Forums

On Robustness of LMS and RLS

Started by Huan Tan February 16, 2004
Dear All,

I am currently simulating the performances of LMS and RLS adaptive filters.
Specifically, I am looking at their robustness. Theoretically, in H-infinite
sense, LMS is a robust filter whereas RLS is not. Through my simulation, I would
like to show or prove this theory.

Can anyone provide me with suggestions on how, in my simulation, I can create
scenarios through which I can show the superiority of LMS over RLS in terms of
robustness? For example with the filter input, what should I do to the filter
input such that I can show that RLS is less robust than LMS?

Thanks in advance for your advice.

Huan.



Hi Huan,

This about what "RLS is not so robust" really means, and that should
tell you how to highlight its weaknesses. Of course, LMS has plenty of
weaknesses for echo cancelling (strong spectral lines, wandering, etc.)
so make sure you hit the weakness you want, or LMS might look just as
bad. :-)

Regards,
Steve Huan Tan wrote:

>Dear All,
>
>I am currently simulating the performances of LMS and RLS adaptive filters.
Specifically, I am looking at their robustness. Theoretically, in H-infinite
sense, LMS is a robust filter whereas RLS is not. Through my simulation, I would
like to show or prove this theory.
>
>Can anyone provide me with suggestions on how, in my simulation, I can create
scenarios through which I can show the superiority of LMS over RLS in terms of
robustness? For example with the filter input, what should I do to the filter
input such that I can show that RLS is less robust than LMS?
>
>Thanks in advance for your advice.
>
>Huan.


Le Lundi 16 Frier 2004 9:39 PM, Huan Tan a rit :
> Dear All,
>
> I am currently simulating the performances of LMS and RLS adaptive filters.
> Specifically, I am looking at their robustness. Theoretically, in
> H-infinite sense, LMS is a robust filter whereas RLS is not. Through my
> simulation, I would like to show or prove this theory.
>
> Can anyone provide me with suggestions on how, in my simulation, I can
> create scenarios through which I can show the superiority of LMS over RLS
> in terms of robustness? For example with the filter input, what should I do
> to the filter input such that I can show that RLS is less robust than LMS?
>
> Thanks in advance for your advice.
>
> Huan.

When you say RLS, do you mean the generic matrix-inversion lemma RLS, Fast
Transversal Filters (FTFs), QR-based RLS, ..?

Regards
Sigmund.