989 High-speed Simultion of the GPRS Link Lyer J Gozlvez nd J Dunlop Deprtment of Electronic nd Electricl Engineering, University of Strthclyde 204 George St, Glsgow G-lXW, Scotlnd Tel: +44 4 548 206, Fx: +44 4 552 4968 j. ozlvez @ comms.eee. s trth. c. uk, i.dun OD @ eee. s trth. c.uk ABSTRACT The Generl Pcket Rdio Service (GPRS) is due to be implemented in current mobile networks, introducing whole new rnge of mobile dt pplictions nd services. Considerble effort hs been devoted to test its cpbilities nd system performnce. n this pper, modulr nd high-speed GPRS simultor is presented. n order to vlidte the simultion method, the GPRS link lyer performnce is ssessed. The performnce, for single slot pcket dt chnnel, is evluted in typicl urbn environment for rnge of mobile velocities. The impct of frequency hopping on link qulity mesures is lso evluted. re shown nd discussed in section V. Finlly, conclusions re drwn, nd some possibilities of future reserch re evoked. 2. GPRS CHANNEL CODNG Prior to trnsmission, GPRS dt pckets re segmented into smller dt blocks. The segmenttion process is depicted in figure.......... NTRODUCTON The Generl Pcket Rdio Service (GPRS) [l] hs been developed s stndrdised system for the provision of pcket dt services for both evolved GSM nd TDMN36 networks. ts higher bndwidth efficiency compred to circuit switched systems, such s GSM, is due to the introduction of cpcity on demnd nd the sttisticl multiplexing of users in single slot. GPRS lso provides incresed dt rtes through the lloction of multiple slots to single user, reching mximum throughput of 7.2 kbits/s when considering eight slots per user. GPRS introduces three new chnnel coding schemes. The fourth GPRS coding scheme, the more robust one, is lredy used by GSM for the Slow Associted Control Chnnel. These coding schemes provide different degrees of error protection. Their impct on the rdio link qulity needs then to be ssessed. For this purpose, n enhnced softwre version of simultor previously used to evlute the impct of Link Adpttion in GSM [2], hs been used. By mens of this high-speed simultor, the GPRS rdio link performnce in typicl urbn environment, under different operting conditions, is evluted. The work hs considered the lloction of single slot per frme. The outcome of this work not only ~ represents n independent verifiction of the GPRS link lyer performnce, but lso n ssessment of the cpbilities of the simultor for the study of pcket bsed systems such s GPRS or EDGE. This pper is orgnised s follows. n section, the chrcteristics of the GPRS chnnel coding schemes re described. The simultor nd method used for this work re presented in section. The GPRS link lyer results Figure Pcket trnsmission flow The dt pckets, received from network lyers, re first split into Logicl Link Control (LLC) frmes. The LLC frmes re then segmented into Rdio Link Control (RLC) blocks. The resulting RLC dt blocks re then coded nd block-interleved over four norml bursts in consecutive TDMA frmes. The RLC block s dt field length will depend on the chnnel coding schemes used. Four chnnel coding schemes, CS to CS4, re specified for the GPRS pcket dt trffic chnnels [3]. Ech scheme hs been designed to provide different resilience to propgtion errors under unfvourble rdio conditions, s depicted in figure 2. Throughput vs CR lor the different GPRS coding schemes Crrier to nterference Rtio (db) Figure 2 Performnce of the GPRS coding schemes 0-7803-6465-5/00 $0.00 Q 2000 EEE
990 The different coding schemes offer trde-off between throughput nd coding protection, pving the wy for the ppliction of Link Adpttion to GPRS. Coding rtes of 2, 23, 34 nd re provided. CS corresponds to the more robust scheme while CS4 does not use ny error correction. CS to CS3 re bsed on hlf rte convolutionl encoder. However, they differ on the puncturing schemes pplied to the output of this encoder. Block Check Sequences re used in ll the schemes to fcilitte the error detection t the receiver. The outcome of the coding process is block of 456 bits. The encoding process is illustrted in figure 3 while the chrcteristics of the different coding schemes re summrised in tble. been modelled s single continuous strong interferer, following the chrcteristics of the testbed used by ETS [4]. The therml noise t the receiver hs lso been included. Figure 4 Trnsmission chin Figure 3 GPRS coding process Tble GPRS chnnel coding prmeters USF USF %; The GPRS chnnel codingldecoding functions use the error dtbse s illustrted in figure 5. When simulting the physicl lyer, the chnnel coding output is first interleved nd then the rdio propgtion effects re dded. The output of this sum is then de-interleved before being pssed to the chnnel decoding process. However, de-interleving the error ptterns nd dding them to the chnnel coding output is equivlent. This lst solution hs been dopted for this simultor for the ske of simplicity. A common chrcteristic of ll the coding schemes is the presence of the Uplink Stte Flg (USF) in the RLC blocks heder. The USF is trnsmitted in the downlink to indicte which Mobile Sttion (MS) should trnsmit in the next uplink slot. The USF comprises three bits t the beginning of ech downlink rdio block. t enbles the coding of eight different USF sttes used to multiplex different MSs. Given its importnce, precoding process is pplied to the three bits representing the USF; s shown in tble. 3. SMULATON METHOD This section describes the simultion method used for this work. The significnt reduction in simultion time is emphsised. A. Testbed An enhnced softwre version of the demonstrtor reported in [2] hs been used in order to study the performnce of the GPRS Link Lyer. This simultor models the trnsmission chin through the use of dtbse of error ptterns produced with the bit level simultion pckge COSSAP [5]. Figure 4 illustrtes the GPRS trnsmission chin. Co-chnnel interference hs Figure 5 GPRS simultor The derivtion of n error dtbse significntly reduces the simultion time whilst mintining ccurcy of rdio link qulity representtion. n fct, the error dtbse is independent of the dt bits trnsmitted [5] mking it possible to be reused whenever the rdio pth effects hve to be tken into ccount. The production of the error dtbse (representing pproximtely the trnsmission of 50000 RLC blocks) tkes round 5 dys of COSSAP simultions using Sun ULTRA Sprc worksttion. Tht is pproximtely the time tht will be required to nlyse the GPRS coding schemes performnce if the trnsmission chin nd the chnnel coding were simulted using COSSAP. By decoupling the simultion of the coding schemes from the trnsmission chin, nd once the error dtbse hs been produced, the sme study only needs round hours with 450MHz PC (with 52 Mbytes of RAM). This results in significnt reduction of 97% in simultion time. The effect of ny chnnel coding nd interleving schemes cn therefore be quickly studied. Also, due to the design of the simultor, other frequency
99 hopping ptterns cn be simulted without the need to simulte gin the trnsmission chin. B. ndependent replictions method A simultion run represents prticulr relistion of the environment being simulted. Different simultion runs cn then differ significntly from the true response of the system. n order to reduce the vrince between the output of different simultion runs, number of simultion runs must be mde before mking ny conclusion bout the relibility of the informtion extrcted. The method selected for the simultions presented in this pper is the ndependent replictions method [6]. The independence between replictions is introduced in the production of the error dtbse. Seven independent replictions were done, ech one simulting the trnsmission of 2000 RLC blocks. The results presented in the following section represent the men vlues from the seven independent replictions with 95% confidence intervl. To illustrte the sttisticl ccurcy of the results, the estimted reltive error [6], defined s the rtio of the hlf-length of the confidence intervl to the smple men, will be specified. 4. LNK LEVEL SMULATON RESULTS n order to vlidte the simultion pproch previously described, the CPRS link lyer performnce is estimted in terms of the Block Error Rte (BLER) nd Uplink Stte Flg (USF) Error Rte. The performnce is ssessed for the four GPRS coding schemes, CS to CS4. The sttisticl ccurcy of the results presented in this section is lso specified. (i.e., t lest one bit error detected by the Block Check Sequence during the chnnel decoding) to the totl number of rdio blocks. Simultion results for both speeds nd without frequency hopping, re shown in figures 6 nd 7. The corresponding results when implementing frequency hopping re depicted in figures 8 nd 9. The BLER reference performnce for pcket dt chnnels (lo%), estblished in [4], is lso plotted in the figures. s 0. 0.0 BLER vs W for TU50 without FH 0.00 CS U cs2 --*-- cs4 - ref performnce le-005 J 0 5 0 5 20 U Figure 6 BLER vs CA, without FH t 50 k" - BLER vs Cn lor TU5 without FH A. Assumptions A single slot per frme lloction policy hs been ssumed for this study. The simultions presented in this pper were conducted for typicl urbn chnnel model, s defined in [4], t two prticulr speeds, 5 km/h nd 50 kmh, corresponding to pedestrin nd cr scenrio. The cpcity limiting fctor for cellulr system in highly populted re, such s n urbn environment, is the co-chnnel interference. Therefore only simultions for n interference-limited cse re considered in this pper, hence yielding the performnce versus Crrier to nterference Rtio (CA). For the purpose of this work, the crrier frequency ws set to 900 MHz. Since GPRS hs been considered for the evolution of GSM nd TDMN36, simultions were conducted with nd without frequency hopping. An idel frequency hopping scheme (hopping between four frequencies), where perfect decorreltion between bursts is ssumed [4], hs been simulted. As proposed in [4], the interference nd the wnted signls hve the sme frequency hopping sequence. B. Block Error Rte d 0.00 '0 5 20 cn Figure 7 BLER vs C/, without FH t 5 km/h BLER vs CA for TU50 with FH4 0.m' 0 5 0 5 20 W Figure 8 BLER vs C/, with FH t 50 km/h The BLER gives n estimte of the rdio link qulity. t is defined s the rtio of rdio blocks received in error
992 BLER vs U for TU5 with FH4 high speeds) will decrese the performnce since errors will ffect more rdio blocks, nd the scheme will not be ble to correct them. 0 0 0 Comprison of BLER for CS4 for TU5 nd TU50 without FH nd with FH4 5=Zz-------=*- --*.-..- 0 00 o.cqo 0 5 0 5 20 U Figure 9 BLER vs C/, with FH t 5 kmh CS4-TUSnoFH CS4-TUSFH4 - *-- CS4-TUSOnoFH --*- CS4-TUSOFH4 ref performnce Figures 6 to 9 demonstrte the different robustness of the four coding schemes for vrying rdio conditions. These results re in ccordnce with those presented in [7]. By extrcting the CA t which the 0% reference performnce is reched, nd compring them with the minimum requirements imposed by ETS [4], it my be concluded tht the GPRS simultor respects ETS stndrds. Figure 0 illustrtes the impct of speed nd frequency hopping in the BLER performnce (in this cse CS). 5 d Comprison of BLER for CS for TU5 nd TU50 without FH nd with FH4 0. 0.0 0.00 0.000 9-005 0 5 0 5 20 U Figure 0 mpct of speed nd frequency hopping on the CS BLER performnce The performnce is significntly improved by the use of frequency hopping, independently from the mobile speed. This improvement is due to the rndomistion (or spred), when using frequency hopping, of the propgtion errors. With robust scheme, spreding the errors hs positive effect s even if more blocks re ffected, the length of errors is smller nd the coding scheme cn correct them. The sme effect is lso obtined by n increse in the speed when frequency hopping is not used, but the gin in performnce is lower. These chrcteristics re vlid for CS, CS2 nd CS3 but not for CS4 s depicted in figure. CS4 corresponds to prticulr cse s speed of 5 km/h without frequency hopping gives the best performnce. This is explined by the fct tht CS4 hs no coding. Spreding the errors (either by frequency hopping or 0 0 0 5 20 U Figure mpct of speed nd frequency hopping on the CS4 BLER performnce The estimted reltive error of the presented men BLER vlues is below 5% for CAS in the rnge of 4 to 7 db for ll the operting conditions. The use of the CSl coding scheme, frequency hopping nd high speeds reduces the precision of the results. This is explined by the fct tht the precision is dependent on the number of error events, nd using CS or procedures tht spred the propgtion errors decrese the mount of errors nd therefore the precision. C. USF Error Rte As previously stted, the USF Error Rte influences the Medium Access Control lyer performnce. t is defined s the rtio of incorrectly interpreted USFs to the totl number of received USFs. The output of the USF precoding process is identicl for CS2, CS3 nd CS4 so only two curves will be presented in the following figures. Simultion results for both speeds nd without frequency hopping re shown in figures 2 nd 3. The corresponding results when implementing frequency hopping re depicted in figure 4. Due to the importnce of the USF, ETS proposes % reference performnce for the USF Error Rte. 0. : E 0.0 U CO 0.00 o.wo' USF Error Rte vs C/ for TU5 wilhoul FH 5 0 5 c/ ' cs -%-- cs2-cs3cs4.--..---- Figure 2 USF Error Rte without FH, t 5 km/h
993 0. USP Error Rte vs C/ fr TU50 without FH The estimted reltive error of the presented USF Error Rte men vlues is below 0% for CAS of round 9 db for ll the operting conditions. The estimted reltive error constrint ws relxed becuse of the strong coding schemes pplied to the USF. - m U m 0 5 U 0) 2 0.0 0.00 0 2 4 6 8 0 2 4 cil Figure 3 USF Error Rte without FH, t 50 km/h 0. 0.0 0.00 USF Error Rte vs cil fr TU5 nd TU50 with FH4 0.000 0 2 4 6 0 C/ Figure 4 USF Error Rte, with FH, t 5 km/h nd 50 km/h The CS2lCS3lCS4 USF coding shows better robustness under ll operting conditions. The USF Error Rte results lso respect the minimum interference rtio for which the % reference performnce for co-chnnel interference shll be met [4]. A better view of the impct of the velocity nd frequency hopping is depicted in figure 5. The sme conclusions s for the BLER results cn be drwn. Cmpnson of USF Error Rte lor CS2 tor TU5 nd TU50 without FH nd wfih FH4 0 2 4 6 8 0 2 4 6 cn Figure 5 mpct of speed nd frequency hopping on the CS2 USF Error Rte performnce 5. CONCLUSONS AND FUTURE RESEARCH n this pper, high-speed simultor using n error dtbse hs been used to estimte the GPRS link lyer performnce by mens of two link qulity mesures. The impct of the speed nd frequency hopping on the GPRS performnce hs lso been presented. The results show the different robustness of the GPRS chnnel coding schemes for vrying rdio conditions. High speeds nd the use of frequency hopping increse the performnce, except for CS4 due to its lck of error protection. The results presented in this pper re in ccordnce with ETS requirements nd with results presented in other ppers, vlidting then the simultion method used. The present study hs ssumed the lloction of single slot per frme. Further reserch will be to ssess the GPRS link lyer performnce when llocting severl slots per frme, tking into considertion the impct of correltion between contiguous slots. Also, s GMSK modultion hs been dopted s fll bck mode in EDGE [8], the error dtbse will lso be reused for future evolved versions of the simultor including EDGE. 2 6 7 8 6. REFERENCES ETS, GSM 03.64; Overll description of the GPRS rdio interfce. Stge 2, version 6..0, Relese 997. J Dunlop, J Pons, J Gozlvez, nd P. Atherton, A Rel-time GSM Link Adpttion Hrdwre Demonstrtor, to be presented t EEE VTC2000- Spring, My 5-8 2000, Tokyo. ETS, GSM 05.03; Chnnel coding, version 7.0., Relese 98. ETS, GSM 05.05; Rdio Trnsmission nd reception, version 7..0, Relese 998. J Pons, Appliction of Link Adpttion to evolved TDMA mobile communictions systems. PhD thesis. October 999. University of Strthclyde. Simultion modelling & nlysis. Averill M.Lw nd W. Dvid Kelton. McGrw-Hill. 2 d edition, 99. P. Schrmm et l, Rdio nterfce Performnce of EDGE, Proposl for Enhnced Dt Rtes in Existing Digitl Cellulr Systems. Proc of EEE VTC, My 998, Ottw. Furuskr, Mzur, Muller nd Olofsson, EDGE: Enhnced Dt Rtes for GSM nd TDMAl36 Evolution. EEE Personl Communictions mgzine, June 999.