Glówka T., Figwer J.

In the paper a new approach to the on-line identification of electro-acoustic plant models for active noise control (ANC) system, based on the higher-order spectra (HOS) is presented. Adaptive ANC algorithms are parameterized with the models of the secondary and feedback paths. If these paths are time-varying, the models have to be updated during ANC system operation. Several problems like inherent feedback loops, low signal-to-noise ratio, and correlation of the identified path input and disturbance signal make on-line identification very difficult. In this case classical identification algorithms often give biased and inconsistent estimates. The proposed solution employs HOS and signal averaging. HOS contain some additional information about processed signals that is not available to extract using second-order spectrum i.e. power spectral density. The most useful identification property of HOS follows from their definitions: they are identically zero for some processes, including Gaussian processes. So if the disturbance signal is Gaussian, it theoretically doesn’t influence the identification results, and in practice this influence is significantly reduced. The proposed technique requires a special, non-Gaussian, excitation signal. In the experiments the excitation sequence is repeated several times and the data is averaged. Averaging enhances signal-to-noise ratio without deteriorating noise attenuation, and improves Gaussian properties of the disturbance. This allows to obtain unbiased estimates. In the paper results of the proposed identification methods are provided and compared with results of classical methods. The estimates are computed on the basis of data acquired in the laboratory experiments as well as in the computer simulations.