Our implementation sucessfully achieved system idetification and noise cancellation. Specifically, time varying noise was reduced. The effects of varying different paramters in the algorithm were also observed. It was found that the stepsize mu affects the rate of convergence, a larger mu leads to faster convergence, but too large a mu can produce divergence. The order of the filter used afffects the distortion of the desired signal. Because of the adaptive filter updates its coefficients to minimize the error between the primary and reference siganls, we get poor performance if the desired signal is similar to the reference signal. This method of noise cancellation is most useful when the reference noise that we have access to is large or delayed relative to the noise that is actually in the primary signal.

The next area to explore would be to implement different algorithms for updating filter coefficients, for example the Normed Least Mean Squares algorithm was briefly experimented with. The system gave an acceptable output for white noise reduction, but its performance could be greatly improved. This noise was much more difficult to eliminate since there exists little correlation between the reference nosie and the noise in primary, given that they are both completely random. Finally, we can customize our system for various real world applications.

<< Implementation <<

>> References >>