[ntp:questions] Accuracy of audio tones via VOIP

unruh unruh at invalid.ca
Tue Jul 16 15:34:58 UTC 2013

On 2013-07-16, Dave Baxter <spam at goes.nowhere.com> wrote:
> In article <51e48ee8.37077834 at news.eternal-september.org>, no-
> one at notreal.invalid says...
>> It is my experience working with audio applications that the only
>> oscillator that affects reproduced audio frequency is the one in the
>> soundcard.  The software merely builds blocks of data samples and
>> delivers them to the soundcard drivers which play them out at a rate
>> determined by the soundcard clock oscillator.  The "system" clock has
>> no affect on this process.
><Serious snippage...>
> Bear in mind however...
> That Skype, along with a lot of other VoIP systems, AFIK use a "Psycho 
> Accoustical Model" codec (much like MP3) to compress the audio, so the 
> contents of the data packets are not just sampled audio data, there is 
> much less data transfered than that, and it make ones eyes water 
> thinking of just how it somehow recreates (suprisingly good) legeble 
> speech, and in any language.
> Tones, and music do propagate, but do tend to get mangled, badly at 
> times, and not only caused by "lost packets" on the wire.  The OS/CPU 
> might get bogged down doing something else it thinks is important for a 
> few milisec's at times, causing stutters etc, plus the codec software 
> may decide that something does not need to be reproduced for 
> intelegability needs at the recieving end.   Or, encoded and sent from 
> the sending end for that matter.
> For those reasons alone, I wouldnt trust any common or garden VoIP 
> system to transfer "Frequency" (or "Time") Standard signals in any way 
> shape or form.   They might to the untrained ear sound OK, but in 
> reality, all bets would be off, to say the least, unless as someone else 
> said, +- a minute was good enough <<G>>
> The PSTN PCM systems (at least when I was working for them in the UK, 
> way back when) use special AD/DA systems with fixed clocks (locked to a 
> local reference, that in turn was compared to a national reference) from 
> vague memory, so the difference between them and VoIP is like chalk and 
> cheese.  (Look's the same at a distance, but, you wouldnt want to eat 
> chalk!)
> (I have no idea at all, what they use now for digital PSTN codecs, my 
> guess is like everything else, technology has moved on somewhat.)
> If you *NEED* a local high accuracy frequency standard, then get 
> something like a Thundabolt GPSDO (maybe put it in an airconditioned 
> room too?) Then divide down from that to get whatever audio reference 
> tone you need..  (And 1PPS pulses!)
> If they are good enough for global cellular system base stations (among 
> others) that are very critical on timing, they should be good enough for 
> the rest of us.

Except what he wants is a way for Joe bloggs, your everyday piano tuner,
to calibrate his piano tuning software so that it gives 10PPM accuracy
for frequencies. Since it is a software solution, it means that his
clients cannot and will not have Thundbot GPSS0 or anything else, and
any hardware purchase they are forced to make will mean that they will
go out and buy an all-in-one piano tuning device instead. 
He has been going on with the question "How can his software reliably
self calibrate itself to 10PPM absolute frequency standard". His idea is
to use the test tones sent out by phone, and is worrying (as he should )
the result of some of his clients using voip services to get those
tones instead of standard wired phones. 

Now, I strongly suspect that his competition, who claim 10PPM accuracy
on their ability to measure the frequency (against an absolute time
standard) of a piano string, are just lying. And that furthermore, the
necessity of them making those measurements to that accuracy is not
there. But that is the challenge he has set himself. Note that he wants
his clients to connect to that tone and to have that tone delivered
reliably, for many minutes, in order to calibrate to that accuracy,
while a piano string vibrates with a decay time of a second or so, and
that he (or anyone) can then measure its to 100PPm accuracy in a second
or so. 

> Cheers.
> Dave B.  (Back under my rock.)

More information about the questions mailing list