We got some pretty big orders and we figured that we'd need about 1000 hours of production test time to get the stuff out. So I am designing a couple of new test sets and the kids will grind out a pile of Python, to replace classic manual test procedures with more automated stuff.
A couple of the product tests need measurement of time jitter. We had been using old Tek 11801C sampling scopes to measure jitter, but they are gigantic and slow. So we got to wondering if a cheap Rigol could measure jitter.
The DS1204B is a 4-channel 200 MHz, 2Gs/s scope for $1395. It doesn't do any jitter measurements. It will measure the time between rising edges of two channels, so we'll work with that.
It's always hard to figure out what scopes are doing internally; that's not often discussed in manuals. Experimentally, the DELAY A->B measurement seems to work well in both free-run and single-shot mode. In ss mode, one can measure a delay, say 5 ns from rising A to rising B, and it's right, and one can change that to 6 ns, fire one more acquisition, and that will be right too.
So we wrote some Python to do a bunch of DELAY A->B measurements and compute the RMS jitter. We used one of our DDGs as the source, which has about 6 ps RMS jitter between outputs. The Rigol reported different jitters depending on whether we were measuring continuous or single-shot. Single-shot takes longer to run, about 100 ms to fire each shot and read the delta-T. Continuous mode lets us snoop the time delta in about 10 ms (this is Ethernet) but reports more jitter, so I'm not sure what it really means.
Taking measurements in single-shot mode and doing the math seems to have a jitter measurement floor a bit below 30 ps RMS, which is amazing for a $1600 scope with a 2 GHz ADC clock, a 500 ps sampling interval. Shannon was one smart dude.
We generated some test signals with known jitter. Once the input jitter gets above 500 ps RMS or so, it doesn't seem to matter whether we use continuous or s/s mode for the time-interval measurement.
For lower jitter measurements, 100 acquisitions in s/s mode might be enough, and that will take about 10 seconds, which is tolerable.
Scopes that internally measure jitter are fabulously expensive, which is silly because it's just a little math.