Improving the iterated MAP decoder by remapping.

Abstract

In this thesis, we present a method of rectifying a deficiency intrinsic to iterative MAP decoding. The decoding of multidimensional product codes using an iterated symbol-by-symbol maximum a posteriori (MAP) decoder was recently shown to have good performance. Although iterating the MAP process was found to improve the BER performance, it causes overly optimistic estimates of bit reliability to be output. This inaccuracy reduces the usefulness of the iterative MAP decoder, when it is used as the inner decoder in a concatenated coding scheme. We investigated several modifications to the basic MAP algorithm in an attempt to improve the accuracy of the reliability estimates. The most effective approach found was one termed "remapping" in which the reliability estimates output by the MAP process are mapped to more realistic values using functions generated from error statistics. Remapping was found to significantly improve the accuracy of reliability estimates output by the iterative MAP decoder. However, this improvement comes at a small cost in BER performance.