1. Congestion Control Based on Priority Drop for H.264/SVC Multimedia and Ubiquitous Engineering, 2007. MUE '07. International Conference on (2007), pp. 585-589. Reporter : 陳志明

2. Outline Introduction Related work PID control Priority dropping Simulation Conclusion

3. Introduction(1/2) Scalable H.264 Coding (SVC) extends the coding approach of H.264/AVC Spatial scalability Spatial scalability Temporal scalability Temporal scalability SNR scalability SNR scalability For network applications based on SVC, congestion control is a crucial issue.

4. Introduction(2/2) The basic idea is that use an active queue management (AQM) algorithm, such as RED, PI, PID, etc. For ensuring the quality of decoding video. This paper presents a new AQM algorithm PID_PD based on priority dropping and PID algorithm.

5. Related work S.Floyd et al. proposed the first AQM algorithm, random early detect (RED). Hollot et al. proposed proportional integral (PI) algorithm. Fan YF et al. presented a proportional integral derivative (PID) algorithm.

6. PID control(1/4) q 0 is expected queue length q is instantaneous queue length e = q − q 0 is error signal p is packet dropping rate

7. PID control(2/4) The input e(t) and the output p(t) satisfy the follow relation. K p is proportion coefficient K i = K p / T i is called as integral coefficient K d = K p * T d is called as derivative coefficient Let t = kT

8. PID control(3/4) We get discrete PID expression : We have the probability dropping formula by getting increment both sides from above expression :

9. PID control(4/4) The pseudo code is as follow : Because the probability p always is between 0 and 1, we define p as follows :

10. Priority dropping(1/3) We will present a new active queue management schema PID_PD based on priority dropping PID. Suppose SVC has m layers, every layer has n priority ranks, and then we can define the follow priority matrix to express the SVC priority

11. Priority dropping(2/3) We firstly define packet priority number as follow : For a new arriving packet, firstly we calculate its dropping probability according to PID control. When the current packet is determined to drop, we look for the packet whose priority number.

12. Priority dropping(3/3) The process is expressed as follow by using pseudo code :

13. Simulation(1/4) JSVM4-11 is used to create bit stream. NS2 is used to build the network simulation environment. The router queue management algorithm is PID_PD.

14. Simulation(2/4) Simulation model Simulation model The delay of all links is a random number between 5ms and 10ms. The max queue length is 200 packets, and the expected queue length is 100 packets. The a = 0.000014, b = -0.00002228 and c = 0.000008729. The H.264/SVC bits stream in the experiment includes two spatial layers, 4800 frames. The first layer (80x44) and the second layer (176x144) have the same frame rate, 30fps. GOP size of the bit stream is 16.

15. Simulation(3/4) The whole bit stream (BL+EL) PSNR using PID The whole bit stream (BL+EL) PSNR using PID_PD

16. Simulation(4/4) The basic layer (BL) bit stream PSNR using PIDThe basic layer (BL) bit stream PSNR using PID_PD

17. Conclusion PID_PD schema can prevent the high priority layer or frame from dropping, provides high video quality to the end users.

18. Spatial scalability 上一頁

19. Temporal scalability 上一頁

20. SNR scalability 上一頁

21. Simulation model 上一頁