Prev: Porch
Next: IFFT of custom frequency array - confusion with the time domainsampling frequency
From: Rune Allnor on 4 May 2008 14:14
On 26 Apr, 21:39, Vladimir Vassilevsky <antispam_bo...(a)hotmail.com>
wrote:
> DigitalSignal wrote:
> > Hi there,
>
> > Here is a topic that we are looking for the solution.
> > We are looking
> > for a convenient way to implement the digital filters with arbitrary
> > amplitude shape.
> > The filter must be implemented in the time domain, preferred with
> > linear phase.
>
> Sounds like a perfect application for Parks-McClellan algorithm.

Depending on the nature of the frequency response, you
might find Ken Steiglitz' papers on the design of arbitrary
response IIR filters interesting. Don't remember the details
but I know he wrote some articles in the early '70s.

Rune
From: Andor on 5 May 2008 11:34
On 28 Apr., 18:19, DigitalSignal <digitalsignal...(a)yahoo.com> wrote:
> On Apr 28, 6:48 am, Mark <makol...(a)yahoo.com> wrote:
>
>
>
>
>
> > On Apr 26, 10:14 pm, "Fred Marshall" <fmarshallx(a)remove_the_x.acm.org>
> > wrote:
>
> > > "Vladimir Vassilevsky" <antispam_bo...(a)hotmail.com> wrote in message
>
> > >news:3yLQj.1175$506.1160(a)newssvr27.news.prodigy.net...
>
> > > > DigitalSignal wrote:
> > > >> Hi there,
>
> > > >> Here is a topic that we are looking for the solution. We are looking
> > > >> for a convenient way to implement the digital filters with arbitrary
> > > >> amplitude shape.
> > > >> The filter must be implemented in the time domain, preferred with
> > > >> linear phase.
>
> > > > Sounds like a perfect application for Parks-McClellan algorithm.
>
> > > The existing Parks-McClellan program will do it if the number of passbands
> > > is large enough for you.  Otherwise, the same type of Remez algorithm
> > > program implementation but one using a continuous frequency amplitude
> > > function as the objective function / filter design will certainly do what
> > > you want.
>
> > > With the Parks-McClellan program you can maybe specify frequency response in
> > > octaves .. or whatever .. that is, you may want do it in such a log
> > > frequency sense.
>
> > > I'd be a little concerned that the impulse response would be what you want
> > > unless you limit the length of the filter.  The length of the filter and the
> > > degree to which you match the desired response go together.
>
> > > Fred
>
> > consider also the phase (group delay) response...that may or may not
> > be important to your application...
>
> > Mark- Hide quoted text -
>
> > - Show quoted text -
>
> Yes phase is important. Thanks for mentioning it. I will do some
> research about Remez filter.

Hi James

you can use FDLS to design linear-phase FIR or arbitrary phase FIR and
IIR filters. Further, you can modify FDLS to return minmax solutions,
similar to Remez, for all classes of filters (linear-phase FIR,
arbitrary FIR and IIRs).

FDLS can be downloaded as Matlab script from here:

http://apollo.ee.columbia.edu/spm/external/tipsandtricks/files/TandT_Jan2007..zip

Here I describe how to modify FDLS to return minimax filter
coefficients:

http://groups.google.ch/group/comp.dsp/browse_frm/thread/d8aaa0438e620cf2/151a77cd29dc12c9?#151a77cd29dc12c9

Regards,
Andor
From: Greg Berchin on 5 May 2008 12:43
Andor,

Can you give an assessment of the typical performance difference
between least squares FDLS and minimax FDLS (which, I guess, should be
renamed to FDMM in this context)? Or is it just too dependent upon
the particulars of any given problem?

Thanks,
Greg Berchin
From: DigitalSignal on 5 May 2008 14:04
An IIR filter invented in early '70s? Are you kidding?

James
www.go-ci.com
From: DigitalSignal on 5 May 2008 14:06

Andor, nice job!

James
www.go-ci.com
 |  Next  |  Last
Pages: 1 2
Prev: Porch
Next: IFFT of custom frequency array - confusion with the time domainsampling frequency