Application of adaptive noise cancellation to Coast Guard voice communications

A variety of approaches to the digital filtering of voice signals corrupted by background engine noise are presented. These approaches include the standard Least Mean Square adaptive noise cancellation algorithm, an optimum fixed weight filter, a special type of notch filter and frequency domain adaptive noise cancellation. The filters have been implemented both off-line using FORTRAN programs on an LSI-11/2 microcomputer, and in real time using thee Texas Instruments TMS 320 microprocessor and in PDP-11 assembly language using an LSI-11/2. Frequency domain adaptive filtering was seen to be superior to the LMS time domain algorithm because its much greater computational efficiency allowed the analysis of much longer filters. A digital filter that exploits the engine noise by having a notch at each harmonic was seen to be more effective than any of the adaptive filters. The digital filter equations are derived, starting with a particular periodic reference input. This adaptive filter is shown to be, in reality, an infinite impulse response, time invariant filter.