A generalized turbo principle for approximate detection and decoding

Optimal joint detection and decoding is infeasible in all but the simplest of communication systems due to the required exhaustive search. A highly popular approximate method for solving this problem is the so-called Turbo principle, in which single-variable beliefs are exchanged between the individual components of the system in an iterative manner. In this paper, the Turbo principle is generalized to allow the exchange of multi-variable beliefs between components by specifying an overall region graph for the entire system. As a result, a larger part of the interdependency between components can be captured leading to a better overall approximation to the exact marginals. Consequently, the resulting soft-information is of a higher quality and a lower error rate closer to that obtainable from the exact marginals may be achieved at any decision point.