CDMA2000 1xEV-DV Reverse Link Performance in the Presence of Voice Users

Transcription

1 CDMA2 1xEV-DV Reverse Link Performance in the Presence of Voice Users Tao Wu, Patrick Hosein, Young C. Yoon, Rath Vannithamby, Shiauhe Tsai, Anthony C.K. Soong Ericsson Inc., U.S.A. {tao.wu, patrick.hosein, young.yoon, rath.vannithamby, shawn.tsai, Abstract One of the main design goals of CDMA2 1xEV-DV is the efficient support of packet data services without detriment to existent voice services. This paper investigates the performance of the 1xEV-DV RL in the presence of voice users. In particular, several algorithms are investigated to support the three MAC mechanism defined in the 1xEV-DV standard. The results show that for the common rate control mechanism, the algorithm introduces no cost to mixing voice and data at low voice loads. However with the common rate control mechanism operating with higher voice load or the dedicated rate control and grant based scheduling mechanisms, the mixing of data and voice results in modest loss in performance. The amount of loss varies and is a result of the interplay of incrementally increasing resources for voice service and the ability of the system to target higher system loads as voice load increases. Nevertheless, these results show that 1xEV-DV can efficiently support both voice and packet data traffic. Keywords - CDMA2; 1xEV-DV; Reverse Link; Mixed Data and Voice I. INTRODUCTION There has recently been increased interest in the cellular communication industry on the efficient support of high-speed packet data service. The impetus for this area of research and development is the high-speed data demand expected in applications running on 3 rd generation wireless networks. This activity has resulted in the development of several evolutionary systems. One of which is the, so-called, 1xEV-DV (1x Radio Transmission Technology Evolution for Data and Voice) systems. This is an evolution of the cdma2 system and was incrementally standardized as cdma2 Revision C (forward link (FL) enhancements) [1,2] and Revision D (FL and reverse link (RL) enhancements) [3]. These revisions introduced high-speed packet data support for the FL with a peak rate of 3.1 Mbps and the RL with a peak rate of 1.8 Mbps. Unlike some other 3G systems, 1xEV-DV maintains backward compatibility with existing networks. Although high-speed data services are expected to generate an increasingly greater percentage of system operator revenue, voice services are presently still the dominant revenue generator. Consequently, many operators, currently, view high-speed data service primarily as a means to keep existing voice customers and/or to attract new customers for the voice service. In other words, they see data services as a way of maintaining or expanding their voice revenue rather than as a revenue generator onto itself. There is, thus, a strong desire among the operator community to be able to offer high-speed packet data service without hindering the profitability or quality of the voice services. Although the ability to offer packet data service without sacrificing voice service is one of the design goals of 1xEV-DV, its performance in the presence of voice users have, to the best of the authors knowledge, not yet been addressed in the literature. Several recent papers [5,6,7] have only considered the RL performance of 1xEV-DV systems in the presence of only data users. Their results have shown that the average RL sector data throughput is roughly double that of existing cdma2 systems. In this paper the performance of the 1xEV-DV RL in the presence of voice users is investigated. The paper is organized as follows: Section II gives a brief description of the RL medium access control (MAC) mechanisms defined in the standard [3]. It should be noted that [3] defines the mobile s response to instructions from the base station, but does not define how the base station determine these instructions. Therefore, in Section III, we the base station algorithms used for each MAC mechanism investigated in this paper. Section IV discusses the performance of these algorithms in the presence of voice users. Section V contains our conclusions derived from the results. II. 1XEV-DV REVERSE LINK MAC MECHANISMS The 1xEV-DV reverse link design supports fast and flexible MAC mechanisms that are needed for various QoS requirements anticipated in the near future for data users while maintaining legacy services such as voice. The rate assignment mechanisms supported by the 1xEV-DV reverse link MAC can be categorized as autonomous, differential rate assignment, and absolute rate assignment. In the autonomous operation, the MS can transmit up to a rate configured by the base station (BS) whenever it has data to send. In differential rate assignment, the forward rate control channel (F-RCCH) carries a differential rate assignment command from the BS to adjust the transmission rate of each MS. The F-RCCH is time multiplexed and re-uses the forward common power control channel. The rate control commands, which can be UP, DOWN or HOLD, may increase, decrease or keep the current data rate, respectively. A deterministic rate control algorithm run by the MS decides the data rate for each time instance based on the commands received from the BS and the pre-negotiated rate adjustment tables stored at the MS. In the absolute rate assignment, the forward grant channel (F-GCH) is used by the BS to specify the maximum rate that the MS may use for subsequent transmissions. The F-GCH carries the grant message including the identification number of the scheduled mobile, the maximum allowed rate, and two indicators showing /5/$2. (c)25 IEEE

2 whether or not this grant message applies to all H-ARQ channels and persists for following packets. Differential rate assignment can be performed with either the common rate control (CRC) mechanism or the dedicated rate control (DRC) mechanism. In CRC, multiple users are controlled by a common rate control subchannel, while in DRC, each user is controlled by a dedicated rate control subchannel on the F-RCCH. CRC can be operated in per-sector or per-group mode, i.e., a single CRC subchannel can control all MS s in the sector or a group of MSs. DRC can be operated only in per-user mode. Though CRC is very efficient in terms of overhead (only one rate control subchannel), it cannot provide any rate guarantees needed for QoS applications. The per-group feature in CRC allows a slight enhancement with its capability of supporting user classes (e.g., Gold, Silver, etc.). Since CRC may control MS s in the entire sector and a large variation of reverse link load should be avoided, a CRC command should not change the data rates of all MS s in the same way. The mobile s reaction to a CRC command should be dependent on its present rate. For MS s at high data rates, a DOWN CRC command should be more forceful because these users contribute more to the reverse link load. For MS s at low data rates, an UP CRC command should be more effective since they utilize less reverse link resources. Different UP/DOWN step sizes corresponding to different data rates are specified in a table configured by the BS. This rate dependent reaction to control commands ensures a stable system. The value of the step size is associated with the transmitted traffic-to-pilot ratio (TPR) of the MS s current data rate. When a mobile receives a UP CRC command, instead of directly increasing its data rate by one step, it adjusts a reference value, called the authorized TPR, by the corresponding UP step size. If the reference value exceeds the transmit TPR of the next data rate, the mobile will then increase its data rate by one step and apply another step size at the next data rate. The DOWN command is processed in a similar fashion. Generally, MS s generating more reverse link interference should be powered down more forcefully. Therefore, the higher the data rate, the larger the DOWN step size but the smaller the UP step size. DRC provides a finer differential rate assignment since each MS is directly controlled by the BS. The finer control comes at the expense of increased reverse link feedback and more forward link overhead. On the reverse link, a one-bit mobile status indicator is used to tell the BS if the mobile has enough buffered data and power headroom to effectively use a higher data rate. On the forward link, the BS computes each MS s priority according to certain criteria and sends the rate command through each mobile s forward rate control subchannel. In DRC, the MS will increase or decrease its rate to the next level respectively by one UP or DOWN command, without computing a reference TPR value as done in CRC. In absolute rate assignment, the BS adjusts the rate of a MS by sending an explicit rate using a grant message on the F-GCH. This Grant Based Scheduler (GBS) provides the fastest mechanism to reach a desired rate. However, it is also the most difficult for the BS to manage since it allows MSs to make large changes in their transmission power and hence can potentially result in greater reverse link load variance. In addition, GBS requires each MS to provide more detailed report, in particular its data buffer level, its power headroom and the service reference ID that indicates the traffic type and the QoS requirement. The BS then schedules the mobile with the grant message. III. BASE STATION MAC ALGORITHMS The base station algorithms used to generate the various control commands for CRC, DRC and absolute grants for GBS are not standardized. Novel MAC algorithms for each of the three mechanisms have been developed and are presented in this section. Most studies and simulation results to date have concentrated on data-only sector populations. However, in the presence of voice traffic, the operation of the various RL control mechanisms need to be dynamically adjusted to ensure that voice traffic quality is not adversely affected. We will illustrate how this can be effectively done. In the CRC mechanism, the BS periodically broadcasts a one-bit command (representing the sector load information) to all (or a subset of) the mobiles. The BS generates and sends the CRC commands based on the following algorithm: A UP command is sent when the current reverse link load measure is lower than a specified lower threshold, and a DOWN command is sent when the current load measure is higher than a specified upper threshold. A HOLD command is sent when the current load measure lies between these two thresholds. Each mobile then reacts to this command by suitable adjustment of its rate. As the number of voice users in the system increases, the average rate of the data users decreases and hence a larger percentage of them are maintained at their autonomous value (which could actually be zero). Since fewer data users have the ability to change then the variance of their contribution to the reverse link loading will also decrease. This effect is not offset by the variance due to the increased number of voice users because voice user loading is much less variable. Therefore, as the number of voice users increases, the reverse link loading operating point used to control the data users can be increased since the lower variance implies fewer excursions beyond the outage limit. The DRC generates rate control commands based on the RL load measure, the one bit MS status indicator (MSIB) and a priority function that is based on the user channel condition and the user effective throughput. The priority function is periodically updated. The BS sets a target load that satisfies the RoT outage criteria. The BS periodically (every 1 ms) sorts the MS s based on the priority and initially assumes all mobiles decrease their rate by one level in the next frame unless that frame will be used for a re-transmission in which case the rate is not tentatively changed. The algorithm then repeats the following procedure starting with the highest priority MS. It determines the expected sector load if the mobile requested a rate increase or hold. If the estimated load is not larger than the target load, then a UP or HOLD command will be granted. Otherwise a DOWN command will be generated. This procedure is repeated for all the MS s. The GBS algorithm works in a similar manner to the DRC. The absolute grants are generated based on the RL load measure, explicit mobile requests and the priority function that is same as the one used for the DRC command generation procedure. The BS periodically sorts the MS s based on the priority and assumes all the MS will transmit at the autonomous rate if the MS is not in the middle of H-ARQ operation in which the rate will not be changed. The algorithm

3 then repeats the following procedure starting with the highest priority MS. It predicts the sector load by allowing the minimum of the highest possible achievable rate and the requested rate so that the predicated load is less than the target load, and this rate is granted to the MS. This procedure is repeated for all the MS s. As the reverse link loading increases, the average throughput per data user decreases. At some point this throughput becomes inadequate for data applications and it is best to terminate one or more data sessions in order to adequately support the remaining ones. The best candidates for removal are the ones that are power limited. We investigate how these users can be detected and terminated and the affect on sector throughput. Note that this results in a greater degree of unfairness since we are essentially setting the throughput of the terminated users to zero. The performance of these algorithms, the impact of mixing voice and packet data users, and the impact of shedding users on the sector throughput are investigated in the next section. IV. SYSTEM PERFORMANCE The performance of a cdma2 1xEV-DV system with multiple cells is too complex to be evaluated analytically. Therefore, simulations are used to demonstrate the performance of mixed data and voice service. A. Simulation Assumptions The simulation configuration and assumptions are detailed in the 1xEV-DV Evaluation Methodology [4], which was developed by 3GPP2 for CDMA2 1xEV-DV standardization. The layout consists of 19 hexagonal cells with 3 sectors in each cell, which are wrapped around so that the simulation statistics can be efficiently collected from all the cells. The data or voice mobiles are randomly dropped in this layout such that each sector has the same number of mobiles. Mobiles are randomly assigned to one of the five channel models with the specified probabilities as presented in Table 1, while the multi-path models are shown in Table 2. A mobile is either providing the voice service or the packet data service, i.e., no concurrent services are supported. Inner/outer loop power control and soft handoff are also simulated. Channel Probability (%) Table 1 Channel s Multipath No. of Fingers Speed (km/h) Fading A 3 Ped A 1 3 Rayleigh B 3 Ped B 3 1 Rayleigh C 2 Veh A 2 3 Rayleigh D 1 Ped A 1 12 Rayleigh E 1 Single Path 1, f D =1.5 Hz Rician K=1dB Table 2 Multi-path s Finger 1 Finger 2 Finger 3 Unrecovered Power Ped A Ped B Veh A For voice mobiles, the traffic is transmitted over the Reverse Fundamental Channel (R-FCH) using a 2ms frame duration. A Markov model is used to emulate an EVRC vocoder for voice traffic with an activity factor of.43, which is based on 29% full rate 9.6kbps, 4% half rate 4.8kbps, 7% quarter rate 2.4kbps, and 6% eighth rate 1.5kbps. The voice capacity is defined as the maximum number of simultaneous voice users that can be supported with an outage rate of at most 1% where outage is as defined in [4]. For packet data traffic, with a 1ms frame size, the Reverse Packet Data Channel (R-PDCH) transmits at one of the mandatory data rates 19.2, 4.8, 79.2, 156, 39.6, 463.2, 616.8, 924, or 1,231.2 kbps. For this analysis, the data mobiles are not allowed to transmit in the autonomous mode. There are 4 physical-layer H-ARQ channels and each packet is allowed maximum of 3 transmissions. Full buffer traffic is assumed, in which case each mobile always has data for transmission on the reverse channel. The performance metric of interest is the average data throughput, which is defined as the ratio of the total number of correctly received bits and the total simulation time. Several constrains must be imposed upon the simulation to ensure that the simulations accurately represent realistic systems. Firstly, in order to maintain stability and voice quality, the Rise over Thermal (RoT), which is defined as the ratio of the total received power to the thermal noise, must be at most 7 db for at least 99% of the time in order to maintain the quality of voice users. The 7dB limit was determined from experience with currently deployed CDMA systems. Secondly, a certain level of fairness must be maintained to ensure that disadvantaged users are still given service. This constraint requires that less than x% of the data users have a throughput that is less than x% of the average mobile data throughput. B. Sector Throughput with Voice Users When the system only contains voice users, the capacity was determined via simulation to be around 42 users per sector under the specified outage criterion. To investigate the impact of voice users, the mixed data and voice service is simulated with all three rate control algorithms supported in Rev. D, i.e., CRC, DRC and GBS. We fix the number of data users at 5 per sector and increase the number of voice users from zero to 4. The voice users are always treated with higher priority than the data users to guarantee their service quality. The data traffic uses the remaining resource and manages to achieve a higher packet data throughput with different rate control schemes. With the CRC algorithm, the average sector data throughput with 5 packet data users as a function of the number of voice users is shown in Figure 1. The equivalent data throughput for both data and voice users is also included, where each voice user is assumed to achieve and average throughput of 3.87kbps (9.6kbps with an activity factor of 4.3%). It can be seen from the figure that without any voice users, the sector throughput is 623 kbps. The sector throughput decreases with increasing voice user loading as expected. One can observe that even with a significant number of voice users, the system still provides substantial sector throughput for packet data users (e.g., larger than 3kbps for 2 or less voice users). When voice traffic becomes intensive, the packet data throughput is heavily affected.

4 To help with determining if there is a mixing gain or loss, a reference line is plotted in Figure 1, which represents data throughput that decreases proportionally with the voice load. This line shows where there is neither a mixing gain nor loss. Consequently, it can be interpreted that if the data throughput is above the line, there is a mixing gain while if the data throughput is below the line, there is a mixing loss. From the Figure 1, when the voice load is low, there is neither a mixing loss nor gain. However, as the voice load increases beyond 2 users, there is a mixing loss. However, it should be emphasized that the loss is always modest and reaches a maximum of 2% at 25 users. To understand why the mixing loss has this behavior, the average RoT values from the above simulations are plotted in Figure 2 along with the average RoT for the voice only case. It can be seen from the figure that the voice only RoT goes up monotonically as the number of voice users increases. Moreover, the curve is not linear but rather shows that incrementally more and more resources are necessary to maintain the quality of the voice service as the voice load increases. Note also that the average RoT for the mixed traffic case is increasing as the voice load increases. This is due to the lower transmit data rates of packet data users which result in smaller power variation at the sectors. Consequently, the average load of the system can increase because less load margin is needed to accommodate the load variations Figure 1. Mixed data and voice performance with CRC The characteristics of the mixing loss can now be elucidated. Recall that the goal of mixed data and voice traffic is to maximize the utilization of the reverse link resource by using the resources left over after servicing voice for the data transmission. At low voice loads, say less than 15 users, the RoT caused by the voice users increases very nearly linearly with increasing voice load. Consequently, the resources left over for data users is decreasing linearly and the maximum decrease is less than 1.25 db. Thus it is expected the aggregate packet data throughput decreases more or less linearly. At high voice loads, the RoT due to the voice load incrementally increases with increasing voice load. This results in incrementally less resources for packet data service which alone would result in incrementally increasing mixing loss. This loss, however, is mitigated by the ability of the system to have a higher and higher target load as the voice load increases. This has the effect of increasing the data throughput. Overall it can be seen that at high voice load, this increase in data throughput more or less offsets the incremental loss part. Figure 3 shows the histograms of transmit data rates for 5 data users with, 15 and 3 voice users. Without voice traffic, the highest percentage of the rate is at 39.6kbps, while in a few cases, the data users are able to achieve the peak rate kbps. With 15 voice users, the histogram shifts to the left and the rate with the highest percentage moves down to 156kbps. With a voice capacity of 3 users, more than half of the reverse link resource is taken by voice traffic, the majority of rates for packet data transmission are around 4.8kbps and the highest rate a data user can reach is only 463.2kbps. This figure clearly shows that the histogram is biased more and more towards the lower data rates as the voice load increases. Average Rise over Thermal Voice Only Voice with 5 Data Users Figure 2. Average sector Rise over Thermal for mixed and voice only cases The same simulations are repeated with DRC and GBS under the RoT constraint, and the performances are shown in Figure 4 and 5 respectively. Comparing with CRC, with tighter control, DRC utilizes the reverse link resource more efficiently and achieves a higher sector throughput when there is no voice load. However, unlike the performance with CRC, the DRC performance shows that the mixing loss gradually increases as the voice load increases beyond zero. This is mainly due to the greater rate variation from the BS DRC algorithm in this paper compared to that of CRC. This additional rate variation means that with DRC the load margin cannot be decreased as quickly as with CRC which reduces the data throughput and causes the increase mixing loss. Percentage of Each Rate (%) w/o voice users w/ 15 voice users w/ 3 voice users Transmit Data Rate (kbps) Figure 3. Packet data rate distributions While the GBS gives the most control over the mobile

5 rates, it results in larger rate fluctuation and is not efficient in terms of RoT management. Thus the aggregate data throughput without any voice load is lower than that of DRC. Moreover, it can be seen from Figure 5 that the mixing loss is greater than that for the DRC. This is because the effects of the GBS algorithm on voice traffic are even more pronounced. Firstly assignment of high rates will result in high intra-cell interference on the sector s voice users as well as inter-cell interference on voice users in adjacent sectors. Secondly, the voice users must use higher average power than in DRC or CDC because of the larger variation in the RoT introduced by data users being control by GBS. Therefore the presence of RL voice users will force an even more conservative (than CRC or DRC) operation of the GBS algorithm. This will result in reduced sector throughput for the packet data users with DRC Figure 4. Mixed data and voice performance with DRC Figure 5. Mixed data and voice performance with GBS C. Improved MAC Algorithm for Higher Sector Throughput With voice load, the average mobile data throughput as well as the average sector throughput is limited by the reverse link resource. For example, with 2 voice users and 5 data users, the aggregate sector throughput is 317kbps. Based on the idea proposed in Section III, we can identify a portion of the mobiles with better channel conditions and only assign them the R-PDCH. Consequently, fewer users that are in better radio conditions will share the resources. This will result in a more efficient utilization of the radio resource and give the operators flexibility to improve the data throughput at the expense of fairness. The simulations have been demonstrated in Figure 6 with the given example, where the number of active data users decreases from 5 to 1. With fewer active data users, the improvement in sector throughput is evident. This is because the data transmission is more efficient for mobiles with better channels. However, the sector throughput reaches the peak with two data users per sector because of the lack of multi-user diversity gain with one user Number of Active Data Users with 2 Voice Users Figure 6. System performance of data user selection algorithm V. CONCLUSIONS In this paper, the performance of mixed voice and data service in the reverse link of 1xEV-DV systems was obtained from simulations. The results show that with low voice load and using the CRC algorithm, the voice and data services can be supported without any mixing loss. Under other operating conditions studied in this paper, there is a modest mixing loss that is dependent upon the voice load and MAC mechanism. This dependency is a result of the interplay between incrementally increasing resources needed by the voice service and the ability of the system to target incrementally increasing system load as voice load increases. It should be noted that although the results in this paper show that DRC and GBS have larger mixing losses than CRC, the DRC and GBS BS algorithms can possibly be further optimized. Future research will focus on methodologies in optimizing these algorithms to reduce the mixing loss. REFERENCES [1] EIA/TIA/IS-2, cdma2 Standard for Spread Spectrum Systems, Revision C, May 22 [2] A. C. K. Soong, et al. Forward High-Speed Wireless Packet Data Service in IS-2 1xEV-DV, IEEE Communication Magazine, Aug 23. [3] TIA-2, cdma2 Standard for Spread Spectrum Systems, Revision D, February 24. [4] 3GPP2 TSG-C WG3, 1xEV-DV Evaluation Methodology Addendum (V14), 3GPP2 Contribution C R1, June 16, 23. [5] P. Hosein, et al. Medium Access Control for Reverse Link High-Speed Packet Data Service in CDMA2 1xEV-DV Network, Submitted to IEEE Communications Magazine. [6] R. T. Derryberry, Overview and Performance of the CDMA2 1xEV-DV Enhanced Reverse Link, ISSSTA 4 (Sydney, Australia), Aug 3 to Sept 3, 24. [7] T. Chen, et al. CDMA2 Revision D Reverse Link Enhancements, PIMRC 4 (Barcelona, Spain), Sept 5-6, 24.