Frequency estimation from proper sets of correlations

As a complement to the periodogram, low-complexity frequency estimators are of interest. One such estimator is based on Prony's method and rely on phase information of the auto correlations. Without prior knowledge of the frequency (e.g., a given frequency interval), the frequency cannot be unambiguously estimated from a single correlation only. We introduce a new method of phase unwrapping using an arbitrary number (more than one) of correlations. From this arbitrary set of correlations, we propose a weighted average estimator. We derive the asymptotic performance and show how the correlation lags should be properly chosen. From a design aspect, there is often a restriction of using a fixed number of computations. In addition, we therefore propose a strategy to find a proper set of correlation lags subject to a given computational complexity. Finally, simulation results that lend support to the theoretical findings are included.