EEVblog #456 - CSIRO Rubidium Frequency Standard

[EEVblog]

Possibly, but I reckon not probably. Dave did a video on this before using resistors instead of crystals. However, I could be wrong; for all I know, drift could always be in one direction for crystal oscillators.

It's not, when you skew end point correct the drift it can look something like this (real data I've measured):

That graph is at a constant temp in a thermal chamber.

Shock and vibration can "reset" the drift characteristic of crystals. I've done experiments on this.

I was under the impression that TCXO is an oscillator that is either kept at a specific temperature with an oven, or contains housing (or possibly even is a certain cut of crystal) that is insensitive to frequency variations due to temperature. Do these oscillators actually come with active circuitry built in?

A TCXO is one that has extra (analog) circuitry that tries to correct for the (measured) drift of the crystal against the ambient temperature.
A DTCXO (MTCXO) does the same thing, but with a digital lookup table and a microprocessor that adjusts the frequency based on the current ambient temperature. These cost a lot because each one unit has to be individually tested at each frequency point (say 1degC steps) and a reference adjustment figure programmed in for each temp.

[chickenHeadKnob]

Out of curiosity, has anyone ever tried making a circuit that takes the average frequency of say 10 or 20 standard (pick of the bunch) crystal oscillators to get higher accuracy at low power levels? How about circuits that compensate for the temperature-frequency variation curve?

Take a bunch of oscillators - crystals, pendulums, what have you, and then put them in close physical proximity to each other so that they can influence one another. You will find they tend to drift either in phase or completely anti-phase (-180).
 as this is the lowest 'system' energy state. It is a real fundamental physical effect with many natural examples.

have a look at this youtube:http://youtu.be/kqFc4wriBvE

[max-bit]

that's my frequency standard

[codeboy2k]

that's my frequency standard

Nicely done max-bit 

Would love to see bigger pics, and a description of whats inside and what's brought out on the output connections. You seem to have a rotary selection switch?? what's that for?

I think I see a Rb reference, and maybe a GPS too?? please describe more !

Thanks

[max-bit]

[max-bit]

Display bought in VK4GHZ
Other parameters
Pattern Trimble Thunderbolt GPS (bought on eBay)
Distribution Amplifier based on BGA616 (siemens)
4 out every 16 dBm at 50 OM

[bingo600]

that's my frequency standard

Nicely done max-bit 

Would love to see bigger pics, and a description of whats inside and what's brought out on the output connections. You seem to have a rotary selection switch?? what's that for?

I think I see a Rb reference, and maybe a GPS too?? please describe more !

Thanks

I would have guessed Tbolt , the grey thing inside it looks like one my Tbolts ...
Ohh and it also might look like a Cisco switcher ... I use those for Tbolts.

What's the relay for on the distamp board ?

/Bingo

[max-bit]

The relay switches the power supply is delayed.
So this power supply Cisco
And it was at the start of the power they give some noise so used a delay
In addition, the system suppressor, the thyristors. As the voltage exceeds 10% thyristor turns on, and the fuse is burning
Simple and effective

[hairykiwi]

It's certainly better than no reference at all but to be fair 1ppm is only really useful for a straightforward crystal oscillator, or perhaps a TCXO. Good ovened oscillators are stable enough to try to set them up to 1x10^-8 to 1x10^-9, Rubidium units should be in the in the 10^-9 to 10^-11 range.

I agree that 1ppm is pretty useless.
You can get a lot better than 1ppm by a DIY  TXCO design (but it's messy of course).
I don't really consider anything a decent reference standard unless it's at least rubidium. e.g. > x10-9
Sure you can get some good oven based oscillators, but they just aren't really the same level.

Thanks for the thoughts guys. To be fair, Kasper does say the Network Tranfer Frequency (NTF) project is better than 1ppm.
And after all Dave, you're on record as saying, "But hey, yeah, point six ppm - Good enough for Australia."    OK, I admit, I'm taking your comment slightly out of context - but equally, a lot of hobbyists probably rely on a calibration lab to make the best of an "awful stock oscillator" in equipment like an Agilent 53131A. If they could achieve the same level of accuracy as you achieved with that Agilent Universal Counter using the NTF, like you say, "good enough".

1ppm is a few orders of magnitude less than ideal for a true standard, I agree, but for those of us just starting out, or on limited budget, or only building 'hobby' projects, I'd argue it's all about making do. So it would be really interesting to see how much better than 1ppm the NTF is - if anyone with a rubidium standard could spare the time please

If anyone's willing, here's the NFT project file links again:
n1.taur.dk/nft/nft.pdf
n1.taur.dk/nft/nft.exe

And if you could elaborate on how many order's of magnitude better a frequency standard really must be to satisfactorily calibrate a given counter for a desired level of accuracy, that would be useful info too.

Just to put things in perspective, ±1ppm is ±0.0864 sec/day for a standard (typically 10 - 20 ppm) 32.768 kHz quartz crystal found in nearly all electronic wrist watches. That's ±2.6 sec/month - and although most electronic watch manufacturers don't advertise products with a spec that good, many achieve it. So everything's relative to its intended use. Speaking of which, the reason I'm really interested in all this 'making-do' is for calibrating the 32.768 kHz quartz oscillator in the Open source Time Machine Project #2 (OTM-02) - a watch module for hobbyists and makers using the latest 128 x 128 Memory LCD and EFM32 ARM Cortex-M3.
Here's where the project's at as of a few days ago:


I'll make a new thread in the EEVBlog Open Source Projects forum when I get a moment - and after getting the first code running later this week, all going well.

Cheers guys - look forward to maybe seeing some results comparing Rubidium and NFT.

[EEVblog]

1ppm is a few orders of magnitude less than ideal for a true standard, I agree, but for those of us just starting out, or on limited budget, or only building 'hobby' projects, I'd argue it's all about making do. So it would be really interesting to see how much better than 1ppm the NTF is - if anyone with a rubidium standard could spare the time please

A rubidium can be as low as $50 or so on ebay, so certainly well within the real of the hobbyist. How much does this NTF one cost?

[hairykiwi]

1ppm is a few orders of magnitude less than ideal for a true standard, I agree, but for those of us just starting out, or on limited budget, or only building 'hobby' projects, I'd argue it's all about making do. So it would be really interesting to see how much better than 1ppm the NTF is - if anyone with a rubidium standard could spare the time please

A rubidium can be as low as $50 or so on ebay, so certainly well within the real of the hobbyist. How much does this NTF one cost?

Thanks Dave, I'll start keeping an eye out for one.

Any recommendations for brands/models to steer clear of? Or are they all likely to have reasonable life left in them for hobbyist use? The PRS10 I linked to earlier was on for £260.

--
The NFT doesn't involve any physical equipment purchase beyond buffering / level-shifting to a PC serial port.

Here's an excerpt from the last page of Kasper's pdf instructions:

"Generating the timing pulse
To be understood by a 115200N81 uart, the pulse must have a high (>+3V) time of 8-78?s
when arriving at the PC RXD pin. If you have a max232 or similar between what you are
measuring and the PC, remember that it inverts the signal.
If you generate the pulse using an output pin, it must be 8-78 us long, and less than 11kHz
repetition rate. A square wave of 6.4-11 kHz meets this, so 32768/4=8192 Hz would be a
good choice.
If you generate the pulse using a uart in a microcontroller, send a 0xFC character. The
baud rate does not need to match as well as it does for data; A baud rate between 230400
and 28800 will generate a usable character on the PC end."

I've not yet had a chance to play with it - it's just been lurking on my things to investigate list and this frequency standard thread seemed like a good place to share it.

BTW, In the example Kasper uses to illustrate the instructions, he achieves ±0.12ppm

[kcozens]

A rubidium can be as low as $50 or so on ebay, so certainly well within the real of the hobbyist. How much does this NTF one cost?

I'd be interested in one for $50. I have yet to see one for that price. The cheapest ones I've seen on eBay are the FE-5680 units for around $80+ plus shipping which pushes the combined price to around $100 and up.

[rf-loop]

A rubidium can be as low as $50 or so on ebay, so certainly well within the real of the hobbyist. How much does this NTF one cost?

I'd be interested in one for $50. I have yet to see one for that price. The cheapest ones I've seen on eBay are the FE-5680 units for around $80+ plus shipping which pushes the combined price to around $100 and up.

Time ago there come suddenly lot of these to markets, more than enough. (just as before it happend with LPRO-101, and also with Trimble Tb GPSDO) Then they try fast change these lots of junk to money and there was many sellers...  and they start competite with each others.. and market is saturated..  until there come buyers who start buy these after they find there is these available... now price is low...  then sellers see that lot of peoples want buy... and here also Dave help sellers to rise prices in this FEI case. (junk do not mean they are bad or not useful but with sellers eyes and how they get these....)


Sometimes in business is (in buyers side) wise to think how to share information to potential other buyers and how looud and much need open mouth.


Perhaps  some times ago it was gift from Dave to sellers they find these can sell more high price.

As long as they can sell with higher price why they sell low price?

[daddario]

Wow, the use of VDA's for splitting the reference signal really came as a suprise to me. But then again, I've never had any encounter in practice with these systems. But what really makes me wonder is... video uses a 75? system, but as far as I've understood the reference clock is distributed over a 50? system. Can someone perhaps enlighten me a bit or give some kind of a link explaining that?
Pardon my ignorance.

[Bored@Work]

But what really makes me wonder is... video uses a 75? system, but as far as I've understood the reference clock is distributed over a 50? system. Can someone perhaps enlighten me a bit or give some kind of a link explaining that?

It is the external circuitry around the video opamp that determines the whole amplifier's output impedance. In fact, for line driving it is typically a simple series resistor at the opamp output. Make it 75 Ohm and you can drive 75 Ohm lines. Make it 50 Ohm and you can drive 50 Ohm lines.

[ve7xen]

So without a GPS antenna and the PPS light blinking on the Rb, is it phase-locking to a GPS receiver's unlocked clock? I don't see any alarm or 'GPS invalid' lights to indicate it's in holdover mode.

I'd hope the CSIRO designed this properly, but have you checked Dave that you're not just calibrating against a TCXO disciplined Rb?

[daddario]

It is the external circuitry around the video opamp that determines the whole amplifier's output impedance. In fact, for line driving it is typically a simple series resistor at the opamp output. Make it 75 Ohm and you can drive 75 Ohm lines. Make it 50 Ohm and you can drive 50 Ohm lines.

But that would assume modifying the stock unit. And, oddly enough, they seem to be using the original 75? connectors on the distribution outputs, as opposed to the correct 50? ones on the custom rubidium card(with the 1PPS, etc outputs).

[SeanB]

Video distribution amplifiers can drive 50R systems easily, you just change the drive resistor at the output from 75R to 50R, and note that you then have a fan out of 2/3 of before. So if each could do 30 then afterwards you can only drive 20 outputs.

If it has a high impedance at the receive end so using a 75R terminator and the original wiring will work just as well, or a 25R series resistor just at the input of 50R terminations and then it still is terminated at the drive end and receive end correctly cable wise, just has a lower voltage level which is not a worry in this case which is using the zero crossings only.

[LaurenceW]

I bought one of the FE5680 rubidium oscillators of FleaBay a while back (which STILL seem to be in plentiful supply), and put it in a nice case. From the outside, it all appears to work just  fine. Once the "lock" light comes on, it is certainly locked to "something" (I have no higher reference to compare it against), and certainly the stability against my (20+ years old?) Option-fitted oven crystal osc in my HP5316 counter is impressive! Either that, or they both drift together... The two stay in step by better than two parts in 10^8. I'm impressed with the HP OCXO!

In the UK, there used to be an impressive and free off-air standard of 15.625KHz, that being the line frequency (64uS) of good old analogue PAL TV transmissions. This was maintained  for fun by the BBC to Better-than-I-ever-needed-it, and I used it to set up my HP OCXO a couple of years back. Sadly, this signal has dissapeared from the airwaves. So imagine my surprise to find that said HP OCXO was still within half a Gnat's Bollock of the FE5680, when I got that powered up!

My "need" for accuracy extends to setting TCXO Temperature compensated crystal oscillators in regular time-of-day clocks to the best I can. With ~30,000,000 seconds in a year, if I can set a clock to keep better than 5 seconds drift per year, then I am happy (the TCXOs themselves in my designs are no better than that). So there it is right there - anything better than one part in 10^7 is all I need, personally. No - that won't win any awards - except the "entirely and sensibly  fit for purpose" award, which is good enough for me.

Some questions about Rubdium oscillators -

> I hear talk of Rubidium tube life. Is that power-on hours, or total elapsed hours?

> If the device is at least locking reliably (which seems to be the case), can I have confidence it must be working?

> What might be the absolute error of an apparently locked oscillator?  I have not attempted any rubidium adjustments.

[grumpydoc]

> I hear talk of Rubidium tube life. Is that power-on hours, or total elapsed hours?
Power on hours for the Rubidium bulb.

> If the device is at least locking reliably (which seems to be the case), can I have confidence it must be working?
AFAIK, yes

> What might be the absolute error of an apparently locked oscillator?  I have not attempted any rubidium adjustments.
Difficult since you don't know the provenance of one from the 'bay - the adjustment range is in the order of ± 2 x10^-9 so if someone's "tweaked" it, it at least shouldn't be more than that out.

The one I have is within 1-2x10^-11 of my GPSDO,

[bingo600]

AFAIK it's not wise to put a led directly on the FE-5680A "lock" pin.
Someone mentioned that it will kill the 1-PPS , as 1-PPS "enable" is gated via this leg , and the led will pull the gate to low for the 1-PPS to go active (enable). They recommend some kind of driver for the LED.

/Bingo

[kkp]

> If the device is at least locking reliably (which seems to be the case), can I have confidence it must be working?

I have repaired a couple of those. One failure mode in the FE5680A is the crystal oscillator drifting out of range. On the board, it's the crystal with a PTC soldered to it. On a good 5680A, during warmup, the frequency ramps between 4-8 ppm high, and many ppm low. So if the unit fails to lock when power cycled hot, you may need to tweak the trimmer to compensate for the aging of the crystal.
Beyond that, if it locks in less than 5 minutes from room temperature, and never loses lock, it's good.

I have one that is weak: It won't lock when hot if the frequency ramps more than 6ppm high during search, but once locked, it stays locked. I've put about a year's worth of runtime on it since then, and it still behaves exactly the same.

One I tested was 10^-9 out, and the thing I learned from that is that you have to have it screwed to something. Not just for cooling, but to clamp the top and bottom mu-metal shields together.

If you don't have a GPSDO, the next thing to get after an Rb is any kind of GPS receiver with pps output. Sometimes your friends will even give you  one .
Use your counter if you can, or cook up something with your favorite microcontroller to timestamp the pulses, and even a 250ns-jittering gps will give you 10^-12 over a day.

/Kasper

who is at the moment doing adev plotting on a free GPS from someone in this thread.

[LaurenceW]

Thanks for those replies.  Certainly, I load buffer all of the outputs from the reference, just in case of shorts, etc. I find myslef surrounded by GPS enabled devices, but none that specifically output a timing pulse! I've got a couple that stream bursts of Ascii data, and I recall that these bursts start at precise second intervals, so i will look at that. Otherwise, my oscillator locks fine, hot or cold, every time.

[casinada]

Sync Wirelessly coming soon
http://www.ecnmag.com/news/2013/05/nist-demonstrates-transfer-ultraprecise-time-signals-over-wireless-optical-channel

[carloscuev]

I got a HP 53131A with 010 option, that is an OCXO reference, "High Oven" as HP calls it in the data sheet. Stability is 2.5 PPB (page 3 http://cp.literature.agilent.com/litweb/pdf/5967-6039EN.pdf) but I don't know when was calibrated for the last time. Cal count is 7, the seller unfortunately removed stickers in an attempt to "clean it", apparently didn't know what he was doing. It was a bargain though. It is in good state, everything seems to work correctly, just little cosmetic damage.

What I want to know how much I can rely in the measurements. The latest chip datecode is from 5th week of 1999, that was 14 years ago, data sheet states 15 ppb drift over 1 month, so i guess worst case scenario it has drifted (14*12*15) ppb, that is 2520 ppb, or 2.52 PPM? Is this right to assume?

I'm thinking on getting a FE5680A to test them against each other, do you think this is a good idea, or the FE5680A could be even worse? Of course if I find that both of them are very close to each other I could be certain that both are on spec, but if difference is significant who could I blame for the difference, the HP "High" OCXO oven or the FE5680A?

Any ideas are welcome, I just want to be confident about my measuring device.

Thanks.