Performance of sample-covariance-based adaptive sonar detectors

This work investigates the performance of sample-covariance-based adaptive algorithms within a passive sonar detection framework. In passive sonar processing, training data for adaptive weights is necessarily the same as test data. Because of this, adaptive sonar detectors often suffer from target self-nulling in the presence of mismatch. This suboptimality of adaptive sonar detectors is quantified by comparing their performance with that of corresponding adaptive detectors that do have target-free training data available for adaptive weight computation. Additionally, the effect of diagonally loading the sample covariance matrix to reduce target self-nulling in adaptive sonar detectors is also quantified.