[time-nuts] FTS1200, OSA8600, OSA8601 phase noise and ADEV measures
magnus at rubidium.dyndns.org
Fri Aug 3 14:44:45 EDT 2012
On 08/03/2012 05:18 AM, John Miles wrote:
>> The benefit is that I can de-correlate the reference oscillator noise,
>> and measure near or even below it.
>> I have just started doing this, so this is really my first sloppy
>> measurements for you to see where I am heading.
>> I expect John to chime in and comments on all my mistakes.
> Nope, you've pretty much got it. :)
> The idea is to fool TimeLab into
> thinking that the signal at the TimePod Ch0 and Ch2 inputs is the
> 'reference,' and the signal at the hardwired Ch1 and Ch3 splitter is the
> 'input.' That is the reasoning behind swapping the channels in the
> subtraction expression.
I forgot to mention that, as I knew it was the point about that move.
> You then connect two uncorrelated oscillators of similar frequency to the
> Ch0 and Ch2 SMA input jacks. The noise contribution from each oscillator is
> removed by the same cross correlation process that gets rid of the ADC
> Technically it's not necessary for the two oscillators to be at exactly the
> same frequency, and it's also OK if they phase-wrap during the measurement.
OK. Good to know.
> In theory they could be as much as a couple kHz apart, but certain
> calculations that the TimePod driver makes to cancel internal clock drift
> may be confused if they are more than a couple of Hz apart. I also have not
> looked carefully for instrument spurs and artifacts that might show up in
> that case, so I usually tell people to tune the two references as closely
> together as possible.
OK. Good to know.
It would help if the frequency between the reference oscillators could
be monitored one way or another.
> They do need to have the same amplitude, within a dB or two, for the same
> reason (to avoid triggering some error-detection code.)
Good to know. Need to look at that then.
> What is critical is that there is no crosstalk between the two oscillators.
> If so, it will show up as a beatnote. All of the warnings in the manual
> about using double-shielded cables at the input and reference ports apply to
> the use of uncorrelated references as well.
I have good cables, but can improve separation, as several of the OCXOs
hang of the same +24V source and is physically close to each other.
> In the case of your FTS measurement, there's a very clear beatnote or other
> artifact at about 3 Hz. You can see this in the frequency difference view
> by zooming in (see attached), in the ADEV plot at t=0.33s, and in the phase
> noise plot as the dominant spur, reported at 2.9 Hz.
> There is also a spur at half that frequency. The AM plot says that the 1.4
> Hz spur and 2.9 Hz spurs are about equal in magnitude, while the phase noise
> plot says that the 2.9 Hz spur is about 20 dB stronger. (That's why it
> dominates the ADEV and frequency-difference views.)
The FTS1200 is about 1.5 Hz below the reference, so maybe that reflects
around the reference? I will fiddle with that in a moment.
> So you should look for any possible differences in frequency between either
> your two reference sources *or* between the reference source(s) and the FTS
> 1200. In the presence of crosstalk between RF cables or power supplies,
> either of these scenarios could happen.
Cross-talk on the power-supply is highly suspected in this case. I'll
toss in a spare power-supply for separation.
> Given perfect isolation between your DUT and reference, you should not see
> any PN spurs or ADEV artifacts at their difference frequency. If you do,
> they should be so faint as to be barely noticeable. (I can say that fairly
> confidently because I spent a lot of time chasing beatnote problems during
> the R&D process. However, I have *not* spent much time looking at what
> happens when you use 2 reference oscillators that are not very close to the
> same frequency, so I can't guarantee what might happen then, spur-wise.)
I'll have a look at how they are tuned and see if I can improve on the
situation. Meanwhile the oscillators remains heated. :)
> Bottom line, the large low-frequency spurs in the FTS plot at 1.4 and 2.9 Hz
> will be caused by one of these conditions:
> 1) A problem with the reference source(s)
> 2) A problem with the FTS oscillator itself
> 3) A normal characteristic of the FTS oscillator (maybe its spur specs
> weren't very good to begin with?)
It's interesting to note that these spurs is unique to the FTS1200
measurement. Maybe it need some service. Schematic anyone?
> 4) An RF crosstalk or leakage problem with the cables/adapters used
> 5) A power-supply regulation issue
Would not explain why only the FTS1200 experience this, while it shares
supply with other oscillator not having this property.
> 6) Coupling between inadequately-bypassed power leads. This is a big
> problem with some OCXOs where they apparently forgot to use bypass
> capacitors inside the can. I usually solder a 0.1 uF ceramic chip cap right
> at the point of entry, if in doubt.
> 7) Some as-yet-unexplored effect related to beatnotes in dual-reference
Will keep looking at it.
> It may be possible to rule out cases (2) and (3), and definitely case (7),
> by temporarily switching back to the normal single-reference configuration.
True. The signal is so strong that it should be clearly visible anyway.
More information about the time-nuts