Neural Filtering for Active Noise Suppression for Diesel Engines

A new system to active noise suppression for small diesel engines is proposed. The novelty of the proposed system lies in its neural filter, which is used in lieu of the commonly used linear adaptive transversal and recursive filters. The neural filter has a more genral structure and has been proven to approximate any nonlinear dynamic system to any desired degree of accuracy, provided that the neural filter has enough hidden neurons. This property is expected to generate a very good antinoise for a noise process that can not be closely approximated by a linear autoregressive moving average process in a short sliding window of time. The neural filter can be implemented on commercially available neural network chips (e.g. Intel's 8OI7ONX ETANN), which are small and energy frugal. The extremely fast processing speed of these chips will allow cancellation of high frequency components of the diesel engine noise, eliminating the need for passive noise suppression measures. The neural filter can be trained to ignore strong acoustic interferences picked up by the residual error microphone(s) . This robustness is very important for an application on an UAV, because the loud propeller noise can upset and confuse such conventional algorithms as filtered-X LMS and filtered-U RLMS.