[time-nuts] OCXO retrace, how long is reasonable?

Sat Sep 5 22:52:24 UTC 2020

I don't agree with the statement about meter pricing:
> One can buy 5 digit LED meters on ebay for $5 each +shipping.
The lowest cost "5 digit" voltmeter I can find listed on any site is the AN870, which is sold for a multiple of 5X to 9X of the price you quote. I can find no published accuracy specifications for that model. And I'm not sure why "LED" meters were specified, since various display technologies are used in commercial meters. In my experience, most 5 digit meters will have some type of LCD display.

In contemplating the results of voltage measurements taken at different times, we need to be careful about confusing several terms which often get thrown around.

Accuracy - This is a catch-all term which should only be trusted after you read all the footnotes. If there are no footnotes you should not not trust the claimed accuracy. Factors which should be fully declared include temperature, humidity, and time after calibration at a standards lab.

Resolution - You would thing this term would be easy to interpret, but for digital voltmeters that's not the case. You might find several different meters which all appear to have 5 displayed digits, but the maximum reading on these different meters (including a comma to make it easier to interpret) might be: 
10,000
11,999
19,999
29,999
99,999
So some might claim that all of these have "5 digit" displays, while the effective resolution on some is only 1 part in 10^4 for positive values. The industry has marketing terms such as "4.5 digit" which I won't go into here. There is also no guarantee based on the number of displayed digits that the resolution is truly one digit and monotonic. So a slowly increasing voltage might display this display over time:
0.00
0.02
0.01
0.04
0.03
0.05
That shows a display which appears to have 10 mV raw resolution, but the response is not monotonic or it includes some noise. It's common for high resolution ADC and DAC IC's to exhibit a small amount of mon-monotonic behavior. In many cases noise causes products to be specified as +/- 1 or +/- 2 digits around the expected value, and that noise might largely hide the non-monotonic behavior.

Uncertainty - You can speak of absolute and relative uncertainty due to both systematic and random error sources. For example, your "5 digit" meter might show a 2.0000 V reading on January 1 and 2.0020 V on July 1 on a source which is known to be stable to 1 ppm (1 x 10^-6) relative accuracy over that time range and changes in ambient conditions. The difference in the readings might be due to drift over time (which might vary as time proceeds in direction and slope due to aging), temperature or humidity changes, or other sources of drift which can't be determined. I'm describing test instrument errors which many users don't understand because they have never carefully read the specifications section of the manual.

The suggestion of using multiple low cost meters doesn't help. There is no reason to believe that the multiple results will be correlated or uncorrelated with respect to time, temperature, humidity, or unknown factors. For example, if you had 4 meters and they all showed identical readings of 2.0000 V at 9 AM and 2.0020 V at 4 PM, the actual voltage might have decreased over that time interval but the room temperature increased by 5 C and the meters all responded identically to that change. Don't assume that all of those digits can be trusted!

If you measure the EFC voltage (part of a feedback loop) driving a varicap (varactor or variable capacitance diode) in an OCXO every other day and find that it's changing over time, your results could be due to:
(1) The DVM is not stable to that degree of uncertainty.
(2) The varicap is not stable to that degree of uncertainty. It might drift due to aging, temperature, mechanical stresses, air pressure. The sensitivity of the varicap to EFC voltage changes might change over time due to various factors.
(3) The crystal itself might be aging or drifting due to environmental effects.
(4) The oven or associated circuitry might be drifting due to aging or an environmental effect. For example, there might be a resistive network between the EFC test point and the varicap diode. Any change in that network might cause the EFC voltage required to keep the OCXO on frequency to change.
(5) Power supplies might be drifting over time or due to environmental changes.
(6) The frequency you are locking to might be changing.
(7) You might be improperly probing the system with your meter -- more on that below.

My point is that the EFC voltage measured on a real meter may drift over time due to various effects. As long as the range of EFC voltages exhibited are well within the allowable control range of both the EFC generating circuit and the frequency control device (such as a varicap), there should be no reason for concern. 

For example, if the EFC voltage changes by 1 mV as the temperature drops 10C, you might assume that the oven temperature combined with the temperature coefficient of the crystal is causing that change. But the true cause of the measurement change might be the temperature or humidity coefficient of the meter, or placing the low meter test lead on the metal chassis without realizing that the oven current flows through a wire with 10 m ohm resistance to chassis ground, so that a 100 mA increase in oven heater current causes that 1 mV EFC change due only to your improper meter probing. The feedback loop corrects properly even with that ground loop offset, but you can't assume that the EFC voltage change is related to changes in the crystal temperature.

Sorry for the long dissertation. I worked at an electronic calibration lab for 4 years, designed some test equipment, and worked as an Application Engineer in test equipment sales for 32 years before retiring late last year. Unfortunately, it's common for users to improperly assume that the reading displayed on the test instrument is the actual precise value of the device under test.
--
Bill Byrom N5BB

On Sat, Sep 5, 2020, at 2:33 PM, Perry Sandeen via time-nuts wrote:
> Learned members,
> Wrote: 
> Well, I could measure the EFC voltage every other day and if it'schanging, then it's the OCXO that’s drifting, if not (or not inaccordance to the DAC graph) it's the DAC itself.
> 
> One can buy 5 digit LED meters on ebay for $5 each +shipping.
> 
>  While I wouldn't necessarily trust their accuracy, at that price one could buy 3 or 4 and continuously measure  several variables.
> Regards,
> Perrier
> _______________________________________________
> time-nuts mailing list -- time-nuts at lists.febo.com
> To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
> and follow the instructions there.
> 