A Fast and Stable Direct-MAP Implementation of Turbo Decoding

Abstract|Most implementations of turbo decoding make use of the so-called \log-MAP algorithm", which has been widely viewed in the literature to be faster and more numer-ically stable than a \direct-MAP " implementation. In this paper, it is demonstrated that the direct-MAP approach can be made numerically stable, and that the iterative pro-cess for this direct-MAP implementation is faster than the log-MAP approach on modern microprocessors, even if a lookup table is used to simplify the log-MAP calculations. Furthermore, since the direct-MAP algorithm does not rely on any approximations, the bit error rate performance of the direct-MAP algorithm is slightly better than that produced using the log-MAP algorithm. I