I am using an adaptive RLS adaptive filter for noise cancellation. My sampling freq. is 500 Hz, but I am interested in only frequencies of up to 60 Hz. I filter the input and the reference signal to the desired frequency range and then apply the adaptive filter. The adaptive filter does a good job at removing the noise. But, since I do not need such a high sampling freq. when my highest freq. is 60 Hz. I decimate by a factor of 4 after running an anti-aliasing filter and before applying the adaptive filter. Even if the input and the reference signal still display high correlation at the decimated rate, the adaptive filter does not perform well and produces spikes in the output. I wonder what the reason could be? Would it help if I use conditioning on the input, i.e. add some white noise before adaptive filtering (I am thinking in eigen value analogy and the smallest corresponding to white noise)?
Thanks in advance.