Usual way to drive JFETs is to pullup gate to source. The enhancement going to positive Vgs isn't substantial, and if you need lower Rds(on) just choose a bigger JFET.
I know, but there is no way to pullup gate to source, or maybe - can I use the output of U6 in some way? Maybe through an analog switch?
Pull to the low impedance: T1 to ground, T2 to U1b output.
I wonder what LM311 output capacitance is, or in general how its output switching depends on load current and capacitance. Its output capacitance isn't specified. D3 probably isn't useful in the end, but I wonder if a cascode would be (e.g. a 2N3904 with R + 2 x diodes voltage divider for base bias, maybe with a Baker clamp as well). In other words, keep the LM311 output voltage swing small, assuming its current swings faster. It can also be biased up to run at higher current (mind to put in a voltage limit so it doesn't pull the 3904's emitter into breakdown).
As for generally faster comparators, I don't have a good recommendation offhand; the next faster option in my box is MCP6561 which is 5V max (and CMOS output). I'm sure there is something available; I do recall there's a big gap between classic parts (~100s ns response time) and fast ones (<10ns), just don't be alarmed if it happens to be far faster than you expected.
You mean a JFET with lower Cg(on)? Suggestions?
Well, J112 and friends, kind of the next step down I guess; PN4391 similar; or the few remaining RF JFETs (be careful not to make them oscillate
) with Rds(on) around there. 2N5486 might be on the small side, BF862 (or its newer replacement CPH3910) is probably about right.
Rds(on) far below 100 ohms doesn't seem all that useful, as the TL072 isn't capable of much more current than an equivalent resistance around there. For T2 at least.
I think the required sample window (essentially 3*Rds(on)*Csample, say) versus charge injection (ΔVg * Cg(on) / Csample) is a roughly constant amount. There isn't much performance difference among JFETs, it's mainly about how wide the channel is; and a wider channel has proportionally more capacitance and less resistance. The main difference is in fine optimizations, matching channel width to circuit impedance (hence ~100s ohms or ~10s mS parts are most practical at RF), and how much extra capacitance comes from wiring/pad connects, I think.
I have considered this, but that will require higher gate voltage, ie. minimum Rds for a MOSFET requires Vgs > 4V, right?
Yes, well you have +/-9V supplies so that's no problem at all.
DG401, DG612, TC4W66FU, even the bog standard 4066 may do.
CMOS switches work over the full supply range. At low voltages (near VSS), the NMOS is doing the work, at high voltages (near VDD) the PMOS. There is an peak in the middle, in Rds(on), particularly exaggerated at low supply voltages -- IC MOS are more like Vgs(th) = 1-2V, so for supply voltages around twice that, they stop working (Rds(on) just too high at middle voltages). At 18V supplies, Rds(on) will be pretty stable, and as low as it gets for most parts.
Tim