David noted something about the AM687 comparator. I remember a story about one of AMD's fast comparators, maybe the AM686 - the one with TTL outputs. In the mid 80's I had just been hired at a startup and a contractor was battling with one of these comparators for a different project. AMD was not kidding when they talked about sensitivity to layout and choice of bypass caps. My contribution was to suggest adding a shield over the pin side of the PCB which helped. I believe the ultimate solution was to go surface mount when the part became available in that form. The DIP package leadframe just had too much mutual inductance.
Comparators faster than the LM311 often have stability problems making them difficult to use in practical circuits. One solution to this problem is to include (and use) a latch enable input on the comparator; some parts are simply never stable in their active region unlike the LT1016 and similar later parts which are much easier to use.
pretty much every circuit that has been referred to seems to use a JFET input buffer. is this just to create a high impedance input or are there any other reasons? I'm not sure about the input impedance of comparators but op amps have better than good enough input impedance on their own so I'm not sure why the JFET buffer is used.
As I mentioned, if you need a high impedance input and lower bandwidth is acceptable, then you can get away with using a JFET or CMOS input operational amplifier in a voltage follower configuration instead of a discrete JFET or MOSFET. In the past, fast JFET and CMOS input operational amplifier simply were not available.
Also I wonder if there is a better way of creating the hysteresis where it doesn't need adjusting the gain based on the input
the schematics that have been provided so far don't seem to have that but they're a bit too complex for me to see how they have gotten around the trigger point problem. maybe the hysteresis points are chosen close together at something like 50mv and 100mv? rcbuck mentioned that even the professional ones that include sensitivity adjustment don't seem to do much anyways ...
The HP 5314/15/16 simply relies on the built in hysteresis in the AM687 comparator that it uses for normal operation and adjusts it in a rather dastardly way in the alternative mode of operation and it definitely has an effect.
The first example I gave above has fixed positive feedback around the last differential amplifier stage. The second discrete example has fixed DC positive feedback around the first differential amplifier stage and AC positive feedback around the last differential amplifier stage which is only there to increase switching speed.
Usually, hysteresis is fixed in the design and there is no need to adjust it. If you use an integrated comparator like an LT1016, it has no hysteresis built in but you can add it easy enough with a resistor from one of its outputs to one of its inputs. Making the hysteresis adjustable would be tricky unless you can pull off what HP did with the HP 5314/15/16; offhand I do not know what modern parts are suitable for that kind of implementation.
But I think the above is largely irrelevant because you can get great performance up to 25 or even 50 MHz with just two parts. Pick your favorite fast JFET or CMOS operational amplifier for use as a voltage follower to buffer the high impedance inputs, maybe an AD8065, and use it to drive an LT1016 or similar with an external positive feedback resistor to add fixed hysteresis; it should only require millivolts to 10s of millivolts for good operation.
You will probably want a compensated divider before the buffer to increase the input voltage range and to protect the input from over-voltage conditions. A high impedance input also allows the use of a x10 oscilloscope probe for added protection and input voltage range.
The LT1016 or similar will be easier to use than the TI part I mentioned because it has complementary outputs; use one to drive the logic and the other to provide hysteresis. TI has some similar parts I think.
If a 50 ohm input is acceptable, then you can drive the input to the comparator directly without a high impedance buffer.