Non recursive multi-harmonic least squares fitting for grid frequency estimation

A new method to approximate the least squares multi-harmonic fitting is proposed. The basic idea is to expand in Taylor's series the derivative on omega of the least squares cost function around a central value, so reducing the frequency estimation to a calculation of a root of a polynomial. In this way the method provides a frequency estimation in a closed form avoiding the recursion that is necessary in the classical approach. The results show that the proposed algorithm reaches the Cramer-Rao bound in a narrow range of frequency around a pre-estimation. Increasing the approximation orders of the Taylor's expansion the range of maximum accuracy widens. This method is particularly suitable in grid frequency estimation due its low variability. The proposed algorithm, preserving the accuracy, requires an execution time up to 8 times lower compared to a single iteration of the classical recursive approach. (C) 2015 Elsevier Ltd. All rights reserved.