VOLUME 5 007 BLAGOEVGRAD, BULGARIA SCIENTIFIC Research ISSN 131-7535 ELECTRONIC ISSUE
IMPROVING FAIRNESS IN CDMA-HDR NETWORKS Valentn Hrstov Abstract. Improvng throughput and farness n Cellular Data Networks s a problem of present nterest. Non hgh throughput and unfarness of data streams and sharng the network resources greatly lmt puttng nto practce and the commercal market wnng of such networks and technologes. In artcle are consdered man realzed n practce way of rasng throughput and farness n wreless networks. Usng smulaton results obtaned wth an avalable onlne smulator, I present some advantages of usng more accurate estmaton of the SINR combned wth H-ARQ n Cellular Data Networks. 1. INTRODUCTION Cellular Data Networks are becomng more and more popular nowadays. In these networks, tme s dvded nto many tme slots and each wreless termnal-wt s transmttng packets (to a base staton). WTs use one or more tme slots to transmt ther payload, but each tme slot can only accommodate one WT. Multple WTs are accommodated usng the tme dvson multple access- TDMA technque. As a result, each WT can use the maxmum power of the entre base staton-bs. In the wreless moble envronment, the RF condton changes sgnfcantly wth tme. When the RF condton s good, lttle codng protecton s needed and modulaton wth hgh constellaton can be used, makng t possble to transmt at a hgh data rate n a gven tmeslot. The data from each tmeslot are scrambled and spreaded usng a computer generated pseudo-random -sequence (chp) unque to the sector of cell (Code-Dvson Multple Access- CDMA). Cellular Data Networks- CDNs combne the advantages of both technques- TDMA and CDMA. Ths combnaton s well suted to the bursty nature of packet data, as well as has the advantage of beng able to have frequency reuse n every sector. CDNs as CDMA system are known to be nterference lmted, whch means that ther capacty can be ncreased by reducng the mnmum requred (for stable recepton of data at the recever) energy of sgnal wth a gven Sgnal to Interference and Nose Rato-SINR. The SINR s a functon of several factors such as path loss, shadowng, fadng, nose, and ntercell nterference. Such CDNs, as CDMA-HDR (Hgh Data Rate), 1xEVDV (1x Evoluton for Hgh-Speed Integrated Data and Voce), and other systems realze procedures for power control []: The WT measures the SINR of the receved sgnal through the plot channel and sends feedback to the BS; Based on the feedback from the WT, the BS adjusts ts transmtted power level. The structure of the reverse lnk traffc channel [5] s shown n Fgure 1. The plot channel ads coherent demodulaton and trackng. The reverse rate ndcator channel s used to nform the base staton about the data rate beng transmtted on the reverse lnk. If the base staton decdes to transmt a packet for a gven WT, t s requred to transmt t at the data rate specfed by the DRC request from that WT. The acknowledgement channel s used to support early completon of forward lnk packets.
Fg.1 Whle the data rate adaptaton dscussed above sgnfcantly mproves spectral effcency, further gan can be obtaned by dvdng the total packet energy n each packet nto several portons and ncrementally transmttng the packet wth a port of the energy usng multple subpackets n separate tmeslots, and termnatng the transmsson as soon as the packet s decoded correctly at the other end. Ths gan exsts because the estmaton of the RF envronment at the WT s not perfect, e.g. the data rate requested by WT va DRC channel for the base staton to transmt s usually conservatve. Ths s realzed by Hybrd Automatc Repeat Request- H-ARQ and Early Packet Termnaton. The Hybrd Automatc Repeat Request- H-ARQ (or more sophstcated Hybrd ARQ/FEC) s usually used to mprove data throughput [] and to allow for early packet termnaton. The general procedure of the H-ARQ s as follows: The packet s coded, nterleaved, added CRC and formed nto subpackets that are transmtted. The recever decodes the packet and checks the CRC. If the CRC check passes/ does not pass, an acknowledgment /negatve acknowledgment s sent back to the transmtter. Early packet termnaton refers to a successful recepton of a packet before the nomnal packet duraton when the channel condton s good. In H-ARQ, the nformaton bts are conveyed usng several subpackets. If the packet can be decoded correctly before all the subpackets are transmtted, the packet transmsson s termnated early and there s no need for addtonal transmssons. The repetton of the packet bts n the subpacket s accomplshed by means of channel codng to obtan further codng gan. To allow tme for the WT to process each subpacket and feed back the nformaton to the base staton, each subpacket s transmtted dsjontly n tme. In terms of energy, H-ARQ can be thought of as a scheme where addtonal energy for each subpacket s transmtted untl the requred SINR for the entre packet s reached [1]. The effect of H-ARQ s qute smlar to that of the fast power control technque snce t mnmzes the total nterference to other WTs by controllng the power used to transmt packets. The H-ARQ s used to mprove network performance n presence of nter-cell nterference [5]. In [1], Kwon et al. consder the problem of controllng the transmtted power. H-ARQ combned wth fast power control acheves a larger reducton n the nterference on the CDMA reverse lnk compared to the system only wth fast power control, however ths knd of gan s obtaned at the expense of larger packet delay. Other works such as [5] have studed the effect of nter-cell nterference over the forward lnk, and the benefts of H- ARQ n counterng the nterference, or rather a msmatch between the data rate that should be transmtted and the data rate that s requested by the WT s effectvely mtgated by H-ARQ and early packet termnaton. In [3], Mhatre et al. study the mpact of network load n the neghborng sectors on the nter-cell nterference n a cellular data network. The observaton that sgnal receved by a WT over the
forward lnk contans nterference from the neghborng base statons s used by the termnal to predct ts SINR more accurately. The sgnal receved by a WT over the forward lnk n a cellular data network contans nterference from the neghborng base statons. In Fg. s shown a WT n sector 0 of cell C receves nter-cell nterference from sector 1 of cell C and sector of cell C. Fg. In [3] s show that the SINR s: G0 A TC (1) SINR = 1 A TC(G1 ρ1 + G ρ 3 ) + N 0 where A s the sgnal ampltude, Tc s wdth of pulse (chp), G s the channel gan from the base staton of the nterferng sector to the termnal, and ρ s the traffc load on the forward lnk of -th sector (ρ s the probablty that a tme slot on the forward lnk of -th sector s busy). The SINR gven by (1), s a functon of G, and gven by: G = cd.w n ξ / 10. 10 () In (), the frst term n the product s determnstc (for a fxed WT locaton), and corresponds to path loss, whle the second term s a random varable correspondng to lognormal shadowng loss. Here, ξ s a Gaussan random varable wth mean 0, and varance σ G. Shadowng s correlated over each tme slot dependng on the speed of the WT as per Gudmundson model [6]. Last term- Raylegh fadng s accounted through W. However, n the actual mplementaton of CDMA-HDR [3], the BSs are GPS-synchronzed and all the BSs transmt ther plot sgnal at the same tme. Hence, the SINR measured by the termnals contans the worst case nter-cell nterference, snce the nterferng sgnals are transmted constantly durng the measurement phase. Referrng to (1), ths amounts to measurng the SINR wth ρ 1 = ρ = 1,.e: G0 A TC (3) SINR = 1 A TC(G1 + G ) + N 0 3 If the termnal has the nformaton about the network loads n all the sectors that are n ts actve set, t can calculate the actual SINR usng (1):
(4) SINR = 1 3 A T C (G 1 G 0 ρ 1 A T C + G ρ ) + N 0 Thus, usng more accurate estmaton of the SINR, WT wll be able to mprove data throughput. Moreover, H-ARQ also mproves throughput, even n spte of ntal conservatve SINR estmates [5], because as mentoned above, H-ARQ adjusts to network loadng n the adjacent sectors. The classcal ndex of farness [7] dsplays level of satsfacton of each user/wt, respectvely far sharng of the network resources, and s gven by: (5) Farness = ( X ) n / X So, Farness =1 speaks of qute farness sharng of the network resources between users, then and Farness =0 corresponds to absolutely opposte stuatons. The performance metrc s based on data throughput (X ). I expect that more accurate ntal estmaton of the SINR and H-ARQ can mprove throughput and farness. In next secton I wll verfy through smulatons whether these mprove the farness.. SIMULATION RESULTS I present smulaton results of farness when three WTs are served over the forward lnk,as run two sets of smulatons wth two subsets. In the frst set, all WTs use (3) n order to estmate SINR, and H -ARQ for early packet termnaton (Prmary Scheme), whle n the second set, all WTs use (4) to estmate SINR, and also use H-ARQ (Secondary. scheme ). In the frst subset, all the three WTs are pedestran, whle n the second- all WTs are vehcular termnals. Smulaton parameters are lsted n Table I, and have been taken from [6]. Each smulaton s run for 0,000 tme slots, and 600 ndependent smulatons are run to gather WT throughputs wthn 90% confdence nterval. I select WT locatons so that the termnal s equdstant from the two nterferng base statons. The locatons from the servng base staton of WTs are selected to be 0.4R 0, 0.7R 0 and R 0. The cell radus R s 1 Km. I use ITU path loss models for vehcular (10 kmph) and pedestran (3 kmph) WTs. Shadow correlaton dstance s the same for both vehcular and pedestran models- 0m, and depends only on the envronment (urban or suburban). Table 1 Carrer frequency, fo 000 MHz Log-normal Shadowng varance, G 10 db Shadow correlaton dstance 0.0 m Nose spectral densty, No -174 dbm/hz A, ampltude of transmt waveform 5.48 (base staton transmt power of 15W) Chp duraton, T c (1.5 Mcps) 0.8 us Radus of the sector, R 1 Km Ro for 90% cell coverage 0.95R = 0.95 Km Mscellaneous gans: antenna gans, 15. db body loss, cable loss Buldng penetraton loss, (only for 1 db pedestran WTs) Pedestran path loss n db, 101og 10 (cd n ) 301og 10 (f 0 ) + 49 + 401og 10 (d) (f 0 n MHz, d n Km) Vehcular path loss n db, 10log 10 (cd n ) 1log 10 (f 0 ) + 58.83 + 37.6 log 10 (d) (f 0 n MHz, d n Km) Fracton of mult-path power captured by the recever (vehcular WT) 0.784
In all our smulatons, although the tme-varyng shadowng and fadng, I assume that the WT locaton remans unchanged durng the course of the smulaton. In all the smulatons, I assume for smplcty that the network loads n both the nterferng sectors are the same,.e., ρ 1 = ρ = ρ, and I vary ρ to study the farness of Prmary and Secondary Schemes for dfferent network loads n the nterferng sectors. Table X1 X X3 Farness 1.0 991 43 149 0.68831 0.9 999 48 151 0.6894 0.8 1007 43 153 0.690167 0.7 101 436 155 0.69173 0.6 1018 439 158 0.693303 0.5 103 44 160 0.694448 0.4 107 445 16 0.695846 0.3 1034 450 165 0.69783 Table 3 X1 X X3 Farness 1.0 99 43 148 0.6875 0.9 99 43 156 0.69633 0.8 990 443 163 0.705844 0.7 987 453 17 0.716467 0.6 985 468 18 0.78971 0.5 980 480 193 0.741667 0.4 973 496 08 0.758444 0.3 963 517 8 0.780033 In tables and 3, I present the throughput (kbps)receved by each WT as a functon of the nterferng network load under the both schemes for pedestran model. Table gves the ndexes of throughput and farness for standard CDMA-HDR networks (Prmary Scheme). In columns ttled as X1, X, and X3 are presented end to end throughput for WTs 1,, and 3. In the last column s calculated the ndexes of farness usng (5). Smulaton results for Secondary Scheme are gven n table 3. I note that as the nterferng network load decreases, both the schemes result n hgher throughput for WTs and 3. Unlke Prmary Scheme, where the throughput of WT 1 ncreases wth decreasng network load, n Secondary Scheme, the throughput of WT 1 decreases wth decreasng network load. Ths s because both the schemes are desgned to mprove the throughput of a WT when the nter-cell nterference s lower. However ths beneft s especally more pronounced for WTs located near the cell boundary (WTs and 3). Table 4 X1 X X3 Farness 1.0 980 95 75 0.576896 0.9 987 300 77 0.579541 0.8 989 305 80 0.584005 0.7 995 310 85 0.589046 0.6 998 315 87 0.5943 0.5 1006 30 9 0.596884 0.4 1009 35 97 0.60399 0.3 1016 330 10 0.606917 Table 5 X1 X X3 Farness 1.0 980 90 75 0.5745 0.9 969 310 84 0.5949 0.8 961 30 9 0.607488 0.7 957 330 10 0.6168 0.6 957 360 11 0.643369 0.5 961 390 14 0.66471 0.4 968 430 14 0.69183 0.3 978 490 170 0.7979 I observe smlar results for the vehcular model n (tables.4 and 5). Near the cell boundary the throughput degrades more rapdly, due to a lower path loss exponent (more n vehcular than pedestran model),.e. more serous. ntercell nterference. In Fg. 7, I plot the farness as a functon of the nterferng network load under Prmary and Secondary Schemes for pedestran model (See respectvely table, and table 3).
Fg. 7. Fg. 8. In fg. 8 s compared the farness of prmary scheme (squares), and secondary scheme (dots) for the vehcular model. As expected, I observe smlar results n Fg. 8, but the farness mprovement of WTs and 3 s even hgher n the vehcular case than the pedestran case. It can explan wth fact, that the Secondary Scheme benefts the WTs located far from the servng base staton, as well as penalzes near located WTs As t has been shown above proposed mechansm (Secondary Scheme) s better than standard CDMA-HDR, takng on account throughput as well as accordng farness. 3. CONCLUSIONS Ths artcle shows the advantages of usng more accurate estmaton of the SINR combned wth H-ARQ n Cellular Data Networks, as well as presents some smulaton results obtaned wth an avalable onlne smulator [4]. Usng the proposed n the present work method the farness can rase and thus, avod beatdown effect.
REFERENCES 1. Kwon et al., Power Controlled H-ARQ n cdma000 1xEV-DV, IEEE Communcatons Magazne, Aprl 005, pp. 77-81. Lee K. and Samuel C., Analyss of a Delay-Constraned Hybrd ARQ Wreless System, IEEE Transactons on Communcatons, Vol. 54, No. 11, November 006, pp.014-03 3. Mhatre V. et al., Impact of Network Load on Forward Lnk Inter-Cell Interference n Cellular Data Networks, IEEE Transactons on Wreless Communcatons, Vol. 5, No. 1, December 006, pp. 3651-3661 4. Mhatre V., and C. Rosenberg, A smulator for CDMA-HDR data networks wth Hybrd- ARQ and opportunstc schedulng functonalty. Onlne avalable: http://mn.ecn.purdue.edu/~mhatre/cdmahdr_sm.tar.gz 5. Q. B, A forward lnk performance study of the 1xEV-DO rev. 0 system usng feld measurements and smulatons, Lucent Technologes, Mar. 004. Onlne avalable: http://www.cdg.org/resources/whte_papers.asp 6. ITU-MTR M.15, Gudelnes for Evaluaton of Rado Transmsson Technologes for IMT-000, 000. 7. Осипов Е.А., Проблемы реализации надежной передачи данных в самоорганизующихся и сенсорных сетях, ISSN 0013-5771. сп. Электросвязь, 6, 006, с.9-3.