Recursive weighted robust least squares filter for frequency estimation

A novel weighted robust least squares estimation approach to the design of recursive frequency estimator for single tone sinusoid is presented. The frequency estimation problem is reformulated as the identification of slowly varying parameters subject to a linear time-varying system which contains the stochastic parametric uncertainties in the measurement matrix. By employing the statistical compensation scheme, the proposed robust frequency estimator successfully eliminates the scale-factor error of nominal weighted least squares frequency estimator. The algorithm shows accurate frequency estimation performance and wide range of robustness in the presence of severe sensor measurement noises. By incorporating the forgetting factor to the estimator, the algorithm can achieve fast convergency and adaptability. Moreover, since it requires small amount of computations compared to the existing estimators, it is attractive for real-time implementation. Theoretical basis and performance evaluation results of this technique are described.