Abstract
Multimedia transmission over time-varying wireless channels presents a number of challenges beyond existing capabilities conceived so far for third-generation networks. Efficient quality-of-service (QoS) provisioning for multimedia on these channels may in particular require a loosening and a rethinking of the layer separation principle. In that context, joint source-channel decoding (JSCD) strategies have gained attention as viable alternatives to separate decoding of source and channel codes. A statistical framework based on hidden Markov models (HMMs) capturing dependencies between the source and channel coding components sets the foundation for optimal design of techniques of joint decoding of source and channel codes. The problem has been largely addressed in the research community, by considering both fixed-length codes (FLC) and variable-length source codes (VLC) widely used in compression standards. Joint source-channel decoding of VLC raises specific difficulties due to the fact that the segmentation of the received bitstream into source symbols is random. This paper makes a survey of recent theoretical and practical advances in the area of JSCD with soft information of VLC-encoded sources. It first describes the main paths followed for designing efficient estimators for VLC-encoded sources, the key component of the JSCD iterative structure. It then presents the main issues involved in the application of the turbo principle to JSCD of VLC-encoded sources as well as the main approaches to source-controlled channel decoding. This survey terminates by performance illustrations with real image and video decoding systems.
Highlights
The advent of wireless communications, in a global mobility context with highly varying channel characteristics, is creating challenging problems in the area of coding
The assumption prevailing so far was essentially that the lower layers would offer a guaranteed delivery service, with a null residual bit error rate: for example, the error detection mechanism supported by the user datagram protocol (UDP) discards all UDP packets corrupted by bit errors, even if those errors are occurring in the packet payload
When the coder can be modelled as a finite-state automaton, maximum a posteriori (MAP), MPM, or minimum mean square error (MMSE) estimation of the sequence of hidden states X0N can be performed on the trellis representation of the automaton, using, for example, BCJR [19] and soft-output Viterbi algorithm (SOVA) [18] algorithms
Summary
The advent of wireless communications, in a global mobility context with highly varying channel characteristics, is creating challenging problems in the area of coding. The resulting dependency structures are well adapted for MAP (maximum a posteriori) and MPM (maximum of posterior marginals) estimation, making use of soft-input soft-output dynamic decoding algorithms such as the soft-output Viterbi algorithm (SOVA) [18] or the BCJR algorithm [19] To solve this problem, various trellis representations have been proposed, either assuming the source to be i.i.d. as in [20, 21], or taking into account the intersymbol dependencies. In [35, 36, 37], the authors remove the memory assumption for the source They propose a turbo-like iterative decoder for estimating the transmitted symbol stream, which alternates channel decoding and VLC decoding.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have