ACTIVE ACOUSTIC NOISE CANCELLATION WITH AUDIO SIGNAL ENHANCEMENT BASED ON AN ALMOST-SYMMETRICAL TIME-VARYING AUTOREGRESSIVE-MOVING AVERAGE MODEL

The basic idea of active noise cancellation (ANC)is to generate a cont rollable secondary source which compensates a noise in a certain spati al region. In headset applications, it is desirable to generate: a sec ondary source which cancels the acoustic noise field, while preserving tile useful audio field. This requires that the cancellation algorith m possess an ability to differentiate between the useful audio signal and the cancellation residual at the secondary-sensor output. This is a formidable task for the conventional linear time-invariant filtering approach. since the Fourier spectra of acoustic noise and the audio s ignal occupy the same frequency bandwidth. In this paper, a two-sensor ANC scheme is presented with application to a noise-reduction headset . The active noise cancellation carried out is based on the concept of the separation of superimposed signals with complete overlapping spec tra by using an almost-symmetrical time-varying autoregressive-moving average model (ASTV-ARMA), The significant advantages of our proposed selective noise cancellation approach are the cancellation of the unde sirable noise field and the preservation of the useful audio wave fiel d. Simulation results indicate that a noise reduction of over 30 dB ca n be achieved, thus resulting in a high-duality audio in the presence of a strong environmental acoustical noise. (C) 1996 Acoustical Societ y of America.