Overlapped arithmetic codes with memory

This paper describes a family of codes based on arithmetic coding with overlapping intervals. These codes are not uniquely decodable but the presence of correlated side information at the decoder can be exploited to achieve a vanishing decoding error probability, making them well suited for the Slepian-Wolf problem. Since these are source codes, they are also particularly well suited for sources exhibiting non-uniform symbol probabilities or memory; and the provided experimental results support this assertion. The construction of the codes is described, along with the corresponding soft decoding algorithm with side information. Finally, simulation results are given which compare very favourably against turbo codes in the presence of source memory.