[time-nuts] Curve fitting

Ulrich Bangert df6jb at ulrich-bangert.de
Wed Dec 23 11:24:47 UTC 2009


Mark,

this is how I have done it:

1) Record EFC voltage and temperature over a time of at least some days so
that you record some of the typical night/day temperature changes in a
normal flat. In my case this has been in the order of 3-4 degree C.

2) Run a cross correlation between the two sets of data. The result will
show you the characteristic delay of your DUT how it reacts to temperature
changes. This delay can be in the order of minutes for example with
rubidiums which react less to temperature changes than xtal oscillators but
react FAST because their physics package is coupled well to the surrounding
due to their heat sink. With a good OCXO (a FTS1200 for example, which
features a deware inside for temperature de-coupling) this time can be in
the order of some hours! 

3) After you have found this delay apply it to your data to shift one of
your data sets in time so that the delay is compensated, i.e. as if the DUT
would react immediately to temperature changes without any delay. Use the
EFC gain figure to compute frequency changes from EFC changes and draw the
frequency changes as a function of temperature in a x/y diagram. This will
not exactly give you a straight line. Instead, what you will see looks more
like a "cloud" of points. Nevertheless you will notice the clear correlation
between temperature and frequency even if the tempco is small. Run a linear
fit to compute the DUT's sensivity to temperature changes.

Note that beneath the measurement itself all actions can be performed with
my PLOTTER utility.

Best regards
Ulrich   

> -----Ursprungliche Nachricht-----
> Von: time-nuts-bounces at febo.com 
> [mailto:time-nuts-bounces at febo.com] Im Auftrag von Mark Sims
> Gesendet: Mittwoch, 23. Dezember 2009 09:48
> An: time-nuts at febo.com
> Betreff: [time-nuts] Curve fitting
> 
> 
> 
> Hello Bruce,
> 
> Yes,  there are all sorts of pregnant packages and web sites 
> that will solve the problem in a jiffy.  But LAPACK,  for 
> instance,  is over 45 megabytes long.   
> 
> What I am looking for is a targeted solution to the 
> problem...  a routine or set of routines that I can drop into 
> Lady Heather with a minimum of muss and fuss that will spit 
> out those two magic numbers.
> 
> The idea is to collect data for a while,  press a key,  and 
> Heather will characterize your oscillator.  I have code in 
> there that does this if you have the active temperature 
> control working (stabilize temperature to get osc drift rate, 
>  slew temp to get osc tempco).  I would like to be able to do 
> it for the more general case where the unit is not under 
> temperature control.
> ---------------------
> Try one of the Lapack derivatives such as clapack, Lapack++ etc
> However you will need the optimised BLAS as well. 		
>  	   		  
> _________________________________________________________________
> Hotmail: Trusted email with powerful SPAM protection. 
> http://clk.atdmt.com/GBL/go/177141665/direct/01/
> _______________________________________________
> 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.




More information about the time-nuts mailing list