Oven Xtal vs Rubidium vs Cesium, and with/without GPS

[tkamiya]

I've wondered about this for quite some time.

I often hear Good ovenized Crystal is the best in short term.  Rubidium and Cesium have low/no drift but pretty bad short term.  GPSDO is.... <insert many things>.

Most Rubidium, Cesium, and GPDSO has pretty good ovenized Crystal oscillator providing output.  Whatever technology used is to steer and maintain stable frequency for mid to longer term.  I would imagine manufacturers set time constant so that combination works to best of advantage for short term and long term.

Then....  why do I hear what I said in second sentence?  Or are my assumptions leave too much for imagination?  I wonder this because I have extra HP Oven Crystal oscillators.  I can just as easily make a setup for short term only.  Then again, why not just use Rubidium? 

Can someone help me understand this concept?

[jpb]

This is not just true of OCXOs, Rubidium, GPSDO but also for more esoteric things like passive hyrdrogen masers.

For the details visit leapsecond.com
http://www.leapsecond.com/

But to over simplify things, a good OCXO will have good phase noise and may be below 10^-13 ADEV at 1second and may remain good ( low 10^-12) out to up to a 1000 seconds but deviate off after that.
A Rubidium will probably be around 10^-11 at 1 second but get better going out to a few thousand seconds but will (eventually drift). Rubidiums also tend to have spurs and be noisy phase wise.

The GPS system will be accurate over long periods (it is kept that way using atomic clocks) so for the average of a day or more it is likely to be heading for 10^-14 but in the short term the 1PPS is going to be a few nanosecs out - probably more like 10 nanosecs so approximately 10^-8 at 1 second.

The aim of a GPSDO is to combine the best of both worlds whether it be a disciplined Rubidium or disciplined OCXO.

Even a Caesium standard clock (the definition of a second is based on it), though very good longer term is not as good short term as a hydrogen maser.

For short term probably the best is a BVA OCXO which were very expensive when made and I don't think they are even manufactured anymore (Oscilloquartz used to make them).

[tkamiya]

OK, but when you characterize Cesium or Rubidium, are you characterizing those components individually or in combination with ovenized crystal?

For example, my Cesium 4040A by Datum has Cesium tube (of course) but it also has a very good and large ovenized Crystal.  Does this drift like Cesium or does it drift like ovenized Crystal?  I'm sure there is a time constant between the two so noise generated by Cesium does not steer Crystal.

I read leap-time over and over....

[jpb]

I'm not an expert but I think all Caesium/Rubidium oscillators have a crystal that is phase-locked to a divided down microwave frequency from the physical oscillation.

But a tightly phase-locked crystal is not going to behave the same as one that is free running.

Having said that, the HP Caesium source uses a very good quality HP10811 OCXO I think so the time constant may be quite long for that (I really don't know the details but I'm sure they are covered on leapseconds site or elsewhere).

The same is true of a GPSDO but here the phase locking is via a very long time constant but even so for some time averaging periods the locked GPSDO will be worse than the free running one.

One option is to have a GPSDO based on Rubidium and then phase lock a low noise OCXO to this with a long time constant - I think quartzlock used to (or still do) produce a commercial system for this.

To answer your question regarding your Datum Caesium 4040A, I would guess that the time constant is selected somewhere around where an ADEV plot of the free-running OCXO crosses the ADEV plot of the Caesium and for short time averaging it drifts like a good OCXO, for long time averages it drifts like a Caesium standard and for inbetween periods it does a complicated mixture of the two.

[tkamiya]

I can't wait for TICC to arrive, so I can see for myself how they compare to each other. 

I was fortunate enough to get one of each type of frequency standard, so I'm trying to get a good grasp on each.  I know there is no standard that is good for all cases, so making an educated judgement may become critical.

[5065AGuru]

OK here is some info derived empirically about stability over many years of using the following standards in my lab:

EFOS2 Hydrogen Maser Beats everything at Taus > than 8 Sec. (below 1X10-14 at 20 Sec)
Low drift

HP5065A optimized beats everything but Maser at Taus > 10 Sec.(1X10-13 @32 Sec.) low drift (the "gold standard" of Rubidiums)

FTS 1200 quartz  The 3 Quartz excel at Taus <10 Sec. but do drift
HP 10811 quartz
HP 106A quartz   106A flat out to 1000 Sec (<5X10-13th)

HP5061B Std. tube Pretty crummy until you reach 8X10-13th at 10000 Sec but no
appreciable drift. If you place the 5061A/B in open loop (only using its internal 10811 Quartz) you usually can get in the area of 5X10-13th from 1 to 100 Sec. This being 10 to 100 times better than in operate. Even with a high performance tube the Cesium is the worst of the lot for the short term stabilities!

There are now some GPSDO that can give you a pretty good reference.

So what does it mean to someone testing oscillators?

Here is a typical test run I make on HP 5065A units after repair and optimization.

Using a DMTD setup run a long term (24 hour) plot against the Maser.
Then run another shorter (20 minutes is usually enough) plot against one of the Quartz. (I usually use the FTS 1200)

now you can make a combined Allan Deviation plot using the FTS 1200 data till it crosses the Masers trace  and the rest of the plot with the data from the Maser.

You now have a plot from 1 Sec to around 9000 Sec that shows the performance of the 5065A being tested.

No one oscillator can do it all and most time-nuts hoard their cherry picked Quartz oscillators. After you have tested 30 or 40 10811 units the really good ones are segregated and treated well!

Cheers,

Corby

[Mr. Scram]

That's why I love this forum. You ask about something not that common and someone with <device>guru in his name comes along and divulges a metric ton of knowledge and experience about the subject.

[tkamiya]

Thank you for the detailed discussion.  I really appreciate it. 

I didn't realize we are pushing top-notch-technology this hard with the equipment we can actually obtain.  (although I will never see Hydrogen Maser sitting in my lab)  Maybe something better will come along in my life time. 

By the way, what does DMTD stands for?

[5065AGuru]

Hi,

DMTD (dual mixer time difference) systems use the heterodyne approach to multiply the measurements resolution by a large factor.

For instance if you take the device under test (DUT) at 10Mhz  and mix it with your reference frequency offset at 10.000001 Mhz you will get a 1Hz output that once amplified and shaped can be counted on a time interval or period counter. If your counter has a 100ns resolution (10Mhz clock rate) then the frequency resolution is the original 1X10-7th multiplied by the number of digits (in this case for a resolution of 1X10-14!

This describes a single mixer channel, a dual mixer uses a separate offset oscillator feeding into two mixer channels that accept the DUT and REF frequencies and output two 1Hz signals. The phase between these signals reflects the multiplied phase difference between the DUT and REF.

Fed into the start and stop inputs of  time interval counter the counts are logged and processed with the timelab or plotter software to plot deviations or frequency.

The beauty of the DMTD is that any jitter introduced by the offset oscillator is cancelled out. (This assuming your measurements are made while the DUT and REF are in close phase coincidence.)

DMTD systems can be easily homebuilt see

http://www.stable32.com/A%20Small%20DMTD%20System.pdf

Also a digital version has shown up that might be a good starting system, see.

https://www.eevblog.com/forum/metrology/turn-dynamic-systems-fsa3011-frequency-stability-analyser/

Cheers,

Corby

[awallin]

By the way, what does DMTD stands for?

dual mixer time difference, see e.g.
http://www.stable32.com/A%20Small%20DMTD%20System.pdf

time-interval counters tend to have a single-shot noise of 10-20ps which when measuring 1PPS leads to a noise floor of ~2e-11 @ 1 s (e.g. 53230A) or so.
When you want to go below that you need a DMTD, or if you want to be more modern an SDR..

[Theboel]

Hi,
let me add some my experience with OCXO

1. Don't go cheap with Your OCXO power supply
2. Stabilize and record Your temp
3. Don't bring Your cell phone near Your measurement desk
4. grounded everything I mean the chassis.
5. Use UPS
6. DMTD is a must.

are You thing I am nut ?

[tkamiya]

Nah, you are not a nut, as long as you don't leave this room!  Funny because except for #6, I already do those things.  In fact, my whole lab is on a giant UPS.  I went a bit further.  Pulled in two dedicated circuits.  Enhanced house ground system, and put two layers of whole-house surge protectors.  Grounding, we kind of have to be careful so that we don't create unintended ground loops.

For timing related stuff, I use linear power supplies or lab power supplies only.  In near future, I may try switcher + linear regulator combination.
 
I have #7 for you.  Use ONLY quality cables and connectors. 

[Theboel]

for no 7, sure I use to buy some cables from decommissioning BTS equipment and aircraft usually the cable is PTFE

about maser don't worry I start use panasonic VP-8122 for my reference and today I have dozens GPSDO, Dozens OCXO, Dosen Rb, My PHM and AHM are very close to use 

[testpoint1]

you can tear down the OCXO, also is crystal with control circuit, what about CS and Rb clock? the combiner is a big OCXO, but the control by FLL atomic clock.

[tkamiya]

A big problem with re-purposing control circuit designed for certain technology to another is time constant.  Rubidium and Crystal do not drift at the same rate, long and short term.  You can easily end up with less stable oscillator than without the steering control.

My Rb oscillator, PRS10 actually has an external PPS input.  It also has an internal SC cut crystal oscillator.  So theoretically, I can hook up my cesium to PRS10 and do exactly what you suggest.  I'm afraid though, I might end up with overly complex, maintenance heavy setup with too much uncertainty.  Too many things moving independently.l