In the quest for an accurate frequency standard there are many options depending on your budget, but one of the most affordable is an oven controlled crystal oscillator (OCXO). [RF Burns] has a video looking at one of the cheapest of these, a sub ten dollar AliExpress module.
A crystal oven is a simple enough device — essentially just a small box containing a crystal oscillator and a thermostatic heater. By keeping the crystal at a constant temperature it has the aim of removing thermal drift from its output frequency, meaning that once it is calibrated it can be used as a reasonably good frequency standard. The one in question is a 10 MHz part on a small PCB with power supply regulator and frequency trimming voltage potentiometer, and aside from seeing it mounted in an old PSU case we also are treated to an evaluation of its adjustment and calibration.
Back in the day such an oscillator would have been calibrated by generating an audible beat with a broadcast standard such as WWV, but in 2024 he uses an off-air GPS standard to calibrate a counter before measuring the oven crystal. It’s pretty good out of the box, but still a fraction of a Hertz off, thus requiring a small modification to the trimmer circuit. We’d be happy with that.
For the price, we can see that one of these makes sense as a bench standard, and we say this from the standpoint of a recovering frequency standard nut.
Is the board mislabelled? Both when he first shows the board as well as when describing the switching of the sinusoidal output, he points to the left output and for TTL he points to the right. However, the silkscreen labels show the opposite. Or am I hallucinating?
I think he just misspoke. It looks like the wire on the black wire on the square wave BNC is connected to the TTL pads.
When adjusting OCXO’s etc, I use a variant of the beat method described.
Put your reference (in my case I use a GPSDO) into one channel of your oscilloscope and trigger from it.
Then put your unknown oscillator to be adjusted into the second channel of the oscilloscope.
Tweak the adjustment on the unknown oscillator until it’s trace is as statonary as possible.
The frequency mismatch can be calculated from the oscilloscope timebase and the time the trace takes to cross graticule lines.
The oscilloscope version of tuning two notes on an instrument by adjusting a string until the warbling ceases
When adjusting OCXO’s etc, I use a variant of the beat method described.
Put your reference (in my case I use a GPSDO) into one channel of your oscilloscope and trigger from it.
Then put your unknown oscillator to be adjusted into the second channel of the oscilloscope.
Tweak the adjustment on the unknown oscillator until it’s trace is as statonary as possible.
The frequency mismatch can be calculated from the oscilloscope timebase and the time the trace takes to cross graticule lines.
Weird, I hit the reply button once but we got two posts. Yay WordPress!
You should add a thermal insulation around the OCXO, much lower power consumption and better thermal stability.
Yeah well, it is like a couple of PPB, for most applications good enough but I prefer the oscilloquartz ine
They are very sensitive to air currents inside an enclosure. A bit of polyester batting inside does the trick. Otherwise the frequency will depend upon the orientation.
Additional insulation will reduce the power drain, but won’t improve the accuracy.
I have a pair,BNC and SMA which are 0.1 ppb after 10-15 minutes. And 20x more bare modules.
Again? https://hackaday.com/2024/11/03/oscillator-needs-fine-tuning/
This is the season for re-runs!
B^)
I’ve always been confused about people obsessing over the accuracy of frequency standards. Accuracy’s pointless – GPS can give you accuracy to essentially any precision you need. Combine it with any pullable frequency standard and you’re done. Heck, with modern techniques you rarely even need it perfect.
It’s stability that matters, and measuring Allan deviation/phase noise is a royal pain in the neck, and I practically never see hacks for that. People still recommend doing goofy stuff like envelope mode on a scope to estimate the max/min cycle-to-cycle variation. It just doesn’t matter if your frequency standard has ppb accuracy if its phase noise is so bad it’s unusable as a carrier.
And yes, obviously, there’s “well do it yourself” but I don’t have access to a decent phase noise analyzer (because they’re stupid expensive), so it’d still be handwavy “looks good” without anything to calibrate it against. I’ve seen like, one homebrew-ish phase noise analysis kit and it doesn’t have design details (so not buildable) and isn’t easily buyable.
Wondering out loud if multiple high-accuracy clocks could be combined somehow to average out the phase noise.
Phase noise is noise, you can’t beat it: you just need a good reference. Plus with multiple clocks, some of the drift is going to be common due to environmental concerns, hence the reason that OCXOs and DOCXOs exist: to isolate and control the thermal environment the oscillator’s in.
Kinda why it’s frustrating to see an OCXO qualified in terms of its frequency accuracy. I can buy a pullable oscillator for a buck and a half – that doesn’t mean it’s a decent oscillator.
They also make these chip-scale atomic clocks that my lab bought shortly after they came out because hey, rubidium clocks have great Allan deviation, right? And they were pretty spot on 10 MHz right out of the box, easily ppb accurate – but their short-term stability was terrible, to the point where we ended up using them to discipline a relatively cheap (~$10) oscillator to generate a sampling reference (they’ve improved a ton, apparently, although I wouldn’t be surprised if they just added a reference internally and disciplined themselves).
In theory, isn’t phase noise expressible in terms of a Fourier transform of the waveform? So if one could record a sample of waves, do a Fourier transform, and look for higher frequency components, one could get a description of the phase noise?
Is there any way to put that theory into practice?
Yeah, but:
1) the oscillators driving sample rates in a scope aren’t anywhere near stable enough to do it
2) they don’t have anywhere near the number of bits to extract useful levels of phase noise (much like how a network analyzer/spectrum analyzer is better than a ‘scope because of the lower noise floor)
The phase noise of the oscillator used in an ADC (like in a ‘scope) actually puts a lower limit on the effective number of bits (ENOB) – so you can actually flip it around and say “given X number of real bits what’s the phase noise/jitter floor” and for a scope, that’s an extremely soft requirement, like 1-10 ps integrated jitter for typical clock frequencies.
For really bad oscillators you could probably measure it, but a 1 ps integrated phase jitter is pretty large.
The other issue is that if you’re trying to do Allan deviation (so really low-frequency phase noise measurements) you need a very long trace, and at that point the ‘scope oscillator stability starts to be the issue. It’s kinda fun taking super-long traces on newer scopes that support it and watching the scope’s oscillator drift.
The 5v regulators on these boards can be noisy, causing the sine output to have sidelobes at the switching frequency. In one case I had to ditch the regulator alltogether and run the board from 5v
WWV may be gone with the next administration. Trump tried the last time to get rid of both WWV and WWH. I’ve used the 3rd minute of the hour for it’s A440 tuning reference to check or tune a tuner-strobe.
$100 gets you Rubidium clock or GPS disciplined oscillator, is fiddling with $8 Chinese OCXO worth it if you really need precision?
Rb/GPS and OCXOs have different phase noise/Allan deviation behavior. Rb/GPS have low long term error (well, depending on the Rb device, but often) and high short term error, OCXOs are the reverse. High-quality crystals are actually the extreme reverse, that’s why they’re used in jitter filters: to provide a low-phase noise reference.
He’s literally creating a GPSDO by using the GPS to trim the OCXO to a specific value, just with a human in the trimming loop.
The HP 5060A cesium standard uses an OCXO locked to the cesium beam filter. It would seem as though the short and long term error would be quite good as a result.