Re: Does "remez" work for anyone?

[Ozzy Lash]

There are comments in the remez.cc source code that indicate that there may be some issues, specifically with the Search() routine. For example, near the top:

/**************************************************************************
 *  There appear to be some problems with the routine Search. See comments
 *  therein [search for PAK:].  I haven't looked closely at the rest
 *  of the code---it may also have some problems.
 *************************************************************************/

and in the Search routine itself:

        // PAK: we sometimes get too many extremal frequencies

and

   // PAK: we sometimes get not enough extremal frequencies

I think the last case is the one you are hitting. It appears that the code is expecting to find between r and 2r extrema (where r is the number of filter coefficients), and if it doesn't it throws an error.

I'm not familiar enough with the Parks-McClellan algorithm to understand what is wrong in in these cases, but for your example, if you change your 32*N in the remez call to 16*N it doesn't throw an error, (although I am probably using a very old version from 2008), and gives a result that looks like it may be much more reasonable.

Bill