Cross Evaluation of Detection Schemes for Sparse Signals

Abstract

We consider environments where sparse signals are embedded in additive white noise. We consider specific signal models and cross-evaluate previously derived parametrically optimal, robust and tree-search policies for the detection of signal presence, in terms of the a posteriori probabilities of correct detection they induce. We specifically present numerical results for the case of a constant signal embedded in additive white Gaussian noise and the signal presence per observation being generated independently by a Bernoulli variable, in both the presence and the absence of data outliers.