Joint fundamental frequency and order estimation using optimal filtering

Recently, two optimal filter designs for fundamental frequency estimation have been proposed with the first being based on a filterbank and the second on a single filter. The two designs are related in a simple manner and are shown to result in the same residual when used for cancelling out the harmonics of periodic signals. We propose to use this residual for estimating the number of harmonics by combining a noise variance estimate with an order dependent penalty term. This leads to a joint estimator of the fundamental frequency and the order based on the same criterion. Via Monte Carlo simulations, the estimator is demonstrated to have good performance in terms of the percentage of correctly estimated orders.