1. Multicast and Unicast Real-Time Video Streaming Over Wireless LANs Abhik Majumdar, Daniel Grobe Sachs, Igor V. Kozintsev, Kannan Ramchandran, and Minerva M. Yeung IEEE Transactions on Circuit and Systems for Video Technology

2. Introduction  Addressing the problem of real-time video streaming over wireless LANs.  Unicast Forward error control (FEC) Automatic Repeat ReQuest (ARQ)  Multicast ARQ Optimization of a maximum regret cost function  Experimental results

3. Challenges for wireless streaming  Fluctuations in channel quality  High bit-error rates  Heterogeneity among receivers Each user will have different channel conditions, power limitations, processing capabilities, etc.

4. Packet-erasure probability in 802.11b

5. Source coding  Fixed-rate (single resolution) coding Compressing the data to a target size. High compression ratio Ex: DPCM, JPEG, MPEG1  Progressive (scalable) coding Data is divided into coding units. The decoding of the data within a coding unit can be partial. More data can be decoded implies the better quality. Ex: MPEG4-FGS

6. Rate distortion characteristics

7. Dependencies between data units

8. Communication protocols  Asynchronous Reliable but have unbounded delay Need acknowledgment Such as ARQ  Synchronous Bounded delay No feedback Such as FEC

9. FEC coding  Can protect data against channel erasures by introducing parity packets.  Cannot guarantee that the receiver receives all the packets without error.  This paper employs Reed-Solomon (RS) codes. Described by two numbers (n, k)  n is the length of the codeword.  k is the number of data symbols in the codeword. The original data can be recovered if at least k of the original n symbols are received.

10. MDFEC  MDFEC Converts a prioritized multi- resolution bitstream into a nonprioritized multiple description bitrstream.  An (n, i) RS code is applied to it to form the N packets.  The ith resolution layer can be decoded on the reception of at least i packets.

11. MDFEC conversion packet layer

12. Hybrid ARQ  Algorithm Split data into “ packet groups ” consisting of k packets each. For each packet group, append n-k RS parity packets. Transmission  Transmitter initially sends only the first k data packets.  Transmitter starts sending parity packets until: An ACK is received The deadline of the transmission is reached  Once at least k packets are received, the receiver sends an ACK.

13. Coding schemes

14. Advantages of HARQ  Require less parity packets than FEC.  Require less acknowledgements than ARQ.  When acknowledges are lost, the transmitter simply assumes that more parity is needed.

15. Block diagram of experimental system

16. Throughput for FEC, ARQ, and HARQ n=150 k=100

17. Discussion  The throughput is defined as (time to send k data packets / the average time actually need to send them). The probability of successfully sending a data packet.  HARQ method is better than that of the ARQ system because fewer ACKs are sent.  Both ARQ and HARQ outperform FEC in this setup.  FEC will become optimal as the block size (n) increases.

18. A multicast case  ARQ-based schemes are less appropriate for the multicast case.  Problem formulation for multicast Arriving at an overall quality criterion for the multi- user case is difficult. This paper focuses on a maximal regret criterion.   R is the rate partition  E[d i ] min is the minimum expected distortion for the ith client.  E[d i (R)] is the expected distortion for the used coding scheme.

19. Minimax regret

20. Comparison of penalty in distortion

21. Distortion penalty