I'm using an AD8421 instrumentation op-amp, and I need to adjust the DC level of the output. As per the datasheet, the VREF input pin needs to be driven by something that asserts a voltage (low output impedance), and they recommend a voltage follower amp. So, I'm doing just that:


simulate this circuit – Schematic created using CircuitLab

The 0.1uF caps were mostly out of paranoia/habit.

I'm wondering: do I really need bypass capacitors for this op-amp? The op-amp is supposed to do a good job at filtering power supply noise/ripple at low frequencies, and this op-amp will not be subject to sharp transitions or any other high frequency signals; at high frequencies, any other op-amp that would produce noise have their own bypass caps so that noise does not propagate to the rest of the circuit.

Now, speaking of power supply noise (since I'm using a DC/DC switching power supply), I then realized that the noise will directly leak to the output of the AD8421 through the input signal of the LF356. I'm thinking I should replace the simple voltage follower with a low-pass filter at, say, 5Hz or so (e.g., place 10uF instead of the 0.1uF caps in the input circuit). However, the fact that the 8421 datasheet does not mention anything like this makes me wonder whether I'm being too paranoid? Any comments?


2 Answers 2


I'm wondering: do I really need bypass capacitors for this op-amp?

Yes, you need capacitors to filter the voltage at the output of the potential divider because you don't want to transfer power supply noise through to the reference pin of the AD8421. You will probably only need one capacitor from the non-inverting input to ground; i.e. you don't need two as you have indictated.

(Alternative) Yes, you need op-amp power rail decouplers.

  • \$\begingroup\$ with just one capacitor as you suggest, the cutoff frequency changes as I move the pot to adjust the offset. Maybe not a big deal --- roughly speaking, in the range 1/2*piC*(15k/2) to 1/2*piC*(5k||25k), which should be fine. \$\endgroup\$
    – Cal-linux
    Dec 19, 2016 at 21:57

When you consider the unity gain bandwidth of the LF356. Then combine the 10 Tera-ohm equivalent input resistance of the LF356 op-amp. You have everything you need to make a nice wide band noise buffer.

Rule one is decoupling noise in a high impedance network is much easy than in low impedance. The exception is a feedback network to help dampen noise problems. Note the phrase "dampen" or time constant.

So first off keep both 0.1uF caps on both input sides of the 20K pot. Also be sure both caps are NPO grade 50VDC for good high frequency performance as well as low DC leakage. As all power supply decoupling caps in the system should be.

And for the last thought. Is the 20K pot 10 turn or 20? Now you have a RCL filter to model. The trim pot is a small inductor.

If your system has any digital clock signals. And/or RF any mixing frequencies. Be sure your random choices for resistor and capacitors values do not become a tuned resonator circuit!


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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