Research On Improved Distributed Arithmetic Coding

On the basis of the Slepian-Wolf theory and Wyner-Ziv theory, distributed source coding(DSC) is one of the hot topics in data coding. Distributed arithmetic coding (DAC) is a DSCimplementation which introduces arithmetic coding as the core of the distributed codec.According to symbol probabilities, the DAC encoder divides the current interval into partiallyoverlapped sub-intervals, which causes the ambiguity in decoder and brings large amounts ofbranches. Side Information (SI) is used to select the right decoding path from those branches.Limited by available memory, only partial branches will be saved and the right decoding pathmay be mis-pruned. The research of upgrading the DAC codec to enhance DAC decoding correctness is far from perfect up to now. By studying the procedure of branch metric and thecorrelated branches in decoding tree, two improved schemes are proposed which are based onthe DAC spectrum and random mapping respectively. The main research contents of thisstudy are as follows:(1) The DAC spectrum is introduced to improve the branch metric. Branches are createdwith equal probabilities to save possible decoding symbols when there is an ambiguity duringthe traditional DAC decoding. The a posteriori probability between SI and source is used tomeasure the importance of each branch. We take advantage of the DAC spectrum to estimatethe probability distribution of codewords by analyzing the intrinsic feature of the bitstream.The closed form of initial spectrum under certain conditions is solved and the probabilitydensity function of each decoding symbol is obtained through the function transform of initialspectrum. Then the intrinsic importance of each decoding branch is computed. Combing theDAC spectrum with the a posteriori probability, the importance of branches is consideredboth from the intrinsic feature of the bitstream and SI, and the branch metric is improved.(2) Random mapping is used to alleviate the influence of correlated branches. Thedecoding error can be caused by outputting a correlated path as the right decoding path. Wepre-process the original source with random mapping and input the mapped source into theDAC codec. The temporary decoding results are inversely mapped and the right decodingpath is selected. The tree structure of traditional decoding branches is disarranged by randommapping and the influence of correlated branches is weakened. The random mapping shouldbe implemented in a segmented way. The source is divided into several segments and each segment is randomly mapped. The binary sequences after mapping are handled as a whole bythe codec. The temporary decoding results should also be inversely mapped in a segmentedway.(3) The improved implementation based on DAC spectrum and random mapping isproposed to reduce the decoding error ratio with better branch metric and less correlatedbranches during the decoding. The experimental results show that the decoding error ratio ofthe proposed implementation is lower than the traditional DAC codec. The influence ofdifferent parameters such as the segment size for random mapping is analyzed through thecomparison experiment.