Adaptive Filter Parameter Space and Validation


  1. To test and validate the LMS adaptive filter algorithm for noise cancelation
  2. To explore the parameter space for the algorithm
  3. To test the algorithm under simulated real world (white noise) conditions


Test, validation and exploration:

  1. The LMS algorithm from Active Noise Control Systems and Digital Signal Processing: A Practical Approach was coded in C using Borland C++ V3.1
  2. The "noise" source was a 64 sample waveform generated by using a sine wave of period 8 samples and amplitude 10 units
  3. The "quiet" signal was the same sample waveform delayed by 4 bins
  4. The noise and quiet waveforms were fed into the LMS algorithm.
  5. The LMS output was the "error" waveform.
  6. The waveform was then plotted using Excel 95.
  7. A graphic showing convergence vs number of filter taps was generated
  8. A graphic showing convergency vs XXX parameter was generated

Simulated real world test:

  1. White noise was generated by Cool Edit and saved as a text file
  2. The LMS algorithm was then applied to the white noise
  3. The original noise, a 16 tap solution and a 64 tap solution were graphed


Convergence vs filter taps
For this data there seemed to be a law of diminishing returns after about 8 taps (the period of the noise source).

Convergence vs u-size
The u size experiment was conducted with a 8 tap filter.

white noise results
The filter settings were very touchy. 128 and 256 tap settings were attempted but they did not converge. I suspect that we are getting cumulative errors in the algorithm.


dak 10/12/97