A new approach to single-tone frequency estimation via linear least squares curve fitting

Abstract

Presented is a new algorithm for estimating the frequency of a single-tone noisy signal using linear least squares (LLS). Frequency estimation is a nonlinear problem, and typically, methods such as Nonlinear Least Squares (NLS) (batch) or a digital phase locked loop (DPLL) (online) are employed for such an estimate. However, with the linearization approach presented here, one can harness the efficiency of LLS to obtain very good estimates, while experiencing little penalty for linearizing. In this paper, the mathematical basis of this algorithm is described, and the bias and variance are analyzed analytically and numerically. With the batch version of this algorithm, it will be demonstrated that the estimator is just as good as NLS. But because LLS is non recursive, the estimate it produces much more efficiently than from NLS. When the proposed algorithm is implemented online, it will be demonstrated that performance is comparable to a digital phase locked loop, with some stability and tracking range advantages.