MAMEWorld Forums - The Loony Bin - Xtal compensation theory 101: Part 2

> Now, on to part 2 and why this won't work in other situations...

(Forgive the borrowed data plot below, the results are consistent with expectations.)

Shown is a plot of a compensated TCXO. This is a more common and less expensive compensation method. You ~still~ won't find these in most clocks, at $2-10 a pop. This compensation method usually utilizes some form of reverse error generation applied to a crystals natural error curve. Generally you would run a crystal over temperature, capture it's frequency response, and then generate a 4th or 5th order polynomial equation for that response. To compensated for the error in any given crystal, you then would calculate a second polynomial equation to generate a cancellation signal to bring the crystal as near to nominal as possible. There are several means by witch to apply that correction to the crystal. All of them have their pluses and minuses.

When you are done, you have something that looks like the attached image below. Note that the response isn't as predictable and smooth as the last attachment. You might be thinking "I still don't understand why I can just store a table of values to compensate it further?" The answer to that lies in aging, drift, hysteresis, and the compensation method in itself.

Aging/drift - As the crystal is used and stressed, subtle changes will happen over time. You may still retain most of the compensatory effects, but all crystals do move. the amount of error at 25C today, may not be the amount of error at 25C next month. You'd have to re-train you values every so often, which isn't in the realm of possibility for most. You are being guaranteed +/- a certain spec over temp, not at defined error at a given temperature point.

Hysteresis - The crystal may not respond in the same manner when traveling forward on a temperature scale as it does in reverse. The amount of error you have compensated for in one direction, may be off a few degrees in the reverse direction. Due to the varied nature of the compensated curve this could be a minor issue, but could also produce very unexpected results.

Compensation - The nature of the beast itself is unpredictable. You are applying a polynomial response curve to a waveform taken once, twice if lucky. Your cancellation/correction curve will never change. Your crystal response will. This could lead to perturbations (random spikes), or even phase shifts. Depending on how the response curve is generated, there could also be hardware limitations to consider. Bottom line, even though you will stay within a certain range of accuracy, you won't always get the same response each time you sampled your compensated crystal over temp.

[ATTACHED IMAGE]

Attachment