<d1b2>
<theorbtwo> That does sound fun. I just hope it doesn't become a block chain...
bvernoux has joined #scopehal
someone--else has joined #scopehal
<azonenberg>
theorbtwo: lol. people will find ways to throw a blockchain at everything
<sorear>
I have a very poor understanding of what does and doesn't count as traceability
<sorear>
if a device is producing physical units that are _at all meaningful_ it must have been calibrated, at least at a type or component level
<azonenberg>
sorear: The most commonly used definition is that a lab with an ISO 17025 certificate measured that specific device against a standard that was itself calibrated by another 17025 accredited lab, eventually chaining up to either physical constants or a national standards lab
<sorear>
which makes me wonder if they're upselling you on a more precise / individual / system-level calibration, or if they're upselling you on a piece of paper that describes the QA process that they use always
<azonenberg>
It's mostly the extra recordkeeping
<azonenberg>
as an example, with Micro Precision, the lab that I usually use
<azonenberg>
a standard "traceable calibration" is just a piece of paper that says "yep, we checked, it's in tolerance" and listing the standards they checked it against
<azonenberg>
a "traceable calibration with data" costs more, and is the exact same process
<azonenberg>
except they also write down the measurements they took on the certificate
someone--else has quit [Quit: Connection closed]
<azonenberg>
so e.g. you know that when given 1.00000 volts, your multimeter measured 1.00001 volts rather than "some unknown value within the datasheet tolerance band of the instrument"
<azonenberg>
and then if you want them to fill out all of the compliance paperwork needed for you to use that instrument as a standard for calibrating other stuff (assuming your lab has a 17025 accreditation too) it costs even more
<sorear>
mm
<sorear>
kind of interested to see what a "to physical constants" cert looks like and how much detail they have to give on the design and operation of the test
<azonenberg>
Good question. I don't have an answer as none of my stuff has ever been cal'd that way
<azonenberg>
in my case everything has been against another, more accurate instrument
<sorear>
i take it these aren't like TLS certificates and you don't get a copy of the full chain
<azonenberg>
No. They're pieces of paper or PDFs that just list specific gear used to cal your stuff
<azonenberg>
I believe you generally have a right to go to the accrediting lab and ask for a copy of their documentation for the standard used on your equipment
<azonenberg>
so you could hypothetically *get* the full chain
<azonenberg>
but it's not normally provided by default
<azonenberg>
This is why they list serial numbers and such on the cert, though
<azonenberg>
So if there is any question, you can get the data
<azonenberg>
alternatively, if it's discovered that a standard malfunctioned since the last cal, you can go back and re-audit anything cal'd against it
<azonenberg>
since the lab keeps records of which customer devices they calibrated using which standard
<sorear>
does the cert go into any detail about how they controlled for parasitics between the standard and the DUT?
<azonenberg>
No, see the example screenshot. Generally the cert is only going to list a cal procedure by reference
<azonenberg>
the procedure will then describe details about which signals you apply to which inputs, using what fixture, etc
<azonenberg>
and what the tolerance bands for each measurement are
<azonenberg>
oh, another thing that higher grade cal certificates include is expanded error bands
<azonenberg>
i.e. rather than just reporting a single data point, they include the tolerance band of the cal standard all the way up the chain
<azonenberg>
So say you apply a 1 GHz tone to a DUT, but your signal generator has a specified accuracy of +/- 0.1% in frequency
<azonenberg>
So the DUT was actually calibrated with a tone somewhere between 0.999 GHz and 1.001 GHz
<azonenberg>
in reality it's more complicated because the signal generator itself has been calibrated by equipment with its own error bounds, etc
<azonenberg>
Ultimately what you end up doing is starting from physical constants at NIST and gradually increasing error bars each hop down the chain
<azonenberg>
There is also going to be offsets, because your equipment won't be perfectly adjusted
<azonenberg>
say your signal generator was measured during cal to have a -0.005% frequency offset from nominal
<azonenberg>
so now your DUT was calibrated with 0.99995 GHz +/- 1 MHz, etc
<azonenberg>
And none of these error bands are absolute either, as they are actually probability distributions due to noise etc
<azonenberg>
normally certs will report a 95% confidence interval or something
<sorear>
i've read some of the literature on measurement processes of the sort that NIST does but that doesn't really translate into the ISO doc requirements
<azonenberg>
i.e. the lab is 95% confident that the actual output of your generator when set to 1.000 GHz is somewhere between 0.99995 and 1.00005 GHz
<sorear>
can wind up being a web if you need to measure multiple things to calibrate a measurement, or if you're using physical constants that don't have a defined value and need to get a value of them from somewhere...
<azonenberg>
Correct
<azonenberg>
But normally you use a single standard for a particular metric
<azonenberg>
so say you use the timebase of a signal generator as a reference for frequency, but an rf power meter as a reference for power
<azonenberg>
then when calibrating a specan the X and Y axes will have different error bounds tracing to difference pieces of gear
<azonenberg>
which is why cal certs normally specify what a device was used for in the procedure
<sorear>
but both of those need to be derated for ambient temperature, so now in principle you need a traceable measurement of the lab temperature
<azonenberg>
That's why the certificate includes the temperature on it
<azonenberg>
but i think it's normally just a pass/fail check, in that the standard is known to be within some tolerance band at 25 +/- 5C
<azonenberg>
and they don't extrapolate within that range
<azonenberg>
although at higher levels of the chain like at NIST they might well do that
someone--else has joined #scopehal
<someone--else>
I wonder how far up the chain are Josephson junctions.. NIST certainly, but perhaps some other labs too
<sorear>
you can just buy josephson voltage standards
<someone--else>
yes, but they are rather uncheap..
<azonenberg>
well a josephson junction still needs calibration for frequency iirc, right?
<azonenberg>
doesn't it turn a frequency into a voltage?
<azonenberg>
So in order to get exactly 1V out you have to put in exactly X Hz
<azonenberg>
also it looks like nist sells josephson standards
<azonenberg>
so presumably there is a demand at major cal labs for them
<sorear>
supracon too
<d1b2>
<j4cbo> high-precision time is generally available though
<d1b2>
<j4cbo> only two kinds of device contain a subcomponent called a “physics package”, and only one of them can be bought on eBay 😛
<sorear>
i am inclined to suspect that an uncalibrated COTS cesium clock has better fractional uncertainty than most voltmeters
<azonenberg>
j4cbo: lol
<azonenberg>
sorear: well i know some hardcore voltnuts who have 3458A's at home as well as their own in house voltage standards
<azonenberg>
One of them actually had his home volt standard calibrated directly to NIST
<azonenberg>
and is already making plans for a home josephson standard
bvernoux has quit [Quit: Leaving]
<sorear>
to be clearer, I think the best voltage standards *that exist* have a fractional uncertainty around 1e-8 due to leakage paths etc, while that new SMD cesium clock was 1e-11 and the commercial tubes are somewhat better