1. Three-layer scheme dominates previous double-layer schemes Distortion-diversity tradeoff provides useful comparison in different operating regions Layered Source-Channel Schemes: A Distortion-Diversity Perspective: S. Jing, L. Zheng and M. Medard Diversity can be achieved through source coding techniques, like multiple description codes We characterize source-channel schemes with distortion-diversity tradeoff Distortion-diversity tradeoff better characterizes layered source-channel schemes MAIN ACHIEVEMENT: A three-layer source-channel scheme, which includes previous multi-resolution-based and multi-description-based schemes as special cases HOW IT WORKS: Multi-description source code with a common refinement component Superposition coding with successive interference cancellation Joint source-channel decoding exploits source code correlation ASSUMPTIONS AND LIMITATIONS: Quasi-static block-fading channel Receivers have perfect channel state information, but the transmitter only has statistical knowledge of the channel Conventional source-channel scheme achieves a single level of reconstruction Diversity is usually achieved in the channel coding component Extend multi-description-based source-channel scheme while preserving the interface between source and channel coding More general channel model IMPACT NEXT-PHASE GOALS ACHIEVEMENT DESCRIPTION STATUS QUO NEW INSIGHTS

2. Sheng Jing, Lizhong Zheng, Muriel Médard ITMANET Mar 2009

3. Motivation Multiple user groups (eg. PDAs vs. Laptops) Accuracy: image resolution Reliability: successful image loading probability Different preferences of accuracy vs. reliability How well can we serve multiple user groups simultaneously? 2015-10-193

4. Background [Diggavi et al ’03] Layered channel codes (“diversity-embedded codes”) [Diggavi et al ’05] Tradeoff between diversity orders of 2-layer channel code [Effros et al ’04] [Laneman et al’05] Source coding techniques can also improve diversity for certain reconstructions We previously looked at the tradeoff between distortion and diversity for SR and MD schemes [Jing et al ’08] In this talk, we present a unifying scheme that matches the distortion and diversity (D-D) tradeoff of SR and MD schemes 2015-10-194

5. Outline Problem formulation Review of two schemes –SR with superposition coding –MD with joint decoding Unifying scheme: MD with common refinement Performance comparison Concluding remarks 2015-10-195

6. Problem Formulation Source: i.i.d. unit-variance complex Gaussian Quadratic distortion measure Quasi-static parallel fading channel where and Power constraint: SNR per subchannel No channel state information at transmitter Perfect channel state information at receiver 2015-10-196

7. Problem Formulation (cont.) At high SNR, for each source reconstruction –Distortion exponent: –Diversity order: 2015-10-197

8. Problem Formulation (cont.) Distortion-Diversity (D-D) tradeoff: achievable distortion exponent & diversity order tuples –Example: three reconstructions (partial, full, and refine), D-D tradeoff includes all achievable –Alternative performance metric: average distortion 2015-10-198

9. Two schemes SR with superposition coding : two-layer successive refinement source code matched to the distortion levels : superposition channel code, with power and : successive interference cancelation channel decoder 2015-10-199

10. Two schemes (cont.) MD with joint decoding [Laneman et al ’05] : symmetric El-Gamal-Cover (EGC) code [El Gamal et al ’82] matched to distortions : joint source-channel decoder –Use correlation between source codewords to identify unlikely pairs of channel codewords 2015-10-1910

11. Performance Comparison SR scheme: D-D tradeoff along the direction of MD scheme: D-D tradeoff along the direction of 2015-10-1911

12. Performance Comparison (cont.) Compare the D-D regions at 2015-10-1912

13. Common Refinement Scheme : symmetric EGC code with common refinement matched to 2015-10-1913

14. Common Refinement Scheme (cont.) Treating refinement layer as noise, form candidate lists of, resp. Search for a jointly typical candidate pair –If find only one pair, subtract the corresponding channel codewords and decode for –Otherwise, search each candidate list for 2015-10-1914

15. Connection with SR and MD Common refinement scheme includes the MD scheme as a special case (set ) Common refinement also includes SR scheme? –Simple approach requires huge codebook –We show, makes the common refinement scheme as good as SR in D-D tradeoff, and avoids exploding 2015-10-1915

16. Performance Comparison D-D tradeoff:, Extreme case 1: 2015-10-1916

17. Performance Comparison (cont.) D-D tradeoff:, Extreme case 2: 2015-10-1917

18. Concluding Remarks 3-level unifying scheme –Dominates both SR and MD schemes –Smooth transition between SR and MD schemes –No strictly superior performance On-going work –Parallel channel MIMO channel –Unifying source-channel scheme that also preserves the digital source-channel interface 2015-10-1918