Attempts to perform self-noise reductin of own combat vehicle and weapon system acoustic signatures in aero-acoustic sensor signals using classical noise cancellation algorithms have not been very effective. Classical algorithms are based on estimation of autocorrelation and cross correlation functions from the data under a stationary (non-time-varying) assemption or under the assumption of a very slowly time-varying signal (for example, the Widrow LMS algorithm makes this assumption). However, a time-frequency analysis of the acoustic signature of an actual combat vehicle shows that these assumptions are not valide, so it is no suprise that the classical methods do not work well. It is shown in this proposal that not only are the self-noise signatures highly time varying, but there are further exploitable signature characteristics that can be used to improve the reduction or cancellation of the self noise. These include a high degree of spectral redundancy (or spectral correlation) and evidence of nonlinear effects in the generation of the combat vehicle exhaust acoustic signature. This proposal addresses signal processing alforithms that exploit the highly time varying nature of the signature and the spectral redundancy inherent in the signal.
Keywords: adaptive noise cancellation adaptive beamforming spectral redundanch spectral correlation
