A robust model-based detector for active sonar

Abstract

The detection of targets using active sonar is often hindered by the adverse effects of propagating through an ocean environment and reflection off of the target. If the resulting received target echo is characterized as simply being spread, a detector that integrates the normalized matched filter intensity is appropriate. However, when the target echo has components that are separated in time delay rather than clumped, this leads to integrating large portions of reverberation and poor performance. A more appropriate characterization describes the received target echo as being composed of a number of paths with unknown amplitudes and delays. Two detector structures based on generalized likelihood ratio tests (GLRT) over the unknown path amplitudes and delays are compared with the post matched filter integrator (PMFI) detector. The GLRT detectors are designed by exploiting different levels of information from a propagation model so that they are robust to modeling inaccuracies. In particular, only the number of paths and the regions in delay space to search for them are determined from the propagation model. Estimating amplitudes and delays in the GLRT detector and increasing the integration region of the PMFI detector result in a loss if the SNR on an added path is not above some minimum level. Theoretical analysis illustrated that this estimation loss, quantified by the SNR on an additional path required one to retain constant probability of detection and false alarm, was not inhibitive and in fact decreased as the number of existing paths increased. ROC curve analysis using a propagation model to simulate target echoes illustrated how the GLRT based detectors can provide improved performance over the PMFI detector, with greater improvement as bandwidth and water depth increase and as range decreases.