Speex dev - Jun 2010 - Sound card problem in acoustic echo cancellation

From: Steve Underwood <steveu at coppice.org>
> It seems some cards use a PLL for their ADC, so they can lock to an
> incoming SPDIF signal, but always use a local crystal clock source for
> their DAC. These cards do not have their ADC and DAC synchronised.
Do common on-board or PCI sound card lock to some incoming signal?
Yes, there is a crystal oscillator and a PLL or divider to generate
assigned clock signal. But if playing and capture are assigned the 
same sampling rate, why clock of ADC and DAC are not synchronised?
1. If it is a pure hardware sound card (no incoming SPDIF signal).
Why not send ONE clock signal to ADC and DAC?
2. if it is a AC97 or HD sound card, clock frequencies of their hardware
ADC and DAC are all fixed to 48/196KHz. Then why their are still not
synchronised?
> It seems like Skype's speakerphone works OK on these machines, so maybe
> they do something tricky to measure the slip rate and resynchronise in
> software. On the other hand, they might just fall back to a crude echo
> suppression scheme. Skype seems to do a reasonable job, though.
It is difficulty to measure the accurate difference between two sampling
rate by a software in a short time. So I don't believe Skype do some kind
of resynchronise or re-sampling work.

From: Guilherme Balena Versiani <guibv at
comunip.com.br>
> Yeah, the ADC and DAC work at different rates. I really don't know why
these
> soundcards are designed this way, but I can tell you that this is very
> common. In fact, I didn't ever find a soundcard with the same capture
and
> render rates.
If it is very common. I have tested more than 20 sound cards. Only 2 sound cards
have exactly the same capture and render rates.
> If you want to sinchronize capture and render parts, you need to implement
a
> kind of buffer control, referenced as "skew control" (check it
out). There
> is a very common control that removes or increases the delay when there is
> not enough energy on the output stream (i.e. the voice being acquired by
the
> microphone).
Sorry, I don't think so. It will cause sudden change of the echo delay which
is harmful to the adaptive filter of AEC.
> But these kind of controls seem to put AEC of Speex crazy... When the skew
> control occurs, normally the AEC stops to work fine.
Yes, of course.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.xiph.org/pipermail/speex-dev/attachments/20100610/7de9780f/attachment.htm

On 06/10/2010 10:32 AM, LiMaoquan2000 wrote:
>
> From: Steve Underwood <steveu at coppice.org <mailto:steveu at
coppice.org>>
>
> > It seems some cards use a PLL for their ADC, so they can lock to an
> > incoming SPDIF signal, but always use a local crystal clock source for
> > their DAC. These cards do not have their ADC and DAC synchronised.
>
> Do common on-board or PCI sound card lock to some incoming signal?
>
I already said they do, and what the lock to - SPDIF ports. Most modern 
machines have one, and if they don't its generally still there in the 
chipset, and there is simply no socket provided.
>
> Yes, there is a crystal oscillator and a PLL or divider to generate
> assigned clock signal. But if playing and capture are assigned the
> same sampling rate, why clock of ADC and DAC are not synchronised?
>
There seems no good reason why the DAC rate should not be synchronised 
to the ADC rate. In reality, this frequently doesn't
happen.
>
> 1. If it is a pure hardware sound card (no incoming SPDIF signal).
> Why not send ONE clock signal to ADC and DAC?
>
Because the ADC always runs from the same PLL source, presumably for 
simplicity.
>
> 2. if it is a AC97 or HD sound card, clock frequencies of their hardware
> ADC and DAC are all fixed to 48/196KHz. Then why their are still not
> synchronised?
>
Not really. They are *approx* 48kHz or *approx* 192kHz.
>
> > It seems like Skype's speakerphone works OK on these machines, so
maybe
> > they do something tricky to measure the slip rate and resynchronise in
> > software. On the other hand, they might just fall back to a crude echo
> > suppression scheme. Skype seems to do a reasonable job, though.
>
> It is difficulty to measure the accurate difference between two sampling
> rate by a software in a short time. So I don't believe Skype do some
kind
> of resynchronise or re-sampling work.
>
They have plenty of time. It is possible to do the necessary resync 
*provided* the PLL doesn't drift around too much when it is not being 
driven by a proper source. Concern about that point might make them 
avoid it completely.
>
> From: Guilherme Balena Versiani <guibv at comunip.com.br 
> <mailto:guibv at comunip.com.br>>
>
> > Yeah, the ADC and DAC work at different rates. I really don't know
> why these
> > soundcards are designed this way, but I can tell you that this is very
> > common. In fact, I didn't ever find a soundcard with the same 
> capture and
> > render rates.
>
> If it is very common. I have tested more than 20 sound cards. Only 2 
> sound cards
> have exactly the same capture and render rates.
>
> > If you want to sinchronize capture and render parts, you need to 
> implement a
> > kind of buffer control, referenced as "skew control" (check
it out).
> There
> > is a very common control that removes or increases the delay when 
> there is
> > not enough energy on the output stream (i.e. the voice being 
> acquired by the
> > microphone).
>
> Sorry, I don't think so. It will cause sudden change of the echo delay 
> which
> is harmful to the adaptive filter of AEC.
>
> > But these kind of controls seem to put AEC of Speex crazy... When 
> the skew
> > control occurs, normally the AEC stops to work fine.
>
> Yes, of course.
>
>
It is quite possible that most hardware actually has the ability to 
select clocking arrangements that will ensure the ADC and DAC operate in 
lock step. Maybe the facility just isn't exposed in a usable way. In the 
end the reasons don't matter all the much. You have to live with what 
you get.

Steve