Optimization technique for frequency estimation: Avoiding local minima

Abstract

A common practical problem when implementing signal model-fitting procedures is that of the frequency local minima. Unfortunately, conventional optimization methods, like Steepest Descent, Newton's Method and Conjugate Gradients (CG) are subject to this problem. What is worse, if the estimated frequency is not correct, the estimated signal amplitude and decay rate will be incorrect. In this paper, we propose a new optimization technique to find the global optimum in frequency estimation. Our method takes advantage of the attenuation characteristics of the signal and is incorporated into the CG algorithm. Various simulation examples illustrate the validity of the proposed approach: it can achieve the global optimal solution of frequency efficiently.