Effect Of Correlation On Signal Detection In Arctic Under-lee Noise

Signal detection in a large segment of nonGaussian and non-stationary Arctic under-ice noise, which contains both high power narrow-band and impulsive components, is examined. It is shown that the correlation functions of sub-segments of data change significantly, and if ignored, can degrade the performance of a detector. For a false alarm probability of 0.05 and a known constant signal, the matched filter was on average 17.6% better than a detector designed assuming independent noise sam- ples. It is also shown that pre-processing the data with an adaptive notch filter, then us4ig the matched filter will result in a further improvement of about 6%. Additionally, the effect two different signal shapes have on the performance of the matched filter is examined.