Bayesian design of decentralized hypothesis testing under communication constraints

We consider a distributed detection system under communication constraints, where several peripheral nodes observe a common phenomenon and send their observations to a fusion center via error-free but rate-constrained channels. Using the minimum expected error probability as a design criterion, we propose a cyclic procedure for the design of the peripheral nodes using the person-by-person methodology. It is shown that a fine-grained binning idea together with a method for updating the conditional probabilities of the joint index space at the fusion center, decrease the complexity of the algorithm and make it tractable. Also, unlike previous methods which use dissimilarity measures (e.g., the Bhattacharyya distance), a-prior hypothesis probabilities are allowed to contribute to the design in the proposed method. The performance of the proposed method is compared to a method due to Longo et al.'s and it is shown that the new method can significantly outperform the previous one at a comparable complexity.