probabilities, can be obtained from the state transition probabilities by solving the equation, NPiP NPiNPiNPjiPPp, Figure 7-3: State transition diagram for no MPDU overhead case From the steady state probabilities the packet loss event probability and average packet loss per packet loss event can be computed based on the set of transitions that result in a packet loss. Figure 7-4 shows the excess allocation required to achieve a packet loss event probability of 10 -5 for N = {50,100, 200} and L = [1, 2, 3]. Note that a PLEP of 10 -5 will result in one packet loss event in an hour and hence is a reasonable

PAGE 93

79 choice for most multimedia applications. Also, N = 50, 100 and 200 correspond to application data rates of 6Mbps, 12Mbps and 24Mbps, respectively when the beacon period is 33.33msec. The excess allocation is measured as the difference between the actual and the minimum allocations required to ensure stability (i.e., pN1 ). These results show that for L = 1, a large amount of over allocation is required to achieve low packet loss event probability. However for L > 1, the amount to over allocation required is negligible. Figure 7-5 shows the average number of packets lost per packet loss event for N = {50,100, 200} and L = [1, 2, 3]. Results also show that the APL tends to be small and show a slight increase as N, L or p increases. Figure 7-4: Excess allocation as a function of FEC block error probability when no transmission overhead present

PAGE 94

80 Figure 7-5: Average packet loss per packet loss event when no transmission overheads are present 7.3 Allocation Requirements with Transmission Overheads Results from previous section show that when no MPDU overhead is incurred, for latencies greater than one, the allocation requirement are very close to the optimal allocation i.e., pN1 In the presence of transmission overheads, the retransmission strategy used in the previous will result in a Fixed Allocation and Average Allocation of HpN11 It can be readily seen that for the range of values of H we are interested in, this can result in gross inefficiencies and hence is not viable. A better approach is to use a hybrid of immediate and delayed retransmission strategies where in at each transmission opportunity, as many Segments as possible are transmitted starting from the oldest pending Segment based on the time available in the TDMA allocation. We are interested in determining the duration of Fixed Allocation and

PAGE 95

81 the Average Allocation for this retransmission strategy. Unfortunately, due to the intricate interaction between the allocation size, FEC errors, and transmission overheads this retransmission strategy is not amicable to direct analysis. In theory, it is possible to conduct a Markov chain analysis by defining the Markov chain state as a L-tuple of backlogs, (b 0 ,b 1 ,,b L-1 ), where b i indicates the backlog of Segments that expire at the end of i th beacon period following the current beacon period. Thus, b 0 will indicate the backlog of Segments that expire at the end of the current beacon period and so on. The number of state for such a Markov chain will be (N+1)L. Due to the computational requirement for obtaining steady state probabilities of the Markov chain, this method will only be useful for small values of N and L. To overcome these problems, we use a combination of simplified analysis and simulations to provide approximate estimates of the Fixed Allocation and Average Allocation. 7.3.1 Fixed Allocation Consider a multimedia stream with a data rate of N Segments per beacon period being transmitted over a channel with FEC Block Error probability of p. If each of the Segments is provided with transmit opportunities, the probability that a given Segment gets successfully delivered is totalT totalTp1 Since FEC Block errors are uncorrelated, the probability that all N Segments will get successfully delivered is Hence the packet loss event probability, PLEP, is given by NTtotalp1 NTtotalpPLEP11 The above equation can be used to determine the number of transmission attempts required to achieve a given PLEP as shown below. Note that represents the ceil operation.

PAGE 96

82 pPLEPTNtotal10110log11log Figure 7-6 shows the number of transmission opportunities required to achieve a PLEP of 10 -5 for N = [50, 100, 200] and various values of p. These results shows that the number of increases with N and p. These results also show that is not highly sensitive to N. totalT totalT Figure 7-6: Transmission opportunities required to achieve PLEP = 10 -5 at various FEC block error probabilities Since a minimum of T transmit opportunities needs to be provided in L TDMA allocations, the number of transmit opportunities per TDMA allocation is, LTTtotalalloc Based on this, an approximate estimate of Fixed Allocation (FA_Theory) can be obtained as,

PAGE 97

83 pNHLTTheoryFA1_ The first term in this equation reflects the contribution of transmission overheads to the Fixed Allocation. The second term shows the contribution due to multiple retransmissions of Segments. We basically use the minimum Segment allocation required for stability as a first order approximation. In reality, due to the statistical nature of the FEC block errors there will be a complex interaction between the FEC Block errors and number of retransmissions. For example, in some TDMA allocations the FEC block errors can be small resulting in lager number of retransmission opportunities and vice versa. Simulations were used to validate how well the theoretical estimate matches the actual allocation requirement. Simulations were performed for N = [50, 100, 200], L = [1, 2] and p = [0.001, 0.005, 0.01, 0.05, 0.10, 0.15, 0.20]. Each simulation was run for 10 7 beacon periods with varying allocations to determine the Fixed Allocation (FA_Sim) that would guarantee PLEP less than 10 -5 Figure 7-7 shows the Fixed Allocation estimation error in for N=100. Estimation error is measured relative to the theoretically computed Fixed Allocation. These results show that the estimation error is large for small values of latencies and large values of FEC block error probabilities. This can be attributed to the use of average value of allocation for Segments in the theoretical estimate. Figure 7-8 shows the probability distribution of the estimation error for all the simulation runs. These results show that the theoretical estimates remain within 5% of the actual Fixed Allocation in 90% of the data points. In general, for L > 3, our theoretical estimates accurate showed an error of less than 5% in 99% of the data points and an error of less

PAGE 98

84 than 2% in 97% of the data points. Our simulations also showed that the estimation errors remain similar for PLEP of 10 -3 and 10 -4 Figure 7-9 shows that average packets lost per packet loss event as a function of FEC block error probability for all the simulations. Results show that the average number of packets lost per packet loss event tends to remain very close to 1 in all the simulations except for H = 1, L = 1. In the later case, the average packet loss showed slight increase with N or p. Figure 7-7: Fixed Allocation estimation error for N = 100 and L = 1, 2, 3, 4, 6, and 8

PAGE 99

85 Figure 7-7: Continued Figure 7-8: Probability distribution function of the Fixed Allocation estimation error

PAGE 100

86 Figure 7-9: Average packet loss per packet loss event when transmission overheads are present 7.3.2 Average Allocation To successfully transmit each Segment, an average of 11p number of transmission opportunities will be needed. Hence the contribution of Segment transmit times to the Average Allocation is exactly pN1 The average time used by transmission overheads will depend on the probability distribution of the transmission attempts. To obtain this, we assume that the number of transmission attempts will be upper bounded by i.e., the minimum number of transmission attempts per TDMA allocation required to achieved the required PLEP. We further assume that all pending Segments will be transmitted at each transmission opportunity (i.e., no truncation effects due to fixed size allocation). Since the probability that a Segment gets delivered allocT

PAGE 101

87 successfully in an allocation is allocTp1 the average number of Segments pending at the start of each TDMA allocation ()is given by, 'N allocTpNN1' To obtain the distribution of transmission attempts, we first note that the probability of delivering Segments in exactly i transmission attempts is 'N ''111NiNipp where i [1,]. Since the maximum number of transmit attempts is upper bounded by the condition distribution of transmitted attempts is given by, allocT allocTjNjNjNiNippppitattemptstransmit111''''1111,Pr Thus, the average time required for transmission overheads ( H ) is given by, allocTiHitH1Pr From this, the theoretical estimate of Average Allocation (AA_Theory) can be obtained as, pNHTheoryAA1_ We again use simulations to determine how well the theoretical estimate matches the actual Average Allocation (AA_Sim). Figure 7-10 shows the error in estimation of Average Allocation (i.e., AA_Sim AA_Theory) for N=100. These results show that the theoretical estimates for Average Allocation are very accurate. This is further validated in Figure 7-11, where we plot the probability distribution of the estimation error for all the simulations.

PAGE 102

88 Figure 7-10: Average Allocation estimation error for N = 100 and L = 1, 2, 3, 4, 6 and 8

PAGE 103

89 Figure 7-11: Probability distribution function of the Average Allocation estimation error 7.4 Effect of Transmission overheads on Allocation Requirements In this section we use the simulation results and analytical estimates of the Fixed Allocation and Average allocation to determine the impact of transmission overheads on on the allocation requirements efficiency. We first note that in the presence on transmission overheads multimedia streams whose latency L is much larger than the number of transmission opportunities required to achieve the PLEP, the Fixed Allocation and Average Allocation required are given by, pNHMinAApNHMinFA1_1_ These represent the minimum allocation requirement, as large allocations will be necessary at lower values of latencies due to transmission overheads.

PAGE 104

90 Figure 7-12 shows the percentage difference in the Fixed Allocation (FA_Sim) and minimum Fixed Allocation (FA_Min) for N = [50,200], and p = [0.01, 0.10]. These results show that for low latencies, significantly large TDMA allocations are required to support the stream For example, for H=10, and p=0.01, 16% additional allocation will be needed to support a Latency of 3 beacon periods compared to the Latency of 4. The percentage increase in TDMA allocation requirement reduces as the data rate of the stream increases. This can be expected as the transmission overhead does not change significantly (refer to Figure 7-6) with the data rate of the stream. Figure 7-13 shows the percentage difference in the Average Allocation (AA_Sim) and minimum Average Allocation (AA_Min) for N = [50,200], and p = [0.01, 0.10]. Comparing these with Figure 7-12, we see that at low latencies, the utilization of the TDMA allocation will remain poor. For latencies larger that four, the differences between the Fixed Allocation and Average Allocation are minor. The analytical results on Average Allocation can be used to determine the effective transmission overhead (i.e., H /H) incurred for multimedia streams. Figure 7-14 shows the effective transmission overhead incurred for various values of latencies and FEC block error probabilities for N=50, 200. These results show that, Latencies greater than 10 and 7 will be needed at 20% and 10% FEC Block error rates, respectively, to support streams with N = 200 (or 24Mbps application data rate), and PLEP=10 -5 with minimal overheads. Providing one additional transmission overhead of allocation will enable support of latencies greater than 5 and 3 at 20% and 10% FEC Block error rate, respectively Latencies greater than 10 and 7 will be needed at 20% and 10% FEC Block error rates, respectively, to support streams with N = 200 (or 24Mbps application data rate), and PLEP=10 -5 with minimal overheads. Providing one additional transmission overhead of allocation will enable support of latencies greater than 5 and 3 at 20% and 10% FEC Block error rate, respectively

PAGE 105

91 Figure 7-12: Percentage difference in Fixed Allocation and Optimal Allocation for N = [50, 200], p= [0.01, 0.10] Figure 7-13: Percentage difference in Average Allocation and Optimal Allocation for N = [50, 200], p= [0.01, 0.10]

PAGE 106

92 Figure 7-13: Conclusions Figure 7-14: Effective transmission overhead for Average Allocations with N = 50 and 200 7.5 Conclusions In this section we investigated the effect of FEC Block errors on the allocation requirements for multimedia stream. To overcome the complexities incurred in direct analysis and Markov chain based analysis, we presented an approximate analytical estimate of the Fixed Allocation and Average Allocation. Simulations were used to validate the accuracy of the analytical estimates. Our results show that, for supporting low latency applications the average allocation size has to be increased by one or more

PAGE 107

93 transmission overheads. Further, the increase in allocation requirement does not show strong dependency on the application data rate. Thus, low data rate applications can suffer significant loss in efficiency under her FEC block error. Hence, channel adaptation should take into consideration the application data rate and QoS requirement to increase the overall system capacity.

PAGE 108

CHAPTER 8 BEACON SCHEDULE PERSISTENCE TO MITIGATE BEACON LOSS HomePlug AV uses a beacon based Time Division Multiple Access (TDMA) technique for providing guaranteed bandwidth to multimedia streams. Beacons in HomePlug AV carry TDMA allocation (or scheduling) information for various active multimedia streams. Since a station that missed a beacon will not know the schedule for that beacon period, it will not be able to transmit. Due to the large attenuation and impulse noise, beacon detection cannot always be guaranteed. To overcome occasional beacon loss, HPAV uses persistent beacon schedules. Persistent schedules are valid for multiple beacon periods, thus enabling a station to transmit even if it misses one or more beacons. In this section we use a Markov chain based analysis to investigate the impact of beacon loss on systems with and without persistent beacon schedule. Results show that significant loss in efficiency, and in the level of guarantees on QoS, can occur when no schedule persistence is used. The analysis further shows that the use of persistent current and preview schedule or persistent preview schedules can enable HPAV systems to tolerate up to 10% beacon loss probability with minimum impact of MAC efficiency and QoS guarantees. A subset of the results presented in this chapter are also published in [46] 8.1 Introduction HomePlug AV Medium Access Control (MAC) layer is based on a hybrid of Time Division Multiple Access (TDMA) and Carrier Sense Multiple Access (CSMA). Each 94

PAGE 109

95 HPAV network has a designated device, the Central Coordinator (CCo), that coordinates access to medium. The CCo periodically generates a beacon with information about the temporal location of TDMA and CSMA allocations for the stations in the network. TDMA is used by applications requiring QoS, while CSMA is used for transport of regular data. In HPAV, CCo beacon is synchronized to the underlying AC line cycle and is transmitted once every two AC line cycle periods. Thus in North America, CCo beacon has a period of 33.33msec, while in Europe a 40msec beacon period is used. Stations with multimedia traffic continuously communicate with the CCo regarding the required duration of their TDMA allocations. This continuous communication is essential as traffic characteristics and channel characteristics can change with time. Based on this feedback, the CCo updates the schedules. In general, the schedule can change on each beacon Period. Hence, a station that failed to receive beacon cannot transmit in the corresponding beacon Period. The CCo beacons in HPAV are 136-octets long. To enable robust reception under a variety of channel conditions, they are encoded using rate turbo convolutional codes, spread in frequency and time by using 5-time repetition code, and are modulated use QSPK modulation. Under AWGN conditions, this encoding will enable reliable beacon reception up to -3dB Signal-to-Noise ratio. In typical homes, the beacon loss probability can be expected to be less that 1%. However, due to the large dynamic range of attenuation and (impulse) noise over power line, beacon loss can be large than 1% in some cases. To overcome such scenarios, HPAV provides two mechanisms, 1. Proxy Networking, and 2. Beacon Schedule Persistence

PAGE 110

96 Using the proxy networking procedure, a CCo can appoint one or more stations as Proxy Coordinators. Proxy Coordinators are provided with TDMA allocations to retransmit the CCos beacon. This repetition of the CCos beacon enables propagation of schedule information to stations that cannot reliably hear the CCo but can hear the proxy station(s). Proxy Networking in HPAV is primarily suited for handling hidden stations; however this function can also be used for enhancing beacon schedule reliability. HPAV standard uses beacon schedule persistence to deal with occasional beacon losses. Beacon schedule persistence enables the CCo to publish a schedule that is valid for multiple beacons. This will enable station(s) that misses to beacon to transmit in a beacon period where the schedule information is known based on previously received beacons. In this section, we provide an in depth analysis on the effect of beacon loss in HPAV networks and investigate the performance gain that can be achieved by using persistent schedules. The organization of the remaining sections is as follows. Section 8.2 provides details on the various Persistent Beacon Schedules allowed in HomePlug AV. We investigate the effect of beacon loss on efficiency in Section 8.3. Section 8.4 provides a frame work for detailed study of the complex interaction between application QoS, MAC Efficiency and beacon loss probability. The effect of beacon loss on the lower bound of packet loss probability for various scheduling strategies is analyzed in section 8.4. Detailed Markov chain based analysis of the effect of beacon loss on QoS and efficiency is presented in Section 8.6. Comparison of the performance of various Persistent Beacon Schedules is presented in Section 8.7 and conclusions are presented in Section 8.8.

PAGE 111

97 8.2 Persistent Beacon Schedules Beacons in HomePlug AV are 136-octets long and can carry variable length scheduling information. The scheduling information consists of the start and end times of allocations for various multimedia stream, along with the information on the beacon period(s) in which the scheduling information applies. Binding the scheduling information to the beacon period(s) in which it is applicable is achieved by means of a Current Schedule Count Down (CSCD) and a Preview Schedule Count Down (PSCD) fields. These fields allow three different types of persistent schedules, 1. Persistent current schedules, 2. Persistent preview schedules, and 3. Persistent current and preview schedules. Persistent current schedules are indicated by setting the PSCD field to zero. In this case, the schedule is valid in the current beacon period and the CSCD indicates the number of up coming beacon periods for which the schedule remains valid (i.e., unchanged). Current schedules enable a station that receives a schedule to transmit in all subsequent beacon periods during which it remains valid. Figure 8-1 shows an example of a 2-persistent current schedule. In this example, beacons B1 carries schedule A that is indicated as being valid in beacon period B2 also by setting CSCD to 1. Beacon B2 also carriers schedule A, and indicates that it is valid only in beacon period B2 by setting CSCD to 0. Similarly, beacons B3 and B4 carry schedule B that is valid in beacon periods B3 and B4, and so on. In general, an r-persistent current schedule remains valid for (r-1) beacon periods subsequent to the beacon period in which it is first published. When r-persistent current schedule is used, the maximum latency incurred by the CCo from the time a schedule change is planned to when it becomes effective is (r-1). The

PAGE 112

98 main disadvantage of persistent current schedules is that new schedules are most venerable in the beacon periods where they are first published. Persistent preview schedules are indicated by setting the PSCD to a non-zero value. In this case, the PSCD field indicates the number of beacon periods (including the current beacon period) in which schedule become active. The CSCD field indicates the number of beacon period for which the preview schedule will remain active minus one. preview schedules enable the CCo to repeat a schedule certain number of times before it becomes active. Figure 8-2 shows an example of a 2-persistent preview schedule. In this example, beacons B1 carriers Schedule B and indicate that it becomes valid in beacon period following beacon B3 (i.e., PSCD = 2) and remains valid for two beacon periods (i.e., CSCD=1). Beacon B2 also repeats the same scheduling information. Similarly, beacons B3 and B4 carry schedule C that is valid in beacon periods B5 and B6, and so on. In general, an r-persistent preview schedule is repeated in r beacon periods before it becomes active and it also remains active for r beacon periods. The main advantage of this method is that a station is only required to receive any one of the r-beacons to be able to transmit in a beacon period. This increases resilience to beacon loss. The drawback of this approach is the additional latencies incurred for the CCo to change schedules. For r-persistent preview schedules, a latency of up to (2r-1) beacon periods can elapse from the time the CCo decides to change a schedule until the new schedule becomes active. HPAV allows the CCo to include one preview schedule and one current schedule in the same beacon, if there is sufficient space available. Note that HPAV beacons always carry 136 octets of information and the scheduling information for each TDMA

PAGE 113

99 allocation typically requires 3 octets. So, if is very likely to have sufficient space to include both current and preview schedules into a beacon. Figure 8-3 shows an example of 2-persistent current and preview schedule. In this example, beacon B1 and B2 carry current schedule A that is valid in their respective beacon periods and preview schedule B that is valid in the beacon periods following B3 and B4. Similarly, beacons B3 and B4 carries current schedule B and preview schedule C, and so on. In an r-persistent current and preview schedules, each schedule is transmitted times, thus enhancing its reliability. The maximum latency incurred in changing r-persistent current and preview schedule is the same as the maximum latency of the corresponding r-persistent preview schedule. B1(A) B2(A) B3(B) B4(B) B5(C) B6(C) Schedule A Schedule A Schedule B Schedule B Schedule C PSCD = 0CSCD = 1PSCD = 0CSCD = 0PSCD = 0CSCD = 1PSCD = 0CSCD = 0PSCD = 0CSCD = 1PSCD = 0CSCD = 0 Figure 8-1: Example of 2-persistent current schedules B1(B) B2(B) B3(C) B4(C) B5(D) B6(D) Schedule A Schedule A Schedule B Schedule B Schedule C PSCD = 2CSCD = 1PSCD = 1CSCD = 1PSCD = 2CSCD = 1PSCD = 1CSCD = 1PSCD = 2CSCD = 1PSCD = 1CSCD = 1 Figure 8-2: Example of 2-persistence preview schedules

PAGE 114

100 B1(B,A) B2(B,A) B3(C,B) B4(C,B) B5(D,C) B6(D,C) Schedule A Schedule A Schedule B Schedule B Schedule C PSCD = 2CSCD = 1PSCD = 1CSCD = 1 PSCD = 2CSCD = 1PSCD = 1CSCD = 1PSCD = 0CSCD = 1PSCD = 0CSCD = 0 PSCD = 0CSCD = 1PSCD = 0CSCD = 0 PSCD = 0CSCD = 1PSCD = 0CSCD = 0 PSCD = 2CSCD = 1PSCD = 1CSCD = 1 Figure 8-3: Example of 2-persistent current and preview schedules 8.3 Efficiency Analysis A station that misses a beacon may not be able to transmit in its allocation. This reduces the overall utilization of an allocation, thus reducing the efficiency. In this section we use simple analysis to study the effect of beacon loss on efficiency for various scheduling strategies. 8.3.1 Non-persistent schedule For non-persistent schedules, every time a beacon loss occurs, the corresponding allocation cannot be used. Hence the efficiency of non-persistent schedules, is NP bNPp1 8.3.2 Persistent current schedules For r-persistent current schedules, effect of a beacon loss will depend on the location of the corresponding beacon period (i.e., beacon period where the beacon loss occurred) relative to the beacon period in which the schedule is first published. To miss an allocation in a beacon period that is t-beacon periods from the first beacon period in which the schedule was published, a total of beacons have to be missed. Hence, the efficiency of the t th beacon period, is given by, rCSt,

PAGE 115

101 1,1tbrCStp Thus, the efficiency of an r-persistent current schedule, is rCS, brbbrttbrCSprpppr11111101, 8.3.3 Persistent preview schedules For r-persistent preview schedules, an allocation can be utilized if any one of the r-beacons in which the schedule was previewed is successfully received. Hence, the average efficiency of a r-persistent preview schedule, is given by, rPS, rbrPSp1, 8.3.4 Persistent current and preview schedules For r-persistent current and preview schedules, effect of a beacon loss will again depend on the location of the corresponding beacon period relative to the beacon period in which the schedule is first published. Further it will also depend on the reception station of the beacons carrying the preview schedule. To miss an allocation in a beacon period that is t-beacon periods from the first beacon period in which the schedule was published, a total of beacons have to be missed. Hence, the efficiency of the t th beacon period, is given by, rPSCSt,, 1,,1trbrPSCStp Thus, the efficiency of an r-persistent current and preview schedule, is rPSCS,, brbrbrttrbrPSCSprpppr111111101,,

PAGE 116

102 8.3.5 Efficiency comparison A comparison on efficiency loss for various scheduling strategies under different beacon loss probabilities is shown in Figure 8-4. These results can be summarized as follows, The efficiency loss for non-persistent schedules in proportional to the beacon reliability. Thus, 10% loss in efficiency will be experienced in environments where the beacon loss probability is 10%. Persistent current schedules do not provide significant improvement in efficiency compared to non-persistent schedules. Further, most of the efficiency loss can be attributed to the poor utilization of the beacon period in which the schedule is first published. In environments with 10% beacon loss probability, current schedules with persistences of 2, 4, and 8 show approximately 5.5%, 2.8% and 1.4% loss in efficiencies, respectively. Significant improvement in efficiencies can be obtained by using persistent preview schedules. 2-persistent preview schedules show a 1% loss in efficiency in environments with 10% beacon loss. Preview schedules with persistent greater than 2, will show negligible loss in performance Persistent preview and current schedules provide the best performance in terms of efficiency. A 2-persient current and preview schedule shows 0.06% loss in efficiency in environments with 10% beacon loss. Figure 8-4: Efficiency loss comparison for various scheduling strategies

PAGE 117

103 8.4 QoS Considerations Apart from reduction in efficiency, beacon loss can also affect the QoS. For example, when non-persistent schedules are used, a burst of beacon errors that span a duration larger than the latency of the multimedia stream will result in a packet loss event. In this section we study the interaction between QoS and beacon loss. In general, the effect of beacon loss on multimedia stream will depend on the following factors, 1. Data rate of the multimedia stream, 2. Latency requirement of the multimedia stream, 3. The maximum packet loss probability tolerated by the multimedia stream, 4. Beacon loss probability observed over the power line channel, 5. The schedule persistent mechanism used by the CCo, and 6. The allocation strategy used by the CCo to clear the backlog created at the station due to beacon loss In the following sections we derive the relationship between these parameters based on the following assumptions, We first assume that the multimedia stream has a constant data rate. Further we assume that the application requires transmission of N Segments (of data) in each beacon period. In HPAV, Segments are 512-octet data units that carry application payload. Hence, N is a function of the data rate of the multimedia stream. In the remainder of the section, we refer to each Segment and packet, Since multimedia streams are provided with one allocation each beacon period, we assume that the latency (L) of the application is in terms of a certain number of beacon periods. This does not limit our analysis in any manner, however it simplifies presentation of the results and makes them independent of the beacon Period (with can be either 33.33msec or 40 msec), Beacon loss events tend to create burst of packet losses. Hence we measure the QoS in terms of packet loss event probability (PLEP) and the average number of packets lost per packet loss event (APL), We assume that the source of the multimedia stream is experiencing uncorrelated beacon loss events with a beacon loss probability of p b Beacons in HPAV are synchronized with respect to the AC line cycle, and power line channel characteristics tend to remain static with respect to the AC line cycle [37, 38], justifying the uncorrelated beacon loss assumption. As we are only interested in the effect of beacon loss, to simplify the analysis we further assume that all transmitted

PAGE 118

104 packets will be successfully received (i.e., no packet loss due to errors in the reception of physical layer protocol data units), Our analysis studies all Scheduling strategies allowed in HPAV, Packets that did not get delivered due to beacon loss should be transmitted in subsequent beacon periods. There are several strategies that can be used by the CCo to enable the station to clear the backlog. These include, 1. Always provide a fixed (X) number of packets of over allocation, 2. Provide a fixed (X) number of packets of over allocation, whenever there is backlog at the station, 3. Provide variable number of packet of over allocation based on backlog, with an upper bound (X) on the maximum over allocation. Approach (a) is by far the simplest. However, this is also the most inefficient approach as the over allocation will most likely remain unused. Approaches (b) and (c) use dynamic over allocation, and they require the CCo to track the backlog for each stream. In HPAV, all streams provide an indication of their backlog to the CCo, thus making approaches (b) and (c) practical. For our analysis, we use choose approach (b), as it is more efficient than (a) and is simpler to analyze than (c). We also assume that the CCo can always track the backlog at the station. In practice, it is possible for CCo to miss transmissions from the stations (i.e., similar to the station missing beacons). However, since the CCo only needs to determine whether a station transmitted in an allocation (irrespective of whether it properly received or not) this can be very highly reliable, thus justifying our assumption. 8.5 Lower Bound on PLEP In this section we derive lower bounds on the PLEP that can be achieved for a given Latency requirement (L) and Beacon Loss Probability (p b ). The derivation of the lower bound is motivated by two reasons. Firstly, as we will see in Section 7, the

PAGE 119

105 derivation of the accurate relationship between {N, p b L, PLEP and Alloc} is quite involved and requires solving complex Markov chains. In contrast, a closed form expression can be obtained for the upper bound. Secondly, as will be shown in subsequent sections, upper bound does provide a good estimate on the actual performance of two of the more important scheduling schemes i.e., persistent current and preview schedules, and persistent preview schedules. We derive the lower bound on the PLEP by assuming that the CCo provides sufficient amount of over allocation in each beacon period to completely eliminate any backlog. This assumption essentially makes PLEP a function of the number of consecutive beacons missed by the station, thus simplifying the analysis. 8.5.1 Non-persistent schedules For non-persistent schedules, a minimum of L consecutive beacons has to be missed to incur a packet loss event. When exactly L consecutive beacons are lost, one packet loss event will occur. When exactly (L+1) beacons are missed, two packet loss events will occur, and so on. For the general case of consecutive beacons being missed, packet loss events will occur. Since consecutive beacons can be lost with probability the packet loss event probability for non-persistent schedules, 1L 21bbpp UBNPPLE P , can be obtained as, LbLbbLBNPpLppPLEP112, 8.5.2 Persistent current schedules For r-persistent current schedules, the PLEP will not only depend not only on the number of consecutive beacons missed, but also on the location of the first missed beacon relative to the beacon in which the corresponding schedule is first published. For a burst

PAGE 120

106 of beacon losses starting with a beacon in which the schedule was first published, the packet loss event probability can be obtained in a manner similar to that of non-persistent schedule, i.e., LbLbbLBrCStpLppPLEP112,,0 For the case of a burst of beacon losses starting with a beacon that is t beacon periods (with t [1 r-1]) from the beacon in which the corresponding schedule was first published, (r-t) additional beacons have to be lost to incur the same number of packet loss events as that of t=0. Hence, can be obtained as follows, LBrCSrtPLEP,,]1,1[ trLbtrLbbLBrCSrtptrLppPLEP112,,]1,1[ Thus, the average PLEP of r-persistent current schedule, LBrCS P LEP,, is, brbLbrttrLbLbrtUBrCStLBrCSprpppprPLEPrPLEP11111110,,,, 8.5.3 Persistent preview schedules As with r-persistent current schedules, the packet loss event probability for r-persistent preview schedules will show dependency on the location of the first beacon lost beacon relative the beacon carrying new schedule. For this analysis, we represent Latency L as, rawherearkL 1,; Consider the case in which the first beacon missed is the beacon in which a new schedule is first published. The first set of packet loss events will occur when (k+1)r consecutive beacons are missed. Figure 8-5 shows and illustration of the relationship

PAGE 121

107 between latencies and persistent preview schedules. In this example, for a packet loss event to occur all beacons carrying Schedule#0 through Schedule#k have to be missed. It can also be seen that beacon loss bursts of length in the range [(k+1)r, (k+2)r-1], will all result in (r-a+1) packet loss events. Similarly, beacon loss bursts of length [(k+2)r, (k+3)r-1], will result in (2r-a+1) packet loss events, and so on. Hence, the packet loss event probability, is given by, UBrPStPLEP,,0 11111111111112,,0aprppamrpppamrppPLEPrbbrkbmrbbrmkbmrmkrmkbbLBrPSt 12 r 12 r 12 r......... L Schedule # 0 Schedule # 1 Schedule # k 12 r a...... Schedule # k+1Time BeaconPeriod Figure 8-5: Illustration of relationship between latency and persistent preview schedule Analysis of packet loss event probability for the case a burst of beacon losses starting with a beacon that is t beacon periods (with t [1 r-1]) from the beacon in which the corresponding schedule was first published, is similar to the case with t = 0, expect that additions beacons have to be missed before the first PLE occurs. Thus, the packet loss event probability, is given by, LBrPSrtPLEP,,]1,1[

PAGE 122

108 1112,,]1,1[aprppPLEPrbbtrkbLBrPSrt Hence the average packet loss event probability, LBrPS P LEP,, is given by, rappPLEPPLEPrPLEPrPLEPrbrkbrtLBrPStLBrPStrtLBrPStLBrPS11111111,,0,,010,,,, 8.5.4 Persistent current and preview schedules The lower bound for r-persistent current and preview schedules can be determined in a manner similar to that of the r-persistent current schedules. For a burst of beacon losses starting with a beacon in which a new schedule was first published, a minimum of (r+L) beacons has to be missed to cause the first packet loss event. When consecutive beacons are missed with, rL a total of 1 rL packet loss events occur. Thus, the packet loss event probability, is given by, LBrPSCStPLEP,,,0 rLbrLbbLBrPSCStprLppPLEP112,,,0 For a burst of beacon losses starting with a beacon that is t beacon periods from the beacon in which the corresponding persistent schedule was first published, (r-t) additional beacons have to be missed to incur similar number of beacon loss events. Hence, the packet loss event probability, is given by LBrPSCSrtPLEP,,,]1,1[ trLbtrLbbLBrPSCSrtptrLppPLEP222,,,]1,1[121 Thus, the average PLEP of r-persistent current and preview schedule, LBrPSCS P LEP,,, is,

PAGE 123

109 brbrLbrttrLbrLbrtLBrPSCStLBrPSCSprpppprPLEPrPLEP111111210,,,,,, 8.5.5 Lower Bound comparison Some interesting properties of the scheduling strategies can be observed by looking at the lower bound. For non-persistence schedules, packet loss event probabilities better than (p b ) L cannot be guarantees. This indicates the degradation in QoS when persistent schedules are not used. For persistent current schedules, the packet loss event probability is primarily effected by ((p b ) L /r). This shows that significantly large values of r should be used to reduce PLEP. Thus, at least from an upper bound perceptive, we can expect no significant improvement in the level of guarantees on QoS, compared to the non-persistent schedules. Packet loss event for persistent preview schedules shows non-linear behavior with L for a given r. Large improvement in performance can be expected when L = kr + 1. However, when L = kr persistent preview schedules does not provide any improvements in upper bound. For persistent current and preview schedules, the packet loss event probability is primarily effected by ((p b ) [L+r] /r). Thus from a PLEP point of view, r-persistent current and preview schedules effectively increase the latency of the stream by r, compared to that of the non-persistent schedule. The goodness of the upper bound will be studies in subsequent sections, by comparing the result obtained in this section to the ones obtained using Markov chain based analysis. 8.6 Markov Chain Analyses In the following sections we use Markov chain based analysis to investigate the effect of beacon loss on a system using various persistent scheduling strategies. Our goal is to obtain the dependency of the packet loss event probability, PLEP, on {N, L, p b and

PAGE 124

110 X}. Note that X is a measure of inefficiency incurred at the MAC layer to overcome beacon loss. Since the minimum value of X needed to overcome any beacon loss is 1, in the subsequent analysis we refer to X=1 as the optimal over allocation. 8.6.1 Non-persistent beacon schedules In this section we investigate the effect of beacon loss on a system that uses non-persistent schedules. Since non-persistent schedules are valid only in the period in which they are published, a station that misses a beacon cannot transmit in the corresponding beacon period. First, consider the buffer backlog of the stream at the end of each beacon period. It can be readily seen that the variation of the buffer backlog is only dependent on the reception status of the beacon in the corresponding beacon period. Furthermore, the maximum buffer backlog is bounded by N(L-1). Hence a discrete-time, finite state Markov chain model can be used to determine the steady state probabilities of buffer backlog from which the PLEP can be computed. Figure 8-6 shows the state transition diagram of buffer backlog when non-persistent (NP) beacon schedules are used with L > 2 and X < N. The state transition probabilities, is given by NPjip, 0,max,01,min,0,XijNNLipNNLNijNNLippbbNPji The steady state probabilities, can be obtained from the state transition probabilities by solving the equation, NPiP NPiNPiNPjiPPp,

PAGE 125

111 0 NL -2N-1 X X+1 N N+1 1-pb 1 1-pb 1-pb 1-pb pbpb pbpb pb pb 1-pb 1-pb NL-2N NL-N-1 NL-N pb NL-2N+1 pbpbpb 1-pb 1-pb 1-pb 1-pb 1-pb pb Figure 8-6: State transition diagram for non-persistence beacon schedules The packet loss event probability ( NPPLE P ) and average packets lost per packet loss event () can be obtained from the steady state probabilities () as follows, NPAPL NPiP NNLNNLibNPiNPpPPLEP12 NPNNLNNLibNPiNPPLE P NNLipPAPL122 One important point to note is that not all buffer backlog states are possible (i.e., the Markov chain is not irreducible). For example, if X and N are even, odd backlog states never occur. Another interesting property that can be observed from the state transition diagram is the effect of scaling N and X by an integer value. This will result in a scaled state transition diagram with the same transition probability between scaled states. This suggests that the ratio NX primarily determines the NP P LE P 8.6.2 Persistent current schedules The performance of persistent current schedules can be carried out in a manner similar to that of non-persistent schedules except with one important difference. Instead

PAGE 126

112 of tracking the buffer backlog at the end of each beacon period, we track the buffer backlog at the end of the last beacon period in which a published schedule remains valid. This choice will eliminate the dependency of buffer backlog states on the schedule persistent, thus simplifying the analysis. Note that Markov chain analysis can also be carried out by tracking the buffer backlog at the end of each beacon period. The disadvantage of the later approach is that the number of states will increase by r times. Irrespective of the approach used, the exact relationship between PLEP, N, L, p b and X can be obtained. Figure 8-7 shows the state transition diagram of buffer backlog when 2-persistent current schedules are used with L > 2 and X < N. A generalized state transition probability for r-persistent current schedule, can be computed by first initializing to zeros and following the procedure given in the following equation, rCSjip,, rCSjip,, NNLNrijNNLipptrXNNLNtijNNLippprXNNLNijNNLipppXrijNNLippprbrCSjibtbrCSjibbrCSjibrCSjirCSji,min,0...0,,minmax,01...0,1,minmax,010,max,01,,,,,,,,,, In the above equation, the first term corresponds to the transition when the first beacon carrying a new schedule is properly received. Subsequent terms correspond to the transition where the first t beacons carrying a new schedule are not properly received. The last term corresponds to the case where all beacons carrying a new schedule are lost. It is important to note that more than one of the above conditions can result in the same

PAGE 127

113 final state. In such cases, the transition probability will be the sum of all events that result in the same final state, as shown in the above equation. Figure 8-7: State transition diagram for 2-persistence current schedules The packet loss event probability, rCSPLE P , can be obtained from the steady state probabilities, as follows, rCSiP, NNLirCSirCSirCSPLTPPPLEP0,,, Where, packet loss transition probability from state i, is given by, rCSiPLTP, rNNNLNriceilprNNNLNticeilppPLTPrbrtbtbrCSi0,max0,max111, The average PBs lost per packet loss event, is given by, rCSAPL,

PAGE 128

114 rCSNNLirCSirCSirCSPLEPrAPLPAPL,0,,, Where, average number of packets lost during transitions from state i, is given by, rCSiAPL, 0,max0,max110,NNLNripNNLNtippAPLrbrtbtbrCSi 8.6.3 Persistent preview schedules The performance of persistent preview schedule can be carried out in a manner similar to that of persistent current schedule, by tracking the buffer backlog at the end of the last beacon period in which a published schedule remains valid. However, since there is latency incurred from the time a new schedule is published to the time it becomes effect, each state in the Markov chain should incorporate three state variables, 1. Buffer backlog, 2. The presence or absence of over allocation in the current schedule (represented by C and NC respectively), and 3. The present or absence of over allocation in the schedule following the current schedule i.e., the immediate future schedule (represented by F and NF respectively). Figure 8-8 illustrates the changes in Markov chain states for the case of 2-persistent preview schedule with an over allocation of N packets (i.e., X = N). In this example, it is assumed that beacons carrying schedule A2 are lost. At the end of beacon periods in which schedule A2 becomes effective, the backlog increases to 2N. At this time CCo publishes schedule A4 with N packets of over allocation (since inferred backlog is greater than zero). Due to the latency incurred for a published schedule to become effective, the backlog at the end of beacon periods in which schedule A3 becomes effective remains at

PAGE 129

115 2N. This in turn causes the CCo to publish a schedule A5 with N packets of over allocation. The backlog becomes zero at the end of beacon periods in which schedule A4 becomes effective, however over allocation is provided until the end of schedule A5. Note that over allocation can be chosen to not provide over allocation in Schedule A4. For simplicity, we do not consider such a scheme in our study. Schedule A1Published Schedule A1EffectiveSchedule A2Published Schedule A2EffectiveSchedule A3Published Schedule A0Effective Backlog = 0NC, NF Backlog = 0NC, NF Schedule A3EffectiveSchedule A4Published Backlog = 2NNC, F Schedule A4EffectiveSchedule A5Published Backlog = 2NC, F Backlog = 0C, NF time Schedule A5EffectiveSchedule A6Published Backlog = 0NC, NF MarkovChainStates Figure 8-8: Illustration of state changes in Markov chain for 2-persistent preview schedule The number of states in the Markov chain can be reduced by noting that, 1. Backlog is always going to be zero in a state where there is no over allocation in the current and immediate future schedules (i.e., [NC, NF]), 2. Backlog is always going to be zero in a state where there is over allocation in the current schedule and no over allocation in the immediate future schedule (i.e., [C, NF]), 3. Backlog is always going to be min(Nr, N(L-1)), in a state where there is no over allocation in the current schedule, but there is over allocation in the future schedule (i.e., [NC, F]) Figure 8-9 shows the state transition diagram of buffer backlog when 2-persistent preview schedules are used with L > 2 and X < N.

PAGE 130

116 Figure 8-9: State transition diagram for 2-persistence preview schedules

PAGE 131

117 A generalized state transition probability for r-persistent preview schedules, can be computed using the follow equations, Transition form state with no current and no future over allocation, rbrbpFNCrNNFNCpNFNCNFNC],,'[],,,0[Pr1],,0[],,,0[Pr Where, 1,min'Lrr Transition form state with no current over allocation, but over allocation is provided in the future schedule 1,min',],,''[],,,'[Pr1],,'[],,,'[PrLrrwherepFNCrNFNCrNpFCrNFNCrNrbrb Where, and 1,min'Lrr 1,*2min''Lrr Transition from state with current over allocation, but no future over allocation rbrbpFNCrNNFCpNFNCNFC],,'[],,,0[Pr1],,0[],,,0[Pr where, 1,min'Lrr Transition from states with current and future over allocations 11;],,1,[min],,,[Pr11,min;1],,0,[max],,,[Pr1,min1;1],,0[],,,[PrLNipFNCLNrNiFCiLNiLNrXpFNCrXiFCiLNrXipNFCFCirbrbrb The packet loss event probability, rPSPLE P , can be obtained from the steady state probabilities, as follows, rPSyxiP,],,[ NNLirbrPSFCirPSprNNNLNriceilPPLEP1,],,[,0,max

PAGE 132

118 rbrPSFNCLNNrrbrPSNFCrbrPSNFNCprNNNLNrLNNrceilPprNNNLNrceilPprNNNLNrceilP0,1,minmax0,max0,max,],,1,[min,],,0[,],,0[ The average PBs lost per packet loss event is given by, rPSrPSrPSPLEPrTPLAPL,,, Where is given by, rPSTPL, rbrPSFNCLNNrrbrPSNFCrbrPSNFNCNNLirbrPSFCirPSpNNLNrLNNrPpNNLNrPpNNLNrPpNNLNriPTPL0,1,minmax0,max0,max0,max,],,1,[min,],,0[,],,0[1,],,[, 8.6.4 Persistent current and preview schedules The performance of persistent current and preview schedules can be carried out in a manner similar to that of the persistent preview schedules, with the addition of one additional state variable to the Markov chain. This variable tracks whether at least one beacon was successfully received in the beacon periods where the previous schedule was published. Note that if at least one of the beacons is properly received (represented by G), then the schedule is known and all allocations can be successfully utilized. If none

PAGE 133

119 of the beacons were properly received (represented by AB) the utilization will depend on when the current schedule will be received (based on current schedule). Thus, eight different states are possible for the same value of backlog, 1. No Current, No Future, At least one good (NC, NF, 1G) 2. No Current, No Future, All Bad (NC, NF, AB) 3. Current, No Future, At least one good (C, NF, 1G) 4. Current, No Future, All bad (C, NF, AB) 5. No Current, Future, At least one good (C, NF, 1G) 6. No Current, Future, All Bad (C, NF, AB) 7. Current, Future, At least one good (C, F, 1G) 8. Current, Future, All Bad (C, F, AB) Reduction in the number of states in the Markov can be done by noting that, 1. Backlog will always be zero in (NC, NF) 2. Backlog will always be zero in (C, NF) 3. Backlog will be limited to 1,min,...,2, LNNrNN when in (NC, F, 1G) 4. Backlog will always be 1,min LNNr when (NC, F, AB) Figure 8-10 shows the state transition diagram of buffer backlog when 2-persistent current and preview schedules are used with L > 2 and X < N. A generalized state transition probability for r-persistent preview schedules, can be computed by first initializing all transition probabilities to zero and following the procedure, Transition probabilities from (NC, NF, 1G) rbrbpABNFNCGNFNCpGNFNCGNFNC],,,0[],1,,,0[Pr1]1,,,0[],1,,,0[Pr Transition probabilities from (NC, NF, AB) rbrbkbkbpABNFNCjABNFNCrkforppGNFNCjABNFNCpGNFNCABNFNC ],,,[],,,,0[Pr11;1]1,,,[],,,,0[Pr1]1,,,0[],,,,0[Pr Where, 1,min LNNmjm

PAGE 134

0 0 1-(pb)2 (pb)21-pb NC,NF,1G 0 NC,NF,AB N pb(1-pb) 2N NC,F,1G NC,F,AB C,NF,1G 0 1-(pb)2 (pb)2 1-pb pb(1-pb) (pb)2 C,NF,AB 1 2X N N-2X 2N-X 2N 3N NL-N 1 2X N N-2X 2N 3N NL-N 4N 1-(pb)2(pb)21-(pb)2(pb)21-pb1-pbpb(1-pb) (pb)2 pb(1-pb) (pb)21-(pb)21-pbpb(1-pb)(pb)2 1-(pb)2 1-(pb)2 1-(pb)2 1-(pb)2 1-(pb)2 (pb)2 (pb)2 (pb)2 (pb)2 (pb)2 pb(1-pb) 1-pb (pb)2(pb)2(pb)2 (pb)2pb(1-pb) pb(1-pb) pb(1-pb) (pb)2 (pb)2 1-pb 1-pb 1-pb 1-pb C,F,1G C,F,AB(pb)21-(pb)21-pbpb(1-pb)(pb)2(pb)2 120 Figure 8-10: State transition diagram for 2-persistence current and preview schedules

PAGE 135

121 Transition probabilities from (NC, F, 1G) 1,min1;],,,[],1,,,[Pr1,min1;1]1,,,[],1,,,[PrLrkforpABFCjGFNCjLrkforpGFCjGFNCjrbkkrbkk Transition probabilities from (NC, F, AB) rbrbkbkbpABFCjABFNCLNNrrkforppGFCjABFNCLNNrpGFCLNNrABFNCLNNr ],,,[],,,,1,[minPr11;1]1,,,[],,,,1,[minPr1]1,,,1,[min],,,,1,[minPr Where, 1,1,minmin LNNmLNNrjm Transition probabilities from (C, NF, 1G) rbrbpABNFNCGNFCpGNFNCGNFC],,,0[],1,,,0[Pr1]1,,,0[],1,,,0[Pr Transition probabilities from (C, NF, AB) rbrbkbkbpABNFNCjABNFCrkforppGNFNCjABNFCpGNFNCABNFC ],,,[],,,,0[Pr11;1]1,,,[],,,,0[Pr1]1,,,0[],,,,0[Pr Where, 1,min LNNmjm Transition probabilities from (C, F, 1G) 11,min;],,,0,[max],1,,,[Pr11,min;1]1,,,0,[max],1,,,[Pr1,min1;],,,0[],1,,,[Pr1,min1;1]1,,,0[],1,,,[PrLNiLNrXpABFCrXiGFCiLNiLNrXpGFCrXiGFCiLNrXipABNFCGFCiLNrXipGNFCGFCirbrbrbrb Transition probabilities from (C, F, AB) 11;],,,[],,,,0[Pr1111;1]1,,,[],,,,[Pr1,min1;1]1,,,0,[min],,,,[Pr1,min1;1]1,,,0[],,,,[Pr,,LNiforpABFCjABFCLNiandrkforppGFCjABFCiLNrXipGFCrXiABFCiLNrXipGNFCABFCirbirbkbikbb Where, 0,1,minmax,XkrLNkNijik

PAGE 136

122 The packet loss event probability, rPSCSPLE P ,, can be obtained from the steady state probabilities, as follows, rPSCSzyxiP,,],,,[ NNLirbrkbkbrPSABFCirPSCSprNNNLNriceilpprNNNLNkiceilPPLEP111,],,,[,,0,max10,max rbrkbkbrPSABNFNCprLrpprLkP0,1max10,1max11,],,,0[ rbrkbkbrPSABNFCprLrpprLkP0,1max10,1max11,],,,0[ rbrkbkbrPSFNCNrprLrrpprLkrP0,1'max10,1'max11,],,'[ Where, The average PBs lost per packet loss event is given by, 1,min'Lrr rPSCSrPSCSrPSCSPLEPrTPLAPL,,,,,, Where (total PBs lost) is given by, rPSCSTPL,, NNLirbrkbkbrPSABFCirPSCSpNNLNrippNNLNkiPTPL111,],,,[,,0,max10,max

PAGE 137

123 rbrkbkbrPSFNCNrrbrkbkbrPSABNFCrbrkbkbrPSABNFNCpNLNrNNrppNLNkNNrPpNLNrNppNLNkNPpNLNrNppNLNkNP0,'max10,'max0,max10,max0,max10,max11,],,'[11,],,,0[11,],,,0[ 8.7 Comparison of Persistent Scheduling Mechanism In this section we compare the performance of various scheduling strategies under different traffic conditions (i.e., N, L, PLEP) and channel conditions (i.e., p b ). The first results of interest is to compare the results from the lower bound analysis in Section 6, with the corresponding results obtained from the Markov chain analysis. We do this by comparing the lower bound on the latency that can be supported for given PLEP and p b It is important to note that for Markov chain analysis with over allocation set to N(L-1), the state transition diagram does not depend on the choice of N. Hence these results are independent of the application data rates (N). Figure 8-11 to Figure 8-14 show the minimum latency that can be supported by various scheduling strategies for p b in the range [0.001, 0.1], and for PLEP of 10 -5 and 10 -7 Note that PLEP of 10 -5 and 10 -7 correspond to packet loss events that occur approximately once an hour and once in 100 hours. It is reasonable to assume that most applications will require packet loss event probability in this range. Figure 8-15 and Figure 8-16 compare the minimum latency that can be supported by scheduling strategies for p b in the range [0.001, 0.1], and for PLEP of 10 -5 and 10 -7 respectively. Figure 8-17 and Figure 8-18 show the schedule persistence required to support Latency of 2, for p b in

PAGE 138

124 the range [0.001, 0.1], and for PLEP of 10 -5 and 10 -7 respectively. These results can be summarized as follows, The results for lower bound obtained from analysis match very closely with those obtained based on the Markov chain analysis. The minor differences observed in case of persistent preview schedules can be attributed to the latency effects of our over-allocation strategy. As expected, the results show that non-persistent schedules limit the levels of QoS (in terms of L, PLEP) than can be guaranteed to applications. As can be see from Figure 8-15, latencies less than 4 cannot be supported when p b >1% at PLEP = 10 -5 Furthermore, latencies less than 6 cannot be supported when p b >1% at PLEP = 10 -7 These results justify the need for schedule persistence over power line channels. Current schedules does not show any significant improvement in the minimum latency supported even at large values of persistence. This is clearly evident in Figure 12, which shows that an 8-persistent current schedule will only reduces the minimum latency supported by at most 1, when the persistence is changed for 2 to 8. Further, as can be seen in Figure 8-15 and Figure 8-16, 8-persistent current schedule does not provide any significant improvement in minimum latency compared to that of non-persistent schedules. While in theory, improvements in performance can be achieved by using large values of persistence for the current schedule; large persistences are not practical as they limit dynamic adjustment of schedules with channel and traffic changes. So, it is safe to conclude that in practice current schedules does not enable significant improvements in QoS compared to non-persistent schedules. In contrast to current schedules, preview schedules enable significant improvements in QoS with modest values of persistence. For example, a persistence of 4 should be sufficient to enable L=2 for up to 6% beacon loss probability at PLEP = 10 -5 (Figure 8-15). A persistence of 6 should be sufficient to enable L=2 for up to 7% beacon loss probability at PLEP = 10 -7 (Figure 8-16). Current and preview schedules provide the highest performance. From Figure 8-15 it can be see that a persistence of 2 is sufficient to handle up to 6% beacon loss probability at PLEP = 10 -5 A persistence of 4 should be sufficient to handle up to 8% beacon loss probability at PLEP = 10 -5 (Figure 8-16). For practical systems, apart from the QoS that can be supported the amount of over allocation that is needed is also important. The over allocation represents the amount of bandwidth the needs to be set aside for enabling speedy reduction of backlog. Hence over allocation will effectively reduce the maximum amount of time that can be allocated for

PAGE 139

125 multimedia streams. Figure 8-19 to Figure 8-22 shows variation of over allocation with latency for various scheduling strategies for p b values of [0.001, 0.005, 0.01, 0.05, 0.1], N=100 and PLEP = 10 -7 These results show that, Non-Persistence schedules require large amounts of over allocation the support their minimum latencies. Furthermore, the reduction in the over allocation with latency is slow. Hence, it can be concluded that non-persistent schedules can result in large inefficiencies especially at beacon loss probabilities greater than 1%, Current schedules enable significantly faster reduction in over allocation compared to non-persistent schedules. So, although current schedules do not enable significant improvements in QoS, they do enable significant improvements in efficiencies at the QoS levels that they can support, Both preview schedules and current and preview schedules require minimum amount of over allocations at and near the minimum latencies that can be supported. Our analysis has also shown that the average number of packets lost per packet loss event is large. For non-persistent schedules and persistent current schedules, they are typically 30%-50% of the number of packet to be delivered in a beacon period (N). For persistent schedules, average packet loss tends to be 50% to100% of N. Since an over allocation of X=1 is the minimum required to deal with any beacon loss, we consider latency that can be supported with X=1 as the optimal latencies. In contrast to minimum latencies, optimal latencies depend on application data rates (i.e., N). Figure 8-23 to Figure 8-26 show the optimal latency that can be supported by scheduling strategies for p b in the range [0.001, 0.1], N = [50, 75, 100] and for PLEP of 10 -5 and 10 -7 Figure 8-27 and Figure 8-28 compare the optimal latency that can be supported by scheduling strategies for p b in the range [0.001, 0.1], N=100, and for PLEP of 10 -5 and 10 -7 respectively. Figure 8-29 and Figure 8-30 show the schedule persistence required to optimally support Latency of 2, for p b in the range [0.001, 0.1], N=100, and for PLEP of 10 -5 and 10 -7 respectively. These results can be summarized as follows,

PAGE 140

126 The optimal latency than can be supported decreases as that application data rate (N) increases. This can be expected, as the backlog resulting from a packet loss event is larger. The optimal latencies for non-persistent schedules are very large. For pb loss greater than 1%, latencies below 14 cannot be supported. Persistent current schedules provide significant improvement in the minimum latency compared to non-persistent schedules. For persistent preview schedules and persistent current and preview schedules, the optimal latency and minimum latency are the same for the data points we have tested. These results show that whenever there is sufficient space available in the beacon, persistent preview and current schedules should be used. In scenarios where there is no space available, implementations should use persistent preview schedules. The choice of persistence should be based on the QoS requirement of the application and the beacon loss probabilities experience by the stream. Implementations can use the analytical lower bound to determine the values of minimum value of persistence that needs to be used. Figure 8-11: Minimum latency required to achieve PLEP of {10 -5 10 -7 } at various beacon loss probabilities in a system with non-persistent schedules

PAGE 141

127 Figure 8-12: Minimum latency required to achieve PLEP of {10 -5 10 -7 } at various beacon loss probabilities in a system with r-persistent current schedules Figure 8-13: Minimum latency required to achieve PLEP of {10 -5 10 -7 } at various beacon loss probabilities in a system with r-persistent preview schedules

PAGE 142

128 Figure 8-14: Minimum latency required to achieve PLEP of {10 -5 10 -7 } at various beacon loss probabilities in a system with r-persistent current and preview schedules Figure 8-15: Comparison of minimum latency supported by various scheduling strategies at various beacon loss probabilities for PLEP = 10 -5

PAGE 143

129 Figure 8-16: Comparison of minimum latency supported by various scheduling strategies at various beacon loss probabilities for PLEP = 10 -7 Figure 8-17: Minimum schedule persistence required to support L = 2 for various scheduling strategies at PLEP = 10 -5

PAGE 144

130 Figure 8-18: Minimum schedule persistence required to support L = 2 for various scheduling strategies at PLEP = 10 -7 Figure 8-19: Over allocation required to support various latencies in a system with non-persistent beacon schedules, N = 100, PLEP = 10 -7

PAGE 145

131 Figure 8-20: Over allocation required to support various latencies in a system with 2-persistent current schedules, N = 100, r = 8,PLEP = 10 -7 Figure 8-21: Over allocation required to support various latencies in a system with 2-persistent preview schedules, N = 100, r = 4,PLEP = 10 -7

PAGE 146

132 Figure 8-22: Over allocation required to support various latencies in a system with 2-persistent current and preview schedules, N = 100, r = 4,PLEP = 10 -7 Figure 8-23: Latencies required to support a stream with optimal over allocation for a system with non-persistent schedules for PLEP = {10 -5 10 -7 }

PAGE 147

133 Figure 8-24: Latencies required to support a stream with optimal over allocation for a system with r-persistent current Schedules for PLEP = {10 -5 10 -7 } Figure 8-25: Latencies required to support a stream with optimal over allocation for a system with r-persistent preview schedules for PLEP = {10 -5 10 -7 }

PAGE 148

134 Figure 8-26: Latencies required to support a stream with optimal over allocation for a system with r-persistent preview schedules for PLEP = {10 -5 10 -7 } Figure 8-27: Comparison of optimal latency supported by various scheduling strategies at various beacon loss probabilities for N = 100, PLEP = 10 -5

PAGE 149

135 Figure 8-28: Comparison of optimal latency supported by various scheduling strategies at various beacon loss probabilities for N = 100, PLEP = 10 -7 Figure 8-29: Minimum schedule persistence required to support L = 2 with X = 1 for various scheduling strategies at PLEP = 10 -5

PAGE 150

136 Figure 8-30: Minimum schedule persistence required to support L = 2 with X = 1 for various scheduling strategies at PLEP = 10 -7 8.8 Conclusions The new generation of home networking applications requires guaranteed QoS. The recently released HomePlug AV standard is designed to overcome the harsh PLC channel conditions and to provide guaranteed QoS. The harsh channel conditions on power line can result in beacon loss. In this section we investigated the effect of beacon loss on QoS and the enhancement that can be achieved by using persistent current schedules, persistent preview schedules, and persistent current and preview schedules. Our average efficiency study showed that all persistent schedules can significantly improve the efficiency due to missed transmission opportunities. A details study of the QoS enhancements using analytical lower bound and Markov chain analysis have shown that persistent current and preview schedules are highly effective in dealing with beacon

PAGE 151

137 loss. Results showed that a persistent of 4 should enable current and preview schedules to provide PLEP of 10 -7 while tolerating up to 8% beacon loss. Persistent preview schedules also provided significant performance improvements at slightly higher values of persistence compared to persistent current and preview schedules. Results also show that current schedules do not provide any significant improvements in QoS support. However, they do enable better efficiency at the latencies that can be supported. For both persistent preview schedules and persistent current and preview schedules, the analytical lower bound can be used to determine the optimal latencies with very low margin of error. We recommend implementation to use either persistent preview schedules or persistent current and preview schedules. Our analysis assumes constant bit rate traffic and no PB errors. Since allocation strategies to deal with such conditions will include over allocation, and since persistent preview schedules and persistent current and preview schedules are not much sensitive to the size of over allocation, if the over allocations are designed to be slightly more conservation, they should also enable them to deal with beacon loss.

PAGE 152

CHAPTER 9 CONCLUSIONS AND FUTURE WORK Most houses are not equipped with specialized wiring for networking purpose, and retrofitting them with new wiring is prohibitively expensive. Hence, the use of existing power line infrastructure for networking purpose can come a long way in solving the in-home connectivity problem. The challenges to enabling high-speed power line communications (PLC) are many. Power line channels are characterized by large attenuations, frequency selective fading and impulse noise. Often the channel also exhibits a temporal dependence with the underlying AC line cycle frequency. The regulatory environments, while still maturing, also place several restrictions on transmit power levels and frequency bands. One of the major contributions of this work is the evaluation of the viability of PLC based home networking. Extensive simulations and field testing has shown that HomePlug 1.0 based PLC systems can not only provide ubiquitous connectivity, but they also out performs IEEE 802.11b/a based wireless home networking solutions. One of the limitations of this study is the use of proto-type equipment for PLC and off the shelf equipment for wireless. Neither of these captures the true capabilities of these technologies due to implementation choices and limitations. Nonetheless, our evaluation definitively concludes that PLC is a viable option for home networking. At present more that 5 million HomePlug 1.0 devices have been deployed through out the world. This fact re-enforces our conclusions. 138

PAGE 153

139 Another major contribution of this work is the study of QoS design issues involved in enabling multimedia communications over power line channels. This study is motivated in part by the work we did in the development of HomePlug AV standard. As part of this, we investigate three issues that the new generation PLC systems need to overcome to enable multimedia communications: (a) AC line cycle variation in noise, (b) Impulse noise, and (c) Beacon Loss. One of the peculiarities of PLC channels is the cyclic variation of their characteristics with the AC line cycle phase. Exploiting this cyclic variation is crucial for not only obtaining the data rates required to support the multimedia applications, but also for enabling guarantees on QoS. Our channel characterization measurements and analysis have show that a 30% improvement in physical layer (PHY) data rates can be obtained by continuously adapting to, and synchronizing communications with, the AC line cycle. Since continuous adaptation with respect to time is unduly complex for practical systems, we propose dividing the AC line cycle into multiple regions and adapting independently in each region. Results show that using a region size of 1-2 milliseconds provides up to 10% improvements in MAC throughput. Significant work still remains in understanding the AC line cycle based variation of PLC channel characteristics and the mechanisms that can be used by practical systems to optimize the performance gains. This includes, A limitation of our study is the use on only one full AC line cycle duration of noise captures. Larger captures that can span multiple AC line cycles will enable more accurate measurements of channel capacities variations. Our study only takes into consideration the variation of noise with the phase of the AC line cycle. While changes in channel attenuation characteristics are relatively rare, they are definitely possible. Future studies should take this into account to accurately measure the channel capacity variations. In our study we propose dividing AC line cycle into fixed size macro slots for simplifying practical implementations. In general, an optimal channel adaptation

PAGE 154

140 scheme should be able to divide AC line cycle into variable number of regions based on the underlying channel characteristics. Algorithm for doing this is still an open research item. Proper handling of impulse noise events is another design aspect that is crucial for enabling multimedia communications over PLC channels. Impulse noise will produce temporal effects in channel capacity. Between two impulse noise events, the channel capacity can be very high, while during the impulse noise the capacity is very low. To obtain low FEC block error rates, the channel has to be adapted close to the lower end of its capacity (i.e., capacity near impulse noise event) and hence is undesirable. Physical layer adapting to the higher end of the channel capacity hinges on the ability to efficiently retransmit corrupted portions of the data at the MAC layer. One of the major contributions of this work is a novel 2-level MAC framing mechanism that enables efficiency retransmissions even under higher FEC block error rates. 2-level MAC framing enables retransmission of only the corrupt FEC block, thus providing optimal performance. Simulation and analysis based comparison with other viable approaches confirmed the superior performance of 2-level MAC framing. Apart from framing overheads, transmission overhead can also negatively impact the MAC efficiency on channels with high FEC Block Error rate due the need for a large number of retransmissions to successfully deliver packets. One of the contributions of this work is the study of the complex interaction between FEC block errors, transmission overhead, Quality of Service (QoS) requirements of a multimedia stream and its TDMA allocation requirements. We provide an approximate analytical estimate of the Fixed Allocation and Average Allocation requirements for multimedia streams. Our results show that, to support low latency applications at high FEC Block errors, the average allocation size has to be increased by one or more transmission overheads. Further, the

PAGE 155

141 increase in average allocation is not a strong function of application data rate. Hence, low data rate applications can suffer significant loss in efficiency at high FEC Block error rates. Our results also show that using a fixed size TDMA allocation can lead to poor utilization when the latency requirements of the application are less than four beacon periods. These results can be used by the channel adaptation algorithm to optimize the overall system capacity. Derivation of more accurate estimates of Fixed Allocation and Average Allocations is an area of future work in this topic. Ability to handle beacon loss is another design issue that needs to be properly handled to enable QoS over PLC channels. HomePlug AV uses robust physical layer parameters for transmission of beacons. However, due to the large dynamic range of attenuation and (impulse) noise over power line, beacon loss cannot be completely avoided. To overcome occasional beacon loss, HomePlug AV uses persistent beacon schedules. One of the major contributions of this dissertation is the evaluation of performance enhancements that can be achieved by using persistent beacon schedules. Detailed Markov Chain based analysis was used to show that using persistent current and preview schedule or persistent preview schedules, HomePlug AV systems can handle up to 10% beacon loss, with minimal impact on MAC efficiency and QoS guarantees. Schedule persistent limits the frequency with which schedules can be changed. Hence the minimum value of the persistence that enables the required QoS should be used by the Central Coordinator. To simplify practical implementations, we propose a lower bound based schedule persistence estimation mechanism that provides highly accurate estimates of the required schedule persistence under any beacon loss rate and application QoS requirements.

PAGE 156

142 Apart from the three design issues that we investigated in this dissertation, there are several other issues that need to be properly addressed by PLC based multimedia networks. These include, Hidden stations Due to the large attenuation and noise over power line channels, it is possible to have stations that are hidden with respect to each other. Hidden stations are relatively rare in PLC based home networks, but are definitely possible. Channel access mechanism that enable proper coordination of hidden stations and repeating/routing mechanism that enable communications between hidden devices are areas of future work. Neighbor Networks Power line transmissions from one house can interfere with transmissions from neighboring houses. Ability to handle such neighbor networks is one of the major hurdles for PLC. This problem is aggravated due to the absence of multiple frequency bands. Mechanisms for coordination between neighbor networks while achieving maximum spatial reuse is still an open research topic. Encryption and Privacy Since power line transmissions in one house can leak into neighboring houses; there is a need for mechanism that can be used for ensuring privacy of the data. Coexistence with Legacy device New generation PLC systems should be able to peacefully coexist with the widely deployed HomePlug 1.0 systems. Mechanisms that enable such coexistence is another area for future work.

PAGE 157

LIST OF REFERENCES 1. Philipps Holger: Performance Measurements of Power line Channels at High Frequencies, Proceedings of International Symposium on Power Line Communications and its Applications, Tokyo, Japan, 1998, pp. 229-237 2. Brown A. Paul: Some Key Factors Influencing Data Transmission Rates in the Power Line Environment when Utilizing Carrier Frequencies above 1MHz, Proceedings of International Symposium on Power Line Communications and its Applications, Tokyo, Japan, 1998, pp. 67-75 3. Halid Hrasnica, Abdelfatteh Haidine, Ralf Lehnert: Broadband Powerline Communications Networks, John Wiley & Sons, Chichester, England, 2004 4. Dosterr Klaus: Telecommunications Over the Power Distribution Grid: Possibilities and Limitations, Proceedings of the International Symposium on Power Line Communications and its Applications, Essen, Germany, 1997, pp. 1-9 5. Brown A. Paul: Power Line Communications Past Present and Future, Proceedings of International Symposium on Power-line Communications and its Applications, Lancaster, U.K., 1999, pp. 1-5 6. Brackmann Ludwig: Power line Applications with European Home Systems, Proceedings of International Symposium on Power Line Communications and its Applications, Essen, Germany, 1997, pp. 24-31 7. Kaizawa Yasuhito and Gen Marubayashi: Needs for the Power Line Communications, Proceedings of International Symposium on Power-line Communications and its Applications, Tokyo, Japan, 1998, pp. 153-157 8. John Newbury: Technical Developments in Power Line Communications, Proceedings of International Symposium on Power Line Communications and its Applications, Lancaster, U.K, 1999, pp. 66-78 9. John S. Barnes: A Physical Multi-Path Model for Power Distribution Network Propagation, Proceedings of International Symposium on Power Line Communications and its Applications, Tokyo, Japan, 1998, pp. 76-89 10. Yu-Ju Lin, Haniph A. Latchman, Minkyu Lee, Srinivas Katar: A Power Line Communication Network Infrastructure for The Smart Home, IEEE Wireless Communications, December 2002, Volume 9, Issue 6, pp. 104-111 143

PAGE 158

144 11. Minkyu Lee, Richard E. Newman, Haniph. A. Latchman, Srinivas Katar, Larry Yonge: HomePlug 1.0 Powerline Communication LANs Protocol Description and Performance Results, Special Issue of the International Journal on Communication Systems on Powerline Communications, April 2003, Volume 16, Issue 5, pp. 447-473 12. Yu-Ju Lin, Haniph A. Latchman, Richard E. Newman, Srinivas Katar: A Comparative Performance Study of Wireless and Power Line Networks, IEEE Communications Magazine, April 2003, Volume 41, Issue 4, pp. 54-63 13. IEEE 1394B-2002: IEEE Standard for a Higher Performance Serial Bus-Amendment 2, http://www.ieee.org/ April 2006 14. IEEE 1394-1995: IEEE Standard for a High Performance Serial Bus, http://www.ieee.org/ April 2006 15. IEEE 802.11, 1999: IEEE Standards for Information Technology Telecommunications and Information Exchange between Systems Local and Metropolitan Area Network Specific Requirements Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications, http://www.ieee.org/ April 2006 16. IEEE 802.11b, 1999: Supplement to 802.11-1999,Wireless LAN MAC and PHY Specifications: Higher Speed Physical Layer (PHY) Extension in the 2.4 GHz band, http://www.ieee.org/ April 2006 17. IEEE 802.11a, 1999: IEEE Standard for Information Technology -Telecommunications and Information Exchange between Systems Local and Metropolitan Area Networks Specific Requirements Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) SpecificationsAmendment 1: High Speed Physical Layer in the 5 GHz Band, http://www.ieee.org/ April 2006 18. IEEE 802.11g, 2003: IEEE Standard for Information technology Telecommunications and Information Exchange between Systems Local and Metropolitan Area Networks Specific Requirements Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications Amendment 4: Further Higher-Speed Physical Layer Extension in the 2.4 GHz Band, http://www.ieee.org/ April 2006 19. IEEE 802.15.1: IEEE Standard for Information Technology Telecommunications and Information Exchange between Systems Local and Metropolitan Area Networks Specific Requirements. Part 15.1: Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications for Wireless Personal Area Networks (WPANs(tm)) http://www.ieee.org/ April 2006

PAGE 159

145 20. IEEE 802.15.3: IEEE Standard for Information Technology Telecommunications and Information Exchange between Systems Local and Metropolitan Area Networks Specific Requirements Part 15.3: Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications for High Rate Wireless Personal Area Networks (WPAN) http://www.ieee.org/ April 2006 21. WiMedia Alliance: http://wimedia.org/ April 2006 22. UWB Forum: http://www.uwbforum.org/ April 2006 23. HomePNA Alliance: http://www.homepna.org/ April 2006 24. Multimedia over Coax Alliance: http://www.mocalliance.org/ April 2006 25. HomePlug Powerline Alliance: http://www.homeplug.org April 2006 26. Intellon Corporation: http://www.intellon.com April 2006 27. Shu Lin and Daniel J. Costello: Error Control Coding and Fundamentals and Applications, Prentice-Hall, New Jersey, 2004 28. Shannon C. E.: A Mathematical Theory of Communication, Bell System Technical Journal, 1948, Volume 27, pp. 379-423 29. Andrew S. Tanenbaum: Computer Networks, Prentice Hall, Fourth Edition, 2003 30. Prasad Ramjee and Richard Van New: OFDM wireless Multimedia Communications, Artech House, Boston, 2000 31. Thomas Starr, John M. Cioffi, and Peter J. Silverman: Understanding Digital Subscriber Line Technology, Prentice Hall, 1999 32. Baowei Ji, Archana Rao, Minkyu Lee, Haniph A. Latchman, Srinivas Katar: Multimedia in Home Networking, Proceedings of CITSA 2004 / ISAS 2004, Volume 1, pp. 397-404 33. Kaywan H. Afkhamie, Srinivas Katar, Larry Yonge, and Richard E. Newman: An Overview of the upcoming HomePlug AV Standard, Proceedings of the International Symposium on Power Line Communications and its Applications, Vancouver, British Columbia, 2005, pp. 400-404 34. HomePlug AV White Paper: http://www.homeplug.com/en/products/ whitepapers.asp April 2006 35. Srinivas Katar, Brent Mashburn, Kaywan Afkhamie, Haniph Latchman, Richard Newman: Channel Adaptation based on Cyclo-Stationary Noise Characteristics in PLC Systems, Proceedings of International Symposium on Power-line Communications and its Applications, Orlando, USA, 2006, pp. 17-21

PAGE 160

146 36. Osamu Ohno, Masaaki Katayama, Takaya Yamazato and Akira Ogawa: A Simple Model of Cyclo-Stationary Power-Line Noise for Communication Systems, Proceedings of International Symposium on Power-line Communications and its Applications, Tokyo, Japan, 1998, pp. 115-122 37. Masaaki Katayama, Saori Itou, Takaya Yamazato and Akira Ogawa: Modeling of Cyclo-Stationary and Frequency Dependent Power-Line Channels for Communications, Proceedings of International Symposium on Power-line Communications and its Applications, Limerick, Ireland, 2000, pp.123 38. Yuichi Hirayama, Hiraku Okada, Takaya Yamazato, Masaaki Katayama: Noise Analysis on Wide-Band PLC with High Sampling Rate and Long Observation Time, Proceedings of International Symposium on Power-line Communications and its Applications, Kyoto, Japan, 2003, pp. 142-147 39. Canete F. J., Cortes J. A., Diez L., Entrambassaguas J. T. and Carmona J. L.: Fundamentals of the Cyclic Short-Time Variation of Indoor Power-line Channels, Proceedings of International Symposium on Power-line Communications and its Applications, Vancouver, Canada, 2005, pp. 157-161 40. Cortes J. A., Canete F. J., and Entrambassaguas J.T.: Characterization of the Cyclic Short-Time Variation of Indoor Power-line Channels Response, Proceedings of International Symposium on Power-line Communications and its Applications, Vancouver Canada, 2005, pp. 326-330 41. John G. Proakis: Digital Communications, McGraw-Hill International, Third Edition, 1995 42. Srinivas Katar, Larry Yonge, Richard Newman and Haniph Latchman, Efficient Framing and ARQ for High-Speed PLC Systems, Proceedings of the IEEE International Symposium on Powerline Communication and Its Applications, Vancouver, Canada, 2005, pp. 27-31 43. Chan M. H. L., Donaldson R.W.: Amplitude, Width, and Interarrival Distributions for Noise Impulses on Intra-Building Power Line Communication Networks, IEEE Transactions on Electromagnetic Compatibility, 1989, Volume 31, Issue 3, pp. 320-323 44. Zimmermann Manfred, Dostert Klaus, "An Analaysis of the Broadband Noise Scenario in Powerline Network", Proceedings of the IEEE International Symposium on Powerline Communication and Its Applications, Limerick, Ireland, 2000, pp. 131-138 45. Leonard Kleinrock: Queueing Theory Volume 1:Theory, Wiley Interscience, First Edition, 1975

PAGE 161

147 46. Srinivas Katar, Manjunath Krishnam, Brent Mashburn, Kaywan Afkhamie, Richard Newman, Haniph Latchman: Beacon Schedule Persistence to Mitigate Beacon Loss in HomePlug AV Networks, Proceedings of International Symposium on Power-line Communications and its Applications, Orlando, USA, 2006, pp. 184-188

PAGE 162

BIOGRAPHICAL SKETCH Srinivas Katar was born on May 19, 1975, in Vadluru, Andhra Pradesh, India. He obtained his Bachelor of Technology degree from the Department of Electrical Engineering from the Indian Institute of Technology, Kanpur, India, in 1998. In May 2000, he received his masters degree from University of Florida. He is currently a Principal Research and Development Engineer at Intellon Corporation. During his tenure at Intellon, Srinivas Katar was instrumental in the development and standardization of HomePlug AV medium access control layer. 148


Permanent Link: http://ufdc.ufl.edu/UFE0013881/00001

Material Information

Title: Quality of Service Design Issues in Multimedia Communications over Power Line 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: UFE0013881:00001

Permanent Link: http://ufdc.ufl.edu/UFE0013881/00001

Material Information

Title: Quality of Service Design Issues in Multimedia Communications over Power Line 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: UFE0013881:00001


This item has the following downloads:


Full Text












QUALITY OF SERVICE DESIGN ISSUES IN MULTIMEDIA COMMUNICATIONS
OVER POWER LINE NETWORKS















By

SRINIVAS KATAR


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


2006

































Copyright 2006

by

Srinivas Katar


































I dedicate this work to my parents,
Katari Lakshmipathi Raju and Krishna Kumari















ACKNOWLEDGMENTS

I would like to thank my parents for their support and encouragement over all these

years. None of this work would have possible without them. I would also like to express

my heartfelt gratitude to my wife Keerti for her support and understanding throughout

this endeavor.

I would especially like to thank Larry Yonge and Intellon Corporation for allowing

me to work part time on my PhD. I would like to thank my colleagues, Brent Mashburn,

Kaywan Afkhamie and Manjunath Krishnam for their valuable feedback. Working with

them has always been a pleasure.

Finally, I would like to thank my advisors, Dr Haniph Latchman and Dr Richard

Newman, for their valuable guidance throughout my studies. I would also thank Dr Oscar

Crisalle and Dr Tan Wong for being part of my PhD supervisory committee.
















TABLE OF CONTENTS

page

A C K N O W L E D G M E N T S ......... ................................................................................... iv

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

LIST OF FIGURES ......... ....... .................... .......... ....... ............ ix

A B S T R A C T ......... .................................. ................................................... x iii

CHAPTER

1 IN TRODU CTION ...................... .................. .. .. ............ ....... 1

2 BACKGROUND .............................. .. ......... ......... ..........

2.1 Low B andw idth PLC ............................................................................9
2 .1.1 R regulatory constraints ............................................ ....................9
2.1.2 Technologies .................. ................. ............... ...... ........... .... .11
2.1.2.1 X-10 ................... .... ...................... 11
2.1.2.2 C E B us .......................................................... ..... .... 11
2.1.2.3 L onW orks ...... ........ ........ ........................ ........... 12
2.2 H igh B andw idth PL C ............................................................................ ....12
2.2.1 R regulatory constraints ............................ ................................. 13
2.2.2 Channel characteristics ....................................... ............... 13
2.2.2.1 Attenuation characteristics..................... ............... 13
2.2.2.2 N oise characteristics ................................. ............... 15
2.2.3 Technologies .................. ................. ............... ...... ............. ...18

3 HOMEPLUG 1.0 TECHNOLOGY DESCRIPTION AND PERFORMANCE
EVALUATION ......... .. ...... .............................. .... .......................... 19

3.1 H om e N etw working Solutions.................................... ........................ 20
3.1.1 New wire solution ...... ................. ...............20
3.1.2 W wireless solution ........................................ .......... .. ...... .. 2 1
3.1.3 N o new w ire solutions ............................ ................................. 22
3.2 H om ePlug 1.0 Technology ........................................ ...................... 23
3.2.1 Hom ePlug Physical layer.................................. ..................... 23
3.2.2 H om ePlug 1.0 M A C ............................................. ............... 25
3.2.2.1 Fram e form ats ....... .................................... ... .............25


v









3.2.2.2 Channel access mechanism............... ...............26
3.2.2.3 Segmentation and reassembly mechanism.........................30
3.3 HomePlug 1.0 Performance based on Simulations...................................31
3.4 HomePlug 1.0 Field Test Results and Comparison with IEEE 802.1 1b/a.32
3.4.1 Experiment method ..... .................... ...............33
3.4.2 Results34
3.4.2.1 IEEE 802.1 la indoor performance ...............................34
3.4.2.2 IEEE 802.1 lb and HomePlug 1.0......................................37

4 POWER LINE MULTIMEDIA NETWORKING ....................... ................41

5 CHANNEL ADAPTATION BASED ON CYCLO-STATIONARY NOISE
C H A R A C T E R IST IC S ................................................................... ..................44

5 .1 Introdu action ................................................................................. 4 5
5.2 Channel M easurem ents................................................... ................. 46
5.3 Channel A adaptation ............................................................................48
5.4 Computation of Channel Capacity..........................................................49
5.5 D iscu ssion of R esults....................................................................... ... ...5 1
5.6 C conclusions .............................................................................. 53

6 M A C FRAM IN G ............................................................ ................... 56

6.1 MAC Framing and Framing Efficiency................. ............................57
6.2 M AC Framing Strategies ................................................. ............... 58
6.2.1 Sim ple concatenation.............. ............................................... 59
6.2.2 Concatenation with explicit demarcation...........................60
6.2.3 2-L evel fram ing ........................................ ......... ............... 61
6.3 E efficiency A nalysis......................................................... ............... 63
6.3.1 Simple concatenation.............. .............................................. 63
6.3.2 Concatenation with explicit demarcation...........................64
6.3.3 2-level fram ing........... ....... .. ....... .... .... .................64
6.4 Comparison of Performance between various Framing Schemes ............68
6.5 C conclusions ........... ......... ......... ............ ............... ............... 69

7 ALLOCATION REQUIREMENTS FOR SUPPORTING LATENCY BOUND
T R A F F IC ...............................................................................................................7 0

7.1 Introduction ...................................... ............... .......... .............. 71
7.2 Allocation Requirements without Transmission Overheads....................76
7.2.1 Fixed A location, L = 1 ...................................... ............... 76
7.2.2 Fixed A location, L > 1 ............... ....... ............ ............... .... 77
7.3 Allocation Requirements with Transmission Overheads.........................80
7.3.1 Fixed A llocation......... .................................. .. .... ........ .....8 1
7.3.2 A average A location ................................................... ................ 86
7.4 Effect of Transmission overheads on Allocation Requirements...............89
7.5 C conclusions .............................................................................. 92









8 BEACON SCHEDULE PERSISTENCE TO MITIGATE BEACON LOSS .......94

8 .1 In tro d u ctio n ................... .......................................... ..................... 9 4
8.2 Persistent Beacon Schedules.................................. ........................ 97
8.3 E efficiency A nalysis.............................................. ......... ............... 100
8.3.1 N on-persistent schedule ............. .......................... .................100
8.3.2 Persistent current schedules ................... .... .............. .................... 100
8.3.3 Persistent preview schedules............................... .............101
8.3.4 Persistent current and preview schedules ..................................101
8.3.5 Efficiency com prison ......................................... .................102
8.4 Q oS C considerations ........................................................................... 103
8.5 Lower Bound on PLEP ......................... .......................104
8.5.1 N on-persistent schedules .................................. ............... 105
8.5.2 Persistent current schedules ....... ........ ..................... ............... 105
8.5.3 Persistent preview schedules............................... .............106
8.5.4 Persistent current and preview schedules ..................................108
8.5.5 Lower Bound comparison..............................................109
8.6 M arkov Chain Analyses........................... .... ........................ 109
8.6.1 N on-persistent beacon schedules................................................110
8.6.2 Persistent current schedules ............. .................. .................... 111
8.6.3 Persistent preview schedules ............... ................... .............. 114
8.6.4 Persistent current and preview schedules ..................................118
8.7 Comparison of Persistent Scheduling Mechanism ............................... 123
8.8 Conclusions ................................... .......................... ................. 136

9 CONCLUSIONS AND FUTURE WORK ................ .................................... 138

L IST O F R E F E R E N C E S ...................................................................... ..................... 143

BIOGRAPHICAL SKETCH ........................................................... ........148
















LIST OF TABLES

Table page

3-1 Fram e Control inform ation fields....................................... .......................... 28

3-2 H om ePlug 1.0 sim ulation results ........................................ ........................ 32

4-1 Bandwidth and QoS requirements for multimedia applications ...........................41

6-1 M A C fram ing overheads...................... .... .................. ................... ............... 62













LIST OF FIGURES

Figure page

1-1 FCC and CENELEC frequency band allocation................................ ................10

2-1 Pow er line channel im pulse response................................... ........................ 14

2-2 Channel frequency response.......................................................... ............... 15

2-3 N oise generated by light dim m er ........................................................ ............... 16

2-4 Noise generated by yard light........................ ................... .. ........... .... 16

2-5 N oise generated by halogen lam p ........................................ ........................ 17

2-6 N oise generated by hair dryer ............................................................................ 17

3-1 L ong F ram e form at............ ... ............................................................ ....... ............... 27

3-2 Short F ram e form at ......................................................................... ................... 27

3-3 Priority resolution and backoff scheme ............ ..............................................29

3-4 IEEE 802.1 1a indoor performance ............................ ......................35

3-5 HomePlug 1.0 and IEEE 802.1 lb indoor performance........................... .........36

5-1 Channel m easurem ent setup......................................................... ............... 47

5-2 Performance gain in PHY data rates as a function of macroslot size ....................54

5-3 Performance gain in MAC data rates as a function of macroslot size ..................54

5-4 Average gain in M AC and PHY data rates ................................... .................55

6-1 M A C fram ing Process ..... .. .................. ......................................... ............... 58

6-2 Sim ple con caten action .................................................................... .....................60

6-3 Concatenation with explicit demarcation .......................................... ............61

6-4 2-Level M A C fram ing..........................................................................62

6-5 Simple concatenation- efficiency with varying FEC block rrror rate and N = 25 ...65









6-6 Simple concatenation efficiency with varying number of FEC blocks and p = 0.1.65

6-7 Explicit demarcation based concatenation efficiency with varying FEC block
error rate and N = 25 ............ ........................ .. ......................................66

6-8 Explicit demarcation based concatenation efficiency with varying number of
FE C blocks and p = 0.1 .......................... ........................ ........... .............. 66

6-9 2-Level framing efficiency with varying FEC block error rate and N= 25 ..........67

6-10 2-Level framing efficiency with varying number of FEC blocks andp = 0.1 .......67

6-11 Performance comparison of various MAC framing strategies with varying FEC
block error rate, N = 25, Lm sdu = 1518 ................ ............................................. 69

7-1 Transmission overheads in HomePlug AV ........................ .................74

7-2 Illustration of delayed retransmission strategy...... ........ .. .................... ............... 74

7-3 State transition diagram for no MPDU overhead case .......................... ..........78

7-4 Excess allocation as a function of FEC block error probability when no
transm mission overhead present ........................................ ........................... 79

7-5 Average packet loss per packet loss event when no transmission overheads are
p re sen t .............................................................................. 8 0

7-6 Transmission opportunities required to achieve PLEP = 10-5 at various FEC
block error probabilities ........................ ......... ..... ... ............... 82

7-7 Fixed Allocation estimation error for N = 100 and L = 1, 2, 3, 4, 6, and 8 .............84

7-8 Probability distribution function of the Fixed Allocation estimation error.............. 85

7-9 Average packet loss per packet loss event when transmission overheads are
p re sen t .............................................................................. 8 6

7-10 Average Allocation estimation error for N = 100 and L = 1, 2, 3, 4, 6 and 8..........88

7-11 Probability distribution function of the Average Allocation estimation error .........89

7-12 Percentage difference in Fixed Allocation and Optimal Allocation for N = [50,
200], p= [0.0 1, 0.10] .......... ......... ......... ............ ............... . ........... 9 1

7-13 Percentage difference in Average Allocation and Optimal Allocation for N =
[50, 200], p= [0.01, 0.10] ...................... ..... ........ ................ 91

7-14 Effective transmission overhead for Average Allocations with N = 50 and 200.....92









8-1 Example of 2-persistent current schedules.................................... ............... 99

8-2 Example of 2-persistence preview schedules................................... ... ..................99

8-3 Example of 2-persistent current and preview schedules ............. ................100

8-4 Efficiency loss comparison for various scheduling strategies.............................. 102

8-5 Illustration of relationship between latency and persistent preview schedule .......107

8-6 State transition diagram for non-persistence beacon schedules ............................111

8-7 State transition diagram for 2-persistence current schedules ..............................113

8-8 Illustration of state changes in Markov chain for 2-persistent preview schedule..115

8-9 State transition diagram for 2-persistence preview schedules .............................116

8-10 State transition diagram for 2-persistence current and preview schedules ............120

8-11 Minimum latency required to achieve PLEP of { 105, 10-7} at various beacon
loss probabilities in a system with non-persistent schedules ..............................126

8-12 Minimum latency required to achieve PLEP of { 105, 10-7} at various beacon
loss probabilities in a system with r-persistent current schedules........................127

8-13 Minimum latency required to achieve PLEP of { 105, 10-7} at various beacon
loss probabilities in a system with r-persistent preview schedules ...................127

8-14 Minimum latency required to achieve PLEP of { 105, 10-7} at various beacon
loss probabilities in a system with r-persistent current and preview schedules.....128

8-15 Comparison of minimum latency supported by various scheduling strategies at
various beacon loss probabilities for PLEP = 105 ............ ................................. 128

8-16 Comparison of minimum latency supported by various scheduling strategies at
various beacon loss probabilities for PLEP = 107 ............ ................................. 129

8-17 Minimum schedule persistence required to support L = 2 for various scheduling
strategies at PLEP = 10-5 .................................................................... 129

8-18 Minimum schedule persistence required to support L = 2 for various scheduling
strategies at PLEP = 10-7 .................................................................... 130

8-19 Over allocation required to support various latencies in a system with non-
persistent beacon schedules, N = 100, PLEP = 10 .......... ................................130

8-20 Over allocation required to support various latencies in a system with 2-
persistent current schedules, N = 100, r = 8,PLEP = 107 ..................................... 131









8-21 Over allocation required to support various latencies in a system with 2-
persistent preview schedules, N= 100, r = 4,PLEP= 107 ......................... ...131

8-22 Over allocation required to support various latencies in a system with 2-
persistent current and preview schedules, N = 100, r= 4,PLEP = 10 7..................132

8-23 Latencies required to support a stream with optimal over allocation for a system
with non-persistent schedules for PLEP = {10-5, 10-7} ......................................132

8-24 Latencies required to support a stream with optimal over allocation for a system
with r-persistent current Schedules for PLEP = {10-5, 10-7} ............................... 133

8-25 Latencies required to support a stream with optimal over allocation for a system
with r-persistent preview schedules for PLEP = {10-5, 10-7} ...............................133

8-26 Latencies required to support a stream with optimal over allocation for a system
with r-persistent preview schedules for PLEP = {10-5, 10-7} ...............................134

8-27 Comparison of optimal latency supported by various scheduling strategies at
various beacon loss probabilities for N = 100, PLEP = 105 ...............................134

8-28 Comparison of optimal latency supported by various scheduling strategies at
various beacon loss probabilities for N = 100, PLEP =107 ...............................135

8-29 Minimum schedule persistence required to support L =2 with X = 1 for various
scheduling strategies at PLEP = 105 ............................... ...... ...............135

8-30 Minimum schedule persistence required to support L = 2 with X = 1 for various
scheduling strategies at PLEP = 107 ............................... ...... ...............136














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

QUALITY OF SERVICE DESIGN ISSUES IN MULTIMEDIA COMMUNICATIONS
OVER POWER LINE NETWORKS

By

Srinivas Katar

May 2006

Chair: Haniph A. Latchman
Cochair: Richard E. Newman
Major Department: Electrical and Computer Engineering

Home Networking is one of the major hurdles in enabling ubiquitous data and

multimedia distribution. Most houses are not equipped with specialized wiring for

networking purposes and retrofitting them with new wiring is prohibitively expensive.

Hence, the use of existing in-home power line infrastructure for networking purposes can

go a long way towards solving the in-home connectivity problem.

In this dissertation we first investigate the viability of power line communications

(PLC) by evaluating the performance of HomePlug 1.0. Detailed event based simulation

models were used to show that HomePlug 1.0 can provide a maximum UDP and TCP

data rates of 7.86Mbps and 5.90Mbps. Extensive field testing in 20 houses showed that

HomePlug 1.0 provides an average of 2.3Mbps higher throughput compared to IEEE

802.11b.

While HomePlug 1.0 is good enough for data networking, it will not be able to

support the newly emerging multimedia applications like audio and video streaming. We









study three issues that the new generation PLC systems need to overcome to enable

multimedia communications: (a) AC line cycle variation in noise, (b) Impulse noise, and

(c) Beacon Loss.

We use channel characterization measurements on 72 channels to investigate the

performance enhancements that can be achieved by using a channel adaptation

mechanism that is synchronized to the AC line cycle. The results show that a 30%

improvement in physical layer (PHY) data rates can be obtained by continuously

adapting to the AC line cycle. A time slot based adaptation that is more suitable for

practical systems provided an average of 10% improvement in MAC data rates.

To effectively overcome impulse noise, we propose a novel 2-level MAC framing

mechanism that enables efficient retransmissions. Our simulations and analysis show that

this mechanism provides linear degradation in performance even under high FEC Block

errors. Further investigation of the effect of transmission overheads on efficiency showed

that low data rate, low latency QoS applications can incur significant loss in efficiency

under high FEC block errors. Hence, PLC channel adaptation should take into account

the application QoS requirements to optimize the overall system capacity.

Due to the unreliable nature of the power line channels, beacon reception cannot

always be guaranteed. We investigate various persistent scheduling approaches to

mitigate beacon loss. Our analysis shows that using persistent current and preview

schedules, or persistent preview schedules, up to 10% beacon loss can be tolerated with

minimal impact on MAC efficiency and QoS.














CHAPTER 1
INTRODUCTION

The explosive growth of the Internet is driving the need for ubiquitous data and

multimedia communications in the twenty-first century. Major hurdles in realizing this

objective are the "last mile (access)" and "last hundred feet (in-home)" connectivity

problems. Approximately 50% of all network investments are spent on providing the

"last mile" infrastructure. The high expense incurred in laying new wiring for broadband

access to individual customers has forced service providers to use the existing cable and

telephone lines. As much of the world still awaits broadband access, there is a need for

alternative viable approaches to solving the last mile problem. In addition in-home

distribution of multimedia content is still a challenge particularly for homes that are not

equipped with proper wiring to support high-speed data and multimedia communications.

Retrofitting the house with new wiring is prohibitively expensive and hence the need

arises for new LAN technologies that enable affordable connectivity within the home.

With more than a century of ongoing deployments throughout the world, electrical power

lines represent by far the most pervasive network of wiring in the world. The use of

power lines as a communications medium can come a long way toward solving the last

mile and last hundred feet problems.

Several attempts were made starting from the early 1920's to develop a reliable

power line communication (PLC) technology. However, the lack of sophisticated digital

signal processing technologies hampered earlier PLC technologies from gaining

widespread acceptance. For several decades, their applicability was limited to operation









and maintenance of the power supply grid. The explosive growth of the Internet in the

recent decades has refueled the interest in developing high bandwidth (i.e., operating in

1-30 MHz frequency band) PLC technologies for in-home networking as well as for

broadband access. However, the challenges to enabling high-speed communication over

the power line are many.

Power lines were originally devised for transmission of power at 50-60 Hz and at

most 400 Hz. At high frequencies the power line is very hostile for signal propagation [1,

2]. Power line networks operate on standard in-building electrical wiring and as such

consist of a variety of conductor types and cross-sections joined almost at random.

Therefore a wide variety of characteristic impedances will be encountered in the network.

This impedance mismatch causes a multi-path effect resulting in deep notches at certain

frequencies. While a typical channel may present an average attenuation of

approximately 40 dB, it is not uncommon for portions of the bands to experience greater

than 60 dB of attenuation. Power line networks are also affected by interference. Electric

appliances with brush motors, switching power supplies and halogen lamps produce

impulse noise that can reduce the reliability of communication signals. Due to high

attenuation over the power line, the noise is also location dependent. Power line channels

also show strong dependence on the underlying AC line cycle. All these factors make the

design of PLC technologies extremely challenging.

In this dissertation we take a practical approach to understanding the issues in

enabling high bandwidth communication over the power lines and how the PLC

technologies based on HomePlug 1.0 and HomePlug AV standards overcome these

challenges. A combination of analysis, simulations and field tests are used to determine









the effectiveness of the various approaches used by these standards. The organization of

this dissertation is as follows.

Chapter 2 builds the background on power line communications. This chapter

introduces the two major categories of PLC devices, low-bandwidth PLC and high-

bandwidth PLC, and describes the various regulations constraints under which PLC

devices should operate. Technology choices currently available for low and high

bandwidth PLC systems are presented. This chapter also includes a detailed description

of the power line channel characteristics at high frequencies (i.e., 1-30 MHz).

The first generation of high-bandwidth PLC technology, HomePlug 1.0, is

investigated in Chapter 3. A detailed description of HomePlug 1.0 standard was followed

by simulation based evaluation of its performance. Simulation results showed that

HomePlug 1.0 provides a maximum throughput of 7.86 Mbps and 5.90 Mbps for UDP

and TCP respectively. The MAC saturation throughput showed graceful degradation with

increase in number of station in the network. Simulations with Voice over IP traffic

showed that the priority contention mechanism in HomePlug 1.0 could guarantee low

latencies for delivery delay sensitive traffic. Extensive field tests were also conducted to

evaluate the performance of HomePlug 1.0 with competing home networking solutions,

IEEE 802.1 lb and IEEE 802.1 la. These results show that the first generation PLC

technologies provide a ubiquitous and reliable home networking solution.

The HomePlug 1.0 standard was designed to transfer data with a maximum MAC

throughput of 8 Mbps. While this is good enough for file sharing and Internet access, it

will not be able to support newly emerging multimedia applications like audio and video

streaming. Multimedia applications require not only significantly larger bandwidths, but









also require guarantees on QoS parameters like latency, jitter and packet loss probability.

To address this new breed of applications, a new generation of HomePlug Standard,

HomePlug AV, was released by the HomePlug Powerline Alliance. In Chapter 4 we

provide a brief overview of the QoS requirements for various applications multimedia

applications and the salient features of the HomePlug AV standard that enables it to

support these applications.

To achieve data rates necessary to support multimedia applications, HomePlug AV

uses a channel adaptation mechanism that is synchronized to the underlying AC line

cycle. In chapter 5, we use channel characterization measurements and analysis to

investigate the performance enhancements that can be received by using AC line cycle

based channel adaptation. These results show that 30% improvement in the physical layer

data rates can be obtained by continuously adapting to the underlying AC line cycle.

Since continuous adaptation with respect to time is unduly complex for practical systems,

we propose dividing the AC line cycle into multiple regions and adapting independently

in each region. Results show that using a 1-2 milliseconds region size will provide

optimal MAC performance on most paths tested.

Impulse noise sources (like dimmers, halogen lamps, etc.,) are very common in

power line networks. Impulse noise will produce temporal effects in channel capacity.

Between two impulse noise events, the channel capacity can be very high, while during

the impulse noise the capacity is very low. To obtain low FEC block error rates, the

channel has to be adapted close to the lower end of its capacity (i.e., capacity near

impulse noise event) and hence is undesirable. Physical layer adapting to the higher end

of the channel capacity hinges on the ability to efficiently retransmit corrupted portions









of the data at the MAC layer. Chapter 6 presents a novel 2-level MAC framing

mechanism that enables efficient re-transmission of corrupt data. Both analytical and

simulation results are used to compare the performance of the 2-level MAC framing with

other viable approaches. These results show that 2-level MAC framing provides linear

degradation in performance even under high error conditions and it outperforms other

approaches.

Apart from the MAC framing overheads, each transmission will also incur

overheads due to physical layer headers, acknowledgments and interframe spacings. The

impact of these transmission overheads on the TDMA allocation requirements and

efficiency of multimedia streams is investigated in Chapter 7. Results show that the

allocation has to be increased by one or more transmission overhead to support low

latency applications. Further the increase the increase in allocation requirements is not

highly dependent on the application data rate. Hence, low data rate applications can incur

significant loss in efficiency. These results show that channel adaptation should take into

consideration the application data rate to improve the overall capacity of the system.

HomePlug AV uses a beacon based Time Division Multiple Access (TDMA)

technique for providing guaranteed bandwidth to multimedia streams. Beacons in

HomePlug AV carry TDMA allocation (or scheduling) information for various active

multimedia streams. Since a station that missed a beacon will not know the schedule for

that beacon period, it will not be able to transmit. Due to the large attenuation and

impulse noise, beacon detection cannot always be guaranteed. To overcome occasional

beacon loss, HomePlug AV uses persistent beacon schedules. Persistent schedules are

valid for multiple beacon periods, thus enabling a station to transmit even if it misses one






6


or more beacons. In Chapter 8, we use a Markov chain based analysis to investigate the

impact of beacon loss on systems with and without persistent beacon schedule. Results

show that by using persistent current and preview schedule or persistent preview

schedules, HomePlug AV systems can tolerate up to 10% beacon loss with minimum

impact of MAC efficiency and QoS guarantees.

A brief summary of the major contributions of this dissertation and the scope for

future work are presented in Chapter 9.














CHAPTER 2
BACKGROUND

Power line communications stands for the use of power supply grid for

communication purpose. Power line network has very extensive infrastructure in nearly

each building. Because of that fact the use of this network for transmission of data in

addition to power supply has gained a lot of attention. Since power lines are devised for

transmission of power at 50-60 Hz and at most 400 Hz, the use this medium for data

transmission, at high frequencies, presents some technically challenging problems.

Besides large attenuation, power line is one of the most electrically contaminated

environments, which makes communication extremely difficult. Furthermore, the

restrictions imposed on the use of various frequency bands in the power line spectrum

limit the achievable data rates.

Power lines connect the power generation station to a variety of customers

dispersed over a wide region. Power transmission is done using varying voltage levels

and power line cables. Power line cable characteristics and the number of crossovers play

an important role in determining the kind of communication technology that needs to be

used. Based on the voltage levels that are used, power lines can be categorized as

follows:

1. High-tension lines: These connect electricity generation stations to distribution
stations. The voltage levels on these lines are typically in the order of hundreds of
kilovolts and they run over distances of the order of tens of kilometers.

2. Medium-tension lines: These connect the distribution stations to pole mounted
transformers. The voltage levels are of the order of a few kilo volts and they run
over distances of the order of a few kilometers.









3. Low-tension lines: These connect pole-mounted transformers to individual
households. The voltage levels on these lines are of the order of a few hundred
volts and these run over distances of the order of a few hundred meters.

High-tension lines represent excellent carriers for RF energy as we only find open

wire equipment with very few crossovers. A transmission power of about 10 watts is

often sufficient to overcome distances of more than 500 kilometers. Around the year

1922 the first carrier frequency system (CFS) began to operate on high-tension lines in

the frequency range of 15-1500 KHz [3, 4, 5]. During the past and even nowadays the

main purpose of CFS was to maintain the operability of the power supply. While in

former times speech transmission was dominant, today we have more and more digital

data communications due to the rapid progress of overall automation. Through the

application of modern digital modulation and coding schemes, a significant enhancement

of bandwidth efficiency could be achieved for CFS.

Medium- and low-tension lines are characterized by a large number of cross

connections and different conductor types (e.g., open wire and cable). Long distance RF

signal propagation is extremely bad in this environment because of high attenuation and

impedance matching problems. Around the year 1930 ripple carrier signaling (RCS)

began to operate on these lines. These used a frequency range below 3 KHz down to 125

Hz with amplitude shift keying (ASK) modulation technique. The data rate achieved by

RCS is in the order of a few bits per second. Load management and automatic

reconfiguration of power distribution networks were among the most important tasks

performed by RCS.

We see that historically the use of power line communications was mainly for use

by the utility corporations (UCs) in maintaining the seamless power supply. The UCs

generally regarded the power distribution wiring as a "natural" medium for their









communication needs, as all important stations are connected. The lack of sophisticated

signal processing techniques at affordable cost points limited the applicability of PLC

technologies in the past.

2.1 Low Bandwidth PLC

In the mid 90's, data communications over low-tension lines regained a lot of

attention for enabling "smart houses." A "smart house" can be defined as a building

equipped with numerous sensors and actuators, where for example heating, air-

conditioning and illumination can be automatically and remotely controlled and

supervised. Furthermore safety systems such as burglar or fire alarms may be included [4,

6, 7]. New generations of low bandwidth PLC systems were designed for addressing this

market.

2.1.1 Regulatory constraints

Frequency bands used by these devices are restricted by the limitations imposed by

the regulatory agencies [8]. These regulations are developed to ensure harmonious

coexistence of various electromagnetic devices in the same environment. The frequency

restrictions imposed in two of the main markets, North America and Europe, are shown

in Figure 1-1. The Federal Communications Commission (FCC) and the European

Committee for Electro technical Standardization (CENELEC) govern regulator rules in

North America and Europe respectively.

In North America frequency band from 0 to 500 KHz can be used for power line

communications. However the regulatory rules in Europe are more stringent. The

spectrum is divided into five bands based on the regulations. They are

4. Frequency Band from 3 9 KHz: The use of this frequency band is limited to
energy provides; however, with their approval it may also be used by other parties
inside consumer's premises.










5. Frequency Band from 9 95 KHz: The use of this frequency band is limited to the
energy providers and their concession-holders. This frequency band is often
referred as the "A-Band"

6. Frequency Band from 95 125 KHz: The use of this frequency band is limited to
the energy provider's costumers; no access protocol is defined for this frequency
band. This frequency band is often referred as the "B-Band"

7. Frequency Band from 125 140 KHz: The use of this frequency band is limited to
the energy providers customers; in order to make simultaneous operation of several
systems within this frequency band possible, a carrier sense multiple access
protocol using center frequency of 132.5 KHz was defined. This frequency band is
often referred to as the "C-Band"

8. Frequency Band from 140 148.5 KHz: The use of this frequency band is limited
to the energy provider's customers; no access-protocol is defined for this frequency
band. This frequency band is often referred to as the "D-Band."



FCC General Use FCC Prohibited
Frequency Band Frequency Band


0 540 KHz

(a) FCC Frequency Band Allocation for North America



A-Band D- CENELEC
B-Band C-Band Ba Prohibited Frequency
nd Band

3 9 95 125 140 148.5 KHz

(b) CENELEC Frequency Band Allocation for Europe


Figure 1-1: FCC and CENELEC frequency band allocation

Thus in Europe power line communications are restricted to operating in the

frequency range from 95 148.5 KHz. Apart from band allocation, regulatory bodies also

impose limits on the radiations that are emitted by these devices. These reflect as


restrictions on the transmitted power in each of these frequency bands.









2.1.2 Technologies

Various protocols have been developed for use by low bandwidth digital devices

for communication on power line. Each of these protocols different in the modulation

technique, channel access mechanism and the frequency band they use. Various products

based on these protocols are available in the market and are mainly used for home

automation purposes. A brief overview of these protocols is presented here.

2.1.2.1 X-10

The X-10 technology is one of the oldest power line communication protocol. It

uses a form of amplitude shift keying (ASK) technique for transmission of information.

Although it was originally unidirectional (from controller to controlled modules) recently

some bi-directional products are being implemented. X-10 controllers send their signals

over the power line to simple receivers that are used mainly to control lightning and other

appliances. Some controllers available today implement gateways between the power line

and other medium such as RF and infrared.

A 120 KHz amplitude modulated carrier, 0.5 watt signal, is superimposed into AC

power line at zero crossings to minimize the noise interference. Information is coded by

way of bursts of high frequency signals. To increase the reliability each bit of information

is transmitted separately, which limits the transmission rate to 60 bits per second. This

represents poor bandwidth utilization while the reliability of transmission is severely

compromised in a noisy environment. These are the main reasons why this technology

has limited applications.

2.1.2.2 CEBus

The CEBus protocol uses peer-to-peer communication model. To avoid collisions a

carrier sensed multiple access with collision resolution and collision detection









(CSMA/CRCD) is used. The power line physical layer of the CEBus communication

protocol is based on spread spectrum technology patented by Intellon Corporation.

Unlike traditional spread spectrum techniques (that used in frequency hopping or time

hopping or direct sequence), the CEBus power line carrier sweeps through a range of

frequencies as it is transmitted. A single sweep covers the frequency band from

100-400 KHz. This frequency sweep is called a chirp. Chirps are used for

synchronization, collision resolution and data transmission. Using this chirp technology

data rate of about 10 KHz can be obtained. The frequency used by this technology

restricts its use in only North American market.

2.1.2.3 LonWorks

LonWorks is a technology developed by Echelon Corporation and provides a peer-

to-peer communication protocol, implemented using Carrier Sensed Multiple Access

(CSMA) technique. Unlike CEBus, LonWorks is a narrowband spread spectrum

modulation technique using the frequency band from 125 KHz tol40 KHz. It uses multi-

bit correlator intended to preserve data in the presence of noise with a patented impulse

noise cancellation. An advantage of the narrowband signaling is that it can be used in

both North American and European markets.

2.2 High Bandwidth PLC

High-speed communication over low-tension power lines has recently gained lot of

attention. This is fueled by the unparalleled growth of the Internet, which has created

accelerating demand for digital telecommunications. High bandwidth PLC devices are

designed to exploit this market. More specifically, these devices use the existing power

line infrastructure within the apartment, office or school building for providing a local

area network (LAN) to interconnect various digital devices. Some of the applications









include high-speed Internet access, multimedia, smart appliances/remote control, home

automation and security; data back up, telecommunications, entertainment and IP-

telephony.

2.2.1 Regulatory constraints

High bandwidth PLC devices for communication on power line use the frequency

band between 1 MHz and 30 MHz. In contrast to low bandwidth digital devices, the

regulatory standards for this region of the spectrum are still being developed. Currently,

United State is the only country where the regulations for High Bandwidth PLC systems

are well defined. In United States, PLC systems operate under FCC Part 15 rules using

the frequency band between 1.8-30 MHz. Several sub-bands within this range have to be

notched to prevent interference with licensed services.

2.2.2 Channel characteristics

Power lines were originally devised for transmission of power at 50-60 Hz and at

most 400 Hz. At high frequencies power line is very hostile for signal propagation [1, 2].

In the section that follows gives a brief overview of power line channel characteristics in

the frequency band between 1 MHz and 30 MHz.

2.2.2.1 Attenuation characteristics

High frequency signals can be injected on to the power line by using an

appropriately designed high pass filter. Received signal power will be maximum when

the impedance of the transmitter, power line and the receiver are matched. Dedicated

communication channels like Ethernet have known impedance, thus impedance matching

is not a problem. However, power line networks are usually made of a variety of

conductor types and cross sections joined almost at random. Therefore a wide variety of

characteristic impedances will be encountered in the network. Further, the network










terminal impedance will tend to vary both at communication signal frequencies and with

time as the consumer premises load pattern varies. The multipath effect resulting from

impedance mismatch and the attenuation caused by the electric wiring makes power line

channels highly frequency selective [9]. While a typical channel may present an average

attenuation of approximately 40 dB, it is not uncommon for portions of the bands to

experience greater than 60 dB of attenuation. Similarly, while most power line channels

have a significant delay spread of 1 to 2 |ts, it is not uncommon for some channels to

exhibit a delay spread larger than 5 hts. Figure 2-1 and Figure 2-2 show a sample power

line channel impulse and frequency response.

One of the unique characteristics of power line channels is the dependence of

channel characteristics on the AC line cycle. Electric appliances may turn on and off,

and/or draw electric power as a function of the AC line cycle. While this may change

both the channel's frequency response and noise profile, it is more common to see

changes in only the noise profile.

x 10




E I


4
I I'I ,





2 3 4 5 6 7 8 9 10 11 12
Time [us]

Figure 2-1: Power line channel impulse response










Channel Frequency Response
0
-10
-20

-30 -

-O40









-1 0.5 1 1 5 2 2.5
Frequency x10'


Figure 2-2: Channel frequency response

2.2.2.2 Noise characteristics

The major sources of noise on the power line are from electrical appliances, which


generate noise components that extend well into the high frequency spectrum. Worse,


there are often impulse noise sources. Impulse noise can be categorized as periodic or


continuous based on when it occurs relative to the underlying AC line cycle. Figure 2-3,


2-4 and 2-5 respectively show periodic impulse noise from light dimmer, yard light and


halogen lamp in addition to the data signal (rectangular shape). Devices that contain


brush motors (Figure 2-6) also produce impulse noise. Induced radio frequency signals


from broadcast, commercial, military, citizen band and amateur stations also impair


certain frequency bands. Because much of the noise experienced by each node may be


highly localized due to attenuation, the noise profile seen by each PLC device may be


significantly different. Therefore, power line channels are not typically symmetric.




















0.4----- -- ---- --- ------ --- --


0.3


0.2


0.1


0


-0.1


-0.2


-0.3


-0.4


-0.5
2 4 6 8 10 12 14 16
x 106




Figure 2-3: Noise generated by light dimmer




0.5


0.4















-05- -------- ------------- ---------------------------------
0.3


0.2


0.1


0


-0.1


-0.2


-0.3


-0.4


-010 12 14 16
2 4 6 8 10 12 14 16


x 105


Figure 2-4: Noise generated by yard light









17






0.5


0.4


0.3


0.2


0.1


0


-0.1


-0.2


-0.3


-0.4


-0.5
2 4 6 8 10





Figure 2-5: Noise generated by halogen lamp




0.5


0.4


0.3


0.2


0.1


0


-0.1


-0.2


-0.3


-0.4


-0.5
2 4 6 8 10


x 10


x 10


Figure 2-6: Noise generated by hair dryer









2.2.3 Technologies

The first generation of high bandwidth PLC systems were developed to enable

high-speed data communications in a Small Office Home Office (SOHO) environments.

These technologies represented a gain leap in exploiting the power lines as a true

communication medium. The most widely deployed first generation PLC systems are

based on HomePlug 1.0 Standard. HomePlug 1.0 was developed by HomePlug

Powerline Alliance; a non-profit industry association that was formed in March of 2000

by a group of industry leading companies to enable standards based power line

networking products. In the next chapter, we take a detailed look at the technology

behind HomePlug 1.0 and how it enabled it to overcome the harsh power line

environment. The performance of this technology against other popular SOHO

networking options was also presented.

As home networks become more and more common, there is increasing demand for

enabling multimedia communications within the home. Multimedia applications not only

require high bandwidth but also need guarantees on various QoS elements like delay,

jitter and packet loss probability. A second generation of HomePlug Standards,

HomePlug AV was developed to address this need. In Chapters 5, 6, 7 and 8, we

investigate some of the design choices made in HomePlug AV and evaluate their

performance.














CHAPTER 3
HOMEPLUG 1.0 TECHNOLOGY DESCRIPTION AND PERFORMANCE
EVALUATION

Affordable broadband Internet communication to residential customers is now

available via Cable Modems and various flavors of Digital Subscriber Lines (DSL). In

turn there is a growing need for in-home networks to share this single full-time Internet

access link, along with network connectivity for various computers and printers in the

home. A good home networking solution should be easy to setup, zero maintenance, cost

effective and should also be able to provide reliable performance throughout the house

With multiple outlets in every room, residential power lines are already the most

pervasive networks in the home. Using this existing infrastructure to provide high speed

networking capabilities provides several benefits. First of all, there is no need for

expensive rewiring of the house. Secondly, almost all devices that need to be networked

are already connected to the AC wiring. Thus, home networking becomes as simple as

plugging the device in the AC outlet.

HomePlug 1.0 Standard was developed to enable ubiquitous, easy to use and

affordable home networking. In this chapter we present detailed description of HomePlug

1.0 technology and evaluate its performance based on simulations and field tests. A more

thorough treatment of the material can be found in the following papers we published

[10, 11, 12]. The organization of this section is as follows. In Section 3.1 we provide a

brief overview of various options for home networking and their pros and cons. Details of

HomePlug 1.0 technology and simulation based evaluation of its performance is









presented in Section 3.2 and Section 3.3, respectively. Field test based performance

comparison of HomePlug 1.0 and IEEE 802.1 lb/a are presented in Section 3.3.

3.1 Home Networking Solutions

A variety of technological alternatives are available to home networking. These

technologies can be grouped into three different categories based on the medium they use

for transfer of data. They are:

1. New Wire
2. Wireless
3. No New Wires

3.1.1 New wire solution

New Wired solutions use specialized cable to connect various devices. There are

the most matured technologies and have found wide spread acceptance in enterprise,

campus and industrial networks. Some examples include Ethernet, Token Bus and Token

ring.

Ethernet is one of the most widely used networking solution inside the home these

days. The major drawback is that it would require retrofitting the house with these

specialized cables. This is quite an expensive proposition as it could involve drilling

holes in the walls. Further, most of the electricians have little or no knowledge on how to

install these cables. For example, a simple mistake of improper termination of Ethernet

wire will make the network useless. Another drawback of Ethernet is the lack of proper

QoS provisions that are very critical for any networking technology to be successful in

the home networking market. The recently developed IEEE 1394b [13] specification also

targets the home networking market. This is an extension of the popular IEEE 1394 [14]

(also known as FireWire or i.Link) that supports cord lengths up to 100 meters. Using Cat

5 UTP cords, it can support data rate of 100 Mbps over 100 meters distance. Using 50km









fiber, data rates of up to 3200 Mbps can be supported for cord lengths of up to 100

meters.

3.1.2 Wireless solution

Wireless technologies use high frequency radio waves to transfer data. The most

predominant technologies include,

1. IEEE 802.1 lb/a/g,
2. BlueTooth,
3. HomeRF, and
4. Ultrawide band.

The main benefit of using wireless networks is the freedom to move around the

house while maintaining network connection. IEEE 802.1 Ix [15, 16, 17, 18] are the most

popular wireless home networking technology. IEEE 802.1 lb uses Direct Sequence

Spread Spectrum (DSSS) in the 2.4GHz Industrial, Scientific and Medical (ISM) band

and provides PHY rates of up to 11 Mbps. The extensions of this technology include

IEEE 802.1 Ig and IEEE 802.1 la. IEEE 802.1 Ig uses the same frequency bands as IEEE

802.1 Ib, but with OFDM modulation and provides PHY rate of up to 54 Mbps. IEEE

802.1 la uses the 5 GHz Unlicensed National Information Infrastructure (U-NII) bands

along with OFDM modulation to provide up to 54 Mbps data rates. The main draw back

of this technology has been coverage and quality of service.

BlueTooth technology [19, 20] uses the 2.4GHz ISM band with frequency hopped

spread spectrum technology. This technology should be treated more like a Personal Area

Networking solution rather than a Home Networking solution due to lower transmit

power required by the kind of applications they serve.









HomeRF also uses the 2.4GHz ISM band with frequency hopped spread spectrum

technology. This technology had limited acceptance in the market. The only advantage of

this when compared to IEEE 802.1 lb is a better support for VoIP traffic.

The use of Ultrawide Band (UWB) technology [21, 22] was recently approved by

the FCC in February 2002. UWB is defined as any radio technology having a spectrum

that occupies a bandwidth greater than 20 percent of the center frequency, or a bandwidth

of at least 500 MHz. There are two competing UWB standards.

1. Multi-Band OFDM (MBOA) UWB standard developed by WiMedia Alliance, and

2. Direct Sequence Code Division Multiple Access (DS-CDMA) UWB standard
developed UWB Forum

UWB technologies can enable data rates of up to 480Mbps. As with Bluetooth,

Ultrawide band signals have limited range and are more suitable for Personal Area

Networks.

3.1.3 No new wire solutions

No New wire solutions use existing wiring inside the house to provide high-speed

communication capabilities. These technologies include

5. HomePlug 1.0, HomePlug AV
6. HomePNA, and
7. MoCA

HomePlug 1.0 and HomePlug AV technologies uses the existing power line wiring

inside the house and provide maximum PHY data rates of 14Mbps and 150Mbps,

respectively. The ubiquity of power outlets is the greatest advantage when compared to

other wired technologies.

HomePNA uses existing phone line as an infrastructure to provide home

networking. The HomePNA 3.0 standard [23] released in 2003 enables data rates up to









128 Mbps with optional extensions to 240 Mbps. It also has deterministic QoS, but

suffers from a limited number of available outlets in the house.

Multimedia over Coax (MoCA) [24] uses existing coax wiring to provide home

networking. To avoid interference with video channels, MoCA uses 850MHz to

1500MHz frequency band. Similar to using existing telephone wiring, cable outlets are

typically limited to 3 or 4 in the average home and are certainly not present in all rooms.

Use of splitters may also limit bi-directional transmission.

As we have seen in this section, there are several technologies that have targeted

the home networking technologies. Most of the technologies have gained limited

acceptance due to coverage or ubiquity problems. HomePlug and IEEE 802.1 Ix are the

most important contenders at present due to their ubiquity.

3.2 HomePlug 1.0 Technology

HomePlug 1.0 technology [11, 12, 25, 26] overcome the harsh channel conditions

on the power lines by using an adaptive approach that uses robust transmission technique

combined with sophisticated forward error correction (FEC) [27, 28], error detection,

data interleaving, and automatic repeat request (ARQ) [29].

3.2.1 HomePlug Physical layer

Orthogonal Frequency Division Multiplexing (OFDM) is the basic transmission

technique used by the HomePlug. OFDM is well known in the literature and in industry

[30]. It is currently used in DSL technology [31], terrestrial wireless distribution of

television signals, and has also been adapted for IEEE's high rate wireless LAN

Standards (802.1 la and 802.1 Ig). The basic idea of OFDM is to divide the available

spectrum into several narrowband, low data rate sub-carriers. To obtain high spectral

efficiency the frequency response of the sub-carriers are overlapping and orthogonal,









hence the name OFDM. Each narrowband sub-carrier can be modulated using various

modulation formats. By choosing the sub-carrier spacing to be small the channel transfer

function reduces to a simple constant within the bandwidth of each sub-carrier. In this

way, a frequency selective channel is divided into many flat-fading sub-channels, which

eliminates the need for sophisticated equalizers.

The OFDM used by HomePlug is specially tailored for power line environments. It

uses 84 equally spaced sub-carriers in the frequency band between 4.5MHz and 21MHz.

Cyclic prefix and differential modulation techniques are used to completely eliminate the

need for any equalization. Impulsive noise events are overcome by means of forward

error correction and data interleaving. HomePlug payload uses a concatenation of Viterbi

and Reed-Solomon FEC. Sensitive frame control data is encoded using turbo product

codes.

The power line channel between any two links has a different amplitude and phase

response. Furthermore, noise on the power line is local to the receiver. HomePlug

technology optimizes the data rate on each link by using channel adaptation. Channel

adaptation is HomePlug 1.0 involves Tone Allocation, modulation selection and FEC

selection. Tone allocation is the process by which certain heavily impaired carriers are

turned off. This significantly reduces the bit error rates and helps in targeting the power

of FEC and Modulation choices on the good carriers. HomePlug allows for choosing

from differential binary phase shift keying (DBPSK), differential quadrature phase shift

keying (DQPSK) on all the carriers, with either 1/2 or 34 FEC code rates. The end result of

this adaptation is a highly optimized link throughput.









Certain types of information, such as broadcast packets, cannot make use of

channel adaptation techniques. HomePlug uses an innovative modulation called ROBO,

so that information is reliably transmitted. ROBO modulation uses a DBPSK with heavy

error correction with bit repetition in time and frequency to enable highly reliable

communication. ROBO frames are also used for channel adaptation.

3.2.2 HomePlug 1.0 MAC

The choice of Medium Access Control (MAC) protocol provides a different set of

challenges. Home networks should be able to support a diverse set of applications

ranging from simple file transfer to QoS demanding applications such as Voice-over-IP

(VoIP) and low data rate streaming media. The HomePlug MAC is built to seamlessly

integrate with the physical layer and addresses these needs.

HomePlug MAC is modeled to work with IEEE 802.3 frame formats. This choice

simplifies the integration with the widely deployed Ethernet. HomePlug MAC appends

the Ethernet frames with encryption and other management before transmitting it over the

power line. A segmentation and reassembly mechanism is used to in cases where the

complete packet cannot be fit in a single frame.

3.2.2.1 Frame formats

HomePlug technology uses two basic frame formats (refer Figure 5-1). A Long

Frame consists of a Start of Frame (SOF) delimiter, Payload and End of Frame delimiter

(EOF). A Short Frame consists of a Response Delimiter and is used as part of the Stop-

and Wait automatic repeat request (ARQ) process. ARQ mechanism causes

retransmission of corrupt packets, thus reducing the packet error rate.

All the delimiters share a common structure. A delimiter consists of a Preamble and

Frame Control information field. The Preamble is a form of spread spectrum signal that is









used to determine the start of a delimiter. This is followed by Frame Control information,

which is encoded using a robust Turbo Product Code and can be detected reliably even at

several dB below the noise floor. Among other things, delimiters convey timing

information that is used by MAC to determine the availability of the medium. The robust

design of the delimiter helps the nodes to obtain a very high level of synchronization,

thus reducing unintended collisions. The details of the various field contained in the

Frame Control are given in Table 3-1. The Payload of the Long Frame delimiter is

encoded based on the channel adaptation. The first 17 bytes of the payload contain the

Frame Header. This field contains the source address, destination address and

segmentation information.

HomePlug technology limits the maximum length of the payload field in the Long

Frame to 160 OFDM Symbols (-1.3 msec). This manifests as better guarantees of QoS as

the delay incurred by higher priority traffic due to on going lower priority transmission is

reduced. If the packet cannot be fitted into a Long Frame, a segmentation and reassembly

mechanism is used to send it in multiple Long Frames. The frame header contains

information that is used by the receiver to properly reconstruct the segmented packet. The

Payload is protected by a Frame Check Sequence (FCS) to detect uncorrected errors.

3.2.2.2 Channel access mechanism

The channel access mechanism used by the HomePlug MAC is a variant of the

well-known CSMA/CA protocol. A typical CSMA/CA protocol would require nodes to

sense the medium (this is the carrier sensing part) for other traffic. If the medium is busy,

nodes will defer from transmitting until the medium becomes idle. When the medium

becomes idle, nodes will wait for a randomly chosen duration (this is the collision

avoidance part). A node will transmit only if it detects no other traffic on the medium













during this randomly chosen duration. HomePlug channel access scheme builds upon this


mechanism by providing prioritized access along with high network utilization. The


overall protocol includes a carrier sensing mechanism, a priority resolution mechanism


and a backoff algorithm.


25 bits


17 bytes


Preamble Frame Control Frame Header


4 symbols



Start of Frame
Delimiter


variable byte variable 2 bytes 25 bits
count byte
count

Frame Body B-PAD FCS Preamble Frame Control


variable 4 symbols
symbol count 4 symbols


Payload End of Frame
20-160 OFDM Symbols Delimiter


Figure 3-1: Long Frame format






25 bits


Preamble Frame Control


4 symbols

A
A [ ---------J

Response Delimiter







Figure 3-2: Short Frame format


The carrier sense mechanism helps HomePlug nodes to synchronize with each


other. At the heart of this mechanism are the delimiters. HomePlug technology uses a


combination of Physical Carrier Sense (PCS) and Virtual Carrier Sense (VCS) to


determine the state of the medium (i.e., if the medium is idle or busy and for how long).









PCS is provided by the HomePlug PHY and basically indicates whether a preamble

signal is detected on the medium. VCS is maintained by the HomePlug MAC layer and is

updated based on the information contained in the delimiter (refer Table 1). Delimiters

contain information not only on the duration of current transmission but also on which

priority traffic can contend for the medium after this transmission. PCS and VCS

information is maintained by the MAC to determine the exact state of the medium.

Table 3-1: Frame Control information fields
Delimiter Type Fields Meaning
This can be SOF with response expected or an
SOF with no response expected depending on
Type whether a Short Frame delimiter is expected at the
end of this Long Frame
When set to 1, this prevents all HomePlug nodes
Contention with packets of priority level equal to or less than
Contention
the current Long Frame's priority from accessing
Control
Start of Frame the channel. However, higher priority nodes can
(SOF) still interrupt this transmission
F e L h This indicates the length of the payload in
Frame Length
multiples of OFDM symbol blocks
This is an index to the channel adaptation
information stored at the receiver. Note that the
Tone Map
Index variable length Payload is encoded using the
maximum transfer rates that can be achieved by
the link.
This can be EOF with response expected or an
EOF with no response expected depending on
Type whether a Short Frame delimiter is expected at the
End of Frame end of this Long Frame.
End of Frame -
SC The information conveyed is same as that
(EOF) Contention .
Control conveyed by this field in SOF delimiter. This
redundancy helps in better synchronization.
Channel Access This field indicates the priority of the current Long
Priority (CAP) Frame.
This can be ACK (positive acknowledgment),
e NACK, (negative acknowledgment indicating
Response (Resp) faulty reception), or FAIL (negative
Responseacknowledgment indicating lack of resources)
Channel Access This field indicates the priority of the preceding
Priority (CAP) Long Frame










The Priority resolution mechanism provides prioritized access of the medium in a

highly distributed manner. Due to the distributed nature of this mechanism, there is no

need for a central node to coordinate access over the medium as in the case of some

networking technologies. HomePlug allow up to four different priority levels. At the

heart of this priority resolution mechanism are the priority resolution slots (PRS) and the

priority resolution signals. Priority resolution signals use a form of spread spectrum

signal that has a high tolerance to delay spread and is also very robust. High tolerance to

delay spread prevents destructive interference when multiple nodes assert in the same

priority resolution slot. After the end of every transmission, two slots are allocated for

priority resolution. Figure 3-3 shows the occurrence of priority resolution signals with

respect to the end of previous transmission and the subsequent contention period. After

priority resolution slots, contention will be only between nodes that have the highest

available priority in the network. For example, when all four priorities are present in the

network, Priority 3 and Priority 2 nodes will transmit priority resolution signal in PRSO.

Priority 1 and Priority 0 nodes will detect this signal, causing them to defer to the higher

priority traffic. Priority 3 nodes will transmit a priority resolution signal in PRS1, which

will be detected by the Priority 2 nodes, causing them to defer. Thus only Priority 3 nodes

will contend in the Contention Period.

Long Frame Short PRSO PRS1 contention Period

V N- -


Previous Transmission


Priority Resolution
Slots


Figure 3-3: Priority resolution and backoff scheme

The backoff algorithm used by HomePlug MAC is designed to provide high

network utilization (which manifests itself as high network throughput) even under


Backoff Slots









heavily loaded conditions. It is also tailored to seamlessly integrate with the priority level

construction and the applications that each of these priorities is expected to support. Note

that only nodes that have the highest priority available in the network will contend during

the Contention Period and all other nodes will defer. From a HomePlug view of the

power line network, all traffic with priority a less than the maximum available traffic

priority in the network is pretty much ignored. Lower priority traffic will have to wait till

all the higher priority traffic is transmitted.

As with other CSMA/CA algorithms, the backoff slot is chosen to be a random

integer between 0 and the Contention Window Size. The growth of the contention

window under HomePlug is controlled by the estimation of traffic on the network as well

as on the priority of the traffic. This helps HomePlug nodes achieve higher network

utilization and also control the latency for higher priority traffic.

3.2.2.3 Segmentation and reassembly mechanism

HomePlug uses an adaptive approach to maximize the throughput on each link.

Furthermore, the maximum size of the long frame is limited to 160 OFDM symbols for

the payload. For cases in which the complete Ethernet packet cannot be fitted in a single

long frame, segmentation and reassembly mechanism is used. The necessary information

for proper reassembly at the receiver is included in the segment control field of the frame

header. Multiple segments can be transmitted in a single burst to obtain high throughputs.

However, each segment is required to go through priority resolution mechanism. This

ensures that a segment burst can be interrupted by higher priority traffic, thus reducing

the latency for higher priority traffic. HomePlug allows extension of this segment

bursting across multiple Ethernet packets to provide contention free access.









3.3 HomePlug 1.0 Performance based on Simulations

In this section we report the measurements observed using an event-based C

program to simulate a HomePlug 1.0 power line network. All scenarios assume QPSK

and a 3/4 coding rate on various links. In this simulation, we use UDP, TCP, and VOIP

traffic. Both TCP and UDP traffic sources are configured to generate 1460 byte packet at

priority 0. TCP and UDP traffic is generated with exponential inter-arrival time with 100

microsecond average. This will ensure that the network will remain saturated. VOIP

traffic is modeled as isochronous with a 20 msec interarrival time. The packet size of

VOIP is 160 bytes and is assigned the highest priority, 3.

In Table 3-2, we provide the of simulation results of a power line network. The

UDP traffic simulation scenario 1 shows the best throughput in our simulations since

there is no contention at all. Table 3-2 also shows channel contention with 2 and 3 UDP

nodes causes a modest reduction in channel throughput. In the TCP traffic simulation,

though scenario 1 has only one traffic source, the bandwidth must be shared with data

and response frames (e.g., TCP ACK packets) thus it provides lower performance than

the UDP traffic simulation. The MAC throughput represents the total number of

transmitted bytes of Ethernet payload transmitted by the MAC divided by the simulation

time. The UDP throughput represents the total number of transmitted bytes of UDP

payload divided by the simulation time. Similarly, TCP throughout represents the total

number of transmitted bytes of TCP payload divided by the simulation time.

The third metric we provide in Table 3-2 is the PLC simulation results of one VOIP

and multiple UDP connections. The high priority VOIP always wins the contention and

the UDP nodes can send packets only when there is no VOIP traffic. In this simulation,

the VoIP Delay indicates the average network delay (i.e., queuing delay and channel









access delay) incurred by the VoIP packets. These results show that the priority

contention mechanism in HomePlug 1.0 can guarantee very low latency for delay

sensitive traffic. The Table 3-2 also shows the simulation results of one VOIP and

multiple TCP connections. The throughput of VOIP is small; the total throughput is

dominated by the TCP component. The slight increase in VoIP delay is due to the

increase in channel access time due to higher number of contending source (i.e.,

contention between TCP Data and TCP ACK).

Table 3-2: HomePlug 1.0 simulation results
Simulation Results
Throughput of multiple UDP Traffic Streams
Scenario 1 (1 UDP) Scenario 2 (2 UDP) Scenario 3 (3 UDP)
MAC Throughput 8.08 Mbps 7.46 Mbps 7.46 Mbps
UDP Throughput 7.86 Mbps 7.26 Mbps 7.26 Mbps
Throughput of multiple TCP Traffic Streams
Scenario 1 (1 TCP) Scenario 2 (2 TCP) Scenario 3 (3 TCP)
MAC Throughput 6.16 Mbps 6.15 Mbps 6.12 Mbps
TCP Throughput 5.90 Mbps 5.89 Mbps 5.87 Mbps
Throughput of one VoIP and multiple UDP traffic streams
Scenario 1 Scenario 2 Scenario 3
1 VoIP, 1 UDP 1 VoIP, 2 UDP 1 VoIP, 3 UDP
MAC Throughput 7.89 Mbps 7.33 Mbps 7.29 Mbps
UDP Throughput 7.53 Mbps 6.99 Mbps 6.96 Mbps
VoIP Delay 2.75 msec 3.00 msec 3.00 msec
Throughput of one VoIP and multiple TCP traffic streams
Scenario 1 Scenario 2 Scenario 3
1 VoIP, 1 TCP 1 VoIP, 2 TCP 1 VoIP, 3 TCP
MAC Throughput 6.04 Mbps 5.85 Mbps 5.77 Mbps
TCP Throughput 5.65 Mbps 5.47 Mbps 5.39 Mbps
VoIP Delay 3.25 msec 3.25 msec 3.25 msec


3.4 HomePlug 1.0 Field Test Results and Comparison with IEEE 802.11b/a

To understand the real-world performance of HomePlug 1.0 and to compare it with

a popular home networking alternative, IEEE 802.1 la/b, we conducted field tests in 20

houses located in the Gainesville, Ocala, Orlando, and Belleview areas of Florida. The









choices of the houses used in the tests were in the medium to large range (1500-5000 ft2),

since larger houses provide a better range on the performance parameters of interest.

The equipment used in this test included the following.

* AP server: A Sony notebook with a 700 MHz Pentium III processor and 128
Mbytes RAM running Windows2000

* Mobile Station: An HP notebook with a 500MHz Pentium III processor and 128
Mbytes RAM running Windows2000

* Linksys HomePlug 1.0-based Powerline-to-Ethemet bridges

* Netgear IEEE 802.1 lb Access Point and PCMCIA Card

* D-Link DWL-5000AP IEEE 802.1 la Access Point and D-Link DWL-A650
PCMCIA card

For PLC testing the two laptops were connected through the power line via

Powerline-to-Ethemet bridges. For wireless testing, a modified infrastructure mode

(MIM) was used. The AP server was connected to an access point using an Ethernet

crossover cable to the built-in Ethernet socket. A PCMCIA slot in the mobile station was

used to connect the wireless card. Note that typical wireless networks use an

infrastructure mode (IM). In this mode, all wireless nodes communicate with each other

through the access point, and must share the bandwidth over two hops. Furthermore,

since these tests had the AP server connected to the access point via Ethernet, there was

no other contention possible, so the test results should represent the best case scenarios in

this regard.

3.4.1 Experiment method

The TCP throughput and distances were measured for various locations of AP and

mobile stations inside the house. The AP server was located close to a phone or cable

outlet, the most probable locations for the home network to be connected to the









broadband access network. The mobile station was located at various places where it

would be likely to find other networked devices in the home. The AP Server antenna and

Mobile Station antenna were placed randomly to minimize the effect of directional

antenna gain. We argue that this is the typical antenna placement since ordinary users

probably don't know how to set antenna directions to maximize throughput. Besides, not

all locations are susceptible to antenna direction adjustment due to the surrounding

environment.

WSTTCP, a popular TTCP implementation ported to Windows sockets, was used.

The TCP buffer size was chosen to be 11,680 bytes (1460 x 8). The number of TCP

buffers transmitted was chosen such that each test ran for approximately 60 seconds. A

single run of WSTTCP involved starting the WSTTCP in receive mode at the receiver on

a selected port. WSTTCP was then started at the transmitter with a specific TCP buffer

size, number of TCP buffers to be transferred, receiver IP address, and receiver port

number. At the end of transmission, WSTTCP (at both the transmitter and receiver)

provided the throughput observed on the link. For several of these tests, real-time packet

capture was also obtained to observe TCP stability. All the procedures were automated

and required minimal human operation.

3.4.2 Results

The amount of data collected is too large to be presented, so only a summary of the

most interesting findings is presented.

3.4.2.1 IEEE 802.11a indoor performance

The performance and coverage results of IEEE 802.1 la are depicted in Fig. 3-4.

Figure 3-4a shows the connectivity (i.e., percentage of good links) as a function of house

area. As expected, the connectivity decreased as the house area increased. Results show










that connectivity is poor even in moderate size (2500 ft2) houses. For larger houses

(>4000 ft2) the connectivity decreased to 20 percent. Figure 3-4b shows a scatter plot of

throughput as a function of distance. It is interesting to note that IEEE 802.1 la

connectivity is almost zero when the distance is larger than 50 ft. Figure 3-4c shows the

percentage of links that exceed the throughput values indicated on the x-axis. Note that

the maximum IEEE 802.1 la throughput obtained from the product being tested was

larger than those expected from theory. This could be because of manufacturer-specific

proprietary enhancements like the use of higher-level modulations. In summary, the

statistics show that:

IEEE 802.1 la failed on at least one link in 19 of the 20 houses tested.

802.11 a failed to connect in 45 percent of the links that were tested.

802.1 la showed close to zero connectivity at distances larger than 50 ft.

For shorter distances, 802.1 la provided excellent throughput in most cases.

Connectivity versus House Area

100%
g 80% --- ---- ------- ---
*- 60% ------- -------
| 40% -- ----
) 20% ------ -- ------------ -- --
0 ----------------------
0%
1400 2400 3400 4400
Area(sq. ft.)
(A)

Figure 3-4: IEEE 802.1 la indoor performance A) Connectivity versus house area B)
Throughput versus distance C) Percentage of links versus throughput











Throughput vs Distance
40.00
30.00 -----. --, .----------.
& *

20.00 -------- .----------.-------

10.00 -^^_^ % ------ r -------
-ti



0.00 t -..
0 20 40 60 80 100
D-tI li ll II I

(B)

Percentage Of Links versus Throughput

100%
8310% ----------------------------------------------
0 60% --------- .--------- .----------.-------------------
I 80% -------




0 10 20 30 40
T L i -lj|.lIi i.r i


Figure 3-4: Continued


Figure 3-5: HomePlug 1.0 and IEEE 802.1 lb indoor performance A) Connectivity versus
house area B) Throughput versus distance C) Percentage of links versus
throughput (MIM) D) Percentage of links versus throughput (MIM)


Connectivity versus House Area

100% -

S80% ----------------

60% ----------------------- ---


S-. *0 HoniePlug 1.0
-8-802.11b
A ii





0/0%
" 60% .------...................-.....---.....




1400 2400 3400 4400
Area(sq. ft.)












Throughput versus Distance
9.00
9 HomePlug 1.0
g, 802.1 lb
6.0 v -- ------ -------- ----------





0.00 Rt 4-
0 20 40 60 80 100

)Di-.riKic I tT.)
(B)

Percentage Of Links vs Throughput (MIM)


100%
80%

60%

40%

20%

0%
0.00


4.00

T lhi ,. iILUI[ Ir ,II ,1 .'N


Figure 3-5: Continued


3.4.2.2 IEEE 802.11b and HomePlug 1.0

To facilitate comparison, the performance and coverage results of IEEE 802.1 lb


and HomePlug 1.0 are shown together in Fig. 3-5. Figure 3-5a shows connectivity as a


Percentage Of Links vs Throughput (IM)
4 100%
S ., HomePlug 1.0
| 80% -- ---- ---o802.11b

U 60% -----------'- ----------- ------------- -- --
S----------------------;------------
cd' I
O
4 0% --------------------

I20% O---------------- -----------F- ---------------

0%
0.00 2.00 4.00 6.00 8.00
T l IIl 111-j 'ut.' 1 M i. ,









function of house area. Both technologies show high connectivity for houses of less than

4000 ft2. For houses larger than 4000 ft2, the connectivity for IEEE 802.1 lb dropped

dramatically to 50 percent in both of the houses tested, while HomePlug 1.0 continued to

show high connectivity. Figure 3-5b shows a scatter plot of throughput as a function of

distance. IEEE 802.1 lb typically provides close to maximum throughput at distances of

less than 50 ft; for distance larger than 50 ft, the performance exhibited large variations.

On the other hand, the HomePlug 1.0 system performance is not correlated with the line

of sight distances measured in this experiment. This is because HomePlug 1.0 signals

have to pass through the convoluted power line cable runs to reach the mobile station.

Figure 3-5c shows the percentage of links that exceed the throughput value depicted on

the x-axis.

Our experiments showed that the overall coverage of 802.1 lb was 92 percent. The

maximum throughput observed in field-testing was 5.13 Mb/s. Figure 3-5c shows that

around 70 percent of the connections operated at more than 4 Mb/s and 10 percent above

5 Mb/s. For HomePlug 1.0, the overall coverage is 98 percent. The maximum throughput

observed in testing was 5.98 Mb/s. For HomePlug 1.0, 58 percent of the connections

operated above 4 Mb/s and 38 percent had throughput above 5 Mb/s. The interesting

crossover phenomena displayed in the graph reflect three aspects of the systems. First,

the paucity of data rates supported by 802.1 lb hurts its performance when channel

conditions are sub optimal. Second, HomePlug 1.0's ability to adapt to the channel

conditions with a nearly continuous selection of data rates allows it to perform better

under mediocre channel conditions. Finally, the higher maximum data rate of HomePlug

1.0 allows it to outperform 802.1 lb when channel conditions are favorable.









Note that, as mentioned earlier, in this experiment, throughput was measured

between the access point and the mobile station in a MIM. However, this may not always

be the way IEEE 802.1 lb stations communicate with each other. IEEE 802.1 lb networks

can be configured in either ad hoc mode or IM. In ad hoc mode, wireless stations

communicate with each other directly. However, typical home networks use an IM in

which each wireless station communicates with the access point, which in turn forwards

the data to the designated receiver. Some of the reasons for using IM include ease of

setup, better coverage, and security. Furthermore, most wireless equipment is configured

in IM out of the box, and must be reconfigured to ad hoc mode by the customer. Thus, in

a typical IEEE 802.1 lb home network, all station-to-station transmissions, other than

those designated to the access point itself (i.e., the access point is the final destination of

the transmission) or those that originate from the access point, will be retransmitted by

the access point. This reduces the effective throughput experienced between such

stations.

We use a simple method to extrapolate the IM throughput from the MIM link

throughput data collected in the field tests. A random sample was chosen from the set of

collected data and used as the throughput (R1) from a TCP source to the access point.

Another random sample was chosen from the sample and was used as the throughput

(R2) from the access point to the TCP destination. The aggregate IM throughput then can

be obtained by assuming that a fixed packet size ofx bits is transmitting through two

links with speeds R1 and R2. The total time to transmit this packet will require x/R1 +

x/R2 s. Thus, throughput can be calculated by R1 x R2/(R1 + R2). Multiple iterations

were used to obtain the distribution of the IM throughput. Although this method is not









fully accurate, it is reasonable to expect the actual performance in IM to be close to the

values obtained.

Figure 3-5d shows the percentage of links that exceed the calculated IM throughput

value depicted on the x-axis. These results show that HomePlug 1.0 stations provide

superior coverage and throughput to IEEE 802.1 lb stations in IM. From the statistics, we

make the following key observations:

* HomePlug 1.0 had a larger maximum throughput than 802.1 lb (about 1 Mb/s
larger).

* On 60 percent of the links HomePlug 1.0 performed better than 802.1 lb links in
MIM.

* On an average basis, HomePlug 1.0 gave approximately 0.2 Mb/s higher TCP
throughput than 802.1 lb in MIM.

* In six of the 20 houses tested, IEEE 802.1 lb failed on at least one link.

* In two of the 20 houses tested, HomePlug 1.0 failed on at least one link.

* On an average basis, HomePlug 1.0 gave approximately 2.3 Mb/s higher TCP
throughput than 802.1 lb would be expected to give in IM.

In summary, HomePlug 1.0 was found to provide better coverage and slightly

better average TCP throughput than IEEE 802.1 lb in MIM, which would be typical for

Internet access. In IM, HomePlug 1.0 was estimated to have throughput about 2.3 Mb/s

greater than 802.1 lb. For shorter line-of-sight distances, 802.1 Ix performed better than

HomePlug 1.0, but for longer distances, the nearly continuous adaptation capability of

HomePlug 1.0 allowed it to make better use of mediocre channels.














CHAPTER 4
POWER LINE MULTIMEDIA NETWORKING

Current generation PLC-based home networking technologies are mainly data

centric. While these are good enough for applications like web browsing and file transfer,

they cannot support multimedia applications such as audio/video streaming and multi-

player gaming as these applications require much higher bandwidths and QoS than those

provided by current technologies [32]. Table 4-1 shows the bandwidth and QoS

requirements for typical multimedia applications. For example, High Definition MPEG2

streams (HDTV streams) will require quasi-error free performance at up to 24 Mbps of

application throughput, and about 300 msec of delay with less than a few microseconds

of jitter. Furthermore, typical home network usage scenarios include simultaneous

transfer of multiple multimedia streams. Hence, home networks should be capable of

providing 50-60 Mbps application level throughput within the network.

Table 4-1: Bandwidth and QoS requirements for multimedia applications
Appl n Bandwidth Latency r Packet Loss
Application Jitter
(Mbps) (msec) Probability
High Definition (HD) Video 11 24 100 300 0.5 |ts to Quasi Error
Streaming several msec Free
Standard Definition (SD) 26 10.5 |ts to Quasi Error
Video Streaming several msec Free
0.5 Cls to Quasi Error
DVD quality Video 6-8 100 300
several msec Free
Internet Video Conferencing 0.1 2 75 100 several msec 10-3
Home Theater Audio6 100 300 0.5 |ts to Quasi Error
(Multiple streams of audio) several msec Free

Voice over IP (VoIP) < 0.064 10 30 1030 10
msec
10 30 Quasi Error
Network Gaming < 0.1 10 30
msec Free









1) msec => milliseconds
2) Quasi Error Free => Less than 1 packet dropped in 2 hours

To support the data rates and QoS required by the new generation of multimedia

applications HomePlug Powerline alliance has released a new generation of HomePlug

Standards, HomePlug AV. The HomePlug AV physical layer is designed to operate as

close as possible to the channel capacity of the Power line. Due to the regulatory

constraints, a spectrally efficient PHY is essential to ensure that multiple multimedia

streams can be supported simultaneously and delivered to the whole house. HomePlug

AV PHY incorporates the following salient features.

1. Bit-loading and higher order modulations (including 8, 16, 64, 256, and 1024-
QAM) to allow each carrier to take full advantage of the SNR it experiences.

2. Coherent modulation, for increased the SNR given a fixed transmit power.

3. Long symbol time to minimize guard interval overhead.

4. Forward error correction based on parallel concatenated convolutional codes, to
decrease the probability of path "no-connects" and ensure high throughput.

5. Transmission synchronized with respect to the AC line cycle, for stable and
improved throughput.

6. MAC-PHY cross-layer design for handling impulse noise.

These enhancements enable HomePlug AV to achieve a maximum physical layer

data rate of 150 Mbps. The exact details and rationale for the choice of various PHY

parameters is beyond the scope of this work. Further details of HomePlug AV physical

layer can be found in [33, 34].

The HomePlug AV Medium Access Control (MAC) layer is based on a hybrid of

Time Division Multiple Access (TDMA) and Carrier Sense Multiple Access (CSMA).

Each network has a designated device, the Central Coordinator (CCo), that coordinates

the access to medium. The CCo periodically generates a beacon with information on









TDMA and CSMA allocation. TDMA is used by applications requiring QoS, while

CSMA is used for transport of regular data. Some of the salient features of the HomePlug

AV MAC are,

1. Beacon Period Synchronized to the AC line cycle to enable AC line cycle based
channel adaptation by the PHY,

2. 2-Level MAC framing to effectively deal with impulse noise,

3. Beacon Schedule Persistence to mitigate beacon loss,

4. Dynamic TDMA allocations to deal with changing channel conditions,

5. Jitter Control based on the use of Arrival Time Stamps and Clock tracking,

6. Proxy Coordination function to deal with hidden stations,

7. Neighbor network mode of operation to deal with overlapping networks,

8. Coexistence with HomePlug 1.0 devices,

For further details of HomePlug AV MAC, refer to [33, 34]. In this dissertation, we

investigate the first three features of the HomePlug AV MAC and evaluate their

performance.














CHAPTER 5
CHANNEL ADAPTATION BASED ON CYCLO-STATIONARY NOISE
CHARACTERISTICS

As power line communication (PLC) systems become more pervasive, there is

increasing demand for new generations of PLC systems that enable bandwidth intensive

applications like audio/video streaming and broadband access. To achieve the data rates

necessary to support such applications within the current regulatory constraints, PLC

systems should exploit the inherent characteristics of the power line channel that enable a

more aggressive and optimized channel adaptation. One such characteristic is the cyclic

variation of the noise power with the phase of the underlying AC line cycle.

In this section, we use extensive channel characterization measurements and

analysis to investigate the performance enhancements that can be achieved by using a

channel adaptation mechanism that exploits the cyclic variation in power line noise

characteristics. The results show that a 30% improvement in physical layer (PHY) data

rates can be obtained by continuously adapting to, and synchronizing communications

with, the AC line cycle. Since continuous adaptation with respect to time is unduly

complex for practical systems, we propose dividing the AC line cycle into multiple

regions and adapting independently in each region. Results show that using a region size

of 1-2 milliseconds provides optimal MAC throughput on most paths tested. A subset of

the results presented in this chapter is also published in [35].









5.1 Introduction

The future evolution of PLC systems is driven primarily by an increasing demand

for higher throughput. In LAN scenarios, the requirement to support multimedia

applications like audio and video streaming is driving the need for higher throughput. In

broadband access, the need to provide higher access speeds to a larger customer base

(sharing the same medium) is causing the demand for higher throughputs. The design of

PLC systems that enable this new generation of high data rate applications should exploit

the unique channel characteristics observed on the power line channels to provide higher

throughput within the current regulatory constraints.

A typical power line network consists of a variety of conductor types joined almost

at random and terminating into a variety of electrical appliances. The conductors were

primarily designed for transmission of power at 50-60 Hz and at most 400 Hz. At higher

frequencies they provide characteristic impedances that not only vary based on the

conductor type but also as a function of frequency. Impedance mismatches in power line

networks result in a multipath effect that often cause deep nulls in certain frequency

bands. Electrical appliances connected to the power line also significantly impact the

PLC channel characteristics. Several common electric appliances turn on/off via

switching power supplies and/or draw a varying amount of electric power as function of

the underlying AC line cycle voltage. During these processes noise is generated that

extends well into the higher frequency bands where PLC systems operate. One of the

unique characteristics of this man-made noise is its cyclic dependency on the phase of the

underlying AC line cycle.

The cyclic nature of noise in power line networks has been well known for several

years. One of the first PLC based home automation technology, X10, exploited the low









noise characteristics at the zero crossing of the AC line cycle to increase the system

robustness. 0. Ohno and M. Katayama [36, 37] were among the first to characterize the

noise in (0-500Kz) frequency band as 'cyclo-stationary'. In [38] Y. Hirayama et al., have

investigated and confirmed the cyclo-stationary noise characteristics in the band from DC

to 25MHz. Apart from noise, the channel response (or attenuation characteristics) can

also change with the AC line cycle. F.J. Canete, J.A. Cortes, et al., [39, 40] has recently

proposed a Linear Periodically Time Varying (LPTV) model for the PLC channel

response. While in general both channel response and noise characteristics can vary as a

function of the AC line cycle, it is more common to see significant variations in noise

profile compared to that of the channel impulse response. In this section, we study the

effect of the cyclic variation of power line noise on the performance of PLC systems.

The organization of the remaining sections is as follows. Section 5.2 describes the

test setup used to measure channel attenuation and noise characteristics. Mechanisms for

channel adaptation over cyclic channels and the various tradeoffs involved are presented

in Section 5.3. Processing of the field test data to compute channel capacities is presented

in Section 5.4. The results obtained are discussed in Section 5.5 and conclusions are

presented in Section 5.6.

5.2 Channel Measurements

A total of 72 channels in two locations were characterized for use in this study.

The characterization of each path involved computation of channel impulse responses

and time domain noise captures spanning a single 60 Hz AC line cycle. The paths

characterized were chosen at random within each location.

The setup for measuring the channel impulse response is shown in Figure 5-1. An

arbitrary waveform generator, functioning as a transmitter with a digital-to-analog









converter (DAC), was loaded with a sounding waveform sampled at 75MS/s. The signal

out of the DAC was amplified and coupled on to the power line. At the receiver side, the

signal is coupled off of the power line medium, passed through a 2 to 30 MHz band pass

filter, digitized at 75 MS/s using an analog-to-digital converter (ADC) card, and passed to

Matlab for subsequent processing. Similarly, noise captures were performed by turning

off the transmitter and simply digitizing a long portion of the ambient noise on the power

line.

The transmit waveform consisted of a repeating OFDM symbol, specifically a

frequency sweeping "chirp" occupying approximately 2 to 30 MHz. After digitizing the

waveform at the receiver, ten consecutive chirps were time domain averaged to reduce

noise. Frequency domain division of the received signal by the transmitted signal (along

with calibration factors to remove the gain and/or frequency response of the amplifiers,

couplers, etc.) was used to estimate the channel frequency response.


Figure 5-1: Channel measurement setup










5.3 Channel Adaptation

Channel adaptation in communication systems enables fine-tuning of the

underlying PHY parameters to optimize the system performance. For periodically time-

varying PLC channels, channel adaptation should also determine the dependency of the

PHY parameters on the underlying phase of the AC line cycle. One simple means of

accomplishing this is by dividing the AC line cycle into multiple regions and determining

the PHY parameters independently for each such region.

Note that there is an important trade-off involved in the choice of number of

regions and the system performance/complexity. By increasing the number of regions, a

finer adaptation to temporal periodic channel characteristics can be obtained. However,

this also increases the complexity of the system. A larger number of regions requires the

transmitter and receiver to determine and maintain a larger set of PHY parameters and to

apply them with higher accuracy within the AC line cycle. The bandwidth and memory

requirements for generation, storing and exchanging the PHY parameters can be

prohibitively expensive. Apart from this, inefficiencies are also incurred at the MAC

Layer due to the need to limit the transmission of each physical layer protocol unit

(PPDU) to a single region. Since most practical implementations will apply a single set of

PHY parameters on a single PPDU, any transmission that spans across two or more

regions will have to be divided into multiple transmissions (or PPDUs) with each PPDU

occupying a single region. The extra overhead incurred in each PPDU transmission, due

to frame synchronization and control overheads at the beginning every PPDU, will

adversely affect the overall efficiency. For the analysis in this section, we use 130ms of

incurred overhead per PPDU, based on the HomePlug AV standard [33, 34].









By decreasing the number of regions, the complexity of the channel adaptation

mechanism can be reduced but the performance of the system suffers. This is due to the

need to choose "worst case" PHY parameters within each region so that reliable

communications can be guaranteed throughout the region. In the following section we

investigate the effect of region size on the PHY and MAC layer data rates.

5.4 Computation of Channel Capacity

To investigate the effect of cyclic variation of noise on the performance of a PLC

system, we first divide the 60Hz AC line cycle into 160 microslots. Each microslot

(@104ts) forms the basic atomic for estimating the variation of signal-to-noise ratio

(SNR) and Shannon Capacity [41] with respect to the AC line cycle. When computing

SNR and Capacity, the bandwidth from DC to Nyquist (37.5 MHz) was divided into 192

subbands. This value was chosen as it provides an intermediate level of frequency

domain resolution while at the same time enabling accurate noise power estimation as

twenty consecutive 384-FFT outputs (5.12[ts per FFT) may be averaged in each

microslot. This entire process is described in the following steps.

First, a fixed subband transmit power, STX, was used for all subbands. This fixed

power was picked to be comparable to the peak power spectral density allowed by

HomePlug standards and hence compliant with FCC regulations. (Since this is merely a

measurement of channel response, and not a data transmission, adherence to FCC

regulations is not strictly required, but resulting SNR measurements are realistic for a

practical system).

Next, the received carrier signal power, SRX is measured across the frequency band

of interest, and used to derive the channel attenuation, ac. Preliminary observations show









that the channel attenuation coefficients typically exhibit little variation throughout the

AC line cycle compared to the variation of the channel noise. Therefore only one set of

coefficients is used in our analysis

SI
a[c]= Sx for c=10,11,12,...,153
SR [c]

Noise power in the jth microslot, N.- was then computed by averaging over twenty

consecutive 384-FFTs, F, of the noise waveform as shown below,

F[i, j = fft(NoiseWf[f1 x 7812.5 + (i x 384: i x 384+383)]
where i = 0,1,2,...,19

19
N, [i, j = 2 F[i, j]
1=0

Finally, the Shannon capacity was used to estimate throughput in the jth microslot,

D[j],

153
D[j] = 17(153 -10 + 1)Slog,(+ )
c=10

For a PLC system that can adapt independently in each microslot, the average

channel capacity can be obtained by averaging the microslot capacities over a full AC

line cycle. In practice, supporting 160 channel adaptation regions is unduly complex. We

study the effect of a smaller number of regions on performance by grouping a varying

number of microslots to form a macroslot. For example, by grouping 20 microslots into a

macroslot, only 8 (160/20) independent regions need to be supported. The average

capacity of a macroslot that contains N microslots is obtained as follows,


D max{im i] ,N i 0..N









Where, {D[0], D[l], ..., D[N-1]} are the capacities of the constituent microslots

arranged in ascending order. This choice can be justified as follows. In each macroslot,

the PHY parameters could be chosen such that reliable trans-mission is guaranteed in all

its microslots. In this case, the minimum of the capacity for all constituent microslots

(i.e., D[0]) will be the capacity of a macroslot. However, for PLC system that can tolerate

errors at the physical layer (for example, by using MAC layer re-transmissions), it is

possible to achieve a larger aggregate throughput (in certain channel conditions) by

choosing more aggressive PHY parameters. In general, PHY parameters based on, D[i]

(where i>0), can be chosen. However, in such cases, transmission in "i" microslots will

not be successful as transmission rate will exceed capacity. Hence the average capacity


will be D[i]N- In our analysis we chose the PHY parameters that provide the
N )

optimal PHY data rate. However, it has to be noted that maximization of the PHY data

rates may not be the best choice under all conditions. For example, in PLC systems that

support Quality of Service (QoS) sensitive multimedia streams, the probability of

physical layer errors should also be bounded. Performance analysis for such cases is

beyond the scope of this study.

5.5 Discussion of Results

For each of the measured paths, we computed the average PHY capacity for various

choices of macroslot sizes. Since we are interested in the performance gain obtained by

AC line cycle based adaptation, we measured the gain in data rate in comparison with the

case when the macroslot spans a complete AC line cycle (i.e., no AC line cycle based

adaptation). Figure 5-2 shows the inverse cumulative distribution (i.e., 1-cdf) of the

percentage gain in PHY data rates ob-served over the measured paths for various choices









of macroslot size. The average gain in PHY data rates over all measured paths as a

function of macroslot size is shown in Figure 5-4 (solid line). These results allow the

following observations:

* Using a 1 millisecond long macroslot, more that 35% improvement in PHY data
rates can be obtained on 30 % of the paths,

* Using a 1 millisecond long macroslot, more that 55% improvement in PHY data
rates can be obtained on at least 15 % of the paths,

* Using continuous adaptation to cyclic noise (i.e., microslot based adaptation), a
30% increase in PHY data rates can be obtained (MAC overhead renders this
adaptation method impractical how-ever, as seen in Figure 5- 4),



MAC layer data rates can be computed from the PHY data rates by taking into

account the additional overheads incurred for transmissions that span multiple regions (or

macroslots). We compute the MAC level throughput as follows,

MacroSlot Size in ps -130/us
MAC Throughput = PHY Throughput M t S i
MacroSlot Size in ps

Note that a 130[ts incurred overhead per PPDU was used for this analysis, based on

the HomePlug AV standard as mentioned earlier. Figure 5-3 shows the inverse CDF of

the percentage gain in MAC layer data rates observed over the measured paths for

various choices of macroslot size. The average gain in MAC layer data rates over all

measured paths as a function of macroslot size is shown in Figure 5-4. It is notable that

MAC performance suffers when the macroslot size is chosen to be either too large or too

small. In the former case, the drop in PHY data rates is due to lack of proper AC line

cycle based adaptation. In the latter case, excessive PPDU overhead incurred due to the

large number of regions negatively impacts the MAC performance. For the paths tested,

these results can be summarized as follows,









* Using 1 millisecond long macroslots, more than 30% improvement in MAC layer
data rates can be obtained on 20% of the paths,

* Using a macroslot that spans 1-2 milliseconds pro-vides the highest MAC
performance,

* AC line cycle based adaptation can provide an average of 10% improvement in
MAC data rates,

* In 40% of the paths, less than 5% improvement in MAC performance was
observed. This suggests that AC line cycle based adaptation does not al-ways
improve performance. The choice of using AC line cycle based adaptation should
be made selectively based on the observed channel characteristics.

5.6 Conclusions

In this section, we investigated the performance enhancements that can be achieved

by using a channel adaptation mechanism that is synchronized to the AC line cycle.

Analysis of channel characteristics on 72 channels has shown that an average of 10%

improvement in MAC data rate can be achieved by using AC line based cycle adaptation.

Furthermore, on 20% of the paths, we have seen more than 30% improvement in MAC

layer data rate. The analysis also showed that using 1-2msec regions provides the largest

gain in MAC performance for most of the channels. Apart from an increase in

throughput, AC line cycle based channel adaptation also enhances the stability of channel

adaptation and allows for better control of physical layer errors. Both of these features are

invaluable for supporting QoS applications over power lines.

The results presented in this section can be considered preliminary as they are

based on 72 paths from two buildings in North America. Further research incorporating a

larger number of channels, as well as taking into account both the frequency response and

noise power variation with the AC line cycle, would help to fully understand the potential

performance gains offered by the use of line cycle based adaptation in PLC systems.








54




100 -
f0.1 msec
90 2-t2_ ----- --- ------------ -- ----0---------- 0 2 msec
: .---- 0.4 msec
8O '-- .-- ---------- --- ------- -- -- 1 0 m sec
-e- 2.0 msec
S70 ------ _-- I -----------... ------ ------ 4 O msec
--- 8.0 msec


0.
4-50 ------ --

S0 -- ----- -- --- -- --- ----------- ------
a-
4j 0 ---_ --- -- -l -- ------ *-- -^- ---- ------ -------------- -______

30 ----- --- --- ^ --* ------ --- ---- -- ---- ------- ------- -------.
c-


20 ------ ----- --- ------- --- ---
10 ...-------------... ..

0
2n-111


0 10 20 30 40 50 60 70 80 1 i0
Percent gain in physical layer data rate


Figure 5-2: Performance gain in PHY data rates as a function of macroslot size



100 -------------------------------- ------ -------------- ------ -------
-- 0 1 msec
90 ---------------- ----- ----- -- -- --- --- I ------- -. 0.2 msec .
S-- 0 4 msec
S0 -- 1.O msec
80 ------ ------ -------------- ------ ---------------- -I 1.0 msec .
-e- 2 0 msec
_U I 1...... ........................................... -0- 4.0 msec
70 -----.------ ------------------ ------ 4...... msec
->-- 8 0 msec
L7-
60 G Fp----------- --------------I-------- ----- ----- --------I-------- ------.



40 -.. ...---- ------- --------..------ I------ .------- -------



20




0 10 20 30 40 50 60 70 80 90 1
Percent gain in MAC layer data rate


Figure 5-3: Performance gain in MAC data rates as a function of macroslot size














50 -----------------------------------------
PHY
40 --------------------- ---- --------- --- ------- MAC

30 -






-10 --
10 ----- ---------- I -------- r-------- ------- ------- ------- -----.- -"------





0--- ------------ ---------------------------------------



-30 ..
-30 ---------------L-------L------- ------- ------- ------- ------- -------

-40 --- --------------------------------------------------

-50 I I I I I I
0 2 4 6 8 10 12 14 16 16
Macroslot Size (msec)



Figure 5-4: Average gain in MAC and PHY data rates














CHAPTER 6
MAC FRAMING

In power line communications (PLC) more so than in many other media, there is a

strong interaction between the Medium Access Control (MAC) layer and the Physical

(PHY) layer. This chapter investigates one of the manifestations of this interaction, that

of MAC framing and error recovery with PHY transmission characteristics. MAC

framing for high-speed PLC systems should be capable of concatenating multiple MAC

Service Data Units (MSDUs, e.g., Ethernet or IP packets) in order to amortize the high

and fixed overheads associated with each MAC Protocol Data Unit (MPDU). By itself,

this is easy. However, transmission errors may damage portions of an MPDU, which can

further rob a system of efficiency. This chapter presents a 2-level MAC framing

mechanism that enables efficient re-transmission of corrupted data. Both analytical and

simulation results are used to compare the performance of the proposed MAC framing

scheme with other existing viable approaches. The results show that the proposed scheme

is significantly superior to existing approaches. A subset of the results presented in this

chapter is also published in [42].

The organization of the remaining sections is as follows. Section 6.1 gives an

overview of MAC framing and provides an intuitive definition of Framing Efficiency. In

Section 6.2, various approaches for MAC framing are presented. Simulation and

analytical results on the performance of each of these approaches are presented in Section

6.3. Performance comparison between various MAC framing approaches and conclusions

are presented in Section 6.4 and Section 6.5 respectively.









6.1 MAC Framing and Framing Efficiency

MAC framing deals with the process of converting Service Data Units received by

the MAC from higher layer (i.e., MAC Service Data Unit, MSDU) into a Protocol Data

Unit that the MAC handovers to the physical layer (i.e., MAC Protocol Data Unit,

MPDU). MAC framing process not only depends on the PHY, but also on MAC features

like channel access mechanism, network architecture, Layer-2 encryption etc. For

example if one of the MAC features is to provide privacy, the MAC framing process will

include fields for passing encryption relevant information. For a token passing MAC, the

MAC framing process will contain fields to indicate busy token etc.

When the underlying PHY is unreliable, MAC framing will also have to add fields

for error detection and retransmissions. Thus the MAC framing process depends heavily

on system details and the approach used at the MAC layer to address them. Once the

MAC framing process generates an MPDU, it is handed over to the physical layer. The

physical layer adds the PHY headers and transmits it to the destination (refer Figure 6-1).

Apart from the physical layer overheads, MAC protocols also incur overheads due

to interframe spacing, transmission of control information, collisions etc. All these

overheads along with the MAC framing overheads will determine the MAC Efficiency.

In this chapter, we are interested in studying the performance of the MAC framing

process with respect to recovery of errors. To decouple analysis from other system level

details, we assume that the only overheads incurred are those required for error recovery

at the MAC layer. We define the MAC framing Efficiency to be the average number of

MSDU bits delivered per MPDU divided by the total bits per MPDU. MAC framing

Efficiency depends on PHY errors, MAC framing approach, MSDU size, and MPDU

size.







58




MSDU(s)



MAC Framing
(Concatenation,
Segrrnitation etc)
MAC







Header PHY Payload Layer
orl ins MeiLayium

---------------Di -----------
Powavrlins Medium


Figure 6-1: MAC framing Process

6.2 MAC Framing Strategies

Concatenation is essential for high-speed PLC systems due to the large and fixed

overheads per MPDU. These occur due to PHY overheads and interframe spacings. In

Home-Plug 1.0 [11], each MPDU transmission will incur a minimum overhead of 351.02

microseconds. If a system is ported to 100 Mbps without any concatenation, the

maximum efficiency (assuming 1518 byte Ethernet packet as the MSDU) that can be

obtained per transmission is -25%, even in the absence of errors. Thus, concatenation is

essential.

Three different packet concatenation mechanisms are investigated here. All three

assume that some of the contents of an MPDU may be deliverable even if the MPDU is

partially damaged. Without this stipulation, the efficiencies are so low in the presence of

errors that these approaches are not considered further. In each case, we assume that the

MPDU body is sent in one or more fixed size FEC blocks of a PPDU.









Given that multiple MSDUs may be concatenated to form an MPDU, we consider

several basic stages. First, the MPDU may be acknowledged as a whole either all of the

constituent MSDUs are received correctly or none are, thereby requiring retransmission

of the entire MPDU contents. This approach has a very low overhead, but is too

inefficient when there are errors. The second main stage is to allow each MSDU to

occupy its own, individually acknowledged, MAC frame. In this case, only the MAC

frames that could not be recovered from a damaged MPDU must be retransmitted. We

consider two variations of this approach; Simple Concatenation and Explicit Demarcation

based Concatenation. Finally we consider two-level concatenation. Here, the individual

FEC blocks hold their own header information, which allows retransmission of

uncorrectable FEC blocks as well as resynchronization of MAC frames.

6.2.1 Simple concatenation

This is a very straightforward extension of the case without concatenation. Each

MSDU is prepended with a Sequence Number (SN) and a Length Field (Len). A CRC

based error detection field in attached at the end of the MSDU. Sequence numbers are

needed to ensure in-order delivery and the Length field enables demarcation of each

packet at the receiver. Thus, multiple MSDUs can be concatenated at the transmitter and

de-concatenated at the receiver. Figure 6-2 shows the MAC Frame format for simple

concatenation. This approach has two main disadvantages. Firstly, each MPDU has to be

padded to fit into the PPDU. Secondly, a single FEC Block error will result in loss of

information contained in all subsequent FEC Blocks.



















MPDU {


F


MSDU
I seU I


SN Len MSU I CRC




I.IAC: Fid'li'i a 1 I.9AC Fi s a 2 N O i-rainlw r., ra |


PPDU PHY- Hader h LUbMK rrWcK rICKd


PHY Payload

Figure 6-2: Simple concatenation

6.2.2 Concatenation with explicit demarcation

One of the main disadvantages of simple concatenation is that all information

following the location of the first FEC block error is lost. This loss is mainly due to the

loss of MAC Frame demarcation. So, it is quite likely that several MAC Frames might

have been received without errors at the receiver but the receiver has to discard them due

to loss in demarcation information. One simple approach is to add a Header Check

Sequence (HCS) for detecting errors on the header fields in the MAC Frame. Thus,

whenever the receiver detects a MAC Frame with errors, it can scan through the received

bits to check for the next header. Once a proper header is confirmed by means of the

HCS, it has the necessary demarcation information for receiving the remaining MAC

Frames within the MPDU. MAC Frame header should also include an Identifier (ID)

field. ID can be set to zero for the first MAC Frame in an MPDU and incremented for

each subsequent MAC Frame. This enables the receiver to use a simple bit map for

indicating the reception status of each MAC Frame within a MPDU. Figure 6-3 shows

the MAC Frame format for Concatenation with Explicit demarcation.










nMSDU


SN Lan ID HCS MSDU CRC




MPDU MACFrame#1 MACFrame#2 ..MAFramem rlI'


PPDU PHY Headet a R | C


PHY Payload


Figure 6-3: Concatenation with explicit demarcation

6.2.3 2-Level framing

PHY errors occur on an FEC block basis. Hence, an optimal framing strategy is one

in which only the corrupted FEC blocks need retransmission. 2-Level framing is based on

this principle. Here, framing is done in two steps.

In the first step each MSDU is prepended with a Length (Len) field. This entity is

referred as a MAC Frame. Multiple MAC Frames that are awaiting delivery are

concatenated together to form a MAC Frame Stream. MAC Frame Stream thus generated

is treated as a stream of bytes and is then divided into segments of fixed size that match

the FEC Block size. Each segment is prepended with a Sequence Number (SN) and a

MAC Frame Boundary Offset (MFBO) field. The SN provides the relative location of the

Segment within the stream while the MFBO field indicates the location of the first MAC

Frame boundary (if any) within the Segment. A CRC based error detection field in

attached at the end. The resulting entity is referred to as a PHY Block. Each PHY Block

is mapped on to FEC Block at the PHY layer (refer Figure 6-4). It might be necessary to


A







62


add a pad to the MAC frame stream when there are not enough bytes to transmit a


Segment. Note that a pad is not necessary if there are enough pending MAC Frames. All


the results presented in this section assume that the padding overhead for 2-Level


Concatenation is negligible.


MSOU


Len MSOU
N A



SMACFrame# MAC Frame #i+l



ISegent Segrrent Segment
k I kI +2
"*nr n~-


SN MFBO Seg t CRC
I Ik+1-


N.
S<


IhYHaad#2s #n#1 #2


PHY Payload


Figure 6-4: 2-Level MAC framing

Table 6-1: MAC framing overheads
SN Len HCS ID CRC MFBO
Simple 2 2 -- 4
Concatenation
Explicit 2 2 2 1 4
Demarcation
2-Level
2 2 4 2
Concatenation


MAC Frame f
Stream








PHY Block



PPDU


MAC Frame #j j Fa


| I I
Il lI

Iun begant"I"'









6.3 Efficiency Analysis

We have derived equations for the efficiency of the Simple Concatenation and the

2-Level Framing schemes. For Explicit Demarcation based concatenation, simulation

results were used to obtain the efficiency. For the results presented in this section, the

overheads shown in Table 6-1 are used. We also assume that all MSDUs are of the same

size. Results are presented for the case of 520-byte FEC Blocks [33, 34]. The FEC block

errors in an MPDU are assumed to be independent of each other.

6.3.1 Simple concatenation

In the following analysis of Simple Concatenation, we assumed that a single FEC

block error would result in loss of all MAC Frames that are contained (either partially or

fully) within it. Due to the loss in demarcation information of MAC Frames, all

information subsequent to the first FEC Block error is also lost. Thus, if kth FEC block is

the first FEC block that was corrupted the number of MAC Frames that will be

successfully delivered is given by kLfr, where Lfec is the length of the FEC Block and
[L-f

Lmfis the length of the MAC Frame. Lmf is the sum of the size of the MSDU, Lmsdu, and

the overhead added per MSDU, i.e., LoH.Sc. Ifp is the probability of an unrecoverable

FEC Block error and Nis the number of FEC Blocks, the efficiency for Simple

Concatenations rs7 is given by,


Nw-1 k L k 1 Lmdu
1se = L (1I- )k Lfec (I fecj du
k1 L -f L f NLfec

Figure 6-5 shows the efficiency of Simple Concatenation as a function of FEC

Block error rate for N = 25 and various MSDU sizes. Figure 6-6 shows the efficiency as a

function ofN forp = 0.1 and various MSDU sizes.









6.3.2 Concatenation with explicit demarcation

For Explicit Demarcation based concatenation, we assume that a single FEC block

error will result in loss of all MAC Frames that are contained (either partially or fully)

within it. The number of MAC Frames successfully delivered will not only depend on the

number of FEC Block errors, but also on how and where they occur. For example, when

two side-by-side FEC blocks have errors, then they tend to produce fewer MAC Frame

errors than if they are far apart. Hence, it is not possible to derive a simple analytical

result for this case. Simulation results were used to study the performance of this MAC

framing Scheme.

Figure 6-7 shows the efficiency of Concatenation with Explicit Demarcation as a

function of FEC Block error rate for N = 25 and various MSDU sizes. Figure 6-8 shows

the efficiency as a function ofN forp = 0.1 and various MSDU sizes.

6.3.3 2-level framing

With 2-Level Framing, only FEC Blocks that were corrupted need to be

retransmitted. Ifp is the probability of FEC Block error, then the average number of FEC

Blocks delivered per MPDU is (l-p) times the number of FEC Blocks, The efficiency

with 2-level concatenation, r2L is given by


72L i L(feL L OHsb2L L du
2ZL = (--\ L
Lfec Lsb )

where LOH, sb, 2L is the overhead per FEC block.

Figure 6-9 shows the efficiency of 2-Level Concatenation and Segmentation as a

function of FEC Block error rate for N = 25 and various MSDU sizes. Figure 6-10 shows

the efficiency as a function ofN forp = 0.1 and various MSDU sizes.












Simple Concatenation
100r---- -- 7 -------- ---------------- 7 --------- -- --------------------
100
1518
-Ksdu
90 ------ --- d 791
90 -------- --- --------- ------ .Lmsdu = 791










140

30 --u
70 --- ^'---- ---------------------- ------ I-------. ---------------I ------ 1

| 60 ------- -- -.$ ------- -------------------- ---- ----- ----- ------

L,, 50 ------ ------- ---- --- ---- ------ I ------- I------- ------ I ....... __ I

40Ar ------- ------ I-------- *s -^ ------ ------ ------. ------- ------ -------




20 -
--- -- ---------- -----------------------
30 ------ ------- -------- - 1- ---:z ^^ -_ __ _


10 --------- --------- -------------. ------- ------ -------,-------,------- --------,

0 --
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
FEC Block Error Rate


Figure 6-5: Simple concatenation- efficiency with varying FEC block rrror rate and N
25


Simple Concatenation


90 ------ ---------- ----- ---------- -- ---- --- .sdu = 791
90 :-- .. d =7

0---- mdu= 64
80 -- ------- ---------=791-
0 ---- --------- ----------- --- ----.-



60 I '

uj 5 --- --\ ------- --------------- T- ------- --------------------- --------

U 50~ --4, -'- ----- -----v~- - ---I ---- T------ -- -- -- -- -- ------- ------ I




30 "~~ ~ ~ ------ I-------- ------------ ------ ------- I-- ---- I --------- -------.

10 -L4- ---- -- ------ ------- ----- --------- -- --


0
0 5 10 15 20 25 30 35 40 45 50
Number of FEC Blocks per MPDU


Figure 6-6: Simple concatenation efficiency with varying number of FEC blocks and p
0.1








66



Explicit Demarcation based Concatenation
100 ------- 7 -------- ---------------- 7 -------- -- --------------------
10 ---

90 T-"-- -- ------ -- ...- Lmsdu = 791
S,- ---_ ---- Lms=du= 64
80 -T -- ----- ------------ -- --- --- ----_ L -791
0 -------------- -- --- ----------- .. L m. .



0 --- -
-- i < '* i

SLU 5 --- --- --------- ---, ----



30 5 ------ -------- --------------- .------ ------, ------. ------------- ------.




20 C --_
0 -- ------ ------ --------------r------- ------ ------ -------- ------- -i -






10
C0 I I I I -

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
FEC Block Error Rate



Figure 6-7: Explicit demarcation based concatenation efficiency with varying FEC
block error rate and N= 25


Explicit Demarcation based Concatenation
100 ------- -- -------------------- -------------------------
100-
Lmsdu =1518
90--- -- Lmsd =791
90 ------- ----- ----------- -------- --- --- T --- L.6 = 791"5
80---- Lmsdu= 64
80 ------- -- .----- ----

S 0 --t -----. --'-------- --- --




S 5------ -- -- -
7 0 ------ ------ ---------- ------ -- -- -- ----------------I ------.



a, I
30 --_r --- ,- ---- -- ------ ----
3U 0 4[- 1- ------ ------- ------- ------ T ------ --------------- r------r ------ I

40 "~~ ~~ ------ ------ I -------- --------- -------, --------------------- -------.

30 -- ---- I-------- --------I ------I ------ I ------ I--------- I------I ------ 1
I

20




0 5 10 15 20 25 30 35 40 45 50
Number of FEC Blocks per MPDU



Figure 6-8: Explicit demarcation based concatenation efficiency with varying number of
FEC blocks andp = 0.1








67



2-Level Concatenation
100 ---

90 ------- ---- ---





L-i



30 -- -------- I --------I
20 -- --.-.-.- T ---- I-- --- -- I ----. -----T------ -- -

0 -C------.------------------- ----- ----------------- .--- -------








K -- du = 64

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
FEC Block Error Rate



Figure 6-9: 2-Level framing efficiency with varying FEC block error rate and N= 25



100 --

90 --
80 ------ ------ -- ------ ------ ------- ------ ------ ------- -------- ------ -
O --- --- ------ --------------------- ------ --------------------- ------








S30 -------------- ------ ------ ------------- ------






.Lsdu = 1518
EU 40 ------- ------ I-- ---------- L-----^ ------ ------ -------,------- L------- L -------,



30 -------------- ------ ---------------.------ -------,-- - -

20 ------- T- -- -- ---- -- -- -- -- ---- -- -- ---- T- -.------

10 ---------------- ------- -------- ---------. -- L-msdu= 791
Ssdu 64

0 5 10 15 20 25 30 35 40 45 50
Number of FEC Blocks per MPDU



Figure 6-10: 2-Level framing efficiency with varying number of FEC blocks and p =0.
,, ,, ,









Nubro.FCEoksprMD

Fiue61:2Lvlfaig-efcec wt ayn ubro E lcsad .









6.4 Comparison of Performance between various Framing Schemes

Using Simple Concatenation, a single FEC block error results in loss of all MAC

Frames that it intersects and all subsequent MAC Frames in the MPDU. This approach is

very sensitive to FEC block error rate, p, and number of FEC blocks per MPDU, N. For

example at 10% FEC block error rate, the efficiency drops to 30%. The effect of MSDU

size on efficiency is small compared to that of the FEC block loss probability. Larger

MSDUs will also incur higher overheads due to padding (since a number of whole

MSDUs are required to fit in each MPDU), thus resulting in lower performance.

Using Explicit Demarcation based Concatenation, a single FEC block error results

in loss of all MAC frames that it intersects. Simulation results show that the efficiency

loss is linear withp and is strongly dependent on the MSDU size. At lowp, smaller

MSDU sizes show poor performance due to framing overhead. However, asp increases,

larger MSDUs sizes suffer higher loss in efficiency since each single FEC block error

results in larger number of MAC frame bits lost.

With 2-Level Concatenation, only FEC blocks that were corrupted need to be

retransmitted. Results show that the efficiency is linear with respect top, and

independent of N. Figure 6-11 shows the relative performance of the three framing

approaches for 1518 byte MSDU size with varying FEC block error rate. The results

show that the performance of 2-level concatenations is superior to that of other

approaches.













100 r- ----- -------- ----- --- --- I -------I------- --I------I- --1




70 ------- ---- ----------- --------- ------------------ -------I------
0 -- -
0 ------ ------ -- ---- --- ------ ----- -



40

30 --------------------- ------------- ------ ------:----------------------




O -- Simple Conc.
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
FEC Block Error Rate


Figure 6-11: Performance comparison of various MAC framing strategies with varying
FEC block error rate, N= 25, Lmsdu = 1518

6.5 Conclusions

This chapter has highlighted the necessity of concatenation when framing at the

MAC level in high-speed PLC communication systems. Five approaches were

considered, with three analyzed and simulated in some detail. Approaches varied

according to their ability to retransmit at the MAC frame or the FEC block level, and in

their ability to resynchronize at the MAC frame level. The results show that 2-Level

Framing was far superior to the others in its efficiency. It was also seen to be

independent of the number of FEC blocks per MPDU, unlike the other two approaches.














CHAPTER 7
ALLOCATION REQUIREMENTS FOR SUPPORTING LATENCY BOUND
TRAFFIC

Impulse noise sources such as dimmers, halogen lamps etc., are quite common in

power line networks. Hence, high-speed power line communications systems should be

designed to deal effectively with them. HomePlug AV uses aggressive channel adaptation

at the physical layer and efficient retransmissions at the MAC layer to deal with impulse

noise. This approach can lead to very high probability (up to 20%) of Forward Error

Correction (FEC) block errors under some channel conditions. HomePlug AV also has

large transmission overhead to ensure reliable communication over the power line.

Transmission overhead can negatively impact the MAC efficiency on channels with high

FEC Block Error Rate due the need for a large number of retransmissions to successfully

deliver packets.

This section studies the complex interaction between FEC block errors,

transmission overhead, Quality of Service (QoS) requirements of a multimedia stream

and its TDMA allocation requirements. Our results show that, to support low latency

applications at high BLERs, the average allocation size has to be increased by one or

more transmission overheads. Further, the increase in average allocation is not a strong

function of application data rate. Hence, low data rate applications can suffer significant

loss in efficiency at high FEC Block errors. Our results also show that using a fixed size

TDMA allocation can lead to poor utilization when the latency requirements of the

application are less than four beacon periods. These results indicate that channel









adaptation in HPAV should take into consideration the application QoS requirements to

optimize the overall system capacity.

7.1 Introduction

The HomePlug AV (HPAV) Medium Access Control (MAC) layer is based on a

hybrid of Time Division Multiple Access (TDMA) and Carrier Sense Multiple Access

(CSMA). Each HPAV network has a designated device, the Central Coordinator (CCo),

which coordinates access to the medium. The CCo periodically (with a period of either

33.33msec or 40msec [33, 34]) generates a beacon with information about the temporal

locations of TDMA and CSMA allocations. Multimedia applications use TDMA

allocations to provide guarantees on QoS while CSMA allocations are used by regular

data traffic. Since power line channels can change with time, each station continuously

provides information on the physical layer (PHY) data rates and the size of its queue (i.e.,

backlog) to the CCo. The CCo uses this information to dynamically adjust the size of

TDMA allocations.

HPAV stations use channel adaptation to choose the best set of physical layer

parameters (carrier modulation, FEC code rate etc.,) for communication with other

stations. Channel adaptation in HPAV is also tailored to effectively deal with impulse

noise [43, 44]. In most communication systems, the channel adaptation mechanism is

used to reduce the PHY layer errors to a range where they can be effectively corrected by

the MAC layer. For example, the channel adaptation can be chosen such that the FEC

block errors are less than 1/100 and the MAC retransmissions can be used to recover the

uncorrectable errors. Such a technique is not viable under power line channel

environments due to the presence of impulse noise. Impulse noise channels will display

temporal effects in channel capacity. In between two impulse noise events, the channel









capacity can be very high, while at the impulse noise the capacity is very low. To obtain

less than 1/100 FEC block error rate, the channel has to be adapted close to the lower end

of its capacity (i.e., capacity near impulse noise events) and hence is undesirable when

high throughput is required. A better approach is to adapt aggressively and use a MAC

ARQ mechanism to correct errors. In some environments, to achieve optimal

performance an HPAV station may need to operate at FEC block error rate as high as

20%.

To effectively recover from large number of FEC Block errors, HPAV uses a novel

2-level MAC framing mechanism and selective acknowledgments (SACK). The 2-level

MAC framing essentially concatenates and then segments higher layer packets (or MAC

Service Data Units) into fixed size block (referred to as Segments) that fit into a single

520-octet FEC block. Since errors at the physical layer occur on an FEC block basis, only

Segments that get corrupted needs to be retransmitted. Further details on 2-level MAC

framing can be found in Chapter 6 [42].

Apart from the MAC framing overheads, each transmission will also incur

overheads due to physical layer headers, selective acknowledgments and interframe

spacings. Figure 7-1 illustrates the overheads incurred in HPAV during TDMA

allocations. Segments are transmitted in Long Physical Layer Protocol Data Units

(PPDUs). Each long PPDU consists of a Start of Frame (SOF) delimiter followed by one

or more FEC blocks that carry Segments. A Delimiter in HPAV consists of a preamble

sequence followed by 128-bit frame control information. Preambles are used to

demarcate the start of each PPDU. A SOF frame control contains broadcast information

that is necessary for proper demodulation of the PPDU payload. The frame control of a









SACK delimiter contains the reception status of the FEC Blocks in the Long PPDU.

Since proper reception of the delimiter is critical for reliable exchange of data, delimiters

in HPAV are designed to be highly robust and hence are quite long (110.48[tsec).

Typically, a Response Interframe Spacing (RIFSAV) of 100[tsec is used between the

transmission of long PPDU and the corresponding SACK. A SACK and the subsequent

Long PPDU are typically separated by 100[tsec (note that HPAV allows variable

interframe spacing to enable flexibility for a variety of implementations). Thus we see

that each PPDU transmission will incur an overhead of around 420[tsec.

Due to the fixed transmission overhead in HPAV, the retransmission strategies can

greatly influence the MAC efficiency. For example, Figure 8-1 also shows a scenario

where Segment contained in FEC Block#2 gets corrupted during the initial transmission

and is retransmitted immediately. Since a 420[tsec overhead is incurred for retransmitting

a single Segment, this reduces efficiency. In this section, we refer to this retransmission

strategy as 'Immediate Retransmission' strategy. A strategy that can overcome the

inefficiencies of immediate retransmissions is shown in Figure 8-2. In this approach, the

transmitter waits until sufficient numbers of new Segments are available before

retransmitting the corrupt Segment(s). We refer to this retransmission strategy as

'Delayed Retransmission' strategy. In essence, the delayed retransmission strategy

reduces the average overhead incurred for transmission of each segment, thus improving

the efficiency. The drawback of this strategy is that, additional delays are incurred while

the transmitter aggregates new Segments. While such additional delays are not an issue

for data traffic, they may not be acceptable for QoS sensitive multimedia streams.











Multimedia streams require guarantees on both latency and packet loss probability.


Since multimedia streams are provided with one TDMA allocation per beacon period,


each Segment will have a fixed number, L, of TDMA allocations within which it need to


be delivered. In contrast, to achieve a given packet loss probability, Segments belonging


to the applications will need to be guaranteed a certain number of transmission


opportunities, TXOPS. In situations where L > TXOPS, delayed retransmissions can be


used. Otherwise, a hybrid of immediate and delayed retransmission strategies has to be


used to meet the packet loss requirement. Inefficiencies will be incurred in the later case,


due to additional transmission overheads.




Control
L--- ID 4is-a]



S' <- "--- -- -Immediately retransmit
-" -... "" corrupt segments


39F FE:. FE. FE.- SACK SOF FEC SACK
D'I.m.-Ir Boi #in BItoijd BloDdcJ Delimiter Delimiter Bl0oc#1 Delimiter


520 Octets FEC Block RIFS AV CFIFS AV RIFSAV
Error (10iU is) (ICopS) (lOOps)


Long PPDU Short PPDU


Figure 7-1: Transmission overheads in HomePlug AV

elaytda reowsnmr.
peg----------------------------- F........... F tat


7": 'E 1e,, ,- e". ....... '' :. 'c", '"1.. -. ...
vi *: v. *I 6 V.


FEr Blrck Errar Wait urrl eufMI fll number Snew Sgrrent arn be FE EC Bnd Eirr
accurnulatd bhci- re rerar'nissin dfcorTup Segmrneis

Figure 7-2: Illustration of delayed retransmission strategy

In this section we analyze the effect of FEC block errors and transmission


overheads on the allocation requirements for the multimedia streams. The characteristics









of the streams that are of interest are its data rate, latency and packet loss tolerance. We

are interested in two different types of allocation requirements,

1. Fixed Allocation This is the duration of the fixed size TDMA allocation that is
required to satisfy the QoS requirements of the multimedia stream. It is important
to note that due to the statistical nature of the FEC errors, it is possible that the
Fixed Allocation may not always be fully utilized.

2. Average Allocation This is the average duration of a variable size TDMA
allocation that will be required to satisfy the QoS requirements of the multimedia
stream.

To simplify our analysis and presentation of the results we make the following

assumptions,

* We assume that the application generates constant bit rate traffic. Application data
rate is measured in terms of the number of Segments (N), that it generates per
beacon period,

* We measure the application latency L in multiples of beacon period. Each Segment
will have transmission opportunities in L number of TDMA allocations before it is
considered stale.

* We measured packet loss probability in terms of the packet loss event probability
(PLEP) and average packet lost (APL) per packet loss event,

* We assume that FEC Block errors are uncorrelated and occur with a probability (p)

* Since the efficiency of a TDMA allocation is primarily dependent on the ratio of
the transmission overhead to transmission duration of a FEC block, we do our
analysis in units of FEC Block transmission duration. For example, at physical
layer data rate of 150Mbps, the transmission duration of an FEC block is 27.7msec.
Hence the ratio of transmission overhead to FEC block transmission time
(represented by H) is 15.16 (i.e., 420/27.7). Since the maximum physical layer data
rate of HomePlug AV is 150Mbps, we are interested in H e (0, 15.16].

The organization of the rest of the section is as follows. Analysis of Fixed

Allocation and Average Allocation requirements without transmission overheads is

presented in Section 7.2. An approximate analysis of Fixed Allocation and Average

Allocation requirements in the presence of transmission overheads is presented in Section

7.3. This section also includes simulation based validation of the analysis. The effect of









transmission overhead on allocation requirements is presented in Section 7.4 and

conclusions are presented in Section 7.5.

7.2 Allocation Requirements without Transmission Overheads

In this section we investigate the allocation requirements when no transmission

overheads are present. In practice, it is not possible to design systems that do not such

overheads. Hence, the results presented are of theoretical importance as they provide

optimal bounds with which to compare actual achieved efficiency.

To successfully transmit each Segment, an average of (1 p) number of

transmission opportunities will be needed. Hence the Average Allocation, T7 ,

requirement for transmitting N Segments is given by,


Tg (N, p)= )

The computation of the Fixed Allocation that can enable the required PLEP is more

involved and depends upon whether L = 1 or L > 1.

7.2.1 Fixed Allocation, L = 1

Fixed Allocation for L=1 can be computed by nothing that the probability that k out

of N Segments remain when an allocation of size A (A>N) is granted on a channel with

FEC Block error probability p, is given by



Pr m Segments remain/ ( )N /N, p, A) N A
1-1 k PA-Nk) N-k; n=O
k=l \k)

From this, the packet loss event probability (PLEP) and Average packet loss (APL)

per packet loss event can be obtained for an allocation A, as follows,










PLEP = (N- m)(1-
mn1 m
NA
APL A= (Nm)(-p) m


Based on the above equation, the Fixed Allocation T can be obtained by

determining the minimum value of A that will achieve the required PLEP.

7.2.2 Fixed Allocation, L > 1

Analytical solution for the Fixed Allocation cannot be obtained for L > 1. Hence

we resort to a Markov chain [45] based analysis. This analysis assumes a retransmission

strategy in which the oldest segment in the queue will be transmitted before transmitting

any other Segments. Since no transmission overhead is incurred, this strategy is optimal

in the sense that it provides the retransmission opportunities to the most urgent Segment.

First, consider the buffer backlog of the stream at the end of each beacon period. It

can be readily seen that the variation of the buffer backlog is only dependent on the

current backlog and the size of the Fixed Allocation for a given FEC block error

probability. Furthermore, the maximum buffer backlog is bounded by N(L-1). Note that

this upper bound is a manifestation of latency bound associated with the Segments. A

discrete-time, finite state Markov chain model can be used to determine the steady state

probabilities of buffer backlog from which the PLEP can be computed. Figure 7-3 shows

the state transition diagram of buffer backlog when L > 2.

The state transition probabilities, p, are given by


P =Pr(N+iA) O 0 otherwise








Where, Pr N i, p, A is given by,


Pr N i, p,A


A A-(N+-J)( p)N+z-j ; 0 < j< N

1-Zi A A(N+,-k) N+k; j
kL- I k--


The steady state probabilities, pNP, can be obtained from the state transition

probabilities by solving the equation,


[NP1][pP]= [p


PL2NL-2N.NL PNL2NL-2 N+1
Figure 7-3: State transition diagram for no MPDU overhead case
From the steady state probabilities the packet loss event probability and average

packet loss per packet loss event can be computed based on the set of transitions that
result in a packet loss. Figure 7-4 shows the excess allocation required to achieve a
packet loss event probability of 10-5 forN = {50,100, 200} and L = [1, 2, 3]. Note that a

PLEP of 10-5 will result in one packet loss event in an hour and hence is a reasonable







79


choice for most multimedia applications. Also, N = 50, 100 and 200 correspond to

application data rates of 6Mbps, 12Mbps and 24Mbps, respectively when the beacon

period is 33.33msec. The excess allocation is measured as the difference between the

N
actual and the minimum allocations required to ensure stability (i.e., ). These results
1-p

show that for L = 1, a large amount of over allocation is required to achieve low packet

loss event probability. However for L > 1, the amount to over allocation required is

negligible. Figure 7-5 shows the average number of packets lost per packet loss event for

N = {50,100, 200} and L = [1, 2, 3]. Results also show that the APL tends to be small

and show a slight increase as N, L or p increases.


40 ..|-:I -I-I---- ....
-4- L=1,N=50
-B- L=1,N=100
35 _-- L=1,N=200 --.- ------- ...
-a- L=2,N=50 /
30 L=2,N=100,200 .. ....
-t-- L=3,N=50,100,200 /
c -25 ------- -------------------

..................--- -- ii ..... ........ --.---- -ii .. ......... i ..... i
20 .. ----- -- .--.--,- ,-,- ,, ... ---.--- .- .- .- --- ... --- -- r- .- -^-^^








103 10.2 10' 100
FEC Block Error Probability

Figure 7-4: Excess allocation as a function of FEC block error probability when no
transmission overhead present







80




,- 4 N=50 --- ----- ------------- --- -- -- ---
B- N=100
-2- -.-- N=200
0 ------ i i i i I- i i i i I
0I
10-3 10-2 10-

4 N=50 ---1 1
0 N=200 ', I ', T ', ', -
10 10 10 1




4 N=50 ----- --'-- ----------- i I--- --- -- -
B N=100 o

0- , ,

103 10.2 10"1



Figure 7-5: Average packet loss per packet loss event when no transmission overheads
are present

7.3 Allocation Requirements with Transmission Overheads

Results from previous section show that when no MPDU overhead is incurred, for

latencies greater than one, the allocation requirement are very close to the optimal

N
allocation i.e., In the presence of transmission overheads, the retransmission
p


strategy used in the previous will result in a Fixed Allocation and Average Allocation of

N
S(1 + H). It can be readily seen that for the range of values of H we are interested
(I p)

in, this can result in gross inefficiencies and hence is not viable.

A better approach is to use a hybrid of immediate and delayed retransmission

strategies where in at each transmission opportunity, as many Segments as possible are

transmitted starting from the oldest pending Segment based on the time available in the

TDMA allocation. We are interested in determining the duration of Fixed Allocation and









the Average Allocation for this retransmission strategy. Unfortunately, due to the

intricate interaction between the allocation size, FEC errors, and transmission overheads

this retransmission strategy is not amicable to direct analysis. In theory, it is possible to

conduct a Markov chain analysis by defining the Markov chain state as a L-tuple of

backlogs, (bo,bi,...,bL-1), where bi indicates the backlog of Segments that expire at the

end of ith beacon period following the current beacon period. Thus, bo will indicate the

backlog of Segments that expire at the end of the current beacon period and so on. The

number of state for such a Markov chain will be (N+1)L. Due to the computational

requirement for obtaining steady state probabilities of the Markov chain, this method will

only be useful for small values of N and L. To overcome these problems, we use a

combination of simplified analysis and simulations to provide approximate estimates of

the Fixed Allocation and Average Allocation.

7.3.1 Fixed Allocation

Consider a multimedia stream with a data rate of N Segments per beacon period

being transmitted over a channel with FEC Block Error probability of p. If each of the

Segments is provided with T, transmit opportunities, the probability that a given

Segment gets successfully delivered is (I p-/oT"). Since FEC Block errors are

uncorrelated, the probability that all N Segments will get successfully delivered is

(1 p'uTti) Hence the packet loss event probability, PLEP, is given by

PLEP=I (1- p~

The above equation can be used to determine the number of transmission attempts

required to achieve a given PLEP as shown below. Note that F[* represents the ceil


operation.







82



Tlogio 1-(1-PLE)
total 1 ( PLEP)N
Tlogl (P)


Figure 7-6 shows the number of transmission opportunities required to achieve a

PLEP of 10-5 for N = [50, 100, 200] and various values of p. These results shows that the

number of T,,, increases with N and p. These results also show that T,,, is not highly

sensitive to N.


-- N=50-
- N=100 i i i i

.... .. .... --- -- - ......--------- ----- ----------- -- ---



---------~~~~~~~~~~~~ ~ 7-- --.-- .-- ---- --.-- -- -- --------i^





--- .-> ^ ^ ---- --- --------
.......-a... "-I *--'--'-i-'-i-'-....---,---------".....



10-2 101
FEC Block Error Probability


Figure 7-6: Transmission opportunities required to achieve PLEP = 10-5 at various FEC
block error probabilities

Since a minimum of T transmit opportunities needs to be provided in L TDMA

allocations, the number of transmit opportunities per TDMA allocation is,





Based on this, an approximate estimate of Fixed Allocation (FATheory) can be

obtained as,










FA _Theory = rH+ N p


The first term in this equation reflects the contribution of transmission overheads to

the Fixed Allocation. The second term shows the contribution due to multiple

retransmissions of Segments. We basically use the minimum Segment allocation required

for stability as a first order approximation. In reality, due to the statistical nature of the

FEC block errors there will be a complex interaction between the FEC Block errors and

number of retransmissions. For example, in some TDMA allocations the FEC block

errors can be small resulting in lager number of retransmission opportunities and vice

versa. Simulations were used to validate how well the theoretical estimate matches the

actual allocation requirement.

Simulations were performed forN = [50, 100, 200], L = [1, 2...8] and p = [0.001,

0.005, 0.01, 0.05, 0.10, 0.15, 0.20]. Each simulation was run for 107 beacon periods with

varying allocations to determine the Fixed Allocation (FASim) that would guarantee

PLEP less than 10-5. Figure 7-7 shows the Fixed Allocation estimation error in for

N=100. Estimation error is measured relative to the theoretically computed Fixed

Allocation. These results show that the estimation error is large for small values of

latencies and large values of FEC block error probabilities. This can be attributed to the

use of average value of allocation for Segments in the theoretical estimate. Figure 7-8

shows the probability distribution of the estimation error for all the simulation runs.

These results show that the theoretical estimates remain within 5% of the actual Fixed

Allocation in 90% of the data points. In general, for L > 3, our theoretical estimates

accurate showed an error of less than 5% in 99% of the data points and an error of less









84




than 2% in 97% of the data points. Our simulations also showed that the estimation errors



remain similar for PLEP of 10-3, and 104.



Figure 7-9 shows that average packets lost per packet loss event as a function of



FEC block error probability for all the simulations. Results show that the average number



of packets lost per packet loss event tends to remain very close to 1 in all the simulations



except for H = 1, L = 1. In the later case, the average packet loss showed slight increase



with N or p.


N=100 L=1


-C- H=1
-B- H=5 ... ----- --------- ----- --4. --. -- --- --------

----:----



.. . . ., - I I . -- I- -, -', .
.. .. -- '--'- .-- -.-.-.-.- -- ----- -- '-- '- -'- '- -. -


........".... "-- -4-'' .---------4- --------".







2 102 101
FEC Block Error Probab lity


N=100 L=3


H=1
H 5 -------------- --------

.H=15 ---,.--,.- -, -, --------- !. --- --- -- --,- -- --- .--------








L -- -L -L L L --- --


102


FEC Block Error Probability


N=100 L=2

-4- H-1
0 H=5 -------------- --------





1 ----








2I
10 ........ ---- ---'--'-'-'-'- .........-- --- '- -- '- '- '- '- I- '- ".- ..... .

15 ......... .. ---'--'-'-'-'- ........---- '--- '--- '-- '-'- ',-'- ....... .




10 102 10'
FEC Block Error Probability


N =100 L=4



15 -e H=15 .--- .-- -- --------- ------- ----- .-.-.-------

6--I
25





10 .........--- -- -----'--- --------- ---- --- --i- .- ...-------'



6 ..-- ------------ -- ------ -4---- --


15 .. --- -- ----' --' -- '-'-- ---------- ---- -- '-- '- '-. ------
10 --- -- --- -- -- --. .-, ---------.---- ----- .---- -- -------_
15 ......... ....."---'--'-'-'-'-" .------- ----'---'--'--' --' -------"

2 0 - - - - - - --: I


in'


102


in1


FEC Block Error Probability


Figure 7-7: Fixed Allocation estimation error for N = 100 and L = 1, 2, 3, 4, 6, and 8















N=100 L=6
25

S.-7-i-7-------- : : : : -----
20 .- H=5 ---'---'- ------- -- -- '- .
20 H=10
15 -e- H=156

-10 -. -- -- --- -----
to ---- - -. .- -.----------.----- --- 1- 1-- 4 -i- ---------

5 -------- ---- -,--, ,- ------- ----- --- -- -, --------

0

-----1- ----- ------- ----

10 ------ ------------ -------

-15 ---- --- '- --- -- -. --- --.- -- -- -------
.20 --- ----- .--- .--. -!- --- --------- ----- -.- -- --------
-20


103 102 101


FEC Block Error Probability



Figure 7-7: Continued


N=100 L=0
25 I
-15 H=1
20 S- H=5 ---------' --- ----- ---- -

15 H=15 --- -- -------- ...------ ----- --- --- -------- ........

to -------- ----- --'--- 2- -------- ----- --- -- --, -------.

- ------- -- : ---,-- -,- --------- ---,--,--,-,-- -,

5 --------- ----

6 ---- -------

-10 ------------- --- ------ --------- ---- --- ---- -- -------



-20 ----- ----------- ----------- i i i i
-20


0 102 102
FEC Block Error Probabllity


04 -







0.3



0.25













0.05 --------- ------------------------------ -------------------- ---------
0.6 I .- I- -



0 1 -- -- --
0.06-------------------------------------------- ---------






0
-5 0 5 10 15 20 25 30
Error in Fixed Allocation Estimation(%)




Figure 7-8: Probability distribution function of the Fixed Allocation estimation error






86



-- N=50,H=1
18 N=100,H=1 -- -' -------- -, --- -- -- -- -------
-B- N=200,H==1 '



12
1 Other N,H ------------ --- ---- --- ---


1 ------
1j 2 ---------------- ----- ------ ---------- ^ ^ -- = -.-- ^ ---





0.4


10 .i-- 10
TO R .- -- -- -- '- --'-- j J--------- L---- J----J--- -L L----------
a- -. --,---- -- - --- -

0J.6 --------- 4 ----- -- I -, --__ __- ___--- --- I-. ------- -

0. ----------*--*- -- ----------- T --- -----

10 .3 10 .2 10 "1
FEC Block Error Probability

Figure 7-9: Average packet loss per packet loss event when transmission overheads are
present

7.3.2 Average Allocation

To successfully transmit each Segment, an average of (1 p) 1 number of

transmission opportunities will be needed. Hence the contribution of Segment transmit

N
times to the Average Allocation is exactly ( The average time used by


transmission overheads will depend on the probability distribution of the transmission

attempts. To obtain this, we assume that the number of transmission attempts will be

upper bounded by T,,,,, i.e., the minimum number of transmission attempts per TDMA

allocation required to achieved the required PLEP. We further assume that all pending

Segments will be transmitted at each transmission opportunity (i.e., no truncation effects

due to fixed size allocation). Since the probability that a Segment gets delivered