Limited in what sense? Sure, there are challenges associated with it, like mitigating noise coupling from the switcher. But, limited suggests to me that it fails to handle certain input conditions, which I don't think it does. Please clarify. Still, I'm more than happy to explore alternatives especially if they're simpler than this solution, which is getting quite complicated.
My first objection is more semantic than practical: it's suggestive of "if-then" procedural thinking, which doesn't map well onto analog circuits. Embrace the continuum! If voltage goes outside of bounds, don't switch it off (what if switching it off causes voltage to rise further? or reverse so it oscillates?), just increase current, diverting it away from your load.
The practical consequences come when considering marginal cases. What if the input is extremely noisy (some volts of AC)? Well, on a practical note, your amplifier is useless (driven to clipping), so maybe it doesn't matter. But randomly activating (and at a high rate) that switch logic won't do any favors for signal distortion, or time dependency of the response.
A simple clipping filter has the benefit that it's not time dependent. It's there only when needed and goes away seamlessly. Distortion increases in an orderly fashion as signal level approaches the threshold. The onset of clipping is easily recognized in the response.
The linearized version might be interesting to ponder, but isn't going to be easy to pull off (you need to synchronize the transfer curves of two switches, potentially while level shifting). That is, suppose the switches were adjustable resistors instead; or some kind of V(I) characteristic curve anyway, since we're using transistors here.
Notice also you're making a negative resistance input characteristic: as voltage rises, current falls [suddenly]. This might not be well-behaved say with a reactive source. Or maybe it's balanced by the shunt switch, but again, it needs to track properly. And it needs to have the same response rate (control theory is involved too!), lest it have an impedance peak that causes ringing or oscillation (between one or the other "switch" and the source, or any other combination thereof).
Maybe those are equally good reasons to reject the continuum, and stick to a logical method instead. I don't know.
10R adds more thermal noise than I'd like. More importantly, though, this exposes the DUT to a prolonged short
Okay but you're already hot-plugging the thing, what did you expect was going to......nevermind.
(Can't you just plug the thing in,
then turn it on? Normal startup transient / soft-start cycle applies?)
and provides significantly less protection to the JFET input, which will now be exposed to elevated voltages and lots of current input for 100s of us. I would expect a JFET gate to be able to handle elevated currents for 10s of ns, but 100s of us seems to be pushing it (again, I will damage test the JFETs to actually make sure they can handle the 10s of ns).
Also can always add source degeneration (reduces gain without increasing noise), or bypass the gate resistor with a capacitor (another old school oscilloscope protection feature, does have a long recovery tail though). (Again, could use depletion MOS for faster recovery while limiting current, I suppose.)
First addressing zeners and clamps: the JFET gate-source is just a pn junction. This doesn't leave a lot of room for devices to kick in before the JFET gate conducts appreciably. Zeners have soft knees, so anything that steals enough current away from the JFET will likely conduct too much at levels where it shouldn't. This greatly lowers the input impedance of the JFET amp and imposes unrealistic demands on the capacitance of the input blocking capacitor. Clamp diodes have similar issues. They provide some protection, but they're not adequate on their own - they only steal some of the current away from the JFET and so the input still needs to be shutoff. Schottky's do a much better job of stealing current from the JFET gate but lower the input impedance too much.
Well, what's it going to be biased at? Surely not Idss (Vgs=0) giving minimal dynamic range. You're already planning on basically a back-to-back diode (SP00R6) (actually I wonder if it's anti-series back diodes* or something, given they show a series diagram internally?) so biasing down a volt or two should be trivial.
*A variant of tunnel diodes, with especially low reverse voltage. Oh but wait, normal forward voltage. Or hm, could tunnel diodes be fabricated with just enough negative resistance to compensate for the forward junction? No that wouldn't make sense, there'd be a shelf in the response. Hmm, maybe they are just punch-through diodes, just tuned for a very low voltage? Dunno. Anyway, laughable that they show anti-series zeners as the diagram, they are absolutely anything but.
Also mind, the gate current flows into the channel, which has significant resistance. It's no rectifier. Actually it's notoriously poor: consider the pico-ampere diodes, which as I understand it are basically diode-strapped JFETs ala 2N4117. (Oh, in fact that's exactly their description(!):
https://www.interfet.com/pad/ ) Your parallel array will be "stiffer" of course, but Rds(on) will be no less a factor in its response.
So I don't see a need to be picky about a volt here or there.
As for the depletion MOS, I'm still exploring this, but I think the fundamental difficulty is that the steady state gate current needs to be kept quite low. I've heard 10mA per JFET thrown around. I don't know if that's accurate - I still need to test this, but let's go with that for now. I'll be paralleling these JFETs, so assuming the current is equally distributed between them that eases things a bit. Let's say 16 JFETs. Total steady state current then needs to be limited to 160mA. The boundary between conduction and non-conduction for depletion MOSFETs isn't very sharp. So, to adequately limit current will require enough series resistance between the MOSFET channel and any external resistors. I've looked at the datasheets for a few depletion MOSFETs and the total series resistance for those would be too high.
I would expect ratings to be driven by thermal and electromigration limits, or something to that effect. 10mA is the DC figure. I would expect pulsed currents of, oh I don't know, 10x for 10s of µs at least, being acceptable. Compare LED pulse ratings, those can have quite small junctions for example. I mean, those are in entirely different material too, so take this for the salt-grained hand-wave it is -- but perhaps your testing will find this out.
Mind to do a good number of cycles, like, thousands, or millions even (should be easy enough to set it going in the background and accumulate that kind of time?), on a reasonable range (preferably a dozen parts or more, from various production dates and manufacturers if possible?), and check the small-signal parameters afterwards. Then re-run at more aggressive settings (higher Ipk, pulse width) until failures are observed.
I suppose noise would be the most critical and least predictable variable here? I don't have a feel for what it should do. Want to say it depends on surface states and spooky stuff like that, which shouldn't be affected by forward bias? Leakage feels more likely to be affected, to me. Or Vbr, or gm or Vgs(off), those are pretty sensitive and something like spot heating or ionic motion (pushed by high current density?) might do it.
Or just outright failure first (short/open), which seems most likely a fatigue or electromigration sort of mode I think.
Will be interesting to know, independently; post that in a new thread even, if you could.
I'm intrigued by this idea, and hadn't thought of it. There are challenges with it, though, relating to the fact that MOSFETs don't have a sharp boundary between conduction and non-conduction.
Well so what... think of it this way: you're going from ~100V to 2-4V. That's a hell of an improvement, eh?
Maybe a multi-stage process is needed; but you're already set on that [series and shunt switches] so it's more a matter of how than what
- Classic, biased diode-bridge current limiter; bias can be from HV BJTs between supply rails.
I'm not familiar with this. Can you explain further or point me to a resource that explains it?
Sure,
Any current sink/source will do, this is just shown with R+D bias for simplicity. Can get closer to the rails with current mirrors (but watch for Early effect!!), or use even just bias resistors if you don't need the currents to cancel out over signal voltages (which is to say, loading the input) (which, really, with a sub-volt range, that should be fine regardless).
Or somewhat higher supplies and generous emitter degeneration, if you need low current noise. This is the subtraction of current sources after all, their noise will be uncorrelated.
There's an additional consideration, which is that I still need to switch in a resistor during an over or undervoltage condition. The lowpass frequency cutoff is low enough that the capacitor would take too long to charge or discharge. This resistor can't always be in because it lowers the input impedance too much. This means that I still need something to switch this in. The requirements on this aren't too tough (and a simple window comparator would do), but I guess my point is that it still needs to be here, so none of the above solutions are sufficient in and of themselves.
Okay but what do you need a high impedance for when you're measuring fractional ohm sources?.....nevermind again.
Oh, or is that so you can save on C in the first place? But that increases noise, at the low end. Because it's the -- a capacitance is lossless, but the implicit resistance that is always present alongside it, making that filter cutoff -- that has noise, and the noise is effectively filtered by it and there you go. It's identical* whether you have 1 ohm and 1mF or 1kohm and 1uF. The only difference is the transition region between those impedances and frequencies, and whatever the high-frequency characteristic should be (apparently very low impedance, to keep voltage noise down?).
*In noise power, but voltage scales with impedance of course, so it's better to use large values here.
Tim