Siglent SDG-2042x 10Mhz reference signal is bad

[pigrew]

checked SDG2000X 10 MHz reference clock quality:  (youtube link was here)

Yeah, that looks to be about the limit of your scope's internal measurement. One small tip is that you can change the soft-menu selection by pressing the soft-key multiple times (instead of going to the multi-function knob). Also, it likely has a statistics display for measurements (from the measure menu), so you can see, for example, the calculated standard deviation of the period. You scope may also have a zoom mode which lets you see the detail better.

I repeated the measurement with a Tek 784D (4GSa/s), and it also seems to be close to the noise floor of the equipment. When directly using the BNC cable between the instruments, the rise time is increased by capacitive loading, allowing the vertical noise of the ADC to dominate the measurement's noise. With an active probe (<1pF) and a short cable, a sharp edge appears at the start of each edge, and then it slows down. I got the lowest jitter numbers by triggering on this small edge (at -1.1V). Be aware that RG-58 coax has about 25 pF/ft, so it will load the output quite a bit. The SDG2042X has a very weak output buffer. Most instruments will trigger at the midpoint of the signal (0V), so will be effected by higher jitter than I'm reporting (because of the lower slope at the midpoint).

I got a one-cycle peak-to-peak of 185ps and a standard deviation of 28ps. This is pretty much the noise floor of my oscilloscope, so the output likely has lower cycle-to-cycle jitter than this.

The oscilloscope had a large enough jitter set to have an 1 second (a few ns pk-pk) that I wouldn't trust it so much to directly measure the 1 second jitter. As a different measurement, I used a Racal 1992 counter to test the 10 MHz output. I set the "period" with a 1 second gate hake a pk-pk of about 0.4 fs. Multiplying by 10^7 yields a pk-pk jitter of 4ns pk-pk over one second. I have reasonably good confidence in this measurement since when using the counter to measure a separate OCXO 10 MHz reference, the last digit was holding steady.

These do not directly translate to the output waveform's jitter since the generator's clock is clocked by a PLL, which could remove some of the high-frequency jitter.

A more accurate measurement would be to use the oscilloscope to measure the phase shift between the SDG2042X and a reference clock (running at some even multiple of 100ns). The oscilloscope can sample this every second, so you can get precise measurement of the long-term jitter. The oscilloscope's resolution is <100ps, so the limiting factor is the reference clock. Note that two SDG2042X could be compared against eachother, which removes the influence of a reference clock. This sort of measurement is what a time-interval-counter does (I've been thinking about buying a TAPR TICC board (or making my own design) to do measurements like this). My counter has a single-shot resolution of only 1 ns whereas the TARP TICC measured with about 30 ps resolution. The data can then be used to calculate Allen deviation plots.

[john95]

thanks for detailed explanation.