Digital Filters - Slow Rolloff - better impulse reponse?

[Cameron]

Hello. Can anyone enlighten me about the deal with
the slow rolloff option available in some digital filters?
Specifically I am looking at the BB/TI PCM1794. I have
seen very vague writings of improved impulse response.
But the stop band attenuation is greatly degraded too.
Does anyone have any experience with this? I am
wondering what, if anything, this might buy. And what
this potential improvement in one aspect might cost
in another aspect of the overall performance....

[Guido Tent]

Cameron,

"a" slow roll off (be it analogue or digital) will always improve impulse response.

It is kind of hard to understand, at least I do not grasp it, but I have seen it, both measured and simulated

regards

[jesper]

Zero oversampling, no digital filtering DAC has perfect pulserespons (before analog filtering that is), but mostly unlinear passband frequency response and fair amount of HF garbage.
An oversampling, interpolating digital brickwall filtering DAC has flat passband frequency response, high stopband attenuation (no HF garbage) but a pulse response as a rather long sinc.
With a slow rolloff digital filter you end up somewhere in-between. The slow rolloff sacrifice some stopband attenuation and passband linearity to get significant shorter sinc in the pulse response. IMO this is a good tradeoff.

/Jesper

[Guido Tent]

Originally posted by jesper
Zero oversampling, no digital filtering DAC has perfect pulserespons (before analog filtering that is), but mostly unlinear passband frequency response and fair amount of HF garbage.
An oversampling, interpolating digital brickwall filtering DAC has flat passband frequency response, high stopband attenuation (no HF garbage) but a pulse response as a rather long sinc.
With a slow rolloff digital filter you end up somewhere in-between. The slow rolloff sacrifice some stopband attenuation and passband linearity to get significant shorter sinc in the pulse response. IMO this is a good tradeoff.

/Jesper

Jesper

Thanks for contributing. The issue however is what final response will be when both filter types are combined

regards

[Pedja]

Originally posted by Cameron
Hello. Can anyone enlighten me about the deal with
the slow rolloff option available in some digital filters?
Specifically I am looking at the BB/TI PCM1794. I have
seen very vague writings of improved impulse response.
But the stop band attenuation is greatly degraded too.
Does anyone have any experience with this? I am
wondering what, if anything, this might buy. And what
this potential improvement in one aspect might cost
in another aspect of the overall performance....

Cameron,

This is not about the chip you mentioned, but almost perfectly answers your question generally.
http://www.dcsltd.co.uk/papers/effects.pdf

Pedja

[jesper]

Originally posted by Guido Tent


Jesper

Thanks for contributing. The issue however is what final response will be when both filter types are combined

regards

The analog filter after the DAC is of low order, probably 2:nd so its pulse response is fairly good. Also, the higher oversampling the higher the breakpoint of analog filter can be, thus shorten its pulse response. As a, result the contribution of the analog filter to the total pulse response is rather small.

/Jesper

[Guido Tent]

Originally posted by jesper


The analog filter after the DAC is of low order, probably 2:nd so its pulse response is fairly good. Also, the higher oversampling the higher the breakpoint of analog filter can be, thus shorten its pulse response. As a, result the contribution of the analog filter to the total pulse response is rather small.

/Jesper

Jesper

That is what I thought a long time

I suggest you simulate or measure on a combination of filters, one with overshoot, in cascade with one that rolls of slow

regards

[jesper]

Originally posted by Guido Tent



Jesper

That is what I thought a long time

I suggest you simulate or measure on a combination of filters, one with overshoot, in cascade with one that rolls of slow

regards

I'll do that when I have some spare time. Or even better, do you have a simulation to send me, Matlab would be great.

/Jesper