Divergence and Bayes error based soft decision for decentralized signal detection of correlated sensor data

Abstract

In decentralized cooperative sensing for cognitive radio, a few secondary users (SUs) sense the spectrum, process individual observation and then pass quantized data to the fusion center (FC), where the decision on signal present hypothesis or signal absent hypothesis is made. When the reporting channels between SUs and the FC are bandlimited and error prone, a quantization scheme was proposed recently based on divergence measures for independent observations. In this paper, we extend the design of quantizers to correlated sensor data. With the assumption that two SUs' observations are jointly distributed as bivariate Gaussian with identical marginals, we design quantizers based on both divergence measures and the Bayes error. Our simulation results demonstrate that a quantizer designed with the knowledge of known joint distributions outperform the quantizer designed with independent sensor data assumption. Thus, it is important to account for correlation in the quantizer design in distributed cooperative sensing.