Adaptive Fast Algorithm Principle And Performance Of Active Control

In the field of active noise control, multichannel active noise control systems are applied widely to expand the spatial extend and enhance sound reduction, however, the corresponding computational load for multichannel adaptive active control algorithm will largely be increased with the increase of the number of the channel. In this dissertation, three kinds of fast algorithms are derived and its performances are investigated theoretically in detail.First, the filtered-x LMS (FxLMS) algorithm, which be applied in an active noise control system, is studied systematically. For the FxLMS algorithm, the principle, implementation and its computational load are given and the effect of the physical parameters on the convergence coefficient and steady performances are studied.Two kinds of fast algorithms, i.e. partial update filtered-x LMS algorithm and combined filtered-e LMS algorithm are derived. In the first kind of fast algorithm, two partial update algorithm and their implementation approaches are examined respectively, and then their range of convergence coefficient and steady state characteristics are investigated. It is shown that their range of convergence coefficient are controlled by partial update coefficient and the delay of modeling secondary path low-pass filter.The combined filtered-e LMS algorithm is given and the implementation approaches and computational load are given respectively, and then its range of convergence coefficient are also investigated. It is shown that its computational load can be increased with numbers of channels, and by comparison of two kinds of fast algorithm, computational load of combined invers algorithm is decreased. Its range of convergence coefficient is controlled by the delay of modeling secondary path low-pass filter and its length.Finally, the convergence performances, computational load and steady state characteristics of three fast algorithms are compared by modeling secondary path. It is shown that the proposed combined filtere-e LMS algorithm is better than the other.