Improving Detection of Unknown Signal with Unknown Duration using an Information Criterion

Detection using a noisy reference signal is a problem that arises frequently, in particular in passive radar applications. In this paper, we go further on the incorrect knowledge of the signal, as its time-support is supposed to be unknown - either too short or too long. We first show and illustrate that this additional error on the signal model deteriorates the detection performance. Two strategies to recover the time-support are proposed, the first one based on the Akaike Information Criterion, and the second one on optional additional constraints on the signal. The improvement in detection performance offered by these strategies is studied through numerical simulations.