I'm doing some sensitive signal measurements in 1~10kHz range where I amplify and filter (multiple active stages) a small voltage signal by hundreds ~ thousands of times and then sample it. A negative, clean power rail will occupy a lot of board space and is a bit harder to implement, so decided to create a virtual ground instead.

I figured that the noise of the virtual ground will be amplified together with the signal, so it's crucial to make the virtual ground noise-free. Now I can either use resistor to split my power rail and buffer it with a low noise op-amp to create the virtual ground, or I can buy a voltage reference, but cannot decide which is better. I found that voltage references usually have noise in the 1uV/rtHz range around 1kHz. However, low noise LDOs can go even lower, and because I'm dividing the LDO voltage, the output noise can be down to 100nV/rtHz range.

So my question is: what's the usual practice for this kind of applications? Should I choose a low noise voltage reference or go for the LDO + opamp solution instead?

  • \$\begingroup\$ Did you consider using differential amplifier by Op amp? \$\endgroup\$ Commented Jan 26, 2023 at 20:10
  • \$\begingroup\$ FWIW if you can fit an opamp and an LDO, you could fit a capacitive voltage inverter to make a negative rail. \$\endgroup\$ Commented Jan 26, 2023 at 20:18
  • \$\begingroup\$ @PCBCrewEngineer You mean full differential amplifiers? Those usually demands matched resistor/capacitors if you need accuracy, and it takes twice the passive components, right? \$\endgroup\$
    – Wjx
    Commented Jan 26, 2023 at 20:22
  • \$\begingroup\$ @Wix Correct. Requires precise resistor value matching. \$\endgroup\$ Commented Jan 26, 2023 at 20:28
  • \$\begingroup\$ @user1850479 True, I guess I'm just trying to avoid negative rails due to some other design constraints. If I make the signal chain dual rail, I have to change a lot of other design too. \$\endgroup\$
    – Wjx
    Commented Jan 26, 2023 at 20:31

1 Answer 1


The usual practise (if you need only 1-10kHz) is to AC couple your signal into a large value resistor-based voltager divider to bias it midway between your supply and ground.

This creates basically no noise at all (far below 1 nV/rtHz), because the capacitor ESR is very small, and resistors contribute essentially no signal current.


simulate this circuit – Schematic created using CircuitLab

  • \$\begingroup\$ In my case the input signal can be considered as a voltage source that can be arbitrarily biased, so the first AC coupled input stage is not needed. I guess in your circuit what's doing the trick is C2, but it needs to be huge, right? \$\endgroup\$
    – Wjx
    Commented Jan 26, 2023 at 20:33
  • \$\begingroup\$ @Wjx yes C2 is important. It prevents, that the opamp stage also amplifies the ground offset. You want its impedance to be well below R3 at the smallest signal frequency, so 100µF will give 1.6 Ω at 1 kHz. Good enough IMO. For this very small value of R3, one could think about a 220µF or 330µF cap instead. They will also have slightly lower ESR than 100 µF. Al electrolytics are fine here, so it won't be huge, but actually rather small and less than 1 € for sure :) R3 will be your main noise source, so if allowable, you can increase its value to relax C2 value. \$\endgroup\$
    – tobalt
    Commented Jan 26, 2023 at 20:39

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.