Optimal subblock-by-subblock detection

We propose a recursive algorithm to compute the joint maximum a-posteriori (MAP) probability of a subblock of N consecutive symbols (i.e., a sliding window of length N) of a finite-state discrete-time Markov process of length K/spl ges/N observed in white noise given the whole block is received. This "optimal subblock-by-subblock detector" (OBBD, "vector MAP") is a generalization of the "optimal symbol-by-symbol detector" (OSSD, "symbol-by-symbol MAP"), which is obtained for N=1. The new algorithm improves applications with outer stage processing. This is indicated by investigating the average mutual information of a convolutional coding system. An example shows that the gain (in terms of average mutual information) by using joint probabilities could even exceed the gain by delivering soft OSSD outputs instead of hard outputs.<>