Hi I am using a classic nlms adaptive filter to minimize the error between an unknown LTI system and the output of an adaptive FIR. However I have no control over the stimulus, which is a mix of sinusoids that change slowly over time. I know that with this input, the filter may not adapt to the true impulse response, but as long as the error goes to a small value I don't really care. If the input contains a mix of large- amplitude sine-waves and low - amplitude sine -waves , I find that the larger- amplitude wave is eliminated very quickly but the smaller amplitude wave has a very long convergence time. I know that I could use multiple frequency bands to solve this, but that would take too many MIPS. Are there any other solutions? I experimented with adaptive whitening using the error output of a linear predictor, but this works better for colored noise than for sinusoids. Jim

# Adaptive filter

Started by ●November 30, 2013

Reply by ●December 1, 20132013-12-01

On Sunday, December 1, 2013 10:05:18 AM UTC+13, Jim craig wrote:> Hi > > > > I am using a classic nlms adaptive filter to minimize the error between an unknown LTI system and the output of an adaptive FIR. However I have no control over the stimulus, which is a mix of sinusoids that change slowly over time. I know that with this input, the filter may not adapt to the true impulse response, but as long as the error goes to a small value I don't really care. > > > > If the input contains a mix of large- amplitude sine-waves and low - amplitude sine -waves , I find that the larger- amplitude wave is eliminated very quickly but the smaller amplitude wave has a very long convergence time. I know that I could use multiple frequency bands to solve this, but that would take too many MIPS. Are there any other solutions? I experimented with adaptive whitening using the error output of a linear predictor, but this works better for colored noise than for sinusoids. > > > > JimRecursive-least squares or Recursive Affine Projection works better on such inputs.