[time-nuts] Testing frequency using NTP

[Bruce Griffiths]

Tom Van Baak wrote:
>> Tom
>>
>> The short term phase instabilities of the sound card LO that occur 
>> during the finite interval between the zero crossing times for one 
>> channel and the zero crossing time for the other channel can be 
>> significant.
>> ...
>>     
>
> Bruce,
>
> I'm all ears for details about the phase stability of sounds cards.
> But you need to define "significant" here. Give me real numbers
> and I'll believe it.
>
> Bear in mind that we're talking about a PC-based application with
> a couple of milliseconds of expected NTP time jitter, and a modest
> goal of measuring frequency to ppm levels even if it takes hours
> to achieve that level.
>
> /tvb
>
>   
Tom

When compared to a few milliseconds of NTP jitter the phase instability 
of a sound card LO is usually insignificant, particularly if it is 
sampling at a frequency directly derived from a crystal oscillator via 
division either explicit or implicit.

For the more demanding task of comparing a couple of reasonably stable 
sources such phase instabilities appear to be relatively small 
(subnanosecond with a high end card).
The simplest way to evaluate this for a particular sound card is to 
divide down the output of a low noise OCXO, low pass filter the 
resultant signal and drive both sound card channels with the same signal.

A stable phase shift between the channels can be produced by driving the 
L and R channels in antiphase.
The simplest way to do this with a stable phase shift between the 2 
signals, at least with a quasi sinusoidal signal, is to use an audio 
transformer with a centre tapped secondary.
With a high end sound card that has balanced inputs it is even simpler 
to drive the 2 channels 180 degrees out of phase.
This allows the measurement noise to be evaluated as a function of the 
time interval between the zero crossings for each channel by varying the 
signal frequency.
To minimise the contribution from sound card input noise (or at least 
maintain this constant) the signal transition times should remain fixed 
as the signal frequency is varied.
It is also essential to use a low phase noise signal or the signal phase 
instability for an averaging time of 1/2 the signal period will dominate.

With an input signal having a finite slew rate at the zero crossing 
sound card noise will produce zero crossing time stamp jitter that 
exceeds any contribution from the LO.
For example, a full scale 1 kHz input signal will have a zero crossing 
time stamp jitter of 1.6ns if the sound card rms noise is 10ppm of full 
scale.
Only a few sound cards are this quiet.
This can be reduced by digital filtering which lowers the effective 
bandwidth.
This isn't so different from the low pass filtering of the mixer IF 
output used in classical dual mixer time difference systems or in such 
instruments as the TS5120A.

It would be useful if the phase stability data for different sound cards 
could be obtained.
I only have a relatively high end card with balanced inputs except for 
motherboard sound systems most of which appear to be rather noisy - lots 
of extraneous signals present in the ADC output spectrum.

Bruce