UFDC Home myUFDC Home  |   Help
<%BANNER%>

# High-Performance Periodic Contention-Free Multiple-Access Protocol for Broadband Multimedia Powerline Communications

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 E20110113_AAAAAT INGEST_TIME 2011-01-13T07:26:55Z PACKAGE UFE0004387_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES
FILE SIZE 6860 DFID F20110113_AAAKOJ ORIGIN DEPOSITOR PATH lin_y_Page_098thm.jpg GLOBAL false PRESERVATION BIT MESSAGE_DIGEST ALGORITHM MD5
4aa02077dee6f4a50eef4997be8d05b9
SHA-1
6a9334421b3cd24cf36358b0f671b99ff0c2ae60
5898 F20110113_AAAKNU lin_y_Page_090thm.jpg
62ac3c48be3770dc1dccdf88dbd3c32c
8cfcca6804789f28864af365fa55c5edcfb01f2a
25271604 F20110113_AAAJLH lin_y_Page_100.tif
f5c2c4c81e958106c2cf2ee78c8d177e
1281a67d92be37b220ae878dec0e1794859be7b8
22995 F20110113_AAAJKS lin_y_Page_087.QC.jpg
e3322e9dc2c6da7ecbec7789819b5980
0212a46fe1f7f507f6b44308d961db5c1d834869
27312 F20110113_AAAKOK lin_y_Page_098.QC.jpg
96030bfcdb5d2e0d1cd6eb1f927aeae8
38f4ffc73966f01b93227acd1492d78711396fb2
22002 F20110113_AAAKNV lin_y_Page_090.QC.jpg
9271bf7808420bdd08093bfde1c1cde2
95de35231b71c85378b1c9cc2d59351a48bc8b9d
33851 F20110113_AAAJLI lin_y_Page_016.QC.jpg
46408 F20110113_AAAJKT lin_y_Page_028.pro
35269650563bd656f9383cfc5ff6a6ac
57f44321316e961218bdcbede354f787900358b2
8187 F20110113_AAAKOL lin_y_Page_099thm.jpg
b0bd3de90b2d2fa05ce30db22f43b0fc
10e331067a26ac2179f8fd954fb6e9e683974e79
6069 F20110113_AAAKNW lin_y_Page_091thm.jpg
e8d9864687d5b719396d589c7e843b36
175f78c9a4f2cb89f30b073a0a7dfc99ca879f07
34021 F20110113_AAAJLJ lin_y_Page_112.QC.jpg
801d5cb65f8d2d701b0ce6fb61ed9a5f
3e159ce81128575cbc93d06df98455a96003f0c8
30808 F20110113_AAAJKU lin_y_Page_055.QC.jpg
79957fb54e43123e860ef3ca813ddd07b438a60c
7978 F20110113_AAAKPA lin_y_Page_109thm.jpg
acc043ef894e032816a96776be26e15c
7033e411438f16efafecb28e9614a1cd28cc7bbc
23078 F20110113_AAAKNX lin_y_Page_091.QC.jpg
03028b34818ee71b1873c9aa6f2b42bf99920f8e
80185 F20110113_AAAJKV lin_y_Page_050.jpg
8a5ce509bb37852109d4311cd22c715a
ea4ccff068f48666bd83697b249969068d636321
30697 F20110113_AAAKPB lin_y_Page_109.QC.jpg
9332bb7fe49be6c6251d09dbb093b99d
bbd66eae6ae26e17582d25a439d8ba389974d58c
9245 F20110113_AAAKOM lin_y_Page_100thm.jpg
05e97f276fb9f67f6857dfb56434e095
b06bf8c2cd2dd24848d984bcc1a68ede887c039b
8063 F20110113_AAAKNY lin_y_Page_092thm.jpg
cb59e577b1106e46cf2b003c3dd3f089288a07e0
1433 F20110113_AAAJLK lin_y_Page_093.txt
9277a1fc99d0134a69eae6e1c9a9a144
1c72c777ec6d6bd551b84a7a3b73edbdf1f279b8
51314 F20110113_AAAJKW lin_y_Page_026.pro
29a11927c6ee60ceb15eb6f832b1b1dc
9b1097a0186cb47b0f5a6f8f09d313be1605fa98
8635 F20110113_AAAKPC lin_y_Page_110thm.jpg
8e0eff30d0c08f50d7e13ab13531f3d2
7da6dc7d1595d2fff780a0303931e251ac8eb77a
36065 F20110113_AAAKON lin_y_Page_100.QC.jpg
7a5dbeff4292817f552dee11aa104513
c53d6b7df5af67500014d966b55ba2b2de1348d7
29475 F20110113_AAAKNZ lin_y_Page_092.QC.jpg
fc7b32341e6ed18a2d50932179515157
2b0f59fe70fdf94b1092628ec4a650159751fda5
21474 F20110113_AAAJLL lin_y_Page_023.pro
470213d90ec4d64b7a2158c57254621a09eb8acf
548631 F20110113_AAAJKX lin_y_Page_038.jp2
16c374f6bea64d23da59a277b97ac0137d9b10c9
2087 F20110113_AAAJMA lin_y_Page_056.txt
b2f93d7dde141cdcc23d6b4abcb66113
92fee4e269fc304a3aa033ecd6937a1fa480cc95
32830 F20110113_AAAKPD lin_y_Page_110.QC.jpg
19f48b5939ffc8b431548d681715a202
7fae656e0979f9661659b2587887fd1fdb44b1bd
8520 F20110113_AAAKOO lin_y_Page_101thm.jpg
3aa8ea68501660430d27dee806ccc22b
ff9637cde4dca08906aba75de8155a4fb8bef4ee
114333 F20110113_AAAJLM lin_y_Page_124.jpg
667dda2361102857d0964da1f9da0db7
ac5e6c845a078058a014de8659405d96b55ae724
31558 F20110113_AAAJKY lin_y_Page_022.QC.jpg
864ea3732094121d184c4095946f8248
a09e7a1161f3e596e651122383a548c52f595ac0
1269 F20110113_AAAJMB lin_y_Page_091.txt
be877f839e5c02b324a4664d7d1cf141
b721ecb08700e9c3cab519deb10c19ede32517fc
8287 F20110113_AAAKPE lin_y_Page_111thm.jpg
ec7ae95c79acfb2b7c0303e979497e22
33222 F20110113_AAAKOP lin_y_Page_101.QC.jpg
f255d36f493eabc6f6586755be65bc36
5fbfa1c0e3974280c221034b89e60177d5fae78e
32858 F20110113_AAAJLN lin_y_Page_028.QC.jpg
fe6084b5f5cb293ced6780d6ca3d5932
1053954 F20110113_AAAJKZ lin_y_Page_064.tif
32bc97f37e4e514ecd9fc084ee278d61
9eb24a9fdecbeec1b2ba67dc686c59fbddd224a8
F20110113_AAAJMC lin_y_Page_097.tif
dd22868f61524b5520f433df507cee52
b6ff479b5da9ca2d5733621a6fd3df3a136e585a
32500 F20110113_AAAKPF lin_y_Page_111.QC.jpg
cb38b651916c5db8fc139f12d25a1e85
fe218ea93deef07f4eb3569c357a4753eecb0eb9
31712 F20110113_AAAKOQ lin_y_Page_102.QC.jpg
F20110113_AAAJLO lin_y_Page_090.tif
b146db8b0166fbd9e2b37c48b245f36d2ffcdd74
F20110113_AAAJMD lin_y_Page_009.tif
1edc797354e8bdcce8ce9b1948ea0471
9b5b48f9063254a95098c58b2081455d3c89c447
8404 F20110113_AAAKPG lin_y_Page_112thm.jpg
90f4eae13867481c83e5371363ffbc51
19c0b21d78f6aab7869568ebd322749463ea32ac
8399 F20110113_AAAKOR lin_y_Page_103thm.jpg
859d1b8ee111c8d50ab1e399f78a1bae
34950 F20110113_AAAJLP lin_y_Page_071.QC.jpg
27e18e0c3c46b2d6cebb81037091a272
c4c05b789eb803f56748159576c749e3904d94d8
1833 F20110113_AAAJME lin_y_Page_049.txt
ed5b2320a0f29515deb1c8d00a14107d
8214 F20110113_AAAKPH lin_y_Page_113thm.jpg
337c14155b6ae7153ed63ac0d4a57d51
7416ba730b2354f2bf3c0f1004922a64a7174c53
30144 F20110113_AAAKOS lin_y_Page_103.QC.jpg
1027a3fba7b960b74ffd3680fe74b62e
afbcbcaf473f07344f1a7c10843d9e1baa1d6e28
1811 F20110113_AAAJLQ lin_y_Page_010.txt
18950893343afc81d66e4209dbf3c913
8c06082b142c646901f572cd7bcec065dc941e82
48679 F20110113_AAAJMF lin_y_Page_024.pro
90af84a2a42090d1bc29708e114881e2
1313efb9a0286f249eb47cabbd08f43a7181cffd
31230 F20110113_AAAKPI lin_y_Page_113.QC.jpg
c6b1caafd586fc313221261ec7f72636
23e94513c0f41df12d0a7ce5600e93edb8b40c47
7454 F20110113_AAAKOT lin_y_Page_104thm.jpg
84f087c1bf6586d5f4bd580702c48faf
99464 F20110113_AAAJLR lin_y_Page_020.jpg
2b29f150016eb1a4c7a2803a92f73851
57595a0a5d3813525ab64122b01daae258f05d05
5885 F20110113_AAAJMG lin_y_Page_003.jp2
e7547a1ae6036cf4e3e7dcf752ef2b58
64953f5ca152c5af947aabc808f379d504937baa
8738 F20110113_AAAKPJ lin_y_Page_114thm.jpg
383064cfeaf6bab7baa5575447d66195
ca1ce0c2c6e1ea5a91c9b16e0276499d9a44f9b2
26775 F20110113_AAAKOU lin_y_Page_104.QC.jpg
c42238eeb79bf2017eb1252f9d2baf73
fa4fb7f05d10cd9bf244744fb09d64d1b3932f36
F20110113_AAAJLS lin_y_Page_061.tif
79c5715f02ffca92a36c1de30814cf78
de166a1ca7e6541ac81eb883385b6e7b24602561
1908 F20110113_AAAJMH lin_y_Page_016.txt
4bd4805ddab960c2e7b60cdb73483cb6
6689551481f4be955cd1d53dd52dc25ab153bb61
33113 F20110113_AAAKPK lin_y_Page_114.QC.jpg
a5075447e4fb46aaa58e94d8edd07096
eabd35a741469d2d824e2cdd40a4657e4a808298
9806 F20110113_AAAKOV lin_y_Page_105thm.jpg
e92bf8d21d29e0c78ec1903b7b011389
5ce55e4f0e2aa526d370111efd03352c0800587d
29002 F20110113_AAAJLT lin_y_Page_027.pro
eb4eb149ffb8c0b90c41f4be8d2d8a8a
F20110113_AAAJMI lin_y_Page_012.tif
c5d2bd5b1547a9d47acf8c3d33db25c3
8500 F20110113_AAAKPL lin_y_Page_115thm.jpg
de3260c55154d748f7a68a0a6f5d54b4
9c6ddc311c6b413c073b6d9d248b97c8828edc34
38193 F20110113_AAAKOW lin_y_Page_105.QC.jpg
a0c99c60856380d1809318d588f09096
e905ba230ffb88a6851f159f8603ddf448dbc30c
6927 F20110113_AAAJLU lin_y_Page_011thm.jpg
ba374f4a3a1caca525439652ed95d7f8e4d5a6fd
7894 F20110113_AAAJMJ lin_y_Page_035thm.jpg
0720115660761175450a92be3536a50f
6464eb02f10b42cc06abce35c439f0638449d3e8
3077 F20110113_AAAKQA lin_y_Page_125thm.jpg
11b719071f7204039c6058901cc97b41a4946451
8808 F20110113_AAAKPM lin_y_Page_116thm.jpg
9b7a0bc8eb12b623ee0321075c960812
8da73e3dcf35775169091f059f557cf3b8f53298
3505 F20110113_AAAKOX lin_y_Page_107thm.jpg
04e0c90f99db4bce171e786bfb8d64be
c253817300f993135e08d2c73fa950b287344fc1
31540 F20110113_AAAJLV lin_y_Page_035.QC.jpg
db3178976cddc997e3e5df256bb4c09d
98d86a8f58d4377ef72a04d014e8bc8c0108b8cb
7648 F20110113_AAAJMK lin_y_Page_077thm.jpg
c379c336e148c17f4ca712938586df17
0aaf18cc6d4ba34d2b81f4f1a32d42ce51856393
142855 F20110113_AAAKQB UFE0004387_00001.mets FULL
4b07edeabec7bf962f32d7ee092d0d2c
a88696dc11759cb84fcf1250ef2de9ab595ea83f
13302 F20110113_AAAKOY lin_y_Page_107.QC.jpg
79508 F20110113_AAAJLW lin_y_Page_047.jpg
049a85ed202d9a80a5613627a1a3f833faab2b63
34213 F20110113_AAAKPN lin_y_Page_116.QC.jpg
4b4a6ba7e542348c177a94c7b3f0b691
86662a46cbdfa1cfeafb8f65d0f80c342a3d2f0d
7017 F20110113_AAAKOZ lin_y_Page_108thm.jpg
e4319bcc1bb5f8989e512161305643ee
5b1df9d0cc45930555deb1a90354a2ac652b794b
53857 F20110113_AAAJLX lin_y_Page_110.pro
5e754853a7537b0e0bb8859d47a21f5a
49dc5870dda1db0ca233225693dd2ea7a82e578f
F20110113_AAAJNA lin_y_Page_043.tif
76c28e731bf510a8d70c47c8c0a4dcaa
9b367fafb082def39c37dda8216a9dfc6616cb28
51719 F20110113_AAAJML lin_y_Page_056.pro
88a5a4b44ed266395ac0a14e84443cf9
82dae5e4bd0b940b541b9d4506b47e577243db97
F20110113_AAAKPO lin_y_Page_117thm.jpg
db211e8fca2455df13ff29422140a87f
32026 F20110113_AAAJLY lin_y_Page_045.QC.jpg
8fa9b74b862d209b7976450818afa54e
650548a709cde158055e8b6ef5ee386b031065ec
50503 F20110113_AAAJNB lin_y_Page_034.pro
1cf24aed87d325f2672067ab768a4b01
c8fdb153403665d69e994421a7ffd165dd85e15a
1703 F20110113_AAAJMM lin_y_Page_108.txt
004b918ef7969fc15960c5d4d318b203
31cb961b38504c12a4d7ffca73320c376e40b56b
8487 F20110113_AAAKPP lin_y_Page_118thm.jpg
c788fbe6b9d2b05c1868101cd9ce16eb
c8f8c9a5da272afe5a176e817a840a894a315660
2049 F20110113_AAAJLZ lin_y_Page_009.txt
b96ea406e4d3a838e2eda96a18235226
F20110113_AAAJNC lin_y_Page_014.tif
74a669e280a56bf22ed1c16022bf87ac
1e4e192484bde135345ba1485f983b5f17e06aa5
94178 F20110113_AAAJMN lin_y_Page_018.jpg
005f851df4104028d39e5b93c58a3f2d
3420c27b187a3228a10dea9c151b3e8cfa1f3623
32169 F20110113_AAAKPQ lin_y_Page_118.QC.jpg
b731921bc7ba54796b7e17d0b72b35ea
2d0976bcd45311d56244f5f38934f4c84460bc55
1506 F20110113_AAAJND lin_y_Page_054.txt
01c01ef068fa0e86f824754c28dfe932
75d68c8a715d784fdc6cfc93e601c863884278bd
102516 F20110113_AAAJMO lin_y_Page_040.jpg
f2ec27ebc99bb1ce85624b4477257f65
c71d5240a6bbfda874c726311fbeb619ee1426bb
8148 F20110113_AAAKPR lin_y_Page_119thm.jpg
fe3b2d74bd4cbfafe6a6f62249d38035
15d6341dd0799de5d698b131fd2ac50068a3017c
101757 F20110113_AAAJNE lin_y_Page_054.jpg
908ae180fd999fe9b12b92b9eefbdc67
9256d08ab4acf30cb103d6f298584d95eeb1a0de
28617 F20110113_AAAJMP lin_y_Page_096.QC.jpg
39c15b63025be3a327735180546f3edd
5f34100ef067a0f5dc6e4d0ec4eebe99434a9c7f
31565 F20110113_AAAKPS lin_y_Page_119.QC.jpg
87b6aa8742a842ce9129c500ed4b0cf3
66a6f266526351b956578bd7986b99f19a5a458c
F20110113_AAAJNF lin_y_Page_118.tif
cd2ffee05640b62d848704f7aa7e8c55
40c46cab3d11e9bf71563d032838385002c43ac2
17207 F20110113_AAAJMQ lin_y_Page_121.pro
8faab6225ba89b380dbff05a1979ac12
4f4e9f61a87471e4cdd043ee7e69dffef24b3f77
32647 F20110113_AAAKPT lin_y_Page_120.QC.jpg
2bb94531b805234be68fc8f1f840dcb8
dfbb71b15804a20172ff727ab85c464b84d15040
34085 F20110113_AAAJNG lin_y_Page_099.QC.jpg
5e6abc4f2f7ce8457e578683a12b3a1bd7d3fa0e
8252 F20110113_AAAJMR lin_y_Page_031thm.jpg
5c7a7a61e0fe8c4c6da2ce950f0f49ee9625c846
3495 F20110113_AAAKPU lin_y_Page_121thm.jpg
3e5c9a69a6a22000de0bc9c3433e99cb
8386db47727a25f078eeb28c95c6d04e6df999d4
F20110113_AAAJNH lin_y_Page_046.tif
c965f25909ce728585fb5eae1f50cd62
ef8673236280f9a7e669f0dd6fa78bc273f68849
F20110113_AAAJMS lin_y_Page_095.tif
220d2aa5d6811433903d7e1c23c79b18
ff5d301de4489ce084db5265fd6d6c64628767b9
7533 F20110113_AAAKPV lin_y_Page_122thm.jpg
ba63f29fab1a001b75bb06b43d4bf778
5d61e58069af7ed3f0d6959e62d7766b1e6681b7
1004 F20110113_AAAJNI lin_y_Page_023.txt
e7c547166a4760d23639981313e895cb
a7a6c354cb29347aa9cf2dc1d3a12ecf480b77fd
13278 F20110113_AAAJMT lin_y_Page_121.QC.jpg
b1eac20874160a1cefab0da25c99c574
a6546d2f5412cf3b406e83ef357687f5564ac4c4
30136 F20110113_AAAKPW lin_y_Page_122.QC.jpg
03af65212c0acf2596facdea6a913c57
eb80bed4e63225b1df6d445f53a20f90e1cdcfdd
7850 F20110113_AAAJNJ lin_y_Page_025thm.jpg
029377d0227a65ca2d33e6125aaebab3
10065012f343638e40369a1d7961df36dd5d9b69
6765 F20110113_AAAJMU lin_y_Page_073thm.jpg
13f9de2a620b159636a4c8140228a7dc88ed4da2
9147 F20110113_AAAKPX lin_y_Page_123thm.jpg
7f020aae625d82b0963709d74cacd6b7
032326030a8fc5031a12d3caecab6a2a1e643b8e
102930 F20110113_AAAJNK lin_y_Page_120.jp2
9a3ee45608af8495c7e3d3259af17a58
33556 F20110113_AAAJMV lin_y_Page_104.pro
048d233b071d2a415562c613abce67f9
99b9b1cff4974e6958b16b24ebd56a022c7748e6
39189 F20110113_AAAKPY lin_y_Page_123.QC.jpg
1c65feb936c684bc68a8c82478576b31dcd0ccea
33232 F20110113_AAAJNL lin_y_Page_024.QC.jpg
8cfcc32eabdb2fea54e41fd704d1b0b3
89749 F20110113_AAAJMW lin_y_Page_092.jpg
247ef741a35f07a6ec723f759b41ee14
12786e5ab1ab6bfca64fc718c6dfe9d41e3d95d4
7915 F20110113_AAAKPZ lin_y_Page_124thm.jpg
c22812778e4f1ceffe24e419af291b5075476432
8613 F20110113_AAAJOA lin_y_Page_058thm.jpg
296ed764d3c70e7b97e028f796c6e56e
80cabe28244ec3d42f8a49985a216b7e7d4100aa
48091 F20110113_AAAJMX lin_y_Page_082.jp2
be7f1bf8ed4fcea86dec5fda332308a4
2e1c54be6cf2cae5410afd754172a12c6834f3b7
8572 F20110113_AAAJOB lin_y_Page_030thm.jpg
2ca511da59ba4ffc3ef6f7b8d54af57d
6128ac434498579745b79bac50c68ff584bd75f5
80195 F20110113_AAAJNM lin_y_Page_080.jp2
db748cc3b258499f36561ab114976e54
2206e160336b16980c6f3fda267e454d40e3eed1
F20110113_AAAJMY lin_y_Page_032.txt
f74388dd1599e1d9390884eebf46ed12
37cccc27d70c28c627e3112ff86753d321696a18
7012 F20110113_AAAJOC lin_y_Page_047thm.jpg
f5782db1030d6da6ccef904f035d2c8a
56945c5f9c6e90c50980b928a195566803884bab
105314 F20110113_AAAJNN lin_y_Page_068.jpg
bc897aa213fc7eff74d2e0f55a243101
b9e6dd670b810cbb32b213c2c9c69014932a7c55
1051960 F20110113_AAAJMZ lin_y_Page_010.jp2
97d2252ec131e238108461dbcd65187b
77922a2f0d652029edcf674ac1903ed9bb547b17
47105 F20110113_AAAJOD lin_y_Page_031.pro
7230a2eeb55a56f324150d5b840a97d1
9c8602bc4efc873ef6560df44aabf3cbae209c8a
1416 F20110113_AAAJNO lin_y_Page_002.QC.jpg
02bb82ca25ab7ca53dabb76cf6aac51b
1a005634091fbe1b0b1ca0d0e78a60e20edcd31e
8077 F20110113_AAAJOE lin_y_Page_120thm.jpg
fd7d502deba349da9172c50df7bd3085
3503ba0f1fb643740620a1139f5f719cf400abff
92602 F20110113_AAAJNP lin_y_Page_083.jpg
c30fbd03a2018457307e609d9dc2d2ac
97031 F20110113_AAAJOF lin_y_Page_035.jpg
91f0d5c5957ac4b4ab079f77393bc8af
a74828bf1306ddabed71b2b67570ec6d164ee4d3
52772 F20110113_AAAJNQ lin_y_Page_004.jp2
6fbc52a9fdd866fbed0fbe0847e4548d
41d2a84be9fa9f5f44e0ace7cfa1027735033d1e
84404 F20110113_AAAJOG lin_y_Page_086.jpg
4d74b3050345af9bd703affbba932ac81713c5db
1828 F20110113_AAAJNR lin_y_Page_064.txt
3b22fbce9ac0571c1f07c68fbb42cf164fbf9846
F20110113_AAAJOH lin_y_Page_079.tif
29db9271a930c0c5f187b17874112b98
394852d5cd1571eb61b514af78dd5de6e175fd4b
5128 F20110113_AAAJNS lin_y_Page_003.jpg
a4c10005d5fa11df5b48a5b91125ffb7
89406e0189e582fe89a88581fc634b085a0ed27b
8205 F20110113_AAAJOI lin_y_Page_076thm.jpg
22fcb22cf59a653822fbd7d25d9a560f
32122 F20110113_AAAJNT lin_y_Page_037.QC.jpg
a13b7c53284ac035c50a93b0d3cc1003
1d05186715ef346eb40368c919d7af2571cd012d
109711 F20110113_AAAJOJ lin_y_Page_032.jpg
7e92f163b177853123d638f9e8dc289c
6390 F20110113_AAAJNU lin_y_Page_087thm.jpg
ebfd7e835e0c58e4f5d3299ec55da976
46243 F20110113_AAAJOK lin_y_Page_064.pro
a60710984c0cee45a726f307ea30e234
7865ca4fc5588373d7fbea302d7119c08f08ac6f
7340 F20110113_AAAJNV lin_y_Page_074thm.jpg
15f2f72c367341cac01c5d20c1d80927
aa6da1a8b06cfd7865596882a9bb2759d53ccc90
1919 F20110113_AAAJOL lin_y_Page_111.txt
3861cfb34cea8450ec4713305c188cdf
e0c30a9facdd9207260d9ccd1b30879ed693f0bf
7126 F20110113_AAAJNW lin_y_Page_053thm.jpg
53309ac3d3424a99b6d4612b1b5babb9
80d9ccf361fda8c990f6bc9a39b054ef2ef711d1
30987 F20110113_AAAJOM lin_y_Page_041.QC.jpg
fcc7dd36cce2a6c54d260f983760da2f
b822243abe5b99cc244f9072535b9b8e72b3b48a
7028 F20110113_AAAJNX lin_y_Page_006thm.jpg
98a7efa8795dba0a05da7f7600530901
45a79234912334bc8ed097b80ff99ce4fe53b6e9
8559 F20110113_AAAJPA lin_y_Page_032thm.jpg
7f14c0ccf3496d590596cabb52723d95
5d4b91f0a58ec889293b0064f2e9ee4f48409bb9
1580 F20110113_AAAJNY lin_y_Page_003.QC.jpg
e74befd3cd5d8245f17d890890807b74
8856d032aa67a891a38d9ffeb95ecb11f860d4f7
1051971 F20110113_AAAJPB lin_y_Page_054.jp2
9e2786e3ab9b8b9ffe6bd27b72cd7a82
F20110113_AAAJON lin_y_Page_109.tif
7751cae2e1e516c12c7116f9852bc631
446f074dc9450cf75c4124a7b168d73ff175f38c
111284 F20110113_AAAJNZ lin_y_Page_034.jp2
a3d659763537c315e9d2a871f147ae30
f5057e3539f759a860485e9b7f461756d798e661
30818 F20110113_AAAJPC lin_y_Page_048.QC.jpg
99e8a7b5e91264c2fc8937c46d8e0c75
27dafa3ed71e3b5ff6a6c59c78effce12831e649
102536 F20110113_AAAJOO lin_y_Page_037.jp2
b781c2a846d207c1f64f6df0cca6f9a37ce5e223
29165 F20110113_AAAJPD lin_y_Page_044.QC.jpg
e9f9460a6e6b5b6cf67a72ee886548f8
c32032b0c1ef4a78a4f76602a9401ebac9ce1960
25824 F20110113_AAAJPE lin_y_Page_085.QC.jpg
b2d8606a54de2018cac0aa877a590526
ac747398e7dc5aff9e92d58b43650765acc46241
1765 F20110113_AAAJOP lin_y_Page_039.txt
3480153a9f8e4bce9b5056263ab8f24d
35936 F20110113_AAAJPF lin_y_Page_117.QC.jpg
95ec6551079c7f90c00f1744d3b68ca8
0155be22d06f83d7e96c1ebfb15a1d6a9d197ddd
58278 F20110113_AAAJOQ lin_y_Page_005.pro
0d723391786a62520f9f903f74ef7835
686ac1167b732489f1c94e461474505b19f8a6f2
36748 F20110113_AAAJPG lin_y_Page_047.pro
1184ef8b8e69e1314fd55cb7607e5c86
8423998 F20110113_AAAJOR lin_y_Page_044.tif
da37293a55e71f6ae11ba89fdff6057f
0f47975b112e1c9874f376086c0a81dee473be5a
29543 F20110113_AAAJPH lin_y_Page_057.pro
7cc4e5761855f4635a66c8d942126745
82605 F20110113_AAAJOS lin_y_Page_046.jpg
487b52b5a03531ab150294ae302a0433
606b639d1c81b105063237edd1284be7fd4027a5
106264 F20110113_AAAJPI lin_y_Page_065.jpg
db467eb38b0e20391e2a529082502305
d1c6e2d6078b1e3287579eaf718499c470fafb6f
1953 F20110113_AAAJOT lin_y_Page_021.txt
10df282147536ff3147ce5a43efea395
702a307285eae81eae75e488d927b1714070e0b9
8013 F20110113_AAAJPJ lin_y_Page_083thm.jpg
531663c4ef5c1835e0a4161c95bba944d9b073d0
103030 F20110113_AAAJOU lin_y_Page_021.jpg
c435272b878276e6daef3be19a36542c
c72be3b486637e1a079153f8af75be696ff36880
F20110113_AAAJPK lin_y_Page_041.tif
d9797862112aef75d295d72170f32bc8081220d8
904831 F20110113_AAAJOV lin_y_Page_053.jp2
5551492a293ca8a79f435737a62438e8
858d093fe9f57b4f4dc37c980eb0f2979bff0466
F20110113_AAAJPL lin_y_Page_028.tif
53a90e7eb1d23a6c49114156637483f4
6a12b2b185ca42048cc6d9092ca24392942bbb54
1889 F20110113_AAAJOW lin_y_Page_118.txt
38bef0fa058bc55de6f1d7d2a9cca8c2
639072212f3ec2a6f48189164a166aa61217025c
1882 F20110113_AAAJQA lin_y_Page_083.txt
89ae3a88a432f94cf5da6b05451ae31b11e0270d
94375 F20110113_AAAJPM lin_y_Page_119.jpg
39e7de206c3cb914ec8e4e41e5897fac
F20110113_AAAJOX lin_y_Page_011.tif
894662a6f4360665ce2d3430e3d0e852
045cfda0b57b2b7ccaa8fa9bd7a298dca52975e7
F20110113_AAAJQB lin_y_Page_120.tif
303cc9f5bf166b4824a58b72b78a0e69
4f79bcd27db57f8e144a5e94529002e44bcb4f06
6749 F20110113_AAAJPN lin_y_Page_051thm.jpg
f514022cdd85b5da8e4578dee2d875a9430221a1
48250 F20110113_AAAJOY lin_y_Page_036.pro
9849bfc2aba8ff12df030d96351b6a3c
de904ccb3e9838bf5054b21c9cfc64dda045c7da
7875 F20110113_AAAJQC lin_y_Page_018thm.jpg
6be7984f5e1f12a29268b0b3ffe841d7
11524 F20110113_AAAJOZ lin_y_Page_125.QC.jpg
9d549bf8bb8ca51bbb9dde4363f185c0
bca3aa070687f1f74d521d4a373d45e527c90bf3
2009 F20110113_AAAJQD lin_y_Page_074.txt
9e3e6cd361a1a8fbf1624e9a0e804f80
F20110113_AAAJPO lin_y_Page_066.tif
4e564ef1ef4c7826e9412af69da0aa15
106991 F20110113_AAAJQE lin_y_Page_069.jp2
679cdaba0814d9c096b80d5c14d83b78
d5ccf3ba2ae21ea004e06351407269a50d0d68d3
80759 F20110113_AAAJPP lin_y_Page_085.jpg
0a20e8a46efd69465867f90772bc1c34
ba55b1fb65a195c43ebd0a50228ee5cd61acf8a3
37820 F20110113_AAAJQF lin_y_Page_096.pro
37aa79837d421903c52e2e1388a5ac94
e75aa364414dfec712a74263a5a5a59234df6958
9523 F20110113_AAAJPQ lin_y_Page_007.pro
d38c54c20a4f924fc0b55b2099306779
7b93e788689475125b9090ffeb5c930db8305af0
F20110113_AAAJQG lin_y_Page_101.tif
2337aeac1bfd786e0ed2ac233141e10f
b6e40dde95d5af076975af6c28a4d99f693c617d
84576 F20110113_AAAJPR lin_y_Page_072.jp2
9b3b0dbc8d889e5ebea85f7e2e30ae55
9e47f6681cf9751402d0e2efeef957340a02a71a
184924 F20110113_AAAJQH UFE0004387_00001.xml
318fca1bb010b8819359aef421a5b5d6
8952 F20110113_AAAJPS lin_y_Page_061thm.jpg
bc73cfd50f770585e9fbed3726c0d890
43f471aa47bc84bc70a925ac77f9d98130de0787
1838 F20110113_AAAJPT lin_y_Page_020.txt
3cc59defc8e5c3fc66dccd37a2183559
f00d264188091c1c9362a1ee9d2c0c884f1dbc6a
15518 F20110113_AAAJPU lin_y_Page_125.pro
f907e0a110267cea0dfcf01d96de0d3f
6cc26be3e9e0678b878f1a9e19e6d26226434c5d
29802 F20110113_AAAJQK lin_y_Page_001.jpg
93fa362baf2a81ebf75248923147bb5d
39928 F20110113_AAAJPV lin_y_Page_044.pro
3cccfb66533813101840ccd25a33062c
450e6fa2c3dff5a64d51ef29856c2901d5776699
50626 F20110113_AAAJQL lin_y_Page_004.jpg
538500a7e35958ced82c0a89e35bf68c
cccac524c17d0542070f789234f4604f276f9ea9
27459 F20110113_AAAJPW lin_y_Page_108.QC.jpg
fb348e7d51ff67f38d56bda972d246c93361e600
100692 F20110113_AAAJQM lin_y_Page_005.jpg
cb6f7b887c4e97d217f2aeac49bfebc3
88100fc15919947481333eb1f36629475fc1e673
925449 F20110113_AAAJPX lin_y_Page_055.jp2
46ef62a9991816b0899f9a4bb10b08a7
6e9c91d5196ea7ecd4581ec75b10b3969ab1d862
90860 F20110113_AAAJRA lin_y_Page_025.jpg
6f031f5bf5c6b2a578ff59f6506da54e
28fab75f31414ef8530c45b08e268fd01d2426d4
141991 F20110113_AAAJQN lin_y_Page_006.jpg
7d24409aa38195d449616916a1caf883
2100 F20110113_AAAJPY lin_y_Page_124.txt
77d9c58758be5c077d4cf7ed975b0009
bd63e0cd1e76ce52738298edca9694e6f47c377e
107378 F20110113_AAAJRB lin_y_Page_026.jpg
0f2fbaf06763f5718df072fa837f8556
6d6dfb67948a9a1bb5c5724a3155f24191983e5e
19425 F20110113_AAAJQO lin_y_Page_007.jpg
e654b3dd2649c858f1be4410b5f6282f
1240 F20110113_AAAJPZ lin_y_Page_051.txt
1d4fa8a22c0f57e8b73971855108697a0589c1ac
99611 F20110113_AAAJRC lin_y_Page_028.jpg
96621 F20110113_AAAJRD lin_y_Page_029.jpg
11263f3cab56f60d30ab7667b919c86a
a68034e8e562172d438c9f241993a5b00a6cd112
53317 F20110113_AAAJQP lin_y_Page_008.jpg
a45aa32db6aa4d24b2bd0339d2686c992824e187
108650 F20110113_AAAJRE lin_y_Page_030.jpg
11e4d718d3f7899133e389b50b402f07
7a5bfbf555d0455d62464242329944b9db832704
99268 F20110113_AAAJQQ lin_y_Page_009.jpg
1260b2fd399a2b8f1472b816b40a3bc78bf3d67e
96745 F20110113_AAAJRF lin_y_Page_031.jpg
2bf77c179dcf2acde964a5e5ab5a06b6
9cbe6c3511775a4db01354199de0bc66dbe66037
84583 F20110113_AAAJQR lin_y_Page_011.jpg
8a123a1a75a2c340313c3a49446ec2fe
94165 F20110113_AAAJRG lin_y_Page_033.jpg
3ed510d7f5a4e7987f8c79f700033033
84035 F20110113_AAAJQS lin_y_Page_012.jpg
301c4febf5e877107c6ef87a42eee772
3926aca6c94eb5e940ccafdac4c5c7bb441631ae
106029 F20110113_AAAJRH lin_y_Page_034.jpg
f83206f3334d270de6ee580d88bc06e7
fb9bb80a3f2b83225e0e48f30ecb9a130a3a769f
96710 F20110113_AAAJQT lin_y_Page_013.jpg
cdf461809c3239b1f194a3a605b832b0
4094b45382cc88ae07e90af30383c83c27319908
100327 F20110113_AAAJRI lin_y_Page_036.jpg
50cebf69283cdcf35621fa5ec3324e00
103443 F20110113_AAAJQU lin_y_Page_014.jpg
58fb97dc16f6c349efe242942d16501f
96964 F20110113_AAAJRJ lin_y_Page_037.jpg
18ccbbc8eed1d242c894ebf8c2494aee
b70eb3fe5df04ea3d22f9c54e0bd9677de58fbc7
100846 F20110113_AAAJQV lin_y_Page_015.jpg
02f5f976cfa2efd9d2fca0ef39c2644d5f225ef9
57018 F20110113_AAAJRK lin_y_Page_038.jpg
1cfce993031f8aa58d8bcd14e60101c4
808b8dcf9298eb408c68243a47c321782c9bf6d9
104540 F20110113_AAAJQW lin_y_Page_016.jpg
dd431b4ffddb3b81cc3551222b994191
91816 F20110113_AAAJRL lin_y_Page_039.jpg
cdaebdf195247f5f358270c4b4c999b2
59f9360135fee57eeea11625a32c849ed7d26169
26754 F20110113_AAAJQX lin_y_Page_017.jpg
bb8b0fdbb41411935420530ecab6bc24
7a65002cd2f781e33a4d3c7277d6418e2c567df1
88396 F20110113_AAAJSA lin_y_Page_059.jpg
5c522d3e1d28a9ca00d57c04114ef2d0
4b7d36c7484e5f7ba84b049322e0139f36110d3a
92247 F20110113_AAAJRM lin_y_Page_041.jpg
0b1a3d466f81fb1218c314c153ea07f8
51449dfcda0285204af8e9af97cb75eb5a50d8d2
102656 F20110113_AAAJQY lin_y_Page_019.jpg
74717407d2ce580cebb91a4a025164ee
5ce8c21f422e41dfa97346ee35774b82b62d17d6
98471 F20110113_AAAJSB lin_y_Page_060.jpg
b49f99ab45f6d04d0b745e4767f979d5
732fe46643b3a434aa339ff73539892fb4156d6b
106675 F20110113_AAAJRN lin_y_Page_042.jpg
c427d69a8392a0464bd89e727b4d377e1f02ed89
84631 F20110113_AAAJQZ lin_y_Page_023.jpg
71cdc940e9a37a6c42bcc54a72e5d08d
e8c8edf31cfa489c8901a33ee20cd0179748885c
101057 F20110113_AAAJSC lin_y_Page_061.jpg
94906bb698ca8a924bc07c365e580588
9d97741d5239cb12011878ed1f69a2d3f3d4cbc2
94184 F20110113_AAAJRO lin_y_Page_043.jpg
8d6a241d9b3da5b7dd1b0afc423b74dd
29fc249930851e442c37058ab9baf39da282268b
80459 F20110113_AAAJSD lin_y_Page_062.jpg
428dc71157d02a2474deb3654ff1eacc
26caedf2ede5de9917a866620538ca72d4e598b9
88399 F20110113_AAAJRP lin_y_Page_044.jpg
951ddf5ff10f52e39378f812c1d98aac
52330913fb9e6d72d466d1e1fbf448b01b880159
103172 F20110113_AAAJSE lin_y_Page_063.jpg
ded07f13fc5bf9842bae5bf823b6e21d
3a513cfa3fb730bf254845ca9954e713dc69c63c
99153 F20110113_AAAJSF lin_y_Page_064.jpg
ea74fd23b8c2d8b0224e525b0df94509
c5e7fb51ec2e9501357cd9d8803f2a353d6ff73a
98742 F20110113_AAAJRQ lin_y_Page_045.jpg
f0e7b49bf09723e4cdfa3278984047bd
7a809dd0cd911db10f0606fd03b8ce3887cc85f2
40514 F20110113_AAAJSG lin_y_Page_066.jpg
3686e65ba5240812251a3ec556f1276f
bfabfdb704764c2c21f0989c3840d1f052b75e86
93475 F20110113_AAAJRR lin_y_Page_048.jpg
b2396c7514c4239d95dae79bd88d5585
6b369f120de583afefdf6ea9a79443366b716b45
89325 F20110113_AAAJSH lin_y_Page_067.jpg
bdd9cc11813ac53f54e9de72b3e15390
7f8813d2fbd8d10249b89ef91444699992b22295
96505 F20110113_AAAJRS lin_y_Page_049.jpg
26497c7f3eb6dfa3c6f11773f1eafe6f
0dcac3ae146a44bb6360c797ec2b8eb7f2b634ff
80257 F20110113_AAAJSI lin_y_Page_072.jpg
58d6c6593a98eff6f2e2b8f351e5629e
7e08a8c142e8de519c78ca3d463e1f816f5dfeeb
73125 F20110113_AAAJRT lin_y_Page_051.jpg
31af6139acf770735959807b53d1ec86
c4c0524c0a39efb65f5c5e9e92251878d0cd8525
72727 F20110113_AAAJSJ lin_y_Page_073.jpg
dcb96de73bce8d636c09e53a6d20ff8d
10bea63aa320622478ba98f1b976352eb3dce599
63183 F20110113_AAAJRU lin_y_Page_052.jpg
ba4e17c113e150713c2eec24e42d20d8f3f572dd
89837 F20110113_AAAJSK lin_y_Page_074.jpg
21d06fec3cb6497ef14dda6437490bfe89e20a6f
90641 F20110113_AAAJRV lin_y_Page_053.jpg
e8929b48f8916658c28ffecbed0a3ef8
152a56c0f956360842ea48bb1671e6117fc1ce65
89522 F20110113_AAAJSL lin_y_Page_075.jpg
0240d58fe338eb22953940de804e32b8
a046289d3e23c2b8730c0c6b13ea4ea991b0f546
93130 F20110113_AAAJRW lin_y_Page_055.jpg
0f376db718198f22d054c57518011818
aa81b130497b5cd94b777dc7730956fd83b8488e
88656 F20110113_AAAJTA lin_y_Page_095.jpg
e46e25d9f997f33d29024bb7037b8c91
51814b7f76c9ccf20e9729aa4b4275963630d392
97260 F20110113_AAAJSM lin_y_Page_076.jpg
105775 F20110113_AAAJRX lin_y_Page_056.jpg
f74bc5dff086f3a1e02f7403783eddc2
4fdb5075be1e45cccf83d3450f47f7f287c63634
84186 F20110113_AAAJTB lin_y_Page_096.jpg
ce1337da4d78e18b52aa097df8fed617
e5b4bfce09d94072882f0504b9ba298e5455c19a
91739 F20110113_AAAJSN lin_y_Page_077.jpg
02bca2c4da18c8d66dbe678cde4f204a
dd3fbcdd7fc0a7b2797b596bfba2c8009918627c
78249 F20110113_AAAJRY lin_y_Page_057.jpg
23e877a16aeb7660a6d861fc3811e977
871e791afe57790056a88659539528a02b38dc64
85407 F20110113_AAAJTC lin_y_Page_098.jpg
27099980248221202e1a6b554efd095a
2b2c79d8f4b1c57cd37e46d0f8c8cd0a561b5c99
78526 F20110113_AAAJSO lin_y_Page_078.jpg
f8f596b72bd225a35554848dde59fd94
9896e04c5afbc1361604fe3e68704bcb2284d522
100930 F20110113_AAAJRZ lin_y_Page_058.jpg
102559 F20110113_AAAJTD lin_y_Page_099.jpg
7e1354bce200c80517459e984b56a547
b1b3ea0d767df5b021ac09c072b06a9861d6f8f0
59834 F20110113_AAAJSP lin_y_Page_079.jpg
88df7bb63c026a083057b087abc097c7
6215af8b2c124d76b4216c73d5485fdf396dd7d5
118143 F20110113_AAAJTE lin_y_Page_100.jpg
914150585464a9825e0eafdfd0e7563d
483e372537afef201b0f63a9db323e6a3f6cfa6e
75831 F20110113_AAAJSQ lin_y_Page_080.jpg
7b885f2e10b1a4d6dca91d4765cd68f3
4fbc1250c72ff3531e752e4c891deb0e6980a412
97631 F20110113_AAAJTF lin_y_Page_101.jpg
891dc1b86ec023b3436f1e2827a1810e
a298523f40be22e495bc429170494da8eb6f6bf3
97333 F20110113_AAAJTG lin_y_Page_102.jpg
b5d4c362a9ff3b8fddecd18864263618
45286 F20110113_AAAJSR lin_y_Page_082.jpg
542543643a9554b4da3e5879e6f4c228
242721fcf1e5eb655160d65d2cc799921ddeeeb1
91279 F20110113_AAAJTH lin_y_Page_103.jpg
e91315fd27fef35cb7ca5b4f59470492
37436cc86cd6d6a01b87a83b4af552a2e27a4479
65830 F20110113_AAAJSS lin_y_Page_084.jpg
76608ba876945cccbafa15d41238d19023a137f7
85839 F20110113_AAAJTI lin_y_Page_104.jpg
77d4b43064d1681f1d8da6752dca6016
0c152d843a0de116c90c6ff0ab84f2a7a71b2c2c
67831 F20110113_AAAJST lin_y_Page_087.jpg
2487ff2858f7d919969d738f8cece8e2
2ed61c2e3b48791b3271b26d549cfe3ed91e781f
134083 F20110113_AAAJTJ lin_y_Page_105.jpg
751f56fccc67bf28666d176fa439b73d
6dc89cf1b48826ef749a979b945c4da6c7b4e446
58279 F20110113_AAAJSU lin_y_Page_088.jpg
0bac4bd57991180f9e2bf64f1c2c604f
9be71faba42437a37c4ae596a762d8526f5baa88
40774 F20110113_AAAJTK lin_y_Page_107.jpg
64e38d459e428c143b4710583a27c86f27e934fb
59587 F20110113_AAAJSV lin_y_Page_089.jpg
7b59ef399f309d3b525aa73b35492972
5d790bba820e083917f8037b99d1fe8c88939c1e
82441 F20110113_AAAJTL lin_y_Page_108.jpg
8e2246f690f76d3cb0f472fb0618afa74c5e4363
63812 F20110113_AAAJSW lin_y_Page_090.jpg
49b0eaff944ce7295051be424a2967f8
abd201a76bda589dcd3221578e15ceeeeb765714
91531 F20110113_AAAJTM lin_y_Page_109.jpg
29534ef88254db49c4c88cf93f81301726299b89
68349 F20110113_AAAJSX lin_y_Page_091.jpg
7de2a7199b02bcdd3398624464ae2a28
86572c9e35738e04fd5caaf3e6144c3ff560d4c7
26327 F20110113_AAAJUA lin_y_Page_001.jp2
91cc8f6cfcf15351e46ba28d03c55443
a1a52e458ab6af0f18285bdf2ef50aae4d18e2cf
86811 F20110113_AAAJSY lin_y_Page_093.jpg
05798a1ff4837fde5b8ede92756ef527
2330758d8fb78c6edd5c25c97c4e332681d8911f
5077 F20110113_AAAJUB lin_y_Page_002.jp2
97720d61510d3a8f9274a766e62e172d
5aa3b77618bf6cd721d8f6430be68f08179bf247
108971 F20110113_AAAJTN lin_y_Page_110.jpg
baba77e6a8caaf337b52663935e67dac
809c7cc1e583fea81bc77cda8dae15eee77afd69
72239 F20110113_AAAJSZ lin_y_Page_094.jpg
5309677a286cac34baaa8182ff604e20
b54ed01547bb8e52c914cd7b7016c90f677e64aa
1051981 F20110113_AAAJUC lin_y_Page_005.jp2
b1fd9a42cf7d3a2fd47c50f406aa481f
c27cf05c95045a5defbc69e0caed3599743ed50e
100984 F20110113_AAAJTO lin_y_Page_111.jpg
22d4c46b5330653bd0096fd369c1dfef
fc74aafe33c8acdb15f2d4c543f545d95b98c6dd
1051979 F20110113_AAAJUD lin_y_Page_006.jp2
c053094726763dbb55c4a6188e5d3493
5b8525885bd35f78fae0a4e1054b4f1c6549d426
103533 F20110113_AAAJTP lin_y_Page_112.jpg
bc2d68a9e50fa1368bedb3b27ec6bcaa
7b22aeb92fd5b680bf7f5da2d43ffcf0a4e4a370
278257 F20110113_AAAJUE lin_y_Page_007.jp2
96f7ae7f7f47c2751e7dc644f463ffc0
94567 F20110113_AAAJTQ lin_y_Page_113.jpg
729dd9b3dfb15c68e55918fd28c4dc56a827324a
976109 F20110113_AAAJUF lin_y_Page_008.jp2
0354ffa22f826cc71f24dae7828e78c9
ec5a1d7fcf00e4b99105647cdd434d9f91e3aa16
102012 F20110113_AAAJTR lin_y_Page_114.jpg
72c6f7bcc6b86c001cd50ece3601b3dc
5275c4293e31108ffbfaedd609baa07d8538dc0b
F20110113_AAAKAA lin_y_Page_073.tif
9a6cfca8499c0d4f1cbc765a70ed86ce
86353 F20110113_AAAJUG lin_y_Page_011.jp2
82e125ca588b0e0ace52d57eae03e5fb
b2f2422ba05b2d1130c05884460c00f79c3a9d0b
F20110113_AAAKAB lin_y_Page_074.tif
0f678273a5b82669b02451d65d6bef33
27be8af3f1fffbf648939a2dfdb26e80556e9527
86919 F20110113_AAAJUH lin_y_Page_012.jp2
e665b2baa173372a97b1d95c18c0d5b6
5383731efd7bcf05e827d25f14984957aedebd4f
101318 F20110113_AAAJTS lin_y_Page_115.jpg
22d1a47a5d396ebe7384747f76272aabc4a57a88
F20110113_AAAKAC lin_y_Page_075.tif
87c3e21655f46440c2c568fa988f1439
8385c34989cf71ef74d70dcbe1c23b8d884983e1
100702 F20110113_AAAJUI lin_y_Page_013.jp2
2b7a3f9bc1137c96a1fb4d00d91912ef
108785 F20110113_AAAJTT lin_y_Page_117.jpg
8467989efcd7a244c7839393cd419c16
3bca07d6039f43ce52f967a4d373eda3034a64da
cb463e1fecc71831ffca59f37e4742c9
107470 F20110113_AAAJUJ lin_y_Page_014.jp2
e28cbae5d6bc36229de7c4c16e4b6048
99308 F20110113_AAAJTU lin_y_Page_118.jpg
d0ce9d948d94eab7cd16a159e6195de4
a089affd785843d98118dfd5a42de4c2e9cf04d9
F20110113_AAAKAE lin_y_Page_078.tif
3fb2c206a461cfc9fb4d9da8e8552bc3
481c5cbc9f88ebdd3afb4781f34f05347e5d55d2
106168 F20110113_AAAJUK lin_y_Page_015.jp2
492fff8af2608990914e26c06093f61f
5ddba98e415f3ae54086f982f3b7764f2d377de6
98296 F20110113_AAAJTV lin_y_Page_120.jpg
e090076fc245f565f153bbed60f59bf94d6cd974
F20110113_AAAKAF lin_y_Page_081.tif
c6caa3a089f88b1d28df62d02e70c4c6
fb4c0300f908ec202ca00c3af0dc16781d2dcc51
28909 F20110113_AAAJUL lin_y_Page_017.jp2
895d9e6abd2e0cd84e1d72eebecc8e63
8b48acca56d20369f7035488e813b4a9f203fd80
38887 F20110113_AAAJTW lin_y_Page_121.jpg
5cb092c49bab7e5f1b28e44803bd7747
F20110113_AAAKAG lin_y_Page_082.tif
10af61d314b27f106ae119f8d35ee6e2
58b39aae78de3b15eb1ca5c0e2d7a1af30f55323
109591 F20110113_AAAJVA lin_y_Page_032.jp2
a48af1b239ee5f2cff37a82f80429aaaeb705faa
97741 F20110113_AAAJUM lin_y_Page_018.jp2
d8572453d7f2858d00368de6f4b0f14a
27687909c3e22f356351ff731c46c5aa69097114
102783 F20110113_AAAJTX lin_y_Page_122.jpg
6d97aa42119fd792fc0de64c9b8aa128
0085b5f80d46c22d75cf92af4c09bf877fc01306
F20110113_AAAKAH lin_y_Page_083.tif
d6ea4b94087e7c2d395b1bbca27cd72e
26f5f9fe719beb4439af05c71603dd3c4479ca67
96165 F20110113_AAAJVB lin_y_Page_033.jp2
07aa651948726f5888ce9e88a6f69e95
107553 F20110113_AAAJUN lin_y_Page_019.jp2
9e7fea2d307178e1ca0773fb2b2b013d
0ef7db03e0c5eaec1340e04852aee02c4ce7d0e7
137874 F20110113_AAAJTY lin_y_Page_123.jpg
8c1978837ef1820a01f5202c733506e0
78a4d729de846b67db76c176a0f60f15fb533efd
F20110113_AAAKAI lin_y_Page_084.tif
56797cd5efc0692dfb61c10756ce8d5a
aede8664b61d812c9da89c9ef65218b5cd295660
96015 F20110113_AAAJVC lin_y_Page_035.jp2
d4a350621c95c5bd97a3ca94812d1c9a
643f624e94454776e0578d567b22239c05662bfb
104609 F20110113_AAAJUO lin_y_Page_020.jp2
25fc88da3b934ef555d50bb6e8cf1d279f30558c
36643 F20110113_AAAJTZ lin_y_Page_125.jpg
9ee367d3e93efb685709ff88f8bb09c7
d08482925cda110d9b82cc706c28949c3cfb982e
F20110113_AAAKAJ lin_y_Page_085.tif
2dddb98ebab205427d8656220ea16a2c
5af310787a6d9b2483a33ab437aea8d61e78f40c
106620 F20110113_AAAJVD lin_y_Page_036.jp2
b7cb24aa9e8905574d1d193bf1d37ae1
d4bb58859b6f15b1b08847ea9d458102354206ca
108590 F20110113_AAAJUP lin_y_Page_021.jp2
c599fc749135aff0bdede3e2fb60cee1
545051c01c5f6b8910e41465773eb0c354e08dcb
95533 F20110113_AAAJVE lin_y_Page_039.jp2
d35c4c33e06ea3253d3bddd26d869ea8
fe981599698e0b47b00be37d78785018cdcd07df
100218 F20110113_AAAJUQ lin_y_Page_022.jp2
9c962baa79c7919a3acf7ac9535565f0
2559dd8951df32e62c9ffcf1098c3616548a78ec
F20110113_AAAKAK lin_y_Page_086.tif
31a2c61120c95a2c5cd65031f3bccbd4
105806 F20110113_AAAJVF lin_y_Page_040.jp2
5e96f9fb8a9ecd9ef55823f37a5a1f6e
67f3f9811cd94fd7d22f5022db7c40f58514c528
829641 F20110113_AAAJUR lin_y_Page_023.jp2
0175bf95a4e53011687f5ec92e6a48cf
3f596866aa06ebd6623daa6599f6161e87cf3a12
F20110113_AAAKAL lin_y_Page_087.tif
b544f7b36e1eef92d0dc29c4265d97cc
06b4bdd99047b19f36e63e7de72a2bfecd9ac50c
98292 F20110113_AAAJVG lin_y_Page_041.jp2
2240f368a6170c556f59237464201cc6
6e19062338e8980b183a23210077f2a82611a67b
106565 F20110113_AAAJUS lin_y_Page_024.jp2
cb90c00385be9a4edc0586724b41361f
a0757db736057ba6b92fb11531b8329cb64016e7
F20110113_AAAKBA lin_y_Page_110.tif
493654fc4a2fe2b2676e5b783f6cce78
5d84aec74bd6fcd8ef67ba52e09765b4ed4b3302
F20110113_AAAKAM lin_y_Page_088.tif
1922e45f0af03bff247ca5aa908e18ff
30727dda9b69906096bcace0525f38abcd52d819
111486 F20110113_AAAJVH lin_y_Page_042.jp2
f4c286cd2b7f656e9de8338c30a2685f
9e624762ae827ae89ffef4accb83b2c3a73386f0
F20110113_AAAKBB lin_y_Page_111.tif
de921899bcb0131daeb5deb3fc89e26f
df015495135ed326ca55db71f21b16fe93a5e709
F20110113_AAAKAN lin_y_Page_089.tif
70115ec3bae0e6cc954485b72d39ccc2
23687db4bdb40863cd9f6d8520b1c1852ab6c4a7
99821 F20110113_AAAJVI lin_y_Page_043.jp2
2cbabbc64564ef2df55a283b250047fc
f5990d6b6aa2e282f330f3e1a76fe80a7d7cabac
93557 F20110113_AAAJUT lin_y_Page_025.jp2
F20110113_AAAKBC lin_y_Page_112.tif
f9181ccaf5ffe94c31bd9e901bafa547
feafd6a817fc398547b463aa1819b46c6a657a38
F20110113_AAAKAO lin_y_Page_091.tif
f446b2a7b94b665ef92612f2203ec22c
0fee975f7f635d3ff4ea68f760b3b314eecf2d7e
930298 F20110113_AAAJVJ lin_y_Page_044.jp2
978468576eed90fdff51a7c6ae42f458
17c0bd9e64ba748326c9db3a40f890ac9a7a8e3b
109910 F20110113_AAAJUU lin_y_Page_026.jp2
80750cff9f886f0e0a77cdc902282cab
c6ede436b71e61440161487777b68e88d11532c0
F20110113_AAAKBD lin_y_Page_113.tif
e04270d90783864120bdc5b5d0758d07
F20110113_AAAKAP lin_y_Page_092.tif
4cac52910684af46ef7b464b39f88bc0
9c195eb4d303342979166cf0693bf161fa515090
102160 F20110113_AAAJVK lin_y_Page_045.jp2
1eeb5b8fc3e192f7096081a4413af721
75bae8802b21cae70abe87d13fe9bdb939fe2525
87345 F20110113_AAAJUV lin_y_Page_027.jp2
68fd3f386b76c3ffa4ca3de525c4c21f
F20110113_AAAKBE lin_y_Page_114.tif
4d9690d3ac34a443d1d7922ac9719071
ec0e170d935b9d77742fb9423e14ce5b429c8246
F20110113_AAAKAQ lin_y_Page_094.tif
469aa629ff8bb67743fff83dfb35e1b1
5a8c568c00b7ba0f0e624ddb0864e550be013dc3
84876 F20110113_AAAJVL lin_y_Page_046.jp2
976b88f711e6be1e3b24bc96a7a37bba
9c1aa3c336ea519ab2182b4d076811d0f54c0593
101964 F20110113_AAAJUW lin_y_Page_028.jp2
409c52ac2d0402a75a0e8ac676707957
179dc3caa9e2d67d522ecce935e4d615e6b55a52
F20110113_AAAKBF lin_y_Page_115.tif
70f1af64b4a14a112a3c588edf67874d
43a625404f9135a15efc5a7a211f8b1476173017
F20110113_AAAKAR lin_y_Page_096.tif
e111482f128cbf4b1246e11777225126
56af3c5819e18c6705a16407d1ca37c8107333b5
83209 F20110113_AAAJVM lin_y_Page_047.jp2
826f71ac2b67afe6749489f5a6fff2d8
ec7168cf075380ce36e7d104ca50e29dc5100c59
100919 F20110113_AAAJUX lin_y_Page_029.jp2
99051ea7492f8a8a1325d4dca5a97225
F20110113_AAAKBG lin_y_Page_116.tif
f9875e06f20e3d79e2faa5e214ca94ba
fbe3618804024f58435cd85240898b751d1abbda
102195 F20110113_AAAJWA lin_y_Page_064.jp2
2093bd9d73498b19cd6b119d71f4ea87
c61e9f27d0f2e4d8002d9d9c9176e8b37968d1c8
F20110113_AAAKAS lin_y_Page_098.tif
b0a8ac7fabc4b0838730b4be8f4b57c3
97763 F20110113_AAAJVN lin_y_Page_048.jp2
e5f1061ec0b556b50319b32da852912f
849e3d79e17c1ab9c530b8a7e19fd754b4bdc2e0
111739 F20110113_AAAJUY lin_y_Page_030.jp2
30d1e3573185ab349f5ec4e161011207
995f0693805bf595534c50b8e7750f5d88e5858c
F20110113_AAAKBH lin_y_Page_117.tif
f47cd20aa281e1bac431fba3ea627eb0
2b0134dd5b8c454d01c551392453425102504632
109686 F20110113_AAAJWB lin_y_Page_065.jp2
67d5cac115814fe8922029cd7397e4d8
F20110113_AAAKAT lin_y_Page_099.tif
86c330446024aa2097dddc82ff9b2430d992283d
102423 F20110113_AAAJVO lin_y_Page_049.jp2
8234839a8c067bfc177f1ec0cdc1731a
0f628352047621356765b220f6dd5fe55e0f8e79
103088 F20110113_AAAJUZ lin_y_Page_031.jp2
49155391c2789238cd48f902290b01a5
e31031df12e7a593f63c982cdc9fa42e05b2ac51
F20110113_AAAKBI lin_y_Page_119.tif
298133a2a5c83b92077155e285f0a144
f009c45ffb1f7a01e52f02f0d46b90776e6d1b70
42648 F20110113_AAAJWC lin_y_Page_066.jp2
7dae0f71fcef28e1e3e6d82612feea90
d498445817fe6501a9125ac6802c25619f2a168d
F20110113_AAAKAU lin_y_Page_102.tif
f66e2d9f6e4fb17e38a5827024f1305a195dd2e6
88898 F20110113_AAAJVP lin_y_Page_050.jp2
586efec634978071f460d0148be74017
5d4d3186bf4ee72b83b69d497ef59011fc12552b
F20110113_AAAKBJ lin_y_Page_121.tif
73e408d7380b263e1601243234def0c9
e8b0a54410665da746ef97c9ba24131417eef153
92007 F20110113_AAAJWD lin_y_Page_067.jp2
57ebc92dfea6baedf34590a80214fb20
30caffe2cb5da444051e8a9188111f5f83783cea
F20110113_AAAKAV lin_y_Page_103.tif
bc062474d7858560426dbb76a6fdc1d5
6d191e55fff8d56054c712bde4ffd15f36e4d98e
722280 F20110113_AAAJVQ lin_y_Page_051.jp2
f53b8992dc6e7e978e4c294c23071546
c8cb5c635c539b69e2121879ea6caa6efe802ed8
F20110113_AAAKBK lin_y_Page_122.tif
3c1bbbb8b8b208cfa55f55a96be84ea7
031f3ab9105c315b87962f525887c02d6ac37b93
109534 F20110113_AAAJWE lin_y_Page_068.jp2
0b94006af04e768a7062e2a2ce786279
3859ee0be548dbea6f2aabd8d12eeef95703a4cb
F20110113_AAAKAW lin_y_Page_104.tif
07ef3d6f930d0e7cd5c6dc9f47c1ac47
1a32d4b853e0f0c0d76149530c014e27a408132c
570300 F20110113_AAAJVR lin_y_Page_052.jp2
729c3914bf67f5d1ff82a416293587d8
4bd5a68e22fbbca25e0757afedd80568006c0928
F20110113_AAAKBL lin_y_Page_123.tif
06da729ca49c9cd3c6c4e1256e2fa66a
fd5cdb7c9e7255c657208181127731e9f865e5ec
108860 F20110113_AAAJWF lin_y_Page_070.jp2
5726f9ab5167e5ca06b55dd2e88f2fc4
f70328166548f2a14de059213f7da77697a5c38c
F20110113_AAAKAX lin_y_Page_105.tif
0877ac133c50628872e0d523a7f329dd
c52e3cc46a2f54f33babcdd34891e6e099dc8a5d
109501 F20110113_AAAJVS lin_y_Page_056.jp2
c3f2597032c14ff2bdaf450f6d9f03f3
affe3b8e3211a1fbf64cc85f31b4466b09c8601a
11508 F20110113_AAAKCA lin_y_Page_017.pro
b0a460b0d8299161c95c45c5f8764ab0
F20110113_AAAKBM lin_y_Page_124.tif
252438f31b7f20396cc430a54bb10be2
d5b55a44e58ecb80e42609b06051b72a8b93c458
108364 F20110113_AAAJWG lin_y_Page_071.jp2
fc344a7853385f143089541159f8541c
93e60fb0056985ba605b3aa5ee9480a32b3d955d
F20110113_AAAKAY lin_y_Page_106.tif
777203 F20110113_AAAJVT lin_y_Page_057.jp2
6ebbd300efa88d6f6fa4df47b0831ebc
b6ef9a7c677cdc4a95b2376e96b8e5ef4da71314
42859 F20110113_AAAKCB lin_y_Page_018.pro
c4fb96933b9f9768748e820391fdd755
9e061e261c8baa34328eaf7cbed7a6e686af48e8
F20110113_AAAKBN lin_y_Page_125.tif
08e296a0abccb42a74a4d7440de21850
dce456b3128a26e25e628cd779cc3a0ca7352eca
100929 F20110113_AAAJWH lin_y_Page_074.jp2
c43bd5fe2305710240834873e9da78ab
c6b9a1f327aff79e308dfd3397335fd76c6e466f
F20110113_AAAKAZ lin_y_Page_107.tif
5a0b94a946205001e08362d42efd26aa
15434bfefe714e4de857b135ec42b9bd490b1b73
49242 F20110113_AAAKCC lin_y_Page_019.pro
1eac57fd271556484af584178f4eac72
bc1a98080065a76959ce24e25bd95b111766a779
8945 F20110113_AAAKBO lin_y_Page_001.pro
3596cb4d9419f406bf49fb1960524b86
93142 F20110113_AAAJWI lin_y_Page_075.jp2
04530dae442db88e2819d1fe6f8a7352133383e6
104455 F20110113_AAAJVU lin_y_Page_058.jp2
43b2823acc6d0877f651409573c1f7ef
387973db244dc8cd15654d771cd05b92f422164b
46225 F20110113_AAAKCD lin_y_Page_020.pro
c2b1d3a40367f7e8111c3fe1e74fee09
eeab6c13d327a41b64cd57fe4dbc0effeebd79b9
1013 F20110113_AAAKBP lin_y_Page_002.pro
c1a7d0cc0d414c97f5f5459e2ddd0781
4d23686c78bd5b8f3677f5972b4e8781f35275ae
100668 F20110113_AAAJWJ lin_y_Page_076.jp2
acc283c2026957af33dc553b59509b52
e69b561f4596c7f5c5454e81085983477dc55004
978398 F20110113_AAAJVV lin_y_Page_059.jp2
108369470dd731d39bf279820b39014878683462
49402 F20110113_AAAKCE lin_y_Page_021.pro
c496e1521844706ed62f76f4a8a16e15
e5d22e68390c6206605deae0bd957818c9864c94
1522 F20110113_AAAKBQ lin_y_Page_003.pro
31922a0d0e559d2196999b688046eabe
f5c550a1dc3c3cc20c33cdaf90eb4b26bcde7326
95103 F20110113_AAAJWK lin_y_Page_077.jp2
11db3abcb5f3d4fcbbb65e43ecaf64ed
7417d270b43e8ffd178fa06c7127575cc76f5d20
1051986 F20110113_AAAJVW lin_y_Page_060.jp2
4a3f82bb7ac58c67dd27afbab0e58717
ab77a83529c55a8e98983e7be8bcd757cce160a3
44853 F20110113_AAAKCF lin_y_Page_022.pro
5edc8883cb81866816b90db4d3846b4d70580915
23040 F20110113_AAAKBR lin_y_Page_004.pro
669f621e7b10c1ab8b20268e1665fb5b41548a09
79895 F20110113_AAAJWL lin_y_Page_078.jp2
cf31fce386c0d75d86d6492e5c7aac88
e9c0e8d849d78196fd15f8ae5c3d00aca3b2385c
1051949 F20110113_AAAJVX lin_y_Page_061.jp2
71cf43f474e8f694929364616ee6061c
587f3b366a0f3464f5cd42314466af01a34b3d91
42531 F20110113_AAAKCG lin_y_Page_025.pro
da423c3d9663ce5060955b664199249f
54f7551990eb3bcd705e1e49b4e7466cc35c7381
918020 F20110113_AAAJXA lin_y_Page_098.jp2
800cc307c8813d568cd983a65e77afe0
29e6be63a428c26a11627f7831366b005e6d44dc
85958 F20110113_AAAKBS lin_y_Page_006.pro
0690294b4002c291b631fc1bb8ffe760
92af38b01c7f7fca6c07969a039d6aa2527d3b70
39683 F20110113_AAAJWM lin_y_Page_081.jp2
77673c5e11e41feddcfe3a47ea5ab9b3
1994a92c914bacaed24e06b5c404c637812d6f92
872644 F20110113_AAAJVY lin_y_Page_062.jp2
82df16b98663a1e6cc2bb55dcd009092
3e876a83161c23c3ef2aa5f434fabd36e2abbe56
45229 F20110113_AAAKCH lin_y_Page_029.pro
109549 F20110113_AAAJXB lin_y_Page_099.jp2
bb52d6ef0e644aef97791438bbdb2631
28316 F20110113_AAAKBT lin_y_Page_008.pro
06080e3e8d34b4f4d08c2ce88181d1cc
f963e65b74168c64d0aaab1c9e082fbd734fa1b5
1040093 F20110113_AAAJWN lin_y_Page_083.jp2
d7d8286d328f87def0b1fc83bfc1bd73
28fb6f29324553f4e3e2277b002a80c400d4d01a
108491 F20110113_AAAJVZ lin_y_Page_063.jp2
fa136880255c1de1ff68e84767fb40d2
b6aa4be6ef904b65e123ca1e9d6cab0ea1652506
49880 F20110113_AAAKCI lin_y_Page_030.pro
fce20616d3295e817682fbe3b045315c
94f63daf00d372d80da044af1ba251d78c0f4e78
1051983 F20110113_AAAJXC lin_y_Page_100.jp2
0b27b592aa18b0ce646ae4352d4c3508
ae03fa5d0a523e168a67a17171cc1a0440943337
50712 F20110113_AAAKBU lin_y_Page_009.pro
36247c01ee8b10a3db004720d389173b
b418fb3c3c2f7500b871820efb3c45a2c5ef9dbd
68645 F20110113_AAAJWO lin_y_Page_084.jp2
81d53dc5190f2a2c67b891386c327ec9
0c8abf8bd144121444fe8f63d445c884f651ae19
48905 F20110113_AAAKCJ lin_y_Page_032.pro
b30ef336c70c4345d78805eb377d07206fd7c94a
102205 F20110113_AAAJXD lin_y_Page_101.jp2
8d9dd2798ffc6cb334e0054ce76c2495
7afbdf03d692e7fe39b8be26bed7d8eee9227591
44687 F20110113_AAAKBV lin_y_Page_010.pro
8948b73c55698970877e7ee38214155f
a9897e7e085a997d36343eb09351dd9d13e4596d
788251 F20110113_AAAJWP lin_y_Page_085.jp2
fa4938bfd21c7f29809d5761635ba5cf
167bd8e683ebe06b25ed60ff09a534053c800e67
43247 F20110113_AAAKCK lin_y_Page_033.pro
9082d3e5c1f3147be9655945cc932cdc
102148 F20110113_AAAJXE lin_y_Page_102.jp2
4097b8ec4bdae9187b87effe1724e633
cfd54ef341787e5f25f140a1c2fb1f4186ffcf1c
38472 F20110113_AAAKBW lin_y_Page_011.pro
37113d73d70ac3ccdd441fc8755067d3
77555 F20110113_AAAJWQ lin_y_Page_086.jp2
85b32b548055451ba199d2bd83018dbc
a1f73bdb6c1ed725d0139604638f7a9d770f071b
41761 F20110113_AAAKCL lin_y_Page_035.pro
54808c45e58265e75249521b73fd6b96
bf2054c27dcdc69ac0a09f7ef632c3e7aae173b0
1051985 F20110113_AAAJXF lin_y_Page_103.jp2
09de2754e9de4256243244ffdbd15367
37809 F20110113_AAAKBX lin_y_Page_012.pro
7389ec5e6cb0a76d069a2615af53e663096ef5a3
669085 F20110113_AAAJWR lin_y_Page_087.jp2
42a235cf3cc1d1ff10bf0ba16d1884a8
61267c9261cb631f7285881db6e6528ffec228a3
33353 F20110113_AAAKDA lin_y_Page_054.pro
cf8cf3c6fc7ab5216cb5bcc1c3a1a884
edb524cf386c2f498a135cb598e5fc4f3ef3fc1b
17329 F20110113_AAAKCM lin_y_Page_038.pro
d9d90956bf239a12b2a93dddb4fe2896
910422 F20110113_AAAJXG lin_y_Page_104.jp2
59ab648ef7c01bb7efccabc344c76574
3f4535e53d0152985bf73ec19fb74d3e68ff6fec
49031 F20110113_AAAKBY lin_y_Page_014.pro
612015f72038b9e31f6ea28e637eef89
2eb5f62b6ef2b9c6c246276681190a8840fab794
60962 F20110113_AAAJWS lin_y_Page_089.jp2
8fa55d14e21eeb808edbf85d3958b619
1ca6b109a6706f1c1cc899be245fec006eac2658
41765 F20110113_AAAKCN lin_y_Page_039.pro
f624bd26396bc11d8b16fe9d1452e7be
844c54f20e689a58e3d32a64c21547f52d589d37
1051973 F20110113_AAAJXH lin_y_Page_105.jp2
eef5df180126dbc698085231b8789033
df1b6cc13f6355c0a9808283988691aec1e8d6bc
48428 F20110113_AAAKBZ lin_y_Page_016.pro
87361a9b7cc3d3772d7968122bb6faca
47ec514532374a8306883bbb47a32c8a5725dddb
52738 F20110113_AAAJWT lin_y_Page_090.jp2
af023db95ca6ba678c9f7dd8b06cb31c
39274 F20110113_AAAKDB lin_y_Page_055.pro
4f81db15a46c6a708c37e77c69abc1ef
93ca4ef678993614691a2d0d5e3a0ae60a51e32c
48522 F20110113_AAAKCO lin_y_Page_040.pro
fb3e0ab43eddfcc41a414627e839c84e
96aea37f670ae8edc7e47ac52760b4cfc201fed5
815092 F20110113_AAAJXI lin_y_Page_106.jp2
c2451a150d703ff12e9761da212fedd4
6da013a7a65f6a2bdc0767ba9b4ecdf31f744e46
58910 F20110113_AAAJWU lin_y_Page_091.jp2
31cdc446c20cf62100d7bfd0f87b7455
27572 F20110113_AAAKDC lin_y_Page_059.pro
2851381a710daa0d739b6457fdefb627
83012a61de8689bf1e8ebee5e87c1c73a7841425
44718 F20110113_AAAKCP lin_y_Page_041.pro
5955c971705140a49a181643701ccabb
68281215d1a5f3ce11dbdca09857c6eb8fac3b16
42937 F20110113_AAAJXJ lin_y_Page_107.jp2
e5dcf3e5d9ced7a4c1c068ae65dd131c
a869b732f9463f5e3530cd79986504be93ff820f
11700 F20110113_AAAKDD lin_y_Page_060.pro
322670d1256b82c63c764c596edd3b63
34fce14bb5a8c6bcc89f2e03f5a87ce94b6757dd
50074 F20110113_AAAKCQ lin_y_Page_042.pro
dc697a2712e2436b959734e1d0b7552c
2aea52d34aac970539e310615e3ecffe7bc6d6ea
87523 F20110113_AAAJXK lin_y_Page_108.jp2
bab036cfae937c6fd0eb4c0cb5f67911
a6b929fdf63f1fd42bc1cdc29791a9cda2032aca
913344 F20110113_AAAJWV lin_y_Page_092.jp2
e47c69e379cbec7f0758205e52241a1665266720
33251 F20110113_AAAKDE lin_y_Page_061.pro
7255574dd2db3b3694c8c91602cea89a
0f83bcd0e9a3d2f933c2e0308066562f4f74bc20
46996 F20110113_AAAKCR lin_y_Page_043.pro
85768aaa2c978a73d4b67be0cb12c1f8
b13f0651db5780490ba478b6cc8780267874dbb0
101719 F20110113_AAAJXL lin_y_Page_109.jp2
6ab2d06b3b90ef6016c03ef5d1c7e596
04f46d32cd5e180dbdcd3ae159b443d52812b6bb
1006079 F20110113_AAAJWW lin_y_Page_093.jp2
78b8d08f8d38cb20a526f79f95270422
8afbab32848ea3d3041131bc0e22a3b96bbb7e93
50216 F20110113_AAAKDF lin_y_Page_063.pro
6df429838fca7cf601d901f69ebda671
9fc9b6bfa2e7b9af9748813bf8c7ceb530d9db2c
F20110113_AAAJYA lin_y_Page_003.tif
3a89d9ae8eaa0db771f129267a4ef25f2f9f66d4
47206 F20110113_AAAKCS lin_y_Page_045.pro
2eec8812f3642ecb2a52f1070b7dc7d8
b30816218b4e373c4b563d25c7bbb8db4842d58e
114769 F20110113_AAAJXM lin_y_Page_110.jp2
b80d7e5571029ec7557576645b2f7597
517490eab40d6f791e20c2fb9b9c26428e374d72
77259 F20110113_AAAJWX lin_y_Page_094.jp2
c306ac3ace7038c67ca0d846326344a6
49f7b35c82ca81bfb4e09777737f11ddf293bd7b
50109 F20110113_AAAKDG lin_y_Page_065.pro
47fbdc79970df7c10672cce75bf1972ec70b4fb0
F20110113_AAAJYB lin_y_Page_004.tif
b96eac5cc94f3b2880c53582c22df4c5e0380469
36424 F20110113_AAAKCT lin_y_Page_046.pro
752c7acb76f6ed0c5cc7821f52597541
b4466358ee9a8be282813c0d24b20a33f0c5d51f
104886 F20110113_AAAJXN lin_y_Page_111.jp2
6bff9ae6fff97ee7bacc6974c52a1e16
8fb31cf718fca39de6bc2bd02b1bc0720cac274b
81397 F20110113_AAAJWY lin_y_Page_096.jp2
558dcfdecc30d3deec03f54becee6d89
67cd797b7cb61ea0b8cb14f2678d2a76521cf420
40971 F20110113_AAAKDH lin_y_Page_067.pro
f0b79db3dae52fb0b91598a5029d6d19
b2440a676b20a90c67a54f51c9788d0d638e2c87
F20110113_AAAJYC lin_y_Page_005.tif
9bf6b02d591412afc111d4f5fee3104c
b422578df2a184733887663ca5387a127bf4ed7e
43768 F20110113_AAAKCU lin_y_Page_048.pro
c13f1b0e7f2597609aac86e4704979cf
8d30dc7c4509770847256b94b79de7abce587257
108325 F20110113_AAAJXO lin_y_Page_112.jp2
704e1e67a727856606ae8113e8fb9eaa
276b1aa674acf2b506a39635e47fc74c76e7d36c
86563 F20110113_AAAJWZ lin_y_Page_097.jp2
d329db402a42b3eac7bdd0c73b670df5
b20caa31abd8a196e8d7a5770702f66bec9a6e08
49221 F20110113_AAAKDI lin_y_Page_068.pro
06efe1c7816e01be2a15f8da756de538
F20110113_AAAJYD lin_y_Page_006.tif
3fe75bf957f19339ac95c78f8f10b210
0b70b9c89a446a1dfaf382a1ccd83fdbf4a31efb
46340 F20110113_AAAKCV lin_y_Page_049.pro
f7636c83e3f93c9b32f7dab016f72420
126846db49633c0cfccf08ce318ffebe3e81636c
102058 F20110113_AAAJXP lin_y_Page_113.jp2
ae8038a0cd32a340b9b68fd7396456c0
f0f240ed692926172b9de50c605b3923f38c9665
50136 F20110113_AAAKDJ lin_y_Page_070.pro
12941c37bfd238bce84190d02bccc3537dc1ef84
F20110113_AAAJYE lin_y_Page_007.tif
9c91b4981d9464f94665c94f1b1a040c
ca46498cbc62b9539dbb869cf9a13f1fb503618d
35765 F20110113_AAAKCW lin_y_Page_050.pro
4c78241bedcd60419aaf12948a5dc8c6
c541c9bc3fafa3841aa476fba6b5eb59babdb3d3
108977 F20110113_AAAJXQ lin_y_Page_114.jp2
b35d883f6bc87c628a1d68efb4b26ba35c200450
49842 F20110113_AAAKDK lin_y_Page_071.pro
4357db07d62a6a381e2a17a1e37964b4
0076e288878a19507528c6d540f2c292e39a962c
F20110113_AAAJYF lin_y_Page_008.tif
e043dbec148db3c33f9414f2cf80f21f0deb64a9
27216 F20110113_AAAKCX lin_y_Page_051.pro
a6a0d7eefa24e65aa657faf023cb517ef7ff4ebf
105513 F20110113_AAAJXR lin_y_Page_115.jp2
25437b7b2bb1af157aa4089c90e91236ab07f39e
26348 F20110113_AAAKEA lin_y_Page_087.pro
42f7ae21831ebfd8fecc17455af24c7c
e0092bf6d72e9e9f98d8f72f4eb870d04d2e0693
28776 F20110113_AAAKDL lin_y_Page_072.pro
89121018ff47ae2bf677848f385f69c7
34b7433c184e1d99abf11aa13ce793f1e0d20b86
F20110113_AAAJYG lin_y_Page_010.tif
f0044599524570cf19891b828d5a6e5fd8c07891
12679 F20110113_AAAKCY lin_y_Page_052.pro
ab8d971f5b8e2383bc5c04ce285da858
b5ceb48bb75ab8fa2954429d30e559d208a366ec
111450 F20110113_AAAJXS lin_y_Page_117.jp2
4d5b6f5c7a6eac83fd8ae698a21085de
5c732e0e3159a020dd432da43e6b81e95205083c
25189 F20110113_AAAKEB lin_y_Page_088.pro
d55c504694cbc1f8344e49727d665c85
426320a546dd76bf63ec2c54b9092eb15aa40ffe
25940 F20110113_AAAKDM lin_y_Page_073.pro
0d4ea8244526c0c6a1eacf45434ab501
6a6ebbd91e5d1fb0801c2766e31d9fe458cd36a4
F20110113_AAAJYH lin_y_Page_013.tif
37e95179619b1a23076cc03990a792dc
92c0974c0cdb988fc590db46222005bc1380a7b5
34788 F20110113_AAAKCZ lin_y_Page_053.pro
4e5c5e34efbc9a5dff60253365c93aa0
abbf06b746d6e6340487df3c7dcaf461056468c2
105261 F20110113_AAAJXT lin_y_Page_118.jp2
7445bb69bc9f5a8c0d5971dbc28c6562
9d71c54ed45d31871c0ab77f106ef93e21121574
38051 F20110113_AAAKDN lin_y_Page_074.pro
0a68e0e6dcd197976a59445b2be1e5e1
12096d1cd1bddf36644f1e409b5d5c8d8f278e4f
F20110113_AAAJYI lin_y_Page_015.tif
99ee3351809626279a5c6f062dd1bfd2
40548 F20110113_AAAJXU lin_y_Page_121.jp2
41f84165f55e985cc1e99b6fc27dd3f1
8790b88ed0c0c88e226b9144e4b81f73ce7a6d3a
26335 F20110113_AAAKEC lin_y_Page_089.pro
ef6b2e43246e0508a4208202ce21e829
5b415acba33ffc80fa58347ec88201f49fa64b86
42284 F20110113_AAAKDO lin_y_Page_075.pro
333949d205f8a154a7360e171120b8f0
172230abcfe449abef256fe027c4dec587350c68
F20110113_AAAJYJ lin_y_Page_016.tif
a1e4b14f9a1252f1a2001c027ea40ea68f3bd348
135908 F20110113_AAAJXV lin_y_Page_123.jp2
3ac9b27bc8035ce6a2f6e4a0740e75b3
f0bd350f941079c99b24b939efd85fed0638882a
19684 F20110113_AAAKED lin_y_Page_090.pro
3447f81df6c3242853baf915c7a7f496
eca115b9f58ef8f9d34c36839ce98e69080c4ba9
45316 F20110113_AAAKDP lin_y_Page_076.pro
6f8c2f21474d519e14a2897d0377e809
d00c520967a4045cb8df13ff6efefe897dc67a14
22904 F20110113_AAAKEE lin_y_Page_091.pro
d03695ed816a4f86c431fe7a5ae6e3c2
42725 F20110113_AAAKDQ lin_y_Page_077.pro
2ebf4103edb1eaaf350e0df17de2955b
2ae66b79be06e1ff6f16cb7fbb5fe5bfb80a8e42
F20110113_AAAJYK lin_y_Page_017.tif
e7515c3e823d0beb839967a887332e5e
97e60cfe8e634211aaef0f8590b7e258eac47f73
111246 F20110113_AAAJXW lin_y_Page_124.jp2
f911fed86d93a86a911ffc620cfb0a0d
0a10f011fbf144d6700b52573c8640079924f65b
36565 F20110113_AAAKEF lin_y_Page_092.pro
2616b3b9b61263b675476e56dc48e9e36ef86a52
35379 F20110113_AAAKDR lin_y_Page_078.pro
15d935547c81de347b316b0f9266d230
f3085502c68c60c1b2e24920bb0d22cb36e50090
F20110113_AAAJYL lin_y_Page_018.tif
9ed345a5dab9fe4c7773c0ef0c8b767e
0fae14a3ef3dfb7b9630c414cdc79bcdfb1a63d7
37213 F20110113_AAAJXX lin_y_Page_125.jp2
970ce545517da5983a386e8487b57e38
41772c4c59790155a149b5d2dc988087e6e9ac35
28131 F20110113_AAAKEG lin_y_Page_093.pro
01e86e867980896aaca6c7a4a2b81692a69376b8
F20110113_AAAJZA lin_y_Page_038.tif
83d2a66e3d646cfe7f9b7f52d8550a5b
9acac4d47bf033fc04e9415edef6222bc28589a4
26196 F20110113_AAAKDS lin_y_Page_079.pro
389762951471a8af1d63715fef562242
F20110113_AAAJYM lin_y_Page_019.tif
24e7896c6b869a4f613197c486008b50
0882c4012d1a035fd0b83cbfe812e2ef68d844d4
F20110113_AAAJXY lin_y_Page_001.tif
b0b9de56b112ab7c904b66a878cedf7a
733d59fd843fed0519c1f7f8ac826c18c5dc9cd4
40780 F20110113_AAAKEH lin_y_Page_095.pro
9911496d87c155e5b0d89de7ffa7500d
517568ef707d0c59ff8ac96170c571e2aceb5604
F20110113_AAAJZB lin_y_Page_039.tif
68a3949d21b37b13227031701eaec9dd
37009 F20110113_AAAKDT lin_y_Page_080.pro
F20110113_AAAJYN lin_y_Page_020.tif
c25aa96fa27a27c1f7ff1cc322b6f5a6
8be7cfbaae8138a7a20de877e604790cc63e5685
F20110113_AAAJXZ lin_y_Page_002.tif
87986f22bfb9ca0bd744b87f6e27f386
49950 F20110113_AAAKEI lin_y_Page_097.pro
d54a9a07499b3502524b1b21d3b579f6
F20110113_AAAJZC lin_y_Page_040.tif
665286c8f7bf3e30dd003d2e7186a2ac
20481e258814b22ce61e5a64c406158fb5294ccf
6210 F20110113_AAAKDU lin_y_Page_081.pro
2d8f4d083d27375ce05a8938efc0ca29
F20110113_AAAJYO lin_y_Page_021.tif
7ca31b44687a54a3cca02d5a368c23fd4939b7ee
53638 F20110113_AAAKEJ lin_y_Page_100.pro
db5cffc779dbc65d050c187b6b6d465e
be4659b85cebd59e1846e873530c6b32d4d733c2
F20110113_AAAJZD lin_y_Page_042.tif
5dbe0f8e94d8467ef0469885f0a0fa12
fe51161811792576cff432e459059d9d73be5119
13200 F20110113_AAAKDV lin_y_Page_082.pro
d46f1c0cde24fd05fd73042cd0ab9f36
43fff21b2bce1a736ac1448c875ba11472b4c207
F20110113_AAAJYP lin_y_Page_022.tif
84fee488e10eb6b02a46ba1766535ee68b6ba929
47229 F20110113_AAAKEK lin_y_Page_102.pro
430996285e141d0fde8d79facb17b1be
f9b0e131b5b502e83c9824046d3b828b2471e0d9
F20110113_AAAJZE lin_y_Page_045.tif
fca5b8def8bfb7e4b038a2f385bb810b
3c5f613c0603b0db1b61799eb5f6347ff996326e
38375 F20110113_AAAKDW lin_y_Page_083.pro
649b2ff30450493e1cbb37f622cdf7d0
c13072b8ca2963cc7aeda2c3f98827dbe2a83900
F20110113_AAAJYQ lin_y_Page_023.tif
b7218e2db1a530b5547c958dd5b03d7a
e130b6931b5169bf82e525298916a8fbac3f4be9
49740 F20110113_AAAKFA lin_y_Page_122.pro
ded00a3d586cf9fdd033c94913473b2c
385f6ee6a0b969235034a5fea4d86f162b6c3bcc
34972 F20110113_AAAKEL lin_y_Page_103.pro
859c43426d82e204ab3fc9ddd8d81f09
cee7e0f43da01f74349fd1c178dd6cd0908d6950
F20110113_AAAJZF lin_y_Page_047.tif
c2ec5489f191bdbe9712b67ce057b6ec
4d75e432c9476cc540fe374a0f241ba54dbcb43a
29791 F20110113_AAAKDX lin_y_Page_084.pro
2536619d64468622bfe39ab1b346493364d8aae8
F20110113_AAAJYR lin_y_Page_024.tif
c074f0d1d5e8565162f1074716947c69
64849 F20110113_AAAKFB lin_y_Page_123.pro
51f8ee2ccd48925d8c94a2a225ea0e70
f20091b09ef9afed66e23c51b413f46bd59d477c
73250 F20110113_AAAKEM lin_y_Page_105.pro
5aac4ebfa8f052432252259c69a8aff4
f2860683d55704d6216f354475d184d5bcfd66a5
F20110113_AAAJZG lin_y_Page_048.tif
73b0d5ed57f1d4efc678f004ca7820c6
e568f2754c20758bd2be090d620999effdcde868
29671 F20110113_AAAKDY lin_y_Page_085.pro
d5042718fc945821429ff1ce60d183c1
6b25654b0ce5e630f9f9aa06b8ca38c4dee27bde
F20110113_AAAJYS lin_y_Page_025.tif
7ac290b23f33447c945edaeba471d91e
51329 F20110113_AAAKFC lin_y_Page_124.pro
ac03a99fc80191a663ca87f12953c247
18025 F20110113_AAAKEN lin_y_Page_107.pro
fc440aa8cd41beddfe3d354e16c9ec7c
F20110113_AAAJZH lin_y_Page_049.tif
1121a8fbd818d1002e567357cea9c706
20594 F20110113_AAAKDZ lin_y_Page_086.pro
1b978d878ea25013f2a1a41f98b16909
7022d87a067faeec6c231332f39ebab6a2824154
F20110113_AAAJYT lin_y_Page_026.tif
352ea01954ef53b3eb4a4b799c5d35c0
dffd160327be2130a065af39dba8cf582d93cc13
39286 F20110113_AAAKEO lin_y_Page_108.pro
5a6ba6c2e3bfcd45a253d05d953b9b82
eb3187cbd2fb2b0d274f347077730840e9ff28b1
F20110113_AAAJZI lin_y_Page_050.tif
b449e66033716a2fbb2e50b726eecf16
785f49272e8c33f9b0581e2e7dc79f71669e360b
F20110113_AAAJYU lin_y_Page_027.tif
1509d226f21693bf33185beee3e06a12
011b649dbc4c18239c599228c4b8c9537386f0f7
486 F20110113_AAAKFD lin_y_Page_001.txt
66d0d6183632c87bd0fda1e0df25a989
aa656e45b569459ec1720dd761d9df596129e5f4
45552 F20110113_AAAKEP lin_y_Page_109.pro
36d4d1b410158a16147a9cb1547a3f92
b392e1180e6cb68a159c90b9026d58d1b1c6a2e4
F20110113_AAAJZJ lin_y_Page_051.tif
759f2ce04330a4cfd02487d8f3fca549
5f6deb8683bbc8cedb729b8fa1726c4982f908c0
F20110113_AAAJYV lin_y_Page_029.tif
2031c39d2e4dde508049e747c6a15c039ee772c0
106 F20110113_AAAKFE lin_y_Page_002.txt
d3d6d5721a48b6bd1165b7cfd30382bf
626b20ae3662ed5b2cfbffe4b9704912ba35287b
48390 F20110113_AAAKEQ lin_y_Page_111.pro
b841430b1f6a1cb12518a58abfcf9b52
c025a8246bce5965a93278be90f5d55dbba6ffd0
F20110113_AAAJZK lin_y_Page_052.tif
6bcfb66ac02b6c9cae45ae0b1830eabb
3a72aa4dd2e51c04af5c8a4fd7c45df0d3deeb2a
F20110113_AAAJYW lin_y_Page_030.tif
cbd5d602f462aa41b8a6209bd71ced05
770d2d6f6d178364d009338ac80e327947471215
126 F20110113_AAAKFF lin_y_Page_003.txt
408df4ddb182034b892b39cc6ab23874
5e217e51dc40001e10cec43fc7cdb64389242580
49033 F20110113_AAAKER lin_y_Page_112.pro
d4d9e044e521d79fc071e0d7f4917990
5453944324544c94d62b6314f9f11db00f03eaf6
F20110113_AAAJZL lin_y_Page_053.tif
646be5b647a7e05ef2c35cd55a2a067c
f4d7a046aee9ce5712c91528b14c6c5e0f58062e
985 F20110113_AAAKFG lin_y_Page_004.txt
188a878ba227dc323cc445fe6222a7fe
46793 F20110113_AAAKES lin_y_Page_113.pro
bbbf84f806c3e440243484571e4806a2
8f53622588d93c511f52ac294ba262f8a56e55cb
F20110113_AAAJZM lin_y_Page_054.tif
aff63a04cd7429874e3942134b1cc78e
bfcd4834d5d09b82fae5cf5d69457925b388628f
F20110113_AAAJYX lin_y_Page_032.tif
c955c6394e7d8b747f3314abab00a8a4
56136ae21381448b3d83659f877b789d64576e2d
48645 F20110113_AAAKET lin_y_Page_114.pro
F20110113_AAAJZN lin_y_Page_055.tif
640928076421d86d150045411d8bc6f1
4ec0cfe19620a6c475269df08d9edca79d5b468f
F20110113_AAAJYY lin_y_Page_034.tif
b5713e354e60a8a2e7f2898d17b206c8
7835bdcd5f652ec7e270b88a4cfc63db04297143
2553 F20110113_AAAKFH lin_y_Page_005.txt
f4656db4c936a3393643d8681c29bf80100504c1
46113 F20110113_AAAKEU lin_y_Page_115.pro
332ed76a4305cbff755d417c2f82e41375db0179
F20110113_AAAJZO lin_y_Page_056.tif
d7262b3b3345e9a3fb24f19fb5f10a6b
99f572d62873b1d3df2f96054268f006e86cd769
F20110113_AAAJYZ lin_y_Page_035.tif
a2574fd0a5ca7b21b1f2c6384d746125
a4a875f195abd8fc12c6237f42efaffc370e6361
3716 F20110113_AAAKFI lin_y_Page_006.txt
0193b7909f8e09ca6398f92127928be2
14dfbc7b0c55089c8db2af408094c0a0e294cfef
49792 F20110113_AAAKEV lin_y_Page_116.pro
fc387b850ce4d1ee233334c36983c199
292cee88d7d4f7915735785e40b673b0c2e3ed80
F20110113_AAAJZP lin_y_Page_057.tif
61bdf7700043736c00d7bae874d61e2a
400 F20110113_AAAKFJ lin_y_Page_007.txt
fa06b99f347272f31acfb676fc4c1c95
49987 F20110113_AAAKEW lin_y_Page_117.pro
08b944824eeb2f3b1c2ed2b5ef4a13e6
2f2a0cd2105a1de6fcac90e36a3a07fe4dbded5d
F20110113_AAAJZQ lin_y_Page_058.tif
83c9a75236625d21070676172b01c0b0
4aedd4b595c1c4c18430235b094782395aa430de
1227 F20110113_AAAKFK lin_y_Page_008.txt
5013e8a61a6d5df9313b21813f313645
bb43a8cd89357d5c671c0a723ec92a38ac65807f
47471 F20110113_AAAKEX lin_y_Page_118.pro
904ccea70ef3da631d1170d1991b5d7c
F20110113_AAAJZR lin_y_Page_060.tif
845a66a4a4c6ef2375d6b42f921f960f
1736 F20110113_AAAKGA lin_y_Page_033.txt
cb914cb87ab7c1f39f31333c6abaed8b
d6606ee1a8d9da030f8bf90304bf41f7804b3ee1
1690 F20110113_AAAKFL lin_y_Page_011.txt
8004a41451f6d384cccdaf028da351c8
7b932022418a5992bbeaac3aae5d7d52020c707b
42667 F20110113_AAAKEY lin_y_Page_119.pro
4dea5c9633d1627243f925da66cc378b
961d140a9ef10df4e6e9bc02f24bdf87fe37ea9d
F20110113_AAAJZS lin_y_Page_062.tif
3b87fdec5f6803c69228acf530cc0956
996c4e844737f64ec822a01f191a9ef01992c752
1999 F20110113_AAAKGB lin_y_Page_034.txt
08fe5ac041cd616819e398ae727f8084
6d54eac694ae131b32155693a9987c2ddb9a2fd1
1514 F20110113_AAAKFM lin_y_Page_012.txt
745c5f1a36988b713cde78e46aa24e1e
45215 F20110113_AAAKEZ lin_y_Page_120.pro
3cc9fd01082de8e8f524b2936ac552cd
F20110113_AAAJZT lin_y_Page_063.tif
2408bcac89ab30c973d18f6bc1c3ef9f
2102 F20110113_AAAKGC lin_y_Page_035.txt
b9649af6e43659be394564aa5ac56b29
6dfda71ec88cd3e2b142bcd11906e88ab252eeb4
1945 F20110113_AAAKFN lin_y_Page_013.txt
8d632f7f5c71b5042409e00487c53a8e
4026035ac53df36362785c974d70ac0659f29947
F20110113_AAAJZU lin_y_Page_065.tif
778e2218aba14c0d4675c8513a841a70
2e4dbe3b74edf3416a4d1fd143cdd3f7910ee8d3
1920 F20110113_AAAKGD lin_y_Page_036.txt
61a4aa8c1c5ef3f773d8f9ea16678e30
f523e0718886bb01e0e1269fa74ae61eeaaed4c1
1963 F20110113_AAAKFO lin_y_Page_014.txt
793c374c2eac1b94071334797951efe3
F20110113_AAAJZV lin_y_Page_067.tif
3a015f153097a00f20c19d2254b0eb65273de117
1902 F20110113_AAAKFP lin_y_Page_015.txt
23a74c654acd6897a79ff9e1c4759f80ac555ab1
F20110113_AAAJZW lin_y_Page_068.tif
5f8509e9a03d2f955c8316eb47289395
869a36467b96c6a25c9d8580ce201d6b1f9fccb2
1876 F20110113_AAAKGE lin_y_Page_037.txt
f1d17764eac20f7b3db03d2d7d827d21
de6102f7c4e503495b15ff870979af818d1d2e90
469 F20110113_AAAKFQ lin_y_Page_017.txt
3f2b7a317025822c0d36f2c177d9738c
6a0f2e587f6355122871e070c78283701f93c5a6
F20110113_AAAJZX lin_y_Page_070.tif
9ef12966a3c64caff78329c4f5598152
9f29c6ebf08ea1d051179daea78a6d18b6bb7efc
734 F20110113_AAAKGF lin_y_Page_038.txt
b04940fd4bbe175761f031431171f6a3b715d12b
1793 F20110113_AAAKFR lin_y_Page_018.txt
577f436fd4dd0dabdf819ae43f8727a3
f0e46b8e5c483d7bcca91d11b950b8bfe340844e
1926 F20110113_AAAKGG lin_y_Page_040.txt
079c83998b9d784fd2f05e22c523bb49
341f9b99c1120966b5ddbab20e3137f5f053926a
1944 F20110113_AAAKFS lin_y_Page_019.txt
b037666336cd2898b7d7e0270ddcc475
F20110113_AAAJZY lin_y_Page_071.tif
2d168e6410a63c00cfd16aff65abb074
6fd5c6341374c06476428528525e958422dc8f79
1819 F20110113_AAAKGH lin_y_Page_041.txt
b52fc11e2e5b86f09a4480a8f4d876e4
e1d3fb651eb29c4c09ab1be36951099832794f18
1790 F20110113_AAAKFT lin_y_Page_022.txt
2522dab0003e5ab18104c4103a9e44e3
f6e1a1e76064a01388bbca214d3eb06a2121d89a
F20110113_AAAJZZ lin_y_Page_072.tif
7ba6f07f6f24751ac1006cf0dcf34be6
6f9c130ce5807bebd780f79f1635c273aac645d7
1969 F20110113_AAAKGI lin_y_Page_042.txt
3c7a0d19f8378737064fe618f4b1ca82
067813990f680b2d7b8a04607b72878f5a6c62ae
F20110113_AAAKFU lin_y_Page_024.txt
8131a829f21ec53d60247964e29b0d57
c5d2955e6936dc6ca1ab5305f291612d19983c2a
1665 F20110113_AAAKGJ lin_y_Page_044.txt
61dde27f9e0a736ebb2c0c419bcc7bb7
1862 F20110113_AAAKFV lin_y_Page_025.txt
c0ed9ddce010f10b3caae539fa2bed61
1fa6f61dd79b57ec2376ef6d37915030d18aa749
1894 F20110113_AAAKGK lin_y_Page_045.txt
7f0fb439d9542baa263b4ece96e9bf1b
e3c87b7b42ebb0d73ee2667345358d1cd8a6244d
2023 F20110113_AAAKFW lin_y_Page_026.txt
0e89dc957b00be86d1ac2a3740736f83
1211 F20110113_AAAKHA lin_y_Page_072.txt
398dafaac2545b3c7990b7ea0542be39
1721d48d20a2b203203becdb89366c1424b20bab
1548 F20110113_AAAKGL lin_y_Page_047.txt
a3dcd83d34f57b449e99eb7fdb6abda0
35671b07c7b76a11fbd6793b86f3452731a2ef84
1197 F20110113_AAAKFX lin_y_Page_027.txt
8b1018273f0d24d6147591d0354d559c
d33033bb63b56f870a6193b0da7c411105d4417f
1184 F20110113_AAAKHB lin_y_Page_073.txt
9dbd1e525e457a4cca16e2683efa0b54
85fc964d5a228e164b96fddbc82b73a817c86971
1784 F20110113_AAAKGM lin_y_Page_048.txt
3fc696f861a2db87ca6553ee35eba9b213bfa252
1879 F20110113_AAAKFY lin_y_Page_028.txt
684dd2ea171e95db9731f9077b178eaa
bae83d0c5483c64a012fae4953393676fac516bc
F20110113_AAAKHC lin_y_Page_075.txt
e76c5794d8ef761779087d07e114179d
dba71f99c9342e16ea82b764ee70b4b98bef5689
F20110113_AAAKGN lin_y_Page_050.txt
47ae8d70958ba80bcdc3572df9fc1b85
00d4b12cc7e75a08e0c5e95ac983bfaa0a6799ff
1797 F20110113_AAAKFZ lin_y_Page_029.txt
48495c5d75b3072467d70598364fea1f
c90a2b2f18135a584c157a280a088b92240d47c7
1805 F20110113_AAAKHD lin_y_Page_076.txt
019fdf31586bd48a829bd07f58e92b5f
729 F20110113_AAAKGO lin_y_Page_052.txt
c691b291fc49b96998175be2048369e9
1709 F20110113_AAAKHE lin_y_Page_077.txt
fd29cb6ccb9a98ab1702c0c076cb7fbc
1457e8789f3519201323ece15e044deb075c69ea
1704 F20110113_AAAKGP lin_y_Page_055.txt
e46dbc54f561818dba067e8fa613920f
1362 F20110113_AAAKGQ lin_y_Page_057.txt
ac28e40e74f41813d7717cc5e4cbd552
5a04c5e2602f8a1e83fc5cd4293c8104cdf9f8dc
1626 F20110113_AAAKHF lin_y_Page_078.txt
046fcd0e4d1fc327eb15bf88d7983b55
80b88587a98d1ef448928edb644299febf9b499b
1237 F20110113_AAAKGR lin_y_Page_059.txt
ea16671019e4bd821e168f6ab629c309
6fbacffb433f43b1b18daec49993d7b759b77777
1837 F20110113_AAAKHG lin_y_Page_080.txt
dbcb3a62819f5ee929f362a077bd0063c93f14b1
692 F20110113_AAAKGS lin_y_Page_060.txt
1de40436339459270e3642052f59cd61
57fff73282e33da6218500e277ca6429b40929d5
358 F20110113_AAAKHH lin_y_Page_081.txt
58396f34d422b288d5e46fcdaa41179a
4120013cb19043b4a543f151ac864734d8a7d133
1501 F20110113_AAAKGT lin_y_Page_061.txt
243c99e5893dc620f8649dd67b86f7e6ab49d5cc
827 F20110113_AAAKHI lin_y_Page_082.txt
63b2a9ffe7c37355eb3b249707581eb9
0a20183abaf62732fee2f4960a69afaecddba51b
2019 F20110113_AAAKGU lin_y_Page_063.txt
c27914e4bb5b56ef7cc5ea9c9b30641c
039598a2a1d3feb94e394f9cb6b006b2a56a4350
1495 F20110113_AAAKHJ lin_y_Page_084.txt
8acdf59a68b17b9435363e083cdfa6d1
1fabdff8446996d48a6f40d10556e4ddb3fd467a
1980 F20110113_AAAKGV lin_y_Page_065.txt
b638e38186732597c54fb5581347c811
ce718446f7607353f19e8f73a55db66bc2ca5365
1329 F20110113_AAAKHK lin_y_Page_085.txt
45ef9f0e64cef279397a3f255343dbdd
ba4a0889af5be0ffda415e5080e0276eb0ee9003
718 F20110113_AAAKGW lin_y_Page_066.txt
f87614a79e2e873e30d2992dee935de3
1191 F20110113_AAAKHL lin_y_Page_087.txt
04ec23565b20a43765c93e1f84eb4404
3f284a125d7b8ceaea44856e3dc5dd3c72a22b07
1938 F20110113_AAAKGX lin_y_Page_068.txt
88d48b7c8c015dd4b8829360b3ff4385
e12cdbfcec986d0f2869b95a2b5fb173a4c3d2e0
2031 F20110113_AAAKIA lin_y_Page_106.txt
c2fbd57a5d406e24cd44ac91e88ded3d
deef45303d0238953191d2df2e7b7f89df5689ba
1498 F20110113_AAAKHM lin_y_Page_088.txt
12d14f6e3089e9d66c8dbb2cb7ca1349
1922 F20110113_AAAKGY lin_y_Page_069.txt
3b77c4f4e0b6378ae375ea6f010cd198
87cf4315c9b42e4992bb99c6bf9e226e658540b4
733 F20110113_AAAKIB lin_y_Page_107.txt
3efcdd3749593291e7e56027912786c9
b3a254609bbabd710ff63a05ee6b380d4eb38410
1599 F20110113_AAAKHN lin_y_Page_089.txt
54a26cc3f728cdc908ce28bb9ed0ef82
1992 F20110113_AAAKGZ lin_y_Page_071.txt
464f4b899f08d6f90bbe21e1e4e463d7
0018b17e25db3ae1aa2770fc4a3a1e25d8d32a35
1821 F20110113_AAAKIC lin_y_Page_109.txt
3cc7ccf25b67cc74359a9cd6b9b85e85
6ee4e97ba9d6121949609769866520e089149a5d
1087 F20110113_AAAKHO lin_y_Page_090.txt
ea1fc75fe11da2b4e06f94a0d852dc87
c6be4a30e20f47d9fcd5109f4642bf5398b34261
2167 F20110113_AAAKID lin_y_Page_110.txt
8f28a013f3ffed756804ffa778412bffe8d2b745
1595 F20110113_AAAKHP lin_y_Page_094.txt
6794579c5dc8c389a1b5268dae66e41f89ea1c7a
1871 F20110113_AAAKIE lin_y_Page_113.txt
9d2124e6214be2be1cd840828736aca8
e6f1292b036fabff073956b4199232b73be418c5
1917 F20110113_AAAKHQ lin_y_Page_095.txt
835a3abfc114dcf22405ffa4d8728fd8
11ea70b2af8a61f7a8aea6f2b753e0a98b3365d6
1925 F20110113_AAAKIF lin_y_Page_114.txt
7f9f00546e42c344c0fe511b2d91de18
3405452884fe36962f302950a5bb31e33f35ea32
1758 F20110113_AAAKHR lin_y_Page_096.txt
471f2d3bc879463e623522fcf06aa210
f9e6b61e60c8891b71c17b5c3f343e877858de9d
2107 F20110113_AAAKHS lin_y_Page_097.txt
430cf162a18db91a19beca2c832f7834
6f1f1662aba191fff0af62eba6b930070c4a4129
1855 F20110113_AAAKIG lin_y_Page_115.txt
3f42b6a60ca151752152c5c08a415672
b92018ed6057b53dcbcbd7b2ed9d0d4f7a35477c
1380 F20110113_AAAKHT lin_y_Page_098.txt
d59f84f20b52d01140ddc7be9df6d898
a997796ae04cb8672557079b5b627a9f4b1f85e2
1967 F20110113_AAAKIH lin_y_Page_116.txt
9b23e305ca25ca932764f86d3893af8a
4e32f7a532fb69054afdb056422f3a974baa04db
F20110113_AAAKHU lin_y_Page_099.txt
80e25390df17f3392dd4885aecd5d15d
e74941cfca40acfb5cef4a6469b4938232799704
2028 F20110113_AAAKII lin_y_Page_117.txt
bcc112a544199c8bf8c685bf93f59f43
a90851c87d830413f472b0ffbc8062412e40b16c
2249 F20110113_AAAKHV lin_y_Page_100.txt
02b04481c43afb72e72f3b8450f46cde
F20110113_AAAKIJ lin_y_Page_119.txt
435c0261109b9bf24a89647c97882a5a
07b9e2bd616de1d1be0f0df8779192d58929a315
F20110113_AAAKHW lin_y_Page_101.txt
3c73c6acabbb6b2377c0b706739c0c53
3fe185e4fbc5e9cb62fd1140de83f40bc502ab0e
1794 F20110113_AAAKIK lin_y_Page_120.txt
03445b7dd6c4461f280c002e2c8df295
1be381e85fe62c78bd93c8b2476b2a6f4fd144fd
1893 F20110113_AAAKHX lin_y_Page_102.txt
1268ab47075103d461805d8718af927c
20f76437b2c28aa9f5ecfe90fbe6e0c8ffef0f7b
6769 F20110113_AAAKJA lin_y_Page_009thm.jpg
1e1419c6e8d4e00a30601cc07ce0807a
15cd9cc02d5549817a142593fa86aaa645184f28
697 F20110113_AAAKIL lin_y_Page_121.txt
04bb568bc9da328ffd2c5425791f5065
60b5b64793d610f9072aa7d39b07e68e8581b59e
1460 F20110113_AAAKHY lin_y_Page_103.txt
18b3104b210b14dde907940fb3f04970
27579 F20110113_AAAKJB lin_y_Page_009.QC.jpg
3f939e37e8e611601164dc1050303ca3
13a6c575fc4bfdc5c6a86a73a39fe6f6107fac8b
2626 F20110113_AAAKIM lin_y_Page_123.txt
de31874339a61cc32f64da09063e5fb25c128ee1
2989 F20110113_AAAKHZ lin_y_Page_105.txt
3b98ccd8095e1a218266b7e5f2c87f0d602f306d
6050 F20110113_AAAKJC lin_y_Page_010thm.jpg
4ec72f5ea07afc55e759962fbddc0013
2034880 F20110113_AAAKIN lin_y.pdf
851d52d48f666c6624be2a896bb3dac8
40d20cffe3f37bc576e1bd37b496e47a26603498
24674 F20110113_AAAKJD lin_y_Page_010.QC.jpg
9459eb546f5388c6441b4d1031329746
de95dd2865328314cc3ed61d150ca19b8a2cdd62
2346 F20110113_AAAKIO lin_y_Page_001thm.jpg
aa3336043a5df7568df409fb3937c802
3b7daf45a3290a07a2d9c9a2c88c11a0f5d1a635
26378 F20110113_AAAKJE lin_y_Page_011.QC.jpg
71d378a00807f354bee7b43cba0fd6e2
8fcf118c164686c5bf4bc807571537965ecc524c
8646 F20110113_AAAKIP lin_y_Page_001.QC.jpg
b37f2126b066a63c9a4f91496b0cb546
aa02bf7510fc7aaf77d0187f4fb4e70958e3c061
7131 F20110113_AAAKJF lin_y_Page_012thm.jpg
68bec5a4c9e622c1258bdd132745087d
25eb83b019e83c3fc900f3129c22e93ea56159b3
563 F20110113_AAAKIQ lin_y_Page_002thm.jpg
aff9887394b03d275d75426da883d703
3516e4cde7899d87b18d437a5ed84f2e6853bf81
27331 F20110113_AAAKJG lin_y_Page_012.QC.jpg
2959381a6958621b30a2c66076c93c67
2e18b97acb738bb4ec28325c8a24f934f34d5fa3
569 F20110113_AAAKIR lin_y_Page_003thm.jpg
250daf5d5571ff1fdb7f1e4d45b2e0a6
6869bfd5dc2a9eb1b4d7506dbe98d6d75a8f63af
4275 F20110113_AAAKIS lin_y_Page_004thm.jpg
7550 F20110113_AAAKJH lin_y_Page_013thm.jpg
65085bf81e8e5012e2654a916be2777e
ef000af77eab4dbd2d289663ef62b6c90d61acaa
16845 F20110113_AAAKIT lin_y_Page_004.QC.jpg
bf1f9837faa8a768d8ba76515b7622de
520908604277dd09b26091ce2b56816fa4a34d4e
29763 F20110113_AAAKJI lin_y_Page_013.QC.jpg
432fca2ceba2ce268a9fd84f45e72d9c
fa40c6321ac40b24ea898990214558f437b90a87
5849 F20110113_AAAKIU lin_y_Page_005thm.jpg
3f64cecf3f4b7a37b11fa239aabcab38
e90784b3f996c0acae054aea6a0e645c1e67e6d3
8486 F20110113_AAAKJJ lin_y_Page_014thm.jpg
859c09c241d2a6c46488fe4ed9263ec8
2e59ba4296a319dfbd31253eb6defe06037dd234
30771 F20110113_AAAKIV lin_y_Page_006.QC.jpg
715e88226097b1ec300bcef32341aa02
f6abc5c7ce25c4e0515e0c6fa7e2bcbf09d147f7
32739 F20110113_AAAKJK lin_y_Page_014.QC.jpg
e188221b55ea4399d37ba625d2b6b1a4
1f5244d7beaf7d072676ac3b33cab2064c19b084
1562 F20110113_AAAKIW lin_y_Page_007thm.jpg
4b80a7fd06c472122a0526762f3214ae
319f7eb73e17fc6da92cd2e8f8e3b14637a602f9
29490 F20110113_AAAKKA lin_y_Page_025.QC.jpg
2f03b81f5c09da1e1370f02849b7f7be
d8a169951ea1e086ce8297e1f620922ce74c4248
8484 F20110113_AAAKJL lin_y_Page_015thm.jpg
e1e2ba0ec33164a81f39ba3b91401546
00cc6494427e4a59998d04f26b43de309a960656
5079 F20110113_AAAKIX lin_y_Page_007.QC.jpg
044c46d71d88f27db772506df23aef02
6a40d1b909135297318b2579c659d4056849fc9c
35871 F20110113_AAAKKB lin_y_Page_026.QC.jpg
2bb93733d202801b08e175166b292c17
e7e5a4b484280046c54451e781028da2aa7d2cd8
33532 F20110113_AAAKJM lin_y_Page_015.QC.jpg
b5593ced28ede11ee2903a624c32f918
0b479b2f65781a14e1169c7f91e191e3d04b4a6b
3967 F20110113_AAAKIY lin_y_Page_008thm.jpg
f5369d9b06bea78d9f6b51373d623844
7294 F20110113_AAAKKC lin_y_Page_027thm.jpg
5110d4b5c86e85b69e1939ac3990eb5a
7c84ebb4e592b0ecaab32b3c30fa35325584a420
8678 F20110113_AAAKJN lin_y_Page_016thm.jpg
32dbd02af283df6d97812ee9740e0680
20082ebb22d76d57306147b0e4c93cf0327b0d57
15302 F20110113_AAAKIZ lin_y_Page_008.QC.jpg
3d338424bd2c05640159482495079594
79717ed5ded224a6ea3d10b4f4ed1d7469502e99
26032 F20110113_AAAKKD lin_y_Page_027.QC.jpg
a6f0f71e6188b76a2dcab88c49b06976
b590cc2475f7a07e89cb7387edd33c67ebc3b51c
9076 F20110113_AAAKJO lin_y_Page_017.QC.jpg
bf1f591a86a41d26d7f2cd2086da0f3bc1ff4f6c
8395 F20110113_AAAKKE lin_y_Page_028thm.jpg
991022ac2a93ab41bfa26f48b2ffb742
30446 F20110113_AAAKJP lin_y_Page_018.QC.jpg
2bc6e807e91ed55911e2320952e7ff17c6d20fe6
8739 F20110113_AAAKJQ lin_y_Page_019thm.jpg
eee1863f030fb3ce7bdf44a6256a469b
8334 F20110113_AAAKKF lin_y_Page_029thm.jpg
85ca5086baa393586e5a31321d08b61c
8faef69e11955cd3d239e11f941d63806f0d1ab1
33381 F20110113_AAAKJR lin_y_Page_019.QC.jpg
455f72bb9868713a869cd48d00e21f03
b64d92f8de3e3e31e4af358213d2a9e4e3db49a4
31960 F20110113_AAAKKG lin_y_Page_029.QC.jpg
2f9e98a34bc689e90514c2952ecaea02
95cde272137026a489de01bcc749413330f8e9bf
8466 F20110113_AAAKJS lin_y_Page_020thm.jpg
7763b78e0fc8605267fb68682d9e65e5
35207 F20110113_AAAKKH lin_y_Page_030.QC.jpg
895b9ca30959fb5bfafa0e021707be9a
31647 F20110113_AAAKJT lin_y_Page_020.QC.jpg
51d92376b2df900a0c7c5c8384bbe835
8ff6e8b986dda0cfc9be531b65510a9f4fdbc25b
8714 F20110113_AAAKJU lin_y_Page_021thm.jpg
7203fec7b74ba3701110fa24c21422fb
b947f1f7422dfbdcb10a3ef3d5c22d8550366711
32840 F20110113_AAAKKI lin_y_Page_031.QC.jpg
24138b64e0c10e34b488a23359b08758
f0f75fe3416d4252d96582edfb71376b14fb2819
33588 F20110113_AAAKJV lin_y_Page_021.QC.jpg
797b60d59577877d941ab1d3dbbe40aa
b8b52c7f7f16c828005cb9ee936a05bf75960a10
33240 F20110113_AAAKKJ lin_y_Page_032.QC.jpg
afef1fec08d374636c756326f762e0ac
ab33280d0c6ee3f93b1214a08b961fce407d84ee
8008 F20110113_AAAKJW lin_y_Page_022thm.jpg
89820185eab4f7ca860ece01168babaf
185b4fbff71bb1fc3ef230e555e355a11489365f
7799 F20110113_AAAKKK lin_y_Page_033thm.jpg
12e3e701c060c869e2424fc40f85e227f7507a53
7833 F20110113_AAAKJX lin_y_Page_023thm.jpg
90e5b8649f80646671ba626751b48049
6ea5ddae000364110b515823b281703f872f37d0
8221 F20110113_AAAKLA lin_y_Page_044thm.jpg
038f617007ee58a06803a1fd7f23a0aa
f9e71f3092af91e2f58667674fed774102b46771
31175 F20110113_AAAKKL lin_y_Page_033.QC.jpg
756ebf726ee708b6b37bcebfb3542034
8d2200aa4de5f70ae7e87751e9a5845c81c23e5e
29125 F20110113_AAAKJY lin_y_Page_023.QC.jpg
b0ab8f1d8a4e4a098122e896ae68dfb1
8234 F20110113_AAAKLB lin_y_Page_045thm.jpg
2f9e2144db0c11488223b673f98ff4da
b32f7f4fa45c965fa52345b12409ff6bcfd88351
8702 F20110113_AAAKKM lin_y_Page_034thm.jpg
6c834229540646f298894e330988df48
48c9f6c0f19e982ff4982fe130c120e604dcf1cb
8533 F20110113_AAAKJZ lin_y_Page_024thm.jpg
d7aaeeb0985ae81d90933faa2bec0da3f7d5a3c5
6795 F20110113_AAAKLC lin_y_Page_046thm.jpg
e78ae77fdfa4a8a85c33a6030973e4c1
5439712cc6a7b75f2e043d1968764a4050dc0512
33969 F20110113_AAAKKN lin_y_Page_034.QC.jpg
2f8ff797a8be32562bda50964605c0c3
2235ffb07254a6b7348bb4ce1fa322285fab6381
26224 F20110113_AAAKLD lin_y_Page_046.QC.jpg
9e3af928b98df03a8157e6403ef24f1b
0f189a195dd31c7b6f71d3f680ed3e8c8e5b549c
33659 F20110113_AAAKKO lin_y_Page_036.QC.jpg
68a240ced73b006da22a8598417d1823c66d979e
26582 F20110113_AAAKLE lin_y_Page_047.QC.jpg
01d738c2220ca8daf23f4bfa2d3f527f
19ed15f360e757a119813ee7bc4e6fdb1a840f7d
8262 F20110113_AAAKKP lin_y_Page_037thm.jpg
1589bbdcf4c7d52913cd3dbfbaa0864b
93f857725701cd08f672f7e6d90b42dbb31fbd3c
8407 F20110113_AAAKLF lin_y_Page_048thm.jpg
877b474d470e9331f93af2a55d0112c9
5378a11b2d2b3292a54fee64c2980902ae4d9797
6094 F20110113_AAAKKQ lin_y_Page_038thm.jpg
2b3549b12bd563e516df978a19da5992
ed441cee31918684984b73b1768a86a7b712b74d
8322 F20110113_AAAKLG lin_y_Page_049thm.jpg
b404940af851118cfebc4132c7e5f05f
2609f7b7f72fd8c96a714dd0d9f0532066a7b0a7
19917 F20110113_AAAKKR lin_y_Page_038.QC.jpg
44c75a0f748ef010cea826d835ef291a
31777 F20110113_AAAKLH lin_y_Page_049.QC.jpg
9b171ea65b114852e52fd28feaa741d8
15f1ae7478ac4bb567b9a3408f0d3cb6b1edafd2
7843 F20110113_AAAKKS lin_y_Page_039thm.jpg
601edc4844ae68f749071f026eda701de50245a0
8801 F20110113_AAAJIG lin_y_Page_063thm.jpg
088bba166793a6bfa989793d75de672a
4ff273c6db9d4ac17795091787a938805f16723c
6664 F20110113_AAAKLI lin_y_Page_050thm.jpg
9e0f077ae153ed1bd60d9972505af5ec
6349b4ff5d8213ca467dda3ae1d0ccd6e53afa64
29323 F20110113_AAAKKT lin_y_Page_039.QC.jpg
1f990673a43601c500a0b554be3c37d4
3de325d6a4d15b37968a58f6c624ff4f96467715
8555 F20110113_AAAKKU lin_y_Page_040thm.jpg
06f0ec8a4a698a28852f7e19078db892
86f461a4fbec274b8134468627dc60bf19d71710
47698 F20110113_AAAJIH lin_y_Page_069.pro
27ace8e3a34376f04266b648930de195
5b96affe49f98dbaf6da6d597d3f58195306b682
25492 F20110113_AAAKLJ lin_y_Page_050.QC.jpg
4eed20ebb3d14b34f970970f6cbecf8c
6b6e589b4c99951d5b19f4a87e4d29734a2aee3c
7805 F20110113_AAAKKV lin_y_Page_041thm.jpg
7a38ff953ee8bbf8773af665403398fb
2d74ba677367c671470c4d4d104f0dbea43cbf51
1949 F20110113_AAAJII lin_y_Page_112.txt
79b4585ff03532d6ab3334c60bac2d00
06577e3621189a88e014edf6b51fb9f5f9a79198
24438 F20110113_AAAKLK lin_y_Page_051.QC.jpg
f83633ed3ec833e2ef9d4f216746855e
8ae71725b0b12f1a15ff5c36f45248d84d4a2b22
8868 F20110113_AAAKKW lin_y_Page_042thm.jpg
90494546bac3cff45490ef846fb6671b
b553d0d78ed6c882a476e311b952724ca5786091
7369 F20110113_AAAKMA lin_y_Page_062thm.jpg
0668705cb539c1d6b8b568c06d9b01a2
656ca19bccc01491c6559ef4ae6bd5fe0638e826
1984 F20110113_AAAJIJ lin_y_Page_070.txt
ef6010f11c04fddd9d44a3e15c0bc157
6091 F20110113_AAAKLL lin_y_Page_052thm.jpg
50a317cb1e406aa5af89c1480703c652
8eb70b877b4f6aecfa6569f258a43966ca2e8385
34838 F20110113_AAAKKX lin_y_Page_042.QC.jpg
7b155cc7ab00a00d8832e713224f9383
d7721d946deba1c0d4e8a9ac040690c27eed5f31
26356 F20110113_AAAKMB lin_y_Page_062.QC.jpg
d3d44f689c58e292f36af4a477437e45
8019133d2d93a895fe6e986dcdc0419600e3f9a9
32642 F20110113_AAAJIK lin_y_Page_115.QC.jpg
74eb30c41aa66ee29730fa5e55d116aa
21073 F20110113_AAAKLM lin_y_Page_052.QC.jpg
9e2207c18a0d3b55ae0f35dab9f1695a
a28a9de60f697bbd9d18230204c7b411fe5378f9
7955 F20110113_AAAKKY lin_y_Page_043thm.jpg
f1caba391b21d706a76cb3a05df1e405
46a871a1386996ff9a8ee8c38caea9220cbcd401
26109 F20110113_AAAJJA lin_y_Page_106.QC.jpg
85c60134f1e198f80b55ec908ae2b994
6163392ed7e126881f435ca9ba72ca783567ec6c
34949 F20110113_AAAKMC lin_y_Page_063.QC.jpg
875f97d79c56d927c4ee4643f49be835
F20110113_AAAJIL lin_y_Page_108.tif
3c45fcce42733c9dcbf409e66bdb30a6
634ac47ea736a90308cf3fac7d9368abb15f7099
28767 F20110113_AAAKLN lin_y_Page_053.QC.jpg
6669e3ff26b3001448a13278b6049f57
12418fa6ee463cd7652a49e321db454b8367c781
30705 F20110113_AAAKKZ lin_y_Page_043.QC.jpg
47ecc7266b9d6f6207747fe7ee6a49dc
b7165bfd6567761f4506eba6814d972aaa422ec8
1971 F20110113_AAAJJB lin_y_Page_030.txt
1a0376356d240e59529758fdd64090d7
1f56a6d150383059f632d5003e58786fd6465df8
8210 F20110113_AAAKMD lin_y_Page_064thm.jpg
1898 F20110113_AAAJIM lin_y_Page_031.txt
3f4f3e8eb1465fb1ddbe3c7e8ba7bbcb
c590fe2ded5001cec8fd038c79a96b8a8f933fd5
8890 F20110113_AAAKLO lin_y_Page_054thm.jpg
d72374c2cc74b4ff9bbd242b6704a155
3aaee8c2faabca184762126bdf86b41b4f92a10a
8506 F20110113_AAAJJC lin_y_Page_036thm.jpg
22034ce5f395eca4ef52bde3f446439a
2909cf8aefe6bbeddc3e7fe3a10da36a91f9b2a7
32718 F20110113_AAAKME lin_y_Page_064.QC.jpg
37c11da0fbe8e5ef791027eaf9ac85c1
ae4889c96ed981a7a0fca18cb5c19080616e1a6a
47922 F20110113_AAAJIN lin_y_Page_015.pro
0dd0d3bf0b9e0118af8f7d3c455cf43e
a9789f11cd9c2d1e3886e2c1da8724987d458cb6
32897 F20110113_AAAKLP lin_y_Page_054.QC.jpg
1661 F20110113_AAAJJD lin_y_Page_046.txt
3599c80f65427d5f022bd217f9b7f295
a1774abd6267f9503735a50a37deb3cebb3bff0d
9110 F20110113_AAAKMF lin_y_Page_065thm.jpg
34284742e7ab5f789f37dbdfd0c00a54
32334d64a6843670cc405cf955aa408f42bacff9
105333 F20110113_AAAJIO lin_y_Page_122.jp2
594cf86601e0b00be9bfc80c875e3dce
366a589e5860f0771906a289c508acbf7df13e6a
7744 F20110113_AAAKLQ lin_y_Page_055thm.jpg
4e4731a80155980a3e487b0f82d63acb
d57922ca8a34d1b616060f98e082f097379b99a8
F20110113_AAAJJE lin_y_Page_098.pro
c56c5c50881591a03d9b2bc27b3d2b99
3a32f007543df8b255b42e1ba0391caf3d6eaaa1
34705 F20110113_AAAKMG lin_y_Page_065.QC.jpg
0935b2ea66cd77fc08ceb122ff876444
492334631df50429f35783ef9a7c59acec3e5349
95158 F20110113_AAAJIP lin_y_Page_022.jpg
bd80db904e474fa37a428e7735ab1b39
dae45cb93b72f17c149a5e82b82f7a220413b26a
F20110113_AAAKLR lin_y_Page_056thm.jpg
13110 F20110113_AAAJJF lin_y_Page_062.pro
dbe945a1b1ed86913dc3457487415319
6f2b8c83c49d408001cf7199d6c771cdb6c7178b
3593 F20110113_AAAKMH lin_y_Page_066thm.jpg
c85f0a17a008a6670c00efe9652fd818
7234738638b510fa527f55cf9fd38dc45bb877c5
37183 F20110113_AAAJIQ lin_y_Page_106.pro
4603fed7f572c97225b10039aed767f8
4a803c7151314bdfd06ffda609e61c8cf7cb0ea8
33278 F20110113_AAAKLS lin_y_Page_056.QC.jpg
d167de9b8a663da0c554d5e7a8364d81
7468d80aa84cdcd68e63a301c4bc367519e3aa86
97655 F20110113_AAAJJG lin_y_Page_097.jpg
472382884552cf7589a73eebf139a0d2
76f2e00e7487b7c44e20c295363c10b90dbc4cf1
13332 F20110113_AAAKMI lin_y_Page_066.QC.jpg
f98bc9e438be4498a8ce4b6c99f4959c
b3931d5848b9492f1a397d743e51d292c3d18e4c
F20110113_AAAJIR lin_y_Page_069.tif
f82355c1cb56554d058c8d136c712c4a
1677e944918072883e057c46485aa7cbf296a780
6861 F20110113_AAAKLT lin_y_Page_057thm.jpg
ba2bfc042cf10916941d105b5ae1715b66c4e1ed
81810 F20110113_AAAJJH lin_y_Page_073.jp2
003e698430f2b56b859110c3cdc7eb2f
587e567e37d141c584e30e5cb3d89e0f4ae33416
7524 F20110113_AAAKMJ lin_y_Page_067thm.jpg
ccc42f03c9f89bed9d9ebfef56a98381
666 F20110113_AAAJIS lin_y_Page_125.txt
6119072488824150c15612e857d4a519
08d6199696c7fe10288248596724fe926abaa798
26052 F20110113_AAAKLU lin_y_Page_057.QC.jpg
4b1bcb62ce47b27f02b667421449ef47
1157433bb5103ca19e64e3521c68781d99992c3a
78097 F20110113_AAAJIT lin_y_Page_106.jpg
eb22f369ef5ea40cefe7ebcc26fce2e2
9fe566f5a1153e00a2fa3a2dc5764721575c8888
7544 F20110113_AAAKLV lin_y_Page_059thm.jpg
de381176effe87b67e3a9068fc9c3f6b
1910 F20110113_AAAJJI lin_y_Page_058.txt
9e031ccb4659d1753802d373cdf55299
29346 F20110113_AAAKMK lin_y_Page_067.QC.jpg
4432454f387c62049647dbcf2eef9b9d
ff2cccb39041f78906f8b99098823cd7f253fa58
35732 F20110113_AAAJIU lin_y_Page_081.jpg
456e9ba0ed356985014081f57c310ddc
4c2f526129f9d2b4d37ab450665566f9688a23be
28392 F20110113_AAAKLW lin_y_Page_059.QC.jpg
3af27e97e53924ff968d7c37c3b434f5
0a4cc33cce1cc7dcb3b20e4e2b52d07e2514a25b
6955 F20110113_AAAKNA lin_y_Page_078thm.jpg
532f146cdc39c6c476616968da0d3d79301631d8
76598 F20110113_AAAJJJ lin_y_Page_027.jpg
3c2028fc84bcc911d888f233bafe0cb3
8792 F20110113_AAAKML lin_y_Page_068thm.jpg
0f42c96552a28011d809a1fcdbb9d881
d16c44a1f14af8d492d99828cb7a1f75e2dd9a3d
1561 F20110113_AAAJIV lin_y_Page_092.txt
2529f9e156382983319eaa6f164b37c3
7e35331f6eef15f02eac17c83932e274b97e9a09
8768 F20110113_AAAKLX lin_y_Page_060thm.jpg
15f0485fb84ac2a72cfb48e587ff5971
f91be602d6d6bcc852f187742e256178e4947e04
24175 F20110113_AAAKNB lin_y_Page_078.QC.jpg
e6a91aae2ea5b867cd9b4039b41b8185
9987ed4aa2082037a802412f3dea6eefdbe5bc47
34615 F20110113_AAAKMM lin_y_Page_068.QC.jpg
95f5e2965a66b5ff12a139f2b28fb80d
9625c81d2ab7e479123d0e51db722538d6cb4b9a
1873 F20110113_AAAJJK lin_y_Page_043.txt
7620fdaca68a38f76b7713011acb55ab17636888
17845 F20110113_AAAJIW lin_y_Page_066.pro
a76b9aa0c89c080fba496e49b8118bd8
31650 F20110113_AAAKLY lin_y_Page_060.QC.jpg
2807c94241c4b81a895773097ee18afa
5740 F20110113_AAAKNC lin_y_Page_079thm.jpg
68eea2692bbf098665c44fb3532ac2da
b59a18755ea7b0ef70ed6ecf7858236da8972e13
F20110113_AAAKMN lin_y_Page_069thm.jpg
e430875d1d5fc905ab3fe5e630f783a9
33510 F20110113_AAAJJL lin_y_Page_040.QC.jpg
e7278a5d8a1415ac38c39dc888f75ffa
dda3b1da444d484741577f22e2f67f8a5ee7bde8
104945 F20110113_AAAJIX lin_y_Page_116.jpg
5b72cb361095a49c20f8cd110dc3c1df
32683 F20110113_AAAKLZ lin_y_Page_061.QC.jpg
f9fb2c3c54f53abcae54370053401fb9
f499a69781872e6cd0d5f2e57cff9cab93f01514
2036 F20110113_AAAJKA lin_y_Page_122.txt
4dd6e7fa102b5e60bf77a6173247f344
6c30815a3aea1db5e051064344d3e77762eb9b4a
19280 F20110113_AAAKND lin_y_Page_079.QC.jpg
ffe4267b8653b5f6daabb862462e9c9f
45a378d7c1f6474d76e3c20ae6d6c391fdf059d1
32244 F20110113_AAAKMO lin_y_Page_069.QC.jpg
38ea3a26b1e2c3b5636a83ca3f1231c3
788f043f01725b341469cba16b13727f294cf0ef
47234 F20110113_AAAJJM lin_y_Page_013.pro
6aa082dfd75fb41c90b6cfdcec697835
5900292d5535940ee025d3eff5a6115aea9b8ee6
1376 F20110113_AAAJIY lin_y_Page_104.txt
4757370533d804c3f420ee58eb5707f6
3e1bbfaa82f0cebe1894a909144a203ac973b7c6
88414 F20110113_AAAJKB lin_y_Page_010.jpg
22406aae6472778730b7fd891b13edda
ee2b64d725f85fd48e6b07e295fe8c1ddaa5acd7
6517 F20110113_AAAKNE lin_y_Page_080thm.jpg
c1db195a5a3e53f5833a95d8bb920a2a
0ffb46b103f2e76aa01c8c9cfa4ce0214184702f
8687 F20110113_AAAKMP lin_y_Page_070thm.jpg
52bf8ca1071ecba3ebae1e069057b47f
629b34d3df2013e4766b3b41a813579b03529577
1051976 F20110113_AAAJJN lin_y_Page_009.jp2
f3ee2788491e653d1c52c7c6c4481033
105644 F20110113_AAAJIZ lin_y_Page_071.jpg
4e06733e424c6c9a5fdaac371949c5bf
dc5ff5895c2183e17ebbec5d335e55da1e7139f6
690 F20110113_AAAJKC lin_y_Page_062.txt
6bfd9920919e84796d38c9aec042a3c7
25689 F20110113_AAAKNF lin_y_Page_080.QC.jpg
7343edeeac564c05515d290f72eed7ff
4b320099cc3545e8244681be3feeb2a02fe0e59f
34665 F20110113_AAAKMQ lin_y_Page_070.QC.jpg
ba9a6c26571d946441803dd8b334f425
F20110113_AAAJJO lin_y_Page_080.tif
215e0f153d07b8b044fa78011fe5e4c1
4237 F20110113_AAAJKD lin_y_Page_002.jpg
2eae102cb1d2fa82c9023842803d2516
3221 F20110113_AAAKNG lin_y_Page_081thm.jpg
fdb6061a04934e655f8d67211530a032
bbb50b63f8aa1424de2efaabd63d928053866018
8658 F20110113_AAAKMR lin_y_Page_071thm.jpg
7e4ba7a137cefcda8488d476900a3f10
e5cf3ff42f2c5470095d6dfa1c5f2832ce05eef1
6934 F20110113_AAAJJP lin_y_Page_106thm.jpg
cbf87337b9686b020f5c9f72e5aaa269
8a9ea9a2bf462d4f2c4be0145e02c502b5976eb8
111626 F20110113_AAAJKE lin_y_Page_116.jp2
88dd33d437843ba6de111153cdf01d10eeb536ef
12079 F20110113_AAAKNH lin_y_Page_081.QC.jpg
5505e55a4a5b4007d534ef42354d1dd8
9036beced719541c338a8212d66c8eb063b162e9
7069 F20110113_AAAKMS lin_y_Page_072thm.jpg
95525eaba2b7131978f47c72dbb76d4c
d72349473aa980c56048c5fd824d6720d79a5580
47055 F20110113_AAAJJQ lin_y_Page_037.pro
d82975c056148da452437030b24432c5
7241fc1d00451d8d023aab64bcf61f92039cc3e4
F20110113_AAAJKF lin_y_Page_036.tif
8798cee809bb6cc2530b797a0c2760cd
85a027ca17cc86f84589718a97d6febe86f72f79
4371 F20110113_AAAKNI lin_y_Page_082thm.jpg
c3874f7472e30803432d29bd055bd755
ab1145e652c0ceb1fd3d6efd76786c8398ee1b1d
26617 F20110113_AAAKMT lin_y_Page_072.QC.jpg
d9337e73da8186f9ddccd34787cf550b
e2ec91b8b02483fcc1b2efa45e1eac12cdb99c35
8545 F20110113_AAAJJR lin_y_Page_026thm.jpg
68b26466d166ef44fb0185b5c18f1403
674ae3bbf9215bdc23690ed15c9a41ae439370cb
F20110113_AAAJKG lin_y_Page_093.tif
c74381fb540121c834564c17730d7c2b
ae6164ecc259d57def95628e7a05fcf633bce418
16050 F20110113_AAAKNJ lin_y_Page_082.QC.jpg
b9ee833f588b526ea2bb938c49eb204c
6a9049c7fd649460f70586c1694695798e72e80f
25037 F20110113_AAAKMU lin_y_Page_073.QC.jpg
77487d25c8f22615b1a6cd2d4b5b4f2a
8918c59f02ac4e3dd76f792a80cc0d0f9ccba06b
1731 F20110113_AAAJKH lin_y_Page_067.txt
856fd6821b50b0a311cd2a64a441da1a
b4f171cc6821056d5bc75e614cdd242990bb224e
33112 F20110113_AAAJJS lin_y_Page_058.QC.jpg
5d746632be43a1d6082b4a125d0835dd
6fd22cf2e33570875d1a089298e73a2ccfd3f18b
29682 F20110113_AAAKNK lin_y_Page_083.QC.jpg
33faa9272fbc816001fc2b3d41cb87a5
6362b0db07c7dec045a8b9591e537eb184b7e3ca
28991 F20110113_AAAKMV lin_y_Page_074.QC.jpg
36dbc1088862f32e70d518d90ac05824
a42f214dcb368a020462d2139c4a658042ed1019
47975 F20110113_AAAJKI lin_y_Page_058.pro
729b01c2dd3e538fef159acb89740d24
9048cd639d0725b98ce23f57a1eb7b56a9b16ea7
90969 F20110113_AAAJJT lin_y_Page_095.jp2
76d83b10165017ca2ea21e82768aa7d9
8032 F20110113_AAAKMW lin_y_Page_075thm.jpg
56995daa1dbe4efe1ea671aa0813f9516d927026
33140 F20110113_AAAJJU lin_y_Page_094.pro
b5f0cc04c48bc588f6e7fa4b5eb1f0fd
c9b7e5eb5684550d7771b83a9974956ddb10b68f
7747 F20110113_AAAKOA lin_y_Page_093thm.jpg
14c446197449ac88f075fc2537decede76ea1292
6004 F20110113_AAAKNL lin_y_Page_084thm.jpg
28654 F20110113_AAAKMX lin_y_Page_075.QC.jpg
e397a6aeecee5909f2a53c3deddea583
b01034eaa90dc7c0046f93f394093b11fd958e56
97214 F20110113_AAAJKJ lin_y_Page_119.jp2
eab80fd5bb0f3c77351b3cc507c8d1ea
a571260be65c6c85e188471e89b3849c866e994e
100798 F20110113_AAAJJV lin_y_Page_069.jpg
e9ef093814a168a8aed75f3d2717a47f
4d612c282fd81f09fccaf2716c25360e421d3566
26591 F20110113_AAAKOB lin_y_Page_093.QC.jpg
886767ea8064f54534d2fc3c3446e2d0
d6b11461a6e0e9ac7b5f795de9620403de624967
22390 F20110113_AAAKNM lin_y_Page_084.QC.jpg
3bf9c00bc7a0b0d5aabb88a5bbf19799
5473687ef0aa98d7470715be40ec1d3d960483e7
31579 F20110113_AAAKMY lin_y_Page_076.QC.jpg
ab879662f8afe54f15c66cc44c38b2d9
dc3258f8078956c2c41e690eaa8a88bbe0e2691a
102655 F20110113_AAAJKK lin_y_Page_024.jpg
291ac8b44715f538a65cce9e9d96c7f9
d93dd0be4469a1bff35daa38e7762f9822f2631c
108197 F20110113_AAAJJW lin_y_Page_016.jp2
6c5dc09e9ec29f9f7fef0491b7647033
21dd43c480e5ec1eda832a561ed8f6a42b0639a6
6852 F20110113_AAAKOC lin_y_Page_094thm.jpg
15762c7d8d8fa1bda50e2022a3c9938d
690c700647554abd0f921bff150390f9e9f310f3
7295 F20110113_AAAKNN lin_y_Page_085thm.jpg
e5e80238a862a3917c02d9e3782c6337
212a1d095e08f79495403dc87bba9c23c3995274
29889 F20110113_AAAKMZ lin_y_Page_077.QC.jpg
cdfb62e8c7d472b95e6dac16f9d45dc2
f3660ea9c0725aa866a73957c030876f0f29ccd8
58414 F20110113_AAAJLA lin_y_Page_088.jp2
386415286bd503d9680296ab1ebbc4bd
0f24439809b9cc98b25f20113e288dae641a6a3b
104500 F20110113_AAAJKL lin_y_Page_070.jpg
212110fd74a55b5797a1906956a272b8
32575 F20110113_AAAJJX lin_y_Page_124.QC.jpg
dccd36160eb49162b0e8ab09b742389f
9ae91d54da75f20517c5d723efeb528a20d0aace
24426 F20110113_AAAKOD lin_y_Page_094.QC.jpg
c58d582e1e6cbef6583e4c549bf42724
4b4a1838cb40f5550aafce9cd0a69b2f02d85707
7277 F20110113_AAAKNO lin_y_Page_086thm.jpg
6c96782290ec0890dc9a57ee1e3aab58
9262d99b081d2ab5924c4f9b37f429b7c7b878a7
2376 F20110113_AAAJLB lin_y_Page_017thm.jpg
27c6e9979e14137d9fae332b8ee2820e
7d4fee649a46cf62f2b1a8f6ed99a771f6e04b1f
1085 F20110113_AAAJKM lin_y_Page_086.txt
8c8f22da275d198d33edf4dba65bb433
abef2758333ae3ef81d7998e206e51e06db04988
7883 F20110113_AAAJJY lin_y_Page_102thm.jpg
92e91176e3607047d05a1b671088594c
7815 F20110113_AAAKOE lin_y_Page_095thm.jpg
1c1da8eac137f4eec30398a703b3dcb3
a01a3f7aa876cbd3b36de062e816fe76912fee1c
28627 F20110113_AAAKNP lin_y_Page_086.QC.jpg
99447cc4497152c362a8b64f543262af
01a7a528d20822aa97b884975da0720a69486858
F20110113_AAAJLC lin_y_Page_037.tif
66103974db16b44639c980baa80034d8f6e774f6
F20110113_AAAJKN lin_y_Page_031.tif
06b1337e3f3c32750d504e185cf187cb982f19b9
F20110113_AAAJJZ lin_y_Page_033.tif
c10db9c8788bd149100d311f02e0b33d24c0e6fb
29058 F20110113_AAAKOF lin_y_Page_095.QC.jpg
3947e61b5fe543f195eb2a3656385695
3a28ff847c521eaf5e59a0cf41ffd2c9e492c2de
5387 F20110113_AAAKNQ lin_y_Page_088thm.jpg
c88312c4d2f8d06883908c209acf9c02
23154 F20110113_AAAJLD lin_y_Page_005.QC.jpg
2f632df78674a2a4b92329cf32db37e8
F20110113_AAAJKO lin_y_Page_076.tif
30eb714e2485cca4397647837a874338
6a3560a3beb397b28c5ec651eae21e007f76c4f8
7166 F20110113_AAAKOG lin_y_Page_096thm.jpg
8e065691de73bf2c2216480399fc5061
19187 F20110113_AAAKNR lin_y_Page_088.QC.jpg
0fec456e18a831c2ea30df70257b33cf
523c6d3c0ecf7aed01101a443061cc74ac5062cd
1436 F20110113_AAAJLE lin_y_Page_079.txt
c876ae98abeaf035ef68201fdcfd19c97e6cf844
49827 F20110113_AAAJKP lin_y_Page_099.pro
d60c1cefaf1067338e60d78778fe180b
8182 F20110113_AAAKOH lin_y_Page_097thm.jpg
a9e54bf33556231475bf50570f789fef
7d2768da271d015f36355f6bdc0663d737d899b3
5407 F20110113_AAAKNS lin_y_Page_089thm.jpg
60cfaf31a7f3f097fb9e5f223e7b9f61
7eba8fef0b881992333996463a87212c70b303db
F20110113_AAAJLF lin_y_Page_059.tif
38ccb5382dec99f2a725756e4e5d46de
c91ef2110fef06f91d68537d33558d57c4736005
47126 F20110113_AAAJKQ lin_y_Page_101.pro
45c3fda52ece520468d411faf05011f7
3c4029c345a42d37845c78ba876ccfe98b61c6df
31928 F20110113_AAAKOI lin_y_Page_097.QC.jpg
d8ac599d9ca16b0d93bae83073a9e54c
d0b5c7dfa06699438fa309a70ce4e16978a8eefe
19395 F20110113_AAAKNT lin_y_Page_089.QC.jpg
969441475af87cf47ce1bf55a51e7ee0
485d76fc52ca869e01f23c47a430c087f40417c4
61395 F20110113_AAAJLG lin_y_Page_079.jp2
4f00d3ce81bb2cd5f311a8337981695c
4329ca5f63b3d8e54a46035188664da8ac7efb55
F20110113_AAAJKR lin_y_Page_053.txt
04b92c00e92a2bca2a910bec9d3e6ea5
220ac5cd936a710bcdbd057d3544d5d3e4b7dfe5

PAGE 1

HIGH-PERFORMANCE PERI ODIC CONTENTION-FREE MULTIPLE-ACCESS PROTOCOL FOR BROADBAND MULTIMEDIA POWERLINE COMMUNICATIONS By YU-JU LIN 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 2004

PAGE 2

PAGE 3

This is dedicated to Parents and to Yen-Li Sun

PAGE 4

ACKNOWLEDGMENTS I thank the Department of Electrical and Computer Engineering for having the best teachers. I would also like to thank LIST lab members for their support and feedback. It was a pleasure to work with them. I also thank Intellon Corp. (Ocala, FL) forgiving me opportunities to work with them and learn about powerline communication from the inside. I especially thank Srinivas Karta for contributing useful discussions and knowledge to my research. I am especially grateful to my parents who raisedme and are longtime supporters. Their full support and love were my stronghold in doing this research. Finally I express my deepest gratitude to my wife, Yen-Li Sun, who has been a source of strength and inspiration over the years. I especially thank her for her support and understandings during many late nights while researching toward this dissertation. iv

PAGE 5

TABLE OF CONTENTS page ACKNOWLEDGMENTS .................................................................................................iv LIST OF TABLES ...........................................................................................................viii LIST OF FIGURES ...........................................................................................................ix ABSTRACT .......................................................................................................................xi CHAPTER 1 INTRODUCTION........................................................................................................1 2 POWER LINE COMMUNICATION NETWORK INFRASTRUCTURE FOR HOME NETWORKS...................................................................................................6 Home-Network Infrastructures.....................................................................................6 Applications over Power Line Communication.........................................................11 Power Line Communication Application in a Home..........................................11 Internet Bridging.................................................................................................14 PLC Design Issues......................................................................................................16 Physical Limitations............................................................................................16 Signal Modulation...............................................................................................16 MAC Layer Protocols..........................................................................................18 Performance Results and Analysis.............................................................................19 Simulation Results...............................................................................................19 Real-World PLC Network Performance.............................................................21 Performance of Delivering Streaming Video...............................................22 Performance of Elastic Data Traffic.............................................................23 Performance of Combined Delay Sensitive Traffic and Elastic Data Traffic24 Conclusion...........................................................................................................25 3 A COMPARATIVE PERFORMANCE STUDY OF WIRELESS AND POWER LINE NETWORKS....................................................................................................27 Introduction.................................................................................................................27 Homeplug 1.0 Protocol...............................................................................................29 PLC Environment................................................................................................29 HomePlug 1.0 PHY.............................................................................................30 v

PAGE 6

HomePlug 1.0 MAC............................................................................................31 Theoretical Performance of 802.11a/b and Homeplug 1.0.........................................33 802.11a/b Theoretical Network Performance......................................................33 Performance Analysis of HomePlug 1.0.............................................................34 Experimental Setup.....................................................................................................36 Experimental Method..........................................................................................37 Results.........................................................................................................................38 IEEE 802.11a Indoor Performance......................................................................39 IEEE 802.11b and HomePlug 1.0........................................................................41 TCP Link Stability...............................................................................................45 Discussion and Conclusion.........................................................................................51 4 PERIODIC CONTENTION FREE MULTIPLE ACCESS FOR POWER LINE COMMUNICATION NETWORKS..........................................................................55 Introduction.................................................................................................................55 Previous Works...........................................................................................................57 Proposed PCF/MA Protocol.......................................................................................59 The Concept of PCF/MA.....................................................................................60 Distributed Admission Control............................................................................60 Data Exchange.....................................................................................................63 MAC Protocol Data Unit.....................................................................................64 Solutions to the Hidden Node Problems and Near Far Effect.............................64 Approximate Performance Analysis and Simulation Results.....................................66 Determine Parameter R.......................................................................................67 Near-Far Effect Modeling...................................................................................71 Extra Allocation...................................................................................................72 Data Stream Model..............................................................................................73 Delay Model........................................................................................................75 Video Traffic.......................................................................................................80 Visualization of Protocol Simulation Data.................................................................87 Introduction.........................................................................................................87 Goal of User Interface.........................................................................................88 Data Presentation Mechanism and System Design.............................................91 Experience Using the Visualization System........................................................95 5 FUTURE WORK AND CONCLUSION...................................................................96 Future Work................................................................................................................96 Voice over IP..............................................................................................................96 Distance Factor....................................................................................................97 Congestion Degree..............................................................................................98 Priority Factor......................................................................................................99 Related Study.....................................................................................................100 Experimental Performance Results...................................................................101 Results Based on the Distance Factor...............................................................103 Results Based on Congestion Degree................................................................105 vi

PAGE 7

Results Based on Priority Factor.......................................................................106 Conclusion................................................................................................................107 LIST OF REFERENCES.................................................................................................110 BIOGRAPHICAL SKETCH...........................................................................................113 vii

PAGE 8

LIST OF TABLES Table page 2-1: Technology comparison...............................................................................................8 2-2: Application traffic amount in a home.........................................................................12 2-3: A power line network simulation results....................................................................20 2-4: Real-world PLC network performance......................................................................22 3-1: 802.11a/b MAC throughput with payload 1500 bytes...............................................34 3-2: List of the houses tested and connectivity..................................................................38 4-1: HomePlug 1.0 and PCF/MA parameters....................................................................61 4-2: Video traffic parameters.............................................................................................84 5-1: Performance comparison for distance factor............................................................103 5-2: Performance comparison based on congestion factor..............................................105 5.3: Performance comparison based on priority factor in day time.................................107 viii

PAGE 9

LIST OF FIGURES Figure page 2-1: Power-line topology in a North American home.......................................................11 2-2: An example of using one of the computers as the PLC, DSL or Cable Modem router15 2-3: Connecting PLC networks to the Internet..................................................................26 3-1: HomePlug 1.0 frame structure and protocol...............................................................32 3-2: IEEE 802.11a indoor connectivity.............................................................................39 3-3: IEEE 802.11a indoor throughput as a function of distance........................................40 3-4: IEEE 802.11a indoor percentage of links...................................................................40 3-5: IEEE 802.11b and HomePlug 1.0 indoor connectivity comparison...........................41 3-6: IEEE 802.11b and HomePlug 1.0 indoor throughput comparison.............................42 3-7: IEEE 802.11b and HomePlug 1.0 indoor percentage of link versus throughput (MIM) comparison...............................................................................................................43 3-8: IEEE 802.11b and HomePlug 1.0 indoor percentage of link versus throughput (IM) comparison...............................................................................................................45 3-9: IEEE 802.11a high speed real-time capture...............................................................47 3-10: IEEE 802.11a low speed real-time capture..............................................................48 3-11: IEEE 802.11b high speed real-time capture.............................................................48 3-12: IEEE 802.11b low speed real-time capture..............................................................49 3-13: Homeplug 1.0 high speed real-time capture.............................................................50 3-14: Homeplug 1.0 low speed real-time capture..............................................................50 4-1: PCF/MA frame structure............................................................................................60 4-2: MPDU process and format.........................................................................................61 ix

PAGE 10

4-3: Reservation process flow chart...................................................................................62 4-4: Probability of successful reservation..........................................................................71 4-5: Maximum over allocation PBs various PB loss rate comparisons.............................73 4-6: PCF/MA various T versus MCSMA/CA throughput comparisons............................74 4-7: PCF/MA delay model.................................................................................................75 4-8: Analytical and simulated average delay with various w, T=25ms.............................78 4-9: Analytical and simulated average delay with various w, T=50ms.............................78 4-10: Analytical and simulated average delay with various w, T=75ms...........................79 4-11: Analytical and simulated average delay with various w, T=100ms.........................79 4-12: Video traffic model...................................................................................................80 4-13: A large video frame in transmit................................................................................81 4-14: Video playback performance comparisons..............................................................84 4-15: Capture of a period of the DVD simulation.............................................................85 4-16: Multiple video streams comparisons........................................................................86 4-17: Text versus graphical event analyzer.......................................................................88 4-18: Turning off unused events can make desired events stands out...............................91 4-19: VPA configuration format........................................................................................92 4-20: Simulator log format.................................................................................................93 4-21: The actual display window is a portion of the whole data.......................................94 x

PAGE 11

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 HIGH-PERFORMANCE PERIODIC CONTENTION-FREE MULTIPLE-ACCESS PROTOCOL FOR BROADBAND POWERLINE COMMUNICATIONS By Yu-Ju Lin May, 2004 Chair: Haniph Latchman Major Department: Electrical and Computer Engineering Applications over PLC networks have drawn much interest in the academic community as well as in the communication industry, not only because of inherent convenience (connecting PLC-capable devices requires no new wires), but also because almost all electrical devices have to connect to a power outlet eventually. This technology makes implementing a digital home entertainment center more realistic than ever. HDTVs in different rooms are now able to share digital content from one set-top box without rewiring, or setting up wireless access points. MP3 players can access music data through PLC networks from different rooms playing different music. PLC networks provide an enabling technology for the smart home. However, PLC technology is still evolving and many problems remain unsolved. The hostile environment of PLC channels makes reliable data transmission difficult. The PLC channel is known for its hostile nature in transmitting electrical signals. Protocols designed for other media may not be suitable for PLC. In some ways, PLC channels are xi

PAGE 12

similar to wireless channels: both of them face hidden node problems, near-far effects and other channel imperfections. However, PLC network nodes tend not to move. It is unlikely that simply applying protocols designed for another medium would result in good performance in the PLC environment; the overhead may be too high or the assumptions about noise may be too optimistic for PLC networks. In light of PLC's unique characteristics, we developed a new protocol called: Periodic Contention-Free Multiple Access (PCF/MA). PCF/MA is an explicit R-ALOHA-like protocol specifically designed for the PLC network. We propose an RTS/CTS-like scheme in the reservation stage to mitigate hidden-node problems, and a delayed NACK mechanism to conquer the near-far effect. Performance of the proposed protocol is evaluated by event-driven computer simulations and by mathematical analysis. Simulation results show that 85 Mbps MAC throughput with a 100 Mbps channel data rate can be obtained, even when there are hidden nodes in the network. To provide smooth video delivery, we propose a mathematical estimation of the required delay in playback time and the amount of playback buffer with tight bandwidth reservation. Our simulation shows that a 100 Mbps channel can deliver up to 9 MPEG-2 video streams simultaneously without dropping any video frames. A visual protocol analyzer was also developed as a tool to study network protocol. xii

PAGE 13

CHAPTER 1 INTRODUCTION The concept of Information Appliances (IA) became a recent reality. Many next-generation appliances come with communication capabilities with embedded processors built right into the devices. For instance, on April 7, 2001, IBM and Carrier announced that they will produce a new air conditioner with JAVA support that can send e-mails to manufacturers for errors; or the user can send commands to the air conditioner to pre-adjust room temperature. We believe that before long our homes will have many kinds of IA devices communicating among them and with the outside. Many of these IA devices are expected to have multimedia capability. Intelligent homes of the future will need multimedia communication support for these IA devices. Providing the right infrastructure for connecting these IA devices will be a major need. For home applications, this infrastructure must be easy to setup, inexpensive to install and maintain, and must perform well. Ordinary people are not network experts, and a typical high-performance network is too complicated for casual daily usage. The supporting infrastructure should be easy to set up, and the effort needed to maintain this infrastructure should be minimal. Differences between home networks and ordinary networks can be summarized as follows: Home networks are different than other types of networks. The data stream in a home network is much smaller than in ordinary public networks and local area networks, because the coverage of a home network is small. With this small coverage, the expected response time is short too. The main purpose of home networks is to share resources. Examples, include sharing a single printer for several computers; using a single Internet connection; 1

PAGE 14

2 sharing files (such as images, spreadsheets and documents); playing games that allow multiple users at different computers; and sending the output of a device (like a DVD player or Webcam) to other devices. Home networks should be easy to setup, maintain, and access. They should be affordable for most families. Many existing networking technologies compete to support this mission. For example, a comprehensive Ethernet network can be constructed by installing UTP-5 special cabling around the house. Alternatively, wireless networks such as 802.11x, Bluetooth, and HomeRF can be constructed by installing multiple interconnected wireless access points (WAP) and base stations within the home. However, the IA devices themselves would need wireless capabilities; and the above three infrastructures all require significant effort and cost to build up the networks externally. Phone-line networks such as HomePNA [1] may seem attractive, but the convenience of mobility is limited by the number of phone sockets available in a home. Hughes and Throne [2] did extensive study of other infrastructure options and technologies appropriate for a home network. In general, for future IA applications, the amount of traffic generated by appliances and computers on a home network is still unknown. Watching digital TV while downloading data from the web should cause no video jitters over home networks. To the best of our knowledge, there is little research to report the performance impact (or at least the likelihood) for supporting multimedia-enabled IA communication over home networks. We are interested in exploring various methods to study the impact of different performance results. We are also interested in analyzing the effect on each individual data streams including continuous media data stream (that is, soft real-time traffic).

PAGE 15

3 In Chapter 2, we studied the performance of multimedia over power line networks using simulation studies and actual measurements on a \emph{Homeplug} 1.0 compliant PLC network. We were particularly interested in measuring the PLC network raw data rate, TCP performance, and the performance impact when QoS support is involved. We were also interested in analyzing network performance with different traffic types, including continuous media data streams (that is, soft real-time traffic). We first built a network simulator that generates various types of traffic, and then applied the same scenarios to a real-world PLC network and to a simulation model. We compared simulation results and real-world performance. A maximum throughput of 8.08 Mbps for UDP was obtained from our simulation, while a 6.21 Mbps TCP throughput was observed in the real-world PLC network experiment. Results show that PLC networks can successfully deliver real-time traffic concurrently with traditional data traffic. Our contributions to this research topic also include modelling human behavior in the use of IA devices, modelling the types of traffic generated by IA devices communicating over the power line, and measuring the performance of real applications over PLC networks. We also describe a practical implementation of the PHY and MAC layers for PLC networks associated simulation results. In Chapter 3, we discuss PLC, 802.11b and 802.11a characteristics, protocol, and theoretical performance. Based on the theoretical analysis, we conducted a series of real-world experiment and performance comparisons between 802.11b and PLC networks. We compared two technologies for implementing a home or small-office networks without new cable installation. For ease of installation, both wireless and power-line

PAGE 16

4 networks were well rated. The wireless network uses air and the power-line network uses the existing power cable as a transmission medium. Current field testing results show that under most circumstances, power line has more stable and reliable connections. Even when the line-of-sight distances between two stations are as long as 69 ft., PLC devices can still generate up to 4.52 Mbps throughput. From 128 testing positions, PLC has 72% (from server to mobile station) and 59.38 % (from mobile station to server) better experimental results than 802.11b. In Chapter 4, we proposed a new Protocol: Periodic-Contention Free Multiple-Access (PCF/MA) Protocol for high-speed Power Line Communications (PLC). PLC networks have impairments similar to wireless networks, but the nodes are largely stationary. There is industrial intent to use PLC networks in the home for delivery of multimedia data, with challenging quality of service (QoS) requirements. Existing protocols for PLC, for wireless networks, or for wired networks cannot meet these challenges efficiently. Our study proposed and analyzed a new protocol designed to provide the high QoS needed for delivering multiple multimedia streams in a PLC environment. The proposed protocol, PCF/MA, addresses hidden nodes and the near-far effect; in addition to addressing the generally noisy medium. PCF/MA performance was analyzed theoretically and was simulated. Simulation showed that 85 Mbps MAC throughput is possible with a channel-data rate of 100 Mbps, even when there are hidden nodes in the network. Through mathematical modeling of buffer space required for tightbandwidth allocation, such a network can deliver up to 9 MPEG-2 video streams simultaneously without dropping any video frames, compared to 7 video streams using Modified CSMA/CA(MCSMA/CA).

PAGE 17

5 Chapter 5 discusses three common factors that influence the Voice over IP quality and future works. Our experiments show that without dedicated networks and QoS enabled schemes, the factors of distance and degree of congestion were proved to be influential on the achieved end-to-end bit rates. Many research issues still need to be addressed before we can guarantee (at least statistically) the quality of the Interne Telephony.

PAGE 18

CHAPTER 2 POWER LINE COMMUNICATION NETWORK INFRASTRUCTURE FOR HOME NETWORKS Low-voltage electrical wiring in homes has largely been dismissed as too noisy and unpredictable to support high-speed communication signals. However, recent advances in communication and modulation methodologies (as well as in adaptive digital signal processing and error detection and correction) have spawned novel media access control (MAC) and physical layer (PHY) protocols capable of supporting power line communication networks at speeds comparable to wired local area networks (LANs). In this chapter we support the use of power-line LAN as a basic infrastructure for building integrated smart homes, where information appliances (IA)ranging from simple control or monitoring devices to multimedia entertainment systems are seamlessly interconnected by the very wires that provide their electricity. By simulation and actual measurements using reference design prototype commercial powerline products, we showed that HomePlug MAC and PHY layers can guarantee QoS for real-time communications, supporting delay-sensitive data streams for smart home applications. Home-Network Infrastructures Many next-generation appliances are being equipped with processors featuring sophisticated communication capabilities. For instance, on April 7, 2001, IBM and Carrier announced plans to produce an air conditioner with JAVA support that can Email manufacturers regarding errors, and will allow users to remotely send commands to the unit to adjust temperatures or shut it down. Smart homes will eventually have many types 6

PAGE 19

7 of information appliances (IAs) communicating among themselves and with the outside world. Soon, many of these IA devices are expected to have multimedia capability. Supporting multimedia communication for these IA devices will be of crucial importance for the smart homes of the future. Providing the right infrastructure for connecting these IA devices will be a major need. For home applications, this infrastructure must be easy to set up, inexpensive to install and maintain, and must perform well. Ordinary people are not network experts, and a typical high-performance network is too complicated for casual daily usage. The supporting infrastructure should be easy to set up, and the effort to maintain this infrastructure should be minimal. Many existing networking technologies compete to support this mission. For example, a comprehensive Ethernet network can be constructed by installing UTP-5 special cabling around the house. Alternatively, wireless networks such as 802.11x, Bluetooth, and HomeRF can be constructed by installing multiple interconnected wireless access points (WAP) and base stations within the home. However, the IA devices themselves would need wireless capabilities, and the above three infrastructures all require a significant effort and cost to build up the networks externally. Phone-line networks such as HomePNA [1] may seem attractive, but the convenience of mobility is limited by the number of phone sockets available in a home. Hughes and Throne [2] did extensive study of other infrastructure options and technologies appropriate for a home network. In this chapter, we advocate direct use of existing electrical wiring and outlets as the medium for data communication within the home. Using power lines as the network infrastructure has many advantages over other technologies. First, no new wires are

PAGE 20

8 needed, since the IA devices will communicate over the very wires that provide their electrical power. Second, there are many access points (power sockets) in a home (4 or more per room). Currently, Power Line Communication (PLC) as specified by the HomePlug 1.0 standard [3] provides a 14 Mbps raw data rate, which is adequate for daily IA-device communication. It also has a built-in QoS protocol, making it attractive for real-time streaming applications. Finally, the cost to build a power-line network is low when compared with other technologies. For example, it was observed that the 802.11x wireless network card has approximately the same street price as the HomePlug network card (about $120). It is expected that with mass production requiring no expensive RF components, the cost of the PLC cards will be about 50% less than comparable wireless cards. Moreover, the cost of a required 802.11x base station is high (more than$250). The 100 Base T Ethernet has the highest performance/cost ratio, but requires new cables and expensive installation. Table 2-1 shows costs and other characteristics of home-network technologies. Installation costs (which are high for 10/100 BT) are not shown. Table 2-1: Technology comparison Technology Media Data Rate(Mbps) QoS Support Cost($) 10 Base T UTP 10 No 20 100 Base T UTP 100 No 80 Bluetooth Wireless 1 Yes 5 HomeRF 2.0 Wireless 10 Yes 110 802.11x Wireless 11 No 125 HomePNA 2.0 Phone line 10 No 80 HomePlug Power line 15 Yes 120 From a marketing perspective, less-expensive and easier-to-use PLC home networks are becoming more attractive, and the potential market is huge. The Yankee Group (Boston, MA) estimates that at least 21 million households in the United States are PAGE 21 9 interested in home networking and that 12.4 million would like to implement in-home networks within the next year. According to Parks Associates (Arlington, VA), 30 million households in the United States will have fast internet connections by 2004, and 17 million of them plan to have home networks. In the past, power lines were considered unacceptable for signal transmission, since the channel contained a lot of noise, interference, and fading. However, the appeal of using the existing power line as a transmission medium for data exchange was too great to be ignored. The advancement of signal-modulation technologies, digital signal processing, and error control coding [4] has minimized the restrictions of channel imperfections; and high-speed signal transmission through power lines is now feasible. Using the existing power-line infrastructure as the medium for supporting IA communication requires a careful design of the overlaid communication systems in order to provide acceptable communication services. It is desired, for example, that when watching digital TV while downloading data from the web, there will be no delay-jitter in the video quality. Current research shows that the maximum raw data rate of first generation PLC is about 14 Mbps. However, the effective data rate is expected to be around 10 Mbps after compensating for impairments and error corrections. On the other hand, research is currently underway to develop PLC chips that operate at 100 Mbps with average throughput 30-60 Mbps. In this chapter, we investigate the performance of multimedia over power-line networks using simulation studies and actual measurements on a Homeplug 1.0 compliant PLC network. We are particularly interested in measuring the PLC network raw-data rate, TCP performance, and the performance impact when QoS support is involved. We are PAGE 22 10 also interested in analyzing the network performance with different traffic types, including continuous-media data streams (that is, soft real-time traffic). We first built a network simulator that generates various types of traffic, and then applied the same scenarios to a real-world PLC network and to a simulation model. The performance comparison between the simulation results and real-world performance are given in this chapter. A maximum throughput of 8.08 Mbps for UDP was obtained from our simulation, while a 6.21 Mbps TCP throughput was observed in the real-world PLC network experiment. Results show that PLC networks can successfully deliver real-time traffic concurrently with traditional data traffic. Our contributions to this research topic also include modelling human behavior in the use of IA devices, modelling the types of traffic generated by IA devices communicating over the power line, and measuring the performance of real applications over PLC networks. We also describe a practical implementation of the PHY and MAC layers for PLC networks as well as associated simulation results. The next section discusses background for implementing power-line communication networks, typical applications of power line communication networks, and the human behavioral model using these appliances. Section 2.3 presents the physical limitations of power line channels, and describes a practical signal-modulation scheme and MAC protocol as used in the HomePlug 1.0 protocol proposed by the HomePlug Powerline Alliance. Section 2.4 compares simulation results with real-world PLC network performance. Finally, Section 2.5 offers conclusions and suggestions for future work. PAGE 23 11 Applications over Power Line Communication Power Line Communication Application in a Home Figure 2-1: Power-line topology in a North American home Traditionally, power lines are used for conveying electrical power to devices. Power lines were not designed for delivering high frequency signals, and so the electrical and frequency response requirements of a power line are not as critical as those of data network cabling. The poor quality of a power line is not ideal for signal transmission because the channel contains noise and interference. The medium is made of different conductor types; therefore a variety of characteristic impedances will be encountered. Further, the network terminal impedance will tend to vary with frequency and time as the consumers load pattern and load types vary. Impedance mismatch causes a multi-path effect resulting in deep notches at certain frequencies. These channel imperfections make signal modulation over a power line difficult [5]. However, the advancement of signal modulation and error control coding techniques now make power line communication possible. PAGE 24 12 The common power line topology of a North American home is shown in Figure 2-1. The figure shows a tree-like power-line topology in a house. Typically, there are two power line trunks: one is 110V and the other is 220V. Each power-line trunk can be divided into several branches. Power-line communication aims to transmit data packets over these branches and trunks. The topology of the power line network and the convenience of its power sockets as potential access points make it a good candidate for smart home IA device networking. Table 2-2 shows the results of a survey from which we inferred usage and traffic patterns generated by typical IAs. The table also suggests some current and future PLC applications. For instance, when merchandize is advertised on a digital TV service, the product information (such as the barcode or webpage) can be downloaded to your computer through a power line. Afterwards, you can send your order information from the computer to the supplier, or you can use the downloaded URL to browse the product web page and get more details. We also anticipate the ability to record music or videos through a power line. For example, when a song is broadcast on TV or a music channel, you can download the song directly to an MP3 player through the power line. Another application is the opportunity to record digital video directly into a PC or even a digital VCR. Other applications of IAs can be easily accomplished using a PLC network. For example, a refrigerator can order food through the power line network according to its inventory, or it can send cooking instructions to the microwave. A smart oven can send predicted environmental temperature information to the air conditioner through a power line, allowing the air conditioner to pre-adjust the temperature and keep rooms comfortable. PAGE 25 13 Table 2-2: Application traffic amount in a home Row No. From node To Node Estimated data size Frequency Possible time period 1 Refrigerator Microwave 160 bytes 2 times a period 7:00-9:00,11:00-1:00, 17:00-19:00,21:00-23:00 2 Microwave AC 72 bytes 2 times a day 7:00-9:00,11:00-1:00, 17:00-19:00,21:00-23:00 3 TV Refrigerator 750 bytes 3 times a day 11:00-1:00,17:00-23:00 4 TV VCR 11KBytes 3 times a day 11:00-1:00,17:00-23:00 5 TV Computer 360 bytes 3 times a day 11:00-1:00,17:00-23:00 6 TV or Settop box PDA or MP3 player 15 Mega bytes 3 times a day 11:00-1:00,17:00-23:00 7 Computer PDA or MP3 player 50 Mega bytes 1 time a day 11:00-1:00,17:00-23:00 8 Computer Computer 60 MB to 180 MB 1 time a day 6:00-24:00 9 Settop box Computer 320 MB to 640 MB 1 time a day 11:00-1:00,17:00-23:00 10 Computer Internet 44 MB to 131 MB 1 time a day 11:00-1:00,17:00-23:00 11 VCR Computer 320 MB to 640 MB 1 time a day 6:00-24:00 12 Front door camera Computer 110 MB to 1100 MB 3 times a day 6:00-24:00 The applications over PLC are not only for novel IA devices. PLC as a home network facilitates data exchange between traditional data processing devices such as PCs and computer peripherals. IA devices that talk with PCs are also possible. For example, sending multimedia data from TVs or VCRs to PCs can be easily done by PLC network, but is difficult with other infrastructure technologies. Home security can also be implemented by PLC so that a digital camera installed on the front door can send video to the TV. PAGE 26 14 Table 2-2 also gives an estimate of the daily traffic volume generated by typical IA applications. These values are based on likely information size. For example, the instruction size that the refrigerator sends to the microwave in Row 1 is estimated by the number of steps required to cook the food (1 byte), the cooking time for each step (4 bytes for each step), the power level for each step (2 bytes), and the packet header size. Added together, the entire instruction size is 160 bytes. Row 7 exemplifies storing digital music from a computer to an MP3 player. The 50 Mbytes traffic volumes is calculated from the number of songs in an album, the length of a song (5 min), the encoded data rate (128 kbps), and the packet header size. The frequency and time period during which each event occurs are also shown. By using this data and typical household dynamics for concurrent events, we can generate a traffic flow for the power-line network for a typical day. Internet Bridging Currently, in-home PLC networks rely on other technologies to send data to the Internet and communicate with mobile devices. Most of the homes in the United States will eventually be equipped with broadband connections like DSL or cable modem services. To share the broadband Internet connection with PLC capable devices, we can add a PLC Internet router to the PLC network. One possible setup is shown in Figure 2-2. In this figure, a desktop computer acts like a data center. Devices that need to communicate with other devices on the Internet will send data to the desktop PC via the power line. The desktop PC decides whether to send it to the Internet. In the future, an IA routing device may be unnecessary. Researchers are developing a solution to make PLC home networks talk directly with other homes, power plants and the Internet using the external distribution power line. Such a network infrastructure for the Internet access PAGE 27 15 would be especially attractive to developing countries, since no additional expenditure is needed for data network infrastructures. Figure 2-2: An example of using one of the computers as the PLC, DSL or Cable Modem router To support data exchange with mobile devices, PLC networks will also need to cooperate with wireless networks. The easiest way to achieve this is to make the 802.11x base station PLC compatible. The base station is treated as an ordinary IA device with a PLC chip built in (see Figure 2-2). Mobile devices with wireless capability can then talk to devices attached to the power line. This is especially ideal when communication is desirable but large coverage areas require multiple interconnected wireless access points (the interconnection is then provided with no new wires using the existing power line infrastructure which would be needed to power the WAPs in any event). The above PLC applications require a properly designed protocol. In addition, to make the PLC network real-time traffic friendly, special care is needed to support delay PAGE 28 16 sensitive traffic. In the following sections, we discuss the physical limitations of power line channels and then describe a robust power line protocol. PLC Design Issues Physical Limitations A power line is used for transmitting 50 or 60 Hz signals but was not designed to convey high frequency signals such as the 20MHz communication signal used in the Homeplug 1.0 protocol. A power line channel is somewhat like a wireless channel both of them suffer from noise, fading, multi-path and interference. Power line noise is produced by the operation of electrical devices. Fading, multi-path and interference are caused by the imperfection of power line channels. Lim C.K., So P.L., Gunawan, E., Chen, S., Lie, T.T.,and Guan, Y.L., [6] describe typical attenuation characteristics in power line channels. The authors report that even when all devices are unplugged, the noise still persists and this drastic variation of attenuation is hostile to power line communication. Furthermore, the Federal Communications Commission (FCC) also limits the available bandwidth for communication purposes. In compliance, the usable bandwidth in the HomePlug standard is 25MHz. Liu Weilin, Widmer H.-P., Aldis J.,and Kaltenschnee T., [7] did an extensive study of the power line channel characteristics and design issues. To conquer these problems, a robust signal modulation and data coding is needed. Signal Modulation To modulate digital signals on to the power lines, we can use many of the same techniques that are widely implemented in wireless communication. Basic modulation techniques such as Phase Shift Keying (PSK), Frequency Shift Keying (FSK), Minimum Shift Keying (MSK), and Gaussian Minimum Shift Keying (GMSK) can be used for low PAGE 29 17 data rate communication. Other more advanced techniques such as M-ary Phase Shift Keying (MPSK), M-ary Quadrature Amplitude Modulation (MQAM), M-ary Frequency Shift Keying (MFSK) and Orthogonal Frequency Division Modulation ( OFDM) can be used when higher data rates are desired. Karl M., and Dostert K.[8] did a thorough study of signal modulation over power lines. OFDM was adapted by HomePlug Powerline Alliance because of its robustness to noise and the fact that it is a parallel data transmission method using a number of parallel frequency division multiplexed subbands. The main problem in using OFDM on wireless networks is frequency offset, caused by the Doppler Effect when the user is moving. The Doppler Effect will cause performance degradation, but in a power line network there are no moving devices, and thus no Doppler effect. The other problem is timing offset, which can be mitigated by offset estimation and compensation. Spread spectrum signal modulation is different. Since the useful bandwidth in the power line channel is under 25 Mhz, the effect of spread spectrum modulation is considered limited. Using a single carrier modulation on the power line is possible but equalizers could be needed to reduce the delay spread effect, and the associated cost is high. In order to cope with the wide variation in channel conditions, the physical layer protocol (PHY) for PLC must be adaptive intelligently using more robust modulation and coding schemes, with lower data rates as needed. In addition, critical protocol management information requires high fidelity forward error correction (FEC) coding to ensure that the protocol functions correctly in the worst case situations. PAGE 30 18 MAC Layer Protocols In PLC home networks, the power line media can be accessed by multiple devices simultaneously. To decide which device gets the floor to send its data, a medium access control (MAC) protocol is needed. There are many existing protocols that can be implemented on the power line network. CSMA/CD, CSMA/CA, TDMA and hybrid protocols such as TDMA+CSMA are all potential candidates. The most popular wired MAC protocol, CSMA/CD, could be also applied on a power line network. However, the large variation in noise on the power line makes collision detection very difficult. This characteristic is again very similar to a wireless network, so some have applied the CSMA/CA protocol as suggested in IEEE802.11 to the power line network. However, the hidden node problem arises when the signal travels through different power lines with highly variable attenuation. To conquer this problem, a RTS/CTS scheme has to be implemented. Though the RTS/CTS scheme solves the hidden node problem, it degrades the network performance. The benefit of using TDMA is that it provides an upper bound of access delay thus QoS is guaranteed. However, the difficulties in generating a synchronized clock signal in power line networks between devices remains a problem. Other hybrid protocols like TDMA+CSMA provide QoS capabilities in nature, but the network efficiency and beacon generation between TDMA slots and CSMA/CA slots remains unsolved. Romans C., and Tourrilhes J. [9] did a detailed discussion of the hybrid TDMA+CSMA/CA protocol. Homeplug 1.0 protocol also provides some level of QoS support in the uses of multiple priority levels that can be used in conjunction with VLAN tagging. The issue of privacy of power line networks is important to their practicality. Like wireless channels, Power line network channels should be treated as open and as with all PAGE 31 19 open channels; nothing prevents a device from receiving signals. To provide a secure network environment, the HomePlug Powerline Alliance defined a 56-bit DES encryption mechanism. Once a signal is encrypted, a device with a different encryption key cannot interpret it and privacy is achieved. This privacy protection seems adequate but stronger encryption may be needed when power line networks are adopted for office environments or apartment building and hotels. We believe that stronger privacy protection should be implemented in the physical layer, so that hackers can not easily break the code. Performance Results and Analysis 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 and a maximum TCP segment size of 1460 bytes. In this simulation, we use UDP, TCP, and VOIP traffic. UDP traffic is generated with an exponential inter-arrival time with a 100 microsecond average. The UDP packet size is assumed to be a constant 1460 bytes with priority 0. TCP traffic is also generated with exponential inter-arrival time with 100 microsecond average and we assume that TCP traffic sources always have data to send. TCP traffic is treated as priority 0 packets. Every time a node has a chance to send, it is allowed to send the maximum segment size of 1460 bytes without headers. VOIP is isochronous traffic with a 20 msec interval. The packet size of VOIP is 160 bytes and is assigned the highest priority (3). Simulation Results In Table 2-3, 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 PAGE 32 20 all. Table III also shows channel contention with 2 and 3 UDP nodes causes a modest reduction in channel throughput. Table 2-3: A power line network simulation results Throughput of multiple UDP traffic Scenario 1(1 UDP) Scenario 2(2 UDP) Scenario 3(3 UDP) MAC Throughput 8.08 Mbps 7.46 Mbps 7.46 Mbps Throughput of multiple TCP traffic Scenario 1(1 UDP) Scenario 2(2 UDP) Scenario 3(3 UDP) MAC Throughput 6.16 Mbps 6.15 Mbps 6.12 Mbps TCP Throughput 5.92 Mbps 5.91 Mbps 5.88 Mbps Throughput of one VOIP and multiple UDP traffic Scenario 1 (VOIP + 1 UDP) Scenario 2 (VOIP + 2 UDP) Scenario 3 (VOIP + 3 UDP) MAC Throughput 7.89 Mbps 7.33 Mbps 7.29 Mbps Queueing Delay 0.25 msec 0.25 msec 0.25 msec Net Delay 2.75 msec 3.00 msec 3.00 msec Throughput of one VOIP and multiple TCP traffic Scenario 1 (VOIP + 1 TCP) Scenario 2 (VOIP + 2 TCP) Scenario 3 (VOIP + 3 TCP) MAC Throughput 6.04 Mbps 5.85 Mbps 5.77 Mbps TCP Throughput 5.72 Mbps 5.54 Mbps 5.45 Mbps Queueing Delay 0.25 msec 0.25 msec 0.25 msec Net Delay 3.25 msec 3.25 msec 3.25 msec In the TCP traffic simulation, though scenario 1 has only one traffic source, the bandwidth must be shared with data and response frames (for example, ACK packets) thus it provides lower performance than the UDP traffic simulation. The MAC throughput represents the total number of transmitted bytes divided by the simulation time regardless of successful delivery. The TCP throughput includes only the successfully delivered data and ACKs. The third metric we provide in Table III 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 queuing delay refers to the time a packet waits in a queue before it enters the transmit PAGE 33 21 buffers. The net delay is the total time for which a packet propagates in the networks. Only low priority packets are considered for this delay because the high priority packets will be delivered as soon as they appear in the queue. The Table 2-3 also shows the simulation results of one VOIP and multiple TCP connections. The throughput of VOIP is only 80 kbps, and hence the total throughput is dominated by the TCP component. Real-World PLC Network Performance In addition to simulating the performance of the HomePlug Powerline Alliance protocol, we were also able to construct a real PLC network using reference designs of actual commercial HomePlug devices. Since there are currently no real IA devices with PLC capability, we used traditional network applications (that is, ftp and streaming multimedia content) as the basis for measuring PLC network performance. In this experiment, there were 4 desktop computers. A 450 MHz Pentium II desktop computer (PC-2 as a file server) is equipped with 128 MBRAM, a 3-COM fast Ethernet card, and a PLC PCI card. Two 700 MHz Pentium III desktop computers (PC-3 and PC-4) are both equipped with 256 MBRAM and PLC PCI cards. A 266 MHz Pentium MMX desktop computer (PC-1) is equipped with 64 MB RAM, and a 3-COM fast Ethernet card. The PC-1 computer is connected to an Ethernet-to-power line bridge, which converts packets generated from the Ethernet card into PLC compatible packets, and vice versa. All computers are connected to power lines. In this experiment, we seek to determine the performance of the PLC network in handling streaming video and large file transfers. PAGE 34 22 Performance of Delivering Streaming Video We first examined the ability of the PLC network to deliver real-time traffic. Four video files are involved in this experiment. The first file is encoded in Real media format with a bit rate of 550 kbps; the second is encoded with bit rate of 1396 kbps, the third is encoded at 2 Mbps, and the fourth is an MPEG2 video file with variable bit rate, and the average bit rate is 8 Mbps. In the first experiment, three client computers simultaneously issued requests for low bit rate (550kbps) video service to the file server. In the second experiment, the same procedure was executed, but a medium bit rate (1394 Kbps) video service was requested. In the third experiment, the 3 clients requested a high bit rate (2 Mbps) video service. Finally, the MPEG2 video service request was issued by the PC-3. The experimental results are shown in Table 2-4. The PLC network successfully delivered both low and medium bit rate streaming videos. We did not observe any packet drops during the experiments. The results met our expectations, since the peak data rate was only 4185 kbps. We did another experiment to further investigate the performance of PLC network in delivering streaming video. A 2 Mbps MPEG-1 file is used in this experiment. As the video begins, a significant video freeze-then-go (halting) phenomenon was observed, causing staccato playback. After several seconds (3-5 seconds) the freeze-then-go phenomenon disappeared. In the case of MPEG2 video file, the average data rate is 8Mbps. During the experiment, a significant video staccato phenomenon was observed. To exclude the possibility that the observed phenomenon was caused by the client computers hardware capability, the experiment was repeated with same configuration, while connected to a fast Ethernet. During that experiment, no such phenomenon (halting playback) was observed. PAGE 35 23 Table 2-4: Real-world PLC network performance Performance of delay sensitive traffic Number of Connections Aggregated bit rate Packet drop delay-jitter 1 550 kbps No No 2 1100 kbps No No Low Bit Rate 3 1650 kbps No No 1 1395 kbps No No 2 2790 kbps No No Medium Bit Rate 3 4185 kbps No No 1 2000 kbps No No 2 4000 kbps No No High Bit Rate 3 6000 kbps N/A Moderate Number of Connections Aggregated bit rate Environment 1 8 Mbps PLC network Variable Bit Rate 1 8 Mbps Fast Ethernet Performance of elastic data traffic Number of Connections Average bit rate 1 6.21 Mbps rate 2 6.15 Mbps rate Elastic data traffic 3 6.27 Mbps rate Performance of combined delay sensitive traffic and elastic data traffic Connections Aggregated bit rate One ftp connection and One video service 6.26 Mbps rate Hybrid data traffic Two ftp connection and One video service 5.92 Mbps rate Performance of Elastic Data Traffic The occurrence of the momentary video freezing phenomenon during playback of variable bit rate streaming is likely because the aggregated data rate was close to or exceeded the PLC network capacity. To understand the real throughput of a PLC network, we conducted another experiment A 215,502,106 byte file was placed on the server running an FTP daemon (The file size was chosen to minimize hardware uncertainty and human error.) Client computers made FTP requests for the file. We tested different numbers of FTP connections, up to 3, using individual client machines in our PLC network. The experimental results are also given in Table 2-4. PAGE 36 24 The aggregated traffic in the table is calculated by adding all observed data rates of all connections. The experimental results show that the real PLC network performance is about 6 Mbps. When there is only one FTP connection, the observed throughput is 6.21 Mbps. By our analysis, one TCP connection will not fully utilize the PLC network, because the server has to stop if no ACK packets are received from the client. Aggregated traffic volume decreased as the number of connections increased from 1 to 2. This phenomenon is because of the ACK packets and the packet overhead increase as the number of connections increased. Although the network utilization improves, the improvement cannot compensate for the loss due to these overheads. When we increased the number of connections from 2 to 3, the PLC network had the highest throughput of 6.27 Mbps. This is because the network utilization increased as the number of connections increased which compensated for the packet overhead and ACK overhead. These experimental results explain the phenomenon of momentary DVD video freezing playback. The requested bandwidth for DVD streaming exceeded the maximum bandwidth the present PLC network can provide. Performance of Combined Delay Sensitive Traffic and Elastic Data Traffic Although we could not explore the QoS service and packet priority provided by the real PLC network, we were eager to learn the effect of mixed traffic on the PLC network. This experiment was conducted as follows: The file server provided two services: one for streaming video with bit rate 550 kbps, and the other one for file transfer with a file size of 215,502,106 bytes. PC-1 requested streaming video while PC-2 and PC-3 requested the file transfer. Each experiment lasted 285 seconds (that is, the length of the video file), PAGE 37 25 after which both video player and FTP client are forced to stop. Table IV shows our experimental results. When the number of FTP connections increased, the observed data rate decreased as was the case in the previous experiment. However the overall average data rate was comparable to the case of multiple FTP traffic. Conclusion The emergence of Information Appliances (IA) for the smart homes of the future will undoubtedly make our lives much more comfortable than ever. However, the infrastructure that supports multimedia traffic and conventional elastic data traffic for communication among IA devices is a critical component of a smart home. We advocate power line as the infrastructure for smart homes based on the convenience of the power sockets and the layout of the power line network existing in every home. At present, 6 Mbps of bandwidth was measured through real-world PLC network experiments. Our studies showed that the PLC network can provide 3 low bit rate or 3 medium bit rate multimedia streams concurrently with no packet drops and jitters. It also successfully delivered one low bit rate multimedia data stream and 2 large ftp file transfer concurrently with no packet drops and jitters. In this chapter, we discussed only the PLC networks for communication within the smart home, but the ultimate goal of PLC network could be the ability to connect to the Internet without dialing up to an ISP server, entirely using electrical wiring only. This can be illustrated as in Figure 2-3. Private home networks are connected to substations, in which a PMTS (Powerline Modem Terminal Service) connects PLC networks within homes to the Internet backbone. The PLC network gateway for a private home network could be installed in the fuse box PAGE 38 26 of that home and then it could be connected to one or more repeaters. Repeaters are for increasing signal strength when the signals level fall below some value. SubstationPowerPlant Internet Backbone ACTACT10M100M 123413141516 567817181920 910111221222324 UPLINK 123456789101112131415161718192021222324 COLCOL PWRSWITCH ACTACT10M100M 123413141516 567817181920 910111221222324 UPLINK 123456789101112131415161718192021222324 COLCOL PWRSWITCH ACTACT10M100M 123413141516 567817181920 910111221222324 UPLINK 123456789101112131415161718192021222324 COLCOL PWRSWITCH Figure 2-3: Connecting PLC networks to the Internet We expect to see higher data rates in power line networks in the future as signal modulation technologies improve; however, issues like network security and the network characteristics with a large number of nodes need further development. Further research on these issues is of critical importance when power line networks are applied to offices and large multi-user buildings. PAGE 39 CHAPTER 3 A COMPARATIVE PERFORMANCE STUDY OF WIRELESS AND POWER LINE NETWORKS Local Area Networks based on the IEEE 802.11a/b wireless networking standards and emerging Power Line Communication (PLC) standards are attractive for establishing networks with No New Wires for in-home and business applications. This Chapter presents a theoretical performance comparison of the 802.11 a/b and the Home-Plug 1.0 PLC protocols. We also presents comprehensive comparative field test results addressing such issues as coverage, channel stability and reliability as well as the associated implications on the capability of these technologies to provide QoS support for multimedia traffic in typical residential settings. Introduction Candidate networking technologies for providing convenient and widespread residential and SOHO networking services may be categorized as Wireless Networks, Wired Networks and No New Wires Networks. Hughes S, and Thorne D.J. [2] did an extensive study of various infrastructure options and technologies appropriate for home networks. Below, we give a short discussion of networks in the above three categories. Wireless Networks such as 802.11x, BlueTooth, and HomeRF can be constructed by installing multiple interconnected wireless access points (WAP) and base stations within target areas. The best benefit of using wireless networks is the freedom to move around while maintaining network connectivity. Blue-Tooth technology is targeted at personal communications and the coverage is expected to be limited. On the other hand, 27 PAGE 40 28 though HomeRF has been on the market for a few years, it is not yet widely accepted. Thus the most interesting and widely accepted wireless networking technologies are the 802.11x family. 802.11b operates in the 2.4 GHz band and provides a maximum data rate of 11 Mbps; 802.11a supports speeds of up to 54 Mbps and operates in the 5 GHz band. Standards for the newer IEEE 802.11g, which should provide data rates up to 54 Mbps in the 2.4 GHz band, have not been finalized, and equipment was not available for testing. For Wired Networks, a comprehensive Ethernet network can be constructed by installing special UTP-5 cabling. While the stability and the security of wired networks are guaranteed, installing new wires in existing home or other buildings may be costly, negating the low cost of the network interface cards. For the No New Wires Networks category, there are phone line networks, cable networks, and power line networks. Using the existing phone line as an infrastructure, as in Frank E.H., and Holloway J. illustrated [2], may seem attractive, but it is limited by available phone sockets in a home. Home Cable Network Alliance (Home-CNA) [10], established in June 2001, and proposes a home network infrastructure using existing coaxial TV cable. There is as yet no standard for HomeCNA and it also suffers from the major drawback of limited convenient connection points. Power Line Communication (PLC) networks such as HomePlug[3] were introduced to the U.S. consumer market in May 2002. European PLC networks have been deployed in recent years. With multiple outlets in almost every room, residential power lines are already the most pervasive network in the home or small office. The HomePlug 1.0 PLC standard supports PHY data rates of 14 Mbps and is thus comparable to the 802.11b declared data rate. PAGE 41 29 A major objective of this paper is to conduct a real-world performance study of the capabilities of wireless (IEEE 802.11b and 802.11a) networks and PLC networks based on the HomePlug 1.0 standard. Our interest is to determine the relative performance of these technologies. This chapter presents a comparative analysis of the TCP performance of power line networks and wireless networks using actual measurements on HomePlug 1.0 compliant PLC networks and 802.11a/b compliant wireless networks. The tests were conducted in 20 houses ranging in area from 1500 to 5000 sq. ft. with an average area of 3000 sq. ft. The paper presents qualitative theoretical and measured throughput performance for 802.11a/b and HomePlug 1.0 PLC. Other issues like the relationship between QoS and channel stability as well as overall coverage are also discussed. The next section briefly describes the HomePlug 1.0 protocol. Section 3-3 presents a theoretical performance analysis of IEEE 802.11a/b and HomePlug 1.0. Section 3-4 describes the experimental setup while Section 3-5 gives our field test results. A summary is given in Section 3-6. Homeplug 1.0 Protocol The parameters and details of 802.11x protocols are well documented in the literature and Internet publications [11]. Here, we briefly describe the HomePlug 1.0 standard. PLC Environment Power lines were originally devised for distributing electrical power using the frequency range of about 50-60 Hz. The use of this medium for high speed communications presents some technically challenging problems. Electrical noise from appliances and the uncontrolled nature of the wiring result in severe signal distortions. PAGE 42 30 The PLC channel is made up of different conductor types; therefore a variety of characteristic impedances will be encountered. Further, the network terminal impedance will tend to vary with frequency and time as the consumers load pattern and load types vary. Impedance mismatch causes multi-path effects resulting in deep notches at certain configuration dependent frequencies. These channel imperfections make signal transmission over a power line very difficult [5]. Reliable data communication over this hostile medium requires powerful Forward Error Correction (FEC) coding, interleaving, error detection and Automatic Repeat Request (ARQ) techniques, along with appropriate modulation schemes as well as a robust Medium Access Control (MAC) protocol. The lack of affordable processing techniques needed to overcome the harsh power line environment resulted in limited success of power line communications in the past. However, both the advances in the ASIC density and speeds, and the advancement of signal modulation, processing and error control coding techniques now make power line communication possible. HomePlug 1.0 PHY To overcome the hostile PLC environment, Orthogonal Frequency Division Modulation (OFDM) with a Cyclic Prefix (CP) was adopted by the HomePlug 1.0 PLC standard. Using OFDM has many benefits. For example, it exhibits excellent mitigation of the effects of time-dispersion, provides excellent Inter-Channel Interference (ICI) performance, and is good at minimizing the effect of in-band narrowband interference. OFDM splits available bandwidth into many small frequency bands called sub-carriers, then may mask out unusable subcarriers and apply the best modulation and coding methods to the usable subcarriers. This approach is used by HomePlug 1.0. A more advanced technique called bit-loading allows use of different modulation and coding PAGE 43 31 schemes for each sub-carrier. In either case, OFDM can adapt bandwidth/data rates according to channel conditions. Unlike 802.11, the bandwidth in HomePlug 1.0 can vary from 1 Mbps to 14 Mbps practically continuously according to the channel conditions 1 Active HomePlug 1.0 nodes perform channel estimation at least once every 5 seconds. This feature allows the PLC network to maximize its data rate adaptively. A preamble and frame control form delimiters used for synchronization and for control. The frame control of start of frame, end of frame, and response delimiters all include delimiter type 2 and contention control information. In the start of frame delimiter, the frame control field includes the tone map information needed by the receiver to decode the rest of the frame, and a length field. The end of frame delimiter contains priority information used for contention control. Response delimiters contain information that allows a sender to verify that the response was indeed sent in response to the frame it just transmitted. An end of frame gap (EFG) of 1.5 s is inserted between the frames frame check sequence (FCS) and the end delimiter to allow for processing. HomePlug 1.0 MAC The HomePlug 1.0 Medium Access (MAC) protocol is a modified CSMA/CA (Carrier Sense Multiple Access / Collision Avoidance) protocol with priority signaling. HomePlug 1.0 devices operate in an ad hoc mode in the sense that devices communicate with each other freely, without any centralized coordination. 1 There are 139 distinct data rates in that range according to number of usable carriers, modulation methods, and coding rate. 2 Response expected is indicated in the delimiter type, and Response delimiter does not have a response expected/not expected indication. PAGE 44 32 The frame structure and protocol of HomePlug 1.0 is depicted in Figure 3-1. CIFS Priority Resolution 0 Priority Resoultion 1 Contention Data RIFS ACK 35.84 us 35.84 us 35.84 us 35.84 us x n 313.5 ~ 1489.5 us 26us 72us Preamble Frame Control Frame Header Frame Body PAD FCS Preamble Frame Control 25bits 17bytes Variable Length 2 Bytes 25bits RIFS 1.5 us End Of Frame Gap Preamble Frame Control Figure 3-1: HomePlug 1.0 frame structure and protocol The HomePlug 1.0 standard uses different terms and stages for inter-frame spacing and for the contention windows than 802.11b. The RIFS shown in the figure is Response Inter-Frame Spacing. Unlike 802.11, there is no SIFS (Short Inter-Frame Spacing) between continued frames. Rather, a frame control bit is used to indicate the desire of a station to continue to send data, allowing preemption only by higher priority traffic. The spacing between the last frame and the incoming frame is CIFS (Contention Window Inter-Frame Spacing). HomePlug 1.0 provides four priority classes CA3, CA2, CA1 and CA0 from highest to lowest. Priority resolution is done by asserting signal of the priority level in the PR0 and PR1 slots. For example, to send a CA2 packet, the PLC device should assert a 1 in PR0, causing any node with CA1 traffic to defer, and not assert 1 in PR1 as it would do otherwise. Nodes with CA3 data assert a 1 in both priority slots and CA0 in neither. This effectively resolves contention between different priority classes. Contention within the same priority class is resolved during the contention period. PAGE 45 33 The contention period is a contention period. The contention period is a form of CSMA/CA with a priority dependent backoff window size schedule. For the lower two priority classes, it is 8-16-32-64 slots, while it is 8-16-16-32 slots for the two higher priority classes. On collision, the range of contention slots over which a transmission is started is increased according to the schedule. Aside from starting with a smaller range (8 slots compared to 32 slots), a major difference from the IEEE 802.11 standard is that when a HomePlug 1.0 node defers (detects another nodes transmission in an earlier slot), it uses this information to back off, but less aggressively than in the case of a collision. This technique serves to reduce costly collisions further. For protocol details, please see the HomePlug 1.0 Specification [3]. Theoretical Performance of 802.11a/b and Homeplug 1.0 To compare the protocol performance of 802.11a/b and HomePlug 1.0, we first analyze the theoretical performance differences between them. In next section the protocol analysis to calculate the theoretical performance of 802.11a/b and HomePlug 1.0 is presented. The analysis assumes that a single station is continuously transmitting frames with 1500 bytes of payload over the medium. 802.11a/b Theoretical Network Performance In the absence of competition, an 802.11b node picks a contention slot between 0 and 31, and starts transmission then. The average contention period delay for a packet without competition is 31/2 = 15.5 slots or 310 sec. The transmitting node will start to send data and wait for the receivers acknowledgment. Each frame is made up of a PLCP header, a MAC header, a DATA field and a CRC field. If it is an ACK frame then the DATA field is not present. PAGE 46 34 In practice, the maximum data payload sent via the 802.11a/b is limited to the Ethernet maximum of 1500 bytes. Table 3-1 summarizes the MAC throughput and efficiency for IEEE 802.11x protocols at various data rates. Packet fragmentation and MAC level packet concatenation are not considered. Table 3-1: 802.11a/b MAC throughput with payload 1500 bytes Technology PHY Data Rate MAC Throughput 1 Mbps 0.91 Mbps 2 Mbps 1.73Mbps 5.5 Mbps 3.99Mbps 802.11b 11 Mbps 6.38Mbps 1 Mbps 0.70Mbps 2 Mbps 1.74 Mbps 5.5 Mbps 3.77 Mbps 11 Mbps 8.08 Mbps HomePlug 1.0 14.1 Mbps 8.08 Mbps 6 Mbps 5.38 Mbps 9 Mbps 7.78 Mbps 12 Mbps 10.02Mbps 18 Mbps 14.1 2Mbps 24 Mbps 17.61 Mbps 36 Mbps 23.74 Mbps 48 Mbps 28.47 Mbps 802.11a 54 Mbps 30.80 Mbps From Table 2-1, the maximum throughput is 6.38 Mbps for 802.11b and 30.8 Mbps 802.11a, representing efficiencies of 58% and 57% respectively. At 1Mbps the MAC efficiency of 802.11b is as high as 91% assuming there is no packet fragmentation. Performance Analysis of HomePlug 1.0 From Figure 2-1, the data transmission time ranges from 313.5 to 1489.5 s, however to transmit a 1500 byte payload at the maximum data rate, 120 symbols are required, taking 1153.5 s. Excluding physical level control overhead results in 1008 s for actual data transmission. The maximum data payload size is limited to the smaller of PAGE 47 35 1500 bytes and 1344R-OH bits, where R is the physical data rate, and OH is the number of overhead bits. Each Ethernet frame incurs an overhead of at least 120 bits for encryption and integrity checking for the corresponding service block. Segment bursting allows a station to send all the segments associated with a service block consecutively, avoiding contention unless it is preempted by a station with higher priority traffic. A service block is broken into physical layer segments, each of which has 19 additional bytes of overhead for addressing and segment control. Additionally, each segment must be a multiple of 20 symbols long, up to 160 symbols maximum, which further complicates throughput analysis. The initial contention window size is 8 slots, so the average contention delay without competition is 3.5 slots. To successfully deliver a data packet of 120 symbols takes 35.84 s + 35.84 s + 35.84 s + 35.84 s 3.5 + 1153.5 s + 22s + 26s= 1484.86s The maximum physical layer data rate is 14.18 Mbps, thus the maximum throughput is 15008 bits/1484.86s = 8.08 Mbps. The efficiency of HomePlug 1.0 at the maximum data rate is 57%. The 70% efficiency of HomePlug 1.0 at PHY rate of 1Mbps is due to the limits of the data transmission time to a maximum of 1484.5s in order to provide better latency and jitter QoS parameters for higher priority traffic. PAGE 48 36 Although we can get up to 8 Mbps maximum MAC throughput in theory, the maximum measured TCP throughput in our field testing so far is 6.3 Mbps, matching earlier HomePlug 1.0 simulation results [12]. Experimental Setup To understand the real world performance of IEEE 802.11a/b and HomePlug 1.0, we conducted field tests in 20 houses located in the Gainesville, Ocala, Orlando, and Belleview areas of Florida. The choice of the houses used in the tests were in the mid-to-large size (1500 sq. ft. to 5000 sq. ft.), since larger houses provide a better range on the performance parameters of interest. The equipment used in this test included the following. 1. AP Server: A Sony notebook with a 700 MHz Pentium III processor and 128k RAM running Windows2000 2. Mobile Station: An HP notebook with a 500 MHz Pentium III processor and 128k RAM running Windows2000 3. Linksys HomePlug 1.0-based Powerline-to-Ethernet bridges 4. Netgear[13] IEEE 802.11b Access Point and PCMCIA Card 5. 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. MIM can be PAGE 49 37 expected for connection from a node to an access point. Since these tests had the AP server connected to the access point via Ethernet, there with no other contention possible, the test results should represent the best case scenarios with respect to this aspect. Experimental 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 outlet or a 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 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 toWindows sockets, was used. The TCP buffer size was chosen to be 11680 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 involves 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, the receiver IP address and the 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 PAGE 50 38 capture was also obtained to observe TCP stability. All the procedures were automated and required minimal human operation. Results The amount of data collected is too large to be presented in detail in this paper, so only a summary of the most interesting findings are presented 3 The location, size and age of each of the houses where testing was conducted is shown in Table 3-2 to 3-4. Table 3-2: List of the houses tested and connectivity Location House Size (Square Feet) House Age (Years) 802.11a Connectivity 802.11b Connectivity HomePlug 1.0 Connectivity Gainesville 1460 4.5 90% 100% 100% Gainesville 2000 5 90% 100% 100% Gainesville 2030 3 75% 100% 100% Gainesville 2100 7.5 100% 100% 100% Ocala 2300 33 42.9% 78.6% 100% Gainesville 2700 4.5 61.9% 100% 100% Gainesville 2700 (two floor) 1 57% 100% 100% Gainesville 2700 10 20% 100% 100% Gainesville 3000 1 91% 100% 100% Gainesville 3000 (two floor) 9 71.4% 100% 100% Ocala 3000 9 54.2% 100% 100% Gainesville 3150 13 55.56% 95.8% 100% Ocala 3500 6 69% 100% 97.6% Ocala 3600 10 41.7% 100% 100% Orlando 3600 4 50% 100% 100% Belleview 3600 4 42% 92% 100% Gainesville 3900 5 56% 100% 100% Gainesville 4000 4 9% 84% 81% Orlando 4200 67 18.75% 50% 100% Ocala 5000 (two floor) 15 17% 50% 100% 3 The complete data set from our field tests is available on request. PAGE 51 39 IEEE 802.11a Indoor Performance The performance and coverage results of IEEE 802.11a are depicted in Figure 3-2. Connectivity versus House Area0%20%40%60%80%100%1400240034004400Area(sq. ft.)Connectivit y Figure 3-2: IEEE 802.11a indoor connectivity Figure 3-2 shows the connectivity (that is, 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 sq. ft.) houses. For larger houses (>4000 sq. ft.) the connectivity decreased to 20 %. Figure 3-3 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 3-4 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 PAGE 52 40 because of manufacturer-specific proprietary enhancements like the use of higher level modulations. Throughput vs Distance0.0010.0020.0030.0040.00020406080100Distance(ft.)Throughput(Mbps ) Figure 3-3: IEEE 802.11a indoor throughput as a function of distance Percentage Of Links versus Throughput0%20%40%60%80%100%01020304Throughput(Mbps)Percentage Of Link 0 s Figure 3-4: IEEE 802.11a indoor percentage of links In summary, the statistics show that PAGE 53 41 1. IEEE 802.11a failed on at least one link in 19 of the 20 houses tested, 2. 802.11a failed to connect in 45% of the links that were tested, 3. 802.11a showed close to zero connectivity at distances larger than 50 feet. For shorter distances, 802.11a provided excellent throughput in most cases. 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 in Figure 3-5 to 3-8. Figure 3-5 shows the connectivity as a function of house area. Both technologies show high connectivity for houses of size less than 4000 sq. ft. For houses larger than 4000 sq. ft., the connectivity for IEEE 802.11b dropped dramatically to 50% in both of the houses tested, while HomePlug 1.0 continued to show high connectivity. Connectivity versus House Area0%20%40%60%80%100%1400240034004400Area(sq. ft.)Connectivit y HomePlug 1.0 802.11b Figure 3-5: IEEE 802.11b and HomePlug 1.0 indoor connectivity comparison Figure 3-6 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 PAGE 54 42 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. Throughput versus Distance0.003.006.009.000204060801Distance(ft.)Throughput(Mbps 00 ) HomePlug 1.0 802.11b Figure 3-6: IEEE 802.11b and HomePlug 1.0 indoor throughput comparison Figure 3-7 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%. The maximum throughput observed in field testing was 5.13 Mbps. Figure 3-7 shows that around 70% of the connections operated at more than 4 Mbps and 10% above 5 Mbps. For HomePlug 1.0, the overall coverage is 98%. The maximum throughput observed in testing was 5.98 Mbps. For HomePlug 1.0, 58% of the connections operated above 4 Mbps and 38% had throughput above 5 Mbps. 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 suboptimal. Second, HomePlug 1.0s ability to adapt to the channel conditions with a nearly continuous selection of data rates allows it to perform PAGE 55 43 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. Percentage Of Links vs Throughput (MIM)0%20%40%60%80%100%0.002.004.006.008.00Throughput(Mbps)Percentage Of Link s HomePlug 1.0 11b Figure 3-7: IEEE 802.11b and HomePlug 1.0 indoor percentage of link versus throughput (MIM) comparison It should also be noted that in this experiment, throughput was measured between the access point to the Mobile Station in a modified Infrastructure mode. 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 Infrastructure mode (IM). In the 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. Further, 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 (that is, the access point is the final destination of the PAGE 56 44 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 Infrastructure Mode (IM) throughput from the MIM link throughput data which was 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 access point to the TCP destination. The aggregate IM throughput then can be obtained by assuming 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 21RxRx second. Thus throughput can be calculated by 2121RRRR 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 3-8 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 compared to IEEE 802.11b stations in IM. From the statistics, we make the following key observations 1. HomePlug 1.0 had a larger maximum throughput that 802.11b(about 1 Mbps larger). 2. On 60% of the links HomePlug 1.0 performed better than 802.11b links in MIM, 3. On an average basis, HomePlug 1.0 gave approximately 0.2 Mbps higher TCP throughput than 802.11b in MIM, 4. In 6 of the 20 houses tested, IEEE 802.11b failed on at least one link, 5. In 2 of the 20 houses tested, HomePlug 1.0 failed on at least one link, 6. On an average basis, HomePlug 1.0 gave approximately 2.3 Mbps higher TCP throughput than 802.11b would be expected to give in IM. PAGE 57 45 In summary, HomePlug 1.0 was found to provide better coverage and slightly better average TCP throughput than IEEE 802.11b in modified infrastructure mode, which would be typical for Internet access. In infrastructure mode, HomePlug 1.0 was estimated to have throughput about 2.3 Mbps 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. Percentage Of Links vs Throughput (IM)0%20%40%60%80%100%0.002.004.006.008.00Throughput(Mbps)Percentage Of Connectivity HomePlug 1.0 802.11b Figure 3-8: IEEE 802.11b and HomePlug 1.0 indoor percentage of link versus throughput (IM) comparison TCP Link Stability QoS algorithms usually deal with admission control and resource allocation. Admission control is concerned with the acceptance of new connections, while resource allocation deals with packet-level throughput, delay, and fairness. In either case, predictability is desirable. PAGE 58 46 Previous studies [14] showed that high channel error rate will reduce the effective bandwidth available for applications, thus negatively affecting application performance. This problem is even more severe for multimedia applications, which typically have bandwidth, delay, and jitter requirements for effective operations; it is important for them that the link remains stationary. However, the 802.11a/b displayed link instability when the PHY data rate was low. This section studies TCP link stability from the realtime capture of 802.11a, 802.11b, and HomePlug 1.0 packets. During testing, when connection speeds were less than 3 Mbps, the wireless network became unstable. This might have been due to problems in rate adaptation. The 802.11b standard indicates around a 4 dB difference in signal strength between 11 Mbps and 2 Mbps mode. Since the signal strength changes continuously with time (for example, due to movement of people), PHY rate adaptation may cause packet drops that make the wireless network unstable when using TCP. During our tests, few links were found with throughputs in the 1 to 3 Mbps range. To find out the cause of this phenomenon, we used real time packet capture to monitor the TCP link stability. TCP throughput was measured at 100 msec intervals. Figure 3-9 and 3-10 shows real time captures of typical high speed and low speed links of 802.11a. High speed links have a mean data rate of 20 Mbps, while it is 5 Mbps for low speed links. Note that the link performance from Mobile station to AP server and from AP server to Mobile station (separated by a 3 second delay) both are shown in the figures. For high speed links, Figure 3-9 shows irregular throughput dropouts during transmission. For AP to mobile link, the throughput differences from one sample to the PAGE 59 47 next can be as high as 22 Mbps. This kind of behavior is exhibited by TCP when packets are dropped. These instabilities would make it very difficult to guarantee QoS. Similar instability was observed for the link from Mobile to AP. A maximum throughput difference of 23 Mbps was observed in this case. Figure 3-9: IEEE 802.11a high speed real-time capture When the data rate of the link was low (Figure 3-10), the throughput also displayed large variations. A maximum throughput difference of 13 Mbps was observed. For Mobile to AP, a maximum throughput difference of 10 Mbps was observed. Figure 3-11 and 3-12 shows the real time capture for a typical high speed and low speed links using IEEE 802.11b. High speed links for 802.11b are links with a data rate over 4 Mbps; low speed links for 802.11b are links with data rate lower than 2 Mbps. The same criteria were applied to the HomePlug 1.0 networks. PAGE 60 48 Figure 3-10: IEEE 802.11a low speed real-time capture Figure 3-11: IEEE 802.11b high speed real-time capture Figure 3-11 shows that the 802.11b links were typically more stable than the 802.11a links. However, there were two dropouts one in the link of AP to Mobile, and the other one is in the link of Mobile to AP as indicated by circles in the figure. This type PAGE 61 49 of dropout often appeared in other captures. In this figure, the maximum throughput differences observed was about 6 Mbps. AP->Mobile Mobile->AP Figure 3-12: IEEE 802.11b low speed real-time capture Figure 3-12 shows the real time capture for a typical low speed of 802.11b link. For AP to mobile link, the throughput differences can be as high as 3 Mbps. For Mobile to AP link, A maximum throughput difference of 3 Mbps was observed in this case. Note that the throughput variation is critical at low data rates. User experience will be poor for links with such throughput variations. For example, under these marginal conditions, a file transfer might halt due to excessive packet drops. Applications can crash or show strange behaviors an extremely unpleasant situation for the user. These links can be considered equivalent to no-connects, in the sense that users will not use these wireless links. The HomePlug 1.0 TCP link stability is depicted in Figure 3-13 and 3-14. The figure shows that HomePlug 1.0 provides a fairly stable TCP link on high speed PAGE 62 50 connections. There are no dramatic dropouts during the test. The maximum variation we observed in the figure was 3.6 Mbps for high speed links. AP->Mobile Mobile->AP Figure 3-13: Homeplug 1.0 high speed real-time capture Figure 3-14: Homeplug 1.0 low speed real-time capture PAGE 63 51 Figure 3-14 shows an interesting stair-wise rate adaptation on the transmission from Mobile to AP. This effect is a manifestation of the channel estimation mechanism. Noise over power lines tends to vary with the line cycle. In HomePlug 1.0 stations, the channel estimation is done asynchronously using a small channel estimation packet. Thus the estimated data rate varies depending on when the packet arrives with respect to the line cycle. Thus different channel estimations (typically done every 5 seconds) will result in a different throughput, and hence this stepped behavior. However, the TCP links can be assumed to be stable between channel estimation cycles. During our testing, we observed that IEEE 802.11b links that operated below 1 Mbps are highly unstable and are frequently marked by disconnects. Discussion and Conclusion The main goal of this chapter is to conduct a practical and theoretical comparison of the IEEE 802.11x and HomePlug 1.0 protocols and their capabilities in providing networking functionalities. This was done through theoretical analysis and by a thorough field test conducted in 20 houses to obtain the performance of IEEE 802.11b/a and HomePlug 1.0 products. From the theoretical results shown in Table I, it can be derived that HomePlug 1.0 and 802.11x have similar maximum efficiency. The significantly higher maximum PHY data rate of 802.11a would indicate that it should perform better than the other two standards, but in field tests its coverage was not as good. In the field tests, the 802.11x products were configured as they came out of the box, and in some cases the wireless links may have used the RTS/CTS (Request To Send/Clear To Send) mechanism needed by 802.11x to handle hidden nodes. Use of RTS/CTS can have significant effects on the performance of the 802.11x protocols. RTS/CTS overhead degrades throughput and is PAGE 64 52 most significant at high data rates, theoretically costing up to 10% in throughput performance. However, use of RTS/CTS can improve performance when there are collisions, as one can expect even on a single link when there are asynchronous, bidirectional exchanges (as TCP does by acknowledging received segments). 802.11x infrastructure mode (IM) also degrades performance, as opposed to modified IM (MIM) or ad hoc mode. In field tests, ad hoc mode was found to be nearly unusable over any but short distances. Fortunately, MIM would be typical for connection to an internet access point, which one might expect to bear the greatest amount of traffic. The wireless protocols have an advantage that these tests cannot show, is their use of multiple channels. With three channels available, one could expect the wireless protocols to perform better under congested conditions, whereas the PLC protocols use all of the available bandwidth for a single channel. Coverage is much harder to predict theoretically, but field tests showed that HomePlug 1.0 had the best coverage, followed by 802.11b, with both trailed significantly by 802.11a. The latter only functioned over line-of-sight distances under 50 ft., and had complete coverage in only one of the 20 houses tested. IEEE 802.11b had severe coverage problems in houses over 4000 sq. ft., and had non-connects or marginal links in more than half of the houses tested. HomePlug 1.0 provided 100% coverage in all but two of the houses tested. This showed convincingly that the absence of an RTS/CTS mechanism in HomePlug 1.0 is not likely to be a problem for single home deployments. Whether or not it is an issue for multiple residences serviced by the same transformer remains to be seen. PAGE 65 53 Throughput showed more interesting behavior. The throughput of IEEE 802.11a was nearly always very high on links less than 50 ft. line of sight, but dropped to zero for links longer than this. 802.11b and HomePlug 1.0 showed crossovers in the percentage of links with data rates meeting some target rate. With its better coverage due to a more robust PHY and MAC, HomePlug 1.0 did better at meeting minimal data rates, but was surpassed by 802.11b at rates around 4 Mbps. For data rates of 5 Mbps and higher, HomePlug 1.0 retook the lead due to its slightly higher maximum data rate. HomePlug 1.0 also showed a much more gradual curve, due to its greater range of PHY data rate selections for adapting to the channel. The continuum of link speeds also allowed HomePlug 1.0 to exhibit greater link stability (as measured by short term variability in TCP throughput). Effects of channel estimation at 5 second intervals and variability of the channel due to the 60 Hz line cycle were evident in these tests. IEEE 802.11b showed greater variability in both lowand high-speed links, most likely due to its sparser choices in adaptation to channel conditions. 802.11a had tremendous and frequent speed fluctuations, which brings into question its ability to offer QoS guarantees for multimedia applications. Both PLC and wireless technology have scope for improvements over the exist standards as evaluated in this paper. For PLC networks, larger bandwidth, the use of higher order modulation, more powerful forward error correction technique, and improved channel estimation can substantially improve performance. Wireless technologies can also invoke similar enhancements along with larger transmit power and antenna diversity to achieve significant improvement. However, increasing crowding in the 2.4 GHz ISM bands used by 802.11b/g are likely to degrade their performance in PAGE 66 54 locations where they must coexist with competing transmitters (including BlueTooth, HomeRF, 2.4 GHz phones, microwave ovens, etc.). Despite its problems, there are still situations for which wireless technologies are needed. Mobile users with handheld devices and nomadic users without access to power outlets will require wireless connectivity. For nomadic users who are able to plug into the home power distribution system, however, PLC offers a robust, stable, and speedy alternative. PLC solutions will be even more desirable for providing QoS support for multimedia applications with the future and emerging PLC technologies offering data rates in excess of 50 Mbps. PAGE 67 CHAPTER 4 PERIODIC CONTENTION FREE MULTIPLE ACCESS FOR POWER LINE COMMUNICATION NETWORKS Introduction Recently, applications over Power Line Communication (PLC) networks have drawn much interest in academe and industry, not only because of their convenience (connecting PLC capable devices requires no new wires), but also because all electrical devices have to connect to a power outlet eventually. This technology makes implementing a digital home entertainment center more realistic than ever HDTVs in different rooms are now able to share digital content from one set-top box without rewiring or setting up wireless access points. MP3 players can access music data through PLC networks from different rooms playing different music. PLC networks make the smart home possible [12]. However, PLC technology is still evolving and many issues remain unsolved. The hostile environment of PLC channels makes reliable data transmission difficult. Much effort is required to ensure data transmission is correct and efficient. The HomePlug Alliance set standards for 14Mbps class data transmission. This resulted in a variety of PLC devices for computer-oriented network communications. Their performance and reliability is comparable to wireless networks [15]. However, the HomePlug 1.0 protocol is not suitable for video playback because of its limited network throughput. To support multimedia streaming for homes, the HomePlug Alliance is now developing the standard for a second generation of PLC devices capable of delivering 55 PAGE 68 56 multiple HD-Videos through newly designed chipsets called HomePlug AV, which supports up to 200Mbps raw data rate. The goal of the HomePlug AV is to make PLC devices capable of delivering two hours of HD-Video without video frame drops, while simultaneously delivering one or more data streams of various data rates and traffic types. To make above mentioned goals possible, the efficient cooperation of high speed PHY and MAC protocols becomes important. However, current existing MAC protocols do not provide such functionalities and they are not suitable for HomePlug AV. The need for a new protocol is urgent. The PLC channel is known for its hostile nature in transmitting electrical signals. Protocols designed for other media may not be suitable for PLC. PLC channels are in some ways similar to wireless channels both of them face hidden node problems, near-far effects and other channel imperfections. However, PLC network nodes tend not to move. It is unlikely that simply applying protocols designed for another medium would result in good performance in the PLC environment; the overhead may be too high or the assumptions about noise may be too optimistic for PLC networks. In light of PLCs unique characteristics, we developed a new protocol Periodic Contention-Free Multiple Access (PCF/MA). PCF/MA is an explicit R-ALOHA-like protocol specifically designed for the PLC network we propose an RTS/CTS-like scheme in the reservation stage to prevent hidden node problems, and a delayed NACK mechanism to conquer near-far effect. Performance of the proposed protocol is evaluated by event driven computer simulation and by mathematical analysis. The simulation results show that 85Mbps MAC throughput under 100Mbps channel data rate can be obtained, even when there are hidden nodes in the network. To provide smooth video PAGE 69 57 delivery, we propose a mathematical estimation of the required delay in playback time and the amount of playback buffer with tight bandwidth reservation. Our simulation shows that an 100Mbps channel can deliver up to 9 MPEG-2 video streams simultaneously without dropping any video frames, however, using Modified CSMA/CA (MCSMA/CA) on the same environment supports 7 video streams only because of its unfairness, unpredictable behaviors and high overhead. This chapter is organized as follows. A brief survey of existing protocols is given in section 2. Section 3 provides the PCF/MA methodology, analysis, and performance evaluation and simulation results. The discussion and conclusion is given in section 4. Previous Works Packet contention techniques such as Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA) and ALOHA find widespread use in data communications, including the first generation PLC networks. Both have the ability to serve a large number of terminals. While they function with little to no coordination, packet contention results in unpredictable behaviors such as unfairness and possibly long delays, which make them unsuitable for delay-sensitive packet delivery. To solve these problems, one can use packet scheduling or reservation-based methods. Reservation-based methods were designed to remove delay-sensitive, high-rate connections from the random-access competition for channel time. The time domain is partitioned into a reservation and data session; each session is then again partitioned into slots. In these methods some slots are reserved for specific stations. Other stations are restrained from using a reserved slot. Generally, these MAC schemes can be categorized according to whether the reservation is done implicitly or explicitly. PAGE 70 58 Packet Reservation Multiple Access (PRMA) is a centralized and slotted multiple access protocol that allows voice and data sources to share the same access channel at the talkspurt level, targeted for wireless local area networks, and was originally proposed by Goodman D. J., Valenzuela R. A., Gayliard K. T., and Ramamurthi B. [16]. PRMA utilizes the speech on-off activity to improve bandwidth efficiency and system capacity. Though PRMA is an implicit reservation-based algorithm, between talkspurts stations must contend for reservation, leading to packet loss and degraded speech quality. It also relies on central control to broadcast an ACK at the end of each slot, hence it may not be appropriate for a home ad-hoc environment. A few modified PRMA protocols such as D-PRMA were proposed to support mobile ad-hoc environments. These emphasize talkspurt-level packet reservation without relying on a central entity. They also try to deal with hidden/exposed node problems by asserting an RTS/CTS-like dialog between sender and receiver. Hidden node and near-far effects make implicit reservation difficult in PLC networks. Impaired receivers may not agree on the current network state. Further, as the PLC network is targeted to be an in-home networking infrastructure, it is not desirable to have a central control device just for medium access, especially if the central control device may become a performance bottleneck and a single failure point of a network. Lack of a central control device makes collision detection even more difficult since PLC devices are not able to detect signal collisions during transmission. Explicit reservations like Five-Phase Reservation Protocol [17] try to implement a dynamic parallel reservation with arbitrary scalable network size. The five phases mentioned in the literature are: Reservation Request (RR); Collision Report (CR); PAGE 71 59 Reservation Confirmation (RC); Reservation Acknowledgement (RA); Packing and Elimination (P/E) phase. However, this protocol makes assumptions inapplicable to the PLC environment. Firstly, the protocol works on a 2 hops radius network, with perfect timing. Secondly, near-far effects do exist in the PLC network especially when a robust signal modulation is applied. The above discussion leads to the new protocol design based on an explicit reservation scheme. Roberts reservation scheme and R-TDMA are good candidates; their merit is that a few slots are dedicated for reservation purpose and the rest of them are for data transmission. However, these schemes lack distributed control, hidden-node prevention and ignore near-far effects, which makes them inappropriate for PLC networks. Careful re-design of Roberts and the R-TDMA protocols results in the new PCF/MA protocol as stated in the next section. Proposed PCF/MA Protocol Though the power line channel is similar to a wireless channel, there are some unique characteristics specific to the power line channel. Firstly, the characteristics of the PLC channel in a home are in general steady over time rather than dramatically changing as in the wireless channel, that is, while the attenuation of the signal may be affected by near-by electronics in the short term, from a long term point of view the attenuation is almost identical. This makes long-term bandwidth scheduling possible. Secondly, the devices in PLC networks are quasi-stationary, that is, the stations in the PLC network are not moving as constantly as devices in the wireless network, thus the bandwidth fluctuations are much less erratic [15]. These two characteristics place PLC networks between Ethernet networks and wireless networks in that they have the properties of a fixed network topology but with more noise and attenuation. PAGE 72 60 The Concept of PCF/MA The PCF/MA protocol is a distributed, contention-free protocol which uses a two-way handshake reservation process to establish TDMA slot assignments. The reservation process for a given node only involves nodes within a one hop radius. Figure 4-1: PCF/MA frame structure Fig. 4-1 shows the protocols frame structure. Time is divided into several TDMA sessions with duration of Ts. Each session is partitioned into a Reservation Frame (RF) and one or more Data Exchange Frames (DEFs). Before the RF is a Starting of TDMA (SOT) frame issued by the leader station on the network that lasts 38.4s 4 Table 4-1 summarizes the parameters used in this research. Distributed Admission Control The RF is divided into R Reservation Slots (RS). An RS does not correspond to a data slot but rather serves as a period of time for the making a reservation. This process involves exchange of Reservation Packets (RP) and ACKs. An RP contains the Source 4 The parameters we used in this paper follows those of HomePlug 1.0 as published in [19] PAGE 73 61 Address (SA), Destination Address (DA), Starting reserved Slot number (SS) and Total of reserved Slots (TS) as depicted in Fig.4-2. To illustrate the reservation process more clearly, we also depict the reservation flowchart in Fig.4-3. Table 4-1: HomePlug 1.0 and PCF/MA parameters HomePlug 1.0 PCF/MA MCSMA/CA SYNC 38.4s 38.4s 38.4s FC 33.6s 33.6s 33.6s RIFS 26s 26s 26s(SIFS) DEL 72s 72s 72s Max MPDU Duration 1.5ms 1.5ms 1.5ms CRS 35.84s 200s(RS) 35.84s DIFS 35.84s 35.84s TDMA Session Length Ts Reservation Slots R SOT 38.4s(SYNC) 38.4s Priority Resolution 35.8s 35.84s In Fig. 4-3, the dashed box represents a state, the rectangle and lattice boxes represent processes and decisions. The oval boxes represent terminals, which lead to other states. Figure 4-2: MPDU process and format PAGE 74 62 If a node wants to reserve a DEF, it first listens to the network for at least one TDMA session. During this period, it monitors the network activities and learns the reservation of each station as depicted in Fig.4-3 CONDIRIONING state. When the RF begins, it chooses a random RS and broadcasts an RP to the networks. Figure 4-3: Reservation process flow chart The destination receives and compares the request with its internal table. If the reservation does not conflict with previous reservations, the node sends a Reservation Acknowledgement (RACK) to the transmitter and completes the reservation process. This process is illustrated in Fig.6.3 IN RECEIVING state. Other stations in the network also listen and broadcast a Reservation Negative ACKnowledgements (RNACK) packet to jam the possible RACK signal if they detect a conflict. To ensure the requester receives the RNACK packet successfully rather than interpret the signal as noise, we use PAGE 75 63 a delayed NACK methodology. We will illustrate this method in hidden node resolution section. If a collision/RNACK happens, the requester will move to backoff stage with random backoff time. The backoff algorithm is the same as in the CSMA/CA, that is, after each unsuccessful reservation, the backoff window size is doubled, up to a maximum value 2 m W, where W equals to (CWmin + 1) and 2 m W equals to (CWmax + 1) and the backoff time is uniformly chosen in the range (0;CW-1). Since PCFMA is a persistent reservation scheme, the competition between stations suppose to be small, thus the maximum window size is set to 16. Once the backoff window size reaches CWmax, it will stay at the value of CWmax until it rests. Data Exchange The data exchange period is divided into N data slots(DSs), each with a duration of Ss. A DEF is composed of several DSs. The length of a DEF is decided by the TS field in owners RP. A DEF always begins at the start of a DS. If a node successfully reserved a period of time for transmission, it listens and waits until its DEF time, then starts to transmit MAC Protocol Data Units (MPDUs). If the receiver receives a successful MPDU, it will send an ACK. The sender can continue until the end of its reserved DEF. At the end of a transmission, the sender broadcasts a Last Frame MPDU to the network. The receiver also broadcasts a Last Frame MPDU to eliminate possible hidden node problems and allow all nodes to update their internal tables. When a station ungracefully terminates its connection, affected stations (usually it is the receiver) also PAGE 76 64 broadcast a Last Frame MPDU during the reserved slots to synchronize internal tables with other stations. MAC Protocol Data Unit MPDUs combine several Ethernet packets belonging to the same path into a jumbo packet to increase overall efficiency. The original Ethernet packet header and trailer is removed and a new small header called PHY Block Header (PBH) containing the sequence number of the original packet is added to indicate the order of the packet. After this process, the original Ethernet packet becomes a new block called a PHY Block (PB). The process is depicted in Fig. 4-2. Several PBs are then combined with the original Ethernet packet sequence into a jumbo packet with a common header to become an MPDU ready for transmission. Based on the current transmission speed, as many Ethernet packets as possible are combined until the transmission duration reaches 1.5ms. Solutions to the Hidden Node Problems and Near Far Effect A robust signal modulation is required due to the attenuation and noise on PLC channels. However, this makes detecting packet collision difficult the signal modulation process may interpret contending packets as noise and remove the noise to restore the intended signal! This phenomenon causes near-far effects and hidden-node problems. A common solution to the hidden-node problem is the use of a RTS/CTS handshake before data transmission. The purpose of RTS/CTS is to notify nearby stations of the incipient data transmission period so that those who are not involved in the data exchange will avoid the channel during that period of time. Signal reception of mobile devices in wireless networks usually suffered from nearby activities, reception is PAGE 77 65 expected to be constantly changing and thus exchanging RTS/CTS packets before every data transmission is required. In PLC networks, we adopt the similar RTS/CTS scheme but only in the reservation stage. In the RF, nodes want to make reservations broadcast RPs to the network. The RP also act as Request to Send signals as in the RTS/CTS scheme. Nodes that receive this packet avoid the reserved period. Nodes outside the broadcasting range may not aware of the reservation resulting in inconsistent databases. If they do not want to make new reservations, the inconsistency becomes irrelevant. If one of the nodes wants to make a new reservation those conflicts with the scheduled reservations, all nodes that hear this RP broadcast RNACK. To deal with the possibility that the RNACK is not heard by the transmitter because of near-far effect, a delayed RNACK scheme is applied as shown in Fig 6.1. The delayed RNACK scheme use Priority Resolution Signal (PRS) to inform the requester the received RACK is incorrect. All stations object to the reservation should wait 98s and broadcast a PRS after the PR packet. The requester should listen to the network after it broadcasts PR packets. If a RACK is received, it should keeps listening to the network for a PRS packet duration (35.84s). If it senses the PRS packet, it determines the reservation is invalid; otherwise it is a successful reservation. If the intended receiver also recognizes the reservation is invalid, it should reply the requester a RNACK with corrected SS and TS fields to make requester database more consistent to the network status. PAGE 78 66 Approximate Performance Analysis and Simulation Results To calculate the maximum throughput, we assume there are always data to send for each node. We define efficiency as the ratio of time spent on transmitting payload to the total time spent on the whole data exchange process. The parameters we used in the calculation is summarized in Table 6.1. The minimum reservation slot time can be obtained by adding a RP duration, two RIFS, a Response, and a PRS which leads to (72 + 26 + 72 + 35.84 + 26)s = 231.84s. A successfully packet transmission process requires a MPDU, an ACK and two RIFSs. The total time required for this process is 1624s when sender sends a maximum length MPDU. The total DAE slots in a TDMA session can be calculated by s29-231.84R-T assume R reservation slots. If there are m transmitters, and the bandwidth is evenly distributed to all transmitters, then a transmitter can have b slots, where msRTb2984.231 slots The total allowed transmitting time for each data stream can be calculated by b s. A sender can transmit p maximum size of MPDUs in a TDMA session, where mssRTp16242984.231 Since the maximum MPDU has duration of 1:5ms and the overhead of a MPDU is 72s, thus the total time spent on data transmission for each node t can be calculated by PAGE 79 67 142816242984.231mssRTt s Since there are m nodes, the protocol efficiency E can be calculated by TmmssRTE142816242984.231 (4-1) If we ignore the floor () operation in Eq.4-1, we can calculate the maximum efficiency Emax as follows: %88when )2984.231(88.0142816242984.231maxmaxETTRTTmmssRTE (4-2) One must note that each DEF starts from the beginning of a DS, if a node does not fully utilize the reserved DS, the resulted efficiency will decrease. From Eq.6.1 we learned that the efficiency of PCF/MA protocol is determined by R, m and T, where m is dynamic. To obtain the optimal performance, we derived a few mathematical forms as follows. Determine Parameter R To minimize fixed overhead, a small R would be desirable considering the number of contenders is relatively small since all reservations are persistent. However, a small R could make the system unstable when the number of contenders increases. PAGE 80 68 It is obvious that when the number of new data streams n is larger than the available reservation slots r, at least two stations will have a reservation packet collision. We call this over-saturation. However, since the reservation session comes every T s, as long as the number of new data streams is less than 1 every TDMA session and at least 1 data stream successfully makes a reservation, the waiting queue should converge. When there are n new data streams, this problem can de described as the probability that at least one station does not conflict with other stations. To determine this probability, we derive the following calculations. The permutations P of choosing n numbers from r numbers is n r P (4-3) The generation function G(x) corresponding to the random choice of n numbers from r numbers with unlimited repetition can be written as rnxnrexxxxGnnnrx where, !)!3!21()(032 We want to know the probability that at least one RP does not conflict with others. That is, at least one node chooses a random number that is different than the others. We can first calculate the permutations p of choosing n numbers from r numbers such that a number is chosen 0 times, 2 times, 3 times...., then calculate (1p)/P to get the desired probability. The generation function of p with unlimited repetition can be written as rxrxexxxg)()!3!21()(32 PAGE 81 69 (4-4) Since we are looking for the coefficient of x n /n!, rearrange Eq.4-4 yields PAGE 82 70 (4-5) The coefficients of x n /n! becomes lnlnlnrlnrlnp)()1(!!0)1( (4-6) From Eq.4-6 and 4-3, we can get the desired probability as nlnln r lnrlnrlnPp)()1(!!110)1( (4-7) To verify formula 6.7, we simulated a constant number of contenders. The simulator simulates one million reservation sessions. The number of contenders is the same from session to session. The simulator counts the event that at least one contender PAGE 83 71 successfully makes its reservation. The number of reservation slots is increased from 2 slots to 9 slots. Simulation results is shown in Fig.4-4 Figure 4-4: Probability of successful reservation Fig.4-4 shows that even r = 2, n = 4, the probability that at least one station successfully makes its reservation is about 75%. When r = 9, even if n = 20, the probability is as high as 90%. To verify our assumption, we simulate a Poisson arrival of =1, r=8, PCF/MA remains stable for one million reservation sessions. When we increase the inter-arrival rate to 3, r have to increase to 16 to make PCF/MA stable. However, 1 new streams for each reservation session is equal to 40 new streams per second if T = 25. We believe 40 new streams per second are more than enough for a home network. From the simulation results, we chose r = 8 for the optimal parameter for PCF/MA. Near-Far Effect Modeling Hadzi-Velkov, Z. and Spasenovski, B. [18] successfully modeled the 802.11b near-far effect under a Rayleigh Fading Channel. It is known that different modulation and PAGE 84 72 transmitting power under different channels will have different near-far effects. To investigate the performance impact with and without the near-far effect, we choose a near-far effect model described as the probability a receiver captures the desired packet (Pcap) in the presence of interfering packets (Ri) THdmiidmiidcapCRRRRRP00when yProbabilit (4-8) where Rd is the received power level of the desired packet and the CTH is the capture threshold. Extra Allocation Since the packet loss rate is high (10 -3 packets per second or higher), a silent retransmission is required when the receiver receives corrupt packets. This requires data streams to allocate extra bandwidth despite the estimated bandwidth. Assume packet loss rate is P and the number of transmission attempts is R. Suppose at one instance, the sender sends N PBs in a MPDU, the amount of over allocation for next transmission is NP. Thus for quasi-error free environment with unlimited retries, we need to over allocate (O) resources by PNPPPPNPNPNPNPNPORR111(1232 (4-9) When P << 1, Eq.4-9 becomes NP PAGE 85 73 We depict the computer simulation and theoretical calculations in Fig.6.5. We simulated 10 6 TDMA sessions to collect meaningful statistics. The simulation results closely match to the mathematical calculation. Figure 4-5: Maximum over allocation PBs various PB loss rate comparisons Data Stream Model Event driven simulation was used to investigate PCF/MA performance and to compare PCF/MA with a CSMA/CA adapted to PLC. The parameters of a PLC network are assigned path by path to simulate real world PLC environment. To estimate the maximum throughput, we applied Always On data stream to both PCF/MA and MCSMA/CA. By Always On data stream we mean the sender has unlimited input buffer and always obtains maximum duration of MPDUs once it has chance to transmit. We assume all channel data rates are 100Mbps and the received power levels from a node to other nodes are randomly generated with Poisson PAGE 86 74 distribution with = 50. There are no hidden nodes in this simulation, the packet loss rate is 0.001 and CTH = 0.3, the simulated results is depicted in Fig.4-6. Figure 4-6: PCF/MA various T versus MCSMA/CA throughput comparisons Fig. 4-6 shows PCF/MA has maximum performance gain over MCSMA/CA about 100% when the number of contenders is large (m60, T = 100). It also shows that PCF/MA with large T not only increases throughput, but also increases the number of possible contenders (m). The reason that T decides m can be inferred from Eq. 4-1 which shows that for a given T, the floor operations cause the useful time of each MPDU to become less as m increases. There are also some dropouts in the PCF/MA throughput chart which are caused by the floor operations. PCF/MA benefits greatly from large T. However, with large T the packet delay also increases. PAGE 87 75 Delay Model To find out the timed average delay of PCF/MA we propose the following model depicted in Fig. 4-7. Figure 4-7: PCF/MA delay model Since PCF/MA divides time into periods each data stream transmits at its reserved period. In Fig. 6.7 we show a stream that has opportunities to transmit at times Z, A, B, C, D, and E. Each transmission lasts for a portion of a frame. The duration is represented as symbol w. Assume a MPDU arrived at time t 1 and departed at time t 2 Assume the system is a steady state system. Thus, the number of MPDUs in the queue at time A is the number of leftover MPDUs, l. Assume y MPDUs arrived during the time from A to t 1 This period of time lasts a portion of a frame represented as symbol x 1 From the above assumption we found that at time t 1 the queue length is the sum of l leftover MPDUs and y arrivals. Assume PCF/MA can deliver MPDUs per frame for PAGE 88 76 current stream. Thus, delivering l+y MPDUs takes r+x 2 frames where x 2 is a portion of w. The delay of the new arrival MPDU thus can be calculated by )())(1(21xErxEd frames (4-10) where E(.) represents the expectation of a variable. During this time all MPDUs in the queue must be processed. This amounts to )())(1()()(21xErxEyElE MPDUs (4-11) Assume the arrival rate is arrivals in a frame. Since the system is a steady state system the inter-arrival time of MPDUs can be treated as a uniform distribution. The expected value of x 1 is 212)1(321)(1 xE (4-12) In a similar way we can calculate E(x 2 ) as wxE21)(2 (4-13) Using queuing theory we can calculate E(y) as 21)()(1xEyE (4-14) From Eq. 4.11 ~ Eq. 4-14 we can obtain r as wlEwlExExEyElEr(21)(1212121)()()()()(21 (4-15) Substituting Eq. 4-15 to Eq. 4-10 we get the average delay PAGE 89 77 21)(12121lEwrd (4-16) Since the service time for each maximum MPDU is the same the system can be modeled as a M/D/1 system, where D means deterministic, we have 221X From the Pollaczek-Khinchin (P-K) formula we denote the residual queue length l as )1(21)1(2)(2222XlE (4-17) where Thus the average delay d becomes )1(22121)1(211222d (4-18) The simulation and analytical results are shown in Fig. 6.8 ~ 6.11. When the reserved duration approaching to the full frame (in this case, w=T), the simulated results closely match to the estimation from M/D/1. However, when the reserved duration is short, the simulated results becomes approaching to the estimation from our analytical model. For long reserved duration, most of the MPDUs arrived during the transmission duration yielding shorter waiting delay. For short reserved duration, however, most of the MPDUs arrived outside the reserved period resulting in a long waiting time. PAGE 90 78 Delay VS Transmission Duration, T=25ms0.00%10.00%20.00%30.00%40.00%50.00%60.00%70.00%80.00%90.00%100.00%0.00%20.00%40.00%60.00%80.00%100.00%Utilization(Lamda/Mu)Delay(Frames) w=1/4T w=1/3T w=1/2T w=1T Analytical M/D/1 Figure 4-8: Analytical and simulated average delay with various w, T=25ms Delay VS Transmission Duration, T=50ms0.00%10.00%20.00%30.00%40.00%50.00%60.00%70.00%80.00%0.00%20.00%40.00%60.00%80.00%100.00%Utilization(Lamda/Mu)Delay(Frame) w=1/4T w=1/3T w=1/2T w=1T Analytical M/D/1 Figure 4-9: Analytical and simulated average delay with various w, T=50ms PAGE 91 79 Delay VS Transmission Duration, T=75ms0.00%10.00%20.00%30.00%40.00%50.00%60.00%70.00%80.00%90.00%100.00%0.00%20.00%40.00%60.00%80.00%100.00%Utlilzation(lamda/Mu)Delay(Frame) w=1/4T w=1/3T w=1/2T w=1T Analytical M/D/1 Figure 4-10: Analytical and simulated average delay with various w, T=75ms Delay VS Transmission Duration, T=100ms0.00%20.00%40.00%60.00%80.00%100.00%120.00%140.00%0.00%20.00%40.00%60.00%80.00%100.00%Utilization(Lamda/Mu)Delay(Frame) w=1/4T w=1/3T w=1/2T w=1T Analytical M/D/1 Figure 4-11: Analytical and simulated average delay with various w, T=100ms PAGE 92 80 Video Traffic The main focus of this section is to discover the performance of PCF/MA in delivering DVD video streams. To do that we build a video traffic model as depicted in Fig.4-12. Figure 4-12: Video traffic model A PLC station acts as a video server. It supports several DVD streams to different destinations. To simplify our discussion we assume that the channel data rate remains stable throughout the whole video display length. We assume the video source is able to supply one video frame immediately after the request is received and that the server will start to transmit MPDUs once it gets permission to transmit. The client moves the received MDPUs to the playback buffer and re-assembles them to a normal video frame before playback. If a video frame is still in transmitting when it is time to display, the video frame is counted as a miss deadline video frame. The client will try to display the video frame 33.3ms later. The miss deadline count keeps incrementing until the video frame is successfully displayed. To smooth video streams, a certain amount of playback buffer (B) is required. When the buffer level is lower than the predefined buffer threshold (BTH), the client sends requests to the server to ask for more video frames. If at time T, we have BT > BTH, the server stops transmitting MPDUs. PAGE 93 81 To make sure the missing video frames are not caused by the protocol in use but by another factor, we investigate the relationships between the playback buffer size at a specific time T (BT ), playback delay (D), average frame size (Va), maximum video frame size (Vmax) channel bandwidth (Bc) and the effective bandwidth (Bw) reserved for this video stream. We classify the way the playback buffer accumulates data into 3 categorizes: Depleting, Accumulating and Balanced. Usually it is the Depleting situation that clients drop video frames. Consider the captured time frame of a DVD movie depicted in Fig. 4-13. At time T, the buffer level at the client side can be calculated by (4-19) TiiwwTVTBDBB301 where Vi is the data size of frame i. If we assume T is relatively large, such that Vi = Va and Eq.4-19 becomes awwTTVTBDBB30 (4-20) Average Video Frame Size in A GOP0100002000030000400005000060000TimeVideo Frame Size(byte s Currently Displaying GOP Currently transferring GOP T Figure 4-13: A large video frame in transmit PAGE 94 82 At time T, the server encounters a Group of Picture (GOP) with large video frames of size Vmax bytes as shown in Fig.4-13. Assume the number of video frames in a GOP is 15, then it takes wBVtmax15 second to deliver this GOP. During this t time frame, the client consumes wBVNmax1530 video frames, that is, the client consumes wBVVN2maxmax450 bytes. Meanwhile it accumulates max15V bytes. Suppose there are G such GOPs, the Eq.4-20 becomes 045015302maxmaxwawwTBVVGTVTBDBB (4-21) In order to prevent video frames from missing their deadlines, we must make Eq.4-21 always larger than zero. To help us analyzing Eq.4-21, we re-arranged it as follows: 03015450max2maxTVBVBVGDBBawwwT (4-22) The term max2max15450VBVGDBww in Eq.4-22 is the accumulated buffer size during the playback delay time plus buffer loss during transmitting large video frame GOPs. The term is the difference of the reserved bandwidth and the average DVD title bit rate. In real world applications, Bw, Vmax and Va are DVD statistics thus can be easily defined. awVB30 To make BT always larger than zero, we conclude the following sufficient, but not required, conditions: 015450max2maxVBVGDBww (4-23) and PAGE 95 83 030 awVB (4-24) If we assume the reserved bandwidth is the video average bit rate, the D can be calculated by 22max2max2max2max2max2)(30)450450()15450(aaaawwVVVGVVVVGBVBVGD (4-25) From Eq.4-23 ~ 4-25 we observed that when reserved bandwidth is equals the average DVD bit rate, the startup delay D should be large enough to absorb the buffer loss due to large video frames. However, when we reserve a large bandwidth and a small playback buffer, the buffer will build up quickly. Once BTH is reached, the server has to stop sending further video frames to prevent buffer overflow. In this case, the term Bw in Eq.4-24 becomes over-estimated. To prevent this problem, a large playback buffer is desirable. Since the minimum playback buffer size should satisfy Eq.4-23 and Eq. 4-24. awVBVBVGB30 and 30max2max (4-26) Since the playback delay D and playback buffer is for the worst scenario in the DVD playback, the G should be large enough to satisfy this situation; thus we set G = 60. To verify our formula, we simulated a video server/client assuming no other data streams on the network where the client has an unlimited buffer, the video parameters is listed in Table 4-2. The efficiency of PCF/MA is about 80% when T = 25ms. The reserved bandwidth for PCF/MA is 9.44Mbps, the observed throughput is 7.45Mbps which is close to average bit rate. We slowly increase delay playback time D, and observe if the simulation drops video frames. If it drops video frames, we record the PAGE 96 84 maximum buffer it accumulates before dropping. The total simulated time is 20 minutes which is about one sixth of the original movie length. Table 4-2: Video traffic parameters Video Average bit rate 7.1Mbps Vmax 36Kbytes Va 30Kbytes In simulating MCSMA/CA protocol, we follow the same procedure except the channel data rate is assigned as 15.5Mbps and the observed throughput is around 7.58Mbps. Simulation results are shown in Fig. 6-14. Figure 4-14: Video playback performance comparisons The simulation results show that without competition, the MCSMA/CA performs slightly better than the PCF/MA (though at worse efficiency). In Fig.6-14, when D = 1, the client drops about 320 video frames and the maximum buffer accumulated is about 2Mbytes. The result shows that the delay time is too short to build up a safe buffer for future video bit rate fluctuation even with unlimited buffer. The figure also shows that the video stops dropping after we increase D to 12 for PCF/MA and 11 for MCSMA/CA. In both cases, the maximum accumulated buffer size is more than 20Mbytes. This result PAGE 97 85 suggests that when the delay playback time and buffer is large enough, both protocols have ample time to deliver large video frames thus no video frames were dropped. From Eq.4-25 and 4-26, we obtain B = 6.48 Mbytes and D = 11 second. The results though different from the simulated results, are acceptable closely. The difference between the simulated and analytical results is because the channel efficiency did not maintain at 80% in the simulation. The problem can be illustrated by the Fig. 4-15. Lots of small video frames Figure 4-15: Capture of a period of the DVD simulation Figure 4-15 shows the DVD video frames in transmit and the time takes to deliver a video frame. Real world DVDs usually have lots of small video frames especially B frames. These small video frames can not fill-up a MPDU result in low channel efficiency. One has to note that Eq.4-25 and 4-26 do not estimate the exact B and D since channel efficiency used is the maximum channel efficiency. Besides, there are too many variations in real world DVD video streams. But Eq. 4-25 and 4-26 gives us an idea of how much bandwidth is required with a given set of video parameters. PAGE 98 86 To demonstrate PCF/MA in delivering multiple video frames at the same time, we conducted multiple DVD server-client simulations. The simulate time is 20 minutes with D = 1 and B = 10000Bytes. The simulation results are depicted in Fig.4-16. In Fig.4-16, the simulation result showed that PCF/MA supports more streams when the channel data rate is high. However, under the same channel data rate, the MCSMA/CA supports less DVD streams. This shows MCSMA/CA suffered from competitions between MPDUs and protocol unfairness. Though PCF/MA is capable of delivering up to 11 DVD streams in theory, the real work DVD streams contains too many small video frames makes channel efficiency much lower than expected. From our simulation, we found the efficiency drops to about 64%! To increase the efficiency, a new streaming protocol that supports striping several small video frames into a MPDU is required. Because of the time constrain, we did not further investigate this issue. Figure 4-16: Multiple video streams comparisons PAGE 99 87 Visualization of Protocol Simulation Data When developing or debugging network protocols, it is necessary to understand both behavior in the aggregate, such as throughput or delay, and specific, pathological scenarios. For the former, graphs are useful ways to present data, but for the latter, it has been more challenging to process the data into a form convenient for human consumption. This section describes a system that represents local area network (LAN) simulation data in the form of time-space diagrams, with station states shown in color. The interface is rich in that it not only allows time scale expansion and compression, but also allows the user to click on specific times and events to obtain detailed information about interesting events. The authors have found the program most useful for understanding and improving various protocols types of LAN protocols. Introduction Traditional approaches to visualization of local area network (LAN) simulation data mainly involve graphing aggregate behavior as a system parameter is varied. This is very useful and has been employed to optimize system settings as well as to select between alternative approaches. Histograms and waterfall charts showing the distribution of behavior classes has also been employed, particularly for determining behaviors relevant to Quality of Service (QoS) metrics. However, these forms of data representation have limitations, especially when attempting to understand detailed behaviors that give rise to the observed performance. When simulating complex protocols such as PCF/MA or MCSMA/CA protocols, it becomes difficult to judge if the simulator works as the desired way using the text based log files. The protocol designers have to look into each text generated by the simulator and analyze the state transition imaginarily. This method works fine when simulating PAGE 100 88 small scale networks or simple scenarios, whereas when simulating large scale networks or complex scenarios, the huge text logs becomes un-manageable. To improve the quality of interpretations of simulation results, Visual Protocol Analyzer (VPA) is designed for efficient analysis of the results of simulations. Goal of User Interface Humans are more sensitive to graphs rather than text. Especially when analyzing complex situations, people usually use drawings to help themselves to reason. For example, Fig. 4-17 shows two modes of the same event transition of a simulation results. People can easily understand the state transition displayed as a diagram on the figure in the right hand side rather than the texts shown on the figure in left hand side. This comparison gives us a clear view of the need of a VPA. To effectively analyze the results of simulations, the desirable VPA should contain the following features Figure 4-17: Text versus graphical event analyzer Easy and efficient to communicate with the analyzer. The way to communicate with the analyzer decides the efficiency of analyzing a simulation result. We choose to use mouse cursor as the major User Interface(UI) because: First, using mouse cursor is an intuitive way of operating modern operating systems Thus people can become more acquainted to the mouse operation. Secondly, PAGE 101 89 using mouse is more effective if the User Interface (UI) is reasonably designed. For example, people do not have to memorize various instructions in order to execute a Save operation. However, we did not completely abandon the usage of a keyboard. It is undeniable that under some circumstances the keyboard is more efficient than mouse operations. For example, to jump to a specific time stamp (if the value of the time stamp is known) simply type the value into an edit box and that would be faster than using mouse to browse to the desired time mark. Easy to zoom in and out. Some events rarely happen except when simulated for a certain amount of time. Thus usually protocol designers will usually simulate a scenario for a long time in order to find out the possible of design flaws. In this case, the time line in the analyzer would become lengthy. Thus an analyzer should be able to zoom into a specific time frame to reveal detailed protocol state transition. It also should have the ability to zoom out to display a global view of the state transitions to find out the correlations between events. To reach this goal, we designed mouse gestures to represent zoom in and out. Designers can use mouse to mark a range of time which represents the time frame for operation. If designers click within the range, then the analyzer assumes it is a command of Zoom In. Otherwise it will zoom out so that the marked time frame occupies half of the window. Designers can easily mark various lengths of time frames to achieve various zoom factors. If designers want a specific zoom factor, they can use the keyboard and type the value into the zoom factor edit box. To prevent ambiguity, the marked time frame is also shown as a text display on the display window. PAGE 102 90 User should be able to mark a specific time. When examining simulation results, designers usually have to look at various time stamps to understand the causes of aberrant events. Without the mark function, it would be difficult to move back and forth between time frames. In VPA system, designers can achieve this effect by simply clicking on the display window. To help designers understand the marked time, VPA also shows the marked time value in the display window. Easy to navigate throughout whole time line. Since the simulation results are usually large, a display window width is not able to display all events at a time. To improve the experiences of browsing events, we use scroll bars to travel between time frames. Since scroll bars are elastic to the mouse movements, designers can drastically move the mouse to achieve large step jumps or click on the end arrow box to achieve small scale time line either forward or backward. The analyzer should be flexible in changing event display colors or event names. Since humans are sensitive to colors, different events should use different colors to represent them. Meanwhile, different protocol designs have different event naming. The VPA should be elastic enough to adapt these minor changes without recompiling the source code. To achieve the above goals, the VPA will read a configuration file before it starts to parse log files. Designers can modify the configuration file to suit their needs. For example, if a protocol has a new state called Preparation, designers can open the configuration file using a text editor, and add a new line of the event name and the associated representing color. The VPA will automatically add the new event into its control panel and in the representing diagram. PAGE 103 91 VPA is not only elastic to the offline event representing modifications, but also elastic to the real time changes. For example, if designers are looking for a specific event, they can turn off other events to make desired event stands out in the final diagram as shown in the Fig.4-18. Check desired events Figure 4-18: Turning off unused events can make desired events stands out The above mentioned VPA features makes using VPA for analyzing protocol events efficient and easy. It can largely reduce the time required in debugging a simulator. Data Presentation Mechanism and System Design As we mentioned earlier, the VPA requires a configuration file for state presentations. The file format is shown in Figure 4-19. PAGE 104 92 Figure 4-19: VPA configuration format The first line in Figure 4-19 shows a state name: INITIALIZATION, the second line shows the color that representing this event. Each statement occupies one line. VPA supports 100 different states and 255 nodes. Users do not have to specify the total number of states in the configuration file. VPA will judge the effective state names and number of states. To translate text log files into diagram forms, VPA has to parse the recorded information. The sample log file format is shown in Fig.4-20. The first line of the log file in Fig. 4-20 defines the total number of nodes in the simulation. Starting from the second line, it contains the state transitions of each node of each time stamp. Whenever a node changes its states, the protocol simulator will record the transition. Other nodes that are not involved in the state transition is not recorded in order to reduce the file size. The protocol simulator first records the event time stamps, then record the states of each node. If the state of a node remains unchanged, the protocol simulator skips that node. PAGE 105 93 Figure 4-20: Simulator log format The tokens are defined in the configuration file. When VPA sees a text that is equal to a token, the index of the token is stored into an internal state queue associated with the starting time stamps. If a text is not recognizable during parsing, the VPA will skip that text. The VPA will keep parsing every text in the log file until the end of the file. When a token is recognized, the ending time stamp of the previous state is also decided. After a whole file is parsed, the VPA will start to display each states with the colors defined in the configuration file. Since the window height and width is limited and is usually smaller than the simulation duration, the real data displayed is a small portion of the whole data as shown in the Fig.4-21. To help designers browse, a Zoom In and Zoom Out command is required. To implement Zoom In and Zoom Out, we need to calculate the Zoom Factor. Initially the Zoom Factor is defined as one. Before calculating the zoom factor, the VPA PAGE 106 94 first translates the marked time frame to coordinates of the display. The translation can be done by Eq. 6.29. Actual display area (x1,y1)(x2,y2)Max_timeMin_timeDisplay Windows WidthDisplayWindowHeight Mark_Time1(M1) Mark_Time2(M2) Figure 4-21: The actual display window is a portion of the whole data FactorZoomtimeMinTimeMarkMFactorZoomtimeMinTimeMarkM______2211 (4-27) If the mouse gesture is a Zoom In command, that is, user makes a mouse click within the marked time frame, the zoom factor can be calculated by dividing the current display time frame with the marked time frame as shown in Eq. 4-28. 1212_MMxxFactorZoom (4-28) If the command is a Zoom Out, the zoom factor is calculated by dividing the marked time frame with the current display time frame as shown in Eq. 4-29. 1212_xxMMFactorZoom (4-29) Since the simulation log file is huge, we need a proper data structure to represent state transitions. An event queue is used in VPA. The queue is dynamically allocated and de-allocated in order to minimize the system memory requirement. To reduce window flickering due to the updating of display contents, we record the areas that are affected by the current operation and update those areas only. This way, only a small portion of the display areas needs to be updated at a time. PAGE 107 95 Experience Using the Visualization System Because of the help of the VPA, we largely reduced the protocol design process and the time spent on debugging the simulation result. For example, in the process of designing the MCSMA/CA protocol, we found that searching for the causes of nodes moving to the backoff state using the text based log file is painful since there are so many different state transitions occurring at the same time. However, using VPA we can easily find the cause using the displayed diagram. In Fig. 4-18, we found the cause of the backoff is because of the RTS collisions. The experience of using VPA is positive; it is efficient and intuitive to operate. PAGE 108 CHAPTER 5 FUTURE WORK AND CONCLUSION Future Work Though the PCF/MA protocol proved to be high performance protocol, there is much work to be done. Currently, we do not implement priority classes and treat the incoming queue length of the PCF/MA protocol as unlimited. We will address these issues in the future publications. We expect to see higher data rates in power line networks in the future as signal modulation technologies improve; however, issues like network security and the network characteristics with a large number of nodes need further development. While we expect the IA devices communicating with other devices on the Internet, the behaviors of these data flows remains unknown. Further research on these issues is of critical importance when power line networks are applied to offices and large multi-user buildings. Currently we are in the process of investigating the above mentioned issues. Next section, we investigated the factors affecting the Voice over IP quality as the pioneer study of our future work. Voice over IP The ultimate goal of VOIP is to provide the communication quality closed to one provided by (circuit-switching) conventional phone systems [20, 21, 22, 23, 24]. However, since the voice conversations are digitized, compressed and packetized into IP packets, to guarantee the same level of QoS as conventional phone networks is challenging. There are many factors that affect the overall end-to-end Internet Telephony 96 PAGE 109 97 delivery. For instances, many background traffic (not necessarily voice packets) can arrive anytime from different places. Since IP-based Internet will accept any packets with the best effort service, these background traffic packets will affect the voice packets intended for Internet Telephony. We are interested in exploring these factors that influence the end-to-end Internet Telephony QoS. Note that these factors certainly will be affected by the compression/decompression algorithms that the system has adopted. Therefore, we provide two tracks of experimental study with either PCM or GSM schemes. We intend to investigate these factors in a simple (while realistic) environment. In order to present the average performance trend, we also decide to investigate the performance results without a dedicated network. Thus, we hope to answer the ultimate question on what QoS should be accomplished in order to satisfy the quality requirement in an experimental setting. This chapter presents the preliminary results from our first set of ongoing investigation. The study was wholly performed by real experiments, thus presenting valuable data/information to be shared with research community. The rest of this section introduces these factors in short while the next section demonstrates the experimental results in greater details. Distance Factor Many reasons with the distance can affect the quality. Without a dedicated network, distance influences voice quality (at least) in the following aspects: Because of the long distance, usually conventional 56-Kbps (or less) modems are the usual devices used by the average users. Therefore physical bandwidth, though sufficient, is always limited. PAGE 110 98 Usually modem connections need to communicate with ISPs modem server. In our case, students access the general modems server provided by the university. Since this line has been open to all students, the load on the modem server can be varied, thus further affecting the quality. The total number of routers can affect QoS in a significant degree. The probability of congestion (thus extra delay) is proportional to the number of routers packet traveled. When voice packets travel through the Internet, the packets may be randomly dropped. Therefore, the probability that a voice packet has been affected can be directly related to the distance it travels. Our preliminary results indicate a consistent performance trend. By designing the experiments carefully, we found that the longer the distance the worse the service quality. Depending on the compression codec used, the achieved-bit-rate 5 measured on end system applications (that is VAT) affected by the long-distance communication can be in the range between 40% and 87%, even only within 10 miles. Some proprietary products (for example, Net2Phone) did optimize the VOIP software in some degree. However, the know-how still remains to be confidential. Our goal, as academic researchers, is to use the public-domain source code as the research vehicle for investigate different factors (in this chapter) and propose solutions (in the future publications). We intend to publish the optimization methods and open-source solutions once they become available. Congestion Degree In the classical definition, network congestion occurs when the combined bandwidth of all flows destined for a given output link exceeds the links capacity. When congestion occurs, the network (mostly switches and routers) discards packets randomly to reduce the congestion. Our recent study [25] for ATM networks indicates that 5 We have added the time probes into the open-source VAT software. The achieved-bit-rate is thus collected as the sustained rate in the application level. PAGE 111 99 congestion is indeed a great factor to influence the end-to-end multimedia delivery over TCP and UDP protocol stack. Performing these kind of experiments over ATM switches and high-speed routers requires a total control over these equipments. Therefore, we are in the process of working with university authorities for performing these similar experiments in the near future. Meanwhile, we will define the congestion in a slightly different manner. Our voice streams typically go through BellSouths PSTN, Universitys shared modem server, and all-traffic-combined shared campus backbone as part of the Internet infrastructure. Since we did not have a total control over this Internet infrastructure, we defined the congestion degree as the experienced background traffic during either day or night time. In order to rule out the possible congestion on a single modem pool server, separated modem servers were used by the caller and colleen. It is reasonable to assume that the background traffic in the night (for example, 4am) will be significantly less than the daytime (for example, 12pm at noon) for the university setting. The typical total traffic at University of Florida around 4am can be as low as 5 Mbps, while as high as 34 Mbps around noon. Certainly we expect the achieved quality will be better when the background traffic is light. Our experimental results revealed that a 39% to 45% bit rate improvement can be achieved with light load at night compared to heavy load in daytime. Priority Factor It has been recognized [26, 27] that priority support and specification are essential for multimedia communication and control systems. Priority schemes can be designed and enforced at both the end systems (that is, operation systems or middleware), and switch/routers in the middle of delivery paths. PAGE 112 100 Priority specification and enforcement at the end systems seem to be a solution that can be designed and implemented without much difficulty. Since our goal is to provide the voice communication with a quality close to the PSTNs dedicated circuit, higher priority should be assigned to (digitized) voice packets. Other IP data packets from conventional applications (for example, E-mail and FTP) should be assigned as low priority. Certainly the other side of this issue is how to enforce the priority schemes. The software-VAT we investigated does implement RTP (Real-time Transport Protocol) in it. One of the most important features of RTP is that it can provide flexible de-multiplexing. It is defined as application-level framing [28, 29]. Basically RTP uses a range of ports to represent different priorities. The VAT (on top of RTP) provides three priorities: high, medium and low priority at the end stations for the voice streams. VAT/RTP then uses different ports for transmitting and receiving the packets in different priorities. High-priority packets will be served first before low-priority packets (by scanning the ports). Through this chapter, we intend to provide an integrated observation which offers design guide-lines for system designers. The overall recommendations from our performance study seem to indicate that a range of 15-Kbps and 50-Kbps produces acceptable or satisfactory quality using PCM encoding scheme. By using the GSM encoding scheme, the range can be between 6-Kbps and 15-Kbps for acceptable or satisfactory quality perception. Related Study Though the great significance of Internet Telephony has been clearly identified in recent years, very few experimental results have been published in the existing literature. Many draft standard recommendations are available. Toga and Ottv [30] did a good PAGE 113 101 overview of these inter-related standards. The VAT/RTP is one of the leading software solutions that is compliant with these emerging standards. To the best of our knowledge, this chapter perhaps presents one of the first few research results that investigate the impact of different factors in an experimental setting. For instance, Kostas [21] discussed mean delay from different locations in USA and the relationship between delay and hop count. Goyal [31] discussed discussed calling signaling and resource management. Schulzrinne [32] mainly focused on the signaling protocol to initiate a voice session. Rosenberg [33] introduced a unique aspect for programming languages to support Internet Telephony. None of these studies addressed the experimental results by jointly considering distance, congestion and priority factors. Therefore, it is very difficult for us to compare our work with others. The most related work that we could find is perhaps the recent work from [34], but only simulation results were reported instead of real experiments. We not only measured the achieved end-to-end bit rates, but also measure the quality in term of human perception. Experimental Performance Results We have adopted the VAT [35] from LBNL (Network Research Group of Lawrence Berkeley National Laboratory) in order to collect experimental performance in a pure software Internet Telephony setting. VAT is based on IETF RTP and is for real-time, multi-party audio conferencing over the Internet. Since our university environment does not encourage multicast and conferencing, therefore support for IP Multicast 6 and IP 6 6 Many departments choose to disable the MBone routers to avoid the large amount of multicast traffic flowing into their departmental networks. PAGE 114 102 Multicast Backbone (MBone) is not possible. Thus we use standard unicast IP addresses for point-to-point audio chat. Source code was down-loaded, and recompiled over Linux for PC and Solaris 2.6 for Suns Ultra workstations. Sun workstations were connected to the campus Ethernet networks, and PCs were connected with 56-Kbps modem at students home apartments. All other applications (for example Netscape) were terminated to keep a clean environment and no background network services were provided on each end system. We then performed the experiments in a systematic manner. Every experiment was performed a few times to ensure that a steady performance trend can be observed. In addition to the objective measurement of actual bit rate at the host, we also had a few participants to subjectively measure the voice quality. The subjective measurements were recorded down as one of the following results: (Satisfactory, Acceptable, and Unsatisfactory). Satisfactory was the judgment that the achieved quality was excellent for comprehension. Acceptable was the indicator that, though the achieved quality was not excellent, comprehension was still reasonable. Unsatisfactory stands for the situation that the archived voice quality was too bad to comprehend the conversation. Note this subjective measurement technique is consistent to emerging industrial subjective performance measurement. The only difference is that we do not convert them into a number system to calculate the performance average for acceptance. Our measurement requires a majority of reviewers to agree on the resulted recommendation, thus presenting the same degree of consistent judgment. We first presented the effect of distance on the QoS of IP telephone. It was performed in the midnight and packets were assigned with high priority. In this PAGE 115 103 experiment, interesting results were observed. We then fixed distance factor for investigating performance results under congested situation. We then fixed the traffic factor for examining the effect or priority specification. Results Based on the Distance Factor In order to isolate the effect of distance from other factors, the experiments were performed at midnight. We observed that the amount of background traffic was significantly less at midnight compared to the daytime. We also fixed the codec scheme using PCM mechanism first, and then extended to GSM. Short-distance experiments were performed at the campus within a one-mile distance. Longer-distance experiments were performed between two students home apartments, which usually ranged from 3 to 7 miles. Table 5-1: Performance comparison for distance factor Achieved Bit-Rate(Kbps) Codec Used Distance Covered Subjective Evaluation Min Mean Max PCM short Satisfactory 33.5 48.2 59.2 PCM long Satisfactory 24.5 34.4 45.2 GSM short Satisfactory 13.7 14.8 15.8 GSM long Acceptable 7.5 7.9 8.2 By using PCM codec, it can be observed in Table 5.1 that a higher throughput can be achieved by using short-distance communication. The average sustained bit rate was about 40% higher (that is, 13.8 Kbps) than the longer-distance Internet telephony. Note that PCM only samples up to 64-Kbps, thus higher rates are not possible (though LAN connections were used). In addition, the performance difference was also jointly caused by the hardware capability. Long-distance communication was mainly supported by 56-Kbps modems. Nevertheless, experiments using PCM codec demonstrated a satisfactory performance from human judgment independent of the distances. PAGE 116 104 On the other hand, GSM specification usually adopted AMR (Adapted Multi-Rate) encoding, which generates up to 22-Kbps compressed voice streams. We did not use any silence suppression. Our experimental results showed that an average 14.8 Kbps voice stream was measured in LAN environment, while only an average of 7.9 Kbps in longer distance. Therefore, the experimental gain from short-distance communication over long-distance was about 87% (that is, 6.9 Kbps). Note that the potential impact by the distance factor seemed to be significant with GSM coding scheme. For instance, with a longer distance, the subjective quality has been reduced from from Satisfactory to Acceptable. We are in the process to investigate the reason(s) why overall quality of GSM codec suffered in such a degree when distance was long. Baldi M., Risso F., and Torino P., [34] observed a similar phenomenon (though from simulation). The relationship between packetization delay and the achieved bandwidth has been addressed. It has been found that when packetization delay was more than 18ms, with more bandwidth allocation was required in order to keep end-to-end delay below QoS requirement. Our early evidences indicate that the GSM codec in VAT/RTP has a packetization delay with the range of 80ms. We will verify these possible reasons through experiments in the future. It also can be observed in Table 5.1 that the high-bandwidth network within the campus was not able to further increase the sustained bandwidth beyond 60 Kbps. We believe that it was caused by the PCM encoding since it only can support maximally 64 Kbps. Since we use software solution, the system overhead increased a certain percentage (for example, about 4 Kbps in this case) because of the interaction between encoding and network transmission is not good enough. In summary, it is shown by our experimental PAGE 117 105 results that a range of 15-Kbps and 50-Kbps is considered to be satisfactory for voice communications (that is, similar to carrier-level quality with 64-Kbps circuit-switching networks). Results Based on Congestion Degree Congestion degree from the background traffic in the Internet will certainly affect the overall end-to-end performance. The more background traffic, the less achieved bit rate will be. It is because the fixed network bandwidth is shared by all the data flows. In order to perform the comparisons, we ran the experiments separately at midnight and daytime (for example, noon). We expected that the background traffic within the University of Florida in the day time will be significantly higher than night time. Table 5-2: Performance comparison based on congestion factor Achieved Bit-Rate(Kbps) Codec Used Degree of Congestion Subjective Evaluation Min Mean Max small Satisfactory 24.5 34.4 45.2 PCM large Acceptable 19.8 23.7 31.2 small Acceptable 7.5 7.9 8.2 GSM large Acceptable 5.1 5.7 6.2 Table 5-2 lists the performance comparisons between these two experiments. In here, we included the previous results with long-distance voice communication for comparison. We believe, with a longer distance, the background traffic within the campus and all channels connected to it will affect the voice quality in a significant degree. Therefore, all the results were collected over long-distance 56-Kbps modems. The PCM results clearly indicate that the quality of sound has been degraded from Satisfactory to Acceptable because of the background traffic. While an average 34.4-Kbps stream was achieved in the night time (that is, much less background traffic), only 23.7-Kbps was achieved in the day time. The mean bit rate achieved in the night time was as high as 45% compared to the mean bit rate in the daytime. Thus, for domestic Internet PAGE 118 106 phone conversations, the results suggested users should take advantage of the night times if no dedicated circuits were used. The GSM results showed a similar effect from the congestion. An average 5.7 Kbps was achieved in the daytime. Therefore with GSM codec, 38.6% more bit rate was achieved in the night time. Compared with PCM, the potential effect of background traffic seemed to be less influenced with GSM. This was an interesting finding, and we were in the process to further identify the exact reason(s) for this observation. Nevertheless, the subjective measurement has been rated as Acceptable though a lower bit rate was achieved in the daytime. Results Based on Priority Factor Since we do not have the RSVP or WFQ-enable routers within the campus, the specification will only affect the sending and receiving ordering within the hosts. High-priority specification indicates that this voice stream should be served as urgently as possible. On the other hand, low-priority specification on a voice stream will potentially delay or drop the stream within the hosts (either at the sending or the receiving hosts). The results that we presented in the previous subsections were specified as high-priority voice streams. In order to demonstrate the counter effects for using low priority specification, we also performed an extra set of experiments in the daytime with long-distance communication. The experiment results are shown in Table 5-3. The VAT/RTP priority scheme proved to be useful when the degree of network congestion degree was high. Given the usual amount of background traffic in the day time, the voice stream decreased to average 0.6 Kbps with low-priority setting for PCM codec. The achieved subjective quality was Unsatisfactory for human judgment. PAGE 119 107 Therefore, it appeared that we should use high-priority specification for VOIP applications all the time. Table 5.3: Performance comparison based on priority factor in day time Achieved Bit-Rate(Kbps) Codec Used Priority Subjective Evaluation Min Mean Max High Acceptable 19.8 23.7 31.2 PCM Low Unsatisfactory 0.4 0.6 1.0 High Acceptable 5.1 5.7 6.2 GSM Low Unsatisfactory 1.4 1.7 1.9 The voice stream with GSM codec again indicated a similar trend by only achieving 1.7 Kbps experimentally. The subjective measurement was evaluated as Unsatisfactory. Note that GSM was particularly designed for low-bandwidth communication (for example, cellular phone communication). The results thus suggest that perhaps an even more sophisticated encoding scheme beyond GSM is needed for the bit rates less than 2 Kbps. Though interesting, this research topic is beyond the scope of the chapter. Conclusion It took many decades for scientists and engineers to optimize the PSTN until it was widely used by people as a life commodity. Using Internets packet switching for supporting concurrent voice streams with guaranteed quality is still in the early stage. Many environmental factors still affect the end-to-end quality. Only when IP telephones provides as good QoS as PSTN does (that is, much less influenced by the environmental factors), then we can expect it to be widely used as another commodity to the general public. Last section presented our preliminary results on investigating the performance impact from three common factors. Without dedicated networks and QoS enable schemes, PAGE 120 108 the factors of distance and congestion degree were proved to be influential on the achieved end-to-end bit rates. There are still many research issues need to be solved in order to guarantee (at least statistically) the quality of Internet Telephony. This study proposes a new protocol PCF/MA for high speed PLC networks. To lower overhead caused by contention between MPDUs, we choose a contention-free method. Through simulation, we observed the network efficiency as high as 85% at T = 100. Theoretical analysis found that with eight reservation slots, it can provide a good contention/overhead balance. We also conducted a simulation with constant contenders to verify our analysis. For comparison purposes, we modified the widely used CSMA/CA protocol into a PLC version MCSMA/CA protocol. The simulation results show that PCF/MA protocol has the maximum performance gain of 100% over MCSMA/CA when the T = 100. Our protocol also proved to be able to support more than 200 streams at the same time. The high speed PLC is targeted for Audio/Video applications over local area networks, however, the parameters for clients to support variable bit rate data streams is remain uncertain at this stage. We analyzed the behaviors of PCF/MA protocol in delivering commercial DVD titles, and derived relationships between required buffer, reserved bandwidth, delay playback time and video statistics. We then used the estimated parameters to design a video server/client simulator in which PCF/MA and MCSMA/CA as MAC protocols were used for comparisons. The simulation results showed that PCF/MA performed much better than MCSMA/CA protocol because of its predictable behavior and low contention/overhead. PAGE 121 109 The emergence of Information Appliances (IA) for the smart homes of the future will undoubtedly make our lives much more comfortable than ever. However, the infrastructure that supports multimedia traffic and conventional elastic data traffic for communication among IA devices is a critical component of a smart home. We advocate power line as the infrastructure for smart homes based on the convenience of the power sockets and the layout of the power line network existing in every home. The ultimate goal of PLC network could be the ability to connect to the Internet without dialing up to an ISP server, entirely using electrical wiring only. PAGE 122 LIST OF REFERENCES 1 Frank E.H., Holloway J.,: Connecting the home with a phone line network chip set. IEEE Micro, pp.27-37, 39, Vol. 20, Issue 2, Mar.-Apr. 2000 2 Hughes S, Thorne D.J.,: Broadband in-home networks, BT Technol Journal, pp.71-79, Vol. 16, Issue 4, Oct. 1998 3 Intellon: Home of HomePlug Power Line Network Technology, 2001, Intellon, http://www.intellon.com, Dec. 2001 4 Simon M.K., Alouini M-S.,: A unified approach to the probability of error for noncoherent and differentially coherent modulations over generalized fading channels, IEEE Trans. Commun., pp. 1625-1638, Vol 46, Dec., 1998 5 Brown John S.,: Physical Multipath Model for Power Distribution Network Propagation, Proceedings of International Symposium on Power-line Communications and its Applications., pp. 76-89, 1998. 6 Lim C.K., So P.L., Gunawan, E., Chen, S., Lie, T.T., Guan, Y.L.,: Development of a test bed for high-speed power line communications International Conference on Power System Technology, 2000., pp. 451 -456, Vol. 1 2000 7 Liu Weilin, Widmer H.-P., Aldis J., Kaltenschnee T.,: Nature of power line medium and design aspects for broadband PLC system 2000 Proceedings. International Zurich Seminar on Broadband Communications., pp: 185 -189, 2000 8 Karl M., Dostert K.,: Selection of an optimal modulation scheme for digital communications over low voltage power lines IEEE 4th International Symposium on Spread Spectrum Techniques and Applications Proceedings, 1996., pp: 1087 -1091, Vol.3, 1996 9 Romans C., Tourrilhes J.,: A medium access protocol for wireless LANs which supports isochronous and asynchronous traffic, The Ninth IEEE International Symposium on Personal, Indoor and Mobile Radio Communications, pp: 147-152 Vol.1, 1998 10 HomeCNA,: HomeCNA Standards and Protocols, 2002, HomeCNA, http://www.caba.org/standard/homecna.html, July, 2002 110 PAGE 123 111 11 IEEE 802.11 Tutorials,: IEEE 802.11, 802.11a, 802.11b Tutorials and Introductory Information, 2003, PaloWireless, http://www.palowireless.com/i802_11/tutorials.asp, Jan, 2003 12 Lin Yu-Ju, Latchman Haniph A., Lee Minkyu, Katar Srinivas,: A Power Line Communication Network Infrastructure for The Smart Home, IEEE Wireless Communications, Vol. 6, pp. 104-111, Dec., 2002. 13 Netgear,: Home, 2002, Neatgear, http://www.netgear.com/, July, 2002 14 Ng T.S.E., Stoica I., Zhang H.,: Packet fair queuing algorithms for wireless networks with location-dependent errors, INFOCOM Seventeenth Annual Joint Conference of the IEEE Computer and Communications Societies. Proceedings. IEEE, Vol. 3, pp. 1334-1340, Apr., 1998. 15 Lin Yu-Ju, Latchman Haniph A., Newman Richard E., Katar Srinivas ,: A Comparative Performance Study of Wireless and Power Line Networks, IEEE Communications Magazine, pp. 54-63., Apr. 2003, 16 Goodman D. J., Valenzuela R. A., Gayliard K. T., Ramamurthi B., :Packet Reservation Multiple Access for Local Wireless Communications, IEEE Transactions on Communications, Vol. 37, No. 8, pp. 885-890, Aug. 1989,. 17 Zhu Chenxi and Corson M. Scott,: A Five-Phase Reservation Protocol(FPRP) for Mobile Ad Hoc Networks, Seventeenth Annual Joint Conference of the IEEE Computer and Communications Societies. Proceedings. IEEE INFOCOM Vol. 1, pp. 322 331, Mar., 1998 18 Hadzi-Velkov Z. Spasenovski B.: Capture effect in IEEE 802.11 basic service area under influence of Rayleigh fading and near/far effect, The 13th IEEE International Symposium on Personal, Indoor and Mobile Radio Communications Vol.1, pp. 172 176, Sept. 2002 19 Lee M. K., Newman R. E., Latchman H. A., Kartar S. and Yonge L., :HomePlug 1.0 powerline communication LANs protocol description and performance results, International Journal Of Communication Systems 2003, vol. 16, pp.447-473, 2003 20 Huitema C., Cameron J., Mouchtaris P., Smyk D.,: An Architecture for Residential Internet Telephony Service, IEEE Internet Computing, pp.73, 1999 21 Kostas T., Borella M., Sidhu I., Schuster G., Grabiec J., Mahler J., : Real-Time Voice Over Packet-Switched Networks, IEEE Network, pp. 18, Jan., 1998 22 Latchman H.,: Computer Communication Networks and the Internet, McGraw-Hill Companies, Inc., 1998. 23 Rizzetto D., Catania C., : A Voice over IP Service Architecture for Integrated Communications, IEEE Network, pp. 34, May, 1999 PAGE 124 112 24 Schulzrinne H.,: The IETF Internet Telephony Architecture and Protocols, IEEE Network, pp. 18, May, 1999. 25 Tsang R., Du D., Pavan A.,:Experiments with Video Transmission over An ATM Network, ACM/Springer Journal of Multimedia Systems, Vol. 4, pp. 157, Aug. 1996. 26 Guha A., Pavan A., Liu J., Rastogi A., and Steeves T.,: Supporting Real-time and Multimedia Applications on the MERCURI Testbed, IEEE Journal of Selected Areas in Communications (JSAC), Vol. 13, No. 4, pp.749-763, May, 1995. 27 Pavan A., Liu J., Guha A., Pugaczewski J., : Experimental Evaluation of Real-Time Support on the MERCURI Wide Area ATM Testbed, Proc. of IEEE Local Computer Networks Conference, Minneapolis, MN, pp. 82-91, Oct., 1995 28 Clark D. D. and Tennenhouse D. L., : Architecture Considerations for A New Generation of Protocols, Proceedings of ACM SIGCOMM Conference, Philadelphia PA, pp. 200-208,1990 29 Schulzrinne H., Casner S., Frederick R., Jacobson V.,: RTP: A Transport Protocol for Real-Time Applications, RFC 1889, Internet Engineering Task Force, 1996. 30 Toga J., Ott J.,: ITU-T Standardization Activities for Interactive Multimedia Communications on Packet-based Networks: H.323 and Related Recommendations, Computer Network, Vol. 31, pp. 205, 1999. 31 Goyal P., Ramakrishnan K.,: Integration of Call Signaling and Resource Management for IP telephony, IEEE Network, pp. 24, May, 1999 32 Schulzrinne H., Rosenberg J.,: Internet Telephony: Architecture and Protocols An IETF Perspective, Computer Networks, Vol. 31, pp. 237, 1999 33 Rosenberg J.,: Programming Internet Telephony Services, IEEE Internet Computing, pp. 63, 1999 34 Baldi M., Risso F., Torino P.,: Efficiency of Packet Voice with Deterministic Delay, IEEE Communication Magazine, pp. 170-177, May 2000 35 Jacobson V., McCanne S.,: LBL Audio Conference Tool (vat), 2000, Lawrence Berkeley National Laboratory http://www-nrg.ee.lbl.gov/vat/, Oct., 2000. PAGE 125 BIOGRAPHICAL SKETCH Yu-Ju Lin was born in Taiwan. He received the Bachelor of Engineering degree from National Central University, Taiwan in 1990;and the MS degree in computer and information engineering from Chung-Yuan Christian University, Taiwan in 1995. He enrolled in the Ph.D. program at the University of Florida in 1998, doing research on power line communications and multimedia under the direction of Dr. Haniph A. Latchman. His research interests include multimedia communication and computing, power line communication, and high-speed networks. He graduated in May 2004. 113  Permanent Link: http://ufdc.ufl.edu/UFE0004387/00001 ## Material Information Title: High-Performance Periodic Contention-Free Multiple-Access Protocol for Broadband Multimedia Powerline Communications 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: UFE0004387:00001  Permanent Link: http://ufdc.ufl.edu/UFE0004387/00001 ## Material Information Title: High-Performance Periodic Contention-Free Multiple-Access Protocol for Broadband Multimedia Powerline Communications 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: UFE0004387:00001 ## This item has the following downloads: Full Text HIGH-PERFORMANCE PERIODIC CONTENTION-FREE MULTIPLE-ACCESS PROTOCOL FOR BROADBAND MULTIMEDIA POWERLNE COMMUNICATIONS By YU-JU LIN 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 2004 Copyright 2004 by Yu-Ju Lin This is dedicated to Parents and to Yen-Li Sun ACKNOWLEDGMENTS I thank the Department of Electrical and Computer Engineering for having the best teachers. I would also like to thank LIST lab members for their support and feedback. It was a pleasure to work with them. I also thank Intellon Corp. (Ocala, FL) forgiving me opportunities to work with them and learn about powerline communication from the inside. I especially thank Srinivas Karta for contributing useful discussions and knowledge to my research. I am especially grateful to my parents who raisedme and are longtime supporters. Their full support and love were my stronghold in doing this research. Finally I express my deepest gratitude to my wife, Yen-Li Sun, who has been a source of strength and inspiration over the years. I especially thank her for her support and understandings during many late nights while researching toward this dissertation. TABLE OF CONTENTS page ACKNOWLEDGMENT S ........._.._.. ...._... ............... iv... LI ST OF T ABLE S ........._.._.. ...._... ............... viii.. LI ST OF FIGURE S .............. .................... ix AB STRAC T ................ .............. xi CHAPTER 1 INTRODUCTION ................. ...............1.......... ...... 2 POWER LINE COMMUNICATION NETWORK INFRASTRUCTURE FOR HOME NETWORKS .............. ...............6..... Home-Network Infrastructures ................. ............ ...............6....... Applications over Power Line Communication ........._.__............ ........_._.....11 Power Line Communication Application in a Home .......__. ........ ........... ....11 Internet Bridging .............. ...............14.... PLC Design Issues .....__................. ........___......... 1 Physical Limitations ................. ................. 16..._ __..... Si gnal M odulation ................. .............. 16.............. MAC Layer Protocols s....___................ ...............18. .... Performance Results and Analysis .............. ...............19.... Simulation Results ................... ..... ...............19...... Real-World PLC Network Performance .............. ...............21.... Performance of Delivering Streaming Video............... ...............22. Performance of Elastic Data Traffic. ................. ............_ ............___...23 Performance of Combined Delay Sensitive Traffic and Elastic Data Traffic24 Conclusion............... ...............2 3 A COMPARATIVE PERFORMANCE STUDY OF WIRELESS AND POWER LINE NETWORKS .............. ...............27.... Introducti on .................. ...............27................. Homeplug 1.0 Protocol ................. ...............29.......... ..... PLC Environment ................. ...............29................. HomePlug 1.0 PHY ................. ...............30.......... .... HomePlug 1.0 M AC .................... .. .......... ... ..._ _. ...........3 Theoretical Performance of 802.11la/b and Homeplug 1.0............_._. .........._._. ...33 802.11a/b Theoretical Network Performance............... ..............3 Performance Analysis of HomePlug 1.0 .............. ...............34.... Experimental Setup............... ...............36. Experimental Method ........._.__........_. ...............37.... R e sults........._._..... ..... .._. _...... ...............38. IEEE 802.11la Indoor Performance .....__.....___ ..........._ ..........3 IEEE 802.11lb and HomePlug 1 .0................ ...............41.. TCP Link Stability............... ...............4 Discussion and Conclusion............... ...............5 4 PERIODIC CONTENTION FREE MULTIPLE ACCESS FOR POWER LINE COMMUNICATION NETWORKS .............. ...............55.... Introducti on ............. ...... ._ ...............55.... Previous W orks ............... ......._ ...............57.... Proposed PCF/MA Protocol .............. ...............59.... The Concept of PCF/MA. ................. ...._.. ...............60. ... Distributed Admission Control ......__................. ......._.. ......... 6 Data Exchange ....___ ................. .........__..........6 MAC Protocol Data Unit ........._._ .......... ..... ... ......_.. .. ..........6 Solutions to the Hidden Node Problems and Near Far Effect............._.._...........64 Approximate Performance Analysis and Simulation Results ................. ................. 66 Determine Parameter R .............. ...............67.... Near-Far Effect Modeling .............. ...............71.... Extra Allocation............... ...............7 Data Stream M odel ................. ...............73................ Delay M odel .............. ...............75.... Video Traffic .............. ........ ........ ..........8 Visualization of Protocol Simulation Data ................. ...............87........... ... Introduction .............. ...............87.... Goal of User Interface .............. ... .... .. ........ ...............88..... Data Presentation Mechanism and System Design .............. ....................9 Experience Using the Visualization System ................. ................ ........ .95 5 FUTURE WORK AND CONCLUSION ................. ...............96........... ... Future Work............... ...............96.. Voice over IP .............. ...............96.... Distance Factor ................. ...............97........... .... Congestion Degree .............. ...............98.... Priority Factor............... ...............99. Related Study............... ....... ... ..........10 Experimental Performance Results .............. ...............101.... Results Based on the Distance Factor .............. ...............103.... Results Based on Congestion Degree ................. ...............105........... ... Results Based on Priority Factor ................. ...............106........... ... Conclusion ................ ...............107................ LIST OF REFERENCE S ................. ...............110................ BIOGRAPHICAL SKETCH ................. ...............113......... ...... LIST OF TABLES Table pg 2-1: Technology comparison .............. ...............8..... 2-2: Application traffic amount in a home............... ...............12.. 2-3: A power line network simulation results ...._ ................. ............... ....2 2-4: Real-world PLC network performance .............. ...............22.... 3-1: 802.11la/b MAC throughput with payload 1500 bytes ........._.__...... .._._...........34 3 -2: List of the houses tested and connectivity ................. ...............38........... 4-1: HomePlug 1.0 and PCF/MA parameters ....... ................ ..................61 4-2: Video traffic parameters ................. ...............84............. 5-1: Performance comparison for distance factor ................. ............... ......... ...10 5-2: Performance comparison based on congestion factor ................. ......................105 5.3: Performance comparison based on priority factor in day time ............... ..............107 LIST OF FIGURES Figure pg 2-1: Power-line topology in a North American home ................ ....___. ................11 2-2: An example of using one of the computers as the PLC, DSL or Cable Modem routerl 5 2-3: Connecting PLC networks to the Internet ........._..... ...._._ ......._.. .......2 3-1: HomePlug 1.0 frame structure and protocol............... ...............32 3-2: IEEE 802.11la indoor connectivity ................. ...............39.............. 3-3: IEEE 802. 11a indoor throughput as a function of distance ................. ................ ..40 3-4: IEEE 802.11la indoor percentage of links ................. ...............40............. 3-5: IEEE 802.11lb and HomePlug 1.0 indoor connectivity comparison .........................41 3-6: IEEE 802.11lb and HomePlug 1.0 indoor throughput comparison ...........................42 3-7: IEEE 802.11lb and HomePlug 1.0 indoor percentage of link versus throughput (MIM) com prison .............. ...............43.... 3-8: IEEE 802.11lb and HomePlug 1.0 indoor percentage of link versus throughput (IM) com prison .............. ...............45.... 3-9: IEEE 802.11la high speed real-time capture ................. ...............47........_. 3-10: IEEE 802.11la low speed real-time capture ................. ...............48........... 3-11: IEEE 802.11lb high speed real-time capture ................. ...............48........... 3-12: IEEE 802.11lb low speed real-time capture .............. ...............49.... 3-13: Homeplug 1.0 high speed real-time capture............... ...............50 3-14: Homeplug 1.0 low speed real-time capture ....._____ .... ... ..__ ...............5 4-1: PCF/MA frame structure ........._.._.. ...._... ...............60.... 4-2: MPDU process and format ..........._ ..... ..__ ...............61.. 4-3: Reservation process flow chart ............... ...............62......__...... 4-4: Probability of successful reservation............... ..............7 4-5: Maximum over allocation PBs various PB loss rate comparisons ...........................73 4-6: PCF/MA various T versus MC SMA/CA throughput compari sons.............._.._. .........74 4-7: PCF/MA delay model ........._._ ...... .... ...............75.. 4-8: Analytical and simulated average delay with various w, T=25ms.............................78 4-9: Analytical and simulated average delay with various w, T=50ms.............................78 4-10: Analytical and simulated average delay with various w, T=75ms ........................79 4-11: Analytical and simulated average delay with various w, T=100ms............._._... ......79 4-12: Video traffic model............... ...............80. 4-13: A large video frame in transmit............... ...............81 4-14: Video playback performance comparisons .............. ...............84.... 4-15: Capture of a period of the DVD simulation ................. ...............85........... 4-16: Multiple video streams comparisons ................. ...............86.............. 4-17: Text versus graphical event analyzer .............. ...............88.... 4-18: Turning off unused events can make desired events stands out ............... ... ............91 4-19: VPA configuration format ............_ ..... ..__ ...............92. 4-20: Simulator log format. ............ _...... ._ ...............93.. 4-21: The actual display window is a portion of the whole data .............. .............. .94 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 HIGH-PERFORMANCE PERIODIC CONTENTION-FREE MULTIPLE-ACCESS PROTOCOL FOR BROADBAND POWERLINE COMMUNICATIONS By Yu-Ju Lin May, 2004 Chair: Haniph Latchman Major Department: Electrical and Computer Engineering Applications over PLC networks have drawn much interest in the academic community as well as in the communication industry, not only because of inherent convenience (connecting PLC-capable devices requires no new wires), but also because almost all electrical devices have to connect to a power outlet eventually. This technology makes implementing a digital home entertainment center more realistic than ever. HDTVs in different rooms are now able to share digital content from one set-top box without rewiring, or setting up wireless access points. MP3 players can access music data through PLC networks from different rooms playing different music. PLC networks provide an enabling technology for the smart home. However, PLC technology is still evolving and many problems remain unsolved. The hostile environment of PLC channels makes reliable data transmission difficult. The PLC channel is known for its hostile nature in transmitting electrical signals. Protocols designed for other media may not be suitable for PLC. In some ways, PLC channels are similar to wireless channels: both of them face hidden node problems, near-far effects and other channel imperfections. However, PLC network nodes tend not to move. It is unlikely that simply applying protocols designed for another medium would result in good performance in the PLC environment; the overhead may be too high or the assumptions about noise may be too optimistic for PLC networks. In light of PLC's unique characteristics, we developed a new protocol called: Periodic Contention-Free Multiple Access (PCF/MA). PCF/MA is an explicit R- ALOHA-like protocol specifically designed for the PLC network. We propose an RTS/CTS-like scheme in the reservation stage to mitigate hidden-node problems, and a delayed NACK mechanism to conquer the near-far effect. Performance of the proposed protocol is evaluated by event-driven computer simulations and by mathematical analysis. Simulation results show that 85 Mbps MAC throughput with a 100 Mbps channel data rate can be obtained, even when there are hidden nodes in the network. To provide smooth video delivery, we propose a mathematical estimation of the required delay in playback time and the amount of playback buffer with tight bandwidth reservation. Our simulation shows that a 100 Mbps channel can deliver up to 9 MPEG-2 video streams simultaneously without dropping any video frames. A visual protocol analyzer was also developed as a tool to study network protocol . CHAPTER 1 INTRODUCTION The concept of Information Appliances (IA) became a recent reality. Many next- generation appliances come with communication capabilities with embedded processors built right into the devices. For instance, on April 7, 2001, IBM and Carrier announced that they will produce a new air conditioner with JAVA support that can send e-mails to manufacturers for errors; or the user can send commands to the air conditioner to pre- adjust room temperature. We believe that before long our homes will have many kinds of IA devices communicating among them and with the outside. Many of these IA devices are expected to have multimedia capability. Intelligent homes of the future will need multimedia communication support for these IA devices. Providing the right infrastructure for connecting these IA devices will be a major need. For home applications, this infrastructure must be easy to setup, inexpensive to install and maintain, and must perform well. Ordinary people are not network experts, and a typical high-performance network is too complicated for casual daily usage. The supporting infrastructure should be easy to set up, and the effort needed to maintain this infrastructure should be minimal. Differences between home networks and ordinary networks can be summarized as follows: * Home networks are different than other types of networks. The data stream in a home network is much smaller than in ordinary public networks and local area networks, because the coverage of a home network is small. With this small coverage, the expected response time is short too. * The main purpose of home networks is to share resources. Examples, include sharing a single printer for several computers; using a single Internet connection; sharing files (such as images, spreadsheets and documents); playing games that allow multiple users at different computers; and sending the output of a device (like a DVD player or Webcam) to other devices. *Home networks should be easy to setup, maintain, and access. They should be affordable for most families. Many existing networking technologies compete to support this mission. For example, a comprehensive Ethernet network can be constructed by installing UTP-5 special cabling around the house. Alternatively, wireless networks such as 802.11x, Bluetooth, and HomeRF can be constructed by installing multiple interconnected wireless access points (WAP) and base stations within the home. However, the IA devices themselves would need wireless capabilities; and the above three infrastructures all require significant effort and cost to build up the networks externally. Phone-line networks such as HomePNA [1] may seem attractive, but the convenience of mobility is limited by the number of phone sockets available in a home. Hughes and Throne [2] did extensive study of other infrastructure options and technologies appropriate for a home network. In general, for future IA applications, the amount of traffic generated by appliances and computers on a home network is still unknown. Watching digital TV while downloading data from the web should cause no video jitters over home networks. To the best of our knowledge, there is little research to report the performance impact (or at least the likelihood) for supporting multimedia-enabled IA communication over home networks. We are interested in exploring various methods to study the impact of different performance results. We are also interested in analyzing the effect on each individual data streams including continuous media data stream (that is, soft real-time traffic). In Chapter 2, we studied the performance of multimedia over power line networks using simulation studies and actual measurements on a \emph{Homeplug} 1.0 compliant PLC network. We were particularly interested in measuring the PLC network raw data rate, TCP performance, and the performance impact when QoS support is involved. We were also interested in analyzing network performance with different traffic types, including continuous media data streams (that is, soft real-time traffic). We first built a network simulator that generates various types of traffic, and then applied the same scenarios to a real-world PLC network and to a simulation model. We compared simulation results and real-world performance. A maximum throughput of 8.08 Mbps for UDP was obtained from our simulation, while a 6.21 Mbps TCP throughput was observed in the real-world PLC network experiment. Results show that PLC networks can successfully deliver real-time traffic concurrently with traditional data traffic. Our contributions to this research topic also include modelling human behavior in the use of IA devices, modelling the types of traffic generated by IA devices communicating over the power line, and measuring the performance of real applications over PLC networks. We also describe a practical implementation of the PHY and MAC layers for PLC networks associated simulation results. In Chapter 3, we discuss PLC, 802.11b and 802.11a characteristics, protocol, and theoretical performance. Based on the theoretical analysis, we conducted a series of real- world experiment and performance comparisons between 802.11b and PLC networks. We compared two technologies for implementing a home or small-office networks without new cable installation. For ease of installation, both wireless and power-line networks were well rated. The wireless network uses air and the power-line network uses the existing power cable as a transmission medium. Current field testing results show that under most circumstances, power line has more stable and reliable connections. Even when the line-of-sight distances between two stations are as long as 69 ft., PLC devices can still generate up to 4.52 Mbps throughput. From 128 testing positions, PLC has 72% (from server to mobile station) and 59.38 % (from mobile station to server) better experimental results than 802.11lb. In Chapter 4, we proposed a new Protocol: Periodic-Contention Free Multiple- Access (PCF/MA) Protocol for high-speed Power Line Communications (PLC). PLC networks have impairments similar to wireless networks, but the nodes are largely stationary. There is industrial intent to use PLC networks in the home for delivery of multimedia data, with challenging quality of service (QoS) requirements. Existing protocols for PLC, for wireless networks, or for wired networks cannot meet these challenges efficiently. Our study proposed and analyzed a new protocol designed to provide the high QoS needed for delivering multiple multimedia streams in a PLC environment. The proposed protocol, PCF/MA, addresses hidden nodes and the near-far effect; in addition to addressing the generally noisy medium. PCF/MA performance was analyzed theoretically and was simulated. Simulation showed that 85 Mbps MAC throughput is possible with a channel-data rate of 100 Mbps, even when there are hidden nodes in the network. Through mathematical modeling of buffer space required for tight- bandwidth allocation, such a network can deliver up to 9 MPEG-2 video streams simultaneously without dropping any video frames, compared to 7 video streams using Modified CSMA/CA(MCSMA/CA). Chapter 5 discusses three common factors that influence the Voice over IP quality and future works. Our experiments show that without dedicated networks and QoS enabled schemes, the factors of distance and degree of congestion were proved to be influential on the achieved end-to-end bit rates. Many research issues still need to be addressed before we can guarantee (at least statistically) the quality of the Interne Telephony. CHAPTER 2 POWER LINE COMMUNICATION NETWORK INFRASTRUCTURE FOR HOME NETWORKS Low-voltage electrical wiring in homes has largely been dismissed as too noisy and unpredictable to support high-speed communication signals. However, recent advances in communication and modulation methodologies (as well as in adaptive digital signal processing and error detection and correction) have spawned novel media access control (MAC) and physical layer (PHY) protocols capable of supporting power line communication networks at speeds comparable to wired local area networks (LANs). In this chapter we support the use of power-line LAN as a basic infrastructure for building integrated smart homes, where information appliances (IA)--ranging from simple control or monitoring devices to multimedia entertainment systems- are seamlessly interconnected by the very wires that provide their electricity. By simulation and actual measurements using reference design prototype commercial powerline products, we showed that HomePhig MAC and PHY layers can guarantee QoS for real-time communications, supporting delay-sensitive data streams for smart home applications. Home-Network Infrastructures Many next-generation appliances are being equipped with processors featuring sophisticated communication capabilities. For instance, on April 7, 2001, IBM and Carrier announced plans to produce an air conditioner with JAVA support that can Email manufacturers regarding errors, and will allow users to remotely send commands to the unit to adjust temperatures or shut it down. Smart homes will eventually have many types of information appliances (IAs) communicating among themselves and with the outside world. Soon, many of these IA devices are expected to have multimedia capability. Supporting multimedia communication for these IA devices will be of crucial importance for the smart homes of the future. Providing the right infrastructure for connecting these IA devices will be a major need. For home applications, this infrastructure must be easy to set up, inexpensive to install and maintain, and must perform well. Ordinary people are not network experts, and a typical high-performance network is too complicated for casual daily usage. The supporting infrastructure should be easy to set up, and the effort to maintain this infrastructure should be minimal. Many existing networking technologies compete to support this mission. For example, a comprehensive Ethernet network can be constructed by installing UTP-5 special cabling around the house. Alternatively, wireless networks such as 802.11x, Bluetooth, and HomeRF can be constructed by installing multiple interconnected wireless access points (WAP) and base stations within the home. However, the IA devices themselves would need wireless capabilities, and the above three infrastructures all require a significant effort and cost to build up the networks externally. Phone-line networks such as Home- PNA [1] may seem attractive, but the convenience of mobility is limited by the number of phone sockets available in a home. Hughes and Throne [2] did extensive study of other infrastructure options and technologies appropriate for a home network. In this chapter, we advocate direct use of existing electrical wiring and outlets as the medium for data communication within the home. Using power lines as the network infrastructure has many advantages over other technologies. First, no new wires are needed, since the IA devices will communicate over the very wires that provide their electrical power. Second, there are many access points (power sockets) in a home (4 or more per room). Currently, Power Line Conanunication (PLC) as specified by the HonzePhig 1.0 standard [3] provides a 14 Mbps raw data rate, which is adequate for daily IA-device communication. It also has a built-in QoS protocol, making it attractive for real-time streaming applications. Finally, the cost to build a power-line network is low when compared with other technologies. For example, it was observed that the 802.11x wireless network card has approximately the same street price as the HonzePhig network card (about$120). It is expected that with mass production requiring no expensive RF

components, the cost of the PLC cards will be about 50% less than comparable wireless

cards. Moreover, the cost of a required 802.11x base station is high (more than $250). The 100 Base TEthernet has the highest performance/cost ratio, but requires new cables and expensive installation. Table 2-1 shows costs and other characteristics of home- network technologies. Installation costs (which are high for 10/100 BT) are not shown. Table 2-1: Technology comparison Technology Media Data Rate(Mbps) QoS Support Cost($)
10 Base T UTP 10 No 20
100 Base T UTP 100 No 80
Bluetooth Wireless 1 Yes 5
HomeRF 2.0 Wireless 10 Yes 110
802.11lx Wireless 11 No 125
HomePNA 2.0 Phone line 10 No 80
HomePlug Power line 15 Yes 120

From a marketing perspective, less-expensive and easier-to-use PLC home

networks are becoming more attractive, and the potential market is huge. The Yankee

Group (Boston, MA) estimates that at least 21 million households in the United States are

interested in home networking and that 12.4 million would like to implement in-home

networks within the next year. According to Parks Associates (Arlington, VA), 30

million households in the United States will have fast internet connections by 2004, and

17 million of them plan to have home networks.

In the past, power lines were considered unacceptable for signal transmission, since

the channel contained a lot of noise, interference, and fading. However, the appeal of

using the existing power line as a transmission medium for data exchange was too great

to be ignored. The advancement of signal-modulation technologies, digital signal

processing, and error control coding [4] has minimized the restrictions of channel

imperfections; and high-speed signal transmission through power lines is now feasible.

Using the existing power-line infrastructure as the medium for supporting IA

communication requires a careful design of the overlaid communication systems in order

to provide acceptable communication services. It is desired, for example, that when

the video quality. Current research shows that the maximum raw data rate of first

generation PLC is about 14 Mbps. However, the effective data rate is expected to be

around 10 Mbps after compensating for impairments and error corrections. On the other

hand, research is currently underway to develop PLC chips that operate at 100 Mbps with

average throughput 30-60 Mbps.

In this chapter, we investigate the performance of multimedia over power-line

networks using simulation studies and actual measurements on a Homephig 1.0 compliant

PLC network. We are particularly interested in measuring the PLC network raw-data rate,

TCP performance, and the performance impact when QoS support is involved. We are

also interested in analyzing the network performance with different traffic types,

including continuous-media data streams (that is, soft real-time traffic).

We first built a network simulator that generates various types of traffic, and then

applied the same scenarios to a real-world PLC network and to a simulation model. The

performance comparison between the simulation results and real-world performance are

given in this chapter. A maximum throughput of 8.08 Mbps for UDP was obtained from

our simulation, while a 6.21 Mbps TCP throughput was observed in the real-world PLC

network experiment. Results show that PLC networks can successfully deliver real-time

traffic concurrently with traditional data traffic.

Our contributions to this research topic also include modelling human behavior in

the use of IA devices, modelling the types of traffic generated by IA devices

communicating over the power line, and measuring the performance of real applications

over PLC networks. We also describe a practical implementation of the PHY and MAC

layers for PLC networks as well as associated simulation results.

The next section discusses background for implementing power-line

communication networks, typical applications of power line communication networks,

and the human behavioral model using these appliances. Section 2.3 presents the physical

limitations of power line channels, and describes a practical signal-modulation scheme

and MAC protocol as used in the HomePhig 1.0 protocol proposed by the HomePhig

Powerline Alliance. Section 2.4 compares simulation results with real-world PLC

network performance. Finally, Section 2.5 offers conclusions and suggestions for future

work.

Applications over Power Line Communication

Power Line Communication Application in a Home

Rhtribution
Transformer

Fiur -1 owrlietooog i Nrh mria hm
Traditional, pwer ines re ued fo coneyingelecricalpowe to evicesPoelis
werenot esiged or dliveing ighfreqencysignlsand o th elctriacalad feuec

response ~ ~ ~T reqireent ofapwrln r o a rtcla hs fdatantokcbigh
poor qualityof a power lne is not ieal or inltasisoeauetecanlcnan
nois an inerfeenc. Te meiumis adeof ifrn odco ye:teeoeavreyo

mismatch-1 caus -ies muti-path i efec Nreutin Aein dee notces tcranfeunishs

channelnlly imefetonsmae signsal modulaton coverin alcri power linedificul [5. However, thes

adv qancmety of sga p rlin mouaison dand eror cognatrol codsing techniuses now mhakel poerine

commu n icatioernce posibe.mdu smd fdfeetcnuco ye;teeoeavre

The common power line topology of a North American home is shown in Figure 2-

1. The figure shows a tree-like power-line topology in a house. Typically, there are two

power line trunks: one is 110V and the other is 220V. Each power-line trunk can be

divided into several branches. Power-line communication aims to transmit data packets

over these branches and trunks. The topology of the power line network and the

convenience of its power sockets as potential access points make it a good candidate for

smart home IA device networking.

Table 2-2 shows the results of a survey from which we inferred usage and traffic

patterns generated by typical IAs. The table also suggests some current and future PLC

applications. For instance, when merchandize is advertised on a digital TV service, the

product information (such as the barcode or webpage) can be downloaded to your

computer through a power line. Afterwards, you can send your order information from

the computer to the supplier, or you can use the downloaded URL to browse the product

web page and get more details. We also anticipate the ability to record music or videos

through a power line. For example, when a song is broadcast on TV or a music channel,

you can download the song directly to an MP3 player through the power line. Another

application is the opportunity to record digital video directly into a PC or even a digital

VCR. Other applications of IA's can be easily accomplished using a PLC network. For

example, a refrigerator can order food through the power line network according to its

inventory, or it can send cooking instructions to the microwave. A smart oven can send

predicted environmental temperature information to the air conditioner through a power

line, allowing the air conditioner to pre-adjust the temperature and keep rooms

comfortable.

network facilitates data exchange between traditional data processing devices such as

PCs and computer peripherals. IA devices that talk with PCs are also possible. For

example, sending multimedia data from TVs or VCRs to PCs can be easily done by PLC

network, but is difficult with other infrastructure technologies. Home security can also be

implemented by PLC so that a digital camera installed on the front door can send video to

the TV.

Possible time

Table 2-2: Application traffic amount in a home
Row From node To Node Estimated Frequency
No. data size

period
7:00-9:00,11:00-
1:00,
17:00-19:00,21:00-
23:00
7:00-9:00,11:00-
1:00,
17:00-19:00,21:00-
23:00
11:00-1:00,17:00-
23:00
11:00-1:00,17:00-
23:00
11:00-1:00,17:00-
23:00
11:00-1:00,17:00-
23:00
11:00-1:00,17:00-
23:00
6:00-24:00

11:00-1:00,17:00-
23:00
11:00-1:00,17:00-
23:00
6:00-24:00

6:00-24:00

160 bytes 2 times
period

1 Refrigerator

2 Microwave

3 TV

4 TV

5 TV

6 TV or
Settop box
7 Computer

8 Computer

9 Settop box

10 Computer

11 VCR

12 Front door
camera

Microwave

72 bytes

2 times a
day

3 times a
day
3 times a
day
3 times a
day
3 times a
day
1 time a day

1 time a day

1 time a day

1 time a day

1 time a day

3 times a
day

Refrigerator

VCR

Computer

PDA or MP3
player
PDA or MP3
player
Computer

Computer

Internet

Computer

Computer

750 bytes

11KBytes

360 bytes

15 Mega
bytes
50 Mega
bytes
60 MB to
180 MB
320 MB to
640 MB
44 MB to
131 MB
320 MB to
640 MB
110 MB to
1100 MB

The applications over PLC are not only for

novel IA devices. PLC as a home

Table 2-2 also gives an estimate of the daily traffic volume generated by typical IA

applications. These values are based on likely information size. For example, the

instruction size that the refrigerator sends to the microwave in Row 1 is estimated by the

number of steps required to cook the food (1 byte), the cooking time for each step (4

bytes for each step), the power level for each step (2 bytes), and the packet header size.

Added together, the entire instruction size is 160 bytes. Row 7 exemplifies storing digital

music from a computer to an MP3 player. The 50 Mbytes traffic volumes is calculated

from the number of songs in an album, the length of a song (5 min), the encoded data rate

(128 kbps), and the packet header size. The frequency and time period during which each

event occurs are also shown. By using this data and typical household dynamics for

concurrent events, we can generate a traffic flow for the power-line network for a typical

day.

Intemet Bridninn

Currently, in-home PLC networks rely on other technologies to send data to the

Internet and communicate with mobile devices. Most of the homes in the United States

will eventually be equipped with broadband connections like DSL or cable modem

services. To share the broadband Internet connection with PLC capable devices, we can

add a PLC Internet router to the PLC network. One possible setup is shown in Figure 2-2.

In this figure, a desktop computer acts like a data center. Devices that need to

communicate with other devices on the Internet will send data to the desktop PC via the

power line. The desktop PC decides whether to send it to the Intemet. In the future, an IA

routing device may be unnecessary. Researchers are developing a solution to make PLC

home networks talk directly with other homes, power plants and the Intemet using the

external distribution power line. Such a network infrastructure for the Intemet access

would be especially attractive to developing countries, since no additional expenditure is

needed for data network infrastructures.

Figure 2-2: An example of using one of the computers as the PLC, DSL or Cable Modem
router

To support data exchange with mobile devices, PLC networks will also need to

cooperate with wireless networks. The easiest way to achieve this is to make the 802.11lx

base station PLC compatible. The base station is treated as an ordinary IA device with a

PLC chip built in (see Figure 2-2). Mobile devices with wireless capability can then talk

to devices attached to the power line. This is especially ideal when communication is

desirable but large coverage areas require multiple interconnected wireless access points

(the interconnection is then provided with "no new wires" using the existing power line

infrastructure which would be needed to power the WAPs in any event).

The above PLC applications require a properly designed protocol. In addition, to

make the PLC network real-time traffic friendly, special care is needed to support delay

sensitive traffic. In the following sections, we discuss the physical limitations of power

line channels and then describe a robust power line protocol.

PLC Desinn Issues

Physical Limitations

A power line is used for transmitting 50 or 60 Hz signals but was not designed to

convey high frequency signals such as the 20MHz communication signal used in the

Homephig 1.0 protocol. A power line channel is somewhat like a wireless channel both

of them suffer from noise, fading, multi-path and interference. Power line noise is

produced by the operation of electrical devices. Fading, multi-path and interference are

caused by the imperfection of power line channels.

Lim C.K., So P.L., Gunawan, E., Chen, S., Lie, T.T.,and Guan, Y.L., [6] describe

typical attenuation characteristics in power line channels. The authors report that even

when all devices are unplugged, the noise still persists and this drastic variation of

attenuation is hostile to power line communication. Furthermore, the Federal

Communications Commission (FCC) also limits the available bandwidth for

communication purposes. In compliance, the usable bandwidth in the HomePhig standard

is 25MHz. Liu Weilin, Widmer H.-P., Aldis J.,and Kaltenschnee T., [7] did an extensive

study of the power line channel characteristics and design issues.

To conquer these problems, a robust signal modulation and data coding is needed.

Signal Modulation

To modulate digital signals on to the power lines, we can use many of the same

techniques that are widely implemented in wireless communication. Basic modulation

techniques such as Phase Shift Keying (PSK), Frequency Shift Keying (FSK), Minimum

Shift Keying (MSK), and Gaussian Minimum Shift Keying (GMSK) can be used for low

data rate communication. Other more advanced techniques such as M-ary Phase Shift

Keying (MPSK), M-ary Quadrature Amplitude Modulation (MQAM), M-ary Frequency

Shift Keying (MFSK) and Orthogonal Frequency Division Modulation ( OFDM) can be

used when higher data rates are desired. Karl M., and Dostert K.[8] did a thorough study

of signal modulation over power lines.

OFDM was adapted by HomePlug Powerline Alliance because of its robustness to

noise and the fact that it is a parallel data transmission method using a number of parallel

frequency division multiplexed subbands. The main problem in using OFDM on wireless

networks is frequency offset, caused by the Doppler Effect when the user is moving. The

Doppler Effect will cause performance degradation, but in a power line network there are

no moving devices, and thus no Doppler effect. The other problem is timing offset, which

can be mitigated by offset estimation and compensation.

Spread spectrum signal modulation is different. Since the useful bandwidth in the

power line channel is under 25 Mhz, the effect of spread spectrum modulation is

considered limited. Using a single carrier modulation on the power line is possible but

equalizers could be needed to reduce the delay spread effect, and the associated cost is

high.

In order to cope with the wide variation in channel conditions, the physical layer

protocol (PHY) for PLC must be adaptive intelligently using more robust modulation and

coding schemes, with lower data rates as needed. In addition, critical protocol

management information requires high fidelity forward error correction (FEC) coding to

ensure that the protocol functions correctly in the worst case situations.

MAC Layer Protocols

In PLC home networks, the power line media can be accessed by multiple devices

simultaneously. To decide which device gets the floor to send its data, a medium access

control (MAC) protocol is needed. There are many existing protocols that can be

implemented on the power line network. CSMA/CD, CSMA/CA, TDMA and hybrid

protocols such as TDMA+CSMA are all potential candidates.

The most popular wired MAC protocol, CSMA/CD, could be also applied on a

power line network. However, the large variation in noise on the power line makes

collision detection very difficult. This characteristic is again very similar to a wireless

network, so some have applied the CSMA/CA protocol as suggested in IEEE802.11 to

the power line network. However, the hidden node problem arises when the signal travels

through different power lines with highly variable attenuation. To conquer this problem, a

RTS/CTS scheme has to be implemented. Though the RTS/CTS scheme solves the

hidden node problem, it degrades the network performance.

The benefit of using TDMA is that it provides an upper bound of access delay thus

QoS is guaranteed. However, the difficulties in generating a synchronized clock signal in

power line networks between devices remains a problem. Other hybrid protocols like

TDMA+CSMA provide QoS capabilities in nature, but the network efficiency and

beacon generation between TDMA slots and CSMA/CA slots remains unsolved. Romans

C., and Tourrilhes J. [9] did a detailed discussion of the hybrid TDMA+CSMA/CA

protocol. Homeplug 1.0 protocol also provides some level of QoS support in the uses of

multiple priority levels that can be used in conjunction with VLAN tagging.

The issue of privacy of power line networks is important to their practicality. Like

wireless channels, Power line network channels should be treated as open and as with all

open channels; nothing prevents a device from receiving signals. To provide a secure

network environment, the HomePhig Powerline Alliance defined a 56-bit DES

encryption mechanism. Once a signal is encrypted, a device with a different encryption

key cannot interpret it and privacy is achieved.

This privacy protection seems adequate but stronger encryption may be needed

when power line networks are adopted for office environments or apartment building and

hotels. We believe that stronger privacy protection should be implemented in the physical

layer, so that hackers can not easily break the code.

Performance Results and Analysis

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

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

and a 3/4 coding rate on various links and a maximum TCP segment size of 1460 bytes.

In this simulation, we use UDP, TCP, and VOIP traffic. UDP traffic is generated with an

exponential inter-arrival time with a 100 microsecond average. The UDP packet size is

assumed to be a constant 1460 bytes with priority 0. TCP traffic is also generated with

exponential inter-arrival time with 100 microsecond average and we assume that TCP

traffic sources always have data to send. TCP traffic is treated as priority 0 packets.

Every time a node has a chance to send, it is allowed to send the maximum segment size

of 1460 bytes without headers. VOIP is isochronous traffic with a 20 msec interval. The

packet size of VOIP is 160 bytes and is assigned the highest priority (3).

Simulation Results

In Table 2-3, 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 III also shows channel contention with 2 and 3 UDP nodes causes a modest reduction in

channel throughput.

Table 2-3: A power line network simulation results
Throughput of multiple UDP traffic
Scenario 1(1 UDP) Scenario 2(2 UDP) Scenario 3(3 UDP)
MAC Throughput 8.08 Mbps 7.46 Mbps 7.46 Mbps
Throughput of multiple TCP traffic
Scenario 1(1 UDP) Scenario 2(2 UDP) Scenario 3(3 UDP)
MAC Throughput 6.16 Mbps 6.15 Mbps 6.12 Mbps
TCP Throughput 5.92 Mbps 5.91 Mbps 5.88 Mbps
Throughput of one VOIP and multiple UDP traffic
Scenario 1 Scenario 2 Scenario 3
(VOIP + 1 UDP) (VOIP + 2 UDP) (VOIP + 3 UDP)
MAC Throughput 7.89 Mbps 7.33 Mbps 7.29 Mbps
Queueing Delay 0.25 msec 0.25 msec 0.25 msec
Net Delay 2.75 msec 3.00 msec 3.00 msec
Throughput of one VOIP and multiple TCP traffic
Scenario 1 Scenario 2 Scenario 3
(VOIP + 1 TCP) (VOIP + 2 TCP) (VOIP + 3 TCP)
MAC Throughput 6.04 Mbps 5.85 Mbps 5.77 Mbps
TCP Throughput 5.72 Mbps 5.54 Mbps 5.45 Mbps
Queueing Delay 0.25 msec 0.25 msec 0.25 msec
Net Delay 3.25 msec 3.25 msec 3.25 msec

In the TCP traffic simulation, though scenario 1 has only one traffic source, the

bandwidth must be shared with data and response frames (for example, ACK packets)

thus it provides lower performance than the UDP traffic simulation. The MAC

throughput represents the total number of transmitted bytes divided by the simulation

time regardless of successful delivery. The TCP throughput includes only the

successfully delivered data and ACKs.

The third metric we provide in Table III 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 queuing delay refers to the time a packet waits in a queue before it enters the transmit

buffers. The net delay is the total time for which a packet propagates in the networks. Only low

priority packets are considered for this delay because the high priority packets will be delivered

as soon as they appear in the queue.

The Table 2-3 also shows the simulation results of one VOIP and multiple TCP

connections. The throughput of VOIP is only 80 kbps, and hence the total throughput is

dominated by the TCP component.

Real-World PLC Network Performance

In addition to simulating the performance of the HomePhig Powerline Alliance

protocol, we were also able to construct a real PLC network using "reference designs" of

actual commercial HomePhig devices. Since there are currently no real IA devices with

PLC capability, we used traditional network applications (that is, ftp and streaming

multimedia content) as the basis for measuring PLC network performance.

In this experiment, there were 4 desktop computers. A 450 MHz Pentium II

desktop computer (PC-2 as a file server) is equipped with 128 MBRAM, a 3-COM fast

Ethernet card, and a PLC PCI card. Two 700 MHz Pentium III desktop computers (PC-3

and PC-4) are both equipped with 256 MBRAM and PLC PCI cards. A 266 MHz

Pentium MMX desktop computer (PC-1) is equipped with 64 MB RAM, and a 3-COM

fast Ethernet card.

The PC-1 computer is connected to an Ethernet-to-power line bridge, which

converts packets generated from the Ethernet card into PLC compatible packets, and vice

versa. All computers are connected to power lines.

In this experiment, we seek to determine the performance of the PLC network in

handling streaming video and large file transfers.

Performance of Delivering Streaming Video

We first examined the ability of the PLC network to deliver real-time traffic. Four

video files are involved in this experiment. The first file is encoded in Real media format

with a bit rate of 550 kbps; the second is encoded with bit rate of 1396 kbps, the third is

encoded at 2 Mbps, and the fourth is an MPEG2 video file with variable bit rate, and the

average bit rate is 8 Mbps. In the first experiment, three client computers simultaneously

issued requests for low bit rate (550kbps) video service to the file server. In the second

experiment, the same procedure was executed, but a medium bit rate (1394 Kbps) video

service was requested. In the third experiment, the 3 clients requested a high bit rate (2

Mbps) video service. Finally, the MPEG2 video service request was issued by the PC-3.

The experimental results are shown in Table 2-4.

The PLC network successfully delivered both low and medium bit rate streaming

videos. We did not observe any packet drops during the experiments. The results met our

expectations, since the peak data rate was only 4185 kbps. We did another experiment to

further investigate the performance of PLC network in delivering streaming video. A 2

Mbps MPEG-1 file is used in this experiment. As the video begins, a significant video

freeze-then-go (halting) phenomenon was observed, causing staccato playback. After

several seconds (3-5 seconds) the freeze-then-go phenomenon disappeared.

In the case of MPEG2 video file, the average data rate is 8Mbps. During the

experiment, a significant "video staccato" phenomenon was observed. To exclude the

possibility that the observed phenomenon was caused by the client computer's hardware

capability, the experiment was repeated with same configuration, while connected to a

fast Ethernet. During that experiment, no such phenomenon (halting playback) was

observed.

Table 2-4: Real-world PLC network performance
Performance of delay sensitive traffic
Number of Aggregated bit rate Packet delay-j itter
Connections drop
1 550 kbps No No
Low Bit Rate 2 1100 kbps No No
3 1650 kbps No No
.1 1395 kbps No No
Medium Bit
2 2790 kbps No No
Rate
3 4185 kbps No No
1 2000 kbps No No
High Bit Rate 2 4000 kbps No No
3 6000 kbps N/A Moderate
Number of Aggregated bit rate Environment
Connections
1 8 Mbps PLC network
Variable Bit Rate
1 8 Mbps Fast Ethernet
Performance of elastic data traffic
Number of Connections Average bit rate
Elastic data 1 6.21 Mbps rate
traffic 2 6.15 Mbps rate
3 6.27 Mbps rate
Performance of combined delay sensitive traffic and elastic data traffic
Connections Aggregated bit rate
Hybrid data One ftp connection and One video 6.26 Mbps rate
traffic service
Two ftp connection and One video 5.92 Mbps rate
service

Performance of Elastic Data Traffic

The occurrence of the momentary video freezing phenomenon during playback of

variable bit rate streaming is likely because the aggregated data rate was close to or

exceeded the PLC network capacity. To understand the real throughput of a PLC network,

we conducted another experiment A 215,502,106 byte file was placed on the server

running an FTP daemon (The file size was chosen to minimize hardware uncertainty and

human error.) Client computers made FTP requests for the file. We tested different

numbers of FTP connections, up to 3, using individual client machines in our PLC

network. The experimental results are also given in Table 2-4.

The aggregated traffic in the table is calculated by adding all observed data rates of

all connections. The experimental results show that the real PLC network performance is

about 6 Mbps. When there is only one FTP connection, the observed throughput is 6.21

Mbps. By our analysis, one TCP connection will not fully utilize the PLC network,

because the server has to stop if no ACK packets are received from the client.

Aggregated traffic volume decreased as the number of connections increased from

1 to 2. This phenomenon is because of the ACK packets and the packet overhead increase

as the number of connections increased. Although the network utilization improves, the

improvement cannot compensate for the loss due to these overheads.

When we increased the number of connections from 2 to 3, the PLC network had

the highest throughput of 6.27 Mbps. This is because the network utilization increased as

the number of connections increased which compensated for the packet overhead and

These experimental results explain the phenomenon of momentary DVD video

freezing playback. The requested bandwidth for DVD streaming exceeded the maximum

bandwidth the present PLC network can provide.

Performance of Combined Delay Sensitive Traffic and Elastic Data Traffic

Although we could not explore the QoS service and packet priority provided by the

real PLC network, we were eager to learn the effect of mixed traffic on the PLC network.

This experiment was conducted as follows: The Eile server provided two services: one for

streaming video with bit rate 550 kbps, and the other one for fie transfer with a fie size

of 215,502,106 bytes. PC-1 requested streaming video while PC-2 and PC-3 requested

the Eile transfer. Each experiment lasted 285 seconds (that is, the length of the video file),

after which both video player and FTP client are forced to stop. Table IV shows our

experimental results.

When the number of FTP connections increased, the observed data rate decreased

as was the case in the previous experiment. However the overall average data rate was

comparable to the case of multiple FTP traffic.

Conclusion

The emergence of Information Appliances (IA) for the smart homes of the future

will undoubtedly make our lives much more comfortable than ever. However, the

infrastructure that supports multimedia traffic and conventional elastic data traffic for

communication among IA devices is a critical component of a smart home.

We advocate power line as the infrastructure for smart homes based on the

convenience of the power sockets and the layout of the power line network existing in

every home. At present, 6 Mbps of bandwidth was measured through real-world PLC

network experiments. Our studies showed that the PLC network can provide 3 low bit

rate or 3 medium bit rate multimedia streams concurrently with no packet drops and

jitters. It also successfully delivered one low bit rate multimedia data stream and 2 large

ftp file transfer concurrently with no packet drops and jitters.

In this chapter, we discussed only the PLC networks for communication 0I irlhin the

smart home, but the ultimate goal of PLC network could be the ability to connect to the

Internet without dialing up to an ISP server, entirely using electrical wiring only. This can

be illustrated as in Figure 2-3.

Private home networks are connected to substations, in which a PMTS (Powerline

Modem Terminal Service) connects PLC networks within homes to the Internet backbone.

The PLC network gateway for a private home network could be installed in the fuse box

of that home and then it could be connected to one or more repeaters. Repeaters are for

increasing signal strength when the signals level fall below some value.

PowerPlant Substation

Figure 2-3: Connecting PLC networks to the Internet

We expect to see higher data rates in power line networks in the future as signal

modulation technologies improve; however, issues like network security and the network

characteristics with a large number of nodes need further development. Further research

on these issues is of critical importance when power line networks are applied to offices

and large multi-user buildings.

CHAPTER 3
A COMPARATIVE PERFORMANCE STUDY OF WIRELESS AND POWER LINE
NETWORKS

Local Area Networks based on the IEEE 802.11a/b wireless networking standards

and emerging Power Line Communication (PLC) standards are attractive for establishing

networks with "No New Wires" for in-home and business applications. This Chapter

presents a theoretical performance comparison of the 802. 11 a/b and the Home-Plug 1.0

PLC protocols. We also presents comprehensive comparative field test results addressing

such issues as coverage, channel stability and reliability as well as the associated

implications on the capability of these technologies to provide QoS support for

multimedia traffic in typical residential settings.

Introduction

Candidate networking technologies for providing convenient and widespread

residential and SOHO networking services may be categorized as Wireless Networks,

Wired Networks and No New Wires Networks. Hughes S, and Thorne D.J. [2] did an

extensive study of various infrastructure options and technologies appropriate for home

networks. Below, we give a short discussion of networks in the above three categories.

Wireless Networks such as 802.11x, BlueTooth, and HomeRF can be constructed

by installing multiple interconnected wireless access points (WAP) and base stations

within target areas. The best benefit of using wireless networks is the freedom to move

around while maintaining network connectivity. Blue-Tooth technology is targeted at

personal communications and the coverage is expected to be limited. On the other hand,

though HomeRF has been on the market for a few years, it is not yet widely accepted.

Thus the most interesting and widely accepted wireless networking technologies are the

802.11lx family. 802.11lb operates in the 2.4 GHz band and provides a maximum data rate

of 11 Mbps; 802.11a supports speeds of up to 54 Mbps and operates in the 5 GHz band.

Standards for the newer IEEE 802.11lg, which should provide data rates up to 54 Mbps in the 2.4

GHz band, have not been finalized, and equipment was not available for testing.

For Wired Networks, a comprehensive Ethernet network can be constructed by

installing special UTP-5 cabling. While the stability and the security of wired networks

are guaranteed, installing new wires in existing home or other buildings may be costly,

negating the low cost of the network interface cards.

For the No New Wires Networks category, there are phone line networks, cable

networks, and power line networks. Using the existing phone line as an infrastructure, as

in Frank E.H., and Holloway J. illustrated [2], may seem attractive, but it is limited by

available phone sockets in a home. Home Cable Network Alliance (Home-CNA) [10],

established in June 2001, and proposes a home network infrastructure using existing

coaxial TV cable. There is as yet no standard for HomeCNA and it also suffers from the

maj or drawback of limited convenient connection points.

Power Line Communication (PLC) networks such as HomePlug[3] were introduced

to the U.S. consumer market in May 2002. European PLC networks have been deployed

in recent years. With multiple outlets in almost every room, residential power lines are

already the most pervasive network in the home or small office. The HomePlug 1.0 PLC

standard supports PHY data rates of 14 Mbps and is thus comparable to the 802.11b

declared data rate.

A maj or obj ective of this paper is to conduct a real-world performance study of the

capabilities of wireless (IEEE 802. 11b and 802. 11a) networks and PLC networks based

on the HomePlug 1.0 standard. Our interest is to determine the relative performance of

these technologies.

This chapter presents a comparative analysis of the TCP performance of power line

networks and wireless networks using actual measurements on HomePlug 1.0 compliant

PLC networks and 802.11a/b compliant wireless networks. The tests were conducted in

20 houses ranging in area from 1500 to 5000 sq. ft. with an average area of 3000 sq. ft.

The paper presents qualitative theoretical and measured throughput performance for

802.11a/b and HomePlug 1.0 PLC. Other issues like the relationship between QoS and

channel stability as well as overall coverage are also discussed.

The next section briefly describes the HomePlug 1.0 protocol. Section 3-3 presents

a theoretical performance analysis of IEEE 802.11la/b and HomePlug 1.0. Section 3-4

describes the experimental setup while Section 3-5 gives our field test results. A

summary is given in Section 3-6.

Homeplun 1.0 Protocol

The parameters and details of 802.11lx protocols are well documented in the

literature and Internet publications [l l]. Here, we briefly describe the HomePlug 1.0

standard.

PLC Environment

Power lines were originally devised for distributing electrical power using the

frequency range of about 50-60 Hz. The use of this medium for high speed

communications presents some technically challenging problems. Electrical noise from

appliances and the uncontrolled nature of the wiring result in severe signal distortions.

The PLC channel is made up of different conductor types; therefore a variety of

characteristic impedances will be encountered. Further, the network terminal impedance

will tend to vary with frequency and time as the consumer's load pattern and load types

vary. Impedance mismatch causes multi-path effects resulting in deep notches at certain

configuration dependent frequencies. These channel imperfections make signal

transmission over a power line very difficult [5].

Reliable data communication over this hostile medium requires powerful Forward

Error Correction (FEC) coding, interleaving, error detection and Automatic Repeat

Request (ARQ) techniques, along with appropriate modulation schemes as well as a

robust Medium Access Control (MAC) protocol. The lack of affordable processing

techniques needed to overcome the harsh power line environment resulted in limited

success of power line communications in the past. However, both the advances in the

ASIC density and speeds, and the advancement of signal modulation, processing and

error control coding techniques now make power line communication possible.

HomePlun 1.0 PHY

To overcome the hostile PLC environment, Orthogonal Frequency Division

Modulation (OFDM) with a Cyclic Prefix (CP) was adopted by the HomePlug 1.0 PLC

standard. Using OFDM has many benefits. For example, it exhibits excellent mitigation

of the effects of time-dispersion, provides excellent Inter-Channel Interference (ICI)

performance, and is good at minimizing the effect of in-band narrowband interference.

OFDM splits available bandwidth into many small frequency bands called sub-carriers,

then may mask out unusable subcarriers and apply the best modulation and coding

methods to the usable subcarriers. This approach is used by HomePlug 1.0. A more

schemes for each sub-carrier. In either case, OFDM can adapt bandwidth/data rates

according to channel conditions.

Unlike 802. 11, the bandwidth in HomePlug 1.0 can vary from 1 Mbps to 14 Mbps

practically continuously according to the channel conditions Active HomePlug 1.0

nodes perform channel estimation at least once every 5 seconds. This feature allows the

PLC network to maximize its data rate adaptively.

A preamble and frame control form delimiters used for synchronization and for

control. The frame control of start of frame, end of frame, and response delimiters all

include delimiter type2, and contention control information. In the start of frame delimiter,

the frame control field includes the tone map information needed by the receiver to

decode the rest of the frame, and a length field. The end of frame delimiter contains

priority information used for contention control. Response delimiters contain information

that allows a sender to verify that the response was indeed sent in response to the frame it

just transmitted. An end of frame gap (EFG) of 1.5 Cps is inserted between the frame's

frame check sequence (FCS) and the end delimiter to allow for processing.

HomePlun 1.0 MAC

The HomePlug 1.0 Medium Access (MAC) protocol is a modified CSMA/CA

(Carrier Sense Multiple Access / Collision Avoidance) protocol with priority signaling.

HomePlug 1.0 devices operate in an ad hoc mode in the sense that devices communicate

with each other freely, without any centralized coordination.

SThere are 139 distinct data rates in that range according to number of usable carriers, modulation methods, and
coding rate.
SResponse expected is indicated in the delimiter type, and Response delimiter does not have a response expected/not
expected
indication.

The frame structure and protocol of HomePlug 1.0 is depicted in Figure 3-1.

35 84 us 35 84 us 35 84 us 35 84 us xn 313 5 ~ 1489 5 us 26us 72us

CIFS Re1 n0 R ln1 Contention Data RIFS ACK

Preamble er Frame Body PAD FCS Erae Preamble Peml

25bits 17bytes Variable Length Bytes 1 5us 25bits RIFS

Figure 3-1: HomePlug 1.0 frame structure and protocol

The HomePlug 1.0 standard uses different terms and stages for inter-frame spacing

and for the contention windows than 802.11lb. The RIFS shown in the figure is "Response

Inter-Frame Spacing." Unlike 802.11, there is no SIFS (Short Inter-Frame Spacing)

between continued frames. Rather, a frame control bit is used to indicate the desire of a

station to continue to send data, allowing preemption only by higher priority traffic. The

spacing between the last frame and the incoming frame is CIFS (Contention Window

Inter-Frame Spacing).

HomePlug 1.0 provides four priority classes CA3, CA2, CAl and CA0 from

highest to lowest. Priority resolution is done by asserting signal of the priority level in the

PRO and PR1 slots. For example, to send a CA2 packet, the PLC device should assert a 1

in PRO, causing any node with CAl traffic to defer, and not assert 1 in PR1 as it would

do otherwise. Nodes with CA3 data assert a 1 in both priority slots and CA0 in neither.

This effectively resolves contention between different priority classes. Contention within

the same priority class is resolved during the contention period.

The contention period is a contention period. The contention period is a form of

CSMA/CA with a priority dependent backoff window size schedule. For the lower two

priority classes, it is 8-16-32-64 slots, while it is 8-16-16-32 slots for the two higher

priority classes. On collision, the range of contention slots over which a transmission is

started is increased according to the schedule. Aside from starting with a smaller range (8

slots compared to 32 slots), a major difference from the IEEE 802.11 standard is that

when a HomePlug 1.0 node defers (detects another node's transmission in an earlier slot),

it uses this information to back off, but less aggressively than in the case of a collision.

This technique serves to reduce costly collisions further. For protocol details, please see

the HomePlug 1.0 Specifieation [3].

Theoretical Performance of 802.11la/b and Homeplug 1.0

To compare the protocol performance of 802.11a/b and HomePlug 1.0, we first

analyze the theoretical performance differences between them. In next section the

protocol analysis to calculate the theoretical performance of 802.11la/b and HomePlug 1.0

is presented. The analysis assumes that a single station is continuously transmitting

frames with 1500 bytes of payload over the medium.

802.11a/b Theoretical Network Performance

In the absence of competition, an 802.11lb node picks a contention slot between 0

and 31, and starts transmission then. The average contention period delay for a packet

without competition is 31/2 = 15.5 slots or 310 Clsec. The transmitting node will start to send

data and wait for the receiver's acknowledgment.

Each frame is made up of a PLCP header, a MAC header, a DATA Hield and a CRC

Hield. If it is an ACK frame then the DATA Hield is not present.

In practice, the maximum data payload sent via the 802.11a/b is limited to the

Ethernet maximum of 1500 bytes. Table 3-1 summarizes the MAC throughput and

efficiency for IEEE 802.11x protocols at various data rates. Packet fragmentation and

MAC level packet concatenation are not considered.

Table 3-1: 802.11la/b MAC throughput with payload 1500 bytes
Technology PHY Data Rate MAC Throughput
802.11lb 1 Mbps 0.91 Mbps
2 Mbps 1.73Mbps
5.5 Mbps 3.99Mbps
11 Mbps 6.38Mbps
HomePlug 1.0 1 Mbps 0.70Mbps
2 Mbps 1.74 Mbps
5.5 Mbps 3.77 Mbps
11 Mbps 8.08 Mbps
14.1 Mbps 8.08 Mbps
802.11a 6 Mbps 5.38 Mbps
9 Mbps 7.78 Mbps
12 Mbps 10.02Mbps
18 Mbps 14.1 2Mbps
24 Mbps 17.61 Mbps
36 Mbps 23.74 Mbps
48 Mbps 28.47 Mbps
54 Mbps 30.80 Mbps

From Table 2-1, the maximum throughput is 6.38 Mbps for 802.11lb and 30.8 Mbps

802.11la, representing efficiencies of 58% and 57% respectively. At IMbps the MAC

efficiency of 802.11lb is as high as 91% assuming there is no packet fragmentation.

Performance Analysis of HomePlug 1.0

From Figure 2-1, the data transmission time ranges from 313.5 to 1489.5 Cps,

however to transmit a 1500 byte payload at the maximum data rate, 120 symbols are

required, taking 1153.5 Cps. Excluding physical level control overhead results in 1008 ys

for actual data transmission. The maximum data payload size is limited to the smaller of

1500 bytes and 1344R-OH bits, where R is the physical data rate, and OH is the number

of overhead bits. Each Ethernet frame incurs an overhead of at least 120 bits for

encryption and integrity checking for the corresponding service block. Segment bursting

allows a station to send all the segments associated with a service block consecutively,

avoiding contention unless it is preempted by a station with higher priority traffc. A

service block is broken into physical layer segments, each of which has 19 additional

be a multiple of 20 symbols long, up to 160 symbols maximum, which further

complicates throughput analysis.

The initial contention window size is 8 slots, so the average contention delay

without competition is 3.5 slots. To successfully deliver a data packet of 120 symbols

takes

35.84 Cps + 35.84 Cps + 35.84 Cps + 35.84 Cps X 3.5 + 1153.5 Cps + 22Cps + 26Cps=

1484.86Cps

The maximum physical layer data rate is 14.18 Mbps, thus the maximum

throughput is

1500 X8 bits/1484.86Cps = 8.08 Mbps.

The efficiency of HomePlug 1.0 at the maximum data rate is 57%. The 70%

effciency of HomePlug 1.0 at PHY rate of IMbps is due to the limits of the data

transmission time to a maximum of 1484.5Cps in order to provide better latency and jitter

QoS parameters for higher priority traffc.

Although we can get up to 8 Mbps maximum MAC throughput in theory, the

maximum measured TCP throughput in our field testing so far is 6.3 Mbps, matching

earlier HomePlug 1.0 simulation results [12].

Experimental Setup

To understand the real world performance of IEEE 802.11a/b and HomePlug 1.0,

we conducted field tests in 20 houses located in the Gainesville, Ocala, Orlando, and

Belleview areas of Florida. The choice of the houses used in the tests were in the mid-to-

large size (1500 sq. ft. to 5000 sq. ft.), since larger houses provide a better range on the

performance parameters of interest.

The equipment used in this test included the following.

1. AP Server: A Sony notebook with a 700 MHz Pentium III processor and 128k
RAM running Windows2000

2. Mobile Station: An HP notebook with a 500 MHz Pentium III processor and 128k
RAM running Windows2000

3. Linksys HomePlug 1.0-based Powerline-to-Ethernet bridges

4. Netgear[13] IEEE 802.11lb Access Point and PCMCIA Card

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. MIM can be

expected for connection from a node to an access point. Since these tests had the AP

server connected to the access point via Ethernet, there with no other contention possible,

the test results should represent the best case scenarios with respect to this aspect.

Experimental Method

The TCP throughput and distances were measured for various locations ofAP and

Mobile stations inside the house. The AP Server was located close to a phone outlet or a

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 toWindows sockets, was used.

The TCP buffer size was chosen to be 11680 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 involves 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

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

SThe complete data set from our field tests is available on request.

capture was also obtained to observe TCP stability. All the procedures were automated

and required minimal human operation.

Results

The amount of data collected is too large to be presented in detail in this paper, so

only a summary of the most interesting findings are presented The location, size and age

of each of the houses where testing was conducted is shown in Table 3-2 to 3-4.

Table 3-2: List of the houses tested and connectivity

House Size
(Square
Feet)
1460
2000
2030
2100
2300
2700
2700
(two floor)
2700
3000
3000
(two floor)
3000
3150
3500
3600
3600
3600
3900
4000
4200
5000
(two floor)

HomePlug
1.0
Connectivity
100%
100%
100%
100%
100%
100%
100%

100%
100%
100%

100%
100%
97.6%
100%
100%
100%
100%
81%
100%
100%

House Age
(Years)

802.11b
Connectivity

802.11a
Connectivity

Location

Gainesville
Gainesville
Gainesville
Gainesville
Ocala
Gainesville
Gainesville

Gainesville
Gainesville
Gainesville

Ocala
Gainesville
Ocala
Ocala
Orl ando
Belleview
Gainesville
Gainesville
Orl ando
Ocala

90%
90%
75%
100%
42.9%
61.9%
57%

20%
91%
71.4%

54.2%
55.56%
69%
41.7%
50%
42%
56%
9%
18.75%
17%

100%
100%
100%
100%
78.6%
100%
100%

100%
100%
100%

100%
95.8%
100%
100%
100%
92%
100%
84%
50%
50%

IEEE 802.11la Indoor Performance

The performance and coverage results of IEEE 802.11la are depicted in Figure 3-2.

Connectivity versus House Area

100%

p 80%
e 60%
S40%
20%

0%
1400 2400 3400 4400

Area(sq. ft.)

Figure 3-2: IEEE 802.11la indoor connectivity

Figure 3-2 shows the connectivity (that is, 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 sq. ft.) houses. For

larger houses (>4000 sq. ft.) the connectivity decreased to 20 %.

Figure 3-3 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 3-4 shows the percentage of links that exceed the throughput values

indicated on the X-axis. Note that the maximum IEEE 802.11la throughput obtained from

the product being tested was larger than those expected from theory. This could be

CCC

because of manufacturer-specific proprietary enhancements like the use of higher level

modulations.

Throughput vs Distance

40.00

30.00

20.00

10.00

0.00

0 20 40 60 80

Distance e(ft. )

Figure 3-3: IEEE 802. 11a indoor throughput as a function of distance

100

100%

80%

60%

40%

20%

0%

0 10 20

Throughput(1Vbps)

Figure 3-4: IEEE 802.11la indoor percentage of links

In summary, the statistics show that

1. IEEE 802. 11a failed on at least one link in 19 of the 20 houses tested,
2. 802. 11a failed to connect in 45% of the links that were tested,
3. 802.11la showed close to zero connectivity at distances larger than 50 feet.
For shorter distances, 802.11la provided excellent throughput in most cases.

IEEE 802.11lb and HomePlug 1.0

To facilitate comparison, the performance and coverage results of IEEE 802.11b

and HomePlug 1.0 are shown in Figure 3-5 to 3-8. Figure 3-5 shows the connectivity as a

function of house area. Both technologies show high connectivity for houses of size less

than 4000 sq. ft. For houses larger than 4000 sq. ft., the connectivity for IEEE 802.11b

dropped dramatically to 50% in both of the houses tested, while HomePlug 1.0 continued

to show high connectivity.

Connectivity versus House Area

60%

'- 40%

40%

1400 2400 3400 4400

Area(s q. it.)
Figure 3-5: IEEE 802.11lb and HomePlug 1.0 indoor connectivity comparison

Figure 3-6 shows a scatter plot of throughput as a function of distance. IEEE

802.11lb 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.

Throughput versus Distance
9.00
o HomePlug 1.0
a + 802.11b
6.00 a ca B a

0.00

0 20 40 60 80 100

Distance(ft.)

Figure 3-6: IEEE 802. 11b and HomePlug 1.0 indoor throughput comparison

Figure 3-7 shows the percentage of links that exceed the throughput value depicted

on the X-axis. Our experiments showed that the overall coverage of 802.11lb was 92%.

The maximum throughput observed in field testing was 5.13 Mbps. Figure 3-7 shows that

around 70% of the connections operated at more than 4 Mbps and 10% above 5 Mbps.

For HomePlug 1.0, the overall coverage is 98%. The maximum throughput observed in

testing was 5.98 Mbps. For HomePlug 1.0, 58% of the connections operated above 4

Mbps and 38% had throughput above 5 Mbps.

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 suboptimal. 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 802.11lb when channel conditions are favorable.

Percentage Of Links vs Throughput (MIM)

100%
~1 ---. ... HomePlug 1.0
~" 80% -- --. ------- 11b

O 60%

S40%

20%

0.00 2.00 4.00 6.00 8.00

Throughput(Mbps)

Figure 3-7: IEEE 802.11lb and HomePlug 1.0 indoor percentage of link versus throughput
(MIM) comparison

It should also be noted that in this experiment, throughput was measured between

the access point to the Mobile Station in a modified Infrastructure mode. However, this

may not always be the way IEEE 802.11b stations communicate with each other. IEEE

802.11lb networks can be configured in either ad hoc mode or Infrastructure mode (IM).

In the 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. Further, 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 (that is, 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 Infrastructure Mode (IM) throughput

from the MIM link throughput data which was 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) fTOm access point to the TCP destination. The

aggregate IM throughput then can be obtained by assuming a fixed packet size ofx bits is

transmitting through two links with speeds R1 and R2. The total time to transmit this

x x R, x R,
packet will require --+- second. Thus throughput can be calculated by
R, R, R, + R,

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 3-8 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 compared to IEEE 802.11lb stations in IM.

From the statistics, we make the following key observations

1. HomePlug 1.0 had a larger maximum throughput that 802.11b(about 1 Mbps
larger).
2. On 60% of the links HomePlug 1.0 performed better than 802. 11b links in MIM,
3. On an average basis, HomePlug 1.0 gave approximately 0.2 Mbps higher TCP
throughput than 802.11lb in MIM,
4. In 6 of the 20 houses tested, IEEE 802. 11b failed on at least one link,
5. In 2 of the 20 houses tested, HomePlug 1.0 failed on at least one link,
6. On an average basis, HomePlug 1.0 gave approximately 2.3 Mbps higher TCP
throughput than 802.11lb 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 modified infrastructure mode,

which would be typical for Internet access. In infrastructure mode, HomePlug 1.0 was

estimated to have throughput about 2.3 Mbps 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.

Percentage Of Links vs Throughput (IM)
& 100%
HomePlug 1.0
S 80% ---- 802. 11b
0 1
S60%
O
S40% \i

a 20% -t

0.00 2.00 4.00 6.00 8.00

Throughput(Mbps)

Figure 3-8: IEEE 802.11lb and HomePlug 1.0 indoor percentage of link versus throughput
(IM) comparison

QoS algorithms usually deal with admission control and resource allocation.

Admission control is concerned with the acceptance of new connections, while resource

allocation deals with packet-level throughput, delay, and fairness. In either case,

predictability is desirable.

Previous studies [14] showed that high channel error rate will reduce the effective

bandwidth available for applications, thus negatively affecting application performance.

This problem is even more severe for multimedia applications, which typically have

bandwidth, delay, and jitter requirements for effective operations; it is important for them

that the link remains stationary. However, the 802.11a/b displayed link instability when

the PHY data rate was low. This section studies TCP link stability from the realtime

capture of 802.11la, 802. 11b, and HomePlug 1.0 packets.

During testing, when connection speeds were less than 3 Mbps, the wireless

network became unstable. This might have been due to problems in rate adaptation. The

802.11lb standard indicates around a 4 dB difference in signal strength between 11 Mbps

and 2 Mbps mode. Since the signal strength changes continuously with time (for example,

due to movement of people), PHY rate adaptation may cause packet drops that make the

wireless network unstable when using TCP. During our tests, few links were found with

throughputs in the 1 to 3 Mbps range. To find out the cause of this phenomenon, we used

real time packet capture to monitor the TCP link stability. TCP throughput was measured

at 100 msec intervals.

Figure 3-9 and 3-10 shows real time captures of typical high speed and low speed

links of 802.11a. High speed links have a mean data rate of 20 Mbps, while it is 5 Mbps

for low speed links. Note that the link performance from Mobile station to AP server and

from AP server to Mobile station (separated by a 3 second delay) both are shown in the

figures.

For high speed links, Figure 3-9 shows irregular throughput dropouts during

transmission. For AP to mobile link, the throughput differences from one sample to the

next can be as high as 22 Mbps. This kind of behavior is exhibited by TCP when packets

are dropped. These instabilities would make it very difficult to guarantee QoS. Similar

instability was observed for the link from Mobile to AP. A maximum throughput

difference of 23 Mbps was observed in this case.

802.11a

30 AP->Mobile Mobile->AP

20

00:35.5 00:45.5 00:55.5 01:05.5 01:15.5 01:25.6 01:35.6 01:45.6 01:55.6 02:05.6

Time(M:S)

Figure 3-9: IEEE 802.11la high speed real-time capture

When the data rate of the link was low (Figure 3-10), the throughput also displayed

large variations. A maximum throughput difference of 13 Mbps was observed. For

Mobile to AP, a maximum throughput difference of 10 Mbps was observed.

Figure 3-11 and 3-12 shows the real time capture for a typical high speed and low

over 4 Mbps; low speed links for 802.11lb are links with data rate lower than 2 Mbps. The

same criteria were applied to the HomePlug 1.0 networks.

802.11a TCP Stability Low Speed Link

Mobile To AP
25.00AP To Mobile

25.00-

20.00

r 12.00 AP- MobileC

10.00

2.009 90 00~

Time(M :S)

Figure 3-11: IEEE 802.11b high speed real-time capture

Figue 311 sowsthatthe802.11b TC tblinks ig wpered tyical oesabeta h

the oher ne isin te lin of obil to A as ndicaed b cirles nth iur.Tistp

of dropout often appeared in other captures. In this figure, the maximum throughput

differences observed was about 6 Mbps.

802.11b TCP Stability Low Speed LinkI
AP->Mobile

4.5~~~ --- --------Mobile->AP

2 .5 -1

1 .5--

0.

39:25.2 39:45.2 40:05.3 40:25.3 40:45.4 41:05.4 41:25.5 41:46.5 42:06.5

Time(M:S)

Figure 3-12: IEEE 802.11lb low speed real-time capture

Figure 3-12 shows the real time capture for a typical low speed of 802.11lb link. For

AP to mobile link, the throughput differences can be as high as 3 Mbps. For Mobile to

AP link, A maximum throughput difference of 3 Mbps was observed in this case.

Note that the throughput variation is critical at low data rates. User experience will

be poor for links with such throughput variations. For example, under these marginal

conditions, a file transfer might halt due to excessive packet drops. Applications can

crash or show strange behaviors an extremely unpleasant situation for the user. These

links can be considered equivalent to no-connects, in the sense that users will not use

The HomePlug 1.0 TCP link stability is depicted in Figure 3-13 and 3-14. The

figure shows that HomePlug 1.0 provides a fairly stable TCP link on high speed

50

connections. There are no dramatic dropouts during the test. The maximum variation we

observed in the figure was 3.6 Mbps for high speed links.

Powerline TCP Stability High Speed Links

12

10

8-

6

4

2

0
48:59.9

49:19.9 49:40.0 50:00.0 50:20.0 50:40.1

Time(M:S)

Figure 3-13: Homeplug 1.0 high speed real-time capture

HomePlug 1.0 TCP Stability Low Speed Linke

1 101

201 301 401 501

Time(M:S)

601 701 801

Figure 3-14: Homeplug 1.0 low speed real-time capture

Figure 3-14 shows an interesting stair-wise rate adaptation on the transmission

from Mobile to AP. This effect is a manifestation of the channel estimation mechanism.

Noise over power lines tends to vary with the line cycle. In HomePlug 1.0 stations, the

channel estimation is done asynchronously using a small channel estimation packet. Thus

the estimated data rate varies depending on when the packet arrives with respect to the

line cycle. Thus different channel estimations (typically done every 5 seconds) will result in

a different throughput, and hence this stepped behavior. However, the TCP links can be assumed

to be stable between channel estimation cycles.

During our testing, we observed that IEEE 802.11b links that operated below 1

Mbps are highly unstable and are frequently marked by disconnects.

Discussion and Conclusion

The main goal of this chapter is to conduct a practical and theoretical comparison

of the IEEE 802.11x and HomePlug 1.0 protocols and their capabilities in providing

networking functionalities. This was done through theoretical analysis and by a thorough

Hield test conducted in 20 houses to obtain the performance of IEEE 802.11b/a and

HomePlug 1.0 products.

From the theoretical results shown in Table I, it can be derived that HomePlug 1.0

and 802.11lx have similar maximum efficiency. The significantly higher maximum PHY

data rate of 802.11a would indicate that it should perform better than the other two

standards, but in Hield tests its coverage was not as good. In the Hield tests, the 802.11x

products were configured as they came out of the box, and in some cases the wireless

links may have used the RTS/CTS (Request To Send/Clear To Send) mechanism needed

by 802.11lx to handle hidden nodes. Use of RTS/CTS can have significant effects on the

most significant at high data rates, theoretically costing up to 10% in throughput

performance. However, use of RTS/CTS can improve performance when there are

collisions, as one can expect even on a single link when there are asynchronous,

bidirectional exchanges (as TCP does by acknowledging received segments). 802.11x

infrastructure mode (IM) also degrades performance, as opposed to modified IM (MIM)

or ad hoc mode. In Hield tests, ad hoc mode was found to be nearly unusable over any but

short distances. Fortunately, MIM would be typical for connection to an internet access

point, which one might expect to bear the greatest amount of traffic. The wireless

protocols have an advantage that these tests cannot show, is their use of multiple channels.

With three channels available, one could expect the wireless protocols to perform better

under congested conditions, whereas the PLC protocols use all of the available bandwidth

for a single channel.

Coverage is much harder to predict theoretically, but Hield tests showed that

HomePlug 1.0 had the best coverage, followed by 802. 11b, with both trailed significantly

by 802.11a. The latter only functioned over line-of-sight distances under 50 ft., and had

complete coverage in only one of the 20 houses tested. IEEE 802.11b had severe

coverage problems in houses over 4000 sq. ft., and had non-connects or marginal links in

more than half of the houses tested. HomePlug 1.0 provided 100% coverage in all but

two of the houses tested. This showed convincingly that the absence of an RTS/CTS

mechanism in HomePlug 1.0 is not likely to be a problem for single home deployments.

Whether or not it is an issue for multiple residences serviced by the same transformer

remains to be seen.

Throughput showed more interesting behavior. The throughput of IEEE 802.11a

was nearly always very high on links less than 50 ft. line of sight, but dropped to zero for

links longer than this. 802.11lb and HomePlug 1.0 showed crossovers in the percentage of

links with data rates meeting some target rate. With its better coverage due to a more

robust PHY and MAC, HomePlug 1.0 did better at meeting minimal data rates, but was

surpassed by 802.11b at rates around 4 Mbps. For data rates of 5 Mbps and higher,

HomePlug 1.0 retook the lead due to its slightly higher maximum data rate. HomePlug

1.0 also showed a much more gradual curve, due to its greater range of PHY data rate

selections for adapting to the channel.

The continuum of link speeds also allowed HomePlug 1.0 to exhibit greater link

stability (as measured by short term variability in TCP throughput). Effects of channel

estimation at 5 second intervals and variability of the channel due to the 60 Hz line cycle

were evident in these tests. IEEE 802.11b showed greater variability in both low- and

high-speed links, most likely due to its sparser choices in adaptation to channel

conditions. 802.11a had tremendous and frequent speed fluctuations, which brings into

question its ability to offer QoS guarantees for multimedia applications.

Both PLC and wireless technology have scope for improvements over the exist

standards as evaluated in this paper. For PLC networks, larger bandwidth, the use of

higher order modulation, more powerful forward error correction technique, and

improved channel estimation can sub stanti ally improve performance. Wireless

technologies can also invoke similar enhancements along with larger transmit power and

antenna diversity to achieve significant improvement. However, increasing crowding in

the 2.4 GHz ISM bands used by 802.11b/g are likely to degrade their performance in

locations where they must coexist with competing transmitters (including BlueTooth,

HomeRF, 2.4 GHz phones, microwave ovens, etc.).

Despite its problems, there are still situations for which wireless technologies are

outlets will require wireless connectivity. For nomadic users who are able to plug into the

home power distribution system, however, PLC offers a robust, stable, and speedy

alternative. PLC solutions will be even more desirable for providing QoS support for

multimedia applications with the future and emerging PLC technologies offering data

rates in excess of 50 Mbps.

CHAPTER 4
PERIODIC CONTENTION FREE MULTIPLE ACCESS FOR POWER LINE
COMMUNICATION NETWORK S

Introduction

Recently, applications over Power Line Communication (PLC) networks have

drawn much interest in academe and industry, not only because of their convenience

(connecting PLC capable devices requires no new wires), but also because all electrical

devices have to connect to a power outlet eventually. This technology makes

implementing a digital home entertainment center more realistic than ever HDTVs in

different rooms are now able to share digital content from one set-top box without

rewiring or setting up wireless access points. MP3 players can access music data through

PLC networks from different rooms playing different music. PLC networks make the

smart home possible [12].

However, PLC technology is still evolving and many issues remain unsolved. The

hostile environment of PLC channels makes reliable data transmission difficult. Much

effort is required to ensure data transmission is correct and efficient. The HomePlug

Alliance set standards for 14Mbps class data transmission. This resulted in a variety of

PLC devices for computer-oriented network communications. Their performance and

reliability is comparable to wireless networks [15]. However, the HomePlug 1.0 protocol

is not suitable for video playback because of its limited network throughput.

To support multimedia streaming for homes, the HomePlug Alliance is now

developing the standard for a second generation of PLC devices capable of delivering

multiple HD-Videos through newly designed chip sets called HomePlug A V, which

supports up to 200Mbps raw data rate. The goal of the HomePlug AV is to make PLC

devices capable of delivering two hours of HD-Video without video frame drops, while

simultaneously delivering one or more data streams of various data rates and traffic types.

To make above mentioned goals possible, the efficient cooperation of high speed PHY

and MAC protocols becomes important. However, current existing MAC protocols do

not provide such functionalities and they are not suitable for HomePlug A V. The need for

a new protocol is urgent.

The PLC channel is known for its hostile nature in transmitting electrical signals.

Protocols designed for other media may not be suitable for PLC. PLC channels are in

some ways similar to wireless channels both of them face hidden node problems, near-

far effects and other channel imperfections. However, PLC network nodes tend not to

move. It is unlikely that simply applying protocols designed for another medium would

result in good performance in the PLC environment; the overhead may be too high or the

assumptions about noise may be too optimistic for PLC networks.

In light of PLC's unique characteristics, we developed a new protocol Periodic

Contention-Free Multiple Access (PCF/MA). PCF/MA is an explicit R-ALOHA-like

protocol specifically designed for the PLC network we propose an RTS/CTS-like

scheme in the reservation stage to prevent hidden node problems, and a delayed NACK

mechanism to conquer near-far effect. Performance of the proposed protocol is evaluated

by event driven computer simulation and by mathematical analysis. The simulation

results show that 85Mbps MAC throughput under 100Mbps channel data rate can be

obtained, even when there are hidden nodes in the network. To provide smooth video

delivery, we propose a mathematical estimation of the required delay in playback time

and the amount of playback buffer with tight bandwidth reservation. Our simulation

shows that an 100Mbps channel can deliver up to 9 MPEG-2 video streams

simultaneously without dropping any video frames, however, using Modified CSMA/CA

(MCSMA/CA) on the same environment supports 7 video streams only because of its

unfairness, unpredictable behaviors and high overhead.

This chapter is organized as follows. A brief survey of existing protocols is given in

section 2. Section 3 provides the PCFM1A methodology, analysis, and performance

evaluation and simulation results. The discussion and conclusion is given in section 4.

Previous Works

Packet contention techniques such as Carrier Sense Multiple Access with Collision

Avoidance (CSMA/CA) and ALOHA find widespread use in data communications,

including the first generation PLC networks. Both have the ability to serve a large

number of terminals. While they function with little to no coordination, packet contention

results in unpredictable behaviors such as unfairness and possibly long delays, which

make them unsuitable for delay-sensitive packet delivery. To solve these problems, one

can use packet scheduling or reservation-based methods.

Reservation-based methods were designed to remove delay-sensitive, high-rate

connections from the random-access competition for channel time. The time domain is

partitioned into a reservation and data session; each session is then again partitioned into

slots. In these methods some slots are reserved for specific stations. Other stations are

restrained from using a reserved slot. Generally, these MAC schemes can be categorized

according to whether the reservation is done implicitly or explicitly.

Packet Reservation Multiple Access (PRMA) is a centralized and slotted multiple

access protocol that allows voice and data sources to share the same access channel at the

talkspurt level, targeted for wireless local area networks, and was originally proposed by

Goodman D. J., Valenzuela R. A., Gayliard K. T., and Ramamurthi B. [16]. PRMA

utilizes the speech on-off activity to improve bandwidth efficiency and system capacity.

Though PRMA is an implicit reservation-based algorithm, between talkspurts stations

must contend for reservation, leading to packet loss and degraded speech quality. It also

relies on central control to broadcast an ACK at the end of each slot, hence it may not be

appropriate for a home ad-hoc environment. A few modified PRMA protocols such as D-

PRMA were proposed to support mobile ad-hoc environments. These emphasize

talkspurt-level packet reservation without relying on a central entity. They also try to deal

with hidden/exposed node problems by asserting an RTS/CTS-like dialog between sender

Hidden node and near-far effects make implicit reservation difficult in PLC

networks. Impaired receivers may not agree on the current network state. Further, as the

PLC network is targeted to be an in-home networking infrastructure, it is not desirable to

have a central control device just for medium access, especially if the central control

device may become a performance bottleneck and a single failure point of a network.

Lack of a central control device makes collision detection even more difficult since PLC

devices are not able to detect signal collisions during transmission.

Explicit reservations like "Five-Phase Reservation Protocol" [17] try to implement

a dynamic parallel reservation with arbitrary scalable network size. The five phases

mentioned in the literature are: Reservation Request (RR); Collision Report (CR);

Reservation Confirmation (RC); Reservation Acknowledgement (RA); Packing and

Elimination (P/E) phase. However, this protocol makes assumptions inapplicable to the

PLC environment. Firstly, the protocol works on a 2 hops radius network, with perfect

timing. Secondly, near-far effects do exist in the PLC network especially when a robust

signal modulation is applied.

The above discussion leads to the new protocol design based on an explicit

reservation scheme. Robert's reservation scheme and R-TDMA are good candidates;

their merit is that a few slots are dedicated for reservation purpose and the rest of them

are for data transmission. However, these schemes lack distributed control, hidden-node

prevention and ignore near-far effects, which makes them inappropriate for PLC

networks. Careful re-design of Robert's and the R-TDMA protocols results in the new

PCF/MA protocol as stated in the next section.

Proposed PCF/MA Protocol

Though the power line channel is similar to a wireless channel, there are some

unique characteristics specific to the power line channel. Firstly, the characteristics of the

PLC channel in a home are in general steady over time rather than dramatically changing

as in the wireless channel, that is, while the attenuation of the signal may be affected by

near-by electronics in the short term, from a long term point of view the attenuation is

almost identical. This makes long-term bandwidth scheduling possible. Secondly, the

devices in PLC networks are quasi-stationary, that is, the stations in the PLC network are

not moving as constantly as devices in the wireless network, thus the bandwidth

fluctuations are much less erratic [15]. These two characteristics place PLC networks

between Ethernet networks and wireless networks in that they have the properties of a

Eixed network topology but with more noise and attenuation.

The Concept of PCF/MA

The PCF/MA protocol is a distributed, contention-free protocol which uses a two-

way handshake reservation process to establish TDMA slot assignments. The reservation

process for a given node only involves nodes within a one hop radius.

"I II

Figure 4-1: PCF/MA frame structure

Fig. 4-1 shows the protocol's frame structure. Time is divided into several TDMA

sessions with duration of Tyc s. Each session is partitioned into a Reservation Frame (RF)

and one or more Data Exchange Frames (DEFs). Before the RF is a Starting of TDMA

(SOT) frame issued by the leader station on the network that lasts 38.4 pc s4. Table 4-1

summarizes the parameters used in this research.

The RF is divided into R Reservation Slots (RS). An RS does not correspond to a

data slot but rather serves as a period of time for the making a reservation. This process

involves exchange of Reservation Packets (RP) and ACKs. An RP contains the Source

4 The parameters we used in this paper follows those ofHomePlug 1.0 as published in [19]

Address (SA), Destination Address (DA), Starting reserved Slot number (SS) and Total

of reserved Slots (TS) as depicted in Fig.4-2. To illustrate the reservation process more

clearly, we also depict the reservation flowchart in Fig.4-3.

Table 4-1: HomePlug 1.0 and PCF/MA parameters
HomePlug 1.0 PCF/MA MC SMA/CA
SYNC 38.4 pcls 38.4 pcls 38.4 pcls
FC 33.6 pls 33.6 pcls 33.6 pls
RIF S 26 pcls 26 pcls 26 pc s(SIFS)
DEL 72 pcls 72 pcls 72 pcls
Max MPDU Duration 1.5ms 1.5ms 1.5ms
CRS 35.84 pcls 200 pc s(RS) 35.84 pcls
DIFS 35.84 pcls 35.84 pcls
TDMA Session Length -Tyc s
Reservation Slots -R
SOT 38. 4 pc s(SYNC) 38.4 pls
Priority Resolution 35. 8 pcls 35. 84 pcls

In Fig. 4-3, the dashed box represents a state, the rectangle and lattice boxes

represent processes and decisions. The oval boxes represent terminals, which lead to

other states.

Figure ~ ~ ~ ~ ir 4-:MDUpoes n fra

iF
ii T-ilii~ii
i~ I(PACI(

ii
Rmiil-nt(
ilrt.

If a node wants to reserve a DEF, it first listens to the network for at least one

TDMA session. During this period, it monitors the network activities and learns the

reservation of each station as depicted in Fig.4-3 CONDIRIONING state. When the RF

begins, it chooses a random RS and broadcasts an RP to the networks.

itlnrianl ~c

ii,
i "
Nn i-.

Y~r
KLi I(niiiioni
KS :iii: iruc

i -.
i
ii
--------------~----

i.S-nd Rr
1KSiiagiisirp

"i, i

"
Yee

baitnf~
Rcsri~Rliun
).Wir/RM :ndc,:i =iilie
ra Mrr*id=tiul
~ir"l

,, i
i, ii ii
r

i

_______i

-- I';cl:
KFCCiVrJ

,*i

v;i
1

_I
i I

Figure 4-3: Reservation process flow chart

The destination receives and compares the request with its internal table. If the

reservation does not conflict with previous reservations, the node sends a Reservation

Acknowledgement (RACK) to the transmitter and completes the reservation process.

This process is illustrated in Fig.6.3 IN RECEIVING state. Other stations in the network

also listen and broadcast a Reservation Negative ACKnowledgements (RNACK) packet

to jam the possible RACK signal if they detect a conflict. To ensure the requester

receives the RNACK packet successfully rather than interpret the signal as noise, we use

I

ii'uir RIPS

,c_
r~jSSCIVB:i;iC
RSi
r,_fioiiLUii
i
~~~~~~.
_i_

"" ",t

a delayed NACK methodology. We will illustrate this method in hidden node resolution

section.

If a collision/RNACK happens, the requester will move to backoff stage with

random backoff time. The backoff algorithm is the same as in the CSMA/CA, that is,

after each unsuccessful reservation, the backoff window size is doubled, up to a

maximum value 2-W, where W equals to (CWmmn+ 1) and 2-W equals to (CWmax + 1) and

the backoff time is uniformly chosen in the range (0;CW-1). Since PCFMA is a persistent

reservation scheme, the competition between stations suppose to be small, thus the

maximum window size is set to 16. Once the backoff window size reaches CWmax, it will

stay at the value of CWmax until it rests.

Data Exchange

The data exchange period is divided into Ndata slots(DSs), each with a duration of

S pc s. A DEF is composed of several DSs. The length of a DEF is decided by the TS field

in owner' s RP. A DEF always begins at the start of a DS.

If a node successfully reserved a period of time for transmission, it listens and waits

until its DEF time, then starts to transmit MAC Protocol Data Units (MPDUs). If the

receiver receives a successful MPDU, it will send an ACK. The sender can continue until

the end of its reserved DEF.

At the end of a transmission, the sender broadcasts a Last Frame MPDU to the

network. The receiver also broadcasts a Last Frame MPDU to eliminate possible hidden

node problems and allow all nodes to update their internal tables. When a station

ungracefully terminates its connection, affected stations (usually it is the receiver) also

broadcast a Last Frame MPDU during the reserved slots to synchronize internal tables

with other stations.

MAC Protocol Data Unit

MPDUs combine several Ethernet packets belonging to the same path into a jumbo

packet to increase overall efficiency. The original Ethernet packet header and trailer is

removed and a new small header called PHY Block Header (PBH) containing the

sequence number of the original packet is added to indicate the order of the packet. After

this process, the original Ethernet packet becomes a new block called a PHY Block (PB).

The process is depicted in Fig. 4-2.

Several PBs are then combined with the original Ethemnet packet sequence into a

jumbo packet with a common header to become an MPDU ready for transmission. Based

on the current transmission speed, as many Ethemnet packets as possible are combined

until the transmission duration reaches 1.5ms.

Solutions to the Hidden Node Problems and Near Far Effect

A robust signal modulation is required due to the attenuation and noise on PLC

channels. However, this makes detecting packet collision difficult the signal modulation

process may interpret contending packets as noise and remove the noise to restore the

intended signal! This phenomenon causes near-far effects and hidden-node problems.

A common solution to the hidden-node problem is the use of a RTS/CTS

handshake before data transmission. The purpose of RTS/CTS is to notify nearby stations

of the incipient data transmission period so that those who are not involved in the data

exchange will avoid the channel during that period of time. Signal reception of mobile

devices in wireless networks usually suffered from nearby activities, reception is

expected to be constantly changing and thus exchanging RTS/CTS packets before every

data transmission is required.

In PLC networks, we adopt the similar RTS/CTS scheme but only in the

reservation stage. In the RF, nodes want to make reservations broadcast RPs to the

network. The RP also act as "Request to Send" signals as in the RTS/CTS scheme. Nodes

that receive this packet avoid the reserved period. Nodes outside the broadcasting range

may not aware of the reservation resulting in inconsistent databases. If they do not want

to make new reservations, the inconsistency becomes irrelevant. If one of the nodes

wants to make a new reservation those conflicts with the scheduled reservations, all

nodes that hear this RP broadcast RNACK. To deal with the possibility that the RNACK

is not heard by the transmitter because of near-far effect, a delayed RNACK scheme is

applied as shown in Fig 6.1.

The delayed RNACK scheme use Priority Resolution Signal (PRS) to inform the

requester the received RACK is incorrect. All stations object to the reservation should

wait 981s and broadcast a PRS after the PR packet. The requester should listen to the

network after it broadcasts PR packets. If a RACK is received, it should keeps listening

to the network for a PRS packet duration (35.84 pc s). If it senses the PRS packet, it

determines the reservation is invalid; otherwise it is a successful reservation.

If the intended receiver also recognizes the reservation is invalid, it should reply the

requester a RNACK with corrected SS and TS fields to make requester database more

consistent to the network status.

Approximate Performance Analysis and Simulation Results

To calculate the maximum throughput, we assume there are always data to send for

each node. We define efficiency as the ratio of time spent on transmitting payload to the

total time spent on the whole data exchange process. The parameters we used in the

calculation is summarized in Table 6.1.

The minimum reservation slot time can be obtained by adding a RP duration, two

RIFS, a Response, and a PRS which leads to (72 + 26 + 72 + 35. 84 + 26) yc s = 231.84 pc s.

A successfully packet transmission process requires a MPDU, an ACK and two RIFSs.

The total time required for this process is 1624 pcls when sender sends a maximum length

MPDU. The total DAE slots in a TDMA session can be calculated by

L T23.84-2s assume R reservation slots. If there are m transmitters, and the

bandwidth is evenly distributed to all transmitters, then a transmitter can have b slots,

where

ST 231.84R 29s
b = s slots

The total allowed transmitting time for each data stream can be calculated by b X s.

A sender can transmit p maximum size of MPDUs in a TDMA session, where

ST 231.84R 29 1
P=
1624m

Since the maximum MPDU has duration of 1:5ms and the overhead of a MPDU is

72 pc s, thus the total time spent on data transmission for each node t can be calculated by

ST 231.84R 29142 c

1624m

Since there are m nodes, the protocol efficiency E can be calculated by

ST 231.84R 29148

1624m

E =T (4-1)
If we ignore the foor ( ) operation in Eq.4-1, we can calculate the maximum efficiency
Emax as follows:

T T-231.84R -29 1148

1624m

max T (4-2)

0.88(T 231.84R 29)

when T 4 00
Emax. 88%
One must note that each DEF starts from the beginning of a DS, if a node does not

fully utilize the reserved DS, the resulted efficiency will decrease.

From Eq.6. 1 we learned that the efficiency of PCF/MA protocol is determined by R,

m and T, where m is dynamic. To obtain the optimal performance, we derived a few

mathematical forms as follows.

Determine Parameter R

To minimize fixed overhead, a small R would be desirable considering the number

of contenders is relatively small since all reservations are persistent. However, a small R

could make the system unstable when the number of contenders increases.

It is obvious that when the number of new data streams n is larger than the

available reservation slots r, at least two stations will have a reservation packet collision.

We call this over-saturation. However, since the reservation session comes every T ps, as

long as the number of new data streams is less than 1 every TDMA session and at least 1

data stream successfully makes a reservation, the waiting queue should converge. When

there are n new data streams, this problem can de described as the probability that at least

one station does not conflict with other stations. To determine this probability, we derive

the following calculations.

The permutations P of choosing n numbers from r numbers is

P =r (4-3)
The generation function G(x) corresponding to the random choice of n numbers

from r numbers with unlimited repetition can be written as

G(x)= (1+x+--+--+--)
2! 3!
= eYX

n r-x, where n > r

We want to know the probability that at least one RP does not conflict with others.

That is, at least one node chooses a random number that is different than the others. We

can first calculate the permutations p of choosing n numbers from r numbers such that a

number is chosen 0 times, 2 times, 3 times...., then calculate (1- p) P to get the desired

probability. The generation function of p with unlimited repetition can be written as

x x3
g(x)= (1+--+-+--)'
2! 3!
= (ex x)'

~I:o( )(-1)"ze' xi -i )"

o1)(-1)(r 2)(r-itf x

(-1) r 1) \i0 4.t

S+(-1i)'(r) )( -- 1 2 < )'1

(-12()(r- )0~~"" (0+2

( 1)"'" ) r0).

(4-4)

Since we are looking for the coefficient ofx-/n!, rearrange Eq.4-4 yields

g(x) = i(-1)o(B)(r 0)oxo

-1)o(B) r ~ 0)]

(4-5)

The coefficients of x n! becomes

p= n!-)"g r- (n-1)); (4-6)

From Eq.4-6 and 4-3, we can get the desired probability as

p ; n n 1~I
1- 1- (4-7)
P r"

To verify formula 6.7, we simulated a constant number of contenders. The

simulator simulates one million reservation sessions. The number of contenders is the

same from session to session. The simulator counts the event that at least one contender

successfully makes its reservation. The number of reservation slots is increased from 2

slots to 9 slots. Simulation results is shown in Fig.4-4

Probability Of Successful Reservation

1.2

+- 2 Reservation Slots
d- 4 Reservation Slots
--x-- 5 Reservation Slots
a 0.6 C"" eevto lt
S I 'x. -- 6 Reservation Slots
is I-*-- 7 Reservation Slots

0x. x -- -9Reservation Slots

0.2

0 2 4 6 8 10 12 14 16 18 20
Number Of New Streams

Figure 4-4: Probability of successful reservation

Fig.4-4 shows that even r = 2, n = 4, the probability that at least one station

successfully makes its reservation is about 75%. When r = 9, even if n = 20, the

probability is as high as 90%. To verify our assumption, we simulate a Poisson arrival of

h=1, r-8, PCF/MA remains stable for one million reservation sessions. When we increase

the inter-arrival rate h to 3, r have to increase to 16 to make PCF/MA stable. However, 1

new streams for each reservation session is equal to 40 new streams per second if T= 25.

We believe 40 new streams per second are more than enough for a home network.

From the simulation results, we chose r = 8 for the optimal parameter for PCF/MA.

Near-Far Effect Modelinn

Hadzi-Velkov, Z. and Spasenovski, B. [18] successfully modeled the 802.11lb near-

far effect under a Rayleigh Fading Channel. It is known that different modulation and

transmitting power under different channels will have different near-far effects. To

investigate the performance impact with and without the near-far effect, we choose a

near-far effect model described as the probability a receiver captures the desired packet

(Pcap) in the presence of interfering packets (RI)

P,, = Probability 'd "I0
R,
(4-8)

when '"= 2 C,,
R,
where Rd iS the received power level of the desired packet and the CmHis the capture

threshold.

Extra Allocation

Since the packet loss rate is high (10-' packets per second or higher), a silent

retransmission is required when the receiver receives corrupt packets. This requires data

streams to allocate extra bandwidth despite the estimated bandwidth.

Assume packet loss rate is P and the number of transmission attempts is R. Suppose

at one instance, the sender sends NPBs in a MPDU, the amount of over allocation for

next transmission is NP. Thus for quasi-error free environment with unlimited retries, we

need to over allocate (0) resources by

O = NP + NP2 + NP3 + NPX R
= rNP( + P+ P +-+PR-1 +-- (4-9)

When P <<1, Eq.4-9 becomes rNP1

We depict the computer simulation and theoretical calculations in Fig.6.5. We

simulated 106 TDMA sessions to collect meaningful statistics. The simulation results

closely match to the mathematical calculation.

Average Over Allocation with various P, R=3

14
--~- P -0.001

10 x -- -.
Simulated, P= 0.001 *X
8 -Simulated, P=0.01 X
0 ~--Si mulated, P= 0.05
6 Simulated, P=0.1 ------ -

0 20 40 60 80 100 120

Figure 4-5: Maximum over allocation PBs various PB loss rate comparisons

Data Stream Model

Event driven simulation was used to investigate PCF/MA performance and to

compare PCF/MA with a CSMA/CA adapted to PLC. The parameters of a PLC network

are assigned path by path to simulate real world PLC environment.

To estimate the maximum throughput, we applied Always On data stream to both

PCF/MA and MCSMA/CA. By Always On data stream we mean the sender has

unlimited input buffer and always obtains maximum duration of MPDUs once it has

chance to transmit. We assume all channel data rates are 100Mbps and the received

power levels from a node to other nodes are randomly generated with Poisson

74

distribution with 2= 50. There are no hidden nodes in this simulation, the packet loss rate

is 0.001 and CmH= 0.3, the simulated results is depicted in Fig.4-6.

Throughput Comparisons With Various T, Cth=0.3, PL=0.001

100

90

80

10 -H T7

n 0

S0 5 0 5 0
r m

Figure~~~~~~~.~ 4-:PF/Avrou vrusMSA/Atrugptcmprsn
Fi.4-6 shw C/A a aiu promne anoe CSAC bu
100 whn te umbr o cnteder islage m 6, = 10).It ls shws ha
PCF/MA~. wihlreTo nyiceae hogpt u lo nrae h ubro

possible contnder (m).. Th eao ha dcds a b nfre fo E.41 hc
shows ~ ~ ~ ~ ~ thtfr ienT h foroprtos as teuefltm o ah PUt
beoe es sm nress Tee r as sm ro ut intePFMAtruhu
chart ~ ~ ~ ~ ~ "'. whc recue b h foroertos
PC/M benefits gretl from lag oeewt ageTh aktdlyas
mereases.

Delay Model

To find out the timed average delay of PCF/MA we propose the following model

depicted in Fig. 4-7.

Amv~ll rr3 ~arnvalsr

IleftovLetove / rial

Queue Status at
time t,
Figure 4-7: PCF/MA delay model

Since PCF/MA divides time into periods each data stream transmits at its reserved

period. In Fig. 6.7 we show a stream that has opportunities to transmit at times Z, A, B, C,

D, and E. Each transmission lasts for a portion of a frame. The duration is represented as

symbol w. Assume a MPDU arrived at time t, and departed at time t,. Assume the system

is a steady state system. Thus, the number of MPDUs in the queue at time A is the

number of leftover MPDUs, 1. Assume y MPDUs arrived during the time from A to t,.

This period of time lasts a portion of a frame represented as symbol x,.

From the above assumption we found that at time t, the queue length is the sum of 1

leftover MPDUs and y arrivals. Assume PCF/MA can deliver pu MPDUs per frame for

current stream. Thus, delivering l+y MPDUs takes rp+x~frames where x2 1S a portion of w.

The delay of the new arrival MPDU thus can be calculated by

d = (1 E(x,)) + r + E(x2) frames (4-10)

where E(. ) represents the expectation of a variable.

During this time all MPDUs in the queue must be processed. This amounts to

E(1) + E(y) = (1- E(x, )) x p + r x pu + E(x2) x pu MPDUs (4-11)
Assume the arrival rate is 32 arrivals in a frame. Since the system is a steady state

system the inter-arrival time of MPDUs can be treated as a uniform distribution. The

expected value of x, is

123 ii

(1 + il)
(4-12)
2il

In a similar way we can calculate E(x2) as

E(x2) = -w (4-13)

Using queuing theory we can calculate EOy) as

E(y) = E(x, ) xA = A (4-14)
From Eq. 4. 11 ~ Eq. 4-14 we can obtain r as

r=E(1) + E(y) pu + E(x, )p E(x2 ~

111
E(1)+ i-2 p- wpu
2 2 2(4-15)

11
pu\2
Substituting Eq. 4-15 to Eq. 4-10 we get the average delay

1 1
d=-+r+-w
2 2
(4-16)
11

Since the service time for each maximum MPDU is the same the system can be

modeled as a M/D/1 system, where D means deterministic, we have X2 = I. From the

Pollaczek-Khinchin (P-K) formula we denote the residual queue length I as

E (1) =
2(1- p)
1 (4-17)

2(1- p)

where p =

Thus the average delay becomes

d= 1 p 1418
p2(1 p)2

1 p
=p+
2 2pu(1- p)
The simulation and analytical results are shown in Fig. 6.8 ~ 6.11. When the

reserved duration approaching to the full frame (in this case, w= T), the simulated results

closely match to the estimation from M/D/1. However, when the reserved duration is

short, the simulated results becomes approaching to the estimation from our analytical

model. For long reserved duration, most of the MPDUs arrived during the transmission

duration yielding shorter waiting delay. For short reserved duration, however, most of the

MPDUs arrived outside the reserved period resulting in a long waiting time.

78

Delay VS Transmission Duration, T=25ms

100.00%

90.00%

80.00%

70.00%

60.00%

50.00%

40.00%

30.00%

20.00%

10.00%

x \N=1/4T

A \N=1/3T

\N=1/2T

o \N=1T

- Analytical
-- -M/D/1

x

- - a

~cco,
;E1; o--

0.00% 20.00% 40.00% 60.00%

Utilization(Lamda/Mu)

80.00% 100.00%

Figure 4-8: Analytical and simulated average delay with various w, T=25ms

Delay VS Transmission Duration, T=50m1s

-000

80.00%

70.00%

-~~ 5.0

60.00% -

1 0.00% --

E 0.00%

10.00%

I

- ---

-

"

x \N=1/4T

A \N=1/3T

m w=1/2T

o \N=1T

- Analytical
-- -M/D/1

x

x
- -

x

n "

20.00% 40.00% 60.00%

Utilization(Lamda/Mu)

80.00%

100.00%

Figure 4-9: Analytical and simulated average delay with various w, T=50ms

79

Delay VS Transmission Duration, T=75ms

7().)()% -- 1x \N=1/4T

E 6().()(% _XJ _x 1/3T
Tim \N=1/2T
~5().()(%
I ..~rx o \N1=1T
4().()(% x x xx -- x __x__, A ayia

3().()()%~ ~ ~ -- -g - - --M/D/1
2().()(% -

().()(% 2().()(% 4().()(% 6().()(% 8().()(% 1()(.()(%

Utlilzation(lamda/Mu)

Figure 4-10: Analytical and simulated average delay with various w, T=75ms

Delay VS Transmission Duration, T=1()(ms

14().()(%

12().()(%

x 11=1/4T
eA \N=1/3T
x m \N=1/2T
6 o \N=1T

r m' -- -M/D/1
4().()(% -

2().()(% mm m- om

().()(% 2().()(% 4().()(% 6().()(% 8().()(% 1()(.()(%

Utilization(Lamda/Mu)

Figure 4-11: Analytical and simulated average delay with various w, T=100ms

Video Traffic

The main focus of this section is to discover the performance of PCF/MA in

delivering DVD video streams. To do that we build a video traffic model as depicted in

Fig.4-12.

Server Clie~nt

Transfe H u fer'c

Theshold

Figure 4-12: Video traffic model

A PLC station acts as a video server. It supports several DVD streams to different

destinations. To simplify our discussion we assume that the channel data rate remains

stable throughout the whole video display length. We assume the video source is able to

supply one video frame immediately after the request is received and that the server will

start to transmit MPDUs once it gets permission to transmit. The client moves the

received MDPUs to the playback buffer and re-assembles them to a normal video frame

before playback. If a video frame is still in transmitting when it is time to display, the

video frame is counted as a miss deadline video frame. The client will try to display the

video frame 33.3ms later. The miss deadline count keeps incrementing until the video

frame is successfully displayed.

To smooth video streams, a certain amount of playback buffer (B) is required.

When the buffer level is lower than the predefined buffer threshold (BmH), the client sends

requests to the server to ask for more video frames. If at time T, we have BT BmH, the

server stops transmitting MPDUs.

To make sure the missing video frames are not caused by the protocol in use but by

another factor, we investigate the relationships between the playback buffer size at a

specific time T (BT), playback delay (D), average frame size (Va), maximum video frame

size (y;;a) channel bandwidth (Bc) and the effective bandwidth (Bw) reserved for this

video stream.

We classify the way the playback buffer accumulates data into 3 categorizes:

"Depleting", "Accumulating" and "Balanced". Usually it is the "Depleting" situation that

clients drop video frames. Consider the captured time frame of a DVD movie depicted in

Fig. 4-13. At time T, the buffer level at the client side can be calculated by

30T
B, = BU.D + BH:T C 1 (4-19)

where yE is the data size of frame i. If we assume T is relatively large, such that yE=

Va and Eq.4-19 becomes

B, = BwD +BwT -30Tya (4-20)

Average Video Frame Size inA GOP

60000

Currently Displaying GOP
50000
Currently transferring
\$. GOP

3 0000 **

E 20000

10000

T Time

Figure 4-13: A large video frame in transmit

At time T, the server encounters a Group of Picture (GOP) with large video frames

of size E;;m bytes as shown in Fig.4-13. Assume the number of video frames in a GOP is

15V,,
15, then it takes t = mxsecond to deliver this GOP. During this t time frame, the
B,

15V
client consumes N= 30 x mx video frames, that is, the client consumes
B,

450V2
Nx Vma = B .x bytes. Meanwhile it accumulates 15 x Vm, bytes. Suppose there are G

such GOPs, the Eq.4-20 becomes

B, = D+B.T-30V +G15V ax > 0 (4-21)
In order to prevent video frames from missing their deadlines, we must make Eq.4-

21 always larger than zero. To help us analyzing Eq.4-21, we re-arranged it as follows:

B, = Bw D-GC 4 50 O x 15max+( -301-)T >0 (4-22)

The term 5Y~B :D-mx 1V in Eq.4-22 is the accumulated buffer size

during the playback delay time plus buffer loss during transmitting large video frame

GOPs. The term (B 30V ) is the difference of the reserved bandwidth and the average

DVD title bit rate. In real world applications, Bw, Mymx and Va are DVD statistics thus can

be easily defined.

To make BT always larger than zero, we conclude the following sufficient, but not

required, conditions:

450Ya
BD, -G mx 1Vmax > (4-23)

(B .- 30V, M0 (4-24)
If we assume the reserved bandwidth is the video average bit rate, the D can be

calculated by

G(450V~ -15BY .Vmx a
BZ
G(450Vna 450VoV,, )
max max(4-25)

G(Va V )
2V2
From Eq.4-23 ~ 4-25 we observed that when reserved bandwidth is equals the

average DVD bit rate, the startup delay D should be large enough to absorb the buffer

loss due to large video frames. However, when we reserve a large bandwidth and a small

playback buffer, the buffer will build up quickly. Once Bm is reached, the server has to

stop sending further video frames to prevent buffer overflow. In this case, the term Bw in

Eq.4-24 becomes over-estimated. To prevent this problem, a large playback buffer is

desirable. Since the minimum playback buffer size should satisfy Eq.4-23 and Eq. 4-24.

B > G m ax V a n d B > 3 0 ( 4 2 6 )
Since the playback delay D and playback buffer is for the worst scenario in the

DVD playback, the G should be large enough to satisfy this situation; thus we set G = 60.

To verify our formula, we simulated a video server/client assuming no other data

streams on the network where the client has an unlimited buffer, the video parameters is

listed in Table 4-2. The efficiency of PCF/MA is about 80% when T = 25ms. The

reserved bandwidth for PCF/MA is 9.44Mbps, the observed throughput is 7.45Mbps

which is close to average bit rate. We slowly increase delay playback time D, and

observe if the simulation drops video frames. If it drops video frames, we record the

84

maximum buffer it accumulates before dropping. The total simulated time is 20 minutes

which is about one sixth of the original movie length.

Table 4-2: Video traffic parameters
Video
Average bit rate 7.1Mbps
Vmax 36Kbytes
Va 30Kbytes

In simulating MCSMA/CA protocol, we follow the same procedure except the

channel data rate is assigned as 15.5Mbps and the observed throughput is around

7.58Mbps. Simulation results are shown in Fig. 6-14.

Miss Deadline Comparisons Maximum Accumulated Buffer

350 PC~4C F/MABc i 25 ~PCF/MA
300~ --P --- -- -- -------MC.SMA /C.A 20 ---k--MCS M /A __
250 -- --.Bc=15.5 Bc=15.5
15 - -

1 00 -- --
10 - -- E 5 -

0 A0 5000 10000 15000
0 2000 4000 6000 8000 10000 12000 140
D~rns)D(ms)

Figure 4-14: Video playback performance comparisons

The simulation results show that without competition, the MCSMA/CA performs

slightly better than the PCF/MA (though at worse efficiency). In Fig.6-14, when D = 1,

the client drops about 320 video frames and the maximum buffer accumulated is about

2Mbytes. The result shows that the delay time is too short to build up a safe buffer for

future video bit rate fluctuation even with unlimited buffer. The figure also shows that the

video stops dropping after we increase D to 12 for PCF/MA and 11 for MCSMA/CA. In

both cases, the maximum accumulated buffer size is more than 20Mbytes. This result

Figure 4-15: Capture of a period of the DVD simulation

Figure 4-15 shows the DVD video frames in transmit and the time takes to deliver a

video frame. Real world DVDs usually have lots of small video frames especially B

frames. These small video frames can not fill-up a MPDU result in low channel

efficiency. One has to note that Eq.4-25 and 4-26 do not estimate the exact B and D since

channel efficiency used is the maximum channel efficiency. Besides, there are too many

variations in real world DVD video streams. But Eq. 4-25 and 4-26 gives us an idea of

how much bandwidth is required with a given set of video parameters.

III II II II II II III I II II II II II II II II II II II II II II II II II III I II II II II Ill

III II II II II II III I II II II II II II II II II II II II II II II II II III I II II II II Ill
Lots of small video frames
IIIIlllill lillllill 11111111 III I II II II II II II II II II II II II II II II II II III I II II II II Ill

III II II II II II III I II II II II II II II II II II II II II II II II II III I II II II II Ill

III II II II II II III I II II II II II II II II II II II II II II II II II III I II II II II Ill

III II II II II II III I II II II II II II II II II II II II II II II II II III I II II II II III

III II II II II II III I II II II II II II II II II II II II II II II II II III I II II II II Ill

suggests that when the delay playback time and buffer is large enough, both protocols

have ample time to deliver large video frames thus no video frames were dropped.

From Eq.4-25 and 4-26, we obtain B = 6.48 Mbytes and D = 11 second. The results

though different from the simulated results, are acceptable closely. The difference

between the simulated and analytical results is because the channel efficiency did not

maintain at 80% in the simulation. The problem can be illustrated by the Fig. 4-15.

Curaii nt Orie ji liel~e ~~D~iy me~52 11 -~~~3 5 :r

100 90 80 70 60 50 40 30 20

Bc

To demonstrate PCF/MA in delivering multiple video frames at the same time, we

conducted multiple DVD server-client simulations. The simulate time is 20 minutes with

D = 1 and B = 10000Bytes. The simulation results are depicted in Fig.4-16.

In Fig.4-16, the simulation result showed that PCF/MA supports more streams

when the channel data rate is high. However, under the same channel data rate, the

MCSMA/CA supports less DVD streams. This shows MCSMA/CA suffered from

competitions between MPDUs and protocol unfairness. Though PCF/MA is capable of

delivering up to 11 DVD streams in theory, the real work DVD streams contains too

many small video frames makes channel efficiency much lower than expected. From our

simulation, we found the efficiency drops to about 64%! To increase the efficiency, a

new streaming protocol that supports striping several small video frames into a MPDU is

required. Because of the time constrain, we did not further investigate this issue.

Number Of Supporting DVD Streams, B=10, D=2

10

v6
m 4

E 2
0

D PCF/MA
B MCSMA/CA

Figure 4-16: Multiple video streams comparisons

Visualization of Protocol Simulation Data

When developing or debugging network protocols, it is necessary to understand

both behavior in the aggregate, such as throughput or delay, and specific, pathological

scenarios. For the former, graphs are useful ways to present data, but for the latter, it has

been more challenging to process the data into a form convenient for human consumption.

This section describes a system that represents local area network (LAN) simulation data

in the form of time-space diagrams, with station states shown in color. The interface is

rich in that it not only allows time scale expansion and compression, but also allows the

user to click on specific times and events to obtain detailed information about interesting

events. The authors have found the program most useful for understanding and improving

various protocols types of LAN protocols.

Introduction

Traditional approaches to visualization of local area network (LAN) simulation

data mainly involve graphing aggregate behavior as a system parameter is varied. This is

very useful and has been employed to optimize system settings as well as to select

between alternative approaches. Histograms and waterfall charts showing the distribution

of behavior classes has also been employed, particularly for determining behaviors

relevant to Quality of Service (QoS) metrics. However, these forms of data representation

have limitations, especially when attempting to understand detailed behaviors that give

rise to the observed performance.

When simulating complex protocols such as PCF/MA or MCSMA/CA protocols, it

becomes difficult to judge if the simulator works as the desired way using the text based

log files. The protocol designers have to look into each text generated by the simulator

and analyze the state transition imaginarily. This method works Eine when simulating

88

small scale networks or simple scenarios, whereas when simulating large scale networks

or complex scenarios, the huge text logs becomes un-manageable.

To improve the quality of interpretations of simulation results, Visual Protocol

Analyzer (VPA) is designed for efficient analysis of the results of simulations.

Goal of User Interface

Humans are more sensitive to graphs rather than text. Especially when analyzing

complex situations, people usually use drawings to help themselves to reason. For

example, Fig. 4-17 shows two modes of the same event transition of a simulation results.

People can easily understand the state transition displayed as a diagram on the figure in

the right hand side rather than the texts shown on the figure in left hand side. This

comparison gives us a clear view of the need of a VPA. To effectively analyze the results

of simulations, the desirable VPA should contain the following features

2.000ese,CONDIC TIONIWC ,CONDIE T IOHIN G,COND IT NI ,CNI ro NONIT NI ,ONI rHI
29.negeneDETECT~IDEECT IGel slamLDETEC slanLDTEC slmLDEET laNLDTET la LDEEC sa
26 0.800ae,CODE CT _SOI GHALDETCTSlNatE,CODETCIHLDTC~l~tDTCIHLDTC~lN I IS 2I IS CI lS 2I ISI E S

7424.52neseCON DIETIONTW 0NIToIcCNiiOTc0NIToINC,CONDICilNT,0NITo PERSRVTO EN _T EN _T NEPNE AKF
72 .520aTECT LEAmL,OETECT s luNA tDETECT sa LDTC l~tDTC lm ,EETsu
1188.00000DTC SI~l,DETECT S CNIGALDETETSIHLDT T IHL ET TSIHLD ETSIHL ET TSI Nde

2628.3s60 o~aeaTnNS IT 00 PLrETnER luE CoPLTElu conLEEEClu coPLT1nC1u com Nede l11

2726.360aeaRECElUE COnPLE TETRANSM~ITcPL TEEClu o*PEERClEcnLT~E~u

Figure 4-17: Text versus graphical event analyzer

*Easy and efficient to communicate with the analyzer.

The way to communicate with the analyzer decides the efficiency of analyzing a

simulation result. We choose to use mouse cursor as the major User Interface(UI)

because: First, using mouse cursor is an intuitive way of operating modern operating

systems Thus people can become more acquainted to the mouse operation. Secondly,