Joint optimization of VQ encoder and channel encoder in VQ-based communication systems over noisy channels

Abstract

It is well known that an efficient data compression system removes the redundancy in the source and retains the useful part in an effort to reduce the data rate. This removal of redundancy, in turn, can introduce a great deal of sensitivity to the transmission noise or the storage device errors. The transmission or storage noise, if not dealt with properly, can result in significant degradations in the performance of the compression scheme. In this thesis we are concerned with nonredundant coding systems, i.e., coding without explicit error precaution. Several techniques are proposed to ensure robust transmission for digital communication over noisy channels. First, we address the index-assignment problem which is important part of the proper VQ design. Here, we propose an index assignment algorithm for Classified VQ codebook, the algorithm is composed of two parts, Inner Index Assignment (IIA) and Cross Index Assignment (CIA). IIA can be considered as a simplified and advanced version of Binary Switching Algorithm (BSA) and can be used for ordinary VQ codebook. CIA modifies the indices of CVQ codebook, which have been assigned by IIA or BSA, in order to increase robustness for channel error that changes the class information of the transmitted codevector into others. Second, a new fast clustering algorithm is proposed to speed up the design of the Channel-Optimized VQ (COVQ) codebook. The usefulness of the COVQ may be restricted due to the increased training time required for the design of the COVQ codebook. Starting with an initial codebook, the COVQ algorithm performs a partition of the training set, in which an exhaustive search procedure is required to map each training vector to its corresponding codevector in the current codebook. Hence, elevating the speed of searching codebook for the training vector set is one of the most effective ways of reducing the computation time of COVQ. The new algorithm is based on the use of the relation between codevecto...