No Need for Signaling: Investigation of Capacity and Quality of Service for Multi Code CDMA Systems using the WBE++Approach

Transcription

1 No Need for Signaling: Invetigation of Capacity and Quality of Service for Multi Code CDMA Sytem uing the WE++Approach Vladilav V. Kravčenko and Holger oche and Frank H.P. Fitzek and Adam Woliz Abtract Net generation mobile communication ytem will not be voice dominated anymore. A large et of application with different quality of ervice requirement will run on top of the wirele end ytem. The quality of ervice upport for different type of ervice over the wirele link poe formidable challenge to the protocol deign. To guarantee a fat return of invetment for the network provider, the capacity per wirele cell ha to be maimized. In thi paper we will introduce a new approach for deigning preading equence for ynchronou CDMA ytem, that achieve high capacity and avoid overwhelming ignaling. y mean of imulation we can how that our approach lead to a ignificant capacity improvement. Moreover, higher PSfrag quality replacement of ervice value can be achieved uing a recently introduced protocol. Keyword: it Error Calculation; Signaling Overhead; Simultaneou MAC Packet Tranmiion; Spreading Sequence; Synchronou Multi Code CDMA; TCP/IP performance. over time. In other word one could think of a queue that i filled with packet and thee packet are tranmitted over a wirele link with a given capacity correponding to. According to the ARQ trategy, corrupted packet have to be retranmitted, while tored packet have to wait until the retranmiion proce of the corrupted packet i ucceful. Thi trategy lead to a variable delay of the higher protocol packet ( one Internet (IP) packet), which i undeirable for multimedia ervice. Therefore we came up with an idea of multiple erver with PSfrag replacement $ %$ %$ I. INTRODUCTION y introducing mobility, wirele communication ytem have undergone a tremendou evolution in the lat decade. The main difference between future wirele communication ytem and omnipreent ytem i that the former one are not voice dominated anymore. Future G and beyond network provider are intereted in a ufficient return on their invetment, which can be realized by larger number of cutomer (that can be billed per wirele cell) offering a large et of application. Introducing a wide range of ervice with heterogenou quality of ervice (QoS) requirement [FKW 02] in the wirele format poe formidable challenge to wirele protocol. The main problem i the characteritic of the wirele link, which (in contrat to a wired link) i highly error prone and uffer by long outage period [ZR9]. Even the ue of error-correction cheme uch a automatic repeat requet (ARQ) will not overcome the problem of the unreliable wirele link a given in the eample of Figure 1. The figure how a implified ARQ tranmiion cheme. The input to the queue i an arrival proce at a rate and the queue i erved at rate. After the erver a deciion i made a to whether thi ervice wa ucceful or not. In cae of an unucceful tranmiion, the ervice ha to be repeated and thi happen with probability, which varie V. Kravčenko and H. oche are with the Heinrich-Hertz Intitute für Nachrichtentechnik, roadband Mobile Communication Network, Einteinufer, 108 erlin, Germany (e mail: kravcenko, F. Fitzek and A. Woliz are with the Telecommunication Network (TKN) group, Dept. of Electrical Eng., Technical Univerity erlin, Sekr. FT 2, Einteinufer 2, 108 erlin, Germany, (e mail: fitzek, Fig. 1. Single erver with ervice rate "! and arrival rate #. Fig. 2. "& and arrival rate #. Multi erver with ervice rate individual ervice rate (' a given in Figure 2. In thi cae, even with )'+*, no buffer overflow will occur if we could only activate another erver every time the ervice failed before. In contrat to the eample given in Figure 1, the delay and jitter of the tored packet with the approach of multiple erver i bounded. Unfortunately, in real ytem, the number of erver correponding to channel i limited uing orthogonal reource uch a in TDMA, FDMA, or CDMA [Rap96] with orthogonal preading equence. Limited reource have to be aigned and ignaled by a centralized entity among tranmitter to avoid colliion. For cellular ytem the aignment can be eaily done within the down link. ut for the up link the aignment could lead to an overwhelming amount of ignaling. On the other hand, there i the poibility to ue peudo noie (PN) equence in combination with CDMA for the up link. PN equence are non orthogonal reource. In thi cae there would be a high number of reource which could be aigned in a ditributed fahion among tranmitter and therefore PN equence doe not need overwhelming ignaling. In contrat to orthogonal reource, uing PN equence the erver have an impact on each other and will lead to a performance degradation of each tranmitter when more erver are active. Regarding Figure 2, the bundling of three erver with rate ' each will lead to an overall erving rate maller than,.-/, becaue of the interference. Each active erver ha an impact on the performance of the adjacent erver. PN equence can be ued in aynchronou a well a in ynchronou CDMA ytem. In thi paper we introduce a new approach for deigning

2 Œ * m 0 0 X preading equence for ynchronou multi code CDMA ytem. The goal i to deign preading equence with no need for ignaling uch a the PN equence. Our approach i compared with a PN and a full knowledge (bet cae) approach. Thi paper i organized a follow. In the net ection we introduce our ytem with the general aumption. Furthermore, we give interference calculation for ynchronou CDMA ytem uing PN or Welch ound Equality (WE) equence. In Section III we introduce the problem of ignaling in CD- MA ytem and introduce our new approach for the deign of preading equence. Afterward we review the SMPT technique in Section IV. The performance evaluation and the related cenario i preent in Section V. We ummarize our work and give an outlook on our future work. A. General aumption II. SYSTEM DESCRIPTION In thi ection we will decribe and motivate all the aumption for the invetigated wirele communication ytem. We conider a ynchronou CDMA (S CDMA) ytem with one bae tation only. Wirele terminal (WT) that are in the coverage of the bae tation are able to convey data toward it. We aume multi code CDMA capabilitie at the wirele terminal (WT) and the bae tation. Furthermore, perfect power control i applied. In cae a WT i tranmitting packet over multiple CDMA channel, we aume that either all or none of the packet are ucceful received. Thi aumption i jutified in [Fit02]. Within thi S CDMA ytem all the ignal are tranmitted and received chip and bit ynchronou. For our S CDMA ytem, we aume that all the ignal are tranmitted aligned in carrier phae and without delay. We will analyze the S CDMA ytem uing modulation like binary phae hift keying (PSK), becaue in uch ytem a correctly computation of bit error probability (EP) i poible [Ver98]. Otherwie we have to ue uitably approimation for EP evaluation, which can ditort computation reult. Some approimation approache have been invetigated and dicued in [V.V01]. We uppoe a channel with path lo depending on ditance between bae tation and wirele terminal plu additive white Gauian noie (AWGN) are preented. Thi i o called AWGN channel model that i often ued for modeling of CDMA ytem. The hape of tranmiion pule i aumed to be rectangular. The binary preading equence for the k th channel i given by 0 12 * with 1 2 =<(>*+? :; (1) and equence length. The conidered S CDMA ytem ue o called unimodular equence, which have the following propertie. The abolute value of thee equence component are equal 1 2 <(8@A* C D. Hence the unimodular equence have power equal to 1. Thi fact will implify our analyi part in thi work. The perfect power control enure that all ubcriber will be received with the ame power. We denote correponding magnitude by E. Furthermore we aume that the receiver conit matched filter to etract information ignal. The output of the F th matched filter correponding to the F th channel can be written a G 2 HIJ*KEL-9M 2 IHI)NOEQP M ' IHI-8U ' 2 NV< 2 IHIJ (2) Y *KZ\[]@ ^`_a =bc8@ X%d. The correlation wherea the power of noie i W(X reflect the interference from other channel and i defined a Ue' 2 * 1 2 ; - 1 '. The binary data ymbol from the F th channel correponding to the time point H are denoted by M 2 Hf. We note that in our ytem with binary preading equence the F th matched filter ue the ame equence a the F th tranmitter. We will analyze our ytem only in baeband becaue both modulation at the tranmitter and demodulation at the receiver have not bearing on obtained reult.. Interference calculation for ynchronou CDMA ytem 1) PN equence: At firt we conider the propoed S CDMA ytem uing o called peudo noie (PN) equence a preading equence g. Generally PN equence are defined a binary random equence which component value occur with the ame probability, i.e hjik 1 2 =<(J*LN hjil 1 2 =<(A*nm J*. We note that the tranmiion performance i only dependent on the autocorrelation and cro correlation value for the PN equence. The correlation have an direct impact on the inter ymbol interference (ISI) and the multiple acce interference (MAI). The autocorrelation property doe not play a role in our S CDMA model, ince thi model ha been aumed a a ynchronou ytem. Due to the PSK detection by Mo 2 IHIJ*qpr]^) G 2 Hf the bit error probability (EP) can be calculated by the following equation [Ver98]: v/wy where PIz }98\P z $ TR t [ Mo 2 Hf\u z~89 P z ƒ *KM 2 IHI9@ g d * E W Y N P M%' Hf.* E W Y Ue' 2Iˆ () denote the Gauian function. Equation take all poible combination of the information ymbol M 2 Š / into account (for any WT (Fƒ* 98Ž ) under the aumption of equal occurrence probability of N and m ). However, thi manner of computation i not practical for a cloed analyi, becaue there are many part with Q function in Equation. Hence the numerical computation take long time and cannot be done for large Ž. The imulation have hown that the calculation of the Equation i poible only for Ž,. That i the motivation to reduce the compleity of thi Equation by correponding approimation. The mot known approimation have been obtained by Viterbi. Thi i baed on the lower bound of ignal to interference ratio (SNIR) with Gauian approimation for interference: t a Mo 2 *šm u 2 *K œjž Ž m N E X Ÿ " (4)

3 v ± D w œ N 2) Optimum Sequence: Now we will define the optimum criteria according to performance characteritic value like EP. Due to the fact that the EP i dependent on preading equence, the optimum deigned equence are given a the olution of the following problem tatement S 8ª «` ` V ) :±³² µ / = cļ¹ º µ =» ¼¾½ º ÀÂÁ =ÃÅÄ ÆÈÇ`Ä É.Ä Ê. (Ë () There are more than one olution to olve thi problem. In [V.V01] we have preented an approach uing a olution of the equivalent problem. A a reult a o-called Welch ound Equality WE equence [VA99] have been obtained, i.e. thee equence meet the optimum criteria given in Equation. We note thee equence are only optimum in cae of EP minimization, not with repect to other performance characteritic value like packet error probability or total throughput. The EP equation in the optimum WE cae ha been obtained in [V.V01] a tìîí`ï Mo 2 *qm u 2 *K œ ž Ž m N E X Ÿ (6) Comparing with Equation 4 we ee that the WE equence lead to much maller EP than in PN cae. Thi fact ha been hown in Figure (comparing PN with WE) auming the 6v ytem with * and ŽÐ*. Clearly the CDMA ytem uing WE equence have better performance property with repect to EP, if the number of channel i not larger than the preading gain becaue equence are orthogonal in thi cae. For more uer the EP value in WE cae converge to the EP value in PN cae. However, the gain of uing of WE equence i epecially large for mall difference between and Ž. There are three kind of the WE equence: binary, real and comple-valued. It wa hown in [VA99] that real and complevalued WE equence eit for any poible combination of preading gain and number of WT. We note, only binary equence can be applied to ytem uing PSK modulation. The binary WE equence can be contructed uing Hadamard matri. We note that Hadamard matrice with dimenion eit only if remainder after diviion by 4 i zero (ÑÓÒ9ÔÕ Öe *Ø ). Hence we can generate binary WE equence only for cae where the number of channel i diviible by four. We note that the contructing of binary WE equence for every cenario with ŽÚÙ i in general an open problem. We will decribe the contructing procedure under the condition ŽÛÙ and ÑÓÒ6ÔÕ ŽÜ:Öel*L in the following tep. At firt we have to generate Hadamard matri of order K. From thi matri we have to take out N column and put them into auiliary matri H. The row of the matri H preent K binary WE equence. We note that there are eactly IŽÝm identical equence in WE equence et generated uing the procedure decribed above. We can conclude that there are eactly ² w ½ poible WE equence et. In contrat to binary cae the compleed valued WE equence can be generated for every cenario ŽÚÙ [V.V01]. III. SIGNALING IN CDMA SYSTEMS A. General problem tatement Here and hereafter, we conider only ignaling ued for equence allocation or equence update. We note that in CDMA ytem with PN equence, ignaling between the bae tation and the wirele terminal i needed only at the beginning of the communication eion. In CDMA ytem with WE equence, the WE property will be lot in cae the number of active CDMA channel i changed. Therefore ignaling i need everytime the number of active channel i changed. The main quetion remain whether change in the number of active channel will affect the performance in term of EP.. WE++ approach Let wirele terminal be allocated baed on the condition for uage of WE equence. We have already hown that thi i the optimum cae. It i not poible to etend the number of channel without violating the aumption of required performance. Otherwie one or more WT can leave the cell looing the WE condition. We define the WE++ approach in cae, if remaining WT keep the equence which would be optimum in the WE cae. The main quetion i here whether the performance become not wore than in optimum cae. In the following we will obtain EP approimation depending not only on preading gain and number of channel in optimum cae Ž, but alo on number of outgoing channel Þ. At firt, we conider the EP equation coniting of interference and noie part µ J² >¹ º µ 9½ ºVß P &áà â»ã ä8 ÓÄä &kå» æ ç Æ Çæ à æ.è ˆ Ë () We aume a CDMA ytem uing comple valued WE equence, becaue the ummarized interference can be eactly calculated in thi cae a X * Ž m (8) In general, the value of the particular interference are not 12 1 ' X *ní u over î)f, Ô * 898ŽïÔðu *nf. However, the epected value of the ingle interference can be obtained by ZÜ[]@ìé p _ :p:ñ * w ò 'SR T X Žóm * wd m Ž m (9) If Þ wirele terminal are witched off, then the interference part will be Þ time ZÜ[ì@ìé pâ_è:p d maller. According to the lat tatement, the EP can be approimately calculated by tâõ\ Mo 2 *šm u 2 *q œyž Žóm IŽóm}ÞÕm Žóm E X`Ÿ (10) where ÞKö Žøm. We note that Equation 10 doe not give the EP value but only it approimation. /

4 v C. Methodology for EP meaurement We have already mentioned in Section II-.2 that everal optimum equence et are avaliable. We have performed imulation to how that the EP value are dependent on the choen equence et uing the following meaurement procedure. For the defined cenario (given by ŽÜ ) we calculate the EP value correponding to the choen equence et for every WT uing the Equation. Thi procedure i repeated for different equence et. We have limited the number of et to Ž becaue there are too many poible equence et. We conider uch EP meaurement in an eample, given by the following cenario: WE cae with *Qù, Žð* v. The reult of EP computation for eight different equence et are preented in Figure, where gray color hading repreent the correponding EP value. The gray color are caled from white (bet EP in the given cae) to black (wort EP in the given cae). We can conclude, that there are only two EP value which are imilar ditrubuted acro the equence et. According to theoretical olution in Equation 6, all the EP value mut be equal and independent on choen equence et. However, thi tatement hold only for comple valued equence, becaue each equence i unique. In a binary WE cae, a we have already eplained in Section II-.2, there are IŽLm identical equence that lead to different EP value among WT. Let u conider the cae where one WT leave the ervice area or i witched off. We aume that remaining WT ue old equence (WE++ cae). The imulation reult that EP value ditribution acro WT will be dratically changed a how Figure 4. Fig.. et Channel Set of Sequence Channel EP depending on equence Fig. 4. EP depending on equence et. Here: ú8ûcü WT i witched off The net imulation give the anwer to the quetion whether in WE++ cae the performance doe not get wore than in optimal WE cae. We have done it auming following cenario: * 6v, Ž 'ký%þ *. We have calculated EP value for each combination, Ž (Ž *L ô88:žÿ'ký%þ ) according to () a average EP whereby all the poible equence et have been taken into account (ee Figure ). ecaue of the large number of equence et and the high compleity of Equation (), the numerical computation take a long time which rapidly increae for large Ž. The imulation have hown that the calculation of Equation () i poible only for ŽÐ,. We note that the wort average EP value in the WE++ cae i maller than the wort average EP value in WE cae. Thi fact i confirmed by the approimation given in Equati- Set of Sequence bit error probability e number of active channel Fig.. EP depending on number of active channel: magenta dahed dotted dotted line PN cae, green dah dotted line approimation (Equation 10) for WE++ cae, blue dahed line true EP in WE++ cae (Equation ), red olid line WE cae on (10). Otherwie we can conclude that all the EP curve converge in the marginal cae Ž. IV. THE SMPT APPROACH We now briefly review the SMPT approach (referring the intereted reader to [Fit02] for a more detailed dicuion). For illutration purpoe, we aume the tranmiion of MAC packet repreenting one higher protocol egment (e.g. one IP packet). The reulting tranmiion over multiple CDMA channel uing SMPT i given in Figure 6. In contrat to the equential tranmiion, SMPT tart uing additional CDMA channel after detecting a good channel tate at time lot. While in the equential cae tored packet are uffering by the retranmiion proce, packet number, uing SMPT i not aware of any retranmiion proce. Thu, SMPT i able to bound the delay for higher protocol egment. Note, that packet are tranmitted uccefully in time lot. The equential approach could achieve thi only by an error free wirele link. In cae an error occur while building up the ramp, the number of channel i reet immediately to one. Thi behavior i imilar to the TCP fairne approach and i implemented to achieve ytem tability. V. PERFORMANCE EVALUATION For the performance evaluation for the PN, WE++, and the WE approach we ue the treaming of video over the TCP protocol, introduced in [FSW 02]. We note that the amount of ignaling required i not taken into account for the WE approach. Furthermore the impact of error prone ignaling i not invetigated. Therefore reult for the WE approach can be een a bet cae reult. In the imulation each WT randomly elect one out of v video equence, which are obtained from [FR01] including port, movie, and new video equence. Alo, each WT elect an independent random tarting phae into the elected trace to enure the tatitical independence of the tranported video tream. The WT commence the video treaming by filling the receiver ide play out buffer to a pre pecified offet value ƒ* 1 Ò. The receiver ide application tart to play out the video once the play out buffer reache the

5 v offet value. Under normal circumtance, for every frame period (which i typically an integer multiple of Öe mec for H.26 encoded video [FR01]), the receiver remove a frame from the play out buffer, decode it, and diplay it. If at any of thee epoch there i no complete video frame in the play out buffer, the receiver eperience playback tarvation, which we refer to a buffer underrun. When a buffer underrun occur, the receiver temporarily upend the play out of the video. The receiver wait until the play out buffer i filled to the offet value, and then reume the play out of the video. The wirele channel i modeled a a two tate Markov chain with a good and a bad tate, where no communication i poible in the bad tate. The EP in the good tate depend on the total number of active channel. A in [TF00] a CH FEC code i ued and give the packet error probability a a function of the number of active channel. At the data link layer we tudy two tranmiion approache: (1) equential tranmiion baed on a SC CDMA ytem and (2) SMPT which i baed on a MC CDMA ytem. In our performance evaluation we tudy primarily the buffer underrun rate and the average time between buffer under run (in minute). For higher value of and lower value of the ytem perform better. We dicu the ytem behavior for Ö kbp rate controlled H.26 encoded video. To accommodate the overhead of the upper protocol layer and ome retranmiion, we choe a bit rate of ù kbp for the phyical layer. (Note that the over proviioning of the wirele channel allow even the equential tranmiion cheme to perform retranmiion.) The link layer error model i baed on a two tate Markov chain, where one tate reflect the bit error dependent on the actual number of active channel and the other tate reflect the ituation where no communication i poible (burtine of the wirele channel). The different error model between wirele terminal and bae tation act uncorrelated. TALE I RESULTS FOR PN, WE++, AND WE APPROACH IN TERMS OF CAPACITY AND QOS ( ) WITH A SPREADING GAIN OF Spreading Sequential SMPT Sequence Capacity Capacity Type min WTS min WT PN WE WE In Table I the reult for PN, WE++, and WE approach in term of capacity and QoS with a preading gain of v i given. The QoS i repreented by. Generally poken, for thi cenario in contrat to the equential approach, SMPT offer a higher QoS performance acrificing a part of the ytem capacity. If we focu on the WE++ cae the equential tranmiion can upport WT, while with SMPT only WT can be upported. ut the QoS i one magnitude higher than the with the equential cae. A already hown in [FSW 02], SMPT perform better with higher preading gain (reulting in a higher multipleing effect). Thu for higher preading gain not only the QoS parameter but alo the capacity might be better with SMPT. VI. CONCLUSION AND OUTLOOK In concluion we can ay that the WE++ approach offer higher capacity than PN equence without increaing the amount of ignaling. Even if higher capacity value can be achieved with the WE approach, we note that the ignaling overhead for WE wa not taken into account. In further tudie we will invetigate the performance of SMPT with higher preading gain. VII. ACKNOWLEDGMENT Thi work wa done by the grant of the MF (the German cience minitry) within the project TraniNet. We want to thank Robill Supatrio working at acticom for hi outtanding work in imulating the SMPT performance. The imulation environment wa implemented while hi tay at the Technical Univerity of erlin doing hi diploma thei. We thank Patrick Seeling for hi valuable comment. Number of Ued Channel 1 2 Error-free Packet AD STATE Probing Erroneou Packet Fig. 6. Slow Healing SMPT mechanim. Slow Healing 4 REFERENCES GOOD STATE time [Fit02] Frank H.P. Fitzek. Quality of Service Support in Wirele Multi Code CDMA Sytem. PhD thei, Technical Univerity of erlin, Telecommunication Network Group, June [FKW 02] Frank H.P. Fitzek, A. Köpel, Adam Woliz, M. Reilein, and M. A. Krihnam. Providing Application Level QoS in G/4G Wirele Sytem: A Comprehenive Framework aed on Multi Rate CDMA. IEEE Wirele Communication, 9(2):42 4, April Special iue on 4G Technologie and Application. [FR01] F. Fitzek and M. Reilein. MPEG 4 and H.26 Video Trace for Network Performance Evaluation. IEEE Network, 1(6):40 4, November/December video trace available at [FSW 02] F.H.P. Fitzek, R. Supatrio, A. Woliz, M. Krihnam, and M. Reilein. Invetigation of Capacity and QoS for Streaming Video Application over TCP in CDMA baed Network. In Proceeding of the G wirele, May [Rap96] Theodore S. Rappaport. Wirele Communication, volume 1. Prentice Hall, jan. [TF00] A. S. Toun and W.-C. Feng. On Improving Quality of Video for H.26 over Wirele CDMA Network. In Proc. of IEEE Wirele Communication and Networking Conference (WCNC), September [VA99] Pramod Viwanath and Venkat Anantharam. Optimal Sequence, Power Control, and Uer Capacity of Synchronou CDMA Sytem with Linear MMSE Multiuer Receiver. IEEE Tranaction on Information Theory, 4(6): , September [Ver98] Sergio Verdu. Multiuer Detection. Cambridge Univerity Pre, [V.V01] V.V. Kravčenko, H. oche, E.Jugl. Model for EP Etimation of CDMA-Sytem and Sequence Deign. In ûü Aachen Sympoium on Signal Theory, September [ZR9] M. Zorzi and R. R. Rao. The effect of correlated error on the performance of TCP. IEEE Communication Letter, 1(September):12 129, 199.