Hello,

I am in the subject of measuring mV signals with maximum accuracy.

The formulas for calculating white noise is described in several data sheets.

But for 1/f noise I didn't see such a formula.

What's confusing to me, that at DC you get infinite nV/rtHz noise level.

Doesn't this translate to infinite RMS and peak-peak level?

In Opamp data sheet there is a graph - noise in nV/rtHz over frequency.

How can I calculate the rms (and peak-peak) noise if I would brick-wall (low pass) filter the output at 0.1Hz / 1Hz / 10Hz ?

How can I compare the "outcome" of zero-drift OPA versa "normal" OPA (noise) for the above sample rate?

PS:

A resistor doesn't have 1/f noise? So a 350 Ohm resistor has 2,4nV/rtHz.

This would mean that at DC to 0.1Hz it would have 0,8nV rms?

Even though the 1/f noise rises at lower frequencies, the bandwidth across which this noise gets integrated gets smaller and smaller. These effects tend to cancel, so you don't go to infinity.

It is a pain to integrate the noise when you are in the 1/f region. You can do it with Riemann sums with a zoomed in version of the spectral density graph from the datasheet, or by figuring out the slope of the 1/f line. This video has a tiny bit of discussion about it. For our precision amplifiers we typically include a 0.1 to 10 Hz number in the datasheet to give some idea of the integrated noise. This is a good quick way to compare amplifiers (including standard vs. autozero), and is what most people do.

In our engineering calculations we typically assume that metal thin-film resistors do not have 1/f noise. For most practical purposes this is a safe assumption. For all the gory details, this paper is a pretty good reference.

Matt