162 Giovanni Giambene, Cristina P´arraga Niebla, Victor Y. H. Kueh Fig. 5.16: Mean packet delay at RLC buffers for different packet scheduling algorithms. streaming applications have been serviced, given that the detrimental affect is not posing significant degradation on the QoS target of streaming users. Figure 5.17 shows the mean jitter experienced by each individual service when employing MLPQ and DDQ packet scheduling. Obviously, DDQ features much lower jitter for both streaming service and download service than MLPQ, especially for lower-class and lower date rate users. Since the uni- directional streaming service in S-MBMS is quite sensitive to delay-variation (jitter), this result proves that DDQ packet scheduling provides a way to balance all FACH queues in order to get the minimum delay variation for streaming services. Analysis of channel utilization ratio Figure 5.18 shows the average S-CCPCH physical channel utilization for both MLPQ and DDQ. Both schedulers managed to achieve throughput values close to the optimum. For instance, the S-CCPCH channel utilization ratios are 97.8%, 96.2%, 85.4% respectively under MPLQ scheduling; whilst they achieve 98.4%, 96.2%, 86.4% respectively under DDQ scheduling. Therefore, DDQ manages to obtain a slight channel utilization improvement on those S-CCPCHs carrying background traffic. To summarize, the DDQ algorithm achieves the following advantages over the MPLQ scheduling scheme: • Dynamic proportional delay-driven prioritization; Chapter 5: ACCESS SCHEMES AND PACKET SCHEDULING TECH. 163 Fig. 5.17: Mean packet jitter at RLC buffers for different packet scheduling algorithms. Fig. 5.18: S-CCPCH utilization for MLPQ and DDQ. 164 Giovanni Giambene, Cristina P´arraga Niebla, Victor Y. H. Kueh • Highest utilization for the background class without posing significant degradation on streaming class; • Significant improvement on mean delay and mean jitter performance; • Better overall system utilization. 5.3.4 Packet scheduling with cross-layer approach Due to the nature of wireless transmissions, satellite communications suffer from strong variations of the received signal power due to shadowing and multipath fading. Shadowing of the satellite signal is due to obstacles in the propagation path (buildings, trees, bridges, etc). Whereas multipath fading occurs because the satellite signal is received not only via the direct path, but also being reflected from objects in the surrounding area. Due to different propagation distances, the multipath signals may add destructively and leads to deep fades. Due to these variations, the most critical part in satellite communications is the communication link between the satellite and the user terminal (i.e., downlink). The downlink availability could be the limiting factor for the performance of the overall system. Thus, a scheduler employing an explicit cross-layer technique is proposed, where signaling interactions from the phys- ical layer are employed so that the scheduler is aware of the channel state of the users. In this cross-layer design, a multicast packet scheduler is developed that relies on the prediction of the wireless channel conditions to improve the performance of downlink transmissions via satellite for a TDMA-based air interface. The domain architecture for the multicast service under consideration is illustrated in Figure 5.19. The entities in the service provider will provide the interface between RAN and external packet data networks. The scheduler is at the Earth station and a GEO satellite relays the multicast information to all users through Multicast Terminals (MTs) and Terminal Equipment (TE). A reliable multicast transport protocol is assumed to guarantee delivery and congestion control mechanism. A unicast return link will be required for acknowledgments. Based on a TDMA framework, the system under consideration supports scheduled access on both downlink and uplink. Downlink capacity is organized into fixed 80 ms MF-TDMA frames that are composed of a sequence of fixed 20 ms time slots. Channel State Information (CSI) of each user, which is the information from physical (PHY) layer, is considered in the decision mechanism whether to transmit or not the next multicast packet. The CSI parameter is averaged over a total of N frames to become a conditional parameter for the next transmission. CSI parameter is updated periodically. The update through uplink bearer is contained within a 200 kHz sub-band, which is further divided in frequency, and time slots and each slot may contain a burst from a user. Chapter 5: ACCESS SCHEMES AND PACKET SCHEDULING TECH. 165 Fig. 5.19: Domain architecture for a TDMA-based satellite system. It is assumed that the CSI update provides a reasonably accurate prediction of the slowly varying elements (i.e., valid at least within one round-trip-delay) of the channel conditions. The satellite link is modeled using the Lutz’s two-state model with variation of fade duration for open and shadowed en- vironments [45]. The following results indicate that a significant performance improvement is possible by adopting a cross-layer design approach in a fading environment. Description of scheduler’s task Prior to assigning the slot, which is the resource allocation step, the different multicast services need to be prioritized. In our scheme, the prioritization is performed at two levels. The first prioritization is static: the scheduler orders the services according to their QoS classes (streaming and best effort), i.e., streaming service is assigned higher priority than best effort. The second level of prioritization is based upon the cross-layer information provided with CSI for services featuring the same level of priority. This prioritization is more dynamic and confined only for best-effort traffic. The algorithm is described as follows: • For all incoming multicast packets, the packet scheduler aims to serve the packets according to priorities dynamically allocated to them. Streaming traffic packets have higher priority to access to time slots at all times. • For the remaining slots, if best effort traffic packets arrive, the scheduler scans the CSI intended for the multicast group. The acquisition of CSI will be performed for each user in the intended multicast group. The update of channel condition is acquired in every 20 ms, according to the slot and 166 Giovanni Giambene, Cristina P´arraga Niebla, Victor Y. H. Kueh burst definition. The channel state information is contained in the bearer control signaling data unit. In the evaluation of scheduler’s performance, we consider TDMA channels in which each frame is divided into fixed control and data sub-frames. The user channel information is updated through bearer control signaling data unit. The data sub-frame length is the block length which is a data transmission unit size of 125 bytes and it fits into time slots each holding one packet. • The packet scheduler will check for estimated E b /N 0 values of each user i in the multicast group, Γ i . A reference E b /N 0 threshold, γ T , is compared with Γ i and the number of users satisfying this reference will be the decision making parameter for slot allocation. • If within one slot, more than one packets of the same priority arrives, then the scheduler will check for the packet with higher percentage of satisfied users. If the number of satisfied users from a particular multicast group is above a certain threshold, the slot is allocated to that packet. If not, then the packet will be delayed and retried for the next slot, provided that the next slot is not intended for higher priority traffic. For more details, the interested reader is referred to [46]. Performance evaluations The following study assumes stationary users and slowly-varying channels in satellite links where fade duration holds within one CSI update. The results are here based on perfect channel predictions; we assume no channel estimation loss occurs. This assumption might be impractical in a satellite environment where the propagation delay is high, but the results with this assumption permit to have a good indication of the effectiveness of this scheme to achieve a high reliability multicast transmission. In this study, two different scenarios have been examined. The first one, the single environment scenario, assumes that all users are subject to identical channel conditions (the single environment model uses an elevation angle α =80 ◦ and values for µ and σ calculated for urban areas with K factor of 7, where K represents the Ricean factor which is defined as the ratio of the dominant component to the scatter contribution [45]). The proposed technique aims at reducing the number of retransmissions that stem from bad channel conditions. Figure 5.20 [46] exhibits some rather interesting results, where parameter zeta is defined as follows: a packet is retransmitted only if the percentage of users in the multicast group that experience Packet Loss Rate (PLR) higher than a defined PLR threshold is greater than a percentage, denoted by parameter zeta. It should be pointed out that PLR significantly diminishes, by endowing the multicast packet scheduler with CSI (cross-layer approach). Moreover, PLR is hardly affected by an increase in the multicast group size, whereas the greater the parameter zeta is, the lower PLR is. Figure 5.21 [46] illustrates the probability that at least one user will request retransmission versus the multicast group size. Apparently, as the Chapter 5: ACCESS SCHEMES AND PACKET SCHEDULING TECH. 167 Fig. 5.20: Packet loss rate versus multicast group size. See reference [46]. Copyright c 2005 IEEE. Fig. 5.21: Probability of at least one user in a multicast group requesting retransmission (failure rate) versus multicast group size. See reference [46]. Copyright c 2005 IEEE. 168 Giovanni Giambene, Cristina P´arraga Niebla, Victor Y. H. Kueh size of the multicast group increases, so does the probability of retransmission. Furthermore, this probability can be reduced by increasing parameter zeta. What is more important is that the greater the size of the multicast group, the higher the average packet delay, as illustrated in Figure 5.22. The value of zeta used to obtain the results in Figure 5.22 is 0.9. Fig. 5.22: Average packet delay versus target number of users (normalized) in good channel condition. At this point, multi-channel environments are simulated based on the parameters in Table 5.5. Area Rician K factor α ( ◦ ) µ σ % Users Suburban 0dB 20 −1.69 2.70 20 Urban 3dB 20 −13.90 3.06 40 Urban 7dB 80 1.75 0.80 30 Suburban 10 dB 60 0.14 0.40 10 Table 5.5: Simulation parameters for the multi-environment scenario. In this case, multicast users are subject to different channel conditions, and the empirical models presented in [45] were deployed. A user requires retransmission only if the difference between γ ref , which is the reference E b /N 0 from the AWGN channel model to achieve a target PLR of 10 −2 , Chapter 5: ACCESS SCHEMES AND PACKET SCHEDULING TECH. 169 and Γ (t), which is the E b /N 0 value of the signal received from this user, is greater than a given E b /N 0 threshold, γ T : γ ref − Γ (t) >γ T [dB]. (5.13) Figure 5.23 [46] depicts the probability that at least one user will re- quest retransmission versus E b /N 0 threshold. Evidently, the retransmission probability decreases as E b /N 0 threshold increases. It should also be noted that the retransmission probability decreases as parameter zeta increases. As far as the average packet delay is concerned, it becomes clear from Figure 5.24 that as E b /N 0 threshold increases, the mean delay increases since the multicast packet scheduler refrains from transmitting packets to multicast groups typically experiencing bad channel conditions [46]. This approach has been shown to reduce unnecessary transmission of best-effort traffic and hence reduces unnecessary bandwidth usage and retrans- mission requests. However, in achieving relatively good channel utilization for a multicast group, higher average packet delay is expected in the cross-layer scheduler. The average packet delay can be regulated according to the power threshold a user is estimated to receive at the downlink transmission. It is also important to note that this approach consumes an amount of resources in performing the channel prediction algorithm. The accuracy of the channel quality is highly dependent on the channel model used. Fig. 5.23: Probability of at least one user in a multicast group requesting retransmission (failure rate) versus E b /N 0 threshold. See reference [46]. Copyright c 2005 IEEE. 170 Giovanni Giambene, Cristina P´arraga Niebla, Victor Y. H. Kueh Fig. 5.24: Average packet delay versus E b /N 0 threshold. See reference [46]. Copyright c 2005 IEEE. 5.4 Conclusions Satellite communications have a potential market in providing high downlink bit-rate services and in supporting multicast services on broad areas of the Earth. These are the reasons why this Chapter has focused on HSDPA and MBMS provision via a GEO bent-pipe satellite. In both cases suitable network architectures and radio resource management techniques have been investigated to support such services in an appropriate and efficient way. For HSDPA, the results showed by the Proportional Fair scheduler are sub-optimal for mixed traffic classes, since it does not provide any QoS differentiation among diverse applications. The study of proposed enhance- ments to the PF scheduler to support QoS differentiation, such as the Exponential Rule, should be addressed in the future for the satellite case. Furthermore, the impact of the round trip time in the acquisition of channel state information has been shown in the form of packet losses in intervals of misalignments between current channel state and information available at the Gateway. In particular, the simulation results in a simplified scenario (using a GOOD/BAD channel model) show non-negligible losses due to the use of outdated information in the selection of the best suited TFRC for transmission. Hence, if it is desired to reduce the number of retransmissions, delay compensation strategies or larger margins in the selection of TFRCs should be adopted. Furthermore, a more complex channel model should be Chapter 5: ACCESS SCHEMES AND PACKET SCHEDULING TECH. 171 considered in order to take into account the channel variation dynamics typical of S Band (S-UMTS band). For the provision of broadcast and multicast services, it has been shown that packet scheduling is an important element within the RRM framework. Aiming at a more efficient provision of heterogeneous QoS-differentiated MBMS services over S-UMTS, novel packet scheduling algorithms have been proposed. These algorithms take into account the impact of important per- formance factors reflecting service QoS demands in order to provide traffic differentiation and overall system performance optimization. To tackle the deteriorating effect of changing propagation environments in multicast trans- missions, channel estimation can fill the void whilst obtaining the current channel state. Statistical channel models can be used to represent channel variations to be exploited by packet scheduler for its decisions. For traffic with strict delay bound, a negotiation between delay and channel states can be facilitated by a cost function where a trade-off between delay and throughput is expected. . most critical part in satellite communications is the communication link between the satellite and the user terminal (i.e., downlink). The downlink availability could be the limiting factor for. of downlink transmissions via satellite for a TDMA-based air interface. The domain architecture for the multicast service under consideration is illustrated in Figure 5 .19. The entities in the. addressed in the future for the satellite case. Furthermore, the impact of the round trip time in the acquisition of channel state information has been shown in the form of packet losses in intervals of