ABSTRACT :
The focus of this thesis is to develop performance-enhancing algorithms for LMS/NLMS adaptive FIR noise canceller. The wide-ranging applications of adaptive noise cancellation include hearing aids and active noise control systems in aircraft.
In general, the standard LMS adaptive noise canceller suffers from performance deterioration, particularly when the required FIR filter tap length is large and the input signals are highly autocorrelated (e.g. speech). Moreover, in the event when the desired signal ‘leaks’ into the reference sensor (indicated by red arrow), the asymptotic performance of the adaptive filter decreases further.
To resolve the signal leakage problem, a signal separation algorithm known as Symmetrical Adaptive Decorrelation (SAD) was implemented. This algorithm engages the use of an additional adaptive filter with symmetrical algorithm, to extract the desired signal and noise separately.
Next, in a bid to reduce the adverse effects of large parameter dimension on the LMS adaptive filter, a Parameter Detection technique was implemented. This technique was further complemented by a signal prewhitening scheme to handle highly autocorrelated signals.
Finally, a hybrid algorithm was derived by combining the Symmetrical Adaptive Decorrelation algorithm and the Parameter Detection technique. Intuitively, the resultant algorithm should enjoy the advantages provided by each scheme. Furthermore, the use of this hybrid algorithm can be possibly extended to general signal/source separation applications.
No comments:
Post a Comment