WPR.7 - Joint Source and Channel Coding/Decoding

Objectives

This WP addresses issues related to the robust transmission of multimedia (mainly video) over wireless channels (typically at the periphery of a wired IP network). Addressed issues include: (i) providing new tools (and expanding existing ones) for Joint Source and Channel Coding/Decoding (JSCC/D) in either classic one-to-one, or one-to-many (broadcast mode), or distributed contexts; (ii) providing new tools for analysing the efficiency of these schemes; (iii) addressing the compatibility problem with the classical OSI layers separation (to what extent is this separation valid) while identifying cross-layer methods in the context of JSCC/D, for instance, finding the best possible redundancy allocation scheme at various layers for a fixed global rate; (iv) working on practical, long-term situations, to be used as inputs to test-beds, considering Wimax and LTE technologies.

Description of work

Work is broken down into several tasks as per the above Objectives.

Task TR7.1: New tools for JSCC/D

Examples of tools that will be studied are given below:

7.1.1. Application of digital fountain codes and Reed-Solomon codes to streaming and UEP in wireless networks.

Digital fountain (DF) codes are becoming increasingly important for multimedia communications over networks subject to packet erasures. These codes have significantly lower complexity than Reed-Solomon, exhibit high erasure-correction performance, and are well suited to generating multiple equally important descriptions of a source. The use of DF codes for reliable multimedia delivery opens a number of research issues in terms of encoding/decoding delay and required overhead for realistic video streaming applications when the number of information packets may be away from asymptotic optimality. Moreover, DF codes will be studied as a means of providing unequal error protection and joint source channel coding. As a reference, schemes built on Reed-Solomon codes that have optimal erasure correction capability will be used.

Main players: CNIT, Technion, FTW, CNRS

7.1.2. Joint source channel coding on a fading channel with no channel state information at the encoder.

The approach will be based on merging two concepts, namely successive refinements with broadcast channel coding. The allocation of power within the layered model in broadcast settings is to be optimized so as to minimize average distortion. The effort will encompass both discrete fading processes as in a network where interfering users are switched on and off without the knowledge of the desired transmitter, and also for continuous processes, such as on-off fading. On the second, more advanced stage of this research, joint-source channel coding will be considered in the presence of side information available at the decoder, and elements based on Wyner-Ziv strategies will be investigated.

Main players: Technion, CNRS

7.1.3. Joint source channel coding based on source codes.

Solutions of joint source-channel coding based on arithmetic codes with different strategies for introducing redundancy (e.g. via a forbidden symbol or extra a prior source information) will be studied in comparison with channel coding based solutions.

Main players: CNIT, CNRS

Task TR7.2: Tools for multi-terminal JSCC/D

7.2.1. Distributed joint source and channel coding.

In a multi-terminal scenario, correlated sources are sent over a common wireless channel without coordination between the senders. In this context the source-channel separation theorem does not hold, which motivates the design of joint schemes. Since the distributed source coding problem can be viewed as a channel coding one, our approach will be based on designing powerful channel codes for the distributed JSCC/D problem. More precisely, the effort will be to design codes that can operate on a wide range of correlation values. In a second step, we will investigate the case of correlated sources with memory.

7.2.2. Collaborative joint source-channel coding strategies for sensor-data communications.

We consider the general problem of collaborative transmission of distributed samples of random fields. A set of sensors capable of measuring spatial samples of a random field (e.g., interference, temperature, noise levels, etc.) are assumed to be deployed in a region.  They are equipped with transceivers which can convey the samples either to a central collector node which then attempts to reconstruct the field, or they can perform joint processing using distributed estimation algorithms. This setup is a combination of the classical problem of distributed source coding of a random vector, and that of multi-user transmission over a noisy channel. This remains an open problem in the literature. The non-ergodicity of the sensing process forces the sensors to be limited to one-shot transmission of the samples to the collector, instead of jointly coding the sequences of estimates. The work will consider different network topologies and coding strategies: (i) time-invariant correlated sources using finite-energy signal sets over multiple-access channels, (ii) multi-hop transmission of time-invariant correlated sources using finite-energy signal sets via relaying and intermediate processing.

Main players: CNRS, FTW, INRIA

Task TR7.3: Tools for evaluating the efficiency of JSCC/D methods

It has become clear thus far that the problem of tuning all steps of the global scheme is so complex that the use of analytical tools is absolutely necessary. Otherwise, simulations will never be able to provide design rules for the global system.  First examples of situations where efficient performance evaluation tools can be found are given below:

1-Theoretical tools for predicting the performance of JSC codes based on arithmetic coding, as well as new practical code constructions with improved distance properties.

Main players: FTW, CNRS

2-Performance bounds and practical constructions of JSC Wyner-Ziv codes, in particular with Bernoulli-Gaussian correlation models, with applications to relay coding (connection to WPR7).

3-Theoretical bounds and practical JSC code designs for continuously degraded broadcast channels, which may also be used as robust point-to-point JSC codes.

Main players: FTW, CNRS, Technion, CNIT

Task TR7.4: JSCC/D compatible with the OSI layers and study of practical long term solutions

The ultimate goal is to learn, understand, and optimize the various interactions that exist between the various building blocks involved in multimedia transmission over wireless channels, considering Wimax and LTE communication environments. Two situations will be studied. First, broadcasting (or multicasting) video to mobiles is believed to be a killer application for the next few years, and it would be of great interest to provide inputs on the possible performance improvements JSCC/D could bring to such applications. Robust decoding of H264/AVC-H264/SVC in a multi/broadcasting context is thus the first situation of interest. Second, video streaming in an ad hoc network is a much more challenging problem; here, problems that should be solved for such a demanding application will be identified. Compatibility of the JSCC/D strategy with the OSI layers will be studied to address the possible contradiction vis-a-vis the layer-independent model. In particular, compatible re-synchronization strategies of source codes in the presence of channel noise will be developed. Issues of cross-layer optimisation of the redundancy at the various layers will be addressed.

Main players: FTW, CNRS, CNIT