Parametric modeling and reconstruction of acoustic signals

The authors report on the development and performance evaluation of parametric signal processing algorithms for extracting a mixed stochastic signal process from its wideband noise corrupted measurements. The signals under consideration are acoustic with periodic components masked by wideband colored noise. First, the unbiased autocorrelation function (ACF) sequence is estimated from the data. The higher lags of the ACF are used in a generalized version of the least-squares modified Yule-Walker equation estimator to obtain accurate sinusoidal frequency estimates. Then the relative sinusoidal amplitudes and phases are found by maximum likelihood estimation or by modal decomposition. Once the sinusoids have been completely characterized, the wideband components of the signal are modeled using AR or ARMA spectral estimation procedures