Distributed Detection using an Adaptive Fusion Processor

It is well known that the optimal signal detection performance in a distributed system is obtained if both the local processors and the fusion processor are designed using apriori known statistics for the local decisions. In this work, we consider the case when the local decision statistics are not known apriori, but instead deviate from some known nominal value. Specifically, we examine the case where the received local decision statistics are known except for an unknown channel transition probability (an unknown probability that the channel may introduce an error). We develop an estimation procedure for the performance characteristics of the local processors using the local decisions received by the fusion processor. The fusion processor design adapts to account for the estimated local performance characteristics. Performance of the adaptive fusion network is significantly better than the fixed-structure nominal network. Furthermore, the adaptive fusion network performs nearly as well as the optimal omniscient system.