HOS-based harmonic retrieval: a deterministic formulation

Abstract

Given a single record, the authors consider the problem of estimating the harmonics of a signal buried in noise. The observed data is modeled as a superposition of sinusoids plus stationary, zero-mean additive Gaussian noise of unknown covariance. Novel higher-order statistics, referred to as 'mixed' cumulants, are appropriate for certain types of mixed processes. After a consistent HOS-based linear estimate, an improved estimate is computed via mixed-cumulant matching. Issues regarding quadratic coupling are addressed. The performance of the proposed method is examined via simulations.<>