engleski

Adaptive Instantaneous Frequency Estimation of Noisy Nonstationary Signals

Methods for a noisy nonstationary signal instantaneous frequency (IF) estimation from its time-frequency representation (TFR) have been proposed. The IF estimation error exhibits bias and variance dependant on the TFR analysis window width. Due to the opposite tendency of bias and variance with respect to the lag window width, a proper width has been found as the one providing the bias-to-variance tradeoff that reduces the estimation error. The adaptive window widths have been obtained by the improved intersection of confidence intervals (ICI) rule, proposed in the thesis. In the case of multicomponent signals, the components IF estimation was preceded by their localization and extraction from the signal TFR using the modified adaptive algorithm proposed in this thesis. The IF estimation has resulted in error reduction by up to 73% for monocomponent signals and by up to 23% for multicomponent signals in terms of the mean squared error when compared to the original ICI-based method. It has been also shown that the estimates are significantly less sensitive to the initial window widths set selection. Furthermore, the highest improvement in the IF estimation accuracy was achieved in high noise environments, for which many of the existing methods are known to fail.

Instantaneous frequency estimation; Time-frequency representation; Nonstationary signals; Monocomponent signals; Multicomponent signals; Components extraction; Adaptive lag window width; Relative intersection of confidence intervals (RICI) rule; Improved intersection of confidence intervals (ICI) rule

Naslov rada na hrvatskom jeziku: Adaptivna procjena trenutne frekvencije zašumljenih nestacionarnih signala