Multicast and Unicast Real-Time Video Streaming Over Wireless LANS April. 27 th, 2005 Presented by, Kang Eui Lee
Packet-Erasure Model for IEEE LANs □ Two lower layers ▪ Physical layer ▪ Data link layer □ On the application level, ▪ Can not access to those two layers ▪ User application see the wireless channel as an IP packet channel with erasures □ Simplest Model ▪ Erasures are i.i.d with probability of P e
Coding for Packet-Erasure Channels □ Automatic Repeat reQuest (ARQ) ▪ Asynchronous ▪ Reliable, but with unbounded delay ▪ Works well for data communication □ Forward Error Correction (FEC) ▪ Synchronous ▪ Protect data using parity packets ▪ No feedback channel ▪ Original data can be recovered perfectly
Coding for Packet-Erasure Channels (cont.) □ Partially-Synchronous version of ARQ ▪ Still requires low packet loss rate and low RTT
Coding for Packet-Erasure Channels (cont.) □ Reed-Solomon Code ▪ ( n, k ) ▪ ‘n’ is the length of codeword ▪ ‘k’ is the number of data symbols in codeword ▪ RS code can be used for correction and erasures ▪ Correct any ( n-k ) erasures out of n
Streaming Video Over WLAN: A Single User Case-MDFEC □ MDFEC(Multi Description FEC) ▪ Transcoding mechanism to convert a prioritized MR bitstream into a nonprioritized bitstream using efficient FEC ▪ The progressive bitstream is marked at N different positions. (forms N resolution layers) ▪ ‘i’ th layer is split into ‘i’ equal parts and (N, i ) RS code is applied to it to form the N descriptions
Streaming Video Over WLAN: A Single User Case-MDFEC(cont.) ‘i’ th (N, i ) RS code MR bitstream
Streaming Video Over WLAN: A Single User Case-MDFEC(cont.) Descriptions Layers(N, i ) RS codes
Streaming Video Over WLAN: A Single User Case-Hybrid ARQ □ Hybrid ARQ ▪ To combine the reliability of ARQ and bounded delay of FEC ▪ Algorithm main(){ send(first k data packets); while(ARQ is not received && Timeout is not expired){ send(n-k RS parity packets); } send(next k data packets); }
Streaming Video Over WLAN: A Single User Case-Throughput Throughput: 1. FEC :R.V. that represents the number of packet erasures in a group of n packets :Probability of packet erasure
Streaming Video Over WLAN: A Single User Case-Throughput(cont.) 2. ARQ E : R.V. that represents the total number of packets sent in a successful transmission of k packet 3. Hybrid ARQ
Streaming Video Over WLAN: A Single User Case-Experiments
Streaming Video Over WLAN: A Multi User Case □ ARQ vs. FEC ▪ ARQ based schemes are less appropriate ▫ Too many ACKs, ▫ Different user requires retransmission of different packets □ Goal in the multicast scenario ▪ Maximize some composite delivered quality criterion, given the total rate constraint and the transmission profile
Streaming Video Over WLAN: A Multi User Case(cont.) □ Definitions ▪ Rate Partition, ▪ Rate-Distortion function for rate ‘r’, ▪ Transmission profile, ▫ probability of the ‘i’ th client receiving j out of N ▪ Expected Distortion(ED), where ‘E’ is the source variance
Streaming Video Over WLAN: A Multi User Case(cont.) □ Maximal Regret Criterion ▪ Optimal coding scheme is the one that minimizes, ▫ E[d i ] min is the minimum ED for the ith client achieved by using the optimal coding scheme when it is the only client. ▫ E[d i (R)] is the ED for the particular coding scheme being used
Streaming Video Over WLAN: A Multi User Case(cont.) □ Constraints on solution ▪ Total rate constraint of the clients: R tot ▪ Total rate when MDFEC is used,
Streaming Video Over WLAN: A Multi User Case(cont.) ▪ Resource constraint ▪ Embedding constraint □ Proposed solution
Streaming Video Over WLAN: A Multi User Case(cont.) □ Proposed solution ▪ Assuming that rate-distortion function is convex ▫ is also convex ▫ Since infimum/supremum of convex is also convex, is convex ▫ Finding the minimax regret becomes convex optimization
Streaming Video Over WLAN: A Multi User Case(cont.) □ Proposed solution ▪ For 2 clients, ▫ Since is convex and minimum of, we choose R where ▪ For more than 2 clients, ▫ Analyzing the users pairwise, choose the highest point of
Packet-Erasure Model □ Erasures generally model two types of events ▪ An unfortunate noise sequence that the underlying error correcting code could not correct ▪ Collisions at either an intermediate node in a network (packet drop) or over the shared comm. medium □ Recovery of erasures ▪ Knowledge of the erasure comes back to the TX. through either an acknowledgment packet or by the transmitter observing the packet getting mangled over the link.
Streaming Video Over WLAN: A Single User Case-MDFEC(extra) Marked MR bit-stream