How does oscillator frequency stability affect timing?

Question

If, for example, a crystal oscillator is specified as +/- 100ppm what does that mean? That the exact frequency is uncertain to that degree, or that the frequency varies by that degree over normal operation? Also, does that automatically imply that any timing based on it is inaccurate to the same degree?

Answer 1

The "ppm" spec is usually for "initial accuracy". It means that the parts coming from the production line might vary by that much if the frequency is measured right after they are produced (or maybe after some defined burn-in operating time), at 25 C.

If the temperature varies, the frequency will vary in response, over and above what is specified by the initial accuracy spec.

The frequency will also tend to drift over time, an effect that is often specifed as "aging". This effect is also beyond what's specified for initial accuracy.

For example, the part linked in the other question mentioned in Eugene's comment, FX135A, has an initial accuracy of +/- 20 ppm, but it will drift as much as +/- 3 ppm per year due to aging, and .045 ppm per degree C due to temperature.

Also, does that automatically imply that any timing based on it is inaccurate to the same degree?

In crude terrms, yes. But you could imagine testing your device against a more accurate clock and storing calibration data to allow your system to correct for the timing error due to initial accuracy.

Edit

As another question says, some vendors specify the overall accuracy including some temperature and aging effects. Be sure to read your datasheet carefully. If they don't say what effects are included, though, your safest bet is to assume the accuracy spec is for initial accuracy only and that temperature and aging effects will add additional frequency uncertainty.

Answer 2

I'm going to disagree with the other answer and (strongly) suggest you actually read the data sheet.

Often oscillators (not necessarily crystals) have a tolerance specified that includes temperature changes and some aging. For example (the first one I picked) this one specifically says that the stability figure (50ppm is standard) is "Inclusive of initial tolerance at time of shipment, changes in supply voltage, load, temperature and 1st year aging".

Yes, any timing dependent on it may vary by at least that amount. There may be other causes for timing to vary- for example changes with output rise and fall times with temperature could slightly change the timing. Changes with supply voltage and temperature typically assume a slowly changing supply voltage and temperature, and that may not be true. Also, there will always be some jitter- the frequency spec is averaged over many cycles so the timing error might be relatively large (as ppm of the total) if only a few cycles are involved. The part I linked to has a p-p jitter spec of 50ps, which sounds pretty tiny, but it's actually 10x worse than the stability spec over a single cycle at 10MHz (more for higher frequencies).

Answer 3

The other questions get into detail about how the ppm may drift due to temperature changes etc. but I'm not sure they actually explain very well what ppm is and how you apply it.

"ppm" simply stands for "parts per million" and is a different way of stating precision. It's like percent, but based on millionths instead of hundredths. The relationship is 1,000,000 / 100 or 10,000 to 1%. ±20 ppm is the same as ±0.002%, or ±0.00002 The latter is a little harder to get one's head around than 20 ppm but easier to use in calculations since you simply multiply.

Your question asked what 100 ppm would be, it would be ±0.01% or ±0.0001. But tolerances of 20 ppm or 10 ppm are more common for crystals.

A 32.768 kHz crystal with a tolerance of 20 ppm or 0.002% can be off as much as 32768*0.00002 or 0.65 Hz. A 10 MHz crystal can be off as much as 10,000,000*0.00002 or ±200 Hz.

So a crystal with a tolerance of 20 ppm can be off as much as 0.00002*60*60*24*30 or 52 seconds. Notice this doesn't depend on the crystal frequency, just the tolerance.