Near ML Performance for Linear Block Codes Using an Iterative Vector SISO Decoder

Inhalt:
In this paper, we present an iterative soft-decision decoding algorithm applicable to binary linear block codes. The algorithm uses a modified Vector SISO algorithm that utilizes soft input information to form candidate codewords. A threshold test is used to determine whether the codeword is likely to be the ML codeword. If the codeword is not accepted, the decoder biases the input vector “away” from the candidate codeword and the decoding repeats this process up to a maximum number of decodings. Occasionally, this process does not find an acceptable codeword. When this happens the algorithm perturbs the input vector by modification of the sign of the input values and repeats the decoding process. The average complexity of the algorithm at reasonable Eb=N0 is low and the performance is near ML. A detailed description of the algorithm and simulations results are presented for low rate linear block codes. We present results for (80,40,16), (104,52,20) and (168,84,24) quadratic residue codes and (256,191,18) and (128,29,44) extended BCH codes.