Performance comparison of diverse detection methods for wideband active sonar

In an active sonar context, transmitting a wide band code often implies processing gain for target detection in many situations such as in a dominant reverberation context, in a narrow band noises environment or when target backscatters signal specifically in certain frequency ranges. In this paper, performances of three detections methods dedicated to detection are discussed. First detector is built considering that backscattered signal and noise are white in all transmitted bandwidth: it corresponds to threshold signal level after the classical match filtering. Second and third detectors consider that backscattered signal level and noise variance vary from one sub-band to another: - Second detector is derived from the Generalized Likelihood Ratio Test (GLRT) of the following two hypotheses detection problem: H1 = {received signal in each sub band is a a sum of unknown attenuated signal + noise} / H0 : {received signal is made of noise on each sub-band}. Noise in each sub band is supposed Gaussian with diverse variance from one sub-band to another. This method is derived from one proposed in [1]. - The third detection method is based on "P sub-band detections among Q" criteria where Q is the number of considered sub bands. These kinds of methods were evaluated in [2] and [3]. The three methods performances are first compared with a white spectrum backscattered signal in a white noise then in a context where signal is strongly attenuated in some sub-bands.