Asymptotically optimal decentralized type-based detection in wireless sensor networks

We propose a novel decentralized detection scheme based on the method of types and type detectors. Contrary to conventional methods, the proposed decentralized type detection (DTD) scheme asymptotically achieves the performance of centralized detection, provided there exists an arbitrarily small (but non-zero) communication capacity between the sensing nodes and the decision center. The sensing nodes extract and transmit the type information of the observed data sequence, which is then recovered and used by a type detector at the decision center. The key to DTD is that the required network information flow (type information) is of asymptotically vanishing entropy rate but still sufficient for the hypothesis testing problem. The simple structure of type sensors greatly reduces system cost in terms of computational and communication capabilities for sensor nodes. In particular, the sensor nodes need not know the signal statistics under different hypothesis. Theoretical analysis and numerical simulations demonstrate the excellent performance of the proposed DTD scheme.