USA Cal Club Round 3

[bdunham7]

I'm just not enough of a volt nut to chase it down further.  Also problem is smaller in my lab as the temp swings are smaller.

Those were some deliberately induced temperature swings, if I set the HVAC constant and don't open outside doors, it is much better.  However your point about the difficulty of knowing how and where to measure the temperature is the real issue--equipment gets a bit warm and I don't have room to separate everything, for example.  So analyzing tempco has more variables than are first apparent and temperature gradients are often more important.  The Fluke 8846A is the one instrument that I've never been able to figure out any tempco on--I don't even know which way it goes, if any.  What the FX helps with here is giving me a good idea how well I can measure stuff (~10V anyway) just setting my HVAC for constant temperature and using normal connections and no special precautions.  Here's the FX and 8846A overnight using 4 foot banana-plug-ended plain cables (not twisted) and other stuff running on the bench.  I did turn the display off, but I don't think that matters. 

I got an average reading of 9.9999503 (1.39ppm below FX nominal) and a standard deviation of 1.645µV (0.1645ppm).  This is about 0.6ppm lower than previous tests above, but I see no point in trying to improve on that, volt-nut or not.  Just quickly retested, I get about the same result if I set the meter up on a different shelf, if I use different cables and even if I use backup (isolated) power.

[bdunham7]

The 34401A has a problem with rounding (limited resolution on data output). I would try to interchange the inputs to get below 1.0 for the ratio. (so you might gain one digit).

I'm not sure it is exactly a rounding issue, but certainly the extra digit(s) are of dubious value, just like many other meters.  Reversing for a reciprocal ratio resulted in an even more bizarre chart.

[Andreas]

Hello,
interesting.

Internally the 34401A makes obviously several conversions for a ratio measurement (100 NPLC do not last 4 seconds but 12 seconds).
I thought the rounding problem exists only at the interface.
But obviously each individual value above 10V is rounded differently from a value below 10V.

with best regards

Andreas

[Rax]

Being essentially neighbor with bdunham7 and through the generous flexibility of the "hub/host" I was fortunate to be able to overlap a tiny bit the Keysight 3458A I have on loan from a friend (calibrated by Keysight/Rosemont about two years ago) and the FX.

Not sure if anyone will think this is totally black or white magic or what, but upon a fresh ACAL ALL, NPLC = 1000, 25.7C, the 3458A reads the FX at 9.9999647V, which in my book [...counts fingers...] is just at .05ppm of 10V. BaaaAAAaam! 

Speaks volumes of the calibration work on the FX, and the incredible opportunity it provides to the US CAL Club members to have at their bench.

[alm]

upon a fresh ACAL ALL, NPLC = 1000, 25.7C, the 3458A reads the FX at 9.9999647V, which in my book [...counts fingers...] is just at .05ppm of 10V

Wouldn't it make more sense to compare the reading to the most recent measurement of the FX with low uncertainty that you trust. Probably TiN's calibration.

I would suggest transferring the value of the FX to your other standards, particularly the Fluke 5440. The 3458A's 24h stability at 10 VDC is specified at > 1 ppm, so be careful to either do sets of measurements of both standards within 10 minutes (for example collecting 16 measurements using built in statistics) and using the 10 minute transfer stability, or better connect the negative terminals of the FX and F5440 together and measure the difference using the 3458A. A lower end DMM would also do because you're probably only measuring a very low voltage, so even a 1% uncertainty wouldn't be a problem. Measure an equal number of samples at both polarities (swap leads between standards) and average the absolute value.

[Rax]

Wouldn't it make more sense to compare the reading to the most recent measurement of the FX with low uncertainty that you trust. Probably TiN's calibration.

That's exactly what I meant with my note. My bad, probably not everyone is aware that TiN's cal is 9.9999642V.

[Rax]

I would suggest transferring the value of the FX to your other standards, particularly the Fluke 5440. The 3458A's 24h stability at 10 VDC is specified at > 1 ppm, so be careful to either do sets of measurements of both standards within 10 minutes (for example collecting 16 measurements using built in statistics) and using the 10 minute transfer stability, or better connect the negative terminals of the FX and F5440 together and measure the difference using the 3458A. A lower end DMM would also do because you're probably only measuring a very low voltage, so even a 1% uncertainty wouldn't be a problem. Measure an equal number of samples at both polarities (swap leads between standards) and average the absolute value.

Thank you for this, alm. I hope I'll have enough time left with both of those to see if I can do something along those lines. We'll see.

[Rax]

upon a fresh ACAL ALL, NPLC = 1000, 25.7C, the 3458A reads the FX at 9.9999647V, which in my book [...counts fingers...] is just at .05ppm of 10V

Wouldn't it make more sense to compare the reading to the most recent measurement of the FX with low uncertainty that you trust. Probably TiN's calibration.

In case I didn't quite fully clarify my initial point, my comparison was between my own reading of 9.9999647V and the readjusted cal by TiN (this is because the FX recently got broken and it needed recertification) of 9.9999645V. That's the ".05ppm of 10V" I was talking about. My comparison wasn't with 10V (though the difference is technically that many ppms off of 10V), but instead with TiN's reading.

Sorry if this is possibly too unpacked, but my initial point was probably a bit cryptic.

[bdunham7]

The kit is on it's way to the next stop and I just wanted to show two less-famous components of the kit.  The first is the 4W resistor box that seems to be pretty accurate, probably enough to calibrate a 5.5-digit meter if it has the right values.  I measured the values with a recently calibrated Fluke 8846A at:

1.000003k
1.99997k
10.00020k
20.0002k
99.9990k
0.190002M
1.000005M
0.200008k
100.0004R
10.00373R
0.00063R

Then there is a nice little (portable, shippable) GPSDO courtesy of AndrewBCN and vindoline.  I haven't had a GPSDO before and I've relied on cross-checking with various instruments that have come and gone.  I made a OCXO board for my HP5316B counter and IIRC I nulled it to the clock in my CMU200 and checked it with an old HP 8644A that came through--or perhaps the other way around--but I've always wondered how close it was.  Here it is--and this is perfectly stable, not a 1 in 1000 photo!  For those of you not familiar with this counter, it is showing 10.0000000 MHz.  10ppb. 

[Rax]

Here it is--and this is perfectly stable, not a 1 in 1000 photo!  For those of you not familiar with this counter, it is showing 10.0000000 MHz.  10ppb.

In time-domain, it's just so much "easier" to be on one full level below the "volt world" (ppm -> ppb)... Also, these old HP counters (mine is a 5328A) are as aged as they can be.

[mendip_discovery]

ppb = nHz/Hz?

Nice here you are having fun with the cal club kit. I really need to get my 1281 fixed so I can get my 10V ref and some other bit together so there can be a UK cal club effort.

[Conrad Hoffman]

The kit has landed in Upstate NY and is back up to room temperature. One corner of the box took a hit but everything looks fine inside. I'll be getting hooked up over the next day or so. The plan is to cal and track my three Fluke 731s plus a couple of the old mini-metro references, and compare the GPSDO to my Bodnar unit. I have various L&N resistors, but only one with a calibration, so we'll see how those do as well. I have a heat pump now, compared to the previous time with the wood stove, but it's single digits tonight and the lab is on the cool side. Should be coming up to normal temps soon and be more stable than periodic wood fires!

[iMo]

..I'm not sure it is exactly a rounding issue, but certainly the extra digit(s) are of dubious value, just like many other meters.  Reversing for a reciprocal ratio resulted in an even more bizarre chart.
..

I've tried with the RATIO and I get a nice chart. Getting always 8 digits via serial. No bizarre patterns, no missing codes. Below raw data.

PS: For example when having 10V vref and aprox -67mV input I get "-6.7744869E-03" via serial and "-6.774487 m" on the display. With aprox -6mV "-6.3836033E-04" via serial and "-638.3603 u" on the display..

Perhaps it needs those 12secs (at 100NPLC) for 2x switching the ranges, 2x AZ, measuring V1 and V2 against the internal 399, calculating V1/V2, sending in SCI format via serial and printing out in ENG format on the display.

[Conrad Hoffman]

Started with a comparison between the kit GPSDO and my Leo Bodnar unit. It's been running since last night. Quick look at the phase difference before going off to work, the drift is one cycle every 2 minutes 15 seconds. That's already way better than I'll ever need, but I wonder if more running time will see it get better. BTW, that GPSDO is a great piece of kit; having a self contained display is a super feature. Both it and the Bodnar can pick up a good signal in my underground lair, which is surprising in itself.

[Conrad Hoffman]

Running for more than 24 hours, the drift is about 140 seconds per cycle @ 10 MHz. It varies, probably depending on the signal quality. I've never compared two GPSDOs before so it was interesting. I can detect a slight 1 pps feedthru, but it's of no consequence. Naturally I can't tell which unit is closer to "truth." Here's a shot of the screens and such. The scope shot is Bodnar on top supplying the trigger. The kit unit is on the bottom and drifts ever so slowly to the right. Next I'll get on to the voltage stuff, as the lair has stabilized at about 69F.

[guenthert]

Running for more than 24 hours, the drift is about 140 seconds per cycle @ 10 MHz. It varies, probably depending on the signal quality. I've never compared two GPSDOs before so it was interesting.

Do I understand this correctly, there's an error of about 1 in 1e9?
I think it's time the time nuts speak up, but to me this sounds like either GPSDO isn't locked. 

I can detect a slight 1 pps feedthru, but it's of no consequence. Naturally I can't tell which unit is closer to "truth."

Isn't the beauty of a GPS disciplined XO, that you actually can tell, whether it tells the truth?  There might be a jitter of a couple of ns on the PPS signal (on cheap chips, don't know how good high end gear is), but with patience you can reduce the error to almost arbitrary low levels.


EDIT: thinking about this a bit, those devices can't be arbitrarily patient of course.  If the control loop is, e.g. 1000s and the jitter on the PPS signal some 15ns, then I think they can determine the time with an error of 15ns/sqrt(1000) / s ~= 1/2ns/s, i.e. about 1 in 2*e9.  So the observed error isn't all that unexpected after all?

[Conrad Hoffman]

I'm not a time nut, but from what I've read, 1E-9 is probably typical unless the signal is better. I'm amazed these things can work at all in my basement. I have a piece of RG-174 that goes upstairs, but I don't think it would work for GPS signals. Ultimately I'm setting the oscillators in my frequency counters and the limit is the adjustability. I was playing with an old Optoelectronics UTC8030 counter last night and just getting the phase drift down to a few seconds was nearly impossible. My HP counter has the ovenized oscillator that's trimmed by voltage and that's much easier. I also have an old and ugly Fluke 7260A that I've never adjusted because every time I check it, it's about dead on.

[vindoline]

BTW, that GPSDO is a great piece of kit; having a self contained display is a super feature. Both it and the Bodnar can pick up a good signal in my underground lair, which is surprising in itself.

Conrad, I'm thrilled you like the GPSDO. This is completely AndrewBCN's design - hardware and software. My contribution was designing a PCB to fit my favorite little enclosure (and making a spare to add to the Cal Club kit!). From my point of view, it is not a ‘time nut’ device. I just wanted something to check and calibrate my frequency meters. I am very pleased with how it came out.

[Conrad Hoffman]

Yes, it's quite a wonderful thing. I had wanted a GPSDO for several years before buying the Bodnar. I considered building one but found all the info on what GPS module to buy, plus what the circuits should be, way too confusing, so I never did anything. I can't be alone in that. (I feel the same way about the LTZ1000 references, though I understand it much better. What board? What resistors?)

So, my final comparison numbers give a drift of one cycle every 2 minutes and 25 seconds (145 seconds.) That's 0.0069 Hz or 6.9 E-10 of a 10 MHz signal. The Bodnar is the "faster" unit, but I don't own a counter that could resolve the difference. Now, on to the voltage measurements!

[bdunham7]

So, my final comparison numbers give a drift of one cycle every 2 minutes and 25 seconds (145 seconds.) That's 0.0069 Hz or 6.9 E-10 of a 10 MHz signal. The Bodnar is the "faster" unit, but I don't own a counter that could resolve the difference. Now, on to the voltage measurements!

The display of the cal club unit and the Bodnar software should both show an actual frequency, right?  Do they both show exactly 10.000000000000000000000000000000000000000000000000000Mhz? 

The cal club unit was showing me 10.00000002 MHz even after running overnight.

[Conrad Hoffman]

You can see the display in my photos above. The Cal Club unit would read -1 count, or dead on, depending on when you look. The Bodnar unit doesn't have a direct readout, only indication that the signal and PLL are good. As I said above, I think if you want the ultimate performance it would require getting the antenna in a much better location than my underground lair! Regardless, what I'm getting is still much better than I can put to use, even setting my HP ovenized oscillator. A more curious person would mess about with Lady Heather, though AFAIK nobody has gotten it working with the Bodnar units.

[CatalinaWOW]

I am trying to avoid both volt and time nut extremes.  Your roughly 10e-9 accuracy is being off about 1 second in 30+ years.  More than good enough for most of us.

Early in my career our internal cal lab sent our counters back to us with a sticker apologizing that they were only set to 1e-8.  I always chuckled because nothing we were doing required better than 1e-4 and most of it was just fine with a percent or two error.

It is wonderful knowing my instruments are pretty accurate, but my real need for accuracy beyond four digits is very, very small.

[Conrad Hoffman]

In 30 years, when I'm 100 years old, maybe I'll miss that second- somebody can dig me up and check! Now, avoiding voltnuttery; them's fightin' words! I'm still waiting for my inexpensive portable Josephson Junction device. I was hoping to get the lab under tighter temperature control, but it's about 9 degrees F outside and I still need to burn some wood, as the heat pump can't keep up at those temperatures. Don't assume "set it and forget it" when it comes to heat pumps and electric cars in the winter!

Things will improve over the next few days and today I'll get the reference set up. I don't have an 8.5 digit meter or a scanner, so I'll be doing manual comparisons between 5 voltage standards and the LTZ reference using the Fluke 845. Debating whether to hook everything to a small rotary switch for speed, or just move the wires by hand. I'm afraid the switch (small Grayhill gold plated rotaries) will introduce thermocouple issues. I've never been able to beat clean copper telephone wire in a binding post. Also, the 845 has an output that I could data log to record sub-ppm stuff, but I don't know how stable the 845 itself is. Maybe with a manual periodic zero check?

[Conrad Hoffman]

Figured it was time for an update. After doing the GPS comparison I got everything set up for voltage comparisons. Three Fluke 731s, one "A" and two "B" plus two Mini-Metrology standards. One was from an unknown builder and bought at a hamfest and one was my original unit from the article in 1996, 28 years ago! Null meter was my Fluke 845AB. I got my "as found" numbers and expected to adjust everything and start collecting data the next day. Murphy must have slipped in during the night because the next morning all the numbers were off by several hundred microvolts.

It turns out my 845 decided to have a nervous breakdown and develop a leakage path. Not a short, just a subtle guard leak of about 1 megohm that came and went. I put it aside and proceeded using the original Mini-Metrology null meter. That agreed with the initial Fluke numbers and worked well enough for me to start adjusting. It doesn't have the filtering that the Fluke does and I really wanted that Fluke back. Condensing about three days of troubleshooting into two words- tin whiskers. The range switch in the 845 can (apparently) develop tin whiskers from the frame to the connections. All it took was careful brushing to get everything back to normal. I'm fortunate to have an HP 4329A high resistance meter, so can check guards and such at several hundred volts and up to 2E16 ohms. You'd think that would be enough to clear the whiskers just by measuring but the current must be too low.

I've got the kit another few days and will collect a run of data on all the references. It's been years since anything was checked or adjusted but I was surprised that almost everything was within a couple ppm. The hamfest mm standard was within 0.3 ppm and my HP3455A bench meter was within 1 count. I guess when things get old enough they stabilize.

[CatalinaWOW]

I really don't think it is tin whiskers.  Impedance seems way to high for that.  But a bit of dust (or even ash from the wood stove, a problem I have periodically) with a bit of moisture from the air would be consistent with your observations.  Glad you got it resolved though.