Analysis Using Bi-Spectral Related Technique.

摘要

This report investigates the use of Bi-spectral related techniques to extract detection/classification clues from time frequency representations. Earlier results have indicated that 1.5-D spectral techniques (a degenerate version of the Bi-spectrum) has potential Signal to Noise Ratio (SNR) gain over conventional techniques. This is partially due to the rejection of Gaussian like perturbations by the cumulant based techniques. The third order moment for Gaussian zero mean random processes is essentially zero. It is assumed that (1) the independence of the Instantaneous Power Spectrum (IPS) from analytic signal representations and (2) the Gaussian character of the noise causes minimal distortion of the output vaxiables, hence permit signal characterization at low to moderate SNRs. The per rm ce is demonstrated on (1) synthetic signals and (2) on segment(s) of the NOSC data set. Spectral estimation, Instantaneous spectrum, Bi-spectrum, Cumulant based spectrum, Detection.