Tracking and Recovering Multiple Sinusoids by an Adaptive Resonator Bank

N. Chernoguz (Israel)


Adaptive Filters, Sinusoid Tracking/Recovering


The problem of tracking and recovering multiple sinusoids from a composite noisy signal is considered in the context of adaptive notch filter (ANF) approach. The standard ANF, a common frequency tracker, is not appropriate for the signal decomposition due to the underlying high-order polynomial, degenerated ARMA model. Reasonable alternative sinusoid decomposition schemes are the Costas estimator-predictor filter bank, cross-feedback loop, closed-loop resonator bank (CLRB), spectral observer, MIMO signal separation and some others. As shown, these configurations represent different arrangements with an identical transfer function. An attractive implementation form, CLRB, was combined with a standard adaptation technique resulting in a novel filter, adaptive resonator bank (ARB). The latter is quite competitive with the ANF in tracking time-varying frequencies and crucially outperforms it in estimation of stationary (or slowly varying) closely spaced tones. The ARB doesn’t differ in computational cost from the parallel of cascade ANF, while its recovering ability is equivalent to that of the state-space spectral observer.

Important Links:

Go Back