Asymptotic Optimality Theory Of Decentralized Sequential Hypothesis Testing In Sensor Networks

In fact, the decentralized sequential hypothesis testing problem is always studied insensor networks, where first a set of sensors receive observations independently, thensensors send the summary messages to the fusion center, making a final decision.When we study the decentralized sequential hypothesis testing problem In the sensornetworks, according to the sensors whether have full access to their own observations ofinformation, the sensor networks can be divided into two categories: system with full localmemory and system with limited local memory. When study on sensor networks, it is canestablish the corresponding asymptotically Bayes sequential test respectively for these twotypes. From studying these two types, we can get an interesting but very useful inference,and a new minimax formulation i.e. optimal stationary sensor quantizers, which isproposed in the case of additive Gaussian sensor noise.Finally, our results that empirical model checking in the adoption of XuefengMountain highway in Hunan province confirmed the possibility of new minimaxformulation. The studying show that, feedback informations from the fusion center do notimprove the asymptotic performance in the system with full local memory, but, even aone-bit, one-shot feedback also can obviously improve incremental performance in thecase of limited local memory.