I have an AC input as follows:

  1. Can range from ±10V to at least ±500V continuously.
  2. Runs from roughly 1 Hz to 1 kHz.
  3. Needs > 100 kΩ of impedance on it, otherwise its amplitude changes.
  4. May occasionally be disconnected and subject the system to ESD events.

When the input is below 20V, I need to digitize the waveform with an ADC. When it is above 20V, I can ignore it as out of range, but my system needs to not be damaged.

Since my ADC needs a relatively stiff signal, I wanted to buffer the input for further stages (in those, I will bias it, clamp it to 0V to 5V, and feed it to an ADC).

I designed the following circuit for my initial input stage to get a safe, strong output that I can feed to further stages:


simulate this circuit – Schematic created using CircuitLab

My goals are:

  1. Ensure > 100 kΩ of impedance on the source.
  2. Change a ±20V input to roughly a ±1.66V output.
  3. Provide a stiff output.
  4. Safely handle continuous high-voltage inputs (at least ±500V).
  5. Handle ESD events without dumping much current/voltage onto the ±7.5V rails.

Here is my rationale for my circuit design:

  1. R1 and R2 form a voltage divider, reducing the voltage by 12X.
  2. The TVS diode reacts quickly to protect against ESD events on the input, dumping them to my strong ground, without dumping anything onto my (weak) ±7.5V rails.
  3. The TVS diode also handles extreme overvoltage (sustained ±500V) by shunting to ground. It is past R1 to limit current in these cases.
  4. D1 and D2 clamp the divided voltage to ±8.5V so I don't need a high-voltage capacitor for C1; being after R1, the current through them is also limited.
  5. C1 decouples the input signal. It will be a bipolar electrolytic. It needs to have a relatively large capacitance to allow the 1 Hz signals to pass unaffected: $$\frac{1}{2 \pi R_2 C_1} \ll 1 \text{ Hz}$$ $$C_1 \gg \frac{1}{2 \pi \times 1 \text{ Hz}\times220 \text{ k}\Omega} = 8 \mu\text{F}$$
  6. R3 and C2, with R3=R1, compensate for input current bias and offset in the op-amp (rather than just shorting the output to the negative input); also form a low-pass filter: $$f_c= \frac{1}{2 \pi R_3 C_2} = 36 \text{ kHz}$$

Is this circuit optimal for my goals? Can I expect any problems with it? Are there any improvements that I should make, or is there a better way to accomplish my goals?


  1. I'd originally said this needed to handle ±200V continuously, but I think ±500V is a safer target.

  2. In order for the TVS diode to work as is, R1 needs to be split into two resistors, here R1a and R1b, as suggested by @jp314:


simulate this circuit


Here is a revised circuit that incorporates the suggestions received so far:

  1. Zeners across the power supply (@Autistic).
  2. Resistors leading into them (@Spehro Pefhany).
  3. Fast BAV199 diodes (@Master; a lower-leakage alternative to the BAV99 that @Spehro Pefhany suggested, albeit with a maximum capacitance of around 2 pF rather than 1.15 pF).
  4. TVS diode out front and upgraded to 500 V (@Master), so it handles only ESD events, protecting R1.
  5. Dead short from op-amp output to negative input (@Spehro Pefhany and @Master).
  6. Decreased C1 to 10μF (@Spehro Pefhany); this introduces a 0.3% voltage drop at 1 Hz which isn't as good as original the 220μF cap, but will make sourcing the capacitor easier.
  7. Added 1 kΩ resistor R6 to limit the current into OA1 (@Autistic and @Master).


simulate this circuit

  • 2
    \$\begingroup\$ Your clamp is not too bad .Place resistor say 10K in series with the pos opamp input and you have something that wont blow the chip.The TVS is cosmetic in its present position. \$\endgroup\$ – Autistic Apr 21 '16 at 2:39
  • \$\begingroup\$ What makes the TVS cosmetic there? I didn't mention it in my rationale, but I was also considering something such as a sustained ±400V input. That's out of spec, but if that happens I don't want to tax my ±7.5V rails, which are from a tiny little supply. (Don't want to damage that, either.) \$\endgroup\$ – JohnSpeeks Apr 21 '16 at 2:47
  • \$\begingroup\$ Put 8v2 zeners on your tiny supply and lose the TVS and never worry about leakage mucking up accuracy again. \$\endgroup\$ – Autistic Apr 21 '16 at 3:02
  • \$\begingroup\$ Shunting over-voltage into the power supply is a terrible idea. Shunt it to ground, and ditto for under-voltage. You could consider a gas discharge device. \$\endgroup\$ – user207421 Apr 23 '16 at 1:40
  • 1
    \$\begingroup\$ @EJP - I believe the shunting issue has been solved in the current version of the circuit (shown at the end of the question). There are pre-biased Zener diodes which are used to shunt both overvoltage and undervoltage to ground. The TVS diode can of course clamp significantly faster than a GDT, and as the primary source of voltages ≫ 500V will be ESD, it seemed a better choice. \$\endgroup\$ – JohnSpeeks Apr 23 '16 at 3:33

Your D1 & D2 will take the input surges, not the TVS -- split the 220k to 200k + 20k, and put the 20k portion between the TVS and the diodes.

Or just use a 4.7 V zener from that node to GND.

  • \$\begingroup\$ I like the idea of splitting the 220K. That makes sense to me. How would the Zener diode work? Wouldn't that asymmetrically affect the AC input? \$\endgroup\$ – JohnSpeeks Apr 21 '16 at 3:16
  • 2
    \$\begingroup\$ One zener would asymmetrically affect things -- you could use 2 zeners in back-back in series, that may be better than the diodes you have if you had a need to limit the opamp's input to less than the supply. \$\endgroup\$ – jp314 Apr 21 '16 at 4:23

You don't need R3/C2. The non-inverting op-amp input 'sees' R2 (20K) on the bias current DC path (not 220K), so the offset will likely be negligible if you replace it with a short. If you insist on R3/C2, see below for the calculation.

The 220K represents capacitive reactance of 0.7uF at 1Hz, so I think a small and inexpensive (and non-leaky) 10uF ceramic capacitor will be just fine, adding, in quadrature, about 7%, so a total effect of less than 0.3%. However there may be some effects because of the clamping, so best to investigate this depending on how exactly you expect it to behave. When clamping it 'sees' the 20k in series with the low impedance clamp, so the time constant is 11x shorter.

R1 is critical for reliability- virtually all the voltage gets dropped across it- it must be a high voltage type, rated to withstand whatever transients you expect, especially if this input voltage is coming from the mains that may mean a couple kV. Vishay VR25 may be suitable (leaded). Don't skimp here. Unless the last few pennies are more important than reliability, I'm not a big fan of using multiple ordinary resistors for this purpose either- one properly rated part should be okay unless you need to use two properly rated resistors in series for even more reliability.

I would lose the TVS and consider clamping either directly with a shunt (such as a zener pair) or low-capacitance switching diodes like a BAV99 pair to pre-biased shunts, such as Zeners or TL431s (with resistors to the supply rails). The latter will have much less capacitance than using zeners directly and will thus cause less phase shift at 1kHz, if that is important to you. The clamping current is less than 1mA at 200V in, so it's not very taxing, so long as R1 holds up against whatever EMF it is subject to. Both the options I suggested can easily clamp 100mA, at least for a brief time.

R3/C2 do not really form a low pass filter- R3 and the input capacitance of the op-amp form a low pass filter, and C2 would ideally be chosen to be much larger, so if the input capacitance is 15pF you might use 1nF or something like that. You would only run into trouble with 20K alone if you had a wildly inappropriate op-amp (capable of very high frequencies) where the resulting phase shift affected the stability, and of course a short does not have that problem.

  • \$\begingroup\$ The two "R2/C2"s in the first paragraph were both supposed to be "R3/C2," right? \$\endgroup\$ – JohnSpeeks Apr 21 '16 at 4:19
  • \$\begingroup\$ @JohnSpeeks Yes, thanks, changed. Need a bigger monitor (or better memory) I guess. \$\endgroup\$ – Spehro Pefhany Apr 21 '16 at 4:21
  • \$\begingroup\$ Would it change your opinion on the TVS diode if it were likely that there may be long periods (30 seconds or more) of ±300 or ±600 volts? I don't know exactly how high it goes continuously, as one instance was measured in the field with an oscilloscope which clipped the signal to ±150V, and extrapolating the waveform I guessed around ±200V, but it's also possible it could go higher. I perhaps should edit the question to give a higher value there. \$\endgroup\$ – JohnSpeeks Apr 21 '16 at 4:28
  • 2
    \$\begingroup\$ @JohnSpeeks 600VDC would cause 1.6W of dissipation in the 220K resistor so it better be rated for a couple watts, but the zeners or shunt regulators I mentioned could easily handle 2.7mA continuously- that's only 20mW @ 7.5V. Two VR68 1W resistors in series could handle a 20kV transient and 100mA is not too hard to clamp. TVS diodes are good for when you have a low impedance and have to absorb a large spike of energy in the hundreds of watts- they're not especially great at dissipating continuous power. In this case, you don't open the door to the spike so it doesn't have to be absorbed. \$\endgroup\$ – Spehro Pefhany Apr 21 '16 at 4:38
  • \$\begingroup\$ @Sphero Pefhany I have noticed that TVS diode datasheets rarely give any specs for continuous operation... Your point about the dissipation across R1 is well taken, as are your suggestions for resistors. In theory I could increase the value of R1 (and R2) to reduce dissipation across R1 (still using something like VR25/VR68 resistors), but I'd be concerned that might introduce new problems. \$\endgroup\$ – JohnSpeeks Apr 21 '16 at 5:23


simulate this circuit – Schematic created using CircuitLab

The P/N of OP AMP and diodes on schematics mean nothing. Diodes D3 D4 are either one BAV199 or 2 Gate to Channel junctions of jFET MMBF4117. OA1 is OPA365. C3 must be selected to provide sufficiently low pass frequency for filter on C3, R1/2.

R2 and R3 are preferably precise thin film resistors or even two parts of one resistor network. They define your zero drift.

R5 must be rated for 1 kV voltage, you can use several 0603 resistors in series.

And, to be really safe, you can add some 1 kOhm resistor between non-inverting input of OPA365 and mid-point of R1 R2. It helps limit the input current if something goes really bad.

The high power voltage limitor (like TVS diode or varistor) is preferably connected between INPUT and GND. Its voltage is about 600-800 V.

  • \$\begingroup\$ I am going to have to order some of those parts before I can prototype this and compare to the other options. Stay tuned! \$\endgroup\$ – JohnSpeeks Apr 21 '16 at 22:00
  • \$\begingroup\$ Unfortunately the RC part of that (ignoring the diodes and the op amp) rolls off the input by around -1.44dB at 1 Hz (cutting the output by about 15%): Frequency response curve. Increasing the cap to 10 uF fixes that and keeps things pretty flat to 1 Hz, but then it takes something like 30 seconds to charge the cap through the 470k resistors. (And of course decreasing those doesn't work, as it rolls off the low-frequency response again.) \$\endgroup\$ – JohnSpeeks Apr 22 '16 at 20:42
  • 1
    \$\begingroup\$ Sorry for the late reply. Yes. it is true, of course. But you get this problem with any design of the low pass filter. Why do you need C3? May be DC coupling is better? \$\endgroup\$ – Master Apr 26 '16 at 12:12
  • \$\begingroup\$ That's a very good point. I could make this DC coupled. In my particular application, there is no possibility of DC offsets, and I also don't care if the output signal is inverted. So I could use an op amp in an inverting configuration to add the offset voltage. \$\endgroup\$ – JohnSpeeks Apr 27 '16 at 13:51
  • 1
    \$\begingroup\$ Ok, good to know! Your questions are welcome! \$\endgroup\$ – Master Apr 27 '16 at 14:43

What kind of OPA do you use? If it is FET input OP AMP (input currents below 100 pA) then you do not need R3 C2. Also, if you do not care about DC offset, it is much better to remove R3 C2.

I see no value in TVS diode 30 V. Absolutely agree with @Autistic. You can put it straight in parallel to the input (before R1) and change to 500-700 V type. Its function then is: to protect R1 and other electronics from really short spikes over 800 V (I do not know if your application can get into this kind of trouble).

R1 must be either rated for 1000 V or implemented as a series of 0603 or larger resistors, taking isolation gaps into account.

As for "real" clamp: the idea of @Spehro Pefhany of pre-biased BAV199 (two low leakage diodes in one SOT package) looks the best. I would not care too mush about currents to power rails: they are limited by 4 mA (800 V / 200 kOhms), it is probably less than power supply current of one OP AMP you use.

Why not to put R2 (I believe it is a voltage dividor) before C1 and to use very large resistor (1 MOhm) on place of R2 - this allows C1 to be as small as few uF.

  • 1
    \$\begingroup\$ You have to keep in mind that the input bias current of this OPA is as large as 1-4 nA at 70 C. It means (for your design) that the additional offset voltage may be up to 200 uV, it is much higher than its "nominal" offset voltage. This is a common problem of jFET OP AMPs, they are not suitable for high impedance inputs at slightly high temperatures. \$\endgroup\$ – Master Apr 21 '16 at 9:16
  • 1
    \$\begingroup\$ The modern BJT OP AMPs (AD8675) have much smaller variation of their bias current vs temperature, although their input currents are also large (1 nA). \$\endgroup\$ – Master Apr 21 '16 at 9:18
  • 1
    \$\begingroup\$ What range of output voltages do you need? \$\endgroup\$ – Master Apr 21 '16 at 9:20
  • 1
    \$\begingroup\$ Why not to use Rail-to-Rail 5 V OPA? It clams naturally to 0-5 V for ADC. They are much better for input performance than "high" voltage OPAs. \$\endgroup\$ – Master Apr 21 '16 at 9:22
  • 1
    \$\begingroup\$ Sorry, "clamps naturally" \$\endgroup\$ – Master Apr 21 '16 at 9:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.