A Gradient-Based Adaptive-step Algorithm for Magnitude Squared Coherence Spectral Estimation

Conventionally, the signal component frequencies are estimated by spectral peak search process, and suffered with the common signal mismatch problem (SMP). MVDR and CCA are typical nonparametric methods in magnitude squared coherence (MSC) spectral estimation. In this paper, a scalar cost function is developed based on the CCA MSC spectrum, where local peak indicates the estimation of signal frequency. Then, a gradient-based adaptive-step algorithm is presented to find the local peaks. In simulations, the proposed algorithm improves the frequency estimation accuracy, and SMP is avoided.