Multitree search decoding of linear codes

Tree search algorithms have a long history in computer science. In the coding literature, tree search algorithms have traditionally been used for decoding convolutional codes. Convolutional codes are linear codes with a special structure. Classic tree search decoders (most notably the sequential decoder) search one code tree in which the bits are ordered sequentially. We propose a multitree search decoder for arbitrary linear codes. We develop several algorithms for constructing code trees from the parity check matrix of a linear code. We propose algorithms that generate code trees, in which the bits appear in random order. We also show how to generate code trees specially designed to decode a given sequence received from a noisy channel. In such code trees, the ordering of the bits depends on the received sequence. Multiple code trees can be generated for each code and sequence. Specialized code trees for low-density parity check codes are also presented. The different code trees are explored with a new search algorithm. The algorithm is similar to the M-algorithm for convolutional codes; it explores a code tree with limits on the breadth of the explored subtree. An evaluation function is used to decide which node to expand at each depth. We present an evaluation function for general linear codes and an improved evaluation function for low-density parity check codes. Both are optimized for the proposed search algorithm. The proposed multitree search decoder achieves near optimal performance for short block codes. For longer block lengths, we propose to use tree search decoding to