Performance of a Mixed-Traffic CDMA2000 Wireless Network with Scalable Streaming Video

Transcription

1 Submitted to IEEE Transactions on Circuits and Systems for Video Technology Performance of a Mixed-Traffic CDMA2000 Wireless Network with Scalable Streaming Video Joe Huang, Member, IEEE, Yuqi Yao, Member, IEEE, Yong Bai, Member, IEEE, and Szu-Wei Wang Member, IEEE, PacketVideo Corporation Abstract - CDMA2000 1X, one of the world s first commercial third generation (3G) systems, offers high-speed data services in conjunction with the voice service. This high data rate capability of CDMA2000 1X enables delivery of real-time streaming video. To facilitate efficient network planning for wireless multimedia services, it is important to have a solid understanding of the feasibility and impact of delivering streaming video over a CDMA2000 1X network in the presence of voice and data traffic. This paper evaluates video streaming throughputs and wireless system capacities for a CDMA2000 1X network under mixed traffic conditions (including voice, web-browsing and streaming video traffic) through a multi-cell multi-user system-level event-driven simulation. We will also demonstrate the advantages of incorporating scalability and dynamic rate control (DRC) in wireless video streaming services. Index Terms Wireless Multimedia, CDMA, Video Streaming, System-Level Simulation I. INTRODUCTION Wireless multimedia communication allows a user to communicate from a mobile location in multiple formats, e.g., voice/audio, data, image, and full motion video. Since multimedia communication is envisioned to be a significant component of future wireless communication services, various two-and-half generation (2.5G) and third generation (3G) wireless standards and technologies such as GPRS (General Packet Radio Service), CDMA IS-95B, CDMA2000 and W-CDMA have been designed with the capability of providing high speed data services, ranging from more than 100 kbps to several Mbps. Among them, CDMA2000 1X is the first commercialized system based on the CDMA2000 standards developed by the global standards body 3rd Generation Partnership Project 2 (3GPP2). The system provides simultaneous voice and high-speed packet data services up to kbps [1-5]. The high data rate capability offered by CDMA2000 1X network opens the opportunity for rich multimedia services. In 3GPP2, the functional characteristics and requirements of video streaming and video conference services have been defined [6,7,8]. This work has been submitted to the IEEE for possible publication. after which this version will be superseded. Copyright may be transferred without notice,

2 Real-time multimedia streaming enables a user to view and/or listen to rich multimedia content soon after the end user begins receiving the streaming data, without having to download the entire multimedia file. However, transmission of real-time multimedia streams in wireless networks still faces several challenges. Firstly, radio transmission over a wireless channel is highly prone to errors due to multi-path effects, shadowing, and interference. Link layer retransmissions that are commonly used in wireless communication systems to correct the corrupted data can result in high transmission delay and jitter. Secondly, the wireless channel bandwidth can vary significantly over time. The reason is that the amount of bandwidth that is assigned to a user can be a function of the signal strength and interference level that such user receives since more processing gain or heavier channel coding is needed to protect the data under low signal strength or high interference conditions. As a user travels through different parts of the cell with varying signal strengths due to radio wave propagation path loss and fading, different bandwidths may be dynamically assigned to the user. Thirdly, depending on the quality of service (QoS) capability of the wireless network, multi-user sharing of the wireless channel with heterogeneous data types can also lead to significant user channel bandwidth variation. This bandwidth variation can further lead to overflow of network buffer and hence packet loss. Finally, data transmission can be interrupted completely depending on wireless network implementation, e.g., cell reselection/handoff process, resulting in transmission gaps ranging from a fraction of a second to several seconds. This unpredictability of available wireless channel bandwidth introduces high delay jitter for the multimedia streaming data. Quality of Service (QoS) control in wireless networks can help alleviate the bandwidth variation and packet delay/loss problems [9,10]. However, to guarantee a QOS in a wireless network is often costly because the capacity of a wireless network is related to the uncontrollable RF condition. For example, in a CDMA system, the required power to serve a requested bandwidth varies significantly with the user s RF condition. To maintain a reasonable data rate for a user near the cell boundary, a large proportion of power from the base station (BS) needs to be assigned, limiting the capability of the BS to serve other users. Therefore, a QoS-enabled wireless network still could not provide strict performance warranty. In a limited QoS provisioning environment, some video standards like MPEG-4 have been proposed to have features suitable for wireless transmissions. The MPEG-4 standards consist of a set of tools providing improved compression efficiency and error resilience. In addition, MPEG-4 provides scalability for both spatial and temporal resolution enhancements that are implemented using multiple encoding layers, i.e., base-layer and Submitted to IEEE Trans. Circuits Syst. Video Technol. 2

3 enhancement layers. For scalable video streaming, dynamic rate control (DRC) [11-12] can be implemented to better utilize the available network resources by allowing the media rate and quality to be adjusted over a wide range. As more bandwidth becomes available, more of the bitstream can be delivered, resulting in a higher quality multimedia presentation. Even though a variety of DRC algorithms have been proposed for wireline Internet video streaming [13-15], novel algorithms are still being developed to improve video streaming performance over a wireless network [16]. Recognizing the importance of wireless video streaming service, a near real-time (i.e., streaming) video model has been defined by 3GPP2 as part of the overall data traffic model in the performance evaluation methodology document [17] for CDMA2000 1xEV-DV (1X Evolution-Data and Voice), the newest CDMA2000 standards. However, the video model defined in [17] pertains only to non-scalable video without DRC. We will demonstrate through this work that scalability and DRC are critical for a successful deployment of wireless video streaming services. Future multimedia services will likely coexist with the basic voice and relatively delay insensitive data services (such as web browsing and ) for most wireless systems [18,19]. To facilitate efficient network planning and business strategy development for future wireless systems, it is necessary to understand overall wireless system performance (such as capacity) when multiple types of traffic, each with distinct characteristic, are present in the same cell/sector. It has been pointed out [20] that while the analysis of CDMA reverse link capacity can be modeled elegantly [21], forward link analysis is often too complex for analytical modeling and thus system level simulations are required [22]. In general, there are two approaches for forward link simulation, i.e., Monte Carlo simulation and event-driven simulation. Monte Carlo simulation examines the system performance through the study of statistical distributions of the relevant performance measures. Event-driven simulation, on the other hand, reveals the interaction between different internal dynamics (usually of different time scale) of a complex system. Monte Carlo simulation has been the predominant approach for CDMA voice capacity studies. However, when system internal dynamics affects the source characteristics (such as web browsing and scalable video streaming), an event-driven simulation becomes necessary. To the best of our knowledge, a full-scale wireless system event-driven simulation incorporating scalable streaming video, voice, and web browsing traffic has not been reported in the literature. In our work, a multi-cell, multi-user, system-level event-driven simulation is developed to study the video streaming performance and wireless system performance Submitted to IEEE Trans. Circuits Syst. Video Technol. 3

4 in a mixed traffic CDMA2000 1X network. Specifically, the impact of video traffic loading on average video throughput under a specific set of voice and (web browsing) data loading conditions will be presented. The impact of introducing data services (including video streaming and web browsing) on voice capacity of the system will also be addressed. The reminder of this paper is organized as follows. In Section II, we describe the protocols involved in video streaming over CDMA2000 1X networks. In Section III, system-level simulation methodology is described. Performance results are presented in Section IV. Finally, in Section V, conclusions and discussions of this paper are provided. II. PROTOCOL OVERVIEW FOR STREAMING VIDEO OVER CDMA2000 1X NETWORK Future wireless multimedia applications will likely work over an open, layered, Internet-style network with a wired backbone and wireless extensions. IETF (Internet Engineering Task Force) has recommended RTP (Real-Time Transport Protocol) and RTSP (Real-Time Streaming Protocol) based protocols for the delivery of video bitstreams with synchronized audio from a server to a terminal. The same protocols have also been adopted by wireless standards bodies (3GPP and 3GPP2) for video streaming services over wireless networks. Compressed audio and video can be transported, multiplexed, and synchronized by using the services provided by the RTP/UDP/IP stack. RTSP, a session-oriented protocol, can work over TCP to control the server and allow tracking of the streaming session status as video is being served. The RTP/RTSP protocol stack we employ in the simulation is shown in Figure 1. For an end-to-end streaming video over a CDMA2000 1X network, the Internet streaming protocol will work in conjunction with CDMA2000 1X network protocols. Figure 2 shows the CDMA2000 1X layer structure [1]. At the most basic level, CDMA2000 1X provides protocols and services that correspond to the bottom two layers of the ISO/OSI Reference Model (i.e., Layer 1 - the Physical Layer, and Layer 2 - the Data Link Layer). Layer 2 is further subdivided into the Link Access Control (LAC) sub-layer and the Medium Access Control (MAC) sub-layer. Applications and upper layer protocols, corresponding to OSI Layers 3 through 7, utilize the services provided by the CDMA2000 1X LAC services. The link access control (LAC) sub-layer performs the functions essential to set up and maintain a logical link connection. The MAC sub-layer coordinates the resources offered by the physical layer. A main task of MAC is to share the transmission media with different users in a fair and efficient manner. The CDMA2000 Submitted to IEEE Trans. Circuits Syst. Video Technol. 4

5 system includes a flexible and efficient MAC entity that provides two important functions, i.e., best effort delivery, and multiplexing and QoS control. The best effort delivery offers reasonably reliable transmission over the radio link with a Radio Link Protocol (RLP) that provides a best effort level of reliability. While the multiplexing and QoS control provides the enforcement of negotiated QoS levels by mediating conflicting requests from competing services and by the appropriate prioritization of access requests. The MAC layer in our simulation, described in Section III, supports call admission control, burst admission control, rate assignment, and priority control for mixed traffic. The physical layer performs coding, modulation, and spreading for the physical channels. There are two types of logical channels in the current packet data implementations: Fundamental CHannel (FCH) and Supplemental CHannel (SCH) on both reverse and forward links. FCH maintains the physical layer connection and carries both signaling and a portion of packet data traffic. SCH is allocated dynamically based on the traffic demand and carries high-speed packet data. The range of SCH data rate depends on the specified Radio Configuration (RC). There are 5 different radio configurations (RC1 through RC5) specified on the forward link and 4 RCs on the reverse link. For one-way video streaming applications, the forward link data rates are of most interest. Currently, RC3 has been implemented by infrastructure and handset suppliers [23], which includes a 9.6 kbps FCH. The available data rates of SCH are 9.6, 19.2, 38.4, 76.8, and kbps. Therefore, the peak data rate for a data user can be as high as kbps (1 SCH + 1 FCH) in the physical layer. The overall user throughput is expected to be bursty due to the dynamically assigned SCH data rate. The practical SCH data rate allocated for a user at a given instant is determined by the network using a vendor-specific algorithm. situation, the available SCH resources may be shared by all high-speed packet data users. In a multi-data-user The SCH resource scheduler of the wireless network will have full control over when and what rate a SCH will be assigned to a packet data user, if the FCH rate is not enough to support the user data traffic. III. SIMULATION METHODOLOGY A. Air Interface: We consider 19 3-sector cells located on a hexagonal regular grid. The antenna pattern associated with each sector has a 90-degree 3-dB beam width oriented in a clover-leaf pattern. The maximum cell site transmit power is assumed to be 20 Watts (16 Watts can be used to support traffic channels while 4 Watts are reserved for Submitted to IEEE Trans. Circuits Syst. Video Technol. 5

6 Pilot, Paging and Sync channels). For the path loss propagation model, we use the model recommended by [24], which leads to a cell size around 3.4 km based on a voice link budget [25]. The standard deviation of the shadow fading is assumed to be 8 db, the shadow fading correlation length is 200 meters, and the base station to base station correlation is assumed to be 0. 5 [20]. Mobiles move at a pedestrian speed for data users and at a mixture of pedestrian and mobile speeds for voice users. The fraction of total traffic power allocated for the mobile, which is a function of geometry (i.e., the total received power spectral density from desired base station divided by the total received power spectral density from all other base stations plus the thermal noise density) and soft handoff condition, is determined from link-level simulation data [26] (without transmit diversity). The FCH handoff procedure between cells (soft handoff) and between sectors (softer handoff) is based on the dynamic threshold defined in CDMA2000 1X [2]. No soft/softer handoff is considered for the SCH; only the base stations with strongest pilot power at the mobile terminal can transmit on the supplemental channel. This assumption is consistent with many of the CDMA2000 1X system implementations. Lastly, to save simulation time, both handoff decision and perfect power control adjustment are made once a second. B. Traffic Model: The voice traffic arrival is assumed to obey a Poisson process and the call holding time has an exponential distribution with a mean value of 90 seconds. For non-real time data traffic, we consider the webbrowsing model recommended by [27]. The model assumes that the session arrival also obeys a Poisson process. The number of packet calls (web pages) per session is geometrically distributed with mean N pc ; the reading time between two consecutive packet call requests in a session is geometrically distributed with mean D pc ; the number of packets in a packet call is geometrically distributed with mean N d ; the time interval between two consecutive packets inside a packet call is geometrically distributed with mean D d ; Packet size S d follows a truncated Pareto distribution with a lower bound k 1, upper bound k 2 (due to IP packet size limitation) and heavy-tailness parameter α. The parameter values for the considered web browsing traffic model are summarized in Table I. For video traffic, we use a simple model in order to incorporate scalability without incurring too much complexity into the system-level simulation. Similar to the voice traffic, video session arrival is assumed to obey the Poisson process, and the clip duration for the video session has a shifted exponential distribution with a minimum and mean clip durations of 5 and 60 seconds respectively. Both video and audio packets are assumed to have a fixed Submitted to IEEE Trans. Circuits Syst. Video Technol. 6

7 size (here we assume 800 bytes for video packet and 200 bytes for audio packets). For non-scalable video, the streaming rate is assumed to be equal to the total video plus audio encoding rate, while for scalable video, the streaming rate can vary within the scalable encoding rate range to adapt to wireless channel variations. We assume that the scalable video is encoded from kbps to kbps and the non-scalable video is encoded at kbps (both including 4.75 kbps audio). For the simulation, we adopt a rate control algorithm similar to those proposed in [14,15]. The video/audio packet inter-arrival time is therefore the video/audio packet size divided by the video/audio streaming rate for both scalable video and non-scalable video. C. Medium Access Control: The goal of the forward link Medium Access Control (MAC) is to improve the network resource utilization efficiency while maintaining fairness among users. Since quality of service (QoS) for different service classes are still being defined by 3GPP2 [9], current deployment of CDMA2000 1X network is unlikely to support QoS among data users. Therefore, we assume that there is no service differentiation between different types of data traffic. However, we do assume that circuit-switched voice traffic has priority over packet-switched data traffic as voice traffic is still regarded as the basic service that must be supported. Moreover, we assume that each data user will be assigned a queue in the wireless core network and the maximum queue size for each user is 100 kbytes. 1) Call admission: In the simulation, a voice call will not be admitted to the system if the current total traffic power consumed by active FCHs plus the power needed to support the incoming voice user exceed the upper power limit. A data user will not be assigned an FCH if the current total traffic power consumed by active FCHs plus the FCH power that would be needed to support data users with non-empty queues but currently are not assigned data bursts (due to potential overload) plus the power needed to support an FCH for the incoming data user exceeds the upper power limit. Note that the data FCH assignment criteria is slightly more stringent than the voice call admission criteria under power overload condition because we assume that voice users have priority in terms of power assignment in the event of potential power overload. In the simulation, we assume that if a data user (web or video) is denied a FCH channel assignment to prevent overload, the user will have to retry some time later (blocked call cleared). Submitted to IEEE Trans. Circuits Syst. Video Technol. 7

8 2) Burst allocation: In our simulation, we assume that data bursts are assigned synchronously with a fixed burst duration (T burst ) of 1 second. When the data queue is depleted, FCH transmitting at 1/8 rate will still be assigned to that data user until an inactivity timer expires. This is to prevent frequent FCH setups and teardowns in the presence of bursty data traffic. In addition, we assume both SCH and FCH can carry data simultaneously. Table II provides the physical layer data rates that can be assigned to a data burst. 3) Burst admission control: As long as the system is not in overload, at least a 9.6 kbps (full rate FCH) bandwidth is guaranteed for each user with data to transmit. If the total traffic power to be assigned is higher than the desired limit (potential overload) and all the non-voice users have been assigned only an FCH, the burst associated with the user requesting the highest power will be stopped first (for the current burst assignment period). The users with no bursts assigned at this time can still compete in the next period. 4) Rate assignment: For a CDMA2000 system without QoS, two basic fairness criteria are evident: equal rate and equal resource (forward link power in this case). The former tries to ensure that all users receive the same data rate, irrespective of their RF conditions, while the later ensures that all users can evenly share the power resources. The equal rate scheme may result in poorer sector throughput performance than the equal power scheme but is fairer from the user point of view. In our study, we use the equal power scheme as the initial criteria, but also take the minimum bandwidth fairness (i.e., each user will be assigned at least an FCH if the system is not in overload), the amount of backlogged data, and Walsh code limitation (64 for RC3 including overhead channels) into account. For example, the users with poor RF conditions or with a small amount of data waiting in the queue will be assigned a low rate SCHs. In the event when total requested power reaches the power limit, the SCH rate associated with the user requesting highest power will be reduced (to the next lower rate) first. On the other hand, In the event when the Walsh code resource becomes the bottleneck, the SCH rate associated with the user assigned the highest rate (hence requesting the highest number of Walsh codes) will be reduced (to the next lower rate) first. D. RLP Retransmission: Due to RLP retransmission, the user throughput is reduced. frame error rate (FER) associated with each SCH/FCH is maintained. For simplicity, we assume that the target In our simulation, we assume that FER is equal to 1% for both FCH and 9.6 kbps SCH. Higher rate SCHs have an FER of 5%. Note that the in a CDMA2000 1X system, power control, interleaving and channel coding are employed to de-correlate the RLP Submitted to IEEE Trans. Circuits Syst. Video Technol. 8

9 frame errors. Therefore, under nominal operating conditions, the frame errors should occur almost randomly in a fading environment. We thus take into account the user throughput reduction by a factor of (1+FER+FER 2 +FER 3 ) for each burst, assuming a maximum of 3 retransmissions per frame. E. Overhead Header compression is not assumed. The overhead for RTP, UDP, IP, PPP/LAC layers are assumed to be 12, 8, 20, and 6 bytes respectively. Moreover, we assume that 20 bits of RLP overhead and 20 bits of Physical layer overhead are needed for every 152 bits of RLP payload. IV. SIMULATION RESULTS In this section, we present the simulation results regarding video streaming performance (including throughput at application layer and packet loss probability) and wireless system performance (including sector throughput at physical layer, soft-blocking probability for voice users, and the probability of FCH assignment denial for data users) for different mixed traffic conditions. To characterize the system performance, we first study the CDMA2000 1X system performance for each individual traffic type, namely, voice, web, and video. The results for mixed traffic scenarios will then be presented. For each data point, the simulated time is 5 hours. A. Voice Traffic Only In Figure 3, we plot the voice soft blocking probability as a function of voice Erlang loading. By soft blocking we mean that the blocking is the result of potential power overload instead of the unavailability of the voice trunk. Since the maximum number of users allowed into the system depends on the locations and RF conditions of existing users and the loading of neighboring interfering cells, the term soft blocking is used [21]. This is opposed to hard blocking where the maximum number of supported simultaneous users in the system is fixed by the number of available voice trunks. The result in Figure 3 shows that for a 2% target soft-blocking probability, the voice capacity of the simulated CDMA2000 1X system is around 23 Erlangs, which is consistent with that estimated in [28]. B. Web Browsing Traffic Only Next, we consider the performance results for web browsing data traffic. The sector (physical layer) throughput is plotted in Figure 4. It is seen that the sector throughput increases almost linearly with web session Submitted to IEEE Trans. Circuits Syst. Video Technol. 9

10 arrival rate but the throughput increase becomes much slower when the web session arrival rate reaches 14 sessions per minute. Moreover, in Figure 5, we plot the probability of denying FCH assignment upon web session arrival as a function of web session arrival rate. To maintain a low probability of denying FCH (e.g. 2%), the system allows roughly 12 web session arrivals per minute. The corresponding sector (physical layer) throughput is around 105 kbps. The observed phenomenon is discussed as follows. Basically, the sector throughput goes through a phase transition (from sporadic to continuous) when the offered load is between 12 ~ 14 sessions per minute. Because the offered web browsing traffic load is very bursty, the sector throughput is also relatively bursty when the offered load is small (less than 12 sessions per minute). On the other hand, when the offered load is above 14 sessions per minute, the sector throughput bursts coalesce to form a continuous throughput. In the latter stage, channels are efficiently utilized, however, a new user will have be more difficult to find their way into the system due to admission control, resulting in a sharp increase in the probability of denying FCH assignment and the slowing down of sector throughput increases. If the offered load continues to increase the sector throughput will eventually saturate. At the saturation stage, probability of denying FCH assignment will be very high and burst admission control will constantly kick in to withhold data burst for transmission. This will likely slow down the response time of existing web browsing sessions. C. Streaming Video Traffic Only For streaming video, the data flow is naturally more continuous than that for web browsing, thus, the system resources can be continuously utilized. As seen in Figure 6, the sector (physical layer) throughput for streaming video can reach beyond 210 kbps for the scalable case, essentially double that for the web browsing. For the non-scalable case, the throughput is lower, roughly 160 kbps. The sector throughput for scalable video is higher than that of non-scalable video because more enhancement layer information can be sent to the player when network resources are available. By the same token, scalability also leads to a higher average (application layer) streaming throughput, as seen in Figure 7 is bounded by the maximum encoding rate). (Note that the average application layer per-session video throughput Moreover, one can observe in Figure 6 that the sector throughputs for both scalable video and non-scalable video increase initially with loading but begin to drop after the offered load exceed 5 and 6 sessions per minute, respectively. The reason is that in a CDMA 1X system, it is more efficient (in terms of power usage) to transmit the same amount of data in a high rate channel than in multiple low rate channels. When the simultaneous active sessions in the system is small, high rate SCH channels (e.g., Submitted to IEEE Trans. Circuits Syst. Video Technol. 10

11 or 76.8 kbps) still have a reasonable chance to be assigned. On the other hand, when the loading is high, most of the active sessions would be assigned a lower SCH rate or no SCH channels, resulting in a lower sector throughput due to lower power utilization efficiency. Similar to the web browsing traffic, the throughput curves will eventually reach a steady state at a high offered load, in which call admission and burst admission control will be constantly involved to regulate the system traffic. In Figure 8, we plot the packet loss percentage as a function of video session arrival rate. The packet loss presented in this Figure is caused by the overflow of wireless network buffer rather than over-the air packet error. It is apparent that packet loss due to network buffer overflow can be reduced significantly by using scalable video because the scalability feature allows the server to adjust send rate in response to the wireless system rate assignment changes, thereby minimizing the occurrence of network buffer overflow and player buffer underflow. The probability of denying FCH assignment for an incoming video session as a function of video session arrival rate is plotted in Figure 9. One can readily see that using scalable video also reduces the probability of denying FCH assignment. This is explained as follows: When the average throughput for non-scalable video falls below the encoding rate (in this case, kbps), the transmission time for a non-scalable clip is often longer than the clip duration. On the other hand, for scalable video, the transmission time for a clip is closer to the clip duration due to the dynamic server rate control. As a result, a non-scalable streaming session tends to stay in the wireless system longer than a scalable one. The average number of sessions in the wireless system with nonscalable video is therefore larger than that with scalable video (for the same session arrival rate), reducing the chance of access for a new session. Assuming again that 2% probability of denying FCH assignment is a reasonable operating point, the maximum session arrival rate for scalable video is around 6 sessions per minute. Note that although the concept of Erlang does not strictly apply to data traffic, we observed in our simulation that the average session arrival rate times the average clip duration (assumed here to be one minute, which is roughly equal to the service time for scalable video) matched reasonably well with the average number of scalable video streaming sessions in the system. Therefore, for scalable video the streaming capacity of the simulated CDMA2000 1X system can also be termed as 6 Erlangs. The corresponding sector (physical layer) throughput, average per-session (application layer) throughput, and packet loss probability are 215 kbps, 28 kbps, and 1.5%, respectively. The visual artifact Submitted to IEEE Trans. Circuits Syst. Video Technol. 11

12 associated with this low packet loss rate can be alleviated through error resilience and error concealment techniques and/or RTP retransmission. D. Mixed Voice, Web Browsing, and Streaming Video Traffic Based on the results obtained so far, we shall assume that, in the simulation, the system is fully loaded if the voice loading reaches 23 Erlangs, web session arrival rate reaches 12 sessions per minute, or video session arrival rate reaches 6 sessions per minute. In the following, we will study the wireless system and video streaming performance under mixed traffic conditions. Only scalable video will be considered. In Figure 10, we plot the average per-session (application layer) throughput as a function of video loading percentage (relative to an arrival rate of 6 sessions per minute) for a video-voice mixed system and for a video-web mixed system. The corresponding voice/web loading (relative to 23 Erlangs/12 web sessions per minute arrival) is 100% minus the video loading percentage so that the system maintains fully loaded. It is seen that in a fully loaded system, the video throughput is higher with low video - high voice/web traffic mixture than that with high video - low voice/web traffic mixture. The reason is that (scalable) video traffic would always try to make the most out of the available power resources to get higher rate and transmit more data using DRC. Normally, there will be periods of low power usage and high power usage if only voice traffic is present due to the random nature of voice arrival, call hold time and voice activity. Similarly, if there is only web traffic in the system, the power usage will also have high and low periods. Therefore, when voice or web traffic dominates, scalable video traffic can seek to increase their transmission rate (by sending more enhancemet layer information) during the period when the power usage is low so that the system resource is efficiently utilized. As a result, the average video throughput is higher than the case when video traffic dominates. On the other hand, the opposite is true for the sector throughput. That is, the sector throughput is higher (even though average per-session throughput is lower) with high video - low web traffic mixture than that with low video - high web traffic mixture due to the presence of more (scalable) video sessions,. This behavior can be readily predicted from the Figure 4 and Figure 6. To study how voice capacity is impacted by the introduction of data services, the sector (physical layer) throughput for different voice - video - web traffic mixtures, assuming a total loading of 100% is plotted in Figure 11. Here, the total sector throughput is plotted as a function of voice loading percentage assuming that the web and video loading percentages are equal. Trade-off between voice capacity and data throughput for different Submitted to IEEE Trans. Circuits Syst. Video Technol. 12

13 web-video traffic mixtures is clearly demonstrated. The convex nature of the mixed traffic sector throughput curves comes from the multiplexing gain. Multiplexing gain exists because the unserved data (piling up during the time when voice traffic is high) can stay in a queue to be transmitted when voice traffic is low, resulting in a more efficient utilization of the system resources. When scalable video is involved, channel utilization can be further improved because of the source rate adaptation to bandwidth availability. Therefore, as can be seen in Figure 11, a system deploying scalable video streaming service in addition to the web service achieves higher spectrum efficiency (kbps/mhz) than that of deploying web service alone due to the efficient utilization of the power resource. One final remark, for the mixed traffic analysis, we have assumed that there is no service differentiation between different types of data traffic in the considered CDMA2000 1X system because QoS is not expected to be deployed in the near future. Specifically, for call admission control, we have assumed that a new data (web or video) session that is denied an FCH assignment will have to retry later. Although such a block call cleared approach is suitable for video streaming service (to avoid long video starting time), it may be advantageous in terms of system resource utilization to admit web traffic into the system queue even when an FCH assignment is denied. To this end, our simulation results indicate that the amount of sector throughput increase is not significant for a wait-time (the amount of time a web browsing user is willing to wait for the first page before quitting) as long as one minute. For burst admission control, we have assumed that an existing data (web or video) session would give up SCHs (but not FCHs) whenever necessary to let the system serve new voice calls. Alternatively, if we force the existing active data sessions to give up FCHs whenever necessary to serve new voice calls, the multiplexing gain for mixed voice-data traffic can be made larger and the data sector throughput will not be zero even when voice traffic is fully loaded (unlike the result presented in Figure 11). If the burst admission control mechanism can be designed such that an FCH assignment is guaranteed for video traffic (when the system is not in overload) but not for web traffic, higher sector throughput for web traffic can be obtained and minimum bandwidth allocation for video streaming services can be provided. V. CONCLUSIONS AND DISCUSSIONS Streaming video has been identified as part of the CDMA2000 3G wireless data services. Understanding how the introduction of video/web services affects voice capacity will facilitate the planning and Submitted to IEEE Trans. Circuits Syst. Video Technol. 13

14 optimization of 3G wireless networks. We have investigated both wireless system performance and video streaming performance under mixed traffic condition using an event-driven system-level simulation model. Based on the 2% soft blocking/fch denial criteria, the reference voice capacity for the considered system is 23 Erlangs and the capacity of scalable streaming video is found to be 6 Erlangs. The tradeoff between wireless system voice capacity and data throughput for different web-video traffic loading mixtures have been shown in Figure 11. Mixed traffic multiplexing gain is demonstrated by convex nature of the throughput curves. Moreover, with the introduction of scalable video traffic, higher spectrum efficiency (kbps/mhz) can be achieved than that of web only traffic due to the dynamic (source) rate control feature of scalable video. For video streaming performance, simulation results show that a CDMA2000 1X system can support an average per-session (application layer) video throughput around 30 ~ 40 kbps under a reasonably loaded condition. The average per-session video throughput is higher with lower video loading percentage, while the sector throughput is higher with higher video loading percentage due to the presence of more (scalable) video sessions. 3GPP2 currently defines a video traffic model for performance evaluation; however, only non-scalable video is considered. This study demonstrates the importance of scalability and DRC for wireless video streaming. It has been found that both sector throughput and per-session throughput can be improved through the use of scalability and DRC. Moreover, the packet loss rate for scalable video streaming can be much lower than that for non-scalable video streaming. Another important benefit of scalability, although not shown explicitly, is that source rate adaptation can alleviate the infamous player jitter buffer underflow and re-buffering problem in a bandwidth-varying environment such as the one considered in this paper. Note that the scalability concept we adopt in this work is general. It can be in the form of temporal, spatial, SNR, or fine-granular scalability. Alternatively, bitstream switching (i.e., switching among non-scalable bitstreams of different encoding rates), widely adopted in current Internet streaming solutions, can also be used to achieve source rate adaptation. Detailed video streaming performance is sensitive to channel bandwidth variation during streaming. The degree of variation depends on the forward link MAC implementation. In this regard, properly designed admission control, scheduling and rate assignment schemes that differentiate video and other data traffic can further improve the video streaming performance (but may be at the cost of overall system capacity). The impact of different MAC and QoS implementations on video streaming performance and wireless system performance is the subject of future work. Submitted to IEEE Trans. Circuits Syst. Video Technol. 14

15 References [1] TIA/EIA/IS A, Introduction to CDMA2000 Standards for Spread Spectrum Systems, March [2] TIA/EIA/IS A, Physical Layer Standard for CDMA2000 Spread Spectrum Systems, March [3] TIA/EIA/IS A, Medium Access Control (MAC) Standard for CDMA2000 Spread Spectrum Systems, March [4] TIA/EIA/IS A, Signaling Link Access Control (LAC) Standard for CDMA2000 Spread Spectrum Systems, March [5] TIA/EIA/IS A, Upper Layer (Layer 3) Signaling Standard for CDMA2000 Spread Spectrum Systems, March [6] 3GPP2 S.R0021, Video Streaming Services-Stage 1, July [7] 3GPP2 S.R0021, Video Conference Services-Stage 1, July [8] 3GPP2 S.R0021, Multimedia Streaming Services-Stage 1, April [9] 3GPP2 S.R0035, Quality of Service-Stage 1 Requirements, October [10] N. Dimitriou, R. Tafazolli, and G. Sfikas, Quality of Service for Multimedia CDMA, IEEE Commun. Mag., pp , July [11] X. Wang and H. Schulzrinne, Comparison of Adaptive Internet Multimedia Applications, IEICE Trans. Commun., vol. E82-B, no. 6, pp , June [12] D. Wu, T. Hou, and Y. Zhang, Scalable Video Coding and Transport over Broadband Wireless Network, Proc. IEEE, vol. 89, no. 1, pp. 1-16, Jan [13] D. Sisalem, H. Schulzrinne, and F. Emanuel, The direct adjustment algorithm: A TCP-friendly adaptation scheme, Quality of Future Internet Services Workshop, Berlin, Germany, September 25-27, [14] K. Yano, H. Sato, and K. Sezaki, A rate control for continuous media transmission based on backlog estimation from end-to-end delay, Available: [15] S. Jacobs and A. Eleftheriadis, Real-time dynamic shaping and control for Internet video applications, Workshop on Multimedia Signal Processing, Princeton, NJ, June, [16] [17] 3GPP2 WG5 Evaluation AHG: 1xEV-DV Evaluation Methodology Addendum (V6), July 25, [18] D. Knisely, S. Kumar, S. Laha, and S. Nanda, Evolution of Wireless Data Services: IS-95 to CDMA2000, IEEE Commun., Mag., vol. 36, Oct [19] S. Nanda, K. Balachandrean, and S. Kumar, Adaptation Technologies in Wireless Packet Data Services, IEEE Commun. Mag., vol. 38, pp , Jan [20] A.J. Viterbi, CDMA: Principles of Spread Spectrum Communications, Addison-Wesley, [21] A. M. Viterbi and A. J. Viterbi, Erlang Capacity of a Power Controlled CDMA System, IEEE J. Select. Areas Commun., vol. 11, no. 6, pp , Aug [22] K.S. Gilhousen, I.M. Jacobs, R. Padovani, A.J. Viterbi, L.A. Werver, and C.E. Wheatley, On the Capacity of a Cellular CDMA System, IEEE Trans. On Vehicular Technology, vol. 40, no. 2, May [23] G. Li, M. Lu, M. Meyers, D. Patel, J. Stekas, and A. Tonello, Performance of Lucent CDMA2000 3G1X Packet Data Experimental System, in Proc. IEEE VTC 2001 Spring, May 2001, pp [24] RECOMMADATION ITU-R M.1225, Guideline For Evaluation of Radio Transmission Technologies for IMT-2000, Submitted to IEEE Trans. Circuits Syst. Video Technol. 15

16 [25] K. Kim, Edited, Handbook of CDMA System Design, Engineering, and Optimization, Prentice Hall Inc., [26] The cdma2000 ITU-R RTT Candidate Submission, v0.18, July [27] TR v3.2.0, Universal Mobile Telecommunications Systems (UMTS), Selection Procedures for the Choice of Radio Transmission Technologies of the UMTS, April [28] W. Hamby, CDMA2000 1X Performance, in Personal Indoor and Mobile Radio Communications (PIMRC) 2001, San Diego, October Submitted to IEEE Trans. Circuits Syst. Video Technol. 16

17 Manuscript received. The authors are with PacketVideo Corporation, 365 West Passaic Street, Fifth Floor, Rochelle Park, NJ 07662, U.S.A. Submitted to IEEE Trans. Circuits Syst. Video Technol. 17

18 List of Figures Fig. 1. Fig. 2. Fig. 3. Fig. 4. Fig. 5. Fig. 6. Only Video Streaming Protocol Stack CDMA2000 1X Layer Structure Blocking Probability vs. Erlang Load Voice Traffic Only Sector (Physical Layer) Throughput vs. Session Arrival Rate Web Traffic Only Probability of Denying FCH Assignment vs. Session Arrival Rate Web Traffic Only Sector (Physical Layer) Throughput Comparison for Scalable and Non-Scalable Video Video Traffic Fig. 7. Per-Session (Application Layer) Video Throughput Comparison for Scalable and Non-Scalable Video Video Traffic Only Fig. 8. Packet Loss Probability Comparison for Scalable and Non-Scalable Video Video Traffic Only Fig. 9. Probability of Denying FCH Assignment Comparison for Scalable and Non-Scalable Video Video Traffic Only Fig. 10. Per-Session (Application Layer) Video Throughput vs. Video Load % for Video/Voice and Video/Web Traffic Mixes at 100% Combined Load Fig. 11. Sector (Physical Layer) Throughput vs. Voice Load % for Voice/Web, Voice/Video and Voice/Web/Video Traffic Mixes at 100% Combined Load Submitted to IEEE Trans. Circuits Syst. Video Technol. 18

19 Commands/SDP / RST RTSP TCP Audio and Video RTP/RTCP UDP IP Radio Link/Data Link Physical Layer Fig. 1. Video Streaming Protocol Stack. Submitted to IEEE Trans. Circuits Syst. Video Technol. 19

20 OSI Layers 3-7 Upper Layer Signaling Data Services Voice Services LAC Sublayer OSI Layer 2 Signaling to Physical Layer Interface SRBP RLP RLP RLP MAC Sublayer Multiplexing and QoS Delivery OSI Layer 1 Physical Layer Fig. 2. CDMA2000 1X Layer Structure. Submitted to IEEE Trans. Circuits Syst. Video Technol. 20

21 6 5 Soft Blocking Probability (%) Requirement Voice Load (Erlang) Fig. 3. Soft Blocking Probability vs. Erlang Load Voice Traffic Only. Submitted to IEEE Trans. Circuits Syst. Video Technol. 21

22 Average Sector Throughput (kbps) Web Session Arrival Rate (Sessions/Minute) Fig. 4. Sector (Physical Layer) Throughput vs. Session Arrival Rate Web Traffic Only. Submitted to IEEE Trans. Circuits Syst. Video Technol. 22

23 7 Probability of Denying FCH Assignment (%) Requirement Web Session Arrival Rate (Sessions/Minute) Fig. 5. Probability of Denying FCH Assignment vs. Session Arrival Rate Web Traffic Only. Submitted to IEEE Trans. Circuits Syst. Video Technol. 23

24 Average Sector Throughput (kbps) non-scalable scalable Video Session Arrival Rate (Sessions per Minute) Fig. 6. Sector (Physical Layer) Throughput Comparison for Scalable and Non-Scalable Video Video Traffic Only. Submitted to IEEE Trans. Circuits Syst. Video Technol. 24

25 Average Per-Session Video Throughput (kbps) non-scalable scalable Video Session Arrival Rate (Sessions per Minute) Fig. 7. Per-Session (Application Layer) Video Throughput Comparison for Scalable and Non-Scalable Video Video Traffic Only. Submitted to IEEE Trans. Circuits Syst. Video Technol. 25

26 50 45 non-scalable scalable 40 Packet Loss Probability (%) Video Session Arrival Rate (Sessions per Minute) Fig. 8. Packet Loss Probability Comparison for Scalable and Non-Scalable Video Video Traffic Only. Submitted to IEEE Trans. Circuits Syst. Video Technol. 26

27 Probability of Denying FCH Assignment (%) non-scalable scalable Requirement Video Session Arrival Rate (Sessions per Minute) Fig. 9. Probability of Denying FCH Assignment Comparison for Scalable and Non-Scalable Video Video Traffic Only. Submitted to IEEE Trans. Circuits Syst. Video Technol. 27

28 Average Per-Session Video Throughput (kbps) voice video mix web video mix Video Load % Fig. 10. Per-Session (Application Layer) Video Throughput vs. Video Load % for Video/Voice and Video/Web Traffic Mixes at 100% Combined Load. Submitted to IEEE Trans. Circuits Syst. Video Technol. 28

29 Average Sector Throughput (kbps) voice web mix voice web video mix voice video mix Voice Percentage (%) Fig. 11. Sector (Physical Layer) Throughput vs. Voice Load % for Voice/Web, Voice/Video and Voice/Web/Video Traffic Mixes at 100% Combined Load. Submitted to IEEE Trans. Circuits Syst. Video Technol. 29

30 List of Tables TABLE I. TABLE II. PARAMETER VALUES FOR WEB BROWSING TRAFFIC MODEL CDMA2000 1X FORWARD LINK PHYSICAL LAYER DATA RATES Submitted to IEEE Trans. Circuits Syst. Video Technol. 30

31 Web Model N pc D pc (sec) N d D d (sec) k 1 (bytes) k 2 (bytes) α Value TABLE I PARAMETER VALUES FOR WEB BROWSING TRAFFIC MODEL Submitted to IEEE Trans. Circuits Syst. Video Technol. 31

32 FCH (kbps) SCH (kbps) Total (kbps) TABLE II CDMA2000 1X FORWARD LINK PHYSICAL LAYER DATA RATES Submitted to IEEE Trans. Circuits Syst. Video Technol. 32