1. An optimal power-saving class II for VoIP traffic and its performance evaluations in IEEE 802.16e JungRyun Lee School of Electrical and Electronics Eng,Chung-Ang University, Korea DongHo Cho Department of Electrical Engineering and Computer, Korea ACM Computer Communications 2008

2. Outline Introduction Delay model for VoIP traffic The optimal sleep interval decision algorithm Simulation Conclusion

3. Introduction Since an MSS is powered by a limited battery, the energy conservation of an MSS in IEEE 802.16e PSM is a key factor in WiMAX application. Under the PSM, an MS goes to the sleep state, and wakes up during predetermined listening period in order to verify the existence of buffered packet destined for it. Since PSCII can adjust the length of sleep intervals according to the period of VoIP packet generation, we apply a PSC II to VoIP service.

4. Motivation The large packet buffering delay of a VoIP packet may lead to a high packet drop rate, the quality of service (QoS) of VoIP services can deteriorate when long sleep interval length is employed. BS MS BS MS DL/UL

5. Goal To find the optimal length of sleep intervals under the PCS II while satisfying the delay constraint of a VoIP connection.

6. Delay model for VoIP traffic Codec-induced delay ( G.729 Codec ) – Encoding delay / Decoding delay – Packetization delay / depacketization delay Playout delay by de-jitter buffering in a receiver – Playout Delay: 語音播放延遲 – De-jitter : jitter = 0 ( 封包 delay 為定值 ) Network delay – Routing delay of IP Router Variable delay and fixed delay – Buffering delay in BS PSC II leads to another additional packet buffering delay caused by periodic sleep intervals in the BS

7. Definitions τ = 37.5 ms ρ = 22 ms T = 20 ms Reference “Voice over internet protocol”

8. Delay model for VoIP traffic VoIP Sender BS VoIP Receiver Encoding delay Routing delay of IP Router Packetization delay Buffering delay in BS Depacketizati on delay Decoding delay T T p1p2 p1 p2 p1p2 p1 P1 end-to-end delay P2 end-to-end delay Sender-to-BS delay

9. Delay model for VoIP traffic Routing delay of IP Router Sender-to-BS delay FixedVariable Transmission delay FixedVariableFixedVariable Router 1Router 2Router 3 Queuing delay with the rate λ i For the service time in each router, the evaluation begins with an exponentially distributed queuing delay with rate λ i at the i-th router and the sum of the exponential random variables follows a hypoexponential distribution.

10. The Optimal Sleep Interval Decision Algorithm Assumption – Only one party ( VoIP users ) is under the PSC II – Sender-to-BS is N hop – The transmission delay for a router is fixed

11. The Optimal Sleep Interval Decision Algorithm Definitions – T : Packet generation interval – α : the length of listening interval – β : the length of sleep interval – (α + β) : a power-saving unit α + β= kT for positive integer k BS αβ VoIP sender T α + β= 2T

12. The Optimal Sleep Interval Decision Algorithm Definitions – m : the m-th slot – p : the p-th power-saving unit – mT + pkT ( 0 <= m < k) : the m-th slot in p-th power- saving unit BS αβ VoIP sender T α + β= 2T mT + pkT = 1 x T + 1 x 2 x T

13. The Optimal Sleep Interval Decision Algorithm Definitions – ρ : Sum of decoding delay and depacketization delay – Y m : sender-to-BS delay of the VoIP packet generated at the m- th slot – : sender-to-BS delay + packet buffering delay – W m : end-to-end delay of the packet generated at the m-th slot α + β= 2T

14. The Optimal Sleep Interval Decision Algorithm Delay Constraint – W : end-to-end delay of a packet – D thr :delay threshold – δ : the maximum tolerable packet loss probability

15. The Optimal Sleep Interval Decision Algorithm Proposition: – The delay constraint δ <=1/k is satisfied if and only if, for each packet generated at the m-th slot of each power-saving unit, there exists some positive integer l that satisfies the following two inequalities simultaneously:

16. The Optimal Sleep Interval Decision Algorithm

17. l m :indicates the smallest value of n such that the packet generated at the m-th slot of the p-th power-saving unit satisfies Proposition simultaneously in the ( n + p )-th power-saving unit. f Y (t) : Probability Density Function of Y : cdf of f Y (t)

18. The Optimal Sleep Interval Decision Algorithm Prove the position is true

19. Example D thr - ρ is larger than 8T, l = 4 the smallest l satisfying proposition becomes 3 for m = 0. The smallest l satisfying proposition is 4 for m =1,

20. The Optimal Sleep Interval Decision Algorithm Definitions – τ: Sum of encoding delay and packetization delay – a i : Fixed delay for each router – X i :Variable delay for each router – Z i : Delay for each router – Y: Total routing delay for all router

21. Analysis Model The probability with which the packet, generated at the m-th slot of the first power-saving unit, arrives at the i-th power-saving unit

22. Simulations ITU-T defines an object model to estimate the perceived quality of VoIP session D thr is set to 285 ms The packet loss threshold, d, is set to 0.03. a i is assumed to be 10 ms The framing interval in IEEE 802.16e is 5 ms and the length of the listening interval is set to one frame duration. shows an example of the available values of l with increasing k when N=3 and λ i = i / 30. Thus, the longest power saving unit satisfying the delay constraint becomes 3 x 20 = 60ms and the longest sleep interval becomes 55 ms.

23. Simulations As the average end-to-end delay of VoIP packet increases, the optimal sleep interval length decreases in order to maintain the packet drop probability within the given threshold.

24. Simulations As expected, the energy consumption of an MS decreases and the average packet buffering delay increases as k increases.

25. Conclusions In this work, we studied the methodology on how to choose the optimal sleep interval of the PSC II while satisfying the given end-to- end delay constraint of a VoIP connection, in the context of IEEE 802.16e standard.