Capacity of VoIP over HSDPA with Frame Bundling

Transcription

1 Capacity of VoIP over HSDPA with Frame Bundling Yong-Seok Kim Telecommunication Network Business Samsung Electronics Youngheon Kim Telecommunication Network Business Samsung Electronics Abstract In this paper, we evaluate the capacity of voice over internet protocol (VoIP) services over high-speed downlink packet access (HSDPA), in which frame-bundling (FB) is incorporated to reduce the effect of relatively large headers in the IP/UDP/RTP layers. Also, a modified proportional pair (PF) packet scheduler design supporting for VoIP service is provided. The main focus of this work is the effect of FB on system outage based on delay budget in radio access networks. Simulation results show that VoIP system performance with FB scheme is highly sensitive to delay budget. We also conclude that HSDPA is attractive for transmission of VoIP if compared to the circuit switched (CS) voice, that is used in WCDMA (Release 99). I. INTRODUCTION HSDPA is a tremendous performance upgrade to 3GPP Release 99 in WCDMA for packet data that delivers peak rates of 14 Mbps and that is likely to increase average throughput rates to about 1 Mbps, a factor of up to three and a half times over Release 99. HSDPA also increases spectral efficiency by a similar factor [1], [2]. It is necessary to provide user equipments (UE) with a multitude of real-time and interactive services on an internet protocol (IP) based transport and service platform [3]. The transmission of voice using packet data IPs is arguably the hottest attention in telecommunication technology today. One reason is it has high visibility in the consumer space. A long-distance VoIP calling is cheaper to operate, maintain, and upgrade than comparable solutions using switched digital or analogue phone service. In addition, it facilitates the creation of new services that combine voice communication with other media and data applications such as video and file sharing [4]. Early VoIP studies were focused on the wireless local area networks (WLAN) since it provides convenience and achievable high-speed data rate as that of the wireline network [5], [6]. Despite the success of VoIP in wireline and WLAN networks, the most widely held expectation is the latest broadband 3G technologies such as CDMA2000 1X EV-DO [7] and WCDMA HSDPA [8]. This is because of inherent spectral inefficiencies that seemingly would disappear if both voice and data were carried on 3G wireless radio link. Backhaul facilities would be more efficient because voice, data traffic and all signaling protocols would be carried on same IP facilities. Hence, in this work, it is attempted to provide system-level simulation results of VoIP capacity which reflects important factors such as frame-bundling (FB) and delay budget. This paper is organized as follows. In Section II, we present the system of VoIP over HSDPA. In Section III, we describe the performance criteria and simulation assumptions. Finally capacity evaluation and conclusion are presented in Section IV and Section V respectively. II. SYSTEM DESCRIPTION OF VOIP OVER HSDPA A. Traffic model and protocols A telephone conversation can be represented by ON/OFF patterns. ON periods correspond to a conversant talking and OFF periods are due to silences. The duration of both ON and OFF periods is negative exponentially distributed with an average of appropriate seconds. Generally, it can be approximated two state Markov traffic model with a suitable voice activity factor [9]. The adaptive multirate (AMR) voice codec is mandatory for voice services in WCDMA systems. It is also a reasonable assumption for VoIP service. During bursts of conversation, with AMR mode 12.2kbps, the VoIP application generates 32-byte voice payload at 20ms intervals [10]. During silent periods, a 7-byte payload carries a silence descriptor (SID) frame at 160ms intervals. A typical VoIP protocol stack, which employs the real-time transport protocol (RTP), is encapsulated in the user datagram protocol (UDP). This, in turn, is carried by IP. The combined size of these protocols is a 40-byte IPv4 header or a 60-byte IPv6 header. Obviously, a header overhead seriously degrades the spectral efficiency to support VoIP service. Therefore, FB can be used to reduce the effect of relatively large headers in the IP/UDP/RTP layers and to decrease the occurrence of bit-stuffing due to the mismatch between the size of VoIP packet and that of HSDPA media access control (MAC) frame format [11]. This means that several voice packets are sent in one RTP packet. However, there is a trade-off between delay and overhead. That is, the more packets are bundled together the more packetization delay is introduced but at the same time overhead is more and more reduced. For conversational services low delay is crucial, hence only few payloads (typically up to 3 in the case of AMR) are bundled. The FB feature is supported in the AMR payload format. Additionally, efficient and robust header compression (ROHC) techniques can reduce the size of the IP/UDP/RTP headers to as little as 3 or 4 bytes. Maximum compression 1 byte can be achieved by imposing limitations. We can apply the IP/UDP/RTP packet header compression with ROHC protocol by [12].

2 UE NodeB RNC SGSN/GGSN IP Network TABLE I SUMMARY OF END-TO-END DELAY COMPONENT UL delay : 40ms UL Processing delay : 30ms DL Processing delay : 30ms DL delay : Variable Fig. 1. Backhaul delay : 30ms NodeB-GSN:15ms, GSN-NodeB:15ms End-to-end delay component B. RF scheduler of MAC-hs for VoIP IP Network delay : about 42ms The use of effective scheduling algorithm is necessary for improving the performance of system, since the HSDPA system shares resources with multi-users at the same TTI. The MAC-hs sublayer, located in the Node B, handles scheduling using a 2ms transmission time interval (TTI). The HSDPA system was designed for services with high throughput for BE traffic requirements but it can also be used for VoIP. This is the proportional fair (PF) algorithm which was shown to be appropriate for elastic best-effort (BE) traffic. PF scheduling takes advantage of independent temporal channel variation at each mobile, by giving priority to users with transitorily better channel condition [13]. For VoIP service over HSDPA it is beneficial to include time-delay factor in scheduling algorithm. To measure delay, the scheduler puts time-stamps in each packet as it arrives at the priority queue. The priority would be more weighted as the delay increase. Hence, in this paper, modified PF scheduling algorithm is adopted as follow: P (i) VoIP (i) (n) = P PF (n) f (remain y, Que size) (1) where P (i) PF (n) is the priority of user i calculated by PF scheduling scheme for BE traffic. The delay function f ( ) can be designed by f ( ) = (Que size) β /(remain y) γ, where, Que size is the size of VoIP packets that must be scheduled at n-th TTI on queue, remain y is the remaining delay budget from the current n-th TTI to the delay bound at RF scheduler, and β and γ are appropriate weight factor for each one. C. Physical layer in HSDPA HSDPA is the performance upgrade version of WCDMA Release 99, named as 3GPP WCDMA Release 5 [14]. It is shared channel transmission that includes MAC-hs function features such as fast link adaptation, fast hybrid automatic repeat request (HARQ), and channel-dependent Node B RF scheduling. HSDPA was designed for services with high throughput for BE traffic requirements but it can also be used for VoIP. Generally, VoIP and high-throughput services have different characteristics that require different treatment. First of all, VoIP flows must be handled by higher priority than interactive traffic. Given that HSDPA maintains individual scheduling buffers to exploit changes in radio conditions per link, it can also be used to take service characteristics into account. The use of separate priority queues makes it possible to optimize HSDPA scheduling for VoIP. Moreover, some other features of the HSDPA radio access solution which are necessary for efficient VoIP transport are thought such as radio Delay component Delay assumption (ms) - Voice encoder - 20ms for AMR 12.2Kbps - RTP frame bundling -FB0(0ms), FB1(20ms), etc - NodeB scheduling+harq - Variable correspond to FB (Max. 110ms when FB0) - NodeB - Fixed (30ms) ROHC, RLC+MAC processing Downlink propagation - UE scheduling+harq - Fixed (40ms) - UE - Fixed (30ms) processing, buffering, voice decoder Uplink propagation - Backhaul delay (GSN-Node B) - Fixed (30ms) - IP network delay - About 42ms link controller unacknowledged mode (RLC UM), delay-based priority, optimized code multiplexing, and reduction of power overhead, for instance, by means of a fractional dedicated physical channel (F-DPCH) which is the feature of WCDMA Release 6. A consideration of F-DPCH will be included in a future study. III. PERFORMANCE METRICS AND SIMULATION ASSUMPTION A. End-to-end delay budget for QoS support To ensure end-to-end QoS in a packet-switched (PS) network, low delay is one of the most important criteria for maintaining high-quality VoIP service. However, to attain high VoIP capacity, the scheduler must have sufficient time to manage the voice packets. End-to-end delay is the measurement of time that it takes the talker s mouth to reach the listener s ear. Of the assumed 285ms end-to-end delay budget for qualified voice service, about 110ms is available for scheduling in the downlink [15]. The delay in IP and backhaul network is in general bounded 72ms from coast to coast by a fixed value. Such simplification allows us to focus on the delay budget within radio access network. Hence, the end-to-end delay budget in the case of mobile-to-mobile conversation can be assumed as Figure 1. Here, we only consider the impact of downlink transmissions. Therefore, available delay budget in uplink scheduling can be fixed. Although VoIP performance depends on both downlink and uplink performance, we would like to set aside the consideration of both directions comprehensive study to future research. Table 1 summarizes the related end-to-end delay budget. B. Performance criteria for VoIP The main objective of this paper is about the VoIP capacity for the maximum number of VoIP users that can be supported per sector without exceeding a given outage threshold. If PS networks do not guarantee that packets are delivered at a scheduled time, packets would be dropped under network traffic loads congestion. Although some packet loss occurred, the voice quality is not affected as long as the amount of packet loss is less than outage threshold. To proceed this work, we

3 TABLE II SYSTEM SIMULATION PARAMETERS (HSDPA) Parameter Assumption Source traffic - AMR 12.2kbps, packet overheads voice activity=0.32, 2-state Markov - Bundling of frames at source(1-2) - Header size compression to 3 bytes (1 ROHC base header+ 2UDP checksum) Cellular layout - Hexagonal grid, 19 sites, 3 sectors (NB-to-NB 1km) - Carrier frequency 1.9GHz Propagation loss - Path loss= *log(r) Shadowing model - Log Normal Std. dev. 8dB UE speed - 3km/h(50%)+120km/h(50%) Antenna gain - Node B 14dB / UE 0dB - Other loss -10dB Fading Model - Combination Pad.A(5%) +Pad.B(45%)+Veh.B(50%) - Evaluated with 3GPP (TS ) UE Rx diversity - With or without considered Retransmission - No RLC retransmission - HARQ (the number of max. retrial = 6) CQI delay, error - 3TTI (6ms), 1% Scheduling - Modified PF scheduler Reserved - Common channel power overhead 20% - Associated DCH power 0.3% per mobile user - HS-SCCH power 9dB offset via associated DCH - The number of common channel code 10 - Associated DCH code 1 per mobile user - HS-SCCH codes are considered assume that the packet error rate (PER) due to packet loss and packet delay exceeding the target budget is kept within 2%. Moreover, at least 97% of VoIP users in the downlink should meet the above criterion. C. System-level simulation setup To investigate the comprehensive performance of VoIP over HSDPA, a system-level computer simulation is accomplished in this paper. In the following, we will describe the simulation environment and parameter setting. The simulations are carried out with a regular hexagonal 19 cellular model. Mobile terminals should be uniformly distributed on the 19- cell layout for each simulation run and assigned different channel models according to the channel model assignment probability specified in [16]. Note that a realistic model of the wave propagation plays an important role for the significance of the simulation results. One common approach is to use deterministic propagation scenarios (e.g., the number of paths) [17]. Shadowing is modelled by a log-normal fading of the total received power and a basic attenuation is determined by the Hata model. Finally, it has to be mentioned that for the system investigations we simulate 100,000 TTI snapshots in average. Moreover, we reserved the resources for control and common channel overhead factors such as OVSF codes and HSDPA power to obtain the provided simulation results. As mentioned before, we applied the RTP/UDP/IP packet header compression using ROHC protocol by IETF RFC 3059 that the total size of all compressed header with 3 bytes (1byte ROHC base header + 2bytes UDP checksum). The main simulation Percentage of UEs with BLER<2% w/ UE Rx Diversity 5 w/o UE Rx Diversity 60UEs 70UEs 80UEs 5 100UEs 110UEs Scheduler Delay Budget [msec] Fig. 2. Outage versus Delay budget for different UEs w/ or w/o UE receiver diversity Percentage of UEs with BLER<2% UEs w/o UE Rx Diversity 100UEs w/ UE Rx Diversity Fig. 3. w/o FB Scheduler Delay Budget [msec] Outage versus Delay budget for different value of FB parameters are summarized in Tables 2. IV. CAPACITY EVALUATION In this section, we evaluate the capacity of VoIP with the effect of FB in the combination of various fading channel environments (Pad.A 5% + Pad.B 45% + Veh.B 50%). AS above described, capacity is defined as the number of UEs satisfying above outage condition of all UEs, this is more than 97%. If an UE s combined PER is more than 2%, the user is considered in outage. Therefore, the various results are investigated by specific performance parameter such as the percentage of UEs satisfying outage limitation. Figure 2 shows the outage performance as a function of scheduler delay budget for different number of user when there is no FB (FB0), with or without UE receiver diversity. From the figure, we observe that the employment of UE Rx diversity result in the additional capacity over the corresponding UE with single antenna system. This is because the probability of success for the initial transmission of VoIP packet becomes more increase. In Figure 3, we evaluate the effect of FB on outage performance, when the number of users is 60 in case of

4 Percentage of UEs with BLER<2% w/o UE Rx Diversity w/ UE Rx Diversity w/ FB Number of user Cumulative distribution w/o FB 60UEs w/o UE Rx Diversity 100UEs w/ UE Rx Diversity TTI Fig. 4. Outage versus Number of user for different value of FB Fig. 5. CDF of packet success time for different value of FB UE single antenna and 100 with receiver diversity. According to the results, the voice transmission with FB has a higher sensitivity to the delay budget in the scheduler. Especially, it is more intensive in constraint of small delay latency, since FB increases VoIP packet arrival interval and packet size. For example, when aiming for an identical percentage value of, in case of 100 UEs, required delay latency to achieve outage is 35ms, 65ms, and 100ms for FB0, FB1, and FB2, respectively. In summary, these results have shown that the FB can be employed in VoIP service by giving larger scheduler delay budget since reduction of voice packet transmission delay can fully compensate the FB delay. Figure 4 characterizes the outage performance as a function of the number of users when FB=0, 1, and 2. Here, 90ms maximum delay budget is assumed. We note from the figure that for FB=1, the outage performance is little degraded as the number of users increases. By contrast, for FB=2, it did a great deal of performance degradation to outage. In Figure 5, the CDF curves of average success time for VoIP packet are plotted as a function of TTI when the number of users is considered as 60 and 100, without and with UE receiver diversity, respectively. It is also shown that for high FB value, transmission delay suffer from the penalty of FB delay. Finally, table 3 summarizes the capacity of VoIP in various propagation conditions for multi-path fading environments on HSDPA [16], [17]. The results confirm that VoIP over HSDPA provides significantly higher capacity if compared to both VoIP and CS voice traffic on WCDMA (Release 99) [1], [8]. V. CONCLUSION We have examined system capacity in the context of FB for VoIP over HSDPA. Our simulation results show that the outage performance is oversensitive to delay budget by employing FB that is prepared to utilize HSDPA peak data rate effeciently and to reduce the effect of relatively large headers. However, HSDPA is attractive for provision of VoIP service as compared to WCDMA (Release 99) by setting limits to a few number of FB. The consideration of the combination of other traffic TABLE III CAPACITY OF VOIP OVER HSDPA Capacity for VoIP Ped.A Ped.B Veh.B Combination w/o UE Rx w/ FB Diversity w/ UE Rx w/ FB Diversity Remark 3km/h 3km/h 120km/h Ped.A 5% +Ped.B 45% +Veh.B 50% types such as FTP and streaming data in the system may be an interesting issue for future study. REFERENCES [1] H. Holma and A. Toskala, WCDMA for UMTS, John Wiley and Sons, Third Edition, 2004 [2] J. Peisa and E. Englund, TCP performance over HS-DSCH, IEEE Proc. VTC 2002 Spring, 2002 [3] M. Lundevall, B. Olin, J. Olsson, J. Eriksson and F. Eng, Streaming Applications over HSDPA in Mixed Service Scenarios, IEEE Proc. VTC2004, Fall, [4] B. Douskalis, IP Telephone, Englewood Cliffs, NJ: Prentice-Hall, 2000 [5] M. Veeraraghavan, N. Cocker and T. Moors, Support of Voice Services in IEEE WLANs, IEEE Proc. INFOCOM, 2001 [6] W. Wang, S.C. Liew and Victor O.K. Li, Solutions to Performance Problems in VoIP Over a Wireless LAN, IEEE Trans. Veh. Technol., vol.54, no.1, pp , January 2005 [7] Q. Bi, P.C. Chen, Y. Yang and Q. Zhang, An Analysis of VoIP Service Using 1xEV-DO Revision A System, IEEE JSAC, vol.24, no.1, pp.36-45, January 2006 [8] B. Wang, K.I. Pedersen, T.E. Kolding and P.E. Mogensen, Performance of VoIP on HSDPA, IEEE Proc. VTC 2005, Spring, 2005 [9] 3GPP TR , Feasibility Study for Enhanced Uplink for UTRA FDD [10] 3GPP TS26.236, Packet switched conversational multimedia application; Transport protocols [11] O. Komolafe and R. Gardner, Aggregation of VoIP Streams in a 3G Mobile Network: A Teletraffic Perspective, Proc. EPMCC, 2003 [12] IETF RFC 3059, Attribute List Extension for the Service Location Protocol, February 2001 [13] P. Viswanath, D. Tse and R. Laroia, Opportunistic beamforming using dumb antennas, IEEE Trans. On Inform. Theory, vol.48, no.6, pp , June 2002

5 [14] 3GPP TS25.308, UTRAN High Speed Downlink Packet Access: Overall Description [15] ITU-T G.114, Transmission systems and media - General characteristics of international telephone connections and international telephone circuits [16] 3GPP TS25.101, User Equipment radio transmission and reception (FDD) [17] ITU-R M.1225, Guidelines for evaluation of radio transmission technologies for IMT-2000