Earlier this year I too was looking for a way to get <30ps edges for TDR/TDT, with the additional requirement that the stimulus be differential. The solution turned out to be the Colby PG-5000A, a long-out-of-production clipping amplifier built around some Fujitsu GaAs magic. It will routinely put out sub-30ps differential stimulus at rep rates from 10MHz or so to 5GHz, and if you hold it at just the right angle you can get it below 23ps (all times 20-80). Using an ESG-AP to drive it, the jitter is silly low. Measured TIE around 420fs RMS, N-cycle around 1ps RMS, at 2+ GHz rep rates, and at 30MHz rep rates, a still wholly respectable 3ps RMS TIE. The only caveat is that it's AC-coupled, so your rep rate can't realistically go below 5MHz and your driving signal has to have DC balance (though you *can* drive it with a square wave to synergistic effect). But there are a couple bonuses: on each of the two differentially related outputs, it has a 10 position attenuator for the pulse signal and, independently, a multiturn pot for DC offset injection. You can imagine the quality of the attenuators and bias Ts, to pass signals past >12GHz with negligible attenuation and group delay. I've peeked inside it and it looks like a physics experiment, all hand-made of course.
See attached for a pretty comprehensive demo. I won't take the time to describe each image in detail. Upshot -- it or my scope or both peter out around 14Ghz, but there are strong harmonics with intact phase to 12GHz. For the demo I show square waves at 2.0GHz, 2.4GHz, and 2.8GHz, with a scattering of jitter figures, rise and fall times, FFT and odd harmonic amplitudes, and I also show a zoom in on the calculated (scope math) differential rise at 30MHz, showing a 35ps rise time.