Induction Balance Stuff - Single/Multi Frequency Response, GB, Disc, Measurements, Ideas, Fun, etc.

[Aziz]

Hi all,

I have tested several digital filters using Claude Code AI:
- very low bandwidth bandpass filter for the TXref signal before decoding
- IIR 4'th order lowpass filter for the Lock-in amplifier mixer stages

It does not give any significant improvement. The standard Lock-in amplifier is doing a good job. The IIR-lowpass for the Lock-in amplifier does improve it slightly. Maybe.
Overall, there is no free lunch here.

A good tracking is required. Like a motion-mode detector to overcome the slow drift issues. Or automatic tracking.
The fckn sound card is producing too much noise. And the mono coil is also picking up much EMI noise at home. The higher the Q, the higher the EMI noise amplification. There is nothing we can do.
Cheers,
Aziz

[Aziz]

BTW,

VDI can be obtained by the phase angle of the response dIQ. You have to map the full 180° angle range into a VDI index number 0..100 or so.
Look at the response of the 2 EUR coin alone and with hot rock together now.
You see, that the VDI is not reliable in discriminating a target.

Only the shovel will ensure you a 100% reliable discrimination by digging out the target.

Aziz

[Aziz]

And this happens to my reference point. A long term drift happens:



The tracking will help to keep the dIQ zero.

[Aziz]

Let's look at the pure X-response by placing a ferrite or hot rock to the coil:


The GB process is to rotate the dIQ vector to the I-axis:


If we do this, we have made the ground balance procedure.
All pure reactive response will go to the I-axis:


So we are ground balanced now. Lets look at the response of a 2 EUR coin:


Now let's put the 2 EUR coin with a hot rock together:


You see the big difference?

Let's look at a 10x10 cm thin Al-foil only:


Since we are ground balanced, the target signal can be read at dQ part. The imaginary part of our dIQ complex number.
You see, this is the most simplest mono coil VLF, which works with the standard GB projection.

Aziz

[Aziz]

This is my initial state:
IQref is set, so the dIQ is zero (almost due to EMI noise).
Operating frequency is set to 43.1 kHz (almost at resonant frequency).
The demodulator delivers a magnitude M of 265.3 and a phase angle of -59.7 °. If I rotate the dIQ vector back to the I-axis, I can measure the real responses. The red dot in the IQ plane shows the actual dIQ vector. I have implemented a history of the dIQ vectors, so you can see the response changes back in time (angle, magnitude).

So lets look at different target responses, where we are in phase with the TXref signal. The dIQ vector will be rotated to the I-axis next time.
to be continued..

[Aziz]

Hi all,

first of all, there is no big difference in selecting the operating frequency or operating mode. Everything works the same with different phase lags. Best sensitivity is given at resonant frequency of course.

The TX reference signal (TXref) will be measured and the demodulator delivers the I/Q parts (IQcurrent). I/Q can be treated as a complex number or a two dimensional vector representation.

Because we don't know the real phase lag of the TX reference (we don't have an IB coil), we are processing only the difference vector (dIQ) in regards to a reference vector (IQref) obtained by ground balance or setup procedure.

dIQ = IQcurrent - IQref, where
IQcurrent is the current measurement of the TXref signal.

On setup procedure, we set IQref = IQcurrent.

I will show you soon the dIQ plane with dIQ is rotated to the I-axis on setup procedure (0 degree axis). And how different targets show different dIQ vector responses.

And finally projecting ground response to the I-axis, to make the ground balance and reading the target signal at the Q-axis.
Since the dIQ is very tiny, I will scale it 1000+ times so we can see the response in the IQ plane better. So the IQ plane isn't a unity IQ plane.
Cheers for now..
Aziz

[Aziz]

I have meant the spider-web flat mono TX coil of course (it is also a spiral). This coil has extra more stability by design.

[Aziz]

BTW guys,

the mono coil VLF is best operated with a flat spiral low R TX mono coil (R = TX coil's DC resistance). You can even make large flat spiral mono coils.
This gives the best depth performance and sensitivity to small targets too.

[Aziz]

Hi friends,

I have finished the quick & dirty implementation of the I/Q-Plane display presentation. I didn't need a second lock-in demodulator for correct TX in-phase angle determination. Just made it via software (rotating the I/Q output of the demodulator to the I-axis). So I only have one I/Q demodulator (simple Lock-in amplifier).

I have implemented the basic ground balance method by projecting the I/Q vector to the ground response I/Q vector.
And I also have implemented the Cayley-Transform (Moodz's solution). Yep, it works.

I will look at the different operating modes before I publish some graphs.
The operating modes are:
- operating frequency at resonant frequency (f = fr)
- at 100 Hz below resonant frequency (f = fr-100 Hz)
- at 100 Hz above resonant frequency (f = fr+100 Hz)
- Chirp modulation from -100 Hz to +100 Hz around resonant frequency (f= fr-100 Hz to fr+100 Hz)

I think the dual frequency TX operation (f1=fr-100 Hz and f2=fr+100 Hz) with two I/Q demodulators isn't really required yet.
Just give me more time please.

The mono coil VLF makes fully sense to realise it. It delivers decent detection depth for its easy realisation. Even on hot grounds.
Cheers,
Aziz

[Aziz]

Hi Paul,

Aziz I did exactly that ... however the codecs are synchronised in sampling the inputs L and R. BUT the input sampling is not synched to the output sampling ... it might be at the same rate ( eg 44.1 khz ) but the 44.1 of the TX is not phase locked to the 44.1 of the RX so you get phase rollover effects which is really bad at higher frequencies and the result is very poor noise floor when you are trying to resolve some weak target with poor ground cancellation.
Dont get me started on buffer syncronisation so TX sample n matches RX sample n so you do the demod properly.
Thats why I went to the 16 bit ADCs on the STM32H750.... they are clock for clock synchronised with the TX and sample rate of 6.4 MSPS.

Yep. That's the reason for, why I am not able to achieve same results as you have with a truly embedded TX/RX locked in system.
You have a crystal clear clock signal. I have a noisy analog line-out channel. And a noisy analog line-in channel. This will cause more noise and detection depth loss.
Nevertheless, for a basic mono coil VLF it is really a win to realise the project. It will still have enough detection depth to have much fun and success.
I like the idea of mono coil TX.

Cheers

[moodz]

Hi all,

the IQ-graphics is really enlightening.
You know, I don't have a really reference timing yet. I am observing the TX reference signal only. Whenever I start the computer program, I get different phase demodulation. This is caused by OS & HAL of the sound card driver. This can be seen well on the IQ-graphics. To fix this, I need a loop-back signal from one of the transmitter channel to the other line input. And a second demodulator, to correct the demodulation values (rotating the I/Q signal).
And we have on the TX reference signal additional phase shift due to the inherent highpass filter (capacitive voltage divider, another complex voltage divider caused by the input impedance of the sound card). This isn't an issue yet.

Let me fix the reference timing first. Then I can show you some interesting I/Q-graphics. This will take some coding time.
Cheers

Aziz I did exactly that ... however the codecs are synchronised in sampling the inputs L and R. BUT the input sampling is not synched to the output sampling ... it might be at the same rate ( eg 44.1 khz ) but the 44.1 of the TX is not phase locked to the 44.1 of the RX so you get phase rollover effects which is really bad at higher frequencies and the result is very poor noise floor when you are trying to resolve some weak target with poor ground cancellation.
Dont get me started on buffer syncronisation so TX sample n matches RX sample n so you do the demod properly.
Thats why I went to the 16 bit ADCs on the STM32H750.... they are clock for clock synchronised with the TX and sample rate of 6.4 MSPS.

[Aziz]

Hi all,

the IQ-graphics is really enlightening.
You know, I don't have a really reference timing yet. I am observing the TX reference signal only. Whenever I start the computer program, I get different phase demodulation. This is caused by OS & HAL of the sound card driver. This can be seen well on the IQ-graphics. To fix this, I need a loop-back signal from one of the transmitter channel to the other line input. And a second demodulator, to correct the demodulation values (rotating the I/Q signal).
And we have on the TX reference signal additional phase shift due to the inherent highpass filter (capacitive voltage divider, another complex voltage divider caused by the input impedance of the sound card). This isn't an issue yet.

Let me fix the reference timing first. Then I can show you some interesting I/Q-graphics. This will take some coding time.
Cheers

[Aziz]

Hi all,

oh man!, I can't realy resist to do quick & dirty hacks to my software. I want to see quick results.
I just implemented a graphics drawing of the complex I/Q change of the demodulated TX reference signal into the complex unit circle.
This will help to understand the response behaviour. Particularly, for the operating frequency (100 Hz left, exact, 100 Hz right of the resonant frequency).
Depending on the operating frequency, the LC tank behaves also like an off-resonance principle.

The complex graphics display will tell me, whether a GB will work or how to do the GB.
And how it behaves with two different operating frequencies close to the resonant frequency.
I will implement the graphics for showing traces instead of current point.
Aziz

[Aziz]

Hi all,

I have finally built the basic variant (the version without the choke). Even the TX-coil current is kept quite low, the sensitivity for a simple mono coil VLF is acceptable now.
The TX is also very sensitiv to reactive responses (hot ground).

As this is a mono coil version only yet, I don't have a true reference vector (unlike on an IB-system). The software generated internal reference can't be used due to delay caused by the OS and HAL of the sound card. HAL = hardware abstraction layer, OS = operating system. Any phase information can't be referenced therefore.
So I have to use either a relative change or obtained by ground balance (GB) procedure.

Let's look, how much the GB eats detection sensitivity.
This will take some time as I have to make a lot of changes to my software (software sanity project is progressing slowly).
Aziz

[Aziz]

This is the Q-Graph. It is close to 400 with given values (capacitor losses not taken into account however).