MFXLMS for Active Noise Cancelling Documentation & Test Data
Hello,
I was wondering if anyone has used the MFXLMS in Sigma Studio successfully. I have read over the documentation here: https://wiki.analog.com/resources/tools-software/sigmastudio/toolbox/filters/mfxlms but was looking for more information to get the algorithm working in a real world scenario.
I have 2 mics and a speaker bar I would like to use for testing. The wiki says you should do a training period to calculate the path coefficients. How long should this training be and how does the length of the filter affect the performance? Is there an optimal placement of the hardware for better performance?
Does anyone have any test data using the algorithm to see how much attenuation they have achieve? I know this will depend on a lot of factors but I would just like any information at this point.
Thank you for your time,
Tannis
dsp
anc
adau
MFXLMS
I haven't used the MFXLMS function yet, but I can answer (somewhat) your learning time and placement questions.
The time that it takes to adequately learn a secondary path model isn't related to the MFXLMS algorithm as much as it is basic spectrum analysis. You are inputing some excitation sequence into your control actuator and detecting it with your sensor. Your goal is to converge on a linear and coherent model. Your concern is the degree with which you are exciting each frequency bin in your model which - although it is not directly related to model taps - is analogous.
Given that the hardware and plant are sufficiently linear and coherent, the amount of time it will take to converge on a good model will depend upon the energy distribution of the excitation sequence. Impulse, sine sweep, MLS, random, ... They all have different power/freq/time characterstics as well as differing abilities to excite nonlinearities (bad). So... the same thought/planning that would go into a spectrum analysis excitation sequence should be put into play when characterizing secondary path models for active control.
I recommend an old paper by Schoukens called "Survey of Excitation Signals for FFT based Signal Analyzers." It is as good as it gets in the explanation of the various sequences and their strengths/weaknesses.
Regarding placement... Per the above - you want good coherence. Keep that in the front of your mind when placing components.
Please log in to reply.