Presentation on theme: "1 A Brief Review of Joint Source-Channel Coding CUBAN/BEATS Meeting 29th April, 2004 Fredrik Hekland Department of Electronics and Telecommunication NTNU."— Presentation transcript:
1 A Brief Review of Joint Source-Channel Coding CUBAN/BEATS Meeting 29th April, 2004 Fredrik Hekland Department of Electronics and Telecommunication NTNU
2 Outline Traditional Tandem Structure (The Separation Principle). The Source and Channel Coders’ Roles. Joint Source-Channel Coding (JSCC). Review some Approaches to JSCC. Transcoding/Digitizing the JSCC Symbols.
3 The Traditional Approach Source Coder Channel Coder Modulation Channel De- Modulation Input Signal Remove Redundancy Add Structured Redundancy Adapt to Channel Channel Decoder Source Decoder Reconstructed Signal
4 The Traditional Approach (Cont’d) Source coder seeks to remove all redundancy. Channel coder seeks to obtain error-free transmission. Important question: Is this optimal when communicating analogue sources?
5 Pros&Cons with the Separation Theorem +Source and Channel Coders can be optimised independently. +Change source coder without affecting channel coder (and vice-versa). +Optimal for most channels. +Very robust above design CSNR. ÷Optimality requires infinite delay/complexity. ÷Not valid for certain multiuser and packet network channels. ÷Break-down below design CSNR. ÷Does not adapt changing channel qualities (No graceful degradation/improvement, Worst-case CSNR design).
6 Joint Source-Channel Coding (JSCC) Source and Channel Coders co-optimised to some extent. Possible benefits: Can perform better when subjected to a delay/complexity constraint. Provides robustness against changing channel qualities. Broadcast channels when sender has no CSI. Less complex systems can perform optimally without explicit coding. Can allow channel noise to be part of the total distortion. But: All this comes at a cost of reduced flexibility!
7 Some Possible Approaches Rate-Distortion Source-Channel (resource control). Unequal Error Protection (UEP) / Hierarchical Protection. Exploit residual correlation remaining after source coding. Index Assignment. Channel Optimised Vector Quantization. Direct Modulation Organizing Schemes. Channel Code not necessary
8 The First “Obvious” Step towards JSCC Standard coder blocks. Channel capacity is shared “intelligently”. CSI dependent. Not quite “true JSCC”. No joint optimisation except for the rate. For example: 3D sub-band video coder with RCPC (Cheung&Zakhor) Source Coder Channel Coder Channel R(D) optimized resource control CSI
9 Graceful Degradation/Improvement Traditional tandem systems designed for worst-case CSNR. No improvement when the channel is better. Breakdown below the design threshold. Hybrid Digital-Analogue (HDA) Systems (Mittal&Phamdo) The linear analogue part provides robustness and/or improvement.
10 Multi-resolution Modulation Wavelet transform in source coder. Modulation space with three levels of protection. (Kozintsev&Ramchandran)
11 Direct Source-Channel Mappings No explicit channel code. Operate on e.g. –Quantization + Index Assignment –Channel Optimised VQ Distribute total distortion on quantization noise and channel noise.
12 Direct Source-Channel Mappings No explicit channel code. Operate on e.g. –Quantization + Index Assignment –Channel Optimised VQ Distribute total distortion on quantization noise and channel noise.
13 Example – Dimension Expansion Problem: Signal points have more neighbours in the channel space than in the source space. Important to utilise the entire channel space.
14 Example – Dimension Reduction A Dimension Reduction mapping should: 1.Cover the entire source space to lower the approximation noise. 2.Map the most probable symbols to low-amplitude channel symbols. 3.Map close channel symbols back to signals close in the source space. 4.Match the channel symbol statistics to the channel statistics in order to attain capacity.
16 Quantizing the JSCC symbols Digitization necessary for further transmission in transport networks. Transcode instead of recoding. (Avoid re-quantization and complexity at the expense of higher bit-rate) Uniform SQ with entropy coding.
17 Quantizing the JSCC symbols (Cont’d)
18 Quantizing the JSCC symbols (Cont’d) Uniform quantization with arbitrarily number of levels + entropy coding. Quantization step closely linked to spiral arm distance. Constant ratio between quantization step and spiral arm distance gives constant loss.
19 Quantizing the JSCC symbols (Cont’d)
20 Summary Trad. Tandem StructureJoint Source-Channel Structure + Very robust above design CSNR. Each coder optimised independently. Easy to change coders. Optimal given no delay constraints. Can be made optimal under delay constraints. Graceful degradation/improvement. Ideal for broadcasting where all the channels are different. ÷ Must design for worst-case CSNR, (i.e. no adaptation). Requires long code blocks. Break-down below design CSNR. Less flexible (complete re-optimisation when changing source or channel). Harder to optimise.