I finally got round to testing a BNC-equipped unit received from Leo on 1/23/18.
By request, this unit was set up for 1.0Vpp output. Leo noted on its included docs that rise time at this high output amplitude is degraded to ~60 ps. However I also noticed from the included test plot that, as an apparent side effect, aberrations in the first ns appear to be well under 2% p-p with this setup. Leo's CSA803 showed very nearly a perfect step.
Many appear obsessed only with rise time; in my experience, a clean top without overshoot is critical for the risetime to be quantitatively meaningful. A dead flat top is an extremely rare and valuable property in a sub-100ps signal source.
I checked it for myself with a Tek S6 sampling head in a 7S11/7T11/7904 rig, approximately confirming both the reported rise time and measured flatness. I count about 65 ps intrinsic after RSS regressing the nominal S6 response, and about 3% p-p aberration in the first 250 ps. (I suspect my SMA terminator of some mischief at about 300ps; the S6 loops the signal back out to an external termination, to allow for TDR applications.)
First 3 attachments show the rise at 50, 200 and 1000 ps/div. Fourth photo shows the fall, significantly faster at about 40 ps (after RSS regression), but including about 7% p-p aberration.
For kicks, the fifth shot is an S-52 tunnel diode (25 ps), and sixth is a home-brew pulser using a Russian GI308D tunnel diode (about 75 ps).
Bottom line: With 1Vp-p amplitude, this unit gives ~60 ps rise with damned near perfect shape, and ~40 ps fall with a little aberration. Win-win. Many thanks, Leo!
-Mike