A new autoregressive estimator for the estimation of the frequencies of a sinusoidal process corrupted with noise

The authors propose an autocorrelation-based method for determining the autoregressive parameters of a signal containing sinusoids. This method does not use the zero-lag autocorrelation value of the signal and is thus immune to the error caused by an additive white noise present in the signal. The autoregressive parameters are then used to determine the frequencies of the sinusoids. The method requires only 2p autocorrelation values, where p is the number of sinusoids. Thus, this method finds applications where the number of autocorrelations to be calculated is to be a minimum. A comparison of the method with other methods shows that over some frequency range the method has the lowest error. A disadvantage of the method is that the signal should be sampled at a rate at least four times the highest frequency in the signal if sampled data are used to estimate the autocorrelation values.<>