Multistage generalized adaptive notch filter with improved accuracy

Generalized adaptive notch filters (GANFs) are estimators of coefficients of quasi-periodically time-varying systems. Current state of the art GANFs can deliver highly accurate estimates of system variations´ frequency, but underperform in terms of accuracy of the coefficient estimates. The paper proposes a novel multistage GANF with accuracy improved in this aspect. The processing pipeline consists of three stages. The preliminary (pilot) frequency estimates are obtained first, then treated with a specially designed linear ilter and used to guide the coefficient tracking GANF, which works out the estimates of system coefficients. The proposed solution has considerably better performance than a single stage GANF or a simple two-stage approach consisting of the pilot frequency estimator and the amplitude tracking GANF only.