f



Back to the LMS again

Hi all,
  I am currently trying to implement a simple LMS algorithm in C and on an
analog deviced blackfin ez-kit.  The only real problem is that I am trying
to implement it in a streaming manner and as close to real time as possible.
I have an implementation but it isn't working.
    I am wondering how many samples should I be looking at when estimating
the gradient?  The books I have been reading pretty much just use the
samples in the filter e.g. the last 32 samples for a 32nd order filter.
That is how I have been estimating the gradient, should it work?

Thanks for your help,
Zach


0
Zach
7/13/2003 2:18:06 AM
comp.dsp 20324 articles. 1 followers. allnor (8506) is leader. Post Follow

2 Replies
443 Views

Similar Articles

[PageSpeed] 26

It should work. Try playing with the step size. 
If you are doubtful about your implementation of the algorithm try it
in matlab or 'C' code on PC.

Navan

"Zach R." <zrimkunas@verizon.net> wrote in message news:<yj3Qa.11308$Kw1.8216@nwrdny02.gnilink.net>...
> Hi all,
>   I am currently trying to implement a simple LMS algorithm in C and on an
> analog deviced blackfin ez-kit.  The only real problem is that I am trying
> to implement it in a streaming manner and as close to real time as possible.
> I have an implementation but it isn't working.
>     I am wondering how many samples should I be looking at when estimating
> the gradient?  The books I have been reading pretty much just use the
> samples in the filter e.g. the last 32 samples for a 32nd order filter.
> That is how I have been estimating the gradient, should it work?
> 
> Thanks for your help,
> Zach
0
epowerx
7/14/2003 10:30:24 PM
I am a little confused by your ppost.  You seem to be saying that
since the filter is 32 samples long, you are iterating over 32
samples.  But I don't think that's what you mean.  If you are asking
how many samples it will take for the filter to converge, then that is
an exponential exp[-1/(mu  * lambda(m))], where mu is the gain, and
lambda(m) is the minimum eigenvalue.  The number of samples it will
take to converge is dependent on the sampling rate.


Maurice Givens


"Zach R." <zrimkunas@verizon.net> wrote in message news:<yj3Qa.11308$Kw1.8216@nwrdny02.gnilink.net>...
> Hi all,
>   I am currently trying to implement a simple LMS algorithm in C and on an
> analog deviced blackfin ez-kit.  The only real problem is that I am trying
> to implement it in a streaming manner and as close to real time as possible.
> I have an implementation but it isn't working.
>     I am wondering how many samples should I be looking at when estimating
> the gradient?  The books I have been reading pretty much just use the
> samples in the filter e.g. the last 32 samples for a 32nd order filter.
> That is how I have been estimating the gradient, should it work?
> 
> Thanks for your help,
> Zach
0
maurice
7/15/2003 2:21:50 PM
Reply: