Paschedag J., Schwarz J.

This paper suggests some algorithmic improvements and it reports about practical experience in active isolation of engine induced Vibration. The analysed control algorithms are implemented on an experimental setup that resembles the automobile structure of interest. All considered algorithms are of feed forward structure. They can be applied highly stable and effective. Although the causality condition is not met because of high acoustic velocity in metal, the successful implementation is possible because of the periodic nature of occurring signals. The first part of the presented work concentrates on implementation issues of the well known filtered-x LMS algorithm (FxLMS). For a proper functioning, the necessary modifications like step size normalization and introduction of a leaky factor are performed. In the second part, other LMS-based algorithms are explored, that, unlike the FxLMS, employ recursive adaptive filters. These algorithms on one hand potentially permit a reduced computational complexity. On the other hand they have some disadvantages like immanent ability of instability and multiple minima of the adaptation error surface. Guidelines for choosing the systems parameters in a way to avoid these unwanted effects have been determined and are presented. In a third part, alternative feed forward approaches are considered, that work exclusively for periodic excitation. A special condition of the considered application is the availability of a tachometer signal. It provides crucial information about period and spectrum of the disturbance signal. Further on, considered as a reference signal for the adaptive algorithms, it is free of feedback-influences. A detailed analysis of the resulting improvement opportunities for the different algorithms is performed and confirmed by practical results. The work concludes with a comparison of the applied methods regarding stability, computational costs and quality of suppression.