DSPRelated.com
Forums
Online · Live + on demand
February 9-11, 2027
Price goes up Jan 1: register now to lock in today's rate
Register Now →

Linear phase and impulse response symmetry

Started by Unknown ●March 11, 2008
On Mar 11, 1:12 pm, Oli Charlesworth <ca...@olifilth.co.uk> wrote:
> On Mar 11, 9:05 pm, Oli Charlesworth <ca...@olifilth.co.uk> wrote: > > > h(t) = sinc(n - T) > > > where T is non-integer. This is asymmetric, but linear phase. > > Damn, that should be > > h[n] = sinc(n - T)
The transform of a Sinc is bandlimited. Therefore the samples of a Sinc fully represent the continuous frequency response (by the reconstruction theorem). That continuous frequency response (a Sinc function) is obviously symmetric. The phase delay of a fractionally delayed symmetric and bandlimited filter will be linear, but fractional in slope. However the filter will also be of infinite length. (why?) Truncation will introduce some small phase non-linearities. IMHO. YMMV. -- rhn A.T nicholson d.0.t C-o-M
Oli Charlesworth wrote:

> > Not sure what the flaw in your argument is yet (I'm working on it), > but here's a counterexample: > > h(t) = sinc(n - T) > > where T is non-integer. &#4294967295;This is asymmetric, but linear phase.
Yeah, we've discussed this here many times. Linear-phase FIR filters have symmetric or anti-symmetric impulse responses. Linear-phase IIR filters don't have to have symmetric or anti-symmetric impulse responses. Those non-symmetric linear-phase IIR filters would also make for a nice blog ... :-) Regards, Andor
On Mar 11, 9:55 pm, "Ron N." <rhnlo...@yahoo.com> wrote:
> On Mar 11, 1:12 pm, Oli Charlesworth <ca...@olifilth.co.uk> wrote: > > > On Mar 11, 9:05 pm, Oli Charlesworth <ca...@olifilth.co.uk> wrote: > > > > h(t) = sinc(n - T) > > > > where T is non-integer. This is asymmetric, but linear phase. > > > Damn, that should be > > > h[n] = sinc(n - T) > > The transform of a Sinc is bandlimited. Therefore the > samples of a Sinc fully represent the continuous frequency > response (by the reconstruction theorem). That continuous > frequency response (a Sinc function) is obviously symmetric. > The phase delay of a fractionally delayed symmetric and > bandlimited filter will be linear, but fractional in slope. > However the filter will also be of infinite length. (why?) > Truncation will introduce some small phase non-linearities.
Indeed. I'm not sure we can pull the "delay by a non-integer number of samples" trick and ever end up with an FIR response. -- Oli
> Those non-symmetric linear-phase IIR filters ...
^^^^^^ causal
Fred Marshall wrote:

   ...

> If the unit sample response is neither purely even or odd then its Fourier > Transform has a nonzero real even part *and* a nonzero odd imaginary part - > by superposition. > > If the real and imaginary parts are nonzero then the phase can't be linear. > > Why?
Because the phase is arctan(Im/Re). Do we need more than that? ... Jerry -- Engineering is the art of making what you want from things you can get. &#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;
Andor wrote:

   ...

> Linear-phase IIR filters don't have to have symmetric or > anti-symmetric impulse responses.
There are no linear-phase IIR filters because an IIr can't be symmetric. What am I missing? ... Jerry -- Engineering is the art of making what you want from things you can get. &#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;
On Mar 11, 2:23 pm, Jerry Avins <j...@ieee.org> wrote:
> Fred Marshall wrote: > > ... > > > If the unit sample response is neither purely even or odd then its Fourier > > Transform has a nonzero real even part *and* a nonzero odd imaginary part - > > by superposition. > > > If the real and imaginary parts are nonzero then the phase can't be linear. > > > Why? > > Because the phase is arctan(Im/Re). Do we need more than that?
Then why are antisymmetric filters (Im part non-zero) linear?
Ron N. wrote:
> On Mar 11, 2:23 pm, Jerry Avins <j...@ieee.org> wrote: >> Fred Marshall wrote: >> >> ... >> >>> If the unit sample response is neither purely even or odd then its Fourier >>> Transform has a nonzero real even part *and* a nonzero odd imaginary part - >>> by superposition. >>> If the real and imaginary parts are nonzero then the phase can't be linear. >>> Why? >> Because the phase is arctan(Im/Re). Do we need more than that? > > Then why are antisymmetric filters (Im part non-zero) > linear?
They have a constant 90-degree phase shift in top of the linear phase characteristic. Symmetric filters are zero phase if made none causal by centering the impulse response at t = 0. Antisymmetric filters are zero phase +/- 90 degrees. Jerry -- Engineering is the art of making what you want from things you can get. &#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;
Ron N. wrote:
 > On Mar 11, 2:23 pm, Jerry Avins <j...@ieee.org> wrote:
 >> Fred Marshall wrote:
 >>
 >>    ...
 >>
 >>> If the unit sample response is neither purely even or odd then its 
Fourier
 >>> Transform has a nonzero real even part *and* a nonzero odd 
imaginary part -
 >>> by superposition.
 >>> If the real and imaginary parts are nonzero then the phase can't be 
linear.
 >>> Why?
 >> Because the phase is arctan(Im/Re). Do we need more than that?
 >
 > Then why are antisymmetric filters (Im part non-zero)
 > linear?

They have a constant 90-degree phase shift in top of the linear phase 
characteristic. Symmetric filters are zero phase if made non-causal by 
centering the impulse response at t = 0. Antisymmetric filters are zero 
phase &#4294967295; 90 degrees.

Jerry
-- 
Engineering is the art of making what you want from things you can get.
&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;
Jerry Avins wrote:
> Andor wrote: > > &#4294967295; &#4294967295;... > > > Linear-phase IIR filters don't have to have symmetric or > > anti-symmetric impulse responses. > > There are no linear-phase IIR filters because an IIr can't be symmetric. > What am I missing?
Non-symmetric linear-phase IIR filters are trivial to construct (Oli gave the example somewhere in this thread of a fractional delay). However, _causal_ linear-phase IIR filters are not trival to construct. We've discussed these here several times as well, the last time here: http://groups.google.ch/group/comp.dsp/browse_frm/thread/5acdd13fb7999e91/1f84696cf000b407?#1f84696cf000b407 Regards, Andor