A couple of points:
Using extensive averaging on the clock usually means you need deep audio
buffers - so not a solution if you need low latency.
My understanding of software sample rate conversion in the windows system is
that it simply assumes the declared sample rates are correct and applies a
fixed sample rate conversion. Which is the correct solution for the most part
(as opposed to async. SRC with some attempt to measure the actual instantaneous
sample rate). In which case, the quality could be quite good. Of course, an
audiophile would never sample rate convert his playback :)
Best,
Marc
On Oct 18, 2017, at 11:13 PMEDT, Troy Gentry <dmarc-noreply@xxxxxxxxxxxxx>
wrote:
in there lies the truth, the opertative phrase here is "If you have the
proper hardware"... the previous poster is attempting to solve this age old
problem mostly in software... also, a SOC that natively has an Integer PLL
(along with a few other goodies) can actually (has actually) solve this
problem, and for both clock trees, and from a single external crystal... so
one really can pull a rabbit out of a hat ;)
I haven't ran into "many" ears that can do the Pepsi challenge with ASYNC vs
SYNC (properly done in hardware, as Geert lays out below) and actually tell
the difference, but I have had the privilege of meeting one such sole (at a
world renowned music studio) that could hear the difference, it was truly an
amazing encounter... so, have been fortunate enough to have had the chance
to implement ISO SYNC synchronization method "properly" in hardware.
...that said, proper test equipment and measurement methods can easily
"see" the difference...
if you ever get to that level of measurement, I'll leave you with this one
tip... do not make all of your audio test files with 1k-Hz sampling (yes
someone did actually do this), as you will simply mask the traditional 1k-Hz
noise generated by the digital processing "current pumping" inherent in USB
audio systems. tip :: change all of these test files to 1.4k-Hz to validate
that your hardware is not excessively impacting the audio quality.
yes, ASYNC shifts the problem to the USB Host, no doubt about it, luckily I
haven't had to struggle with USB audio on the host side of system... not so
surprising, I guess, is that historically USB Hosts have struggled to
properly implement ASYNC synchronization... fun fun
....nobody was assuming that we live in a perfect world, but rather we're all
just simply following the spirit of the USB spec <grin, as I type this
message to the UAC spec author>.... welcome to the world of extremely low
jitter bit perfect prosumer audio that those customers demand ;)
Oh yeah, Hi Geert and thank you very much for all that you do for us ;)
really Kindest Regards (don't shoot the messenger),
-t
From: "geert@xxxxxxxxxxxxxxxxxxx" <geert@xxxxxxxxxxxxxxxxxxx>
To: wdmaudiodev@xxxxxxxxxxxxx
Sent: Wednesday, October 18, 2017 6:44 PM
Subject: [wdmaudiodev] Re: Windows Clock Compensation
If you have the proper hardware (i.e. a fractional PLL) the adjustments you
can make to the local clock are extremely precise (typically 0.01 ppm) If you
then also implement a low-pass filter behavior on the clock adjustment
algorithm and deploy very extensive averaging, you can achieve audio clock
quality levels that are beyond what current ADCs and DACs require.
Going with the asynchronous mode actually shifts (not solves) the problem to
the Host. If the Host has sufficiently sophisticated asynchronous sampling
rate algorithms available, then that may be a good solution. But do not
assume that all quality concerns evaporate when you switch to async mode.
Kindest Regards,
<http://www.designandadvice.com/> <http://www.jwhouse.org/>Geert Knapen
President/Owner
Design & Advice, L.L.C. <http://www.designandadvice.com/>
1725 Martin Avenue, San Jose CA 95128
e-mail: geert@xxxxxxxxxxxxxxxxxxx <mailto:geert@xxxxxxxxxxxxxxxxxxx> | Tel:
+1-408-297-3731 | Cell: +1-408-507-7852 | Google Voice: +1-408-805-4320
On Oct 18, 2017, at 5:05 PM, Troy Gentry (Redacted sender "tge96" for DMARC)
<dmarc-noreply@xxxxxxxxxxxxx <mailto:dmarc-noreply@xxxxxxxxxxxxx>> wrote:
doing it this way, is as Tim says an "extremely tricky problem" - agreed
doing it this way, will likely produce noticeable audible synchronization
clocking artifacts in the actual audio stream itself (depending on the
actual clock mismatch between the USB Host and the USB device)
if this is all one has for their audio buffer management mechanism, then
switching to ASYNC synchronization method would be highly recommended.
From: Akshaykeerti Sharma <asharma@xxxxxxxxxxxxxxxxxxxxx
<mailto:asharma@xxxxxxxxxxxxxxxxxxxxx>>
To: wdmaudiodev@xxxxxxxxxxxxx <mailto:wdmaudiodev@xxxxxxxxxxxxx>
Sent: Wednesday, October 18, 2017 4:51 PM
Subject: [wdmaudiodev] Re: Windows Clock Compensation
Yes, we are monitoring the micro controller queues live to ensure that there
is no problem. If I remember correctly the usb audio 1.0 spec we are using
transmits 48samples/ms(48khz) in our implementation. Making a very small
change to our I2S reference clock(this is what drives the audio out to a
speaker essentially) controls the FIFO queue. The trim setting on the micro
is accurate enough to ensure precise enough changes to the clock to
sometimes even match the input data stream.
Another way is to sync to the usb frame start of frame which results in
getting the same information. You would sync to the start frame and up/down
the clock appropriately.
The disadvantage is that we are introducing harmonic distortion in the
signal. Albeit it is small enough to have no noticeable difference on the
not-so-great speaker.
Again, this is using the USB synchronous mode.
Since the digital stream on the microconroller side is continuous if you
start over/underflowing data it really messes up the sound. Currently our
implementation works without any discontinuity. We still have some
qualitative tests to do to ensure that the reproduction is top notch.
For a high quality audio application I would recommend ASYNC mode to control
the data stream in software.
Sent from my iPhone
On Oct 18, 2017, at 4:40 PM, Tim Roberts <timr@xxxxxxxxx
<mailto:timr@xxxxxxxxx>> wrote:
Akshaykeerti Sharma wrote:
One possible solution we implemented for usb-audio using synchronous
mode was to adjust the reference clocks on the micro-controller in
realtime by a very small amount to ensure that there are no
underflow/overflow conditions that manifest themselves as
pops/clicks(we did have pops and clicks otherwise)
How would you know? By monitoring the levels in your onboard FIFOs?
This is an extremely tricky problem, because even though audio seems to
be smoothly continuous, at the microscopic level it is rather chunky.
Audio Engine tries to transfer 10ms at a time, and applications commonly
work in buffers even larger than that.
--
Tim Roberts, timr@xxxxxxxxx <mailto:timr@xxxxxxxxx>
Providenza & Boekelheide, Inc.
******************
WDMAUDIODEV addresses:
Post message: mailto:wdmaudiodev@xxxxxxxxxxxxx ;
<mailto:wdmaudiodev@xxxxxxxxxxxxx>
Subscribe: mailto:wdmaudiodev-request@xxxxxxxxxxxxx ;
<mailto:wdmaudiodev-request@xxxxxxxxxxxxx>?subject=subscribe
Unsubscribe: mailto:wdmaudiodev-request@xxxxxxxxxxxxx ;
<mailto:wdmaudiodev-request@xxxxxxxxxxxxx>?subject=unsubscribe
Moderator: mailto:wdmaudiodev-moderators@xxxxxxxxxxxxx ;
<mailto:wdmaudiodev-moderators@xxxxxxxxxxxxx>
URL to WDMAUDIODEV page:
http://www.wdmaudiodev.com/ ;<http://www.wdmaudiodev.com/>
******************
WDMAUDIODEV addresses:
Post message: mailto:wdmaudiodev@xxxxxxxxxxxxx ;
<mailto:wdmaudiodev@xxxxxxxxxxxxx>
Subscribe: mailto:wdmaudiodev-request@xxxxxxxxxxxxx ;
<mailto:wdmaudiodev-request@xxxxxxxxxxxxx>?subject=subscribe
Unsubscribe: mailto:wdmaudiodev-request@xxxxxxxxxxxxx ;
<mailto:wdmaudiodev-request@xxxxxxxxxxxxx>?subject=unsubscribe
Moderator: mailto:wdmaudiodev-moderators@xxxxxxxxxxxxx ;
<mailto:wdmaudiodev-moderators@xxxxxxxxxxxxx>
URL to WDMAUDIODEV page:
http://www.wdmaudiodev.com/ ;<http://www.wdmaudiodev.com/>