[time-nuts] Soundcard sampling Re: Picking a good HP 10811

[Bruce Griffiths]

Bruce Griffiths wrote:
> Lux, James P wrote:
>   
>> On 12/1/08 10:58 PM, "Bruce Griffiths" <bruce.griffiths at xtra.co.nz> wrote:
>>
>>
>>   
>>     
>>> Thus it may well be worthwhile doing this as one then (in principle)
>>> only needs 3 mixers (plus 3 simultaneously sampled sound card input
>>> channels) and no offset source, however the maximum achievable offset
>>> will probably result in beat frequencies that are a little too low for a
>>> sound card ADC.
>>>     
>>>       
>> Most sound cards roll off below around 10-20 Hz, or, at least, you're
>> starting to get into the roll off filter characteristic where the phase is
>> changing rapidly.
>>
>> Re: "simultaneously sampled"...  I would assume that sample jitter here
>> affects the measurement.  Granted, one can do a fit of many samples to a
>> sine wave, and get a sqrt(N) improvement (if it's random jitter, and not
>> systematic), but what's a typical spec for channel/channel jitter on a sound
>> card?  Or, more properly, if you're measuring an Allan deviation of, say,
>> 1E-15 over 100 seconds for a couple 10MHz sources beaten down to, say, 100
>> Hz or thereabouts.. How good does it need to be?
>>
>>   
>>     
> The differential sampling jitter between channels would probably have to
> be less than a 100ps or so a little more if its random.
> Preliminary tests with a high end sound card appear to demonstrate a
> system noise level below 1E-15 at tau = 100 sec.
>
> Bruce
>
>   
Oops I omitted a factor of 100.
The differential sampling jitter would need to be less than about 10ns.
Practically all high end sound cards should easily meet this spec.

The sampling clock jitter spec for a given SNR and oversampling ration is

 jitter < SQRT(OSR)/(2*PI*SNR)
OSR = oversampling ratio
SNR = signal to noise ratio

e.g. with a 100dB SNR a 10KHz input and an OSR of 9.6 (192kHz)
The sampling jitter needs to be < 500ps rms.

with a 100dB SNR a 1kHz input and an OSR of 96
the sampling jitter needs to be < 15.6 ns rms.

Bruce