Clustering Approach for Detection and Time of Arrival Estimation of Hydrocoustic Signals

Abstract

We focus on detection and time-of-arrival (ToA) estimation of underwater acoustic signals. These tasks are key enabling technologies for underwater acoustic sensor applications, including SONAR, acoustic communication, and depth detectors. Detection usually involves comparing the output of the matched filter to a detection threshold, and the first maxima that reaches the threshold is considered the ToA. The detection threshold is set by the expected distribution of the noise and the target false alarm, and detection probabilities. However, due to the strong multipath and the presence of transient noise in the underwater acoustic channel, target performance is hard to guarantee. Considering these problems and with no training data, a clustering approach is offered to identify the samples at the output of the matched filter as signal or noise. Then, detection is verified by testing if, at high likelihood, enough samples of type signal exist, and the ToA is set as the time instance of the first sample identified as signal. For clustering, we utilize the fact that the channel impulse response is long and sparse. The numerical simulations show that, compared with the matched filter test, the proposed approach greatly reduces the false alarm rate and improves the accuracy of ToA. This is obtained at a cost of complexity and only a slight decrease in the detection rate. The proposed method was successfully tested in a sea experiment conducted in the Mediterranean Sea at the water depth of 900 m.