[LAD] Xiph.org - Video:Digital Show and Tell - No difference between analog and digitally processed sound.
ladev6 at jrigg.co.uk
Thu May 23 10:29:03 UTC 2013
On Thu, May 23, 2013 at 09:38:40AM +0200, Jörn Nettingsmeier wrote:
> On 05/22/2013 09:44 PM, John Rigg wrote:
> >Decimation filters
> >that only give 6 or 12dB attenuation at fs/2 (typical in many pro
> >audio ADC chips) can allow audible aliasing too.
> this comment raised my eyebrows a little bit. can you explain what
> you mean by "decimation filter"? the way i understand it, decimation
> means chopping off bits, usually by shifting the data words, and
> possibly adding dither.
Most of the DSP textbooks I own refer to down-sampling as decimation
and up-sampling as interpolation. I guess it depends on context.
> otoh, if you mean sample rate down-conversion, i understand your
> comment, but then you picked an unfortunate term.
I meant the anti-alias filter used in the down-sampling stage
of an oversampling ADC. Many ADC manufacturers use the same term
on their data sheets (not that it excuses my laziness ;-)).
> moreover, i'd expect src circuits with only -12dB at fs/s to be
> unusable in practise, because the aliasing artefacts would be
> obvious. it means the top octave from 10-20hkz would be polluted
> with junk at -24 to -12dB, unless of course there are some
> oversampling tricks going on and the effective fs is higher during
> down conversion. although i must confess i don't know anything about
> DAC and SRC design - if someone can explain this in more detail, i'm
> all ears.
Data sheets for ADC chips commonly used in pro audio applications
typically list stop band attenuation at 0.55fs, where they give
impressive attenuation figures. At 0.5fs the attenuation can be much
As an example, take the TI PCM4222. The data sheet shows a stop
band attenuation of -100dB at 0.546fs for the "classic" filter
response setting (there is also a low group delay setting with
lower attenuation). However, a look at the frequency response graph
shows the attenuation at 0.5fs is only around -10dB. Many of the
commonly used ADCs show a similar response, but not all manufacturers
provide this information.
There's an excellent paper by the late Julian Dunn describing
Section 3 goes into detail on inadequate filter rejection. Some
of the figures listed for real ADCs are pretty shocking. Section 4
discusses related problems in DAC anti-imaging filters.
More information about the Linux-audio-dev