DSP Robotics Support

This is new topic for me (5 minutes old?), and I see not too much in my language. Briefly, it's a method of exposing/measuring very weak signals from noise floor.

Now my question. (Martin? KG?). Is this done rather by specific hardware, or can it be done by specific signal processing within a software? And if in software... could someone give an example schematic to start with?

It can be done in software. Do you have a specific use case in mind?

Electromagnetic resonances of matter (from what I read - it covers nuclei spin resonances and soft inter-molecular bonds and some other stuff) within acoustic spectra - probably fixed constant frequencies, burried in noise. On the hardware side I'm using low noise high sensitivity in-amps, and I'm still prototyping the sensors and measurement methods, but from what I see - I'm on the right track. The goal is to create a design, that a) could be used with regular studio soundcards (from what I see, I can get signals below 0.5Hz, which fits my expectation) and b) would be cheap/easy/accessible hardware solution.

On practical side - I have an audio file at 44.1k or 96k, that contains a lot of noise (for example captured on contact electrodes), and I need to figure out if there are any periodic regularities in such signal and where. While I have on board pretty nice FFT analyzer, something tells me, that these signals may be burried some decibels below the spectral noise floor, and from what I found yesterday - the lock-in amplifier approach - may be helpful to extend the work I do.

But I don't exactly understand yet how this lock-in methods work, so.

First - I'd like to see some modular dataflow schematic that shows how it works, and what I'm getting on output.

Second - I'd like to have something, that allows me to see the spectra of analyzed files with high accuracy (in terms of signals buried in noise).

Is this a way of equalizing/filtering the audio file for further use in FFT analysis or it must be bonded with FFT inside the schematic? If this lock-in technique(s) prepares the file (removes noise, keeps periodic signals), that can be analyzed elsewhere, then I'm happy with it.

From what I understood (or misunderstood), these lock-in methods require some reference signal, that is being applied to input signal, and then at some step size - it acummulates somehow data at increasing frequency steps. I have no idea how this fits the possibility of covering whole acoustic spectra. But even exporting images of the spectra in a loop mode, subband by subband would be good, if this shows something. I'm not sure what variables should depend on user, to give the ability to adjust various parameters of performance vs accuracy.

So I appreciate any help with it.

BTW, I'm not sure if standard stock/wavetable oscillators are good for it, since they don't offer stable freq accuracy.

For example - I may have files similar to attached below (it's a really rough test).

Here is a demo - not sure if that will help you.

tester wrote:For example - I may have files similar to attached below (it's a really rough test).

FFT reveals distinct peaks at 50 Hz, 150 Hz, and 1500 Hz. The rest is just noise
Edit: Although if you analyze it in the time domain, there seems to be some pattern which does not show in the frequency domain.

Sure, but these peaks are rather artifacts. The sample I uploaded was to show how the signal may looks like. The recording was made on unfinished sensor base (in this example - contact pseudo-electrodes in water), not according to any specific destination method, without any shielding, and so on. It's also a part of checkup, i.e. how rude the procedure may be to get anything of use.

But I'd like to push such signals through the lock-in virtual amplifier routine, to see what else will emerge. Plus I'd like to understand the lock-in idea.

As for time domain, I'm aware of that too. I'm thinking to make a sort of peak counter, that will count peaks in general and distribution in time.

//

A simple test design. Eventually, instead of sine osc - some distorted signal (like pack of sines around the same freq) could be used. Of course this is simplified example.

Now my question - how deep this signal can be burried in noise, so a) that it is still possible to get it and b) with lock-in methods?

Either I overlooked or the browser did not refreshed. Thanks a lot, checking on it.