On-line modeling techniques for active noise control in the duct
MetadataShow full item record
An Active Noise Control (ANC) system uses an adaptive cancellation filter to generate a secondary noise to cancel the primary noise. The adaptation algorithm used in the ANC system is the filtered-U RLMS algorithm, which requires the estimation filter to identify the error-path correctly. In order to track the dynamic error-path changes, an adaptive on-line modeling technique is necessary. Since there exists a complicated interaction between the cancellation and estimation filters, the on-line modeling technique is difficult to implement in realtime applications. This thesis proposes a new on-line modeling technique based on an automatic direct on-line modeling control algorithm. This will reduce the interaction between the cancellation and estimation filters such that the on-line modeling technique can be implemented in real-time. This automatic direct online modeling control algorithm uses the cross-correlation estimate ρ(η) between the error signal and the cancellation filter output to control the adaptation of the estimation filter. Since the cross-correlation between the error signal and the cancellation filter output is low after the ANC system has converged, the new algorithm freezes the adaptation of the estimation filter. When the error-path changes, the cross-correlation estimate is increasing due to the reduction of noise attenuation. Therefore the estimation filter updates its coefficients again. The results of the computer simulation and real-time experiment show that the automatic direct on-line modeling control algorithm improves the performance of the direct on-line modeling algorithm. Three other on-line modeling techniques were also evaluated and are as follows: additional random noise on-line modeling, overall system on-line modeling, and robust adaptive on-line modeling algorithms. The acoustic feedback from the cancellation filter output to the input microphone not only causes stability problems for the ANC system, but also makes it difficult for the estimation filter to identify the error-path correctly. Computer simulation results show that neither the additional random noise algorithm nor the overall system on-line modeling algorithm can remove the bias in the estimation filter because of the acoustic feedback. Although the robust adaptive on-line modeling technique shows better performance, the small step-size used in the algorithm makes it difficult to implement in real-time. Therefore, the most important contribution of this thesis is that the proposed new algorithm supports a more robust ANC system to perform on-line modeling in real-time.