Detection of active linear FM sonar signals using the bispectrum

The paper focuses on the problem of a returning active sonar signal masked by environmental background noise. According to theory, if the original signal is sufficiently non-Gaussian the bispectrum of the received signal-plus-noise will contain only information due to the signal, and the white Gaussian noise will be suppressed. A bispectrum detector has been developed which utilizes the a priori knowledge of the input active sonar signal by selecting regions of the bispectral domain where the magnitude exceeds a significance test threshold. The presence of the returning sonar signal in noise is then detected by measuring bispectral content in those regions. Linear FM (chirp) signals of several frequencies and sweep ranges are generated to simulate active sonar input, and corrupted by additive white Gaussian noise to produce various signal-to-noise ratios. Receiver Operating Characteristic curves for these signals demonstrate that this new detector provides gain over a bispectrum detection method previously presented in the literature. Detection results are presented for the original signals with additional multipath returns