Forward-link Throughput Optimization for Wireless Cellular Packet Data Networks

Transcription

1 JCS& Vol. 11 No. Ocober 11 Forward-link hroughpu Opimizaion for Wireless Cellular Packe Daa Neworks Moses E. Ekpenyong Deparmen of Compuer Science Universiy of Uyo, PMB. 117, 3, Uyo, NIGERIA ABSRAC An aemp has been made in his paper o access he conrol layer of he General Packe Radio Service (GPRS nework base saion (BS, using hroughpu opimizaion. We focus on he effec of packe ransmissions on he hroughpu of forwardlink (downlink power balancing, wih specific ineres on he relaive hroughpu of he four GPRS coding schemes for a realisic GPRS nework operaing in Nigeria. We derive a model capable of improving he channel qualiy and simulae he hroughpu and choice of opimum coding scheme over diverse packe sizes using he MArix ABoraory oolki. Simulaion resuls show ha he exising sysem requires effecive nework opimizaion o improve he sysem s hroughpu performance. Keywords: GSM/GPRS coding, payload, SINR, payload lengh, opimizaion 1. INRODUCION hroughpu is defined as he average rae of successful message delivery over a communicaion channel and is herefore a good measure of he channel capaciy of a communicaion link. he message or informaion delivery may be over a physical or logical link, or over a wireless channel passing hrough he cellular nework. he hroughpu of a cellular nework depends on a number of variables namely, he packe size, ransmission rae, number of overhead bi in each packe, received signal power, received noise power specral densiy, modulaion echnique and channel condiions. he key o maximizing hroughpu is mainaining he signal-o-inerference-and-noise raio (SINR a an opimum level [1]. Cellular neworks use a scheme called handoff o minimize signal breakup as a mobile user migraes from one cell o anoher. hey also employ spread-specrum echnology and special coding schemes ( each base saion is ed a code o enable he muliplexing of muliple users over he same physical channel. he effec of addiive noise and inerference from oher cells is common o all cellular neworks. he oal power of hese impairmens is σ was. o be precise, we refer o he impairmens (collecively as noise. Noise appears a he receiver as an addiional signal ha does no conribue o he overall hroughpu. he sysem has o use some of is power and bandwidh resources o overcome he effec of noise. In cellular neworks, ransmission from a base saion say i, o a node j, is seen as inerference by ransmission from any node n o any oher node k, such ha ( n k ( i, j,, and consequenly, reducing he bi service raes for link (n,k. his ineracion inroduces a negaive feedback (ino he sysem in he form of emporary congesion in he nework, caused by flucuaions in he wireless channel qualiy, hus reducing he nework hroughpu around he congesed area and increasing furher he congesion and fading raes, hereby obsrucing he ransmission of packes and hindering he propagaion of nework signals. his paper emulaes a realisic nework by using empirical daa from he field. Firsly, we obain realisic daa measuremens from Airel Nigeria (formally, Zain communicaions limied, using a dashboard Moorola cell monior. he daa were obained from base saion conrollers wihin he souheas region of Nigeria during busy hours. In sudying he nework, we observe he behaviour (of imporan parameers of he sysem, operaing under ideal condiions and derive a sysem model ha depics he sysem.. GPRS: BACKGROUND ISSUES Due o growing ineres in mobile daa applicaions, he global sysem for mobile communicaions (GSM has been exended by he general packe radio service (GPRS. GPRS allows he ransmission of packe daa a moderae bi raes using random access channels wihin he exising GSM infrasrucure. GPRS is a packe daa overlay o GSM, which has he larges number of subscribers in he world among all wireless sysems []. I provides he necessary funcionaliies o handle packe daa in GSM neworks. Mobile erminals such as noebooks, personal digial assisans (PDAs, mobile phones and pagers can give access o a wide range of new mobile daa services. here are hree main parameers ha deermine he link layer performance of a GPRS nework. hey include frequency reuse, error coding scheme (CS and daa packe delay [3]. (i Frequency Reuse Frequency reuse capabiliy is a concep in which, wih precise power conrol, differen cells in a cellular nework can share same frequencies (from 74

2 JCS& Vol. 11 No. Ocober 11 one cell. GPRS sysems ypically use universal frequency reuse, as oppose o earlier neworks, he mobile saions and base saions use he whole bandwidh o ransfer and receive informaion. Since GSM voice services usually require a frequency reuse of 3 or 7, as separae overlaying neworks, GPRS provides beer resource uilizaion and maximizes he number of users fulfilling he specified QoS requiremens. he cellular nework has a predeermined range of frequencies o exchange informaion beween he base saions and mobile saions. herefore, he smaller he frequency reuse facor which a cellular nework can employ, he higher he nework capaciy. (ii Error Coding Scheme (CS able 1 shows he four coding schemes employed by GPRS sysems: Coding Scheme able 1. GPRS Coding Schemes Code Rae PC daa block payload ( Service daa rae kb/s CS 1 ½ 8.8 CS / CS 3 ¾ CS he payload of each Radio ink Conrol (RC block is added o he Radio ink Conrol/Media Access Conrol (RC/MAC header and a block check sequence. For, CS- and, he radio block is convoluionally coded wih code rae ½ and hen puncured o he desired code rae. does no conain forward error correcion. he service daa rae is calculaed from he 4ms duraion of a GPRS super-frame. he super-frame consiss of GSM frames including 48 raffic frames and 4 conrol frames. he 48 raffic frames carry 1 RC blocks. herefore he daa rae is compued as follows: payload( / block 1( block / sup er frame 8( bis / bye 4( ms / sup erframe =.4 payload( ( kb / s Among he 4 coding schemes, has he leas coding rae. I is also he mos robus coding scheme. herefore, is used for all conrol messages. CS- and have lower coding raes and more informaion bis. is he mos daa efficien coding scheme, bu is mos vulnerable o channel impairmens. he base saion sub-sysem can selec any of he coding schemes o opimize he hroughpu and delay performance. he GSM sysem uses a combined FDMA/DMA air inerface, one DMA frame is sen a 4.61ms, in a KHz wide frequency band. Each DMA frame is divided ino eigh slos and a mobile erminal is ofen ed o a slo. A sequence of hese slos is called a channel. In GPRS, four DMA frames are combined wih addiional conrol informaion o form one ogical ink Conrol (C frame. Each of he eigh channels in a DMA frame can be used for GPRS. Figure 1. A simplified GSM frame srucure wih GPRS (packe daa channels. A he link layer, here are a maximum of eigh GPRS packe daa channels (PDCHs per allocaed frequency. Each of hem capable of ransferring one Radio ink Conrol (RC block every ms, providing a raw bi rae of 3Kbps per channel. All mobile saions requiring daa services in a paricular cell share he available PDCHs. 3. REAED IERAURE Several research publicaions exis on hroughpu analysis of cellular neworks [4-]. Also, many models have been developed o sui he resource issues invesigaed. hese include choosing an opimal power level o maximize hroughpu [6]. In [7], coding of GPRS neworks by forward error correcion, using puncured convoluional codes a differen coding raes and reransmission of RC blocks conaining unrecoverable errors, usually referred o as Auomaic Repea Reques (ARR is used o srenghen he hroughpu of cellular neworks. Sung and Wong [8] have presened forward-link hroughpu maximizaion, he inerference is modeled by he aid of he orhogonaliy facor and deermined by he ransmission sraegy for a range of he orhogonaliy facor values and user s channel gains, so as o maximize he sysem hroughpu, subjec o a oal power consrain. Mos of he previous works found have aken specific look a hroughpu in differen cellular nework echnologies (such as DMA, FDMA, CDMA, ec. and consider various sysem parameers in heir analysis. 4. DOWNINK QUAIY MEASUREMEN Due o he success of cellular elephone services in he 199 s, he global communicaion sysem has shifed aenion o daa ransmission [9]. hroughpu and sof handoff provides a good measure of he qualiy of he wireless daa link: 7

3 JCS& Vol. 11 No. Ocober 11 Maximum hroughpu wih No Addiive Noise Before carrying ou a performance analysis for arbirary number of erminals, N, le us examine a wo erminal case (N=, o provide a clue on he effec of power levels on base saion hroughpu. (i wo erminals case According o [8], he simples nonrivial hroughpu maximizaion of γ = i = 1 j j i Qi Q σ G (1 j occurs when N= and σ =. In his case, he normalized hroughpu U, is a funcion of jus one variable, Z = Q 1 / Q and γ = G / 1 Z. However, γ = G. By adoping he noaion U Z = f GZ f G as he normalized ( ( Z ( σ du = a Z = 1. dz his suggess ha U (1 = f (, could be a local maximum, hroughpu when N= and =, i is found ha G depending on he sign of he second derivaive a Z=1. Examining he second derivaive for arbirary S-shaped f ( γ, we find ha U (1 is a local minimum a low values of G, and a local maximum a high values. Specifically, d U Z = = G[ Gf ( G f ( G ] 1 dz ( d = G ( df ( x x= G dx he sign of he second derivaive of U is he same as ha of he derivaive of x f (x. For he class of funcions f ( x, xf ( x has a bell shape reaching a single maximum a some poin x = G *. herefore, he derivaive of x f (x is posiive for any x, o he lef of G*, and is negaive oherwise. his implies ha U (1 is a local maximum when G is large enough o exceed G*, and is a local minimum oherwise. I is deermined wheher here is a global maximum a he boundaries of he feasible region when Z=1 yields a local minimum; i.e. a Z= and Z=. Even when Z=1 is a local maximum, he probabiliy ha he global maximum lies a he boundary of he feasible region mus also be considered. his suggess ha we compare he equal-received-power condiion (Z=1 wih he condiion ha erminal wo, does no ransmi a all (Z=. he comparison reveals ha Z=1 produces higher hroughpu when f ( G > 1, while Z= produces higher hroughpu when f ( G > 1. Accordingly, he criical processing gain G c, is / defined as he value of G for which f ( G =.. G c has he propery ha single erminal ransmission (Z= or Z= is beer han equal-received-power ransmission when G<G c and conversely, when G>G c. Since G is proporional o he bandwidh, we can sae ha he sysem requires a bandwidh corresponding o a leas G=G c o suppor wo daa erminals. (ii Arbirary Number of erminals e s consider and u u = ; = = σ > ; = = σ (3 (4 Equaions (3 and (4 are used o explore power balancing o deermine wheher hey provide maximum or minimum hroughpu. We exend he approach adoped in he case of N=. o do so, we assume ha signals f from k ransmiers arrive a he base saion wih equal power and ha he oher N-k erminals urn off heir ransmiers when = Q for i = 1,,, k and σ = and G ( k = kf k 1 U ( U (k is he power balancing hroughpu and a funcion of k. k are he ransmiers f are he signals ransmied by k ransmiers G is he processing gain For values of G>G c, U(k has a maximum value a k = N*. hroughpu is maximum when N* signals are received wih equal power and he oher N-N* signals are no ransmied a all.. HROUGHPU AND DEAY MODEING: PARAMEERS DEFINIION he expeced hroughpu of he sysem can be represened mahemaically as: N [ ] ( packe_ lengh E hroughpu = lim ( kb/ s (6 N( is he number of packes ransmied in he sysem packe_lengh is he lengh of packes (assuming equal lengh of packes is he expeced service ime e N ( = 1, represen he inverse of he expeced packe service ime. As approaches infiniy, we have packe _ legnh E [ hroughpu] = ( kb / s (7 76

4 JCS& Vol. 11 No. Ocober 11 is he average ransfer delay excluding he random access phase ha akes place before he uplink channel men. o compue he packe ransmission delay, we segmen he oal delay ino several componens saring wih channel men ill he final acknowledgemen: = (8 x error correcion final ack is he duraion of he channel men procedure for ransmission and processing ime. x is he daa block s firs ime ransmission period based on he packe lengh and coding scheme. error-correcion is he handling ime for backward error conrol including reransmission. final-ack is he final acknowledgemen (ACK message including conrol acknowledgemen message processing and ransmission ime. For our model, = (blocks and final-ack = 4 (blocks x = n(blocks (9 packe_ lengh n =, block _ lengh 3 block_ lengh = error-correcion can be spli ino wo: error-correcion = rex nack (1 rex is he reransmission ime of erroneous daa blocks nack is he negaive acknowledgemen (NACK message processing and ransmission imes. recx bler( cs, ci n = ( blocks (11 ( 1 bler( cs, ci bler is he block error rae e m(n be a random variable represening he number of NACK messages successfully sen during he ransmission of a packe consising of n RC blocks, which is equivalen o he number of reransmission cycles for he frame. e one-nack denoe he ransmission ime for one NACK message (urina and udiving, 1997, E[ m(n ] = (1 nack one nack o accoun for errors in he downlink ACK/NACK messages, we noe ha he process of sending one NACK message consiss of pars: iniial ransmission and reransmission of he message. Here iniial-nack = 3. he iniial afer-frame NACK message is ransmied wih. I is correcly received wih probabiliy ( 1 bler( CS 1, Ci. If he NACK is no received by he Mobile Saion (MS, he MS will resend he oldes block wih errors or he oldes unacknowledged block, expecing he base saion (BS o re-send he ACK/NACK message. Our model assumes ha if he daa block is again in error, he whole process is repeaed. Only if boh he resen daa block and he NACK are successful, will he NACK ransmission process complee. he round rip delay is 4 blocks. Assuming here is no limi on he number of ransmissions allowed for one NACK message, 4blerCS ( 1, Ci one nack = iniial nack (13 ( 1blerCS ( 1, Ci ( 1blerCsCi (, 6. HE SYSEM MODE Recall from equaion (6 ha he hroughpu is given by N hr = lim he delay ( P P hr = N( = and P is he packe (14 lengh. Also recall from equaion (8 ha = x error correcion is aken as (blocks final-ack is aken as (4 blocks P x = n = B 3 = B, final ack which represens he block lengh or payload. Now, subsiuing he respecive parameers ino equaion (1 gives: bler n error correcion = onenack E 1 bler = iniial nack ( 1 bler( 1 bler [ m( n ] 4 bler E (1 Hence, equaion (14 can be rewrien as hr= P bler n B 1bler 4bler n ( 1 bler i= 1 n i n bler i P iniialnack [ m( n ] n1 n ( 1bler ( 1bler ( m1 (16 m is he number of messages sen during ransmission he hroughpu can also be relaed o bler and he daa rae as: 77

5 JCS& Vol. 11 No. Ocober 11 hr = ( 1 bler R R is he daa rae Bu P = ( 1 bler R P = ( 1 blerr R =. 4 payload P = ( 1 bler(.4 x payload (17 (18 Average hroughpu y =.x 9.87 R = EXPERIMENA DAA ANAYSIS Figure reveals ha for he duraion under sudy, he sysem hroughpu flucuaes a he various base saion locaions. his could be aribued o nework insabiliy caused by propagaion impairmens ha disor he received daa packes, resuling in poor delivery a he receiver s end. In Figure 3, we plo he average hroughpu as a funcion of he base saions over he sudy period. We observe ha he hroughpu performance is no significanly influenced by he base saions, as eviden from he obained correlaion coefficien (R value (i.e. R =.13. However, oher facors such as increased number of users, fading, inerference due o noise, equipmen malfuncioning or downime, ec., could conribue o a grea exen and explain beer he insabiliy ha has hreaened he sysem s performance. Airel Nigeria herefore requires an opimized echnique o deal wih hese impairmens. hroughp u BSC 1 Figure. Forward-link hroughpu analysis a various BSs during he sudy period Days Duraion (days Figure 3. Graph of average hroughpu vs. number of days 8. SIMUAION INPU o evaluae he proposed sysem model, sample inpus were used o simulae he model. Some of he inpu daa were drawn from he field, under ideal condiions. able shows he parameers used for he research. able. Sample simulaion parameers and heir values Parameer Value he packe lengh (P 1,6 kb/s he block lengh or = payload (B CS-=3 =38 = Duraion of he channel men blocks procedure for ransmission and processing ime ( Average forward-link kb/s hroughpu from he field (hr Block error rae (bler.1-.9 Number of packe ransmied (n Iniial afer-frame negaive acknowledgmen message(s ransmied ( iniial_nack Number of message(s sen during ransmission(m 4 3 blocks 1-1 Sample oupus were produced in he form of graphs and sudied. A discussion of he oupus obained is done in he following secion. 78

6 JCS& Vol. 11 No. Ocober DISCUSSION OF RESUS Figure 4 shows he effec of messages ransmied on he sysem hroughpu using diverse block lenghs or payload sizes. We observed ha as he coding scheme increases, he hroughpu also increases, which validaes our opimizaion echnique. As more messages are sen, hroughpu ends o slowdown for he differen coding schemes, excep -4, which shows improved sysem performance. Shown in Figure, is a relaionship beween he hroughpu and packe lengh. Comparing he hroughpu using varied packe lenghs, we observed ha he packe lengh had dominan effec on he sysem hroughpu. he hroughpus for shorer packes are subsanially lower han longer packes. Again, he hroughpu seems inhibied for oher coding schemes excep, which shows improved sysem performance. hroughpu Graph of hroughpu vs. Number of messages sen during packe ransmission CS No. of message sen during ransmission Figure 4. Graph of hroughpu vs. number of massages sen Graph of hroughpu vs. Packe lengh bler, he hroughpu increases rapidly bu as he bler exceeds., he hroughpu slows down, indicaing nework congesion as a resul of more users accessing he sysem. his may lead o he possibiliy of poor message delivery a he receiver s end. hroughpu Graph of hroughpu vs. Block error rae CS Block error rae x 1-3 Figure 6. Graph of hroughpu vs. block error rae In Figure 7, we examine he performance of he sysem by sudying he effec of packe lengh on sysem s delay using he four coding schemes. We observed ha he longer he packe lengh, he longer (ime i akes o deliver he desired message. However, as he coding scheme is increased (from o, he delay in ransmission shorens. his explains he robusness score of he various coding schemes in dealing wih channel impairmens. herefore, nework operaors should implemen efficien coding schemes ha will opimize he sysem s performance and guaranee effecive message deliveries Graph of Delay vs. Packe lengh 3 3 hroughpu 1 Delay CS Packe lengh Figure. Graph of hroughpu vs. packe lengh As expeced in Figure 6, he hroughpu increases wih he block error rae (bler. A lower. CS Packe lengh Figure 7. Graph of delay vs. packe lengh 79

7 JCS& Vol. 11 No. Ocober CONCUSION Due o sharing of common communicaion channel by users, inerference in form of noise is always generaed and is he main limiaion for deermining he opimum hroughpu. A general overview of GPRS echnology has been presened and a summary of hose aracive feaures, which demonsraes GPRS echnologies superioriy over G cellular nework echnologies and oher 3G echnologies given. his paper has been able o opimize he hroughpu in GPRS cellular nework hrough forward-link (downlink ransmi power balancing by simulaing a model wih relevan parameers. Some pracical issues regarding he applicaion of his research have been considered and series of invesigaions idenifying he key feaures and parameers, as well as he funcionaliy of he GPRS coding schemes have been presened. Simulaion resuls show ha he sysem s hroughpu has been opimized by he proposed model. his echnique will aid in ransmiing signals from he GPRS base saion wih equal possible power. [9] Goodman D. and Maranz Z. (3. Maximizing he hroughpu of CDMA Daa Communicaions. In Proceedings of 8 h IEEE Conference on Vehicular echnology. : REFERENCES [1] Rodriguez V. and Goodman D. J. (3. Prioriized hroughpu Maximizaion via Rae and Power Conrol for 3G CDMA: he wo erminal Scenario. In Proceedings of 4 h Alleron Conference Communicaion, Conrol and Compuing. [] Goodman D. (. Performance Analysis of Radio ink/media Access Conrol of General Packe Radio Service Sysem, Ph.D. hesis Repor. WINAB R 9. [3] Hamalainen D. and Kari P. (199. Handover and Dynamic Channel Allocaion echniques in Mobile Cellular Nework. IEEE ransacion on Vehicular echnology. 44(:9-37. [4] Bali S. A. (1. Sudy of he hroughpu of General Packe Radio Service (GPRS for Inerne Applicaions. M.Sc. hesis, Universiy of Jordan, Jordan. [] indemann C. and hümmler A. (1. Evaluaing he GPRS Radio Inerface for Differen Qualiy of Service Profiles. In Proceedings of 1 h GI/IG Fachagung Kommunikaion in Vereilen Sysemen (KiVS, Hamburg, Germany: [6] arson R. and Olofsson D. (1998. Handover and Dynamic Channel Allocaion echniques in Mobile Cellular Nework. IEEE ransacion on Vehicular echnology. 44(: [7] Cai J. and Goodman D. (1997. General Packe Radio Service in GSM. IEEE Communicaion Magazine. 3: [8] Sung C. W. and Wong W. S. (. Power Conrol and Rae Managemen for Wireless Muli-media CDMA sysem. IEEE ransacions on Communicaion. 49(7: