Hi

1.Cross-Correlation of signal with selected frequency(or band of frequencies) would be a good bet after eliminating the dc component.(Notch filters may give some results)

2.Another possibility is to solve very  high order time domain linear equations using    [A]x =b  . The b is obtained at various instants using assumed frequency range . The x will give the amplitudes. If phase is important then we may use both sin and cos terms.

Cheers

The obvious answer is to first measure the total field then shield one board (earthed copper or aluminium foil to form a Faraday cage around it perhaps) to remove its contribution. Fourier transforms of the two signals (i.e. compare two spectrum analyser traces) will show which frequencies get removed by the shielding so are due to that board directly or due to intermodulation products between that board and others. You may also need to put filters on connecting leads, something like ferrite beads for example.

Hello Don,

It seems that you are experiencing severe noise problem that has those "Thousands" of frequency components in it.

As George has mentioned, a solution like Farady cage would be helpful to you. Or in simple terms I would state to use proper SHIELDING and GROUNDING techniques to eliminate those unwanted frequency components and then to a Spectrum Anallyzer Scan to identify the Peaks of dominant frequencies.

Then use filters to remove or select frequencies of particular interest.

Simple in-line LC filters would help you and ferrite beads to counter the high frequency components.

Best Luck!

At my university, In 1977, I had a project involving gathering small signals from muscle behavior signals in Cerebral Palsy clients. This was a study of myographic signals from partially responsive patients.    

We had radio interference from three radio stations.  The signal was rock-and-roll and extremenly strong as we were directly in the antenna beam path.  This was a second floor lab and we could see the antennas. .  

Our Spectrum Analyzer showed extremely low S/N ratio.

All efforts at simple shielding and groundind failed. 

In 1977, I only had Analog design options, 

( no Digital Signal Processing ). 

Simple Faraday Shielding of the mother-board was not effective. 

Basically, we blocked all unwanted signals, 

and captured only the target signals. 

All this was in Analog Design mode. 

The ultimate method was a multiple approach.

First, we built a 8' x 8' x 8'  Faraday Cage room.  The walls of this shielded "cage" was grounded via #4 AWG copper to a 10' ground rod outside the building. ( Various other electrical "grounds" were far from effective.)

Secondly, the electronic amplifier I designed was based on the  Instrumentation Amplifier  (IA)  approach, with an active ground presented to the IA .   I designed that instrument from scratch, using high performance OpAmps.  Now days, "Analog Devices, Inc."  has some beautiful devices that could speed up the design time greatly.  

(1) our IA circuit had a common mode rejection of > 90 dB, which virtually eliminated the outside noise which was common to all leads entering the equipment.

Since we used eight pick-up leads, we had eight IA devices, attached to the pick-up leads about 1 foot from the sampling point on the patient. 

(2) Each mother-board had a ground plan covering everything, and all grounding was equi-potential to a single point.  

(3) All leads, incoming and outgoing, were threaded through ferrite beads, further blocking the outside noise.

(4) Each was a very tightly built cabinet. 

(5) This equipment was placed inside the Faraday Cage, 

and all leads were kept as short as possible, directly beside the patient. 

Ours solution was not simple, 

but it was very effective in blocking 

the Amplitude Modulated Radio Signal ,

and capturing only the target signal.  

In looking back, it is still evident that

(1) the Faraday Cage 

and

(2) the Instrumentation Amplifier design

would be required to drive

any modern Digital Signal Process computer program.  

Young folks, there is NO magic in the world of research, 

only magical researchers who know how to design. 

Study well, and good luck. 

http://www.geocities.ws/glene77is/

Hi - Thanks to all for your great idea's!  I should have advised that I cannot touch them, nor hook equipment up to them.  This is where it gets difficult......

Hi, let me add my bits ...

IF you know rather exactly what your signal is looking like (in 'waveform', amplitude is less important if it is a constant factor to the waveform): cross-correlation is a way to detect a signal within "a lot of junk". Some kind of 'brute force' approach, but easily working down to 0 dB and even below.

If your signal is not so well-defined: well, cross-correlation is in principle applicable to many 'problems'. But adjusting can be quite daunting.

U. Dreher:  How does that work?  Can you cite an example for me?

'cite': no. I used it about 25 years ago and can say the results were quite impressive

But let me try to explain..

In the image attached (from https://en.wikipedia.org/wiki/Electrocardiography) I marked a section (aka 'object' or' target') of the top-most trace: the part of the trace between the red lines.

Imagine target and trace as just series of amplitude values - sampled at a constant data rate.

The task is to detect 'similar' (no need for 'identical') occurrences of this target within the trace. This can be done by cross correlation - preferably 'normalized cross correlation' (see https://en.wikipedia.org/wiki/Cross-correlation).

In principle, you are 'shifting' the target along the trace, each time one value more to the right, and 'folding' the sample with the a section of the trace of identical length. The numerical result of normalized cross correlation is the correlation coefficient: 1 indicating 'perfect match', -1 'exactly the opposite'. In most cases values exceeding 0.95 indicate a 'sufficient match'. Though the exact threshold may vary from application to application.

The result from shifting the target along the trace is a series of correlation coefficients.

The algorithm works well is your sample has at least two edges crossing the mean value line. It does not perform too well is the deviations from the mean line are very small and It has some limits when applied to periodic signals, where you will get matches every period. (In the case of the ECG, you can get the heart beat rate from the 'distance' between matches.)

Hope the above description is understandable and helps you to dive deeper into the subject...

U Dreher: Thank you for that explanation. Yes, I understand what you mean now. It's a cool idea, but needs a repetitive signal. I need to recover unique signals. Unless I misunderstand - this would not work?

There is no need for a repetitive signal - a single occurrence can be detected. But you have to have prior knowledge: what the signal to be detected is 'looking like' - the waveform of the signal to detect.

Otherwise, the first questions to be answered would be: "What 'makes' the signal to be detected. How can it be described?"

U. Dreher:  I think I have reached the end of where this can go for now.  To answer your question - What am I doing?  I envision monitoring patients in hospitals without needing any wires connected to them.

Slowly, progress is being made in the extraction of EEG signals by remote means in a person.  Current research needs the reading unit to be 4 mm away from the body.  Once a greater distance capability has been reached - we will have to figure out how to isolate one reading from another.  The human heartbeat is unique.  I believe that other human bio-signals are also unique and thus could be read and associated with a person. 

But I cannot find any credible research having been done on the human bio-field - other than just EEG readings based on helmets or MEG (etc.)...   Because those are X,Y,Z specific, I cannot see a unit remotely located reading them with any real level of accuracy.  Thus I am now looking for another, higher-level (more complex) signal that could be read. 

But until I know what could be monitored I do not see how this question can proceed further than it already has.