Computational optimization of preditive filter for noise cancellation

Abstract

Noise cancellation of stationary random signals has been obtained at an isolated point in space. This has been done without using adaptive techniques, an error microphone, or any sort of “upstream” preview of a signal correlated with the noise source. The antinoise speaker is driven by a digital signal processor implementing an IIR filter using sampled input from a microphone placed at the point where noise cancellation is desired. By using a simple physical model of the total noise at the microphone, the coefficients of the DSP IIR filter can be defined. A linear prediction has been incorporated into the IIR filter in order to anticipate the level of sound that will be produced by the noise source in the future. Experimental cancellation results were attained by using an initial guess of the optimal IIR DSP implemented filter. The work presented here involves computationally finding the set of filter coefficients that gives the theoretically optimal reduction in noise. The total reduction in decibels of the entire noise spectrum is minimized subject to the constraint of feedback stability.