Forums

improving an lms filter

Started by mk December 29, 2004
hi,
I have an adaptive equalizer which implements a dfe which works
reasonably well but I am trying to reduce the excess MSE after
convergence. What changes can I make by throwing in some more hardware
at it to get slightly lower MSE ? I am afraid I can afford only 2x~3x
more multipliers only. Any hope ? I am trying to keep my convergence
rate constant.

Thanks.
mk<kal@delete.dspia.com> writes:

> hi, > I have an adaptive equalizer which implements a dfe which works > reasonably well but I am trying to reduce the excess MSE after > convergence. What changes can I make by throwing in some more hardware > at it to get slightly lower MSE ? I am afraid I can afford only 2x~3x > more multipliers only. Any hope ? I am trying to keep my convergence > rate constant.
As far as I know, the only way to improve MSE is to reduce the step-size, but that slows convergence as you imply. Could you oversample the system while reducing step-size so that the convergence rate remains constant in real-time? -- % Randy Yates % "Bird, on the wing, %% Fuquay-Varina, NC % goes floating by %%% 919-577-9882 % but there's a teardrop in his eye..." %%%% <yates@ieee.org> % 'One Summer Dream', *Face The Music*, ELO http://home.earthlink.net/~yatescr
There might be a way by implementing a "data-reusing" LMS algorithm.
This may increase convergence too with little hardware changes.  These
types of algs are discussed in an advanced topics on adaptive filtering
text (the book info escapes me at the moment).

Jim




Randy Yates wrote:
> mk<kal@delete.dspia.com> writes: > > > hi, > > I have an adaptive equalizer which implements a dfe which works > > reasonably well but I am trying to reduce the excess MSE after > > convergence. What changes can I make by throwing in some more
hardware
> > at it to get slightly lower MSE ? I am afraid I can afford only
2x~3x
> > more multipliers only. Any hope ? I am trying to keep my
convergence
> > rate constant. > > As far as I know, the only way to improve MSE is to reduce the > step-size, but that slows convergence as you imply. Could you > oversample the system while reducing step-size so that the
convergence
> rate remains constant in real-time? > -- > % Randy Yates % "Bird, on the wing, > %% Fuquay-Varina, NC % goes floating by > %%% 919-577-9882 % but there's a teardrop in his
eye..."
> %%%% <yates@ieee.org> % 'One Summer Dream', *Face The
Music*, ELO
> http://home.earthlink.net/~yatescr