Least squares based decoding for BCH codes in image applications

BCH codes in the frequency domain provide robust channel coding for image channel coding applications. The underlying problem of estimation of real/complex sinusoids in white additive noise may be formulated and solved in different ways. The standard approach is based on the least squares method and Berlekamp-Massey algorithm (BMA). In this paper we compare the performance of the BMA with other LS based algorithms including: minimum norm solution based algorithm (MNS), forward-backward linear prediction based algorithm (FBLP) and singular-value decomposition based minimum norm algorithm (SVD-MNA). Results of computer experiments show that the introduction of minimum norm solution, forward-backward prediction and the SVD decomposition may significantly improve the performance of the decoder in the case of the relatively low SNR. In selecting between the proposed algorithms a performance/complexity trade-off has to be considered.