[time-nuts] Method for comparing oscillators

[WarrenS]

John

If you want quick answers to Frequency differences, 
You should think and measure Phase change differences Not Freq differences.
Example: 1e-12 freq offset causes a 1ns shift every 1000 seconds,
NO matter what the Frequencies are that you are comparing.

Any decent standard analog scope can see Sub 1 ns differences 
on its fastest sweep with its time base multiplier on when viewing high speed signals.

By using a dual trace scope and syncing the scope to one of the 10 MHz signals 
and watching the two signal's zero crossing drift apart on the Scope it takes no more 
than 1000 seconds and more like a minute or so to be able see 1e-12 freq differences.

In a hour you can see 1e-13 or better easy, assuming of course that the signal's short term jitter 
is not more than a few tens of ns and the scope has warmed up long enough to stabilize.
The same method works even if the two freq are not the same, such as  2 to1 or 4 to5.

The same method (i.e Phase change) can be used with most any counter also, 
and is best when comparing low freq like 1 to 1 KHz. which are hard to see on a analog scope.
The cheap counters have less resolution than 1 ns so you need to do time interval averaging 
if the counter has that and/or just let it run longer between your readings.
Example: 
Start counter Time interval count on one signal, stop it on the other signal, 
record what the difference is (and the uncertainty is), come back 
in a minute, an hour or day or week and see what the time difference now is, 
Calculate the time difference between the two reading in ns and 
the time laps between the two reading in seconds or hours. 

Freq difference between the signals is equal to [1e-9 * (ns_change / Sec) ] 
so that  10 ns_change  / sec = 1e-8,  1ns_change  / 10 sec = 1e-10
Using hours instead of seconds this works out to  
[1e-10 * ((ns_change /Hrs)  / 360ns) ]
so that 3.6 ns / hr = 1e-12

BTW,  Phase change per unit time is the same way frequency error is calculated using WWVB, the low freq 60Khz signal.

warren

***************
----- Original Message ----- 
From: "John Green" <wpxs472 at gmail.com>
To: <time-nuts at febo.com>
Sent: Monday, August 03, 2009 7:58 AM
Subject: Re: [time-nuts] Method for comparing oscillators


>I have studied the dual mixer approach and the consensus is that it is the
> most accurate method. However, it seems pretty difficult to obtain that
> accuracy. I do have some DBMs with IF response down to DC. I don't have  a
> 10811 but do have a pretty good oscillator to use for the offset. The
> problem comes in with the time interval counter. The only thing we have is
> an old 5328A. I believe, at this time, DMTD is just not possible for me to
> do. My oscilloscope method seems to work pretty well. I can't produce graphs
> showing frequency stability but that isn't a big deal for me. I just want to
> be able to compare a Rb source to a GPSDO and look at several OCXOs either
> stand alone or in equipment we have here. If I figure correctly, an error of
> 1e-12 is 1 Hz every 27.7 hours if comparing 2 10 MHz sources. I don't have
> to wait for a full cycle to occur, I can see pretty small phase differences.
> Let's say I can see a 10 degree change. That would cut the observation time
> down to just over 3 quarters of an hour. Not bad. Most OCXOs will move a lot
> more than that so shorter times would work for them. I know from past
> experience that this works pretty well for looking at warm up performance.
> My first experiment will be the hardest. I am going to check a couple of Rb
> sources against a Tbolt. I'll let you know how this works out. Thanks for
> all the input.
> 
>