On Fully-Distributed Composite Tests With General Parametric Data Distributions in Sensor Networks

We consider a distributed detection problem where measurements at each sensor follow general parametric distribution. The network does not have central processing unit or fusion center (FC). Thus, node takes some measurements, processing, exchanges messages with its neighbors and finally makes decision (typically the same for all nodes) about phenomenon of interest. can be formulated as composite hypothesis test unknown parameters where, in general, uniformly most powerful exist. This leads naturally to use Generalized Likelihood Ratio (GLR) test. As distribution (which could model spatial dependence data), implementation fully-distributed procedures demanding resources. For this reason, we study simpler (referred L-MP) which uses product marginals taken node, are easily estimated only local measurements. Although simple proposal still requires network-wide cooperation between nodes, number communications is significantly reduced respect GLR test, making it suitable choice severely resource-constrained networks. exploit full data, becomes important analyze statistical properties potential performance loss. done through analysis L-MP asymptotic Interestingly, despite fact that more efficient implement than obtain conditions under has superior Finally, present numerical results spectrum sensing application cognitive radios, showing gains terms performance, saving resources comparison other well-known approaches application.