[time-nuts] Best phase detector / mixer for 100MHz?

[Bruce Griffiths]

Support HpW-Works.com wrote:
> Hi Bruce,
>
>    
>>> The equivalent phase noise (measured in dBc/Hz) is essentially independent of
>>> the sample size.
>>>        
> The dBc/Hz normalization is based on this (also the sample rate). While the sample
> rate&  sample size = FFT Bin size (resolution or filter band width) is used to get
> the required correction factor of the spectrum to get the dBc / (1Hz) Y scaling
> back.
>
>    
>>> So increasing the sample size isnt particularly useful for reducing the phase
>>> noise floor.
>>>        
> By theory, yes... but we use a sound card with a lot of flicker noise on the lower
> end, also we have the 10...20Hz low freq. cutoff due the usage of a servo / single
> 5V power. Also the raising noise below 100Hz (ADC serve&  noise on the ADC power /
> input circuit) limits the performance.
>
> In my simple test increase of the sample size reduced the noise floor in better way
> than just using sample size with 1-2K and large averaging cycles.
>
> Keep in mind, Bin resolution is sample rate / sample size:
>
> Example:
>
> - 32khz / 32678 = about 1Hz
> - 32khz / 1024  = about 31 Hz
>    
Your sample sizes are far too small to be particularly useful in 
measuring phase noise down to offsets of 1-10Hz or so,
Phase noise is highly coloured at such offsets necessitating the use of 
bin sizes substantially smaller than the lowest offset frequency of 
interest.
However decreasing the bin size beyond a small fraction of the lowest 
frequency of interest is counter productive in that one forgoes the 
effect of averaging to reduce the variance of the noise signal power 
within each bin.
There is a NIST paper on the effect of equivalent filter bandwidth on 
the accuracy of coloured noise measurements.

Indeed one can break the frequency range into a set of bands, the higher 
frequency bands having larger bin sizes and greater averaging and hence 
(lower bin noise signal power variance) than the lower frequency bands.

However for the purposes of spur identification using as small a bin 
size as possible can be useful.

>
>    
>>> However with a sound card plus a mixer a somewhat lower number of samples
>>> should suffice since there is no carrier.
>>>        
>
> Ideally the input circuit&  ADC of the 3562A would be nice but with much larger RAM
> buffer and ASIO interface O:)
>
> Hanspeter
>
>    
Using a high end PC should allow real time signal processing with 
200ksps or greater, this is certainly the case for some external USB 
instruments.

Bruce