Forums

Common practice - Aliasing in transition region?

Started by Unknown June 7, 2009
I'm designing a low pass FIR and want to minimise the number of taps -
no surprises there.
The FIR is an antialiasing filter in a DDC
The target bandwidth is 80% of the available bandwidth eg
Fs = 125MHz
Passband is 50MHz
Nyquist = 62.5MHz
When designing the filter is it reasonable to have the stop band start
at 75MHz?
My rationale for this is that any signals in the range 62.5 to 75MHz
will alias to the transition region
of the filter where any signal is already being distorted by the
filter's roll off.

What is the normal approach to this?
Thanks!
That's all in your textbooks, my friend.

dc.to.uv@googlemail.com wrote:
> I'm designing a low pass FIR and want to minimise the number of taps - > no surprises there. > The FIR is an antialiasing filter in a DDC > The target bandwidth is 80% of the available bandwidth eg > Fs = 125MHz > Passband is 50MHz > Nyquist = 62.5MHz > When designing the filter is it reasonable to have the stop band start > at 75MHz? > My rationale for this is that any signals in the range 62.5 to 75MHz > will alias to the transition region > of the filter where any signal is already being distorted by the > filter's roll off. > > What is the normal approach to this? > Thanks!
On 6/7/2009 4:13 AM, dc.to.uv@googlemail.com wrote:
> I'm designing a low pass FIR and want to minimise the number of taps - > no surprises there. > The FIR is an antialiasing filter in a DDC > The target bandwidth is 80% of the available bandwidth eg > Fs = 125MHz > Passband is 50MHz > Nyquist = 62.5MHz > When designing the filter is it reasonable to have the stop band start > at 75MHz? > My rationale for this is that any signals in the range 62.5 to 75MHz > will alias to the transition region > of the filter where any signal is already being distorted by the > filter's roll off. > > What is the normal approach to this? > Thanks!
What you're thinking works for a decimating filter, i.e., if the initial sample rate supported frequencies up to 75MHz (i.e., 150MHz or greater) and then you decimated down to 125MHz. As it is the only support region with your current sample rate is 0-62.5MHz, so anything at 75MHz is already aliased to 50MHz, and since that's the edge of your passband it won't get any appreciable attenuation.
On Jun 7, 11:03&#2013266080;am, Eric Jacobsen <eric.jacob...@ieee.org> wrote:
> On 6/7/2009 4:13 AM, dc.to...@googlemail.com wrote: > > > > > > > I'm designing a low pass FIR and want to minimise the number of taps - > > no surprises there. > > The FIR is an antialiasing filter in a DDC > > The target bandwidth is 80% of the available bandwidth eg > > Fs = 125MHz > > Passband is 50MHz > > Nyquist = 62.5MHz > > When designing the filter is it reasonable to have the stop band start > > at 75MHz? > > My rationale for this is that any signals in the range 62.5 to 75MHz > > will alias to the transition region > > of the filter where any signal is already being distorted by the > > filter's roll off. > > > What is the normal approach to this? > > Thanks! > > What you're thinking works for a decimating filter, i.e., if the initial > sample rate supported frequencies up to 75MHz (i.e., 150MHz or greater) > and then you decimated down to 125MHz. &#2013266080;As it is the only support region > with your current sample rate is 0-62.5MHz, so anything at 75MHz is > already aliased to 50MHz, and since that's the edge of your passband it > won't get any appreciable attenuation.- Hide quoted text - > > - Show quoted text -
If the goal was to have an accuractely represented decimated signal (no significant aliasing in final result) you would not do this if it was the final filter in a sequence of decimating filters. It would require a further "clean up" filter to get rid of the aliased region, AND/OR adjustment of the specs in the final decimating filter to get to not allow any significant aliasing, and to get rid of any significant aliasing from previous decimating stages (possible if designed correctly). In your example, since you probably objectionably alias right to the edge of your passband, plan on either accepting some undesired aliasing or losing part of your passband to get rid of it (assuming you want a clean signal at the end). Dirk
I'd say it depends on your application.  After decimation, if you're going
to FFT the data and then go from there, you may not care about any aliased
components.  In other words, if you weren't planning on using the bins in
the transition band, it won't matter if anything aliases there.  However,
there are plenty of counter examples where aliased components would/could
cause problems.
On Jun 7, 12:22&#2013266080;pm, "Impoliticus" <swis...@uiuc.edu> wrote:
> I'd say it depends on your application. &#2013266080;After decimation, if you're going > to FFT the data and then go from there, you may not care about any aliased > components. &#2013266080;In other words, if you weren't planning on using the bins in > the transition band, it won't matter if anything aliases there. &#2013266080;However, > there are plenty of counter examples where aliased components would/could > cause problems.
If what was in the transition band was large relative the desired signal in the passband, then it could in fact affect the results even if you were only interested in FFT points within the passband. Dirk
On 7 June, 14:23, Vladimir Vassilevsky <antispam_bo...@hotmail.com>
wrote:
> That's all in your textbooks, my friend.
Any pointers to a textbook that describes actual usage would be helpful, if slightly unusual. Your implication that I'm a student looking for homework help is a little wide of the mark.
>If what was in the transition band was large relative the desired >signal in the passband, then it could in fact affect the results even >if you were only interested in FFT points within the passband. > >Dirk >
Good point. Another reason to conclude that it all depends on the application.

dc.to.uv@googlemail.com wrote:
> On 7 June, 14:23, Vladimir Vassilevsky <antispam_bo...@hotmail.com> > wrote: > >>That's all in your textbooks, my friend. > > > Any pointers to a textbook that describes actual usage would be > helpful, if slightly unusual.
L. Rabiner "Multirate Signal Processing" VLV
On Jun 7, 5:29 pm, dc.to...@googlemail.com wrote:
> On 7 June, 14:23, Vladimir Vassilevsky <antispam_bo...@hotmail.com> > wrote: > > > That's all in your textbooks, my friend. > > Any pointers to a textbook that describes actual usage would be > helpful, if slightly unusual. > > Your implication that I'm a student looking for homework help is a > little wide of the mark.
Please ignore the troll. Your question is very valid. I have seen DDC systems do exactly that, mainly in systems where multiple channels were used and little or no noise was expected between the band limited signals. What about your situation would make this reasonable? Rick