3.1 Simulation Validation
The analytical models have been validated by the simulation assumptions. Based on the system model for HSDPA and the simulation parameters shown in Table 1, Table 2 shows the analytical and simulation results. For VoIP over HSDPA, the simulations yield capabilities of 41 and 51 simultaneous connections for delay budgets 80 and 150 ms, respectively. Clearly, the simulation results match the analytical results well. For HSUPA, we take packet bundling into consideration.
By applying the detailed simulation settings shown in Table 1 into our event-driven simulation model, we validate the correctness of our uplink analysis model.
Table 3 compares the simulation results with those of our HSUPA polling model
Table 1. Simulation Parameters For HSDPA/HSUPA System
Parameter for HSDPA Value Parameter for HSUPA Value
TTI size 2ms TTI size 10ms
Downlink data rate 3.6Mbps Uplink data rate 5.76Mbps
VoIP codes type GSM6.10 VoIP codes type GSM6.10
UE profile Pedestrian A UE Maximum Transmit power 21dBm
UE speed 3kmph (βec/βc)2 3dB
UE receiver structure Single-antenna Rake (βed/βc)2 8dB
320 S.-H. Yang, S.-R. Yang, and C.-C. Kao
Table 2. Comparison Between The Analytical And Simulation Results On HSDPA
Tdelay=80 ms Tdelay=150 ms
Ndown(Analytical) 39 49
Ndown(Simulation) 41 51
Table 3. Comparison Between The Analytical And Simulation Results On HSUPA
Tdelay= 80ms Nb=1 Nb=2 Nb=3 Tdelay= 150ms Nb=1 Nb=2 Nb=3 Nup(Analytical) 63 80 53 Nup(Analytical) 103 127 94 Nup(Simulation) 62 77 51 Nup(Simulation) 99 123 90
under different delay budgets 80 ms and 150 ms and different packet bundle sizes. It is clear that the analytical analysis is consistent with the simulation results.
3.2 Simulation Results and Analysis
Based on the HSDPA and HSUPA simulation models validated against the an-alytic analysis, we design different simulation scenarios to study and compare the VoIP performance over HSDPA and HSUPA technologies. First, we concen-trate on HSDPA. For HSDPA, we have provided the analytical and simulation VoIP capacity results based on RR scheduling in the last subsection. When other factors such as channel quality and throughput are taken into account for scheduling, the VoIP capacity may be influenced by the adopted schedul-ing algorithms. Fig. 2 shows the VoIP capacity performance over HSDPA. Fig.
2(a) represents the VoIP capacities when applying RR, Max C/I, FCDS and PF scheduling schemes under the delay budgets 80 and 150 ms. The specific relative capacity gains from each scheduling scheme compared to PF are shown in Fig.
2(b). PF provides better VoIP capacity performance than RR because it is on average able to schedule users at better channel conditions. Compared to Max C/I, PF is more fair because of taking UE’s average throughput into considera-tion. The results verify that packet scheduling plays a key role in achieving good
80
RR MaxC/I FCDS(0.5) PF
VoIPcapacity[users/cell]
RR MaxC/I FCDS(0.5) PF
VoIPcapacity[users/cell]
RR MaxC/I FCDS(0.5) PF
VoIPcapacity[users/cell]
Schedulingscheme
80ms 150ms
(a) VoIP cell capacities with different scheduler.
(b) Relative capacity gain of PF over the other algorithms.
Fig. 2. VoIP capacities (a) different scheduling schemes (b) relative capacity gain
Analyzing VoIP Capacity with Delay Guarantee 321
20 40 60 80 100
PCapacity[user/cell] elaybudget=80ms)
RR PF 0
20 40 60 80 100
1 2 3
VoIPCapacity[user/cell] (delaybudget=80ms)
Packetbundlesize
RR PF
(a) Delay Budget = 80 ms
40 60 80 100 120 140 160
PCapacity[user/cell] elaybudget=150ms)
RR PF 0
20 40 60 80 100 120 140 160
1 2 3
VoIPCapacity[user/cell] (delaybudget=150ms)
Packetbundlesize
RR PF
(b) Delay Budget = 150 ms
Fig. 3. Number of Supported VoIP Connections on (a) Delay Budget = 80 ms, (b) Delay Budget = 150 ms
VoIP capacity in HSDPA. From Fig. 2(a), we also note that the delay budget is another factor that affects the VoIP capacity. The longer the delay budget, the more the number of supported voice connections.
After discussing the VoIP capacity performance over HSDPA, we study the VoIP capacity over HSUPA. Considering RR, we have already provided the an-alytical and simulation VoIP capacity results over HSUPA. We also take the PF scheduling algorithm into account for observing the HSUPA VoIP performance.
Fig. 3 shows the VoIP capacity applying the RR and PF scheduling algorithms with different packet bundle sizes under 80 and 150 ms delay budgets. From Fig. 3(a) and Fig. 3(b), PF scheduling provides better VoIP capacity than RR scheduling regardless of the delay budget or the packet bundle size. Moreover, the packet bundle size is the other factor that influences VoIP performance. Packet bundle size two provides the better capacity performance than that without packet bundling. The phenomenon is explained as follows. The bundled packet simply contains one header and then the header overhead will be shared among the two packets. Hence, packet bundling decreases the header overhead and pro-vides better bandwidth efficiency. We also observe that applying packet bundling should waste time on waiting enough packets to be bundled and transmitted.
However, when increasing the bundle size from two to three, the VoIP capac-ity decreases. This is because the more the packet bundle size, the longer the waiting time for bundling packets. Hence, the VoIP capacity will decrease when the packet bundling delay is intolerable. From Fig. 3, we know that there is a trade-off between the packet bundle size and the improvement of VoIP capacity.
As a result, bundling two packets into one packet is the best option to enhance the VoIP performance over HSUPA in our simulation parameter settings.
After discussing the VoIP capacity over HSDPA and HSUPA, we know that the VoIP performance is influenced by the adopted scheduling scheme, the de-lay budget, and the packet bundle size. In the above experiments, we obtain the capacity based on the delay budget, one of the QoS requirements. From our simu-lation results, under the 80 ms delay budget, we obtain that the optimal capacity achieved with PF scheduling over HSDPA is 51 users. Due to the restrictions on the UE downlink data reception transmission rate, the peak data rate in our NS2 simulation scenarios is 3.6 Mbit/s while the ideal transmission rate defined in UMTS Release 5 is 14.4 Mbit/s. If the downlink transmission rate can be
322 S.-H. Yang, S.-R. Yang, and C.-C. Kao
Fig. 4. The number of simultaneous VoIP connections under the different packet bundle sizes and the corresponding MOS values when applying (a) RR (b) PF
increased up to 14.4 Mbit/s, fourfold number of voice packets can be transmit-ted during each TTI. Hence, the VoIP capacity over HSDPA may be improved at least two or three times and then the maximum supported number of VoIP connections may be 102 or 153. Similarly, under the 80 ms delay budget and applying the PF scheduling, the optimal capacity achieved with packet bundling size two under HSUPA is 86 users. Thus, when both uplink and downlink pack-ets can be transmitted with the ideal peak date rate, the VoIP performance is bounded by the uplink transmission performance. In the following, we focus on the VoIP performance under HSUPA and obtain the VoIP capacity according to E-model. Fig. 4 shows the number of simultaneous VoIP connections under the different packet bundle sizes and the corresponding MOS values. Clearly, the more restricted the demanding connection quality, the less the number of sup-ported VoIP connections. Compared to the three curves in Fig. 4(a), those in Fig.
4(b) degrade less sharply. This means when applying PF scheduling, the increase in the number of simultaneous VoIP connections leads to less decrease in connec-tion quality. Hence, PF scheduling algorithm provides better VoIP performance than RR scheduling algorithm. Based on E-model, we know that the connection quality of our analytical VoIP capacity is above 3.6 MOS value regardless of the adopted scheduling scheme and packet bundle size. This means that our analysis model approximates the VoIP capacity with guaranteed connection quality with which most of the users satisfy.
4 Conclusion
This paper focused on the VoIP performance in HSPA networks and builded a mathematical VoIP capacity analysis model to obtain the maximum number of supported VoIP connections that still meet transmission delay and voice quality requirements. Using simulation results, this study also validated the correctness of the proposed analytical method and discussed other factors affecting VoIP performance. Experimental results indicated that PF is the most appropriate algorithm for VoIP service, even in a mixed traffic environment or when the UE has poor channel quality. For packet bundling, there is a tradeoff between the packet bundle size and the bundle delay. In the simulated scenario in this study, a packet bundle size of two causes less bundle delay and decreases packet overhead.
Analyzing VoIP Capacity with Delay Guarantee 323
Hence, bundling two packets together improves VoIP capacity tremendously. Our precise analytical model showed that VoIP performance is bounded by uplink VoIP performance. Following to the E-model, this study also examined the effects of other QoS requirements on uplink VoIP capacity and verified that our analysis model approximates the VoIP capacity with guaranteed connection quality with which most of the users satisfy.
Acknowledgment
We would like to thank Chunghwa Telecom. This work was accomplished un-der close discussions with the researchers of Chunghwa Telecom. This work was supported in part by the National Science Council (NSC) of Taiwan un-der Contracts 96-2752-E-007-003-PAE, 96-2221-E-007-025-, 96-2221-E-007-027-, 96-2219-E-007-012- and 96-2219-E-007-011-, and Chunghwa Telecom.
References
1. Chen, T., Kuusela, M., Malkamaki, E.: Uplink capacity of voip on hsupa. In: IEEE 63rd Vehicular Technology Conference, 2006. VTC 2006-Spring, May 2006, vol. 1, pp. 451–455 (2006)
2. Lee, H., Kwon, T., Cho, D.-H.: An efficient uplink scheduling algorithm for voip services. IEEE 802.16 bwa systems 5 (September 2004)
3. Lee, H., Kwon, T., Cho, D.-H.: Extended-rtps algorithm for voip services. IEEE 802.16 systems 5, 2060–2065 (2006)
4. Qualcomm White paper. Air interface cell capacity of wcdma systems (May 2007) 5. Takagi, H.: Queuing analysis of polling models. ACM Comput. Surv. (1988) 6. Wanstedt, S., Ericson, M., Hevizi, L., Pettersson, J., Barta, J.: The effect of f-dpch
on voip over hsdpa capacity. In: IEEE 63rd Vehicular Technology Conference, 2006.
VTC 2006-Spring, May 2006, vol. 1, pp. 410–414 (2006)
7. Yang, S.-R., Yang, S.-H.: Analyzing voip capacity with delay guarantee for inte-grated hspa networks. Technical Report, Natl. Tsing Hua Univ. (2009)
8. Zhao, D., Shen, X.: Performance of packet voice transmission using IEEE 802.16 protocol. IEEE Wireless Communications 14(1), 44–51 (2007)