1 Impact of Iterferece o the GPRS Multslot Lk Level Performace Javer Gozalvez ad Joh Dulop Uversty of Strathclyde - Departmet of Electroc ad Electrcal Egeerg - George St - Glasgow G-XW- Scotlad Ph.: + 8 8, Fax: + 98, e-mal: ABSTRACT The GPRS lk level performace has bee usually assessed cosderg the allocato of a sgle slot per TDMA frame. However, multple slots may be allocated to a sgle user order to crease the trasmsso badwdth. Ths paper compares the GPRS multslot lk level performace uder two terferece scearos that ca be assocated wth performace bouds. Numercal results show that the terferece patter ca have a mpact o the lk level performace.. INTRODUCTION The Geeral Packet Rado Servce (GPRS)  has bee developed as a stadardsed system for the provso of packet data servces for both evolved GSM ad TDMA/ etworks. Its hgher badwdth effcecy compared to crcut swtched systems, such as GSM, s due to the troducto of capacty o demad ad the statstcal multplexg of users a sgle slot. GPRS also provdes creased data rates through the allocato of multple slots to a sgle user, reachg a maxmum theoretcal peak rate of 7. kbts/s whe eght slots are assged at the same tme to a sgle user. GPRS performace s usually assessed two stages: system level ad lk level. The former models a moble rado etwork takg to accout aspects such as moblty, dstace atteuato, shadow fadg ad the behavor of terferers. Its output s geerally expressed terms of the dstrbuto of the Carrer to Iterferece Rato (). O the other had, lk level studes model the rado lk at a bt level provdg the lk qualty (e.g., Block Error Rate) as a fucto of the average. Due to the large smulato tme, a sgle rado lk s usually cosdered. The results for both studes are the merged to aalyse the global performace. Usual procedures to terface both levels are to use the lk level results as a source of formato for the system level. The lk level performace s usually represeted as a set of look-up tables  mappg to qualty for dfferet sets of operatg codtos (e.g., varyg the termal speed, the propagato chael ad cosderg the use of Frequecy Hoppg). The study of multslot systems has bee cocetrated o system level aalyss. Partcular atteto has bee pad to the MAC operato (e.g., ). Curret system level studes rely o look-up tables obtaed through lk level studes based o sgle slot trasmsso. Ths approach assumes that whe trasmttg RLC blocks o dfferet chaels of the same frame, ther lk level performace (.e., whether the blocks have bee receved error or ot) s totally ucorrelated. Ths assumpto ca be justfed whe performg studes that are ot depedet o the tme varablty of the system ad that average out over tme the stataeous lk level performace (e.g., capacty studes, , ). However, whe studyg adaptve techques such as Lk Adaptato or Power Cotrol, the tme propertes of the lk qualty are of paramout mportace. A tal vestgato of the multslot lk level performace s reported . Ths study demostrated that the lk level performace ca be correlated uder certa operatg codtos, wth the degree of correlato depedg o the termal speed ad mea. The spacg betwee slots of a same frame used to trasmt dfferet RLC blocks proved to have a key fluece suggestg that the performace wll deped o the partcular slots assged to a user a multslot trasmsso. The scearo aalysed  cosdered the same domat multslot terferer for all trasmttg slots. However, a more typcal scearo a multslot trasmsso mght experece terferece from the same user some of the slots, whereas other slots mght be terfered by other depedet terferers. Therefore, the terferece may be dstrbuted o the slots a umber of cofguratos. Evaluatg the lk level performace for all of them would be too computatoally expesve. Ths paper complemets the GPRS multslot lk level study  by cosderg two dfferet terferece scearos that provde a lower ad upper boud o the correlato of the terferers. The scearo modellg the terferece as a sgle domat multslot user (scearo studed ) s oe of these boudary scearos. The other scearo, dscussed ths paper, models the terferece o each dvdual slot a multslot trasmsso as a dfferet, depedet user. The comparso wll allow the fluece of the terferece patter o the multslot lk level performace to be assessed.. GPRS RADIO INTERFACE The GPRS rado terface ca be modeled as a herarchy of logcal layers wth specfc fuctos . Pror to trasmsso, data packets are segmeted to smaller data blocks across the dfferet layers, wth the fal logcal ut beg the Rado Lk Cotrol (RLC) block. The resultg RLC data blocks are the coded ad block-terleaved over four ormal bursts cosecutve TDMA frames. The RLC block s data feld legth wll deped o the chael Codg Schemes (CSs) used. Four chael codg schemes, CS to CS, are specfed for the GPRS packet data traffc chaels . Each scheme has bee desged to provde dfferet
2 reslece to propagato errors uder ufavorable rado codtos, offerg a trade-off betwee throughput ad codg protecto. CS correspods to the more robust scheme whle CS does ot use ay error correcto. CS to CS are based o a half rate covolutoal ecoder. However, they dffer o the pucturg schemes appled to the output of ths ecoder. Block Check Sequeces are used all the schemes to facltate the error detecto at the recever. The characterstcs of the dfferet codg schemes are summarsed Table. A effcet utlzato of the spectrum s obtaed usg a multslot chael reservato scheme. The GSM stadard defes 9 dfferet multslot classes, each oe allocatg a dfferet maxmum umber of slots for recepto ad trasmsso, ad mposg dfferet restrctos . Depedg o the multslot capabltes of a MS, the umber of avalable chaels ad the system load, RLC blocks belogg to oe LLC frame ca be set o dfferet physcal chaels smultaeously ad parallel. Usg ths reservato scheme, trasfer delays ca be reduced ad the assged badwdth ca be vared dyamcally. The multple slots allocated for ether recepto or trasmsso eed ot be cotguous. However, the multslot class of the MS wll lmt the combatos ad cofguratos allowed whe supportg multslot commucatos due to for example the type of the MS, the ecessty to perform adjacet cell power measuremets or ay costrats mposed by the servce selected. Some examples llustratg these lmtatos ca be foud Aex B of . The effect of the ocotguous allocato of multple slots mght the fluece the GPRS multslot lk level performace. Scheme Code rate Payload Data rate kb/s CS- / 8 9. CS- / 8. CS- /. CS- 8. Table : GPRS codg schemes parameters. MULTISLOT LINK LEVEL ANALYSIS The coherece tme of the chael  s approxmately te GPRS frames at km/h, oe frame at km/h, ad oe slot ad a half at km/h, for a moble system operatg at 9 MHz. There s therefore a potetal correlato betwee sgals receved dfferet slots, whch mght fluece the stataeous lk level performace o dfferet slots of the same frame. A multslot trasmsso scheme assgg two slots from the same frame to a sgle user s llustrated Fgure. The sgals receved slots ad may the be correlated. Ths correlato ca be due to the correlato betwee sgals trasmtted slots of the same frame (e.g., betwee slots ad of frame ) ad betwee slots of dfferet frames (e.g., slot of frame ad slot of frame ). The correlato betwee slots of the same frame wll be termed ter-slot correlato ad the correlato betwee slots of dfferet frames wll be termed ter-frame correlato. Both correlatos ca the fluece the post-decodg state (that s, whether a block has bee receved error or ot) whch the RLC blocks are receved. The spacg betwee the slots of the same frame used to trasmt the dfferet RLC blocks s a mportat parameter fluecg ths correlato. A spacg of zero slots correspods to the case whch two RLC blocks are trasmtted cotguous slots of four cosecutve frames. Fgure correspods to the case of a fve slot spacg betwee the two physcal chaels used to trasmt the RLC blocks. The maxmum spacg s sx slots. Frame Frame Frame Frame Frame RLC Block Fgure : RLC Blocks trasmsso The aalyss reported  hghlghted that the tradtoal lk qualty metrc, amely mea Block Error Rate (BLER), caot be used to study the effect of correlato multslot lk level results, as t does ot reflect the effect of fast fadg ad tme propertes of the GPRS lk layer performace. A ew parameter termed correlato probablty, P c, was specfcally defed for ths purpose. Ths parameter defes the probablty that a RLC block trasmtted a arbtrary slot, Y, s receved, after chael decodg, wth same state (error/o error) as a RLC block trasmtted aother slot, X, of the same frame. Let X represet the post-decodg state whch a RLC block trasmtted the slot of four cosecutve frames s receved. X takes the value f the block has bee receved error ad the value f the block has bee correctly decoded. Pr[,] s defed as the codtoal probablty that two RLC blocks trasmtted dfferet slots of the same frame are receved wth error, for a slot spacg, gve that the frst block s receved error. Smlarly, Pr[,], Pr[,] ad Pr[,] may be defed as follows: Pr[,] Pr[,] Pr[,] Pr[,] slot spacg Slot Slot = X = X = X = X = ] = ] RLC Block = ] = ]
3 The correlato probablty P c ca the be expressed as follows: P c = Pr[,] + Pr[, ] P c = Pr[,] Pr[,] + wth [,] ad + < 7. Cosderg the example llustrated Fgure, P c wll represet the probablty that RLC block trasmtted slot s receved wth error/o error f RLC block s receved wth error/o error.. SIMULATION ENVIRONMENT A ehaced software verso of the demostrator reported  has bee used order to study the performace of the GPRS Lk Layer. Ths smulator models the trasmsso cha through the use of a database of error patters produced wth the bt level smulato package COSSAP. Fgure llustrates the GPRS trasmsso cha. The effect of the thermal ose at the recever has bee cluded. Iformato bts Burst Buldg Modulato Fgure : Trasmsso cha The GPRS chael codg/decodg fuctos use the error database as llustrated Fgure. Whe smulatg the physcal layer, the chael codg output s frst terleaved ad the the rado propagato effects are added. The output of ths sum s the deterleaved before beg passed to the chael decodg process. However, de-terleavg the error patters ad addg them to the chael codg output s equvalet. Ths last soluto has bee adopted here for the sake of smplcty. The dervato of a error database sgfcatly reduces the smulato tme  whlst matag accuracy of rado lk qualty represetato . I fact, the error database s depedet of the data bts trasmtted makg t possble to be reused wheever the rado path effects have to be take to accout. Iformato bts Chael Model GPRS chael codg Iterferer Modulato & Chael Bt Error database Deterleaver X GPRS chael decodg Comparso Nose Correlator Equalser Fgure : GPRS rado lk smulator Bt Error database Receved blocks & statstcs The smulatos have bee performed for typcal urba ad rural scearos usg chael models followg the recommedatos proposed . The mpact of the terferece has bee evaluated for dfferet speeds. Speeds of km/h, ad km/h have bee cosdered for the urba scearo. The rural evromet has bee evaluated for a speed of km/h. For the purpose of ths work, the carrer frequecy was set to 9 MHz. The capacty lmtg factor for a cellular system s the co-chael terferece. Therefore oly smulatos for a terferece-lmted case are cosdered ths paper. Co-chael terferece has bee modelled as a sgle strog radom, cotuous GSM-modulated terferg sgal followg the gudeles for the testbed descrbed . Moreover,  t has bee demostrated that a sgle terferer geerally domates the terferece. The terferg sgal s ucorrelated wth the trasmttg sgal. The followg results assume the same mea durg the four cosecutve TDMA frames used to trasmt RLC blocks, as the effect of fast fadg s aalysed. A costat mea mght be expereced hghly loaded systems. It mght also be targeted for the applcato of adaptve techques, whch requre a terferece evromet as stable as possble sce large ad fast varatos mght lead to urelable chael estmates, producg a poor performace. Moreover, a costat mea correspods to the scearo where the effect of correlato o the multslot lk level performace should be stroger. The frst step should the be to check whether the correlato has ay effect the lk level performace uder a costat, as f t s ot the case, the t wll be very ulkely that t would have ay mpact uder a varable scearo. As prevously explaed the am of ths paper s to compare the GPRS multslot lk level performace uder two referece scearos that ca be assocated wth performace bouds. The frst scearo (Scearo ), studed , models the terferece as a sgle multslot user. The trasmttg slots wll hece be all terfered by the same user. The secod scearo (Scearo ) models the terferece as eght depedet sgle slot users. The trasmttg slots are therefore each terfered by dfferet ad ucorrelated users.. NUMERICAL RESULTS Ths secto evaluates ad compares the GPRS multslot lk level performace, by meas of the correlato probablty, uder the above metoed referece terferece scearos. Frst of all, t s worth otg that the umercal values of the correlato probablty dffer betwee the two scearos uder the same operatg codtos as depcted Fgures ad. Ths s maly due to the fact that Scearo the correlato probablty s flueced by the correlato wth the sgal receved from the trasmtter ad by the correlato wth the sgal receved from the multslot terferer. O the other had, Scearo, the correlato wth sgals receved from the terferers s reduced ad the mpact of such reducto reflects o the umercal values of the correlato probablty.
4 Despte dfferet umercal values, smlar coclusos ca be reached for both scearos regardg the effect of the mea ad robustess of the dfferet GPRS codg schemes (CS). Fgure ad Fgure plot the probablty P c, uder Scearo ad Scearo, for all four codg schemes, a slot spacg of four ad a speed of km/h. From these fgures t ca be observed that for both scearos ad depedetly of the codg scheme appled the probablty P c tally decreases whe the s creased but evetually creases after a partcular value. The probablty P c s creased whe the RLC blocks set dfferet slots have bee both receved ether error (Pr[,] ) or wth o error (Pr[,] ). P c decreases whe the RLC blocks have bee receved wth dfferet states. For very low s (e.g., = db), the probablty s hgh due to the bg quatty of errors ad therefore the hgh value of Pr[,]. Whe the creases, the umber of trasmsso errors decreases ad so does Pr[,], where as Pr[,] ad Pr[,] crease explag the decrease P c. Whe the CS has a strog error correcto capablty, t wll be able to correct more errors as the creases ad therefore Pr[,] ad Pr[,] wll decrease to the detrmet of Pr[,]. Whe Pr[,] becomes the domat factor P c, P c wll start creasg aga. The pot at whch ths occurs s dfferet for the codg schemes. Ths s due to the fact that for the same mea as CSs less robust are used, fewer errors are corrected ad the probabltes Pr[,] ad Pr[,] take hgher values compared to more robust CSs. Oly whe the has creased to the pot where each CS ca hadle properly the errors, P c starts to crease. O the other had, the effect of slot spacg o the correlato probablty dffers betwee both scearos for certa speeds. As show Fgure the slot spacg has a strog mpact o the correlato probablty uder certa operatg codtos wth Scearo. These codtos correspod to the rage of average s where the error correctg capabltes of the CSs may or may ot correct the trasmssos errors (.e., whe ether Pr[,] or Pr[,] are domat factors P c ). As show Fgures ad ths rage of mea s vary wth the error correctg capabltes of the codg schemes. CS CS CS CS Fgure : P c, cosderg Scearo, for km/h, a spacg of slots ad all the CSs Fgure : P c for varyg s ad slot spacg (at km/h ad CS) uder Scearo At km/h, the effect of correlato o the lk level performace, for both terferece scearos, s such that the slot spacg has o mpact o the correlato probablty as P c does ot vary across the frame. The partcular slots selected for a multslot trasmsso would the have o mpact o the performace for both terferece scearos at km/h. CS CS CS CS Fgure : P c, cosderg Scearo, for km/h, a spacg of slots ad all the CSs Correlato. kmh-sce kmh-sce kmh-sce km/h-sce. Slot spacg Fgure 7: Effect of slot spacg o the correlato probablty (=db ad CS)
5 Correlato s stll preset. Therefore, the partcular slots selected for a multslot trasmsso would stll have a mpact o the lk level performace at hgh speeds uder both terferg scearos.. kmh-sce kmh-sce kmh-sce kmh-sce. Slot spacg Fgure 8: Effect of slot spacg o the correlato probablty (=db ad CS) At hgher speeds, the slot spacg ca play a mportat role. Fgures 7 ad 8 llustrate the effect of slot spacg o the correlato probablty at speeds of km/h ad km/h for the two terferece scearos. At km/h ad km/h, the slot spacg has a cosderable effect o the correlato probablty uder the terferece codtos modelled Scearo. However, ths effect s sgfcatly reduced uder Scearo. I Scearo ad cosderg the operatg codtos reported Fgure 7, the dfferece the correlato probablty betwee trasmttg two RLC blocks wth slot spacg equal to or slot spacg equal to s 7.%, at km/h. Ths dfferece s reduced to.% uder Scearo. I Fgure 8, ths dfferece s reduced from 7.9% to.8% for the same speed. Drect comparso of Fgures ad 9 reveal that, for the whole rage of mea s, the effect of slot spacg s eglgble whe each trasmttg slot s terfered by a dfferet terferer but ot whe they all are terfered by the same mutslot user. For the less robust codg scheme, CS, the effect of slot spacg o the correlato probablty P c s also sgfcatly reduced uder Scearo (Fgure ) compared to Scearo (Fgure ). However, ts effect s stll cosderable ad should be take to accout. At km/h ad for the majorty of the codg schemes, the effect of slot spacg o the multslot lk level performace could the be eglected uder the terferece codtos modelled Scearo. Except for the less robust codg scheme, the lk level performace slots of the same frame wll the be depedet of the partcular slots selected for a multslot trasmsso. The case cosderg a speed of km/h represets a dfferet scearo. From Fgures 7 ad 8, t ca be observed that the correlato probablty tally decreases wth the slot spacg but the begs to crease aga (whe a three slot spacg s reached), possbly due to the creasg mpact of ter-frame correlato. At km/h ad cosderg the operatg codtos reported Fgure 7, the dfferece the correlato probablty betwee trasmttg two RLC blocks wth slot spacg equal to or slot spacg equal to s reduced from % (Scearo ) to 9.7% (Scearo ). I Fgure 8, the same dfferece s reduced from 7.7% to 9.8%. Although the effect of slot spacg s also clearly reduced at km/h uder Scearo, t Fgure 9: P c for varyg s ad slot spacg (at km/h ad CS) uder Scearo Fgure : P c for varyg s ad slot spacg (at km/h ad CS) uder Scearo Fgure : P c for varyg s ad slot spacg (at km/h ad CS) uder Scearo. CONCLUSIONS Ths paper has expaded the aalyss  by comparg the GPRS multslot lk level performace
6 uder two terferece scearos that ca be assocated wth performace bouds. The tal terferece scearo descrbed , whch cosders a sgle strog multslot terferer, has therefore bee augmeted to accommodate multple sgle slot terferers as part of a ew terferece scearo. The results demostrate that the correlato probablty s affected by the terferece patter. Also, the effect of slot spacg o the GPRS multslot lk level performace s flueced by the terferece patter for medum ad hgh speeds. The results also dcate that whe each slot of the multslot trasmtter s terfered by a dfferet user, the effect that slot spacg has o the correlato probablty, for medum speeds, ca be elmated. I ths case, the multslot lk level performace across slots of the same frame s depedet of the resources allocated. Eve though ths effect was also clearly reduce for hgh speeds ts fluece caot be eglected.  J. Pos ad J. Dulop, I-Servce Lk Qualty Estmato for Lk Adaptato Algorthms, Appled to GSM, Proceedgs of the IEEE Iteratoal Coferece o Uversal Persoal Commucatos ICUPC'98, pp 9-7, October 998, Freze, Italy.  ETSI, GSM.; Rado Trasmsso ad recepto, verso 7.., Release 998.  B. Hagerma, O Dowlk Co-Chael Iterferece Cellular Rado Systems, Nordc Rado Symposum, pp -, 99, Salsjobade, Swede. REFERENCES  ETSI, GSM.; Overall descrpto of the GPRS rado terface. Stage, verso.., Release 997.  J Gozalvez ad J Dulop, Hgh-Speed Smulato of the GPRS Lk Layer, Proceedgs of the IEEE Iteratoal Symposum o Persoal, Idoor ad Moble Rado Commucatos, PIMRC, pp , September 8-, Lodo, UK.  P. Taaghol, R. Tafazoll ad B.G. Evas, A Ar Iterface Soluto for Mult-Rate Geeral Packet Rado Servce for GSM/DCS, Proceedgs of the IEEE Vehcular Techology Coferece, VTC97, pp -7, May 997, Phoex, USA.  A. Furuskär, M. Frodgh, H. Olofsso ad J. Sköld, System Performace of EDGE, a Proposal for Ehaced Data Rates Exstg Dgtal Cellular Systems, Proceedgs of the IEEE Vehcular Techology Coferece, VTC98, pp 8-89, May 998, Ottawa, Caada.  M. Flamet ad M. Ubehau, Frequecy Reuse ad Codg for GPRS Mult-Slot Operato, Proceedgs of the Iteratoal Zurch Semar o Dgtal Commucatos, pp 7-, February, Zurch, Swtzerlad.  J Gozalvez ad J Dulop, O the Effect of Correlato Multslot Lk Layer Aalyss for GPRS, Proceedgs of the IEEE Vehcular Techology Coferece, VTC-Fall, pp -, September, Bosto, USA.  ETSI, GSM.; Chael codg, verso 7.., Release 98.  ETSI, GSM.; Multplexg ad multple access o the rado path, verso 7.., Release 998.  T.Rappaport, Wreless Commucatos. Prcples & practce. IEEE press. 99.  J Dulop, J Pos, J Gozalvez, ad P. Atherto, A Real-tme GSM Lk Adaptato Hardware Demostrator, Proceedgs of the IEEE Vehcular Techology Coferece, VTC-Sprg, pp 9-9, May -8, Tokyo, Japa.