[time-nuts] Thunderbolt? (re simple gpsdo.)

[David]

On Mon, 02 Jan 2012 17:14:37 -0800, Hal Murray
<hmurray at megapathdsl.net> wrote:

>
>> Time constant is just R*C.  If you have a 1000uF cap and a 1K resistor you
>> have 1 second.  In theory you could build 100s just by using a 100K resistor
>> but I think real world components are not perfect enough. 
>
>Does anybody know anything about the temperature coefficients of large caps?  
>I found data for ceramic caps, but when I added "electrolytic" all I got was 
>lifetime stuff rather than capacitance change with temperature.
>
>I'm not interested in the frequency shift of the filter as the temperature 
>but the voltage shift due to a fixed charge as the capacitance changes.

That is not something you are likely to find specified because there
is no need to test or guarantee it in the applications that
electrolytic capacitors are generally used in.  I would also expect
problems because of dielectric absorption, leakage, and possibly
noise.

In the best case, you would probably need to qualify the capacitor
yourself.  I ended up doing that anyway with a polypropylene film
capacitor used in a similar application because we needed to control
leakage at high temperatures and that was not something that the
manufacturer tested for or guaranteed.  In that case, the integrator
had a drift below 1 uV/sec at room temperature in a production
environment using a 0.47uF polypropylene film capacitor and an
equivalent time constant of about 5 seconds.  The PC board used guard
rings and the summing node was wired into the air although in a good
environment that would not have been necessary.

With care and the proper environment I think, 100 times better would
not be out of the question but I suspect other factors like 1/f noise
would become an issue.  I think I would try a charge balancing scheme
instead of integrating or averaging the output of the phase/frequency
detector directly but maybe that is getting away from simple.