In Dedicated Channel model, two alternatives are considered: Transmission of one single packet Sequence of packet transmission.

Transcription

1 Department of Signal Theory and Communications - Technical University of Catalonia (UPC) Campus Nord-Edificio D4 - C/ Jordi Girona Barcelona-Spain E_mail: angelah@tsc.upc.es, ferranc@tsc.upc.es In the last few years, the demand for multimedia and packet data services, based on Internet standards, has dramatically increased in wired networks. This will clearly impact on the design of the third generation radio systems, such as the UMTS promoted by ETSI, where it is required to provide service quality for multimedia communications, mainly Internet access and video/picture. Because many multimedia applications are bursty, it will be essential to optimize third-generation techniques for supporting variable bit rate and packet transmission capabilities with quality of service requirements in order to increase the efficiency. Circuit switching should also be supported for the provision of some constant bit rate services or very high quality voice transmission. The UMTS terrestrial radio access-utra,[4], defines two operation modes, namely, frequency-division duplex (FDD) and time division duplex (TDD). The FDD mode is based on pure Wideband CDMA (W- CDMA) while the TDD mode includes an additional time-division multiple-access component according to the TD/CDMA proposal. For packet data transmission based on the W-CDMA mode, there are two defined methods: Short packets, which can be transmitted on Random Access Channel without any reservation scheme. Then the overhead necessary to transmit a packet and the access delay is kept to a minimum. Obviously, the packet can be lost due to collision. Notice that, for short packets only open loop power control is assumed. For the case that the user has large amount of data to be transmitted, these transmissions are based on a Dedicated Channel (DCH) selected according to a demand assignment protocol [2]. In that case, the MS sets up a dedicated code using an initial Random Access request, wherein the type of traffic to be transmitted is specified. Then, the network evaluates the request and decides if the necessary resources can be assigned to the MS. Once the Dedicated Channel is assigned to the MS, the channel is still not allowed to start a transmission. It needs to wait until the network specifies the transport format and the time in which it can initiate the transmission. This procedure will introduce some overhead and delay, which can be as smaller as far the length of packets increases. However, in the Dedicated Channel mode the data transfer is more reliable due to the closed loop power control performed and the absence of collisions. In Dedicated Channel model, two alternatives are considered: Transmission of one single packet Sequence of packet transmission. In the case of real time services, the allocation procedure is the same. The only difference is that in this the mobile station must start transmission immediately once a dedicated spreading code has been assigned.

2 In the context of W-CDMA, the issue of the Quality of Service (QoS) support can be divided into two subjects: Temporal transparency and semantic transparency. The semantic transparency is obtained by means of the power control while temporal transparency is achieved using transmission scheduling. The time scheduling scheme implemented at the base station is responsible for arranging transmission of packets within their specified rate requirements and delay tolerances while the purpose of power control criterion is to meet the Bit Error Rate (BER) of simultaneously transmitted packets. In any case, the basic approach for packet transmission over Dedicated Channels is to exploit the property of delay tolerance, which is a characteristic of many data sources, to improve the system efficiency. The main idea is to schedule the transmission of the packets coming from delay tolerant users so as to reduce, at any time instant, the interference seen by the other users. As the result, the users can transfer information at higher rates, leading to an overall increase in throughput. Besides, in order to support differentiated QoS the base station scheduler must allow a fair sharing of the available resources depending on the requested QoS. Each user has a quality of service requirements that specifies in terms of: maximum transfer delay and delay jitter, guaranteed minimum transmission rate r i, and maximum bit error rates (BER) or frame error rates (FER) mapped into an equivalent E b /N o constraint denoted by γ i, E () = γ = + R η M R Given a set of requirements, we adopt an optimization criterion to assign an optimum level of the transmitted powers for all the users in such as way that their sum is minimized, guaranteeing that the E b /N o requirements of all of the users are met. This criterion minimizes the interference caused to other cells increasing the system efficiency. It can be shown that for a bandwidth W and N transmitters, an optimum value of the transmitted power can be achieved if and only if [3]: (2) < i =..N = + γ If this condition is satisfied for a set of rates and E b /N o values, then the power can be obtained using η (3) = Cres = - + UHV γ γ = + This previous condition can easily be extended to the transmitted power constrained case and cell environment. Now, taking into account the minimum total transmitted power criterion we can consider two transmission modes for delay tolerant users. In the first one, all users admitted in the system are allowed to transmit information, with a rate as higher as the allowed in the system to satisfy minimum power constraint requirements. For the second case, in a given time instant, only a limited number of users are allowed to transmit while the remaining users can not transmit even though they are in contact with the base through a control channel. Given that in W-CDMA only a set of spreading factors (W/r i ) can be used, we choose to analyze in this paper the second option. On the other hand, W-CDMA defines two types of dedicated physical channels: the dedicated physical data channel (DPDCH), used to carry dedicated data, and dedicated physical control channel (DPCCH), used to transmit control information [pilot bits, transmit-power control (TPC) commands, and optional transport format indicator (TFI)]. The DPCCH is transmitted continuously at a constant symbol rate with relatively

3 low power enabling physical maintenance (i.e. closed-loop power control, time synchronization, and up-link channel estimation for coherent demodulation). In the up-link, the DPDCH and DPCCH are transmitted in parallel in phase and quadrature-phase branches, respectively, using different orthogonal codes. Although there is no self-interference among DPDCH and DPCCH, we must consider the effect of the DPCCH channel coming from users. Thus, the condition we must satisfy in order to allow the transmission of M<N users in a given time-slot is: (4) = M= + G, γ + G, ( ) F, γ M F, M + + < + = = F G wherein r d,, r c are the rates, γ d, γ c are the E b /N o constraints for DPDCH and DPCCH respectively, and m is the amount of overhead introduced by DPCCH. A centralized demand assignment protocol is implemented in order to guarantee QoS requirements. Every user who has packets waiting for transmission sends a request over Random Access Channel to setup a dedicated code. This initial Random request includes the type of traffic and the amount of data to be transmitted. Once, the dedicated channel (code) is assigned, users wait the notification of the base station conform they can transmit in the next frame. At the end of each frame, the base station specifies the set of services permitted to transmit simultaneously, together with the transfer format (e.g. the bit rate) to be used for packet transmission. This procedure is done in conjunction with power control in such a way that the QoS requirements of all scheduled services are met. If user has more packets to transmit the mobile station sent an access request on the dedicated channel. Useful information of access request is contained at the beginning of frame to allow the response of the base station at the end of the same frame. This access request is performed with maximum spreading factor (W/r=26) in order to decrease interference originated over the rest of users. So, a minimum capacity is reserved for request in order to agree with minimum transmitted power criterion. To increase the data throughput and decreasing delay, data rates can be increased during periods of low activity. Processing delay of a frame is assumed for all packets. The retransmission strategy used is the selective repeat ARQ scheme with negative acknowledgements. In ETSI W-CDMA the scheduling is a resource allocation function closely connected to the transport format selection (rate of the dedicated channel, coding used, etc). During communication MAC scheduler selects the appropriate transport format within an assigned transport format set for each active transport channel depending on source rate and radio resource limitations. The selection can be done on a ms frame basis or slower. Depending on the selected transport format one or more PDUs (Packet Data Units) from higher layer may be mapped onto a transport block (time period in which a transport format is not allowed to change), consisting of one ms frame. Transport block determines maximum size of PDU. The main objective of scheduler is to integrate traffic sources with different transmission rates, priorities, delays and packet loss requirements optimizing the up-link channel utilization. Packet loss requirements are guaranteed if condition (4) is observed. To ensure QoS requirements in terms of rate and minimum delay, several scheduling strategies based both on static or dynamic priorities can be used. Between service classes static priorities are used while among packet belonging to the same class dynamic priorities can be applied. The simplest approach for achieving QoS is to apply a Round Robin strategy. However this strategy can only guarantee a large upper bound on packet delay. Furthermore, since the maximum delay is governed by the round duration, which is the same for all the sources, it can not provide different delay bound to different services. Then for that reason, in the paper we propose a dynamic priority scheduling algorithm based on lifetime of packet. The scheduler arranges transmission packets in increasing order of priority. Priorities are calculated as:

4 () ( ( ), ) + N D N The lifetime (in frames) of the packet placed on the first places of queue of terminal, when it sends a request for transmit, is calculated as d i -(t-t a ),, where d i is the delay tolerance of packet (normalized to the frame duration) and t and t a are respectively the current frame number and the frame number when packet was generated. The lifetime of the other packets is calculated according to equation (), where T is the estimated packet inter-arrival time. The base station has a request table containing terminal requirements and the lifetimes of next packets to be transmitted. The lifetimes are updated and decremented at each frame. The base station has not to be informed about the arrival of each new packet because it can estimate the time of the next packet applying the same scheme. Only in case of a faulty estimation, the wireless terminal has to transmit an explicit capacity request in order to resynchronize the estimation algorithm. The packets to be transmitted are scheduled in increasing order of lifetime. If vt>d, the next packet to be reserved has not yet been generated. Therefore the reservation will be inserted in service queue as soon as vt=d. Packets with delay constraints are retransmitted until they are correctly received, or their deadlines are violated. As a consequence of the dynamic frame assignment, an error packet can be recovered immediately through a retransmission attempt. Besides, we combine this scheme with assignation of static priorities between classes of services. In this case we contemplate two possibilities: ) all resources are available for all classes of traffic and 2) a minimum capacity is guaranteed for non-delay constrained traffic while the remaining is assigned with preemptive priority to delay constraint traffic. In order to assess the performances of the different scheduling disciplines previously described in the paper, some computer simulations have been done. In particular, for the simulation purposes, two classes of services have been considered: DD and CD services: Data services with low (ms) and long (3ms) delay constraints. Two levels of QoS ( -3, -6 ) have been considered. Only convolutional coding together with a retransmission scheme (ARQ) are used in case of BER= -3, whereas Reed Solomon, outer interleaving and convolutional coding or Turbo codes could be used to achieve BER= -6. UDD: Data services with non-delay constraints and a Block Error Rate (BER)< -. Although circuit-switched mode transmission has been proposed by ETSI for the DD and CD logical channels, packet switched and channel activity<% have been considered in this work, in order to support real time services. This is assumed in order to assess the capabilities of the W-CDMA as a packet radio system. Therefore, in our simulations, the real-time sessions are carried out by means of DD and CD services and from a traffic viewpoint they are modeled by means of ON_OFF traffic sources. An ON-OFF traffic source is modeled as two-state Markov modulated process. In the ON state, packets are generated in a fixed intervals of time t. In the OFF state, no packets are generated. The duration of the ON and OFF state are exponentially distributed with mean t ON and t OFF.. Mean residence time in ON state and OFF state are 3ms and 6ms respectively while a service of 3,4kps is assumed although real transmission rate is 28kbps, so a spreading factor of 32 is assumed. From the viewpoint of the UDD services, which bears non-real time sessions, the traffic source is based on a normal Pareto distribution for the data packet size. Reference [] gives a detail description of the model. In particular 8kbps data rate is assumed. The return channel (downlink DPDCH-PDCCH) is assumed to be error free, whereas receiver and transmitter buffers are assumed infinite. Figures to 3 show the behavior of the system considering two different CD services simultaneously with a maximum delivery time delay equal to msec. and 3 msec respectively. Moreover, it is assumed the same number of simultaneous users for both services. The schedule policies taken into account are a

5 Round Robin strategy and a dynamic priority algorithm based on lifetime of packet. We will call this last scheduling algorithm time-stamp strategy Throughput (Transmitted packets / frame ) Packet generated / class Packet transmitted CD (d=ms) Packet transmitted CD (d=3ms) Packet transmitted CD (d=ms) Packet transmitted CD (d=3ms) Number of users ( %CD+%DD) % Packet Dropping DD (d=ms) CD (d=3ms) DD(d=ms) CD (d=3ms) Number of users (%CD+%DD) Throughput measured as number of packets per frame versus the total number of users. Two DC with maximum delay and 3 msec services simultaneously. Dropping Probability, measured in %, versus the total number of users Two DC with maximum delay and 3 msec services simultaneously. From the viewpoint of the system throughput we can realize that the maximum number of simultaneous user is around 3 users per cell, in which case the total number of packets send by frame is 7 packets per service. Notice that there is not difference between both scheduling strategies in terms of throughput. However, if we look at the dropping probability versus the number of users, see figure 2, we can realize that the time stamp strategy performs more fairly than the Round robin scheme. Certainly, in the time stamp case both dropping probability curves are very close, that is, both services access to the radio resources with more or less the same priority, whereas in the case of round robin scheme the service with a maximum delay of msec is penalized by a higher dropping probability. Notice that for the time stamp case the maximum number of users per cell for a dropping probability equal to % is 34. Finally, in figure 3 the maximum delay for both services versus the number of users is represented. Notice that for a number of total users that guarantees that the dropping probability is lower than % the delay is kept much lower that the maximum allowed. Average packet delay (frames) DD (ms) CD (3ms) DD (ms] CD (3ms) Number of users (%CD+%DD] Delay ( frames ) 4 DD CD) 4 Priority Round Robin DD 3 CD Priority Round Robin + Reserve 3 DD CD Number of www users Delay, measured in terms of number of frame, versus the total number of users. Two DC with maximum delay and 3 msec services simultaneously. Round-Robin strategy. Three simultaneous services. Two DC services with users each one plus the UDD service. Delay of the different services, measured as number of packets per frame versus the number of UDD users. Figures 4 to 6 show the behavior of the system considering three different services simultaneously. In particular, two DC services with user every one and a maximum delivery time delay equal to msec. and 3 msec respectively are assumed. Figures 4 and show the evolution of the service delay versus the number of users of the UDD service (WEB users). Round-robin and time stamp strategies are assumed in figures 4 and respectively. For both strategies, three different algorithms are consider: basic algorithm, two different queues one for the real-time services (CD services) with higher priority and another for the non-

6 real time service (WEB service) and finally the same algorithm but reserving a minimum of % of the capacity for the non-real time service. From the figures we can see that in the case of non-real time service there is a great difference in the delay values when a Round-robin strategy is assumed, whereas for timestamp strategy the three algorithms perform more or less equal. For real time services there is not significant differences between different algorithm mainly for a msec maximum delay service. However, if we take into account the dropping probability, the Round-robin strategy performs significantly worst than the timestamp. Certainly, we can achieve up to 3 simultaneous users if a queue algorithm is considered together with the time-stamp strategy, whereas for Round-robin the maximum number of WEB users is around 8 to. 7, Delay (frames) DD CD Priority Time Stamp DD CD Priority Time stamp + Reserve DD CD Pdrop packets DD services 4, 4, 3, 3, 2, 2,,, Time Stamp Priority Time Stamp Priority Time Stamp + Reserve Round Robin Priority Round Robin Priority Round Robin +Reserve Number of www users,, Number of www users. Time-Stamp strategy. Three simultaneous services. Two DC services with 3 users each one plus the UDD service. Delay of the different services, measured as number of packets per frame versus the number of UDD users. Dropping Probability, measured in %, versus the total UDD users. Three simultaneous services. 2 DC services with 3 users each one plus the UDD service. The paper focuses on the comparative analysis of the performances of different scheduling strategies to guarantee the Quality of Service of the link for the ETSI W-CDMA packet transmission proposal. A Dedicated Channel is assumed, and several scheduling strategies such as Round robin, or Packet ifetime are analyzed. Selective ARQ schemes based on a frame structure of ms are considered. The scope of this study is limited to the up-link. From the obtained results, it can be concluded that the dynamic priority algorithm based on lifetime of packet performs more fairly that the Round robin scheme, with similar implementation complexity [] Universal Mobile Telecommunications System (UMTS). Selection procedures for the choice of radio transmission technologies of the UMTS, UMTS Technical Report 3.3, 997. [2] Christiaan Roobol, Per Beming, Johan undsjö, Mathias Johansson, A proposal for an RC/MAC Protocol for Wideband CDMA Capable of Handling Real Time and Non Real Time Services, proceedings of the IEEE VTC 98. [3] Ashwin Sampath, Sarath Kumar and Jack M. Holtzman, Power Control and Resource Management for a Multimedia CDMA Wirelless System, PIMRC 9. [4] Submission of Proposed Radio Transmission Technologies SMG2, attachment 2 of Circular etter 8/CCE/47. This work has been supported by CYCIT (Spanish National Science Council) under grant TIC98-684