Kernel-based cooperative robust sequential hypothesis testing
Afief Dias Pambudi, Michael Fauß, Abdelhak M. Zoubir · 2018
We consider a scheme for sequential likelihood ratio tests in cooperative sensor networks. Instead of employing parameterized families of probability densities, we estimate the densities under both hypotheses using a kernel-based method that does not require strong assumptions about the distributions. In order to make the test robust, two density bands are constructed that capture the uncertainties in the density estimates under each hypothesis and a pair of least favorable densities within the band model is determined. The likelihood ratio of the least favorable densities is then used as a test statistic based on which the cooperative sequential detection is performed. The performance of the proposed scheme, in terms of error probabilities and the average number of samples, is evaluated via Monte Carlo simulations and compared to alternative approaches.