[time-nuts] Another "atomic" clock question

Jim Lux jimlux at earthlink.net
Mon Mar 3 09:27:51 EST 2014

On 3/3/14 2:18 AM, Hal Murray wrote:
>>> Junk crystals are good thermometers.  Ballpark is 1 ppm/degree-C
> albertson.chris at gmail.com said:
>> So does this mean I can epoxy a sandstone power resister to a junk crystal
>> and keep the frequency exactly perfect by varying the power in the resister?
> Sure, for some values of "perfect" and such.
> I've occasionally thought about building something like this, just for the
> hell of it to see what happens and/or what I learn, and or how good I/we can
> get on a low budget.
> I think there are two problem areas.  One is sensors and control algorithms.
> The other is board layout.
> Where is the sweet spot on complexity vs accuracy?  I'm looking for
> science-fair level of goodness rather than super-expensive to get another 0
> or two.
> What's the best low-cost way to measure temperature?  Many of the obvious
> choices are only good to 0.1 C.  That's great if you are trying to measure
> room temperature or or want to keep your CPU from melting, but it's probably
> leaving a lot on the table if you are interested in the frequency from a
> crystal.
> My straw man would be a thermistor and OP-Amp feeding into the ADC on your
> favorite uProc.  Maybe the other side of a bridge would be adjustable.

A number of microcontrollers have onchip temperature sensors (Freescale 
Kinetis, for instance).  If the controller were bonded to the crystal 
housing, that might be enough coupling.

Could you hold 0.1 or 0.001 degree? the chip has a 16 bid ADC, although 
I wouldn't trust the bottom bit or two because of noise. But in any case 
1 LSB is 3.3/64k or about 50 microvolts.  The temperature sensor slope 
is 1.715 mV/C, so that's in the 0.03 C/LSB range.. On a good day, you 
*might* be able to hold 0.1 degree, assuming there's no systematic errors.

> How much power do you need to keep things warm?  I'm assuming something like
> a watt or 2 with something like a PWM from the uProc.

More information about the time-nuts mailing list