Multiple Model Adaptive Systems For Active Noise Attenuation

Abstract

The characteristics of :most active control systems change with time. In particular, the characteristics of the transfer functions between the secondary loudspeakers and error slensors (the "secondary path") can be time-varying. In many situations, an adaptive scheme to estimate these transfer functions is needed. This is in addition to the adaptive filter implementing the controller. Most adaptive control filters have used FIR structures based on filtered-x LMS algorithms. :Recently, Eriksson er al [ 11 showed that IIR structures are more desirable for the active control of duct noise in order to remove the poles introduced by the acoustic feedback and presented an algorithm to adjust the coefficients of an IIR filter using the recursive least mean square: (RLMS) algorithm of Feintuch [2]. Since both of these approaches require knowledge of the secondary path transfer function, some adaptive algorithms which simultaneoiisly estimate the transfer function of a secondary path have been presented [1,3]. Such adaptive techniques have a tendency to diverge when the parameters vary rapidly and it is difficullt to apply them to the multiple sensor multiple speaker cases [4] because there are too many parameters to be estimated in each step. We present a new algorithm using multiple models to reduce the tendency to diverge compared with previous adaptive algorithms under time-varying conditions. Since this approach requires only a small amount of computation, it may also be used in the multiple channel case. The block diagxim of the multiple model adaptive control (MMAC) technique for noise attenuation is shown in Figure 1.