[time-nuts] need example frequency vs temp equation

Jim Lux jimlux at earthlink.net
Thu Oct 13 05:52:10 UTC 2011


On 10/12/11 10:10 PM, Bernd Neubig wrote:
> Hi Jim,
>
> There are different types of TCXO compensation techniques on the market.
> Each of them generating a different style of f(T) characterisitcs.
> Furthermore the f(T) response varies from unit to unit, because each TCXO is
> usually uindividually compensated (or sometimes in groups of similar quartz
> f(T) characteristics.
> To name a few compensation types:
> The classical types can be broken down in direct and indirect compensation:
> 1. the first one using a thermistor/capacitor/resistor network connected in
> series to the quartz crystal resonator. This network represents a temprature
> dependenta load caopacitants to the crystal.
> 2. the indirect ones using a thermistor/resistor network which generates a
> temperature-dependent DC voltage, which is fed to a varactor diode (in
> series to the crystal) and thus changing f over T. Sometimes the passive
> network is combined with an op-amp to realize a higher voltage swing.
> 3. The modern TCXO (all these small ceramic packaged SMD units) use IC-based
> compensation techniques. There are different TCO on the market which differ
> in their working principle slightly. But in general, most of those IC's
> contain a temperature sensor, from which a DC voltage represented by
> polynomial of 3rd or higher order is generated by analog techniqes:
> The coefficients for the polynomial are
> - a0 = reference voltage
> - a1 = outoput from temperature sensor
> - a2 = output from temperature sensor multiplied by the same with an
> analogue multiplier
> - a3 = output of a2 multiplied with temp sensor output  etc.
> These components are fed into an analogue summing amplifier through analogue
> potentiometers, which are setting the magnitude of each coefficient.
> This summed-up voltage ploynomial feeds one or two varactor diodes in series
> to the crystal.
> In the (still individual, but highly automated)compensation process, the
> coefficient potentiometers are set set through a serial data line such, that
> the f(T) characteristic shows minimum deviation over temperature. This
> process runs through the whole operating temperture range in small
> temperature steps, mostly in both directíons to take into account some of
> the hysteresis of the crystal's f(T) characteristic.
> 4. Besides these techniques there are some other approaches, such like the
> first generation of digitally compensated TCXO, which were using loo-up
> tables for eacht temperature increment (bit), which contains the digital
> word for the necessary compensation voltage. The disadvantage of this method
> are the discontinuities between eacht temperature bit, causing small
> frequency jumps and/or jitter
>
> To conclude: Because of the individual process, TCXO do not show any uniform
> f(T) characteristic. You can fit it by a higher order polynomial, but the
> responses are looking different for each individual unit.
>
>
> Best regards
>
> Bernd, DK1AG
>
> AXTAL GmbH&  Co. KG
> www.axtal.com
>





Thanks a lot.. This was quite informative.

I think all I need for this purpose (it's an example of what you might 
have to deal with) is any old scheme. Given the hidebound reactionary 
conservatism of spaceflight electronics designers and their even more 
conservative review board members, the first one (temperature dependent 
resistor and capacitor, or maybe varactor) is probably the most plausible.

And for the non-temperature compensated case (e.g. the CPU clock), the 
standard AT cut cubic will probably work just fine.

I found a thesis from someone who was modeling this kind of thing 
(actually he was developing an set of tools to design it) and I can 
probably crib his matlab code.  (which is all I really need...)

I'm trying to come up with some illustrative experimental scenarios 
where you have multiple widgets varying in temperature (orbiting around 
something, so they cycle every 90-100 minutes, with a bigger cycle on a 
monthly/annual basis), and you want to do comparisons/ensembles/doppler 
measurements.

So it's not that it has to exactly match any particular scheme, just be 
representative of some scheme in terms of overall magnitude and number 
of wiggles in the curve of f(T).   I could arbitrarily just pick 
something like a moderate order polynomial or spline that "looks good", 
but hey, if someone has a model out there for a real device, I might as 
well use that.





More information about the Time-nuts_lists.febo.com mailing list