Forums

design of equiripple FIR filters with nearly linear phase in passband (complex chebyshev approximation)

Started by Robert Rozman July 20, 2006
Hi,

I'm trying to design N=256 lowpass FIR filter with constant group delay 
constraint in passband (resulting filter has nearly constant group delay in 
passband).

Originally Burnside and Parks described that procedure in IEEE article, but 
I cannot compile their MEX code for Matlab sucessfully - designs above N=250 
cause Matlab crashes. I did that without problems on older computers, but 
cannot reproduce it anymore. There is cremez function in Matlab's Signal 
processing toolbox that should do the same, but it produces complex solution 
despite explicit "real" constraint given. I've heard that this is a bug and 
that cremez can be fixed to produce real results.

Anyone aware of any SW that could be used for such design or how to 
workaround named problems ?

Thanks in advance,

regards,

Rob.


Robert Rozman wrote:
> Hi, > > I'm trying to design N=256 lowpass FIR filter with constant group delay > constraint in passband (resulting filter has nearly constant group delay in > passband). > > Originally Burnside and Parks described that procedure in IEEE article, but > I cannot compile their MEX code for Matlab sucessfully - designs above N=250 > cause Matlab crashes. I did that without problems on older computers, but > cannot reproduce it anymore. There is cremez function in Matlab's Signal > processing toolbox that should do the same, but it produces complex solution > despite explicit "real" constraint given. I've heard that this is a bug and > that cremez can be fixed to produce real results. > > Anyone aware of any SW that could be used for such design or how to > workaround named problems ?
Use remez? I don't understand why you need a complex filter tool when you state you want to constrain it to produce a real filter. I don't get "complex chebyshev approximation." --Randy
Randy Yates wrote:
> Robert Rozman wrote: >> Hi, >> >> I'm trying to design N=256 lowpass FIR filter with constant group delay >> constraint in passband (resulting filter has nearly constant group delay in >> passband). >> >> Originally Burnside and Parks described that procedure in IEEE article, but >> I cannot compile their MEX code for Matlab sucessfully - designs above N=250 >> cause Matlab crashes. I did that without problems on older computers, but >> cannot reproduce it anymore. There is cremez function in Matlab's Signal >> processing toolbox that should do the same, but it produces complex solution >> despite explicit "real" constraint given. I've heard that this is a bug and >> that cremez can be fixed to produce real results. >> >> Anyone aware of any SW that could be used for such design or how to >> workaround named problems ? > > Use remez? I don't understand why you need a complex > filter tool when you state you want to constrain it to produce > a real filter. I don't get "complex chebyshev approximation."
Besides, any FIR filter with end-to-end symmetry has exactly constant group delay. What's the big deal? Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Jerry Avins wrote:
> > > > Use remez? I don't understand why you need a complex > > filter tool when you state you want to constrain it to produce > > a real filter. I don't get "complex chebyshev approximation." > > Besides, any FIR filter with end-to-end symmetry has exactly constant > group delay. What's the big deal? > > Jerry > --
Hello Jerry and Randy, I believe the deal is that if the constant group delay is only forced only in the bandpass portion of the filter's response, you can design a sharper and/or lower order filter than one that has constant group delay across the entire band. Another way of viewing this is why have linear phase in the stopband where nothing (well almost) gets through there anyway? How much gain one achieves by relaxing the linear phase across the whole band constraint I don't know. But in some cases it may prove to be helpful. Clay
Clay skrev:
> Jerry Avins wrote: > > > > > > Use remez? I don't understand why you need a complex > > > filter tool when you state you want to constrain it to produce > > > a real filter. I don't get "complex chebyshev approximation." > > > > Besides, any FIR filter with end-to-end symmetry has exactly constant > > group delay. What's the big deal? > > > > Jerry > > -- > > Hello Jerry and Randy, > > I believe the deal is that if the constant group delay is only forced > only in the bandpass portion of the filter's response, you can design a > sharper and/or lower order filter than one that has constant group > delay across the entire band. Another way of viewing this is why have > linear phase in the stopband where nothing (well almost) gets through > there anyway? How much gain one achieves by relaxing the linear phase > across the whole band constraint I don't know. But in some cases it may > prove to be helpful. > > Clay
I don't really see why one would like to go that way at all. If you can settle for *nearly* linear phase *and* computational complexity is a big deal (i.e. you are pussing the limits of your system), why not use a IIR + an all-pass phase compensator? A low-order IIR (low order compared to the FIR) gets you the sharp transition, and a relatively low order all-pass (again, low order compared to the FIR alternative) squeezes the phase bact towards linear. If you can get there by means of IIRs, why meddle with FIRs? Rune