[time-nuts] Next upgrade

Sun Nov 26 14:39:54 EST 2017

Hi

If you only run over 10% of the EFC range, you only gain 3 bits. If the objective is 
in the 22 bit vicinity, (maybe 20 maybe 22 …) you really don’t get enough bits at a 10% 
span. From a lot of years of playing with control loops, if you need 20 *good* bits, you better
have a few more than that in the design …. Indeed there are converters out there with 1/8
LSB performance. There also are a lot of them with “guaranteed monotonic” as the main spec.
In that case you may get 2 LSB of “jump” as you do this or that….

Indeed another alternative is to let the OCXO warm up for a month. Then adjust a pot to center
things up. Run a “fine range” ADC to keep it happy. Come back in three to nine months and 
tweak the pot again. The main risk is a power outage and waiting a few weeks to get things 
back up and running again. 

A lot of this depends on how much of an EFC range you have and how much aging you expect. 
If you have 4 PPM of EFC and expect 1x10^-9 per month that gets you a pretty small range. If
you have 5x10^-8 of EFC and expect 1x10^-9 per day, the entire EFC may not last you for very
long at all. 

Another factor is temperature. Your OCXO may be happy at 1x10^-11 / C. If your control circuit
is good at the 5x10^-10 / C level that may be ok or it may be a problem. Either way, your control
range needs to accommodate both the OCXO and the rest of the circuit on top of the aging. 

Bob

> On Nov 26, 2017, at 1:05 PM, Azelio Boriani <azelio.boriani at gmail.com> wrote:
> 
> ...and what about shrinking the 16bit over the fraction of the EFC
> range that, for example, the OCXO will be using for the next 5 years?
> 16bit over 10V are as 20 (a little less, OK) over 1V, if I can use my
> 16bit over 1V for the next 5 years, when the DAC will be near full
> scale I can "trim" the aging.
> 
> On Sun, Nov 26, 2017 at 6:25 PM, Bob kb8tq <kb8tq at n1k.org> wrote:
>> Hi
>> 
>> If you sum two DAC’s without any sort of feedback, you get problems when the
>> “coarse” dac is changed. You have no way to know the step size of the coarse
>> dac to (say) 20 bit precision.
>> 
>> As an example : If you are after 20 “good” bits, you might overlap
>> them at the 10 bit point on the coarse dac. That would give you 22 bits on the
>> summed output. It would give you enough extra bits to take care of any odd
>> things that might be going on. You only have 1/1024 of the total range before
>> you must tune the coarse dac. Even with a good set of parts, you *will* be
>> doing coarse tuning.
>> 
>> Bob
>> 
>>> On Nov 26, 2017, at 12:13 PM, Azelio Boriani <azelio.boriani at gmail.com> wrote:
>>> 
>>> Is summing a "fine tune" 16bit DAC and a "coarse tune" 16bit (or less)
>>> DAC with an op-amp not good enough?
>>> 
>>> On Sun, Nov 26, 2017 at 5:53 PM, Bob kb8tq <kb8tq at n1k.org> wrote:
>>>> Hi
>>>> 
>>>> Each time I’ve tried the method in the app note, there has been a tone in the output
>>>> spectrum at the sample rate of the ADC. I’ve never found a way to do the grounding
>>>> that eliminates it. The tone is large enough to show up as a spur on a “typical” OCXO
>>>> when it goes into the EFC port.
>>>> 
>>>> Bob
>>>> 
>>>>> On Nov 26, 2017, at 8:56 AM, Ole Petter Rønningen <opronningen at gmail.com> wrote:
>>>>> 
>>>>> I guess everyone has seen this, but Linear has a nice appnote «A Standards Lab Grade 20-Bit DAC with 0.1ppm/°C Drift»
>>>>> 
>>>>> http://cds.linear.com/docs/en/application-note/an86f.pdf
>>>>> 
>>>>> Ole
>>>>> 
>>>>>> 26. nov. 2017 kl. 13:50 skrev Magnus Danielson <magnus at rubidium.dyndns.org>:
>>>>>> 
>>>>>> Hi
>>>>>> 
>>>>>>> On 11/26/2017 02:26 PM, Attila Kinali wrote:
>>>>>>> Though, if you have a decent 16bit DAC and want to get to 18bit,
>>>>>>> that's fairly simple using delta-sigma modulation... if you can live
>>>>>>> with a low pass fillter after the DAC. But the DNL will be the limiting
>>>>>>> factor here (unless you use some special techniques) and the (absolute) INL
>>>>>>> will not get better, for obvious reasons.
>>>>>> 
>>>>>> I needed 19 bit rather than 16 bit, so I implemented an interpolation scheme. A first degree sigma-delta would also be possible, but for low ratios what I did was more efficient.
>>>>>> 
>>>>>> A first degree sigma-delta is fairly simple thought.
>>>>>> 
>>>>>> The trick is that you want to push the noise high up so it becomes trivial to filter, then the filter will not be hard to design and won't be low enough to cause PLL instability and implementation troubles.
>>>>>> 
>>>>>> Cheers,
>>>>>> Magnus
>>>>>> _______________________________________________
>>>>>> time-nuts mailing list -- time-nuts at febo.com
>>>>>> To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
>>>>>> and follow the instructions there.
>>>>> _______________________________________________
>>>>> time-nuts mailing list -- time-nuts at febo.com
>>>>> To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
>>>>> and follow the instructions there.
>>>> 
>>>> _______________________________________________
>>>> time-nuts mailing list -- time-nuts at febo.com
>>>> To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
>>>> and follow the instructions there.
>>> _______________________________________________
>>> time-nuts mailing list -- time-nuts at febo.com
>>> To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
>>> and follow the instructions there.
>> 
>> _______________________________________________
>> time-nuts mailing list -- time-nuts at febo.com
>> To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
>> and follow the instructions there.
> _______________________________________________
> time-nuts mailing list -- time-nuts at febo.com
> To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
> and follow the instructions there.