Jitter measurement

I want to make a jitter measurement for the PWM output of this STM32F302R8 Nucelo board. Its datasheet is given here. By jitter I mean the histogram of pulse periods which shows how the periods vary from the mean.

If I'm not wrong, there's no jitter spec in these documents. So, in my case I will generate a pulse train of 1024 pulses, where each pulse period is 655 μs.

If I'm asked to quantify jitter in this case, is there a rule of thumb how many pulses is enough to quantify the jitter? (If I take too many pulses I cannot sample the rising edges fast enough)

• refresh your statistics, is a mathematical question, not engineering. you are measuring something that has some uncertainty. Let's assume your jitter distributes normally with zero mean and std. dev. sigma. These are the parameters you want to obtain (mu and sigma) from actual measurements. You will get different levels of confidence depending on how many samples you take. In other words, there isn't a precise number. from experience, I would say a very minimum of 100 and up to 1000 if you want good accuracy (but that would make sense only if your measurement layout is also accurate)
– Joan
Commented Jun 13, 2022 at 8:24
• If you know nothing of the statistical properties of the jitter you cannot calculate beforehand how long sequence is needed to estimate the characteristic numbers with a certain confidence. Data communication standards declare the number of the needed pulses and that depends on application. Unfortunately I do not know this subject well enough, but the wanted numbers of pulses are thousands. You must cumulate the distribution of your pulses until you believe it's stabilized - or know the causes of the jitter to make any predictions.
– user136077
Commented Jun 13, 2022 at 8:24
• My problem is, with the equipment/scopes I use if I record 100 samples or 1000 samples then I cannot catch enough points at rising edges. Do they use special high definition equipment in industry for such measurements?
– cm64
Commented Jun 13, 2022 at 8:43
• If I remember well, the samples don't need to be "consecutive" ... Commented Jun 13, 2022 at 8:53
• Since the jitter is probably constant in nanoseconds, if you don't have hardware to continuously acquire edges, you could reduce the pulse period so that more edges fit into the same interval. You can also repeat the measurement multiple times, which will help you catch long term drift (e.g. the clock frequency changes as the device warms/cools). Commented Jun 13, 2022 at 16:11