Experimental results on the detection of embedded objects by a prewhitening filter

In this paper, a signal-processing technique for the detection of objects buried in the seafloor by exploiting an active sonar system Is proposed and assessed. The technique is based on the modeling of the reverberation phenomenon as an autoregressive process. The detector is composed of an adaptive prewhitening filter and a bank of matched filters. The autoregressive parameters are computed by algorithms (based on the modified-covariance function or on the higher-order statistics) that work on successive short reverberation segments. No echoes of the buried target have been used to arrange the matched filters but only echoes of the same target floating in free water. The proposed technique has been tested on an experimental data set related to a steel cylinder deeply buried in the seafloor and 20 in far from a parametric sonar source. The obtained performances have been compared with those achieved without the prewhitening stage. The results yielded by the described technique are impressive, in spite of the weak signal-to-reverberation ratio and the low (subcritical) grazing angle used during the experiments.