[time-nuts] Re: Phase noise with a lock-in amplifier.

David Kirkby david.kirkby at onetel.net
Mon Apr 18 03:59:52 EDT 2005

John Miles wrote:
>>The point of others about the delay line needing to be huge to get
>>uncorrelated noise close in is possibly valid.
>>However, there is nothing stopping that mixer being two A/D's, with
>>their outputs fed to a CPU that does a simple multiplication. Although I
>>suggested an analogue RF mixer, it could be done in the digital domain.
>>(Conincedently, DSP based lock-in amplifiers implement the multipliers
> Well, there's a gedanken experiment for you.  Why do you need two A/Ds?  How
> would their inputs differ, if you didn't have a delay line?  Why not sample
> the stream only once, then buffer (delay) it digitally and mix it with the
> delayed copy?

Agreed, that would be equivalent.

> Answer: because it won't work.  You'll end up with a comb-filter response if
> you do this, which isn't of much value in phase-noise measurement.

I'll take your word for that.

> Fundamentally, you don't have any information that differentiates the noise
> you're trying to measure from the signal you're trying to ignore.  

That is not true, as the signal you are wishing to measure is AC, 
whereas the clean part one is trying to avoid is converted to DC. Any AC 
component is noise.

Surely, if you had an A/D of infinite resolution, with zero jitter, you 
could work out the phase noise of any oscillator from those samples? An 
FTT would give you the phase noise directly.

I appreciate that to get a A/D with zero jitter you would need an 
oscillator with zero jitter, so one has not avoided the need for a low 
noise oscillator.

I'm *not* suggesting my proposed solution dispenses with the need for a 
low-noise oscillator, as clearly the lock-in would need one. The lock-in 
itself must have an oscillator, and that will always put a lower limit 
to the performance of the system.

I was simply asking whether this method would work, as a point for 
discussion. I'm not stating it will, hence the question mark at the end 
of the first sentence in my original post.

*IF* it would work, it seems a cheap way of measuring phase noise.

Looking at completed auctions on eBay, an EG&G failed to attract a 
single bid at $399 starting price, an EG&G 5209 sold for $408, and the 
Stanford SR830 DSP based lock-in (which is the one of the many I have at 
work) did sell for quite a bit more at $1036.11. One would need to check 
carefully if the lock-in is a dual phase or single phase model - I don't 
think the proposed method has any hope of working with a single phase 
unit. However, I suspect they are all dual phase units. Although I don't 
know the part numbers, the pictures shown on those look like models I 
have seen lying around at work, and all are dual phase. EG&G make a 
cheap lock-in, that relies on a PC to display the results, so has just a 
few BNC jacks on the front. However, the one I know is single phase, and 
its performance is quite poor compared to the more professional units.

> That is
> why PN measurements need a clean(er) reference of some sort, whether it's a
> spectrum-analyzer LO or a separate signal source used to downmix to
> baseband.
> -- john KE5FX
> _______________________________________________
> time-nuts mailing list
> time-nuts at febo.com
> https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts

Dr. David Kirkby,

Please check out http://www.g8wrb.org/
of if you live in Essex http://www.southminster-branch-line.org.uk/

More information about the time-nuts mailing list