N. Kudoh, S. Numahata, and Y. Tadokoro (Japan)
Digital signal processing, active noise control, filtered-xLMS algorithm, adaptive frequency estimation, andnarrow band signals
In many fields of active noise control (ANC), the filtered-x least mean squares ( LMS ) algorithm and its relatives are popular ones[1],[2], because of their simplicity and their relatively low complexity. In these algorithms, the input signal to the algorithms is, not the signal obtained from the reference sensor, but the signal filtered by the plant model, which must be identified in advance. As well known, the usage of the signal filtered by the plant model, causes two major problems [2]: the delay between adaptive FIR filter and the error signal; the well known eigenvalue spread in the autocorrelation matrix of the filtered input signal to update adaptive filter coefficients. These mean that a smaller adaptation gain must be used, namely, the convergence slows down. In this article, the active noise control is proposed for awkward sound generated by rotating machines as ventilation equipment with a duct. The reference signal can be approximated into plural narrow band signals by estimating each resonant frequency and then extracting around its vicinity. As this approximation leads to that frequency characteristics of the plant model around vicinity of resonant frequencies are only taken account, there is no need to identify the overall characteristics of the plant model in advance. In the proposed method, lower order adaptive filters are only needed to adjust to the plant model with on-line manner. Furthermore, It is expected that above two problems are eased, because the approximated narrow-band signals are used in the proposed method. Finally, it is shown that the proposed method has almost the same performance as the filtered-x LMS algorithm with less computational load.
Important Links:
Go Back
