[time-nuts] Re: CTI OSC5A2B02 OXCO testing

Tue Jun 6 01:28:02 UTC 2023

 I was sloppy with my post, but I'd been at it for quite a while and was tired. Next time I'll wait until the next day to post.

OK, time to 'fess up.

I've got a 1990's $500k MSRP T&M lab. Missing the bits to do precision time closer than 1e-10. I was using the 34401A to monitor the voltage. 

I have a Tek 11801 sampling scope and a rich suite of sampling heads. I can measure to 20 femptoseconds, though noise becomes a serious issue at delays that small. But I think I see how to use that to compare OXCOs.

I'm already getting better than 1e-10 accuracy (the limit of the 5386A). So the project is to see if I can achieve 1e-11 or better. What I'd really be pleased with would be 1e-12 from a 4 OXCO reference over the course of a year. based on cals at 1,2, 4, 8 and 16 month intervals. Does it make any sense? No. But I've got an idea for solving a difficult problem and those are the things I enjoy playing with. This is a pure aging correction. The shift for an OXCO operating continuously in a stable environment.

The key to this is using basis pursuit to select from say, 1 million possible aging curves, the 2-3 curves whose sum best matches the available data. My testing of numerical models of the heat equation, an infinite sum of exponentials, showed that I could predict future values to with 1-2% for as far into the future as I had data from the past.

To get acceptable adjustments, you can't have a single pot for the entire divider. So you have to have a small pot between a pair of fixed resistors chosen so the temp cos cancel. Vishay is pretty much what you can get and they are not cheap. The voltages I stated were the EFC voltages measured by the 34401A.

I have AR488s for GPIB. No worries there.

The concept I'm working on is to model the aging of N OXCOs, adjust Vref to keep the output on frequency and at octave time intervals recalibrate the unit.

Let's suppose an OXCO with a 365 ppb first year aging rate. If I predict and correct Vref for the aging rate to ~2% let's call the error 7 ppb. That's a 50x improvement in accuracy by using an MCU to adjust Vref over the course of the year. If the 0.02 ppb/s is random, then having 4 OXCOs feeding a common tuned circuit should reduce the random frequency errors from 0.02 ppb/s to 0.01 ppb/s (1/sqrt(N)).

Have Fun!
Reg
     On Monday, June 5, 2023 at 03:06:40 PM CDT, Tom Van Baak via time-nuts <time-nuts at lists.febo.com> wrote:  

 Reg,

Ah, it wasn't clear the 10 readings were from 10 different oscillators. 
Thanks for the clarification. Ok, what you see then is simply that all 
ten oscillators can be set on-frequency and still be within your desired 
EFC tuning range. That's a good result.

The next step is to determine what your accuracy or stability 
requirements are. There's no point in spending extra money for an ultra 
stable Vref if all you need is just 7 or 8 digits of accuracy, or if 
that's all your 5386A counter can even measure.

I don't quite understand your comment about expensive resistors. Most 
instruments with OCXO have an EFC adjustment, usually implemented as a 
single 10T or 20T panel- or PCB- mounted pot. To first order tempco 
doesn't matter because the pot is being used as a *divider* across Vref. 
Yes, a temperature coefficient in the pot changes the resistance of each 
leg but it doesn't change the ratio.

It's a cheap experiment for you to try. Surely you have a ~10k 
multi-turn pot lying around. Measure the frequency periodically over a 
day and also record temperature. As a baseline do the same with EFC 
shorted. The difference between the two sets of data tells you how much 
of the tempco you see is in the OCXO and how much is in the Vref / 
divider / EFC.

You mentioned you have a hp 34401A as well. You could use that to 
directly monitor the EFC voltage and add that to your plots. Best to 
automate the data collection; do you have a GPIB adapter for your DMM 
and your frequency counter?

/tvb

On 6/4/2023 11:51 AM, Reginald Beardsley via time-nuts wrote:
> Tom,
>
> Thanks. Typo is correct. I didn't describe the data properly.
>
> Those are a single measurement on 10 OXCOs to determine the 10 MHz bias voltage requirement for each. I was adjusting the PSU voltage which is rather coarse, to get close to 10 MHz (within 1 Hz) and measuring the actual voltage with a 34401A.
>
> The purpose is to calculate what precision, low temp co resistors I need to set the Vref for each oscillator. With particular care to try to make it neutral at a convenient temperature. Those resistors are seriously not cheap. The Vishay 50 ppm/C precision foil trimmers ar $25.
>
> The 5386A was running on the internal XOCO as I simply wanted order of magnitude values for each OXCO. So an accuracy of ~2 ppb (5386A and DUT) for time and <50 ppm for voltage.
>
> I'll run the same series with a better reference for Vref so all are tested for frequency at 2 V input to Vref.
>
> Reg
_______________________________________________
time-nuts mailing list -- time-nuts at lists.febo.com
To unsubscribe send an email to time-nuts-leave at lists.febo.com