Test Equipment Anonymous

Zenith wrote: ↑Wed Jun 17, 2026 12:52 pm

tggzzz wrote: ↑Wed Jun 17, 2026 11:09 am I'm going to have to look inside the box containing the OCXO

I suspect it works as well as it ever did and I suggest you'd be better off not messing with it. If it was clearly faulty, it would be different.

My statement was ambiguous. I meant the box into which I put the 10811 plus PSU.

Going inside the 10811 is indeed a no-no!

Might still be worth crudely buffering the output using the partially regulated internal supply (the 10811's PSU is the recommended 723 PSU).

I've opened an HP OCXO to replace a thermal fuse. You can get to it without unfurling the bits that shouldn't be disturbed but the outer metal skin needs to be removed. It's in a socket but that might just be to simplify assembly as a soldering iron would definitely open it. It's a very small wire-ended part with short leads and the nearest modern replacement is slightly longer but can be made to fit.

Useful info, but I'm not breaking and entering since the 10811 is working well

I'm just being ridiculously fussy. I'm prompted by the most interesting words in science: "that's funny", not "eureka".

synx508 wrote: ↑Wed Jun 17, 2026 3:35 pm I've opened an HP OCXO to replace a thermal fuse. You can get to it without unfurling the bits that shouldn't be disturbed but the outer metal skin needs to be removed. It's in a socket but that might just be to simplify assembly as a soldering iron would definitely open it. It's a very small wire-ended part with short leads and the nearest modern replacement is slightly longer but can be made to fit.

I've opened a Racal 04A which was well below the limits of adjustment, and gave up. The 04A is the usual Racal overnised frequency frequency reference.They claimed to be good to about 2 in 10^7.

I've also opened a Marconi 10MHz frequency reference for a 2019, also below the limits of adjustment. That was different. For a start, it was in a plastic case which snapped together and did not involve desoldering. Someone had reverse engineered one and published the circuit on the web. I tweaked a preset about a sixth turn to the right and tried it. It was now adjustable in its working range and the 2019 was sort of on target; within one ppm but settling somewhere different each time it was started. Good enough for most things I wanted it for, but still a bit naff.

I bought five 12V CTI OCXO pulls from Aliexpress as potential replacements. Comparing them against a GPSDO they appear to be surprisingly good. I must get round to that project.

I have a Racal 1992 with the 04E double overnised frequency reference option. It's not far out of agreement with the GPSDO and gets there within a few minutes of starting.

I also have a Racal 9906A, (probably 1980, now around 45 years old) which had a shot frequency reference, I replaced it with a crudely codged board based on a 5V CTI pull OCXO from Aliexpress. It's currently on loan to tggzzz. From what he tells me, it appears to have given a more than adequate account of itself.

A few more "that's funny" results using my 53310a to measure my 10811. The latter is in a diecast box, the PSU is bench28V -> capacitance multiplier -> 723 -> 10811 with the odd three terminal RF capacitor and smoothing capacitor thrown in. The (electronic) frequency control is hardwired to 0V, frequency is adjusted via the (mechanical) trimcap control.

In both pictures the markers are 10mHz/1ppb apart; definitely timenuts territory. Each "peak" is of 100 measurements; that is sufficient to indicate the relative noise.

Here's the "orientation sensitivity" picture with the box resting on each of four sides. From left to right the peaks are -90, 0, 180, +90
The -90/+90 axis shows the 4e-9 variation mentioned in the spec. It also shows much more noise. The 0/180 axis shows less variation and less noise.

Conclusion: measure noise and frequency on all four axes, choose which to use as your reference orientation, then adjust. Fortuitously I had adjusted the frequency in the 0 orientation.

Here's the cable sensitivity picture. The variations are more than I would have expected, but less than due to the orientation. I have not found a correlation with length, and one cable is less noisy that others.
No, I haven't cleaned contacts, measured cable capacitance, etc. I did consider adding a crude extra external buffer before driving the cable, but on reading the 10811 manual and playing around, I decided it was more likely to cause noise problems than to cure frequency variations.

At this point life feels too short, and I don't feel like "putting lipstick on a pig"

Picturing the quartz inside the OCXO as a wobbling bar, I can imagine that if it was horizontal and right way up up, or upside down, gravity would have an effect. On its side either way, or on its end either way, not so much so. Which way it was used I'd guess would have an effect on ageing.

The effect of cables is curious but irrelevant for most purposes. Few are interested in frequency errors in the order of ppb. If you test BNC cables with a spectrum analyser some show strange behaviour. Especially the cheap ones can have dips and peaks that shouldn't be there. I'm pretty sure this must have been noticed before, such as by the designers of equipment like the 53310A and 10811. There's no point spending a lot of time researching something which is already well understood in the right circles. Have you checked the Metrology section at the other place? It might be worth asking a question there.

The other thing that occurs is that labs often have one high quality frequency reference, used by several instruments as an external reference via a distribution amplifier. If different cables between a reference and the MDA have a detectable effect, I wonder what happens when you add a distribution amplifier and more cables.

Interesting stuff. Some numbers on the noise in different orientations would be nice, although the visual comparison is good. I imagine the instrument has cursors, so it should be possible to measure the width of those peaks at half maximum amplitude (FWHM). I would expect the noise in all orientations to be similar, provided that the crystal has had sufficient time to settle its internal stresses in that orientation. That might take significant time.

The cable results are also interesting. As it happens, I've been measuring noise in phenolic insulators recently. I apply 183V (not a special voltage, but it's what I have that is quiet) across an insulator, measure the leakage current with an electrometer, and do an FFT of its noise. What I see is (roughly) 1/f noise. This is excess noise as produced by thick film resistors and is dependent on composition (and applied voltage). If the cable you're referring to has a direct voltage on it, it could be producing excess noise. You can make a given characteristic impedance using a variety of dielectrics, and material purity (shunt R) won't make much difference to Z₀.

Regarding the distribution amplifier scenario, I suppose a lot depends on the PLL at the various destinations. I'm not really clued up on RF, so how often is phase noise more important than absolute accuracy of frequency? Won't the low-pass filter implicit in a PLL ignore most phase noise?

EC8010 wrote: ↑Thu Jun 18, 2026 10:22 am Interesting stuff. Some numbers on the noise in different orientations would be nice, although the visual comparison is good. I imagine the instrument has cursors, so it should be possible to measure the width of those peaks at half maximum amplitude (FWHM). I would expect the noise in all orientations to be similar, provided that the crystal has had sufficient time to settle its internal stresses in that orientation. That might take significant time.

I've been seeing some unexplained variations in the measured stdDev, so I'm reluctant to give figures at the moment.

Those measures were very fast; 100 samples over 50s per orientation, with maybe 30s rest after changing orientation. On returning to the "0" orientation, the frequency and noise were essentially unchanged (and lower). Sufficient to show principle, no more.

What constitutes "significant time" is an interesting question, one which I don't have a good feel for. I can't rule out such changes being behind the unexplained variations, in which case "overnight" would be a starting point. Calls for more investigation.

The cable results are also interesting. As it happens, I've been measuring noise in phenolic insulators recently. I apply 183V (not a special voltage, but it's what I have that is quiet) across an insulator, measure the leakage current with an electrometer, and do an FFT of its noise. What I see is (roughly) 1/f noise. This is excess noise as produced by thick film resistors and is dependent on composition (and applied voltage). If the cable you're referring to has a direct voltage on it, it could be producing excess noise. You can make a given characteristic impedance using a variety of dielectrics, and material purity (shunt R) won't make much difference to Z₀.

The cables are bog-standard whatever I have lying around. The OCXO output is a 511ohm resistor driving a 3:1 transformer, that drives the cable and the 50R load inside the analyser. 1.5Vpp sine wave, mean value 0V.

Observing the the display as the histogram accumulates, the individual measurements do not show a trend; they are scattered across the normal distribution. Each measurement is averaged over 5s, and I accumulate 10 measurements before updating the histogram. That might limit the scope for 1/f noise.

Regarding the distribution amplifier scenario, I suppose a lot depends on the PLL at the various destinations. I'm not really clued up on RF, so how often is phase noise more important than absolute accuracy of frequency? Won't the low-pass filter implicit in a PLL ignore most phase noise?

In a comms system, phase noise translates to clock jitter, and that can badly affect received SNR[1]. Hence people prize low phase noise sources and analysers. Contrariwise, the frequency difference between tx and rx is accommodated by the PLL in the receiver. Larger frequency difference can imply a longer time for a PLL to come into lock, but (like metastability) that is non-deterministic. Not sure about frequency locked loops, though. Overall it is a complex topic, the subject of much engineering

[1] to pick an article/pictures at random, see https://www.edn.com/how-to-relate-eye-m ... ts-to-ber/ and mentally extrapolate from a two-level constellation to a 4 or 8 level constellation.

tggzzz wrote: ↑Thu Jun 18, 2026 7:22 am No, I haven't cleaned contacts, measured cable capacitance, etc. I did consider adding a crude extra external buffer before driving the cable, but on reading the 10811 manual and playing around, I decided it was more likely to cause noise problems than to cure frequency variations.

At this point life feels too short, and I don't feel like "putting lipstick on a pig"

It's just occurred that you could drop IQD technical support an email and run this past them.

They have a place in Crewkerne.

https://www.iqdfrequencyproducts.com/en ... g-overview

They are working with this stuff all the time. I'm sure they'd be familiar with the effects of cables, as well as orientation with respect to gravity, magnetic fields and Feng Shui.

They have a large app note library

https://www.iqdfrequencyproducts.com/en ... nt-library

There's an app note on crystal ageing. It seems rather a pain to investigate, because it obviously involves long waits. They use accelerated ageing at 85C to speed things up. It also seems there's a measure of inherent unpredictability.

tggzzz wrote: ↑Thu Jun 18, 2026 11:19 am
I've been seeing some unexplained variations in the measured stdDev, so I'm reluctant to give figures at the moment.

Those measures were very fast; 100 samples over 50s per orientation, with maybe 30s rest after changing orientation. On returning to the "0" orientation, the frequency and noise were essentially unchanged (and lower). Sufficient to show principle, no more.

What constitutes "significant time" is an interesting question, one which I don't have a good feel for. I can't rule out such changes being behind the unexplained variations, in which case "overnight" would be a starting point. Calls for more investigation.

Intuitively 100 samples over 50s does not seem sufficient.

Isn't there an assumption here that the 10811 in the MDA is near perfect? It may be if you are comparing it with a TCXO or single oven OCXO of similar vintage, but against another 10811?

Zenith wrote: ↑Thu Jun 18, 2026 12:10 pm

tggzzz wrote: ↑Thu Jun 18, 2026 11:19 am
I've been seeing some unexplained variations in the measured stdDev, so I'm reluctant to give figures at the moment.

Those measures were very fast; 100 samples over 50s per orientation, with maybe 30s rest after changing orientation. On returning to the "0" orientation, the frequency and noise were essentially unchanged (and lower). Sufficient to show principle, no more.

What constitutes "significant time" is an interesting question, one which I don't have a good feel for. I can't rule out such changes being behind the unexplained variations, in which case "overnight" would be a starting point. Calls for more investigation.

Intuitively 100 samples over 50s does not seem sufficient.

I agree - but in practice it is surprisingly good at indicating the gross characteristics/artefacts, and where extra attention should be directed. It is how I first noticed the gravity effect and cable effect.

Isn't there an assumption here that the 10811 in the MDA is near perfect? It may be if you are comparing it with a TCXO or single oven OCXO of similar vintage, but against another 10811?

The MDA is indeed assumed to be golden.

Given two external sources A and B, and the MDA indicates A is better than B, there are two possibilities. Either the MDA is better than B, or there is "interference/noise" common to the MDA and A which cancels A's defects. In this case I would consider mains and the low level 96.4HMz that can be seen on some scope traces in my house.

The other issue is that the HP manual does not explain how it makes its measurements in sufficient detail to quantify the measurement artefacts. Very disappointing. ISTR people being disappointed there's no HP Journal article; ought to check if there's a 5371/5372 article.

As with all data, it's important not just to look at the final result, but also how it got there. Histograms are a case in point because random additions (as described) are what you want to see, not a gently changing line of changes (usually due to temperature slowly changing). FFTs are also useful, but you need to observe the time domain data going into them to spot popcorn noise (which doesn't show up very well on an FFT).

A single Gaussian peak is the ideal histogram and should ideally be about nine channels wide at FWHM and contain 5000 counts. Yup. That's a lot.

That IQD link was useful; I still have their data book from 1994. I smiled when I saw that photograph of one of their benches with their poshest equipment on it and all neat and tidy. Still, at least they didn't have a pretty girl there holding a soldering iron by the hot end.

Transformer coupled to the cable rules out a direct voltage being there (unless the far end introduces something), ruling out excess noise.

EC8010 wrote: ↑Thu Jun 18, 2026 2:28 pm As with all data, it's important not just to look at the final result, but also how it got there. Histograms are a case in point because random additions (as described) are what you want to see, not a gently changing line of changes (usually due to temperature slowly changing).

This afternoon's case in point. At the starting point a short histogram was somewhat gaussian, mean 9.99999998441930Hz. Two and a half hours later it had gaussian sides but was more flat-topped. Another short histogram's mean was 9.9999999860998Hz, a shift of 2mHz.

Datasheet gives MDA resolution at 10MHz as 100µHz.

The 10811 OCXO was in its "gravity affected" position, so the shift might or might have been due to the crystal "relaxing" in some way. The room temperature change was ~4C.

I find this slightly disturbing. I'd assumed that if you feed a signal at say 10MHz from a source at 50R into a BNC cable and at the other end is an instrument also terminated at 50R, you get 10MHz out at the instrument. There may be some loss of amplitude, but there's no obvious reason why there should be a frequency change, even in parts in 10^12. Doesn't it cast doubts on the first set of results you presented, because you didn't account for the cable effect?

It's only slightly disturbing, because for just about anything I do a couple of parts in 10^8 is enough. I've lazily assumed my GPSDO is perfect for anything I want to do.

Do the MDA or 10811 manuals give any advice on cables, even the usual patter about using only genuine HP BNC cables (or if you are skint, Pomona etc, who probably made the HP cables)?

Now we are talking about the limits of measurement - well for the money we are likely spend on such things - and otherwise marginal effects start to impose themselves. For instance, the radiometer effect was discovered when Crookes was using a balance in a vacuum (of sorts), and noticed that light impinging on apparatus had an effect. Who normally worries about light affecting a balance? I think people seriously using 8½ digit voltmeters will routinely use practice and materials which most of us wouldn't dream of.

I wonder what people who run caesium beam clocks use for cables? It certainly won't be something they picked up at a swapmeet for a quid.

As I say, people using definitive time standards must have come across this. IQD is one possibility but there are plenty of others.

There is no simple obvious way in which the cable could affect the frequency. I can dream up several convoluted "movie plot" mechanisms, but the problem is associating cause and effect.

Im using "movie plot" in the same sense as Bruce Schneier: convoluted brittle and entertaining. Think of the old TV show "Mission Impossible". A chain of N events had to be predicted precisely and happen; if only one failed then the whole plot was lost.

Just as with voltnuts, first and second order effects are dealt with, leaving third and fourth order effects to ruin your day. Gravity is such a third order effect

Zenith wrote: ↑Thu Jun 18, 2026 8:42 pm
I wonder what people who run caesium beam clocks use for cables? It certainly won't be something they picked up at a swapmeet for a quid.

These.

https://verdantaudio.com/products/wirew ... NsEALw_wcB

MED6753 wrote: ↑Thu Jun 18, 2026 9:05 pm

Zenith wrote: ↑Thu Jun 18, 2026 8:42 pm
I wonder what people who run caesium beam clocks use for cables? It certainly won't be something they picked up at a swapmeet for a quid.

These.

https://verdantaudio.com/products/wirew ... NsEALw_wcB

Of course. They'd also need competent mains cables. Skimp on those and spoil the ship for a ha'porth of tar.

https://www.audioemotion.co.uk/ansuz-d- ... 0472-p.asp

The worrying thing is that there are lots of snake oil audio cables. Fancy mains cables, fancy loudspeaker cables, fancy interconnect cables with phono plugs (not XLRs) at each end. But guess what? The world still uses unbalanced cables from turntables to RIAA stage. The cartridge is a balanced transducer and the ideal cable from arm base to RIAA stage is twisted pair in a proper electrostatic shield. But no, coaxial cable is used and you can see it in measured turntable rumble spectra as a 50Hz line that just doesn't need to be there. It isn't even essential to make the input of the RIAA stage balanced, just that cable. /RANT

We used quite fancy 75 ohm video cable at the Beeb when video was analogue because you could easily send signals a significant distance in a studio centre. It was double braid screen and carefully made for Z₀ = 75 ohm. I would like to think the people using caesium clocks are similarly fussy. However... My experience of most physicists is that they are weak on electronics (despite all their measurements being made via electronics). They treat cables with contempt, have not heard of "minimum bend radius", don't keep connectors clean, tug on cables, and don't segregate different characteristic impedances. All my RG58 cables have PVC caps over their BNCs when not in use and metal caps on chassis BNCs. Clean insulators = low noise. All my cables have clip-on ferrites at each end. It's much easier to presume that you have an RF problem and take steps against it than to discover months of measurements are rubbish. My previous laptop interfered with various bits of kit...

EDIT: I forgot. I do have a fancy mains cable. If you discover that a piece of kit might be sensitive to mains-borne common-mode interference, you can block it with a low C_PS isolating transformer. I have a Topaz transformer that's 5fF. But there's not much point in connecting your sensitive kit to that via an aerial, so I made a twisted pair mains cable with double braid overall screen. (Yes, the braid was stolen from some PSF1/2 analogue video cable.)

... and then there's the linearity of the microphone, and loudspeaker plus room acoustics to consider. We'll ignore the the linearity of the ear-brain combination.

Plus, what is the correct sound anyway, doubly so when the mixing and sound engineer's job is considered. Start by comparing the loved original music with the "remastered" version; in one case I rapidly gave away the remastered version and got a faithful CD transcription from the original vinyl master.

tggzzz wrote: ↑Thu Jun 18, 2026 4:40 pm ... a shift of 2mHz.

Still drifting steadily upwards (about 5mHz in total), despite a vaguely constant room temperature over the past 6 hours. Will see what it is tomorrow morning

No idea what's happening to the other orientations.

A colleague, who'd been an academic electronic engineer, told me of a talk given to students and staff by a peddler of oxygen-free copper loudspeaker cables, extolling their wonders. At the end someone said, "Are you presenting it for belief, that a signal that's passed at least 100 metres of ordinary copper cable, and imperfect amplifiers and been recorded on magnetic tape and all the rest, is improved by going through a couple of metres of oxygen-free copper speaker cable"? The peddler assured him that was exactly true, and any connoisseur could tell the difference immediately. There were smirks from the senior members of staff. Students and junior members of staff burst out laughing.

As for sonically competent mains leads......

This is different in that there appears to be a definite, measurable effect with no obvious explanation.

EC8010 wrote: ↑Thu Jun 18, 2026 9:32 pm The worrying thing is that there are lots of snake oil audio cables. Fancy mains cables, fancy loudspeaker cables, fancy interconnect cables with phono plugs (not XLRs) at each end. But guess what? The world still uses unbalanced cables from turntables to RIAA stage. The cartridge is a balanced transducer and the ideal cable from arm base to RIAA stage is twisted pair in a proper electrostatic shield. But no, coaxial cable is used and you can see it in measured turntable rumble spectra as a 50Hz line that just doesn't need to be there. It isn't even essential to make the input of the RIAA stage balanced, just that cable. /RANT

I would like to reassure you that my turntable has not one, but two sets of balanced XLR outputs; one being main, the other monitor. Obviously they're amplified, buffered and EQ'd but they're balanced. (I will carefully avoid mentioning what is inside the record player itself as input connectors to the RIAA stage )

EC8010 wrote: ↑Thu Jun 18, 2026 9:32 pm We used quite fancy 75 ohm video cable at the Beeb when video was analogue because you could easily send signals a significant distance in a studio centre. It was double braid screen and carefully made for Z₀ = 75 ohm.

Welcome to SDI at 1080p50: Good long distance SMPTE coax takes you perhaps 50m. On a good day. Forget patch-bays. Everything in the machine room is directly wired to the central matrix and there are a disturbing amount of SDI-over-singlemode-fibre converters. Shelves of them. About the only thing that is unbothered by length is SMPTE camera cable using Lemos or Neutrik Opticalcon.. No wonder people are trying very hard to move into video transport over IP, which is practically unlimited in distance.

Timenuts use regular coaxial cable. The cable cannot make small shifts in an applied frequency. It will provide phase shifts, time delays and if not properly terminated have resonances. These might affect a poorly isolated source or some forms of measurement. The better GPS time references allow for compenstion for the delay in the antenna cable.
To progress the cable issue the first thing is to provide the source with some isolation. Add an attenuator, amplifier, attenuator at the source end and an attenuator at the measurement end. Try your different cables. if the effect is still apparent, add isolation to remove any hum loops.
Put a 1:1 transformer between the amplifier output and attenuator. Old ethernet cards, especially older ones with coaxial 10Base2 connectors are a good source of suitable transformers.

The only people who reasonably worry about their cables and pay large amounts of money for them are users of Vector Network Analysers. Talking of which I'm expecting delivery of two Anritsu phase stable 6GHz cables today. I hope they check out OK.

Robert.

New arrival, Panasonic VP-8191A. Works, looks to be in good physical shape and will do a signal to spec.
Thought it would do stereo modulation, but no. That was just me in delirious acquisitions mode.

Now researching stereo MPX modulation generators. NJM 2035 plus quite few extra components will do this. Nice little project that. For later in summer.

mansaxel wrote: ↑Fri Jun 19, 2026 5:51 am (I will carefully avoid mentioning what is inside the record player itself as input connectors to the RIAA stage )

Welcome to SDI at 1080p50:

I thought I might hear from you. My turntable is a Garrard 301, so it needs a very solid plinth, achieved by the plinth being... ...solid. So there's no room for electronics inside, necessitating a cable and connector to nearby RIAA stage. As broadcasters prove all the time with microphones, you can take low-level signals quite a distance provided they're properly balanced.

I left broadcast just as SDI was coming in. Interesting to hear that fibre is the present solution.

Why compensate for delay in a GPS antenna cable? Delay down a cable tends to be of the order of 2/3c, so it's a little slow, but nothing like the time required to receive a signal from a geostationary satellite. Or is that compensation for the delay being slightly different at different frequencies?

I remember going on a course once where VNAs and associated cables were being used and being given stern warnings about using the correct torque on the connectors.