Detection and frequency estimation of narrow-band signals by means of the instantaneous-frequency distribution (IFD)

The instantaneous-frequency distribution (IFD) is an amplitude-invariant spectral representation applicable to signals such as speech. The performance of an IFD with time-averaging and squared-amplitude weighting as a narrow-band signal detector and frequency estimator, in comparison to the classical maximum-likelihood detector/estimator, is evaluated experimentally for a single-complex-pole Gaussian signal of known bandwidth but unknown center frequency, at known signal/noise ratio (SNR) in white Gaussian noise. It is concluded that the IFD is capable of detection and estimation performance approaching the theoretical optimum (i.e. that of maximum likelihood) to within the equivalent of a fraction of a dB degradation in SNR, at least over a range of signal bandwidths comparable to that of speech formants, for example.<
>