Over/Undervoltage protection for ADC

[pmbrunelle]

I have an idea for a microcontroller ADC input protection circuit. Is my idea sound, or is it backwards? The output of an instrumentation amplifier (Analog Devices AD620 or Linear Technology LT1167) must be kept within the safe voltage range of a microcontroller's ADC.

The normal way to do this seems to be with a pair of clamping diodes on each input pin. However, I am concerned that the short-circuit current during an overvoltage condition would force the +5 V rail voltage to rise, and fry any logic connected to that rail.

Therefore, my goal is to sink the short-circuit current directly to ground...

R1 & D2 maintain Vb of Q1 at roughly Vcc - 0.4 V. When the amplifier voltage goes too high, the base-emitter junction of Q1 is forward biased, and the transistor starts sinking current to ground. If D2 were a silicon diode with Vf 0.6 - 0.7 V, Q1 might start conducting prematurely. D1 prevents excess negative voltage from appearing at the ADC input. Excessive negative amplifier voltage would also cause reverse breakdown of Q1's base-emitter junction.

***Component choice/reasoning****

R1:     330 Ohm
Determines quiescient current though D2. Quiescent current should be sufficiently high that Ib does not significantly affect Vb.

D1:     SB520
Must have low Vf.

D2:     BAT46
Small signal Schottky diode. Vf ~ 400 mV @ If = 10 mA

Q1:     KSA708Y
Base-emitter reverse breakdown voltage rating of 8 V, more than the typical 5 V which would be too close for comfort. Power dissipation in the TO-92 package is sufficient. hFE of 120 minimum @ Ic = 50 mA is adequate.

Good? Bad? Likely to work?

[Jon86]

Why would more current cause the rail to rise? That doesn't really make sense...
Also, the circuit should in theory work, but why bother with all the transistor nonsense when you can just add a current limiting resistor in series with the output of the opamp. Oh, and just use a few diodes in series instead of a zener for better leakage specs.

[pmbrunelle]

The rail voltage would not rise with a single zener connected across the adc input.

However, finding tight tolerance (1%) zeners is a pain. As for regular diodes in series, the uncertainty of their Vf (and temp. coefifcients) all add up...

This is the "standard method" (with or without the resistor) I was referring to:

As for series resistance, leakage current and accuracy become an issue  so I'm trying to avoid adding a series resistor.

[codeboy2k]

You're idea is good, but probably not needed.

Go with the standard method as in your imgur drawing. 

Make sure your 5V rail has a 5.1V zener on it.
Make sure your voltage regulator on the 5V rail has a reverse biased diode from output to input.

Also, if you are concerned that the "standard method" will cause the 5V rail to bounce up when the voltage exceeds the top diode Vmax+0.4, why aren't you also concerned that the "standard method" will cause the ground rail to bounce down when the voltage goes below the bottom diode Vmin-0.4V

i.e. ask yourself this: why are you satisfied with a diode D1 as the protection device for the negative excursions but you want active clamping for the positive excursions only?   If you think the standard method top diode will pull the +5V rail above 5V, then why don't you also think the standard method bottom diode will similarily pull the ground rail down below ground??

i.e. why don't you also want active clamping on the negative excursions below ground?

see my attached drawing

I've given you a thought exercise here.

It also helps to think about this: how much current can the the INAMP supply above 5V or below ground?? in most cases, the usual standard clamping diodes do not cause the rails to rise or fall because the energy per clamping event x number of events per second = watts  is less than the total power load of the rail, thus the loads will consume it and not allow the rail to rise (or ground to fall).  Now, IF, in your case, the power load of the rail is much less than the power delivered by the INAMP, then you might have a problem that needs fixing.



^^^ I just drew this to get you to think about it, you likely don't need active clamping like this unless your load on the 5V rail is really, really low and /or the INAMP can supply excessive amounts of current.  Think about the energy in W.s x seconds per event = watts .... and where it can go during an over-voltage event.. if the power exceeds the load power (consumed) it will raise the rail.. if not, it will not raise the rail, but will be clamped at the input to a diode drop below the rail. 

[Baliszoft]

Why dont you protect the input of the opamp instead? Sometimes it is just as easy as a high value current limiting resistor.

[codeboy2k]

Why dont you protect the input of the opamp instead? Sometimes it is just as easy as a high value current limiting resistor.

I think you meant the input of the ADC ?? OP wants to protect the input of the ADC from over-voltage excursions originating at the InAmp.

OP is feeding an ADC from the output of an InAmp; the ADC likely has a S&H cap at the front end that will draw a 100uA or more depending on the size of the S&H capacitor, which could be 20pF up to 40pF.  Thus any resistance higher than 1k will drop 100mV and the ADC value will be wrong, introducing errors.  He may not even be able to tolerate 1mV error, meaning not even 1 ohm of resistance (assuming the ADC draws 100uA during a sample time).

But whether or not he needs active clamping remains to be seen.

[tszaboo]

OP has a microcontroller ADC, 5V. That means that the resolution is probably 10-12 bit. Only few TI and AD MCU has more bit resolution. On 5V LSB is about 1.2mV. So that much voltage drop is probably OK. I would put a 470 Ohm - 1KOhm resistor in series, and then put a sot23 double diode to clamp it to rails. Maybe even the MCU built in diodes can handle it. I mean, in normal operation it should not be like that, only on startup/shutdown for brief time, and fault conditions.

[pmbrunelle]

It also helps to think about this: how much current can the the INAMP supply above 5V or below ground?? in most cases, the usual standard clamping diodes do not cause the rails to rise or fall because the energy per clamping event x number of events per second = watts  is less than the total power load of the rail, thus the loads will consume it and not allow the rail to rise (or ground to fall).  Now, IF, in your case, the power load of the rail is much less than the power delivered by the INAMP, then you might have a problem that needs fixing.

This is what I am concerned about.

Each InAmp can have a short-circuit current of 50 mA each. Since I will have two channels like this in my gadget, the 5 V rail needs to be able to sink 100 mA. I don't think I will have that much load.

It's good that you mentioned the time aspect. It made me think, since the short-circuit current decays with time, perhaps the rail capacitance can absorb the initial current spike, but upon checking the datasheet, the current doesn't decay quickly enough for that.

Whatever the clamping solution is, it should be adequate for continuous duty. The two InAmps maye have their outputs pegged high indefinately while I adjust/calibrate the thing, and the various trimpots are improperly adjusted. It's not to clip the peaks of a waveform half the time.

I would put a 470 Ohm - 1KOhm resistor in series, and then put a sot23 double diode to clamp it to rails.

I probably could tolerate 1K of series resistance, but definately not with the leakage current of Schottky diodes. If you're talking about silicon diodes, is a silicon diode drop of overvoltage okay?

If a silicon drop is okay, then I can tie the base of the transistor directly to the 5 V rail, and then the clamping circuit will consist of one transistor and one diode (per channel).

Since the InAmp is current-limited anyway, adding series resistance doesn't eliminate the clamping problem; it only reduces the magnitude of the clamping problem.

[tszaboo]

If you're talking about silicon diodes, is a silicon diode drop of overvoltage okay?
 it only reduces the magnitude of the clamping problem.

I don't know, you did not disclose the MCU type, nor the power supply voltage your INA is powered. A BAT54 has about 0.5uA leakage at 25 degrees and 5 volt, that would give you 0.5mV error on a 1K resistor.

[matkar]

I don't know enough about your application so this might not be suitable. Perhaps changing the supply for the OpAmp would be an option. Pick a R2R OpAmp and supply it with 0-5V. Of course it all depends on the stuff before the OpAmp. And the fact you actually need to reach full 0 and 5V on the ADC channel. What is your ADC resolution?

[pmbrunelle]

I don't know, you did not disclose the MCU type, nor the power supply voltage your INA is powered. A BAT54 has about 0.5uA leakage at 25 degrees and 5 volt, that would give you 0.5mV error on a 1K resistor.

The InAmp has bipolar +12 V/-12 V supplies, so the current limiting resistor may have to drop some 6 V or so.

I haven't disclosed the MCU type, because I don't know what it is  This is a team project at school, and the MCU guy hasn't chosen one yet... But I've seen one MCU (MC9S12) with 2.5 uA max of adc input current.

I don't know enough about your application so this might not be suitable. Perhaps changing the supply for the OpAmp would be an option. Pick a R2R OpAmp and supply it with 0-5V. Of course it all depends on the stuff before the OpAmp. And the fact you actually need to reach full 0 and 5V on the ADC channel. What is your ADC resolution?

The InAmp reads a wheatstone bridge force sensor (excited by the same 5 V). The common-mode voltage is 2.5 V. Looking at something like an AD623 R2R InAmp (looking at DIP stuff, nothing small), getting most of the 0-5V scale with that common-mode voltage is likely not to work.

Anyway, I think trying to work with R2R amplifiers (and guarantee linearity near the extremes) is a bigger headache than trying to clamp the output.

ADC resolution unknown, but Verror < half a 12-bit LSB would be nice.

It seems like I need to keep thinking about this for a while...