ROBUST DECENTRALIZED DETECTION BY ASYMPTOTICALLY MANY SENSORS

We consider a decentralized hypothesis testing structure with asymptotically many sensors, each collecting a single datum. The sensors deploy robust test functions that are designed for outlier classes of hypotheses. The sensor outputs are transmitted to the fusion center for the global decision. In this paper, we concentrate on sensor-level decision making, and study the asymptotic performance of the decentralized detection scheme described above. In particular, we utilize the asymptotic relative efficiency performance measure, defined as the ratio of the number of sensors needed by the decentralized structure over the number of data needed by the centralized one to attain the same performance level. Our results indicate the superiority and the necessity of using robust statistical techniques.