1. Time Slicing in Mobile TV Broadcast Networks with Arbitrary Channel Bit Rates Cheng-Hsin Hsu Joint work with Dr. Mohamed Hefeeda April 23, 2009 Simon Fraser University, Canada 1

2. Outline 2  Motivation  Problem  Saving energy on mobile devices in mobile TV networks  Solution and Analysis  Efficient approximation algorithm  Evaluation  With simulations and a real testbed  Conclusion

3. Mobile TV 3  Watch TV anywhere, and anytime  Watch more programs  higher revenues for service providers  Broadcast over cellular networks  but they are: (i) designed for unicast, and (ii) narrowband

4. Mobile TV Broadcast Networks 4  T-DMB: Terrestrial Digital Media Broadcasting  Started in South Korea  Limited bandwidth (< 1.8 Mbps)  DVB-H: Digital Video Broadcast – Handheld  Extends DVB-T to support mobile devices  High bandwidth (< 25 Mbps), energy saving, error protection, efficient handoff, ….  Open standard  MediaFLO: Media Forward Link Only  Similar to DVB-H, but proprietary (QualComm)

5. Mobile TV Receivers 5  In contrast to TV sets  Battery powered  Mobile and wireless  Small displays  Energy consumption is critical on mobile devices  Mobile TV chip consumes 40~60% energy  our measurements on Nokia N96 phones  Broadcast standards dictate mechanisms to save energy

6. Outline 6  Motivation  Problem  Saving energy on mobile devices in mobile TV networks  Solution and Analysis  Efficient approximation algorithm  Evaluation  With simulations and a real testbed  Conclusion

7. Problem Statement 7 Optimally broadcast multiple TV channels to minimize energy consumption on mobile devices

8.  This is called Time Slicing (in DVB-H and MediaFLO)  Need to construct Feasible Time Slicing Schedules  No receiver buffer under/over flow instances  No overlap between bursts  Burst scheduling problem for base stations Energy Saving for Mobile Devices Time Bit Rate R r Off Burst Overhead T o 8

9. Burst Schedule  Easy IF all TV channels have same bit rate  Currently assumed in many deployed networks Simple, but is it efficient (visual quality & bw utilization)? TV channels broadcast different programs (sports, series, talk shows, …)  different visual/motion complexity Time R Bit Rate Window p 9

10. The Need for Different Bit Rates  Wide variations in quality (PSNR), a s high as 10—20 dB 10 dB  Encode multiple video sequences using H.264/AVC codec at various bit rates, measure quality 10

11.  Ensure no buffer violations for ALL TV channels  Difficult Problem Burst Scheduling with Different Bit Rates Time R Bit Rate Window p 11

12. Shifting bursts in time can lead to playout glitches Challenge 12 Time Buffer Fullness Time Buffer Fullness Buffer Underflow Time Buffer Fullness Buffer Overflow

13.  Theorem: Burst Scheduling to minimize energy consumption for TV channels with arbitrary bit rates is NP-Complete  Proof Sketch:  We show that minimizing energy consumption is the same as minimizing number of bursts  Then, we reduce the task sequencing problem with release times and deadlines problem to it  We can NOT optimally solve it in Real Time Harness 13

14. Outline 14  Motivation  Problem  Saving energy on mobile devices in mobile TV networks  Solution and Analysis  Efficient approximation algorithm  Evaluation  With simulations and a real testbed  Conclusion

15.  Observation: Hardness is due to tightly-coupled constraints: no burst collision & no buffer violation  could not use previous machine scheduling solutions, because they will produce buffer violations  Our idea: decouple them!  Transform problem to a buffer violation-free problem  Solve the transformed problem efficiently  Convert the solution back to the original problem  Ensure correctness and bound optimality gap in all steps Solution Approach 15

16.  Transform idea:  Divide receiver buffer into two: B and B’  Drain B while filling B’ and vice versa  Divide each scheduling frame p into multiple subframes  Schedule bursts s.t. bits received in a preceding frame = bits consumed in current frame Double Buffering Scheduling (DBS) 16 Buf B Buf B’ Fullness Fill Drain Fill Drain Fill Drain

17. DBS Algorithm: Pseudocode 17 1.// double buffering transform 2.For each TV channel, divide the scheduling frame into multiple subframes based on its encoding bit rate 3.// note that each frame is specified by 4.// burst scheduling based on decision points 5.For each decision point t, schedule a burst from time t to t n for the subframe with the smallest end_time, where t n is the next decision point

18.  Theorem: Any feasible schedule for the transformed problem is a valid schedule for the original problem.  Also a schedule will be found iff one exists.  Theorem: The approximation factor is:  How good is this? Correctness and Performance 18

19.  20 channels (R = 7.62 Mbps), energy saving achieved by the algorithm is 5% less than the optimal Approximation Factor 19

20. Outline 20  Motivation  Problem  Saving energy on mobile devices in mobile TV networks  Solution and Analysis  Efficient approximation algorithm  Evaluation  With simulations and a real testbed  Conclusion

21. Broadcast 12 TV channels Empirical Evaluation  No buffer violations  Notice the buffer dynamics are different 21

22.  Compare against a conservative upper bound  Broadcast channels one by one Near-Optimality in Energy Saving  Gap < 7% 22

23.  Running time for a 10-sec window is < 100 msec on commodity PC for broadcasting channels saturating the air medium Efficiency 23

24. Outline 24  Motivation  Problem  Saving energy on mobile devices in mobile TV networks  Solution and Analysis  Efficient approximation algorithm  Evaluation  With simulations and a real testbed  Conclusion

25.  Broadcast multiple TV channels to minimize energy consumption on mobile devices  A near-optimal algorithm for a NP-Complete burst scheduling problem  Approximation factor close to 1 for typical network parameters  Evaluated with simulations and a real mobile TV testbed Conclusion 25

26. Questions? 26 Thank you! More details can be found online at http://nsl.cs.sfu.ca