Three novel adaptive noise-cancellation techniques

Includes bibliographical references (pages 91-94)This research presents three novel adaptive noise cancellation techniques for reducing periodic interferences, such as engines, blowers, compressors, and fans. (1) The sub-band two-model adaptive HR noise cancellation scheme structure enables the cancellation of multiple sinusoidal interferences in a received speech signal. Compared with the classical LMS adaptive noise cancellation method of B. Widrow et al., this new structure has two major differences. First, the quadrature mirror filter (QMF) banks have been used to decompose signals into sub-bands. Second, a new constrained time-varying HR notch filter and adaptive HR bandpass filter were developed to form a two-model adaptive HR noise canceller to remove the multi-rate noise from a received signal. Computer simulations were performed using a set of actual speech signals corrupted by car engine noise. Experiments demonstrated that more than 20 dB signal-to-noise ratio (SNR) improvements have been achieved. (2) In the case that the reference signal is not available, a new noise control structure using a single-pulse excitation generator, the synchronous adaptive noise control with LPC noise synthesizer, was developed. The linear predictive coding (LPC) technique is employed along with the adaptive noise-cancelling technique to form this new structure. The comparison result shows that the proposed structure converges faster and has a better mean-squared error compared with Elliott and Darlington’s method, in which there is no LPC noise synthesizer involved. (3) An alternative structure, the synchronous adaptive noise control with multi-pulse excitation, using a multi-pulse generator is also presented. Because of the better exciting signal generated by the multi-pulse excitation generator, the LPC synthesizer can be removed to reduce the complexity. Due to the engine noise’s periodic and harmonic properties, the weighting process has been omitted to further reduce computation cost. A faster convergence rate and a more than 10 dB mean-squared error (MSE) improvement have been achieved by using the new structure, as compared with that of Elliott and Darlington.M.S. (Master of Science