rf-design wrote on Mar 9th, 2010, 2:34pm:An RC-oscillator inherit the 50-200ppm/K from the Poly-Res. The clear solution is to adjust the RC every RTC period and second to use the RC also as RTC. Because within the RTC periods the temp drift is very low. The only critical stuff was the initial tune in code and model detail for verification.
Sounds simple!
Now - go do something very
quantitative:1. Find the frequency accuracy needed to hit a tuned circuit (the crystal) with enough accuracy to realize less than 3dB-6dB of passband loss. The Q of your circuit is 0.5 x 10E6. For the sake of argument, lets call it a 10MHz crystal.
I expect you will find an answer that is under 50Hz, and if my memory serves me correctly the answer is around 20Hz.
2. Go model that RC oscillator, put in all the PVT corner variables and determine the frequency variance.
I expect you will find a +/- 40% variance (give or take)
3. Determine how much resolution you are going to need to get an adjust circuit that can be "tuned" within 1/2 of the tuned accuracy stated above. (best guess is 10Hz) - so with an RC oscillator that varies from 14MHz to 6MHz (delta = 8 MHz) your adjust system needs 800,000 adjustment points.
Thats a lot of tuning and adjustment to hit a very narrow frequency window.
Now, after you get that circuit tuned up to hit the right frequency, change something, pretty much anything, but lets say change the temperature, or the power supply voltage.
You just lost your tuning point, and you are outside the frequency window.
Your numbers may vary, but the above is the concept. Problems are seen in simulation, and I have seen this idea in the lab and seen that it has problems there too.