An FDD ideband CDMA MAC Protoco for ireess Mutimedia Networks Xudong ang Broadband and ireess Networking (BN) Lab Schoo of Eectrica and Computer Engineering Georgia Institute of Technoogy Atanta, GA 3332 Emai: wxudong@ece.gatech.edu Abstract A medium access contro (MAC) protoco is deveoped for wireess mutimedia networks based on frequency division dupex (FDD) wideband code division mutipe access (CDMA). In this protoco, the received power eves of simutaneousy transmitting users are controed by a minimum-power aocation agorithm such that the heterogeneous bit error rates (BERs) of mutimedia traffic are guaranteed. ith minimumpower aocation, a mutimedia wideband CDMA generaized processor sharing (GPS) scheduing scheme is proposed. It provides fair queueing to mutimedia traffic with different QoS constraints. It aso takes into account the imited number of code channes for each user and the variabe system capacity due to interference experienced by users in a CDMA network. The admission of rea-time connections is determined by a new effective bandwidth connection admission contro (CAC) agorithm in which the minimum-power aocation is aso considered. Simuation resuts show that the new MAC protoco guarantees QoS requirements of both rea-time and non-rea-time traffic in an FDD wideband CDMA network. I. Introduction To date, mutipe radio transmission technoogies have been proposed for the next generation wireess networks. Some of them are sti under investigation, whie others have been seected as a candidate technoogy for the next generation wireess networks. A typica exampe is wideband codedivision mutipe-access (CDMA). It has been chosen as the basic access technoogy for UMTS/IMT2 []. Compared to the narrow-band CDMA, wideband CDMA can support services with much higher rate. It is aso fexibe to deiver mutimedia traffic. However, in order to fuy make use of such improvements, a new medium access contro protoco (MAC) is required to efficienty manage packet access in wideband CDMA wireess networks. ideband CDMA can be categorized into pure wideband CDMA and wideband timedivision (TD) CDMA. Pure wideband CDMA uses frequency division dupex (FDD) to organize the upink and downink transmissions, whie wideband TD-CDMA uses time division dupex (TDD). TDD mode is we-suited for indoor environments with high traffic density and appications with highy asymmetric traffic [2], whie FDD mode is advantageous for appications in pubic macro- and micro-ce environments. In This work is supported by NSF under grant CCR-99-88532 and by the State of Georgia Yamacraw Project (E2-5). terms of resource aocation, more differences exist between FDD mode pure wideband CDMA and wideband TD-CDMA. In wideband TD-CDMA, resource units incude both time sots in a frame and codes in a time sot due to sotted frame structure. Muti-code (MC) operation and muti-sot operation are normay used to aocate resources [3]. However, FDD mode wideband CDMA does not have the option of mutisot operation because the resource units ony incude codes in a frame. In order to enhance the fexibiity of resource aocation in FDD mode wideband CDMA, orthogona-variabespreading-factor (OVSF) is used [3]. A mobie termina can have mutipe codes and the OVSF of each code can be variabe. Thus, both MC operation and OVSF operation are used in FDD mode wideband CDMA. This paper is focused on the MAC protoco for FDD wideband CDMA wireess mutimedia networks. Severa MAC protocos have been proposed for FDD wideband CDMA networks. In [4], a proposa for an RLC/MAC protoco for wideband CDMA is presented. This proposa does not consider how to aocate resources to different services. An upink MC CDMA system architecture is proposed to support heterogeneous traffic with diverse QoS requirements [5]. The received power eve of each code channe is controed by a power aocation agorithm so that BERs of a services are guaranteed. Moreover, the scheduing scheme for nonrea-time traffic is based on first-in first-out (FIFO) queueing and round-robin queueing. The average message transmission deay of non-rea-time traffic can be very arge when a ower priority is assigned. Another MAC protoco for FDD wideband CDMA is proposed in [6], where the power aocation agorithm and the token bucket traffic reguator are not designed interactivey. In addition, the power aocation agorithm does not consider both MC and OVSF operation. For UMTS/IMT-2 based on wideband CDMA, the performance of a mutipe access protoco for integration of variabe bit rate mutimedia traffic is anayzed in [7]. However, a PRMA-ike MAC protoco is assumed for wideband CDMA networks. The 3rd Generation Partnership Project (3GPP) standardization committee has aso reeased a specification on MAC protoco [8]. In this specification, the MAC architecture, channe structures, services, and MAC functions are defined. -783-7753-2/3/$7. (C) 23 IEEE IEEE INFOCOM 23
However, no specific access scheme is specified for the MAC protoco. Thus, the design of the MAC protoco for wideband CDMA is sti an open probem. Recenty, an access scheme is deveoped in [9] for the voice/data traffic transmissions over common packet channe (CPCH) in a wideband CDMA network. However, it does not incude an access scheme for dedicated channes (DCHs). In this paper, a new MAC protoco is proposed for mutimedia traffic transmission in upink DCHs of wideband CDMA wireess networks. Simiar but simper procedures can be appied to mutimedia traffic transmission in downink DCHs, thanks to the broadcast nature of downink. In the new MAC protoco, a minimum-power aocation agorithm is first derived to determine the minimum received power eve for each code channe when a target signa-to-interferencepus-noise-ratio (SINR) is given. In this agorithm, both MC and OVSF transmissions is considered. Aso, channe fading and noise are taken into account to estabish the reationship between the required BER of a service type and the SINR at the receiver. The minimum received power eves are assumed to be maintained through open- or cose-oop power contro agorithms [], []. How to design a power contro agorithm is out of the scope of this paper. Based on the minimumpower aocation agorithm, a wideband CDMA generaized processor sharing (GPS) scheduing scheme is proposed. This GPS scheduing scheme is nove and different from other GPS scheduing schemes [2], [3], because a the foowing features are considered: ) The system capacity considered in the scheduer varies with respect to the interference eve experienced by CDMA users. 2) Both MC-CDMA and VSF-CDMA are considered in the scheduing scheme. 3) Both power constraint of code channes and the imited number of code channes for each user are considered in the scheduing scheme. The GPS scheduing scheme provides fair queueing to mutimedia traffic with heterogeneous QoS requirements. It aso guarantees the diverse BER vaues of mutimedia traffic. In order to admit rea-time connections, a CAC scheme is needed. One of the chaenging issues of a CAC agorithm is that the bandwidth requirement of a connection cannot be determined at connection admission time due to the traffic burst. Thus, effective bandwidth-based CAC schemes are generay used to sove this probem. For a CDMA system, another chaenging issue is that the system capacity is interference-sensitive. Thus, the effective bandwidth-based CAC agorithm for FDD wideband CDMA networks must consider minimum-power aocation so that both maximum system capacity and BER guarantees are taken into account by the CAC agorithm. Such a CAC scheme improves the system performance by greaty reducing packet oss and packet deay of mutimedia traffic. The CAC agorithm in [4] does not consider minimum-power aocation. Athough the CAC agorithm proposed in [5] takes into account the minimumpower aocation, it does not incude the concept of effective bandwidth, and thus it is not feasibe for connections with bursty traffic. In addition, MC transmission is not considered in [5]. In this paper, simuations are aso carried out to evauate the performance of the MAC protoco. Resuts show that exceent performance is achieved by the MAC protoco. The paper is organized as foows. The overa MAC protoco is described in Section II. In Section III, the minimumpower aocation agorithm is derived for an FDD wideband CDMA network. Based on the minimum-power aocation agorithm, the mutimedia wideband CDMA GPS scheduing scheme is deveoped in Section IV, and the effectivebandwidth CAC scheme is derived in Section V. In Section VI, performance of the new MAC protoco is evauated through simuation. Comparisons between the new MAC protoco and the FDD mode CDMA MAC protoco in [5] is presented in Section VII. The concusion of this paper is drawn in section VIII. II. Protoco Description A. System Specifications In this paper, a FDD mode wideband CDMA system is considered. Due to the broadcast nature, downink transmission of mutimedia traffic can be supported more easiy than upink transmission. Thus, when a MAC protoco is designed, the focus is on the upink transmission. The foowing transport channes defined in the 3GPP specification [8] are used in the MAC protoco: Random access channe (RACH). This channe is used by mobie terminas to send contro packets, e.g., connection requests. Broadcast contro channe (BCCH). This channe is a point-to-mutipoint channe, which is used to convey system information from the base station to mobie terminas. For exampe, the feedback about resource aocation is transmitted in this channe. Dedicated channe (DCH). The DCH is point-to-point channe, which is used to transmit data from mobie terminas to the base station or vice versa. DCHs, RACH and BCCH are mutipexed in the code division. In DCHs of FDD mode wideband CDMA, both MC transmission and VSF transmission are used. A DCH can have variabe transmission rate depending on the spreading factor, and the basic transmission rate of the DCH corresponds to the maximum spreading factor used in this channe. Thus, a variabe-ength MAC packet is accommodated in a DCH. However, in order to simpify packet segmentation in the interface between ink access contro (LAC) ayer and the MAC ayer, a fixed-ength packet, caed radio ink contro (RLC) packet data unit (PDU) is defined [], [4]. The size of a fixed-ength RLC packet, denoted by r, reates to the basic transmission rate r b of a DCH according to r = r b t fr, where t fr is the frame ength. hen a packet is generated in the LAC ayer, it is segmented into mutipe RLC PDUs. -783-7753-2/3/$7. (C) 23 IEEE IEEE INFOCOM 23
Those RLC PDUs may be transmitted in one frame or severa frames, depending on avaiabe DCHs and transmission rates of these DCHs. Thus, the variabe ength of a MAC packet is represented by the variabe number of RLC PDUs transmitted during a frame. In order to reduce the compexity of a mobie termina, the number of DCHs that can be used by the mobie termina is imited. The imited number varies with service type supported in a mobie termina. For exampe, a mobie termina transmitting video traffic needs severa DCHs, whie a mobie termina transmitting voice traffic can be satisfied with one DCH. B. The MAC Protoco The MAC protoco supports both rea-time and non-rea time services: hen a mobie termina wants to support rea-time service, it needs to send a connection request in the RACH. hen this request is received at the base station, an effective-bandwidth CAC scheme, which is based on the minimum-power aocation agorithm as wi be derived in Section III, is used to check the admission of the connection request. If the answer is positive, the connection request is accepted and become ready to transmit rea-time traffic. However, how the packets of this connection are transmitted in each frame is determined by the wideband CDMA GPS scheduing scheme, which wi be derived in Section IV. hen a mobie termina wants to deiver non-rea-time service, no admission contro is used. henever packets become avaiabe in this termina, they are ready to be transmitted as ong as the resources are aocated by the base station. Packet transmissions in rea-time connections and a nonrea-time traffic fows foow the same procedure: ) In a mobie termina, when a packet in the network ayer is generated, its virtua arriva time and virtua departure time are determined by the wideband CDMA GPS scheduing scheme. Since this packet must be segmented into smaer RLC PDUs in the MAC ayer, a those RLC PDUs hod the same virtua arriva time and virtua departure time. It shoud be noted that the wideband CDMA GPS scheduer takes into account the interference-sensitive system capacity in the CDMA networks, because it is derived by considering minimumpower aocation. 2) The priority of packet transmission is determined according to the virtua departure time. Packets in a rea-time connections and non-rea-time traffic fows are schedued from the highest priority towards the owest one, unti system capacity is exhausted or no packet waits for scheduing. In the wideband CDMA GPS scheduing scheme, the imited number of code channes in a connection and the variabe rate of a code channe must be considered. The system capacity in a frame is determined by the minimum-power aocation agorithm through which the interference-sensitive CDMA system capacity is maximized. 3) The received power eves of RLC PDUs that are schedued for transmission are determined based on the minimum-power aocation agorithm. 4) After scheduing is finished in a frame, feedback information is sent back from the base station to mobie termina through the BCCH. It incudes the connection or traffic fow ID, the aocated number of code channes for each connection or traffic fow, the corresponding VSF of each code channe, the received power eves of those code channes, and the number of RLC PDUs to be transmitted in each code channe. 5) Feedback information is received by mobie terminas. The transmitted power eve of a DCH is determined based on the received power eve and estimated channe gain. Then, packets are transmitted in a DCH by using aocated VSF and desired transmitted power eve. 6) If errors are detected but cannot be corrected in non-reatime packets received at the base station, seective-repeat ARQ is used to re-transmit these erroneous packets. III. Minimum Power Aocation As described in Section II-B, both the effective-bandwidth based CAC scheme and the wideband CDMA GPS scheduing scheme are based on the minimum-power aocation agorithm. The objective of this agorithm is, given a number of code channes of different users with heterogeneous BER requirements, to find the minimum received power eve of each code channe such that the heterogeneous BER vaues of different users are satisfied. Compared to other reated work [6], [6], [7], the agorithm considers a the foowing features of FDD wideband CDMA: Mutipe service types with heterogeneous BER requirements are supported. Both MC operation and OVSF operation are taken into account when aocating resources. However, MC operation is not considered in [6], [6], whie OVSF operation is not incuded in [7] where ony one service type is supported. A. The Minimum-Power Aocation Agorithm To focus on the MAC protoco design, a singe ce in an FDD mode wideband CDMA wireess network is considered. There are K type of services supported in the ce, and that service type k requires SINR to be in order to have desired BER. The reationship between SINR and BER wi be discussed in Section III-B. Of service type k, it is assumed that there are N k users. The set of DCHs aocated to user n k is represented by a vector C n k =[C n k,,cn k M ], which must be chosen from an OVSF code tree. Such a OVSF tree has M eves of orthogona codes, and the SF of m-th eve is assumed to be G m = 256/2 m,m =, 2,,M. Thus, the transmission rate of the a DCH using a code at m-th eve is r m = /G m, where is the bandwidth of the wideband CDMA system. P n k =[P n k,,pn k] denotes the received M -783-7753-2/3/$7. (C) 23 IEEE IEEE INFOCOM 23
power eves that corresponds to DCHs of C n k. Thus, the overa transmission rate r nk and received power eve of user n k are M m= Cn k m G m and M m= Cn k m P n k m, respectivey. Considering a specific user n ζ with service type ζ, one of its DCHs at the ν-th eve of the OVSF code tree experiences interference I n ζ ν at the receiver of the base station. I n ζ ν consists of two components: one is the interference from DCHs of other users in the same system, denoted by I I, and the other is the noise, denoted by N o. Thus, the SINR of one of the ν-th eve DCHs, denoted by γ n ζ ν, can be described as γ n ζ ν /r ν ν = (I I + N o )/, () where ν {,,M}, ζ {,,K}, n ζ {,,N ζ }, and r ν is the transmission rate of a DCH at the ν-th eve. Interference I I is contributed by the power eves of DCHs of a users except those of user n ζ,so I I = K N k M k= n k = m= C n k m P n k m } {{ } power eves of a users M = C n ζ } {{ } power eves of user n ζ Combining () and (2) and considering γ n ζ ν ν G ν K Nk M k= n k = m= Cn k m P n k m M = Cn ζ yied. (2) P n γ ζ ζ, + N o (3) where G ν = /r ν. To minimize the power eves of each DCH, the equaity in (3) must hod. This yieds G ν K N k M M ν = C n k m P n k m C n ζ + N o. (4) k= n k = m= Note that (4) is satisfied for a ν {,,M} of user n ζ. Thus, for the -st eve DCHs of user n ζ, the right side of (4) is the same as that for the ν-th eve channe. Thus, ν G ν = G, (5) i.e., ν = G G ν. According to this equation, a power eves in (4) can be represented by the power eve or P n k. Defining Γ nk as M G m and Γ nζ as M and from (5), (4) becomes ( G m= G m C n k +Γ nζ ) = K N k k= n k = = = G G C n ζ, Γ nk P n k + N o. (6) It shoud be noted that (6) is aso satisfied for any user n k, i.e., the eft side of (6) can be ( G +Γ nk )P n k. Thus, i.e., P n k = = ( G +Γ nζ ) =( G +Γ nk )P n k, (7) G γ +Γ nζ ζ G γ +Γ nk k ( G. Putting this into (6) yieds N o +Γ nζ )( K k= Nk n k = Γ nk G ). (8) γ +Γ nk k According to the definition of Γ nk and r nk = M m= Cn k m G m, r Γ nk = G nk and Γ nk G =. Putting these resuts γ +Γ nk + k rn k back into (8), then = N o /G ( + rn ζ )( K k= Nk n k = + ). (9) (9) gives the minimum-required power eve of one of the - st eve DCHs of user n ζ. For DCHs on other eves of the OVSF code tree of user n ζ, this equation aso hods, i.e., m = N o /G m ( + rn ζ )( K k= Nk n k = + ), () as ong as an m-th eve channes is aocated to user n ζ. Since each DCH has power constraint, it is required that the received power eve of one of the m-th eve DCHs be ess than Pm max. Thus, K N k= o/g m Nk n k = + P max m ( m {,,M} and ζ {,,K}. Thus, K N k k= n k = + m=,,m ζ=,,k r nk max Define = max m=,,m ζ=,,k N o/g m Pm max( γ + rn ζ ζ γ + rn ζ ζ N o /G m P max m ( ) for a () + rn ζ ). () becomes ), K N k +, (2) k= n k = r nk which must be satisfied in order to have minimum-power aocation for each DCH and satisfy BER of each user in P a wideband CDMA system. Since m N o/g m is the SNR, so P max m N o/g m must be the maximum SNR that can be achieved by the receiver on the base station. For the base station receiver of current 5 MHz wideband CDMA, it is norma that SNR can be as high as 6dB. Thus, a very sma vaue of in (2) can guarantee that the power eve of each code channe does not exceed its maximum vaue. In (2), + can be viewed as the normaized transmission rate of user n k whose service type is k, transmission rate in a frame is r nk, and required SINR is. Thus, (2) means that the overa normaized transmission rates of a users in a frame cannot exceed, which is caed the normaized system capacity. B. Reationship between BER and SINR In a wireess network, the reationship between BER and SINR is determined by severa factors such as error contro agorithms, moduation schemes, and channe fading. In this paper, the moduation scheme is assumed to be binary phase shifting keying (BPSK), and channes are assumed to have Rayeigh fading. Error contro schemes are different for reatime and non-rea-time traffic. Rea-time traffic. At the receiver, convoutiona, RS, and CRC decoders are used sequentiay to correct the errors in packets. The BER and SINR reationship of rea-time -783-7753-2/3/$7. (C) 23 IEEE IEEE INFOCOM 23
TABLE I SINR Vaues and BER Requirements. Service Type BER SINR (db) 3 2.54 4 2.72 5 2.9 6 3.8 Data or Emai 2.75 traffic is derived using the same formuas for cass-i traffic in [5]. Non-rea-time traffic. Ony the convoutiona and RS decoders are used to correct errors in packets. The residua errors are detected by the CRC decoder. Seective ARQ is used for re-transmission of detected erroneous packets. The probabiity P r of packet re-transmission triggered by the CRC decoder is approximatey equa to the bit error probabiity at the output of the RS decoder [5]. Suppose the transmission rate of a non-rea-time mobie termina is r nrt and its aowed normaized transmission rate is R nrt. Thus, according to the definition of normaized transmission rate in Section III-A, r nrt =, ( R )γ nrt nrt where γ nrt is the required SINR of non-rea-time traffic. Thus, the throughput ρ of the mobie termina is ( P r )r nrt, i.e., ρ =( P r ). (3) ( R nrt )γ nrt P r increases as γ nrt decreases, so there is an optimum SINR γnrt that achieves maximum throughput of the mobie termina. Such an optimum SINR is determined by maximizing ( P r )/γ nrt. Suppose 2 convoutiona coding with d free = and (256, 24, 256) RS coding with error correction capabiity t =8are used, then the SINR vaues corresponding to the typica BERs of different services are isted in Tabe I. IV. GPS Scheduing Scheme for ideband CDMA A. The wideband CDMA GPS scheduing scheme hen packets in a frame are avaiabe for transmission, they need to be schedued according to their QoS and BER requirements. The scheduing scheme must satisfy severa constraints in order to achieve high performance in the FDD mode wideband CDMA system: Power/BER constraint. hen packets are transmitted in a wideband CDMA frame, minimum-power aocation must be used in order to achieve maximum capacity and guarantee BER requirements. Therefore, the constraint in (2) must be satisfied in the scheduing scheme. QoS Requirement. The scheduing scheme must support heterogeneous QoS requirements of mutimedia traffic. It needs to be fair and provide QoS guarantees to mutimedia traffic. Code channe constraint. Not a DCHs on the OVSF code tree can be used simutaneousy by a user. One reason is that the DCHs used simutaneousy must be orthogona. The other reason is that the number of DCHs avaiabe to a user is generay imited, which reduces the compexity of a mobie termina by reducing the transceiver units [8]. Since scrambing codes are userspecific [], code bocking that exists in the downink [8] does not occur in the upink of wideband CDMA systems. In order to take into account code channe constraint, the scheduing scheme must check if the permitted number of DCHs in a mobie termina is exceeded. Having these constraints in mind, a new scheduing scheme, caed wideband CDMA GPS scheduing, is proposed in this section. The wideband CDMA GPS scheduing scheme is based on GPS. The significant feature of GPS is that it treats different traffic types differenty according to their QoS requirements [2]. GPS aso assumes that mutipe traffic fows with variabe traffic rates can be served simutaneousy. This was considered as a drawback of GPS because the cassica packet-based systems are TDMA-based and thus do not permit parae packet transmissions. However, in a CDMA system, it is natura to simutaneousy serve mutipe traffic fows with variabe transmission rates. Thus, GPS heps to design a high performance scheduing scheme for CDMA systems. In [3], a GPS-based scheduing scheme is proposed for hybrid CDMA/TDMA systems. The scheme in [3] assumes that ony VSF transmission is used. The scheduing scheme under this assumption is easy to derive, because the power/ber constraint has a simper form and no code channe constraint exists. However, in FDD mode wideband CDMA systems, both MC and VSF transmissions need to be considered. Thus, the wideband CDMA GPS scheduing scheme designed here is much more genera and fexibe to support mutimedia traffic. The FDD mode wideband CDMA does not have a TDMA frame. Mobie terminas with different BER requirements do not have separate time sots to accommodate their packets. The ack of such a fexibiity resuts that the BER-scheduing schemes used in [9] cannot be appied in FDD mode wideband CDMA networks. The wideband CDMA GPS scheduing scheme is operated as foows: ) Determine the virtua finishing time of MAC packets. hen a LAC PDU of a mobie termina arrives, the base station shoud be informed of the arriva time of this packet. Then, the virtua finishing time of this LAC PDU is determined, which wi be derived in Section IV- B. RLC PDUs beonging to the same LAC PDU have the same virtua finishing time. 2) Serve RLC PDUs according to virtua finishing times. In a wideband CDMA frame, the RLC PDUs from different mobie terminas are serviced from the highest priority towards the owest one. The smaer the virtua finishing time of RLC PDUs, the higher the priority. In addition, two constraints need to be considered: -783-7753-2/3/$7. (C) 23 IEEE IEEE INFOCOM 23
Check system capacity. The power constraint in (2) must be checked. As ong as this constraint is satisfied, packets are schedued unti a RLC PDUs are serviced or no system capacity is eft. Check code channe constraint. Since each mobie termina has a imited number of DCHs, an agorithm is necessary to seect appropriate DCHs and transmit as many packets as possibe by using these DCHs. Such an agorithm wi be discussed in Section IV-C. 3) Cacuate received power eves. After the code channes has been assigned to each mobie termina, the received power eve of each assigned code channe is cacuated according to (). B. Determining Virtua Finishing Time As in Section III-A, a ce in a FDD mode wideband CDMA network can be considered as a queueing system with capacity. Denote Ω i (t,t 2 ) as the amount of traffic of session i served in the time interva (t,t 2 ], and ω i (t) as the work rate of a session, i.e., ω i (t) = d dt Ω i(,t). φ i is a positive number associated with session i. According to the definition of GPS and its work conserving characteristics [2], the GPS for a ce of FDD mode wideband CDMA network must have the foowing two features: The work rate of each backogged session i is guaranteed to be φ i Ω i = j A φ ( ), (4) j where A is the set of a the accepted sessions in the system. Fair resource sharing is guaranteed, i.e., for any two backogged sessions i and j, ωi(t) ω = φi j(t) φ j. Thus, φ i ω i (t) = j β(t) φ ( ), i β(t), (5) j where β(t) is the set of a backogged sessions. For the m-th LAC PDU of session n k, assume it arrives at a m n k, starts service at Sn m k, and finishes service at d m n k. According to the definition of normaized transmission rate in Section III-A, ω nk (t) = during (S + n m k ]. Com- bining this resut with (5) when i = n k, then + = φ n k j β(t) φj ( ), i.e., r nk =, j β(t) φj φ nk ( ) = φ nk ( ) j β(t) f jφ j, (6) where n k β(t) and f j =if j n k ; otherwise, f j =. For any given busy period (t,t 2 ] in GPS, the virtua time v(t) of session i is defined as [2] v(t 2 ) v(t )= Ω i(t 2,t ), i β(t,t 2 ), (7) Ω i where v()=. Thus, the virtua time of the m-th LAC PDU of session n k is v(d m n k ) v(s m n k )= Ω n k (S m n k ) Ω nk (8) Considering that ω nk (t) = + during (S m n k ], Ω i (S m n k ) can be described as Ω i (S m n k ) = d m n k S m n k ω nk (t)dt, = dm n k Sn m k +, r nk = L m n k r nk +, (9) where L m n k = r nk (d m n k Sn m k ) is the ength of the m-th LAC PDU of session n k. Putting (6) into (9) yieds Thus, (8) becomes Ω i (S m n k )= v(d m n k ) v(s m n k )= L m n k φj j β(t) j β(t) fjφj L m n k φj j β(t) Ω nk j β(t) fjφj. (2). (2) From (4), Ω nk = φn k j A φj ( ). Thus, (2) becomes v(d m n k ) v(s m n k )= L m n k φj j β(t) j β(t) fjφj. (22) φ nk ( ) j A φj Because v(s m n k ) is defined to be v(s m n k ) = max{v(d m n k ), v(a m n k )} [2], v(d m n k ) is thus derived as v(d m n k ) = max{v(dn m k,v(a m n k )} + L m n k φj j β(t) j β(t) fjφj where v(a m n k ) is determined from dv(t) dt =, (23) φ nk ( ) j A φj φj j A. It j β (t) φj shoud be noted that β (t) is the set of a backogged sessions before the arriva of the m-th LAC PDU. (23) determines the virtua finishing time of a LAC PDU once it arrives. Such virtua time is used as a priority for fair queueing of packets. In a wideband CDMA frame, a packets are serviced from the queue with the smaest virtua finishing time towards the one with the argest virtua finishing time, unti system capacity is exhausted. -783-7753-2/3/$7. (C) 23 IEEE IEEE INFOCOM 23
C. Checking Code Channe Constraint In order to decrease the compexity of the transceiver of mobie terminas, the maximum number of DCHs for user n k, denoted by M nk, is normay much ess than the avaiabe codes on the OVSF tree. Thus, in the wideband CDMA GPS scheduing scheme, this constraint must be aways checked so that it is not vioated. Given M nk DCHs for user n k and an OVSF tree with M eves of codes, the avaiabe transmission rates for such a user need to be determined by considering the foowing two factors: ) The tota number of DCHs is not arger than M nk. 2) The codes of two DCHs cannot be on the same path to the root of the OVSF tree. One approach to this probem is to check if a transmission rate can be decomposed into ( M k ) smaer transmission rates such that each of these smaer transmission rates is 2 i (i =,,M) times of the basic transmission rate r b, i.e. the transmission rate of a DCH using the M-th eve code on the OVSF code tree. As an exampe, the set of avaiabe transmission rates when M nk =2and M =7is {, 2, 3, 4, 5, 6, 8, 9,, 2, 6, 7, 8, 2, 24, 32, 33, 34, 36, 4, 48, 64}, where the transmission rates are in unit of r b. Such an approach can be used off-ine to determine the set of avaiabe transmission rates. Suppose the number of packets schedued by the wideband GPS scheduing scheme for user n k is g nk. Then, the transmission rate corresponding to g nk packets must be g nk r b, since one packet needs a basic transmission rate. If such a transmission rate ies in the set of avaiabe transmission rates for user n k, a the schedued packets can be transmitted. Otherwise, the maximum transmission rate that is ess than g nk r b is used. Suppose this aowed transmission rate h nk r b, then (g nk h nk ) packets cannot be served in the current frame due to the constraint of avaiabe code channes. However, since some packets are not served, a certain amount of system capacity is not utiized by user n k. Therefore, under this situation, the wideband CDMA GPS scheduing scheme needs to be performed again unti system capacity is fuy utiized or a packets are serviced. V. Effective-Bandwidth CAC Since singe ce is considered when designing a MAC protoco, user mobiity is not taken into account in the CAC agorithm. A. The CAC Agorithm The CAC agorithm is ony appied to rea-time connections. Non-rea-time connections are aways accepted. In order to take into account the contribution of non-rea-time traffic to the overa normaized transmission rates, a minimum normaized transmission rate, denoted by R nrt, is reserved for non-rea-time traffic. R nk = denotes the normaized + transmission rate of a rea-time connection n k, then, from (2), the foowing constraint must be satisfied for rea-time connections: K N k R nk R nrt, (24) k= n k = where N k is the number of rea-time connections in the system. During the ife time of the connection n k, R nk is a random variabe because r nk varies from frame to frame. Thus, a satisfaction factor α is used to evauate the probabiity that (24) is satisfied, i.e., for <α, if K N k Pr( R nk R nrt ) >α, (25) k= n k = then (24) is satisfied with probabiity α. Given a satisfaction factor α, the admission region of reatime connections can be determined based on (25). In what foows, Gaussian approximation [4] is used to derive the admission region. Of the same service type k, different connections are independent and foow the same traffic characteristics. Considering the connection n k, denote the mean and variance of R nk as µ k and σk 2, respectivey. According to the centra imit theorem, when N k is arge, N k n k = R n k can be approximated by a Gaussian random variabe G k with mean and variabe equa to N k µ k and N k σk 2, respectivey. Since {G k,k =,,K} are independent Gaussian random variabes, K k= G k can aso be approximated by a Gaussian random variabe G whose mean and variance are K k= N kµ k and K k= N kσk 2, respectivey. Thus, (25) becomes Pr(G < R nrt ) >α. (26) According to the characteristics of Gaussian random variabe, (26) is satisfied if and ony if R nrt E[G] β, (27) Var[G] where E[G] and Var[G] are the mean and the variance of G, respectivey, and β is defined by e t2 /2 dt = α. (28) 2π β ith E[G] = K k= N kµ k and Var[G] = K k= N kσk 2 and after some agebra, (27) becomes K N k µ k + β K N k σk 2 R nrt, (29) k= k= which determines an admission region (N,,N K ). hen a connection of service type k arrives, N k is increased by one. hether or not this connection can be accepted is decided by checking if the new vaue of N k satisfies (29). If the answer is positive, the connection is accepted. Otherwise, it is rejected. hen a connection of service type k is terminated, N k is decreased by one, which equivaenty reeases the -783-7753-2/3/$7. (C) 23 IEEE IEEE INFOCOM 23
capacity occupied by the connection. Therefore, in the CAC agorithm given by (29), the effective bandwidth of each new arriva connection does not need to be expicity determined, which decreases the compexity of the CAC agorithm. VI. Performance Evauation Simuation is carried out to evauate the performance of the MAC protoco. A. Traffic Modes and System Parameters Six types of traffic modes [9] are considered in the simuation:. The duration of a voice connection is exponentiay distributed with the average equa to 8. s. The average ength of takspurts and gaps are. s and.35 s, respectivey. hen a connection is in the takspurt, the average ength of minispurts and gaps are.235 s and.5 s, respectivey. BER of voice traffic is 3, timeout vaue of a voice packet is 2 frames, and the maximum number of DCHs for a voice connection is.. The duration of an audio connection is aso exponentiay distributed with the average equa to 8. s. The bit rate is 28 kbps. BER of audio traffic is 4, timeout vaue of an audio packet is 4 frames, and the maximum number of DCHs for a connection is 2. CBR video. The duration of a CBR video connection is exponentiay distributed with the average equa to 36. s. The bit rate is 22 kbps. BER of CBR video traffic is 5, timeout vaue of a packet is 4 frames, and the maximum number of DCHs for a connection is 3. VBR video. The duration of a VBR video connection is exponentiay distributed with the average equa to 8. s. A mutipe-state mode is used to simuate the traffic in each connection. Duration of each state of the mode is aso exponentiay distributed with the average equa to 6 msec. The traffic rate in each state is obtained from a truncated exponentia distribution with the maximum and minimum bit rates equa to 2 and 42 kbps, respectivey. BER of VBR video traffic is 6, timeout vaue of a packet is 6 frames, and the maximum number of DCHs for a connection is 3. Computer data. The ength of a computer data message is exponentiay distributed with the mean size equa to 3 kbytes. BER of data traffic is approximatey equa to zero, and the maximum number of DCHs for a ca is. Packets cannot be dropped because of timeout. Emai. The emai mode proposed in [9] is used to simuate emais. BER of emai is approximatey equa to zero, and the maximum number of DCHs for a ca is. Packets cannot be dropped because of timeout. Among a cas, 6%, 5%, 6%, 9%, 5%, and 5% are generated by voice, audio, CBR video, VBR video, computer data, and emai traffic. For the wideband CDMA system, the eve of the OVSF tree M is 7, the basic transmission rate r b is 9.5 kbps, system bandwidth is 5 MHz, satisfaction Average Packet Deay (msec) Average Packet Deay (msec) 55 5 45 4 35 3 25 2 2 5 5 Fig.. Computer Data Emai Average packet deay versus ca arriva rate without CAC. factor α is.99, minimum capacity for non-rea-time traffic R nrt is.2, and and in (2) is.5. The SINR vaues corresponding to different BERs are given in Tabe I. Frame ength t fr is ms. The performance metrics for the MAC protoco are the average packet deay d p, packet oss ratio p, throughput t r, and ca bocking probabiity b c. The average packet deay consists of three components, i.e., d p = d r + d a + d t, where d r is the average time of successfuy sending a packet transmission request, d a is the queueing time before a code channe is aocated to a packet, and d t is the transmission time of a packet after the code channe is aocated. In the simuation, the code channes in RACH are assumed to have a arge number, so the request transmission is coision-free. Thus, the average vaue of d r is equa to haf a frame because the generation time of a request is uniformy distributed in a frame. Since pure CDMA is used in FDD mode wideband CDMA, a packet is transmitted within a whoe frame. Thus, d t is equa to the frame ength. p of a connection is defined as p = N N +N t, where N is the number of ost packets due to timeout, and N t is the number of packets being successfuy transmitted. r t is defined as the tota packets being transmitted in a frame. b c of a service type is defined as the b c = C b C b +C a, where C b and C a are the number of bocked cas and accepted cas, respectivey, of a service type. B. Numerica Resuts ) Experiments without CAC: Three performance metrics such as average packet deay, packet oss ratio, and throughput are used. To show the performance versus traffic oad, different experiments are performed by varying the ca arriva rates. The average packet deay versus the ca arriva rate is shown in Fig.. The resuts of packet oss ratio and throughput are shown in Fig 2 and Fig. 3, respectivey. As shown in, the average packet deay of each service type increases as the traffic oad becomes high. The average packet deay of nonrea-time traffic is much arger than that of rea-time traffic because: a) the virtua finishing times of non-rea-time packets -783-7753-2/3/$7. (C) 23 IEEE IEEE INFOCOM 23
Packet Loss Ratio.45.4.35.3.25.2.5..5 Average Packet Deay (msec) Average Packet Deay (msec) 26 25 24 23 22 2 2 5 5 Computer Data Emai Fig. 2. Packet oss ratio versus ca arriva rate without CAC. Fig. 4. Average packet deay versus ca arriva rate with CAC. 8 6 Overa 2 3 4 Throughput (packets/frame) 2 8 6 4 Packet Loss Ratio 4 5 6 7 2 8 9 9 Fig. 3. Throughput versus ca arriva rate without CAC. Fig. 5. Packet oss ratio versus ca arriva rate with CAC. are much arger; b) no timeout occurs in non-rea-time traffic. Among the service types of rea-time traffic, voice has the smaest average packet deay. The average packet deay of audio is smaer than that of video traffic, because the bit rate of a video connection is arger. VBR video has the argest average packet deay due to its bursty characteristics. It shoud be noted that the packet deay of rea-time traffic is guaranteed to be ower than a bound, because each rea-time service has a certain timeout vaue and the deay of each packet cannot exceed this vaue. As shown in Fig. 2, the packet oss ratio of voice traffic is much arger than those of other service types, which is reasonabe since voice traffic is ess sensitive to packet oss than other types of traffic. As shown in Fig 3, the increment of throughput becomes ower and ower when the traffic oad increases. The reason is that the system capacity is graduay approached as traffic oad increases, and thus, the packet oss becomes higher and higher. This can be iustrated by the comparison between the throughput in Fig. 3 and the packet oss ratio in Fig. 2. As shown in Fig. 2, when the packet arriva rate is very high, the packet oss ratio of a service type becomes too arge to be acceptabe. In order to resove this issue, CAC must be used so that some connections are bocked to ensure that traffic oad in the system does not exceed the system capacity. 2) Experiments with CAC: In this experiment, the CAC scheme proposed in Section V is empoyed to admit connections of rea-time traffic. ith CAC, the average packet deay, packet oss ratio, and throughput are shown in Figs. 4, 5, 6, respectivey. In addition, the connection bocking probabiity of each rea-time service type is shown in Fig. 7. As iustrated by the comparison between Fig. and Fig. 4, the average packet deay is greaty decreased. For rea-time traffic, the average packet deay is amost decreased to 25 msec. Comparisons between the resuts in Fig. 5 and Fig. 2 show that the CAC agorithm greaty reduces packet oss ratio of rea-time traffic. Such improvement is because CAC ensures that the traffic oad does not exceed the system capacity. It shoud be noted that the minimum vaue of average packet deay is 25 msec, because the average packer deay at east consists of haf a frame of request deay, one frame of queueing deay, and one frame of transmission deay. -783-7753-2/3/$7. (C) 23 IEEE IEEE INFOCOM 23
Throughput (packets/frame) 6 4 2 8 6 4 2 Overa Average Message Transmission Deay (frames) 2 8 6 4 2 New Protoco: cass II A New Protoco: cass II B Protoco in [5]: cass II A Protoco in [5]: cass II B 2 3 4 5 6 7 8 9 Offered Load (packets/frame) Fig. 6. Throughput versus ca arriva rate with CAC. Fig. 8. Average message transmission time versus offered oad. Connection Bocking Probabiity.2.8.6.4.2..8.6.4.2 Fig. 7. Connection bocking probabiity versus ca arriva rate. As shown in Fig. 6, the system throughput is ower than that in 3. The reason is that the rejected connections reduce overa offered oad in the system. However, the throughput reduces by ess than % for two orders of magnitude of reduced packet oss ratio. As shown by the connection bocking probabiity in Fig. 7, connections with higher traffic rate are easier to be rejected. Thus, the CAC scheme is fair to different service types. VII. Comparisons The protoco proposed in [5] and the new MAC protoco proposed in this paper have simiar features. For exampe, mutimedia traffic with diverse QoS requirements can be supported by both protocos. Moreover, power aocation to a code channe is considered in the CAC and scheduing schemes. However, in the new MAC protoco, the minimumpower aocation agorithm and the wideband CDMA GPS scheduing scheme are used. Therefore, interference-sensitive system capacity can be utiized more efficienty, and packets of different services can be fairy serviced according to their heterogeneous BER and QoS requirements. In order to show such advantages, the new MAC protoco is compared with that in [5]. Since peak rate is assigned to a rea-time connection in [5], QoS of rea-time connections is aways guaranteed. Thus, the average message transmission deay of non-rea-time traffic (i.e., the cass II traffic in [5]) is the ony performance metrics used in the comparison between two protocos. Assumptions, traffic modes, and system parameters are summarized as foows, which are same as those in [5]: ) There are two types of cass-ii traffic. Cass II-A traffic is deay sensitive, whie cass II-B is deay-toerabe. 2) The number of packets in each message of cass II- A is geometricay distributed with average 2. Same distribution is used for cass II-B, but the average message size is equa to 8 packets. 3) 5 mobies are generated for each traffic type. In each mobie, the message generation rate is a Poisson process. The average generation rate of cass II-A is.9λ/5, whie that of cass II-B is.λ/5. Thus, each traffic type has the equa oad (i.e.,.8λ), and the tota offered oad is.9λ 2+.λ 8=3.6λ. 4) Fading is not considered in signaing and contro channes. 5) The system bandwidth is equa to 28 8 kbps, i.e.,.24 Mbps. 6) 5 out of 7 bandwidth units are reserved for cass II traffic. Rea-time connections (i.e., cass I traffic in [5]) do not change throughout the simuation. Thus, they do not affect the performance of cass II traffic. 7) The average message transmission deay consists of three components: request access deay, queueing deay, and transmission deay. In addition, a arge number of request access code channes are used so that request coision rarey occurs. In the simuation, when this number is equa to 25, the coision is amost free. Under these assumptions and system parameters, the average message transmission deay versus the offered oad is shown in Fig. 8. In our protoco, athough a wideband CDMA GPS scheduing scheme is used, the average message -783-7753-2/3/$7. (C) 23 IEEE IEEE INFOCOM 23
transmission deay of two traffic types is different. The reason is that a arger weight (i.e. φ i in (23)) is assigned to cass II-A traffic than that to cass II-B traffic. As shown in Fig. 8, for cass II-B traffic, the new MAC protoco achieves a ower average message transmission deay. hen the traffic oad is ower than 4 packets/frame, the difference of the average message transmission deay is approximatey between.5 and 3.5 frames. However, when the offered oad is between 45 and 65 packets/frame, the average message transmission deay achieved by the protoco in [5] is more than frames arger than that of the new protoco. For cass II-A traffic, when the offer oad is ess than 9 packets/frame, the average message transmission deay of the new protoco is a itte smaer than that achieved by the protoco in [5]. However, when the offered oad is between 9 and 95 packets/frame, the average message transmission deay of the new MAC protoco is more than frames smaer. The main reasons for the better performance achieved by the new protoco are as foows: Minimum-power aocation agorithms. The new MAC protoco uses minimum-power aocation for each traffic type. Athough the protoco in [5] aocates different power eves to different traffic type, minimum-power aocation is not considered. Therefore, given the same offered oad, the new protoco can transmit more packets in a frame. Fair scheduing schemes. A wideband CDMA GPS scheduing scheme is used in the new MAC protoco to aocate the code channes to mobie terminas. The transmission order of packets are determined based on their virtua arriva times. Thus, packets of cass II-B are not necessariy transmitted ater than those of cass II-A traffic. However, in [5], higher priority is aways given to cass II-A traffic. Thus, the message transmission deay of cass II-B traffic is arge and increases abrupty when the offered oad is as ow as 45 packets/frame. Better re-transmission mechanism. In the protoco in [5], athough the packets to be re-transmitted are put into a FIFO queue and have higher priority than packets in the round-robin queue, packets in the FIFO queue of cass II-B traffic is sti ower than packets in the FIFO and the round-robin queues of cass II-A traffic. Thus, for cass II-B traffic, the packets to be re-transmitted have a arge queueing deay. In our protoco, a packet to be re-transmitted participates in scheduing according to its virtua arriva time. Thus, this packet does not need to be transmitted ater than a cass II-A traffic. VIII. Concusions In this paper, a MAC protoco was proposed for next generation wireess networks based on FDD mode wideband CDMA. First, a minimum-power aocation agorithm was derived with the consideration of both MC and VSF transmissions. Based on this agorithm, a wideband CDMA GPS scheduing scheme was proposed. It took into account different constraints such as interference-imited system capacity, BER and QoS requirements of mutimedia traffic, and imited number of codes in each mobie termina. An effective bandwidth-based CAC agorithm was aso proposed with the consideration of minimum-power aocation. The MAC protoco proposed in this paper guarantees heterogeneous QoS of mutimedia traffic and improves the overa system throughput. Acknowedgment The author woud ike to thank Prof. Ian F. Akyidiz for his advice. References [] E. Dahman et a., ``CDMA - The Radio Interface for Future Mobie Mutimedia Communications,'' IEEE Trans. Veh. Techno., vo. 47, no. 4, pp. 5-7, Nov. 998. [2] M. Haardt et a., ``The TD-CDMA Based UTRA TDD Mode,'' IEEE J. Seect. Areas Commun., vo. 8, no. 8, pp. 375-385, Aug. 2. [3] ``3rd Generation Partnership Project; Technica Specification Group Radio Access Network; Radio Resource Management Strategies (Reease 4),'' 3GPP TR 25.922, v 4.. Mar. 2. [4] C. Roobo et a., ``A Proposa for an RLC/MAC Protoco for ideband CDMA Capabe of Handing Rea Time and Non Rea Time Services,'' IEEE VTC'98, 998, pp. 7-. [5] S. Choi and K. G. Shin, ``An Upink CDMA System Architecture with Diverse QoS Guarantees for Heterogeneous Traffic,'' IEEE/ACM Trans. Networking, vo. 7, no. 5, pp. 66-628, Oct. 999. [6] L. Carrasco and G. Femenias, ``-CDMA MAC Protoco for Mutimedia Traffic Support,'' IEEE VTC', 2, pp. 293-297. [7] R. Fantacci and S. Nannicini, ``Mutipe Access Protoco for Integration of Variabe Bit Rate Mutimedia Traffic in UMTS/IMT-2 Based on ideband CDMA,'' IEEE J. Seect. Areas Commun., vo. 8, no. 8, pp. 44-454, Aug. 2. [8] ``3rd Generation Partnership Project; Technica Specification Group Radio Access Network; MAC Protoco Specification (Reease 4),'' 3GPP TR 25.32, v 4.. Mar. 2. [9] J.-. So and D.-H. Cho, ``Access Schemes for Integrated /Data Transmissions Over Common Packet Channe in 3GPP,'' IEEE Commun. Letters, vo. 5, no. 2, pp. 46-48, Feb. 2. [] T. H. Hu and M. M. K. Liu, ``A New Power Contro Function for Mutirate DS-CDMA Systems,'' IEEE Trans. Vehic. Techno., vo. 47, no. 6, pp. 896-94, June 999. [] L. Song, N. B. Mandayam, and Z. Gajic, ``Anaysis of an Up/Down Power Contro Agorithm for the CDMA Reverse Link Under Fading,'' IEEE J. Seect. Areas Commun., vo. 9, no. 2, pp. 277-286, Feb. 2. [2] A. K. Parekh and R. G. Gaager, ``A Generaized Processor Sharing Approach to Fow Contro in Integrated Services Networks: The Singe- Node Case,'' IEEE/ACM Trans. Networking, vo., no. 3, pp. 344-357, June 993. [3] M. A. Arad and A. Leon-Garcia, ``A Generaized Processor Sharing Approach to Time Scheduing in Hybrid CDMA/TDMA,'' IEEE INFO- COM'98, Apr. 998, pp. 64-7. [4] J. S. Evans and D. Everitt, ``Effective Bandwidth-Based Admission Contro for Mutiservice CDMA Ceuar Networks,'' IEEE Trans. Vehic. Techno., vo. 48, no., pp. 36-46, Jan. 999. [5] T.-H. Lee, J. T. ang, ``Admission Contro for Variabe Spreading Gain CDMA ireess Packet Networks,'' IEEE Trans. Vehic. Techno., vo. 49, no. 2, pp. 565-575, March 2. [6] A. Sampath, P. S. Kumar, and J. M. Hotzman ``Power Contro and Resource Management for a Mutimedia CDMA ireess System,'' IEEE PIMRC'95, 995, pp. 2-25. [7] Z. Liu et a., ``Channe Access and Interference Issues in Muti-Code DS-CDMA ireess Packet (ATM) Networks,'' ACM/Batzer ireess Networks (INET), vo. 2, pp. 73-93, 996. [8] T. Minn and K.-Y. Siu, ``Dynamic Assignment of Orthogona Variabe- Spreading-Factor Codes in -CDMA,'' IEEE J. Seect. Areas Commun., vo. 8, no. 8, pp. 429-44, Aug. 2. [9] I. F. Akyidiz, D. A. Levine, and I. Joe, ``A Sotted CDMA Protoco with BER Scheduing for ireess Mutimedia Networks,'' IEEE/ACM Trans. Networking, vo. 7, no. 2, pp. -3, Apr. 999. [2] S. J. Goestani, ``A Sef-Cocked Fair Queueing for Broadband Appications,'' IEEE INFOCOM'94, 994, pp. 636-646. -783-7753-2/3/$7. (C) 23 IEEE IEEE INFOCOM 23