A Near-Optimal Iterative Algorithm via Alternately Optimizing Sensor and Fusion Rules in Distributed Decision Systems

For parallel distributed sensor systems with statistically independent sensor decision rules, Chair and Varshney in [2] has obtained the optimal fusion rule. On the other hand, under a given fusion rule, the optimal sensor compression rules have been proposed by Zhu et al. in [21], [23] for generally distributed and dependent sensor observations. An open problem is how to simultaneously obtain an optimal fusion rule and optimal sensor compression rules for general parallel distributed sensor decision systems. Obviously, the exhaustive method for searching for the optimal fusion rule is computationally intractable. For general parallel distributed sensor decision systems, we provide necessary conditions of an optimal fusion rule and optimal sensor compression rules and propose a computationally efficient iterative algorithm to simultaneously/alternately search for a fusion rule and sensor compression rules by combining both Zhu´s and Chair and Varshney´s methods. Moreover, the algorithm is extended to multiple bit compression and Network decision systems. Numerical examples show that the fusion rule obtained by the algorithm is in most cases the same as the optimal fusion rule obtained by the exhaustive method, therefore, it is effective and near optimal.