Expectation Maximisation Algorithm for Detection of High Bandwidth Sonar Pulses

The use of high bandwidth pulses in active sonar systems can reduce interference from reverberation and variability due to channel fading. However, the matched filter detection performance is degraded more by acoustic multipath when the bandwidth is increased. The performance degradation due to multipath distortion results when the destructive interference of the paths cancels some frequencies reducing the similarity between the echo and the transmitted pulse required for matched filter operation. Increased bandwidth means that more frequencies are cancelled so distortion and degradation increases. If the acoustic properties of the channel are known well enough to predict the acoustic multipath characteristics, detection can be improved by matched filtering for the distorted signal rather than the transmitted signal. In cases where the channel properties are unknown, acoustic path delays and amplitudes must be estimated from the data itself. In this paper the acoustic path delays and amplitudes are first estimated using the Expectation Maximisation (EM) algorithm. The estimates are then used to integrate the return from each path recovering part of the loss caused by multipath distortion.