Fast and accurate frequency estimation with a gradient-based iterative algorithm

Nonparametric spectral estimation algorithms were applied to frequency estimation for their significant performance. To avoid the signal mismatch problem and to improve the frequency estimation accuracy, a new iterative algorithm was presented based on the minimum variance distortionless response (MVDR) spectrum. With given initial frequency, searching directions and adaptive steps were derived to update the frequency sequence, which converges to a local spectral peak as the scalar gradient function goes to zero. Without the spectral peak searching on predefined analysis frequency grids, the computation is saved. The proposed algorithm was also applied to multiple component frequency estimation with carefully selected initial frequencies.