Performance of VoIP over HSDPA in mobility scenarios

Transcription

1 Performance of VoIP over HSDPA in mobility scenarios Petteri Lundén Nokia Research Center P.O. Box 45, FI-45 Nokia Group, Finland Jussi Äijänen, Kari Aho, Tapani Ristaniemi University of Jyväskylä P.O.Box 35, 414 JKL, Finland jussi.aijanen, kari.aho, Abstract This paper presents an evaluation of Voice over IP (VoIP) system level performance in High Speed Downlink Packet Access (HSDPA) network. The main purpose is to determine how VoIP over HSDPA performance is affected by HSDPA handover i.e. serving High Speed Downlink Shared Channel (HS-DSCH) cell change. This paper studies how sensitive the capacity of VoIP service is to handover delay and User Equipment (UE) velocity. Additionally, the reliability of handover related signaling in downlink is evaluated. The system level performance of VoIP in the mobility related scenarios is evaluated with fully dynamic system level tool in which mobility of users, radio resource management functionalities and the interactions between them are explicitly taken into account. The simulation results indicated moderate sensitivity of VoIP capacity to the UE velocity. High velocities lead to capacity loss for VoIP over HSDPA. Handover delays started to have significant effect only at UE velocity of 5 km/h or more. Reliability of handover signaling does not pose any problems even at high velocity and incurs only 1 % resource overhead when transmitted over HSDPA. I. INTRODUCTION Deploying of Voice over IP (VoIP) over cellular networks was made possible with reasonable quality and efficiency for the first time when Wideband Code Division Multiple Access (WCDMA) networks [1] were introduced. However, the spectral efficiency was not still high enough to exceed Circuit Switched (CS) voice performance until High Speed Packet Access (HSPA) technologies [2] were introduced in Third Generation Partnership Project (3GPP) Release 5 and Release 6 to improve the packet data capabilities of WCDMA networks. Both downlink (High Speed Downlink Packet Access (HSDPA)) and uplink (High Speed Uplink Packet Access (HSUPA)) counterparts of HSPA evolution were originally designed to achieve higher capacities and better coverage for high transmission rates and delay tolerant services. Improvements to capacities and to better coverage are based on innovative techniques such as Hybrid Automatic Repeat Request (HARQ), shortened Transmission Time Interval (TTI) and Node B controlled packet scheduling. However, as the previous capacity and mobility studies regarding the HSDPA have proved (see e.g. [3], [4] [5]), also delay critical small bit rate services such as VoIP can benefit from those new features. Using VoIP in cellular networks brings many benefits and cost saving opportunities for end users, network vendors and operators. VoIP over HSPA introduces, for instance, the possibility to move into All-IP networks instead of using separate CS network elements for voice and Packet Switched (PS) elements for data. This would lead to cost savings as the CS related part of the core network would not be needed anymore. VoIP would also make it possible to offer rich call services [6] with less complexity. These advantages among others have increased the interest towards VoIP performance and performance enhancements. This paper aims to deepen the knowledge of VoIP over HSDPA performance by evaluating the performance in fully dynamic system level simulations where e.g. mobility of users and interactions of the radio resource management functionalities are explicitly taken into account. By conducting these dynamic network simulations we are able to point out some aspects that need to be paid special attention without having costly and time consuming physical network trials. In this study the main focus will be on the impact of HSDPA handovers i.e. impact of serving High Speed Downlink Shared Channel (HS-DSCH) cell change to the VoIP over HSDPA performance. The goal is to determine how sensitive VoIP service is to handover delay, User Equipment (UE) velocity and to find out the conditions where VoIP users are able to smoothly switch from cell to cell while maintaining good Quality of Service (QoS). The rest of the paper is organized as follows. In section two, we discuss the simulator modeling issues especially related to VoIP over HSDPA. In section three, we present the simulation scenario, simulation results and analysis of those results. Finally, conclusions are drawn in section four. II. VOIP OVER HSDPA MODELING We have used a dynamic system level radio network simulator in our VoIP over HSDPA studies. The simulator was presented for the first time in [7] and after that it has been used in various performance evaluations of WCDMA and its enhancements, see e.g. [8] and [9]. Simulation tool enables detailed simulation of users in multiple cells with realistic propagation, fading and mobility models (adopted from [1]). Since the early days, the simulator tool has been developed with e.g. HSDPA and VoIP functionality. Modeling issues which are the most relevant regarding to this study are presented briefly in the following subsections /8/$ IEEE 246

2 Fig. 1. Example of the VoIP traffic model Fig. 2. Serving HS-DSCH cell change, [13] A. VoIP traffic Duration of each VoIP call is random and distributed according to negative exponential distribution (mean length is given as parameter). Discontinuous transmission (DTX) of VoIP call is simulated through alternating, random duration activity and silence periods (see Figure 1). The duration of these periods is distributed according to negative exponential distribution. During activity periods, a new VoIP packet which assumes robust header compression (ROHC) [11] is generated on constant intervals. Size of the packet (including payload and RTP/UDP/IP/PDCP/RLC compressed header) is given in the parameters as well as the interval between successive packets (e.g. 38 byte packet [12] every 2 ms). B. VoIP QoS VoIP quality of service (QoS) is monitored at the receiving end by following quality and delay of the packets for each VoIP user. VoIP user is counted to be in outage if it does not receive correctly at least 95 % of its packets within the delay threshold (e.g. 8 ms). Outage is monitored over a sliding observation window of user s activity. Silence period are therefore not part of the monitoring window. VoIP user can also be dropped if the quality of the transmission drops significantly. Dropped users are also counted to be in outage. Total cell capacity is defined as the maximum number of active VoIP users that can be supported by the cell without exceeding 5 % cell outage. Cell capacity is measured over the whole simulation time. C. VoIP traffic scheduling The details of VoIP over HSDPA specific scheduling have been studied previously e.g. in [4]. We assume the same principles and we use VoIP optimized MAC-hs packet scheduler with dynamic resource allocation (High Speed Physical Downlink Shared Channel (HS-PDSCH) codes and power) in our simulations. Principles of the scheduling are the following: 1) The scheduler selects a user to be scheduled in the next TTI according to the used priority metric (e.g. Round Robin (RR) or Proportional Fair (PF)) fromthe Scheduling Candidate Set (SCS). 2) Then, the scheduler assigns required amount of resources to the selected user (or skips the user if there are not enough resources). 3) Finally, if there are still schedulable users in the SCS and remaining HS-PDSCH resources, the scheduling continues again from step 1. If there are remaining resources but none of the users in the SCS can be scheduled with those, the scheduling continues with more relaxed SCS definition from step 1. The SCS is dynamically updated in each TTI so that it includes: 1) Users that have at least M pkts VoIP packets buffered in the Node B. Here M pkts is a parameter of the scheduling algorithm. (M pkts can be adjusted during the scheduling to allow more users to enter SCS.) 2) Users with pending retransmissions in their Hybrid-ARQ manager. D. HS-DSCH serving cell change reliability Figure 2 illustrates serving HS-DSCH cell change procedure. We assume that the HS-DSCH cell change can be made only to a cell that is already in the active set. Further optimization of the serving HS-DSCH cell change mechanism would be the combined procedure of HS-DSCH cell change and active set update in soft handover. This would result in shorter procedural delay, but is not considered in this paper. During synchronized HS-DSCH serving cell change, physical channel reconfiguration Radio Resource Control (RRC) message is sent from Radio Network Controller (RNC) to UE. We are assuming that these messages are transmitted using HSDPA, i.e. Signaling Radio Bearer (SRB) is mapped on HSDPA. The RRC messages are assumed to fit in PDU size of 336 bits so that they can be sent using a single High Speed Physical Downlink Shared Channel (HS-PDSCH) code. In this paper the following methodology is used for assessing the reliability of physical channel reconfiguration message transmissions on HS-DSCH (and thus the reliability of the cell change mechanism): 1) Es/N data is collected during the HS-DSCH cell changes in the system simulations. 2) The last 1 ms of the Es/N data is used as an input for a separate simulation of RRC message reliability. 3) RRC message transmission is assumed to have the highest priority and it is allocated as large share of the HS-PDSCH power resources as needed. The benefit of this approach is that it allows us to quickly simulate multiple RRC message transmission events for each 247

3 TABLE I SYSTEM SIMULATION PARAMETERS. TABLE II RRC MESSAGE SIMULATION PARAMETERS. Cellular layout 18 hexagonal macro cells Inter-site distance 28 m Pathloss model Modified Okumura-Hata [1] Channel profile Vehicular A UE velocities 3, 3, 5, 9 km/h Average VoIP call length 6 seconds Packet size 38 bytes Packet inter-arrival rate 2 ms Call drop threshold 15 % Delay budget 8 ms Observation window length 5s Simulation time 2 s Power resources HSDPA 15 W Power resources HS-PDSCH 1 W Code resources HS-PDSCH 1 Maximum power resources HS-SCCH 5W Number of HS-SCCH s 4 HS-DSCH cell change procedural delay 1, 2, 4 ms Packet scheduling algorithm VoIP optimized [see section II-C] Mpkts scheduling parameter 2 handover event that takes place in the system simulation. As outputs from this simulation we acquire RRC message residual error rate, average number of transmission attempts made per message and average resource (power and code) consumption per message. The handover delay is not explicitly simulated because it depends also on the uplink performance. For simplicity only downlink performance is considered in this paper. In the uplink, the reliability of RRC messages does not become a problem due to soft handover. III. SIMULATION SCENARIO AND RESULTS ANALYSIS The selected scenario for simulations is a macro cell scenario with 18 cells and 9 base stations. Simulation scenario is illustrated in Figure 3. Main simulation parameters are shortly listed in Table I and the parameters used in handover message reliability simulations are listed in Table II. In the following, the VoIP performance is evaluated using system level simulations. The effect of UE velocity, handover delay and handover signaling reliability is considered. The performance with different UE velocities and handover delays is mainly evaluated using the outage criterion described in section II-B. The cost of handover becomes high with this outage criterion in the case the user has multiple handovers close to each other (within the 5 second observation window). As the handovers are likely to cause additional loss of VoIP packets, having several of them in a short time can push the user over the outage limit. However, this outage criterion is generally accepted metric for VoIP system level performance [14]. A. Impact of UE velocity to the VoIP performance Average VoIP cell capacities are presented in Figure 4. The figure shows the achieved capacities for 1 ms, 2 ms and 4 ms handover delays. To assess the impact of UE velocity without the full cost of handover procedure we compare the Max. Tx attempts per message 4 Number of simulated messages per HO event 1 HS-DSCH BLER target 1% Distance between BSs is 28 m Fig. 3. Simulation scenario Simulation scenario layout capacities with somewhat ideal 1 ms handover delay. From the results it can be seen that the VoIP capacity is highly sensitive to UE velocity. Higher velocities lead to lowered VoIP capacity. When going from 3 km/h to 3 km/h capacity drops by roughly 2 %. With higher velocity the capacity drops even further. This is caused by the fact that the radio channel is changing more rapidly and is thus more challenging with high UE velocity. Moreover, the amount handovers is increased significantly when UE velocity increases, as Figure 5 illustrates. B. Impact of HS-DSCH cell change procedural delay The procedural or handover delay in the HS-DSCH serving cell change is the time between sending the measurement report 1d (change of best cell) from the UE to the point when transmission starts from the new cell. In this study different delay values are studied in order to determine how VoIP over HSDPA would perform in a situations when this procedural delay exists. The delay is caused by e.g. different number of HARQ retransmissions needed to get the RRC messages through. Apart from the procedural delay, additional delay to the handover comes from handover measurement filtering and triggering. Average VoIP cell capacities with different delay values are illustrated in Figure 4. As it can be seen from that figure the capacity becomes sensitive to handover delay only at higher velocities when the performance is compared to the more idealistic 1 ms delay case. When the UE velocity is relatively low, such as 3 or 3 km/h, the degradation caused by the delay stays well within 7 % margin even though the delay would increase to as high as 4 ms. With 3 km/h the

4 VoIP capacity [users/cell] Fig. 4. delays ms 2 ms 4 ms Average VoIP cell capacity with different UE velocities and handover HS DSCH serving cell changes [/user/minute] Fig Number of HS-DSCH cell changes per user per minute performance can be even enhanced in some situations as the delay starts to limit the number of (ping-pong) handovers, however this requires relatively stable channel conditions in order to keep the tradeoff between delay and number of handovers on the positive side. When the UE velocity is higher, e.g. 5 or 9 km/h, radio channel conditions are changing more rapidly and thus any delay before the handover event will have stronger impact. Increased delay leads to increased BLER, as the serving HSDPA sector remains suboptimal for longer period of time while the conditions can get worse and worse when the UE is moving further away from the original point where the event was triggered. With 5 km/h, 2 ms delay causes degradation that stays within 8 % but if the delay would increase to 4 ms then the performance would degrade roughly 24 % when compared to the case with 1 ms delay. With 9 km/h the situation is even worse as then the channel conditions have already degraded the VoIP performance in the given scenario notably. The performance loss with 2 ms in the 9 km/h case is already close to 15 % and if the delay would be around 4 ms then the performance degradation would lead to around 4 % loss in comparison to 1 ms case. All in all, the results show that the handover delay is critical to the performance of VoIP service over HSDPA, especially when the UE velocity gets higher. This is due to the fact that with high UE velocities the performance is already greatly impacted by the worsened radio channel conditions and thus it is even more important to have fast handover procedure. C. Handover reliability In addition to the VoIP capacity results, we investigate the transmission of DL RRC messages on HSDPA during the handover procedure. At high velocity the situation of the UE can deteriorate very quickly while waiting for the handover. We intend to see whether this has an impact on the reliability of the handover mechanism (if the Physical Channel Reconfiguration RRC message does not get through). Figure 6 presents the residual error rate of DL RRC messages for different UE velocities and handover delay of 2 ms and 4 ms. The network load is set to maximum VoIP capacity, approximately at 5 % cell outage point. The results show that the residual error rate after 4 transmission attempts is below.5 % in all the cases. Furthermore, the simulation results show that even in the worst case the average overhead of DL RRC messages (in terms of transmit power) is less than 1 % of the total HS-PDSCH power. Average number of required transmission attempts per RRC message is below 1.5 and most users need only 1 or 2 attempts even in 9 km/h case. In 3 km/h case almost all of the messages go through first time (average number of needed transmission attempts is approximately 1.1). The results show that the SRB transmissions can be made significantly more robust than the VoIP service (as the transmissions can be prioritized and given more resources). There Residual RRC message error rate [%] Fig. 6. messages 2 ms 4 ms Residual error rate of Physical Channel Reconfiguration RRC 249

5 are, however, ways to improve the DL RRC message reliability even further than what is considered in this paper. For example, increasing the maximum number of retransmissions will decrease the residual error rate. Also reduction of the handover delay improves the reliability. Therefore any optimization of the handover procedure has an impact. [14] C. Tao, M. Kuusela, and E. Malkamki, Uplink capacity of VoIP on HSUPA, in Proc. IEEE Vehicular Technology Conference (VTC), Melbourne, Australia, May 26. IV. CONCLUSIONS The purpose of this study was to evaluate the effect of mobility on the VoIP performance in HSDPA network with dynamic system level simulator. The sensitivity of VoIP capacity to handover delay and UE velocity was investigated. Additionally, the reliability of handover related signaling in downlink was evaluated. The system simulation results showed that due to the strict delay and quality requirements, VoIP QoS quickly deteriorates if the serving HS-DSCH cell is not optimally selected. At high UE velocity this effect is more pronounced because of more frequent handovers as well as more rapidly changing radio channel conditions. Therefore it is essential to keep the handover delays as short as possible when running a real-time service such as VoIP. The reliability of DL RRC messages was simulated during HS-DSCH cell change. It was found that those messages go through with high reliability (residual error rate below.5 %) while consuming less than 1 % of the total HSDPA capacity even in the worst case. Hence, the reliability of the handover procedure is not a problem. REFERENCES [1] H. Holma and A. Toskala, Eds., WCDMA for UMTS, 4th ed. John Wiley & Sons Ltd, 27. [2], HSDPA/HSUPA for UMTS, 1st ed. John Wiley & Sons Ltd, 26. [3] S. Wager and K. Sandlund, Performance evaluation of HSDPA mobility forvoiceoverip, inproc. IEEE Vehicular Technology Conference (VTC), Dublin, Ireland, April 27. [4] P. Lunden and M. Kuusela, Enhancing performance of VoIP over HS- DPA, in Proc. IEEE Vehicular Technology Conference (VTC), Dublin, Ireland, April 27. [5] B. Wang, K. I. Pedersen, T. E. Kolding, and P. E. Mogensen, Performance of VoIP on HSDPA, in Proc. IEEE Vehicular Technology Conference (VTC), Stockholm, Sweden, May 25. [6] H. Holma, M. Kuusela, E. Malkamäki, K. Ranta-aho, and C. Tao, VoIP over HSPA with 3GPP release 7, in Proc. IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC), Helsinki, Finland, September 26. [7] S. Hämäläinen, H. Holma, and K. Sipilä, Advanced WCDMA radio network simulator, in Proc. IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC), Osaka, Japan, September 1999, pp [8] S. Hämäläinen, WCDMA radio network performance, Ph.D. dissertation, University of Jyväskylä, 23. [9] J. Kurjenniemi, A study of TD-CDMA and WCDMA radio network enhancements, Ph.D. dissertation, University of Jyväskylä, 25. [1] Selection procedures for the choice of radio transmission technologies of the UMTS, Universal Mobile Telecommunications System (UMTS 3.3) TR , Rev. v3.1., November [11] Robust Header Compression (ROHC); Framework and four profiles: RTP, UDP, ESP, and uncompressed, IETF RFC 395, July 21. [12] R. Cuny and A. Lakaniemi, VoIP in 3G networks: An end-to-end quality of service analysis, in Vehicular Technology Conference, VTC 23-Spring, Jeju Island, Korea, April 23. [13] High Speed Downlink Packet Access (HSDPA); Overall description; Stage 2, 3GPP TSG TS25.38, Rev. 7.3., June