What can be said about the nature of the noise by looking at this spectrum?

Question

I use some industrial amplifier modules where each of them amplifies force transducers. All amplifiers are powered by a single power supply and the output. Amplifier is set to 6kHz filter. And there is no anti-aliasing filter in the system.

Interest of frequency is max 10Hz and the sampling rate is 500Hz. Sometimes when the noise noise appears at the sampled outputs it has a peak around 3.1Hz. So this corrupts the readings below 10Hz.

I took some offset readings to observe the electrical noise better.

I have the following 12 seconds of noisy signal in time series:

And here 1 second view:

And below is the FFT of the signal:

(left-click to zoom in)

And finally comparison when that 3.1Hz noise, green when it is there and the blue plot when it is not there:

I still could not find the source of this noise.

But just by looking at these plots, what can that 3.1Hz noise be? Can that be aliasing? If so, is there a way to estimate the real noise freq. by looking at the peaks at FFT given that we only have 500Hz sampling rate?

Answer 1

Looks like you're in 50Hz-land, observing that there are about 10 cycles in 0.2 seconds in the second graph.

If whatever 50Hz signal you have is rich in harmonic content, the tenth harmonic will be pretty close to your sampling frequency. In fact I'll speculate it's 3.1Hz away (and will move around as the mains frequency drifts under load).

There's a reason sample data systems always need an anti-aliasing filter.

If you're not interested above 10Hz, you can use it to attenuate 50Hz too.

Answer 2

It could be something higher being aliased, but it is in the nature of aliasing that you cannot definitively go from the aliased signal back to the original unless you have knowledge of which nyquest zone the original signal fitted into, and I am guessing that in your case the original did NOT fit in a single nyquest zone.

I would start by hooking up a PC sound card as a digitiser and recording the signal at 48/96 or 192KHz, this will give you a few tens of kHz of bandwidth with correct anti aliasing filters, so you should be able to see what is really going on.

This file (it will be more then a little large) can then be used both for analysis and as a reproducible source of test signals, so that you can actually play the interference out into your measurement system on demand.

One other thing with weirdly intermittent noise sources, look for radios in the area (cell phones, walkie talkies, that kind of thing), a maintenance guy with a handheld radio leaning on a badly filtered control cabinet while he calls the control room can cause chaos that is HARD to track down.

One final thought, can you try a divide a conquer approach (Replace the transducer amplifier on one channel with a resistor to see if the problem goes away, then do the same for the transducer, both at the amp and at the sensor, keep halving the problem until you find where the issue is getting in).