Transient detection using the nonstationary bispectrum

Abstract

A detection statistic is described which exploits features in the three-dimensional response of the nonstationary bispectrum (third-order cumulant spectrum) for an assumed class of transient signals. The detection performance of the proposed detector is investigated for transients of this class in the presence of additive white stationary interference. The results are presented relative to the performance of a conventional power spectrum detector and a detection statistic based on the spectral correlation (second-order cumulant spectrum). The merits of the nonstationary bispectral detector relative to other transient detection approaches are also discussed.

Detection of random transient signals via hyperparameter estimation

“…Here, we compare the detectors developed in the previous section to the "power-law" detector. The situation is of FFT bins; under , each is distributed as unit- are exponential with mean SNR -note that this fixed-situation is that under which (8) was developed and does not match our Bernoulli model of (12). Comparison is on the basis of 100 000 simulations, and the ordering of the signal-containing bins under is relevant to none of the detectors implemented; hence, we arbitrarily choose these as It will be obvious that the proposed detector with unknown (i.e., estimated)…”

“…Most likely due to its relative insensitivity to starting point, the frequency domain appears to have become the home for these. For example, the effects of transients upon higher order spectra are exploited in [10]- [12], and correlation in the frequency domain is the focus of [13]. In [14], the assumed tendency of transient energy to be concentrated both in time and in frequency leads to a Gaussian-mixture timespectrogram model.…”