Non-parametric detection of unknown correlated multichannel transients with sequential and block hypothesis testing in multiresolution spaces

The problem of detecting transient multichannel signals of unknown waveform shape has been widely studied due to the numerous applications associated with it. Nevertheless, the particular case of unknown source waveforms impinging on an array of closely spaced sensors and buried in strong spatially and temporally correlated noise process has received little attention due to its inherent complexity, but yet arises in many array signal processing applications. In this work, we extend our multiresolution generalized likelihood ratio test detector (MRGLRT) to the problem of determining the number of observations needed to make a decision in the statistical hypothesis-testing problem. We reformulate the MRGLRT detector and compare two strategies for the hypothesis testing, namely, sequential and block tests. We show that, on the average, a smaller number of observations are needed to make a decision in either case. The strength of the proposed detector lies in the fact that it is derived under no assumptions about the signal and/or the noise process. We illustrate through simulated and experimental data the superiority of the new detector over the traditional time domain detector in many aspects.