SIGNAL SHAPE RECONSTRUCTION BY DCT-BASED FILTERING OF FOURIER SPECTRUM RECOVERED FROM BISPECTRUM DATA

The noise characteristics of complex valued signal Fourier spectrum recovered from bispectrum estimate are investigated. It is shown that additive Gaussian noise usually observed at the input of bispectrum-filtering signal reconstruction system induces non-stationary and nonGaussian fluctuations in real and imaginary components of recovered Fourier spectrum of a signal of unknown shape to be estimated. For improving this estimate it is proposed to apply a modified DCT-based filtering of recovered Fourier spectrum with adaptation to local variance. Simulation results obtained for different test signal waveforms and for various input signal-to noise ratios (SNRs) are presented. They demonstrate the efficiency of the proposed approach and its applicability in case of limited a priori information about signal and noise characteristics.