JFET series switch gate diode weirdness

[nickcordle]

Hi all, I've got a circuit fragment built on a breadboard, schematic below ... bipolar supply, +/- 9V.  I'm trying to switch a bunch of stuff based on whether a cable is plugged into a stereo jack or not, and one of the things to switch is a series JFET like so:



Anyway this is made of pretty well-worn fragments so far.  Cable goes in, collector of the 2N3906 goes to roughly 9V, JFET gate floats, channel allows signal.  Cable goes out, collector would float but is pulled down to -9V by that resistor, Rds goes high.  All good so far ...

I originally had a BAT46 for the diode thinking I wanted a low Vf, to allow for a bit more negative signal swing, but saw something I wasn't expecting:



Looks kind of ... phased ... or something, just on the top side?  That shot was at 1kHz.  It gets worse with higher frequencies.  So I tried a few different diodes: 1N914, 1N4148, 1N4001, all -worse- than the BAT46, until finally 1N5817 ... this is more what I expected to see.  At 1kHz the JFET will pass pretty much anything until the negative side hits Vgs(off) and the rails of the op amp come into play.  Nice!



But I think with the 1N5817 the strange effect is still present, because if I go up to 10 kHz, there it is again:



What am I seeing here?  This is all with the plug -in-, so the diode is reverse biased the whole time, cathode around 9V.  So ... frequency-dependent reverse leakage currents messing with the JFET gate?  Some implicit capacitance I'm not realizing?  What gives?

[Rob Strand]

What am I seeing here?

I think what is happening here is when the signal goes negative the JFET gate to channel junction acts like a diode and charges up the capacitance of the external diode.   At that point the anode of the external diode is negative.  Then when the signal goes positive enough to reverse bias the JFET junction the charged diode capacitance is left hanging there.  The gate control voltage (from the transistor ckt) is positive but the voltage across the diode capacitance is subtracting from that so the voltage at the gate is less than the control voltage.   The result is the switch isn't fully on.  The only path for the diode capacitance to discharge is via leakage which takes time.   If the time to discharge is relatively fixed that would also explain why the problem is worse at high frequencies.

So the root cause of the problem is that Schottky's have higher junction capacitance than silicon signal diodes.

It's a very interesting problem.  So thanks for posting it.

[Rob Strand]

So I tried a few different diodes: 1N914, 1N4148, 1N4001, all -worse- than the BAT46, until finally 1N5817 ... this is more what I expected to see. 

I had to do some stuff in a hurry before and I missed this part.  What i said can't explain why the 1N914 and 1N4148 are worse.  Unless the key parameter is the leakage to capacitance ratio.

For the case like yours, where the JFET connects to the opamp, it might be possible to use a linearization network to provide a path to dump the diode charge.   That's going to complicate the switch.

So the root cause of the problem is that Schottky's have higher junction capacitance than silicon signal diodes.

The signal level also comes into play.   For signals less than 1.2V peak the capacitance has virtually no effect.  At 1.5V peak the diode capacitance starts to have an impact.