In Reply to: Re: Domain Dementia... posted by Todd Krieger on August 27, 2003 at 12:37:14:
"There is a rolloff, but a non-OS or ZOH DAC does not use a "sinc" filter,"No no no. The mere presence of a zero order hold with period T=1/fs
implies, immediately, by law of nature, that its frequency-domain
transfer function is a low-pass defined by sin(WT/2)/(WT/2).
(W is omega, i.e. 2 x Pi x f; T = 1/fs).Take the absolute value of this to have the amplitude-f response;
take the Hilbert transform of this and then isolate the argument
to get the phase-f response, which happens to be WT/2, i.e. there
is a phase shift present (90 degrees at fs/2), but it conveys to a constant group delay and as such is harmless.Build yourself a NONOS non-filtered DAC and measure it. Then listen to it. You hear this rolloff just as much as you hear it on the Wadia
DACs you like so much (if I remember correctly).But the pleasant news is: plot the Sinc(f) function and observe how most images disappear or at least get an attenuation decent enough not
to cause too many problems in downstream electronics.It is an effect that became immaterial with the application of oversampling (plot it again with an 8 times smaller T), but it is very real when operating a ZOH at the original T. I have to admit I had forgotten about it myself until I started working on NONOS DACs, triggered by Thorsten's Adagio.
Oh, something else ... only a little bit imagination of required to picture this rolloff near fs, actually. Imagine a 44.1k sampled baseband signal of low frequency, say 20Hz. Its first image would be at 44080Hz. But if you look at this signal after the hod stage, you only see a staircase following the 20Hz sine. No wild activity at 44.08k.
If you still doubt it, sample a DC signal and then look at the amount of 44.1k (first image) in the ZOH-output :-)
Hope this helps.
This post is made possible by the generous support of people like you and our sponsors:
Follow Ups
- sigh ... - Werner 08/27/0322:43:03 08/27/03 (0)