Fundamental frequency smoothing for nonstationary multi-harmonic signals

A new estimator of the instantaneous fundamental frequency of a signal with multiple harmonic components is described. The proposed algorithm is a three-step procedure. First, the fundamental frequency is estimated using a bank of frequency tracking algorithms set for different signal harmonics. Then the smoothed estimates are evaluated by means of backward-time filtering of the results of tracking. Finally, the estimate of the instantaneous fundamental frequency is computed as a weighted sum of partial estimates. It is shown that the smoothing and weighting steps significantly increase both accuracy and robustness of frequency estimation.