Non-linear preprocessing of heavy tailed reverberation

Echolocation in shallow water is often characterized by high spurious detection rates generated by heavy tailed (impulse like) boundary reverberation. While persistent echoes that are delay/Doppler consistent over several interrogations are best detected and left to a subsequent classifier stage, impulsive transient events in the time series can be suppressed by preprocessing the data stream with a nonlinear filter. A simple preprocessor consisting of a linear predictor with a thresholded non-linearity has been used to attenuate sporadic impulse like events and successfully reduce the heavy pdf tails of synthesized Rayleigh mixture reverberation matched to actual data. It is shown that the preprocessor was able to restore nearly all of the ROC performance of a matched filter and reduce the false alarm yet maintain the detection rate when applied to real data from a shallow water rocky bottom area.