THE ELUSVE 1us TC TARGET

[Tinkerer]

my mistake was that I removed the top shield, adjusted the balance and then added the top shield again. = Leave it as it is and use the electronic nulling circuit (I described it earlier, check my posts)

Electronic nulling is interesting, as it gives the possibility to re-adjust sealed coils. I tried to find your post with the search function but failed. Could you point me to your post or thread please?

[Tinkerer]

PREAMP.pdf Here is an update: The RX damping and the preamp as it is now.
The following screen shots are the RX signals of the targets:
The red trace is the target signal.
The green and blue traces are the no target signals.

1. 25x25mm foil
2. Nickel / USA 5cents
3. Lead coin, a lead ball hammered into about the shape and size of a Nickel
4. An iron coin of about the same size of a Nickel. Notice how the iron target signal is inverted.

On the schematic I also show the continuation. The input into an ADC.
My questions are:

1. How does the offset of the A and B waveforms fit into the ADC?
2. Do we need to adjust offsets?
3. How does the ADC react to the overlapping waveforms? Most notable on the lead target.

[pito]

Post #6229

TESORO GOLDEN SABRE - Geotech Forums (geotech1.com)

[Carl-NC]

On the schematic I also show the continuation. The input into an ADC.
My questions are:

1. How does the offset of the A and B waveforms fit into the ADC?
2. Do we need to adjust offsets?
3. How does the ADC react to the overlapping waveforms? Most notable on the lead target.

Shouldn't be an issue. The ADC literally takes the differential of the 2 inputs and digitizes, so the offsets will add, but the LTC2338 has gobs of input range. If you want to minimize the additive offset then add a slow autotune circuit to both final opamps. Do a mathematical subtraction of the signals on the oscope to see what the composite waveform will look like.

[Tinkerer]

With the signal in the ADC, we need to start thinking of the digital signal processing. If we sample every slot at 1us per slot, we have a fair representation of the decay curve without target. If we save this curve, we can then subtract it from the decay curve with target and ground.
Why should we look at the whole decay curve and not just a 2 or 3 samples?

The decay curve is an approximate exponential curve. Any target or ground changes this curve. Often the resulting compound decay curve is composed of more than one exponential decay curve. For example: Iron produces a short TC decay curve for the magnetic component and a long decay curve. T he long decay curve is very similar to the long decay curve of silver.
The difference of the decay of iron, the short magnetic decay is at the very beginning. If we only take 2 or 3 samples we can not see this initial short decay signal.
Above is a simulation where I try to simulate this 2TC decay curve of iron.
The red trace has a TC of about 10us like a Nickel (US5c)
The pink trace is the decay of a small silver coin, with a TC of about 50us. Note how slow the decay is. It is far from ending after 100us.
The green trace has an initial fast decay, then the decay continues slowly similar to the decay of the silver, so the 2 traces are on top of each other.

The simulation: [ATTACH]n419443[/ATTACH]

[Gunghouk]

Is it necessary to sample rising and falling sides of the saved 'reference' sample ? If the decaying flyback is most useful then determine the peak point and sample only from that point thereafter thus reducing sample sizes/processing time. The highest sampling speed achievable is preferred for greatest accuracy.

Normalisation of waveforms would also compensate for amplitude variation.

[Tinkerer]

Is it necessary to sample rising and falling sides of the saved 'reference' sample ? If the decaying flyback is most useful then determine the peak point and sample only from that point thereafter thus reducing sample sizes/processing time. The highest sampling speed achievable is preferred for greatest accuracy.

Normalisation of waveforms would also compensate for amplitude variation.

Thank you Gunghouk for the valuable feedback.

With PI time domain detectors (TC= Time Constant) we usually sample the decay of the target signals after the end of the Flyback. The ground is also a target, a very large one and sometimes composed of several distinguishable TC's. Different height of the coil over the ground results in different amplitude of the ground response. Normalization can help with that. The TC's of the ground remain the same, independent of the height from the ground.

Until now, we have not talked about sapling the Flyback, but indeed It can be sampled.
Once we have concluded the digital signal processing of the receive signal, I will return to the Flyback sampling. I did post such a circuit on this forum some time ago.

For the digital signal processing, algorithms, MCU and all that I depend entirely on the help of forum members who are well versed in that art.
So, please help me.

[pito]

help from forum members = show a picture of what you have done so far

[Gunghouk]

I'm wondering if a statistical approach to comparing waveforms is a better way to tackle the problem rather than their absolute difference at any instance in time.
ie. the difference in their means, std deviation, in other words using the correlation coefficient and regression to determine significant difference between the reference (no target but with ground included) and a potential target signal.

Absolute/normalised values would be needed throughout.
The regression (R) value, mean or std deviation or all three could be used as the threshold value for discrimination, identification etc.

I'm sure these basic statistical algorithms/libraries must exist already for most MCU compilers.

The offset from mean would always be different as the return signal contains the target signature. Spread (std deviation) will increase for all targets but significantly so for large or long TC targets.

[Tinkerer]

help from forum members = show a picture of what you have done so far

What would you like to see? Coil? TX board? PSU board? RX frontend breadbord? A bucket full of PCB's of earlier PI experiments? Or a bucket full of coil? STM MCU on the Nucleo board? SD card with parameter Firmware and ADC registry?
I can not show you the firmware, because it is borrowed, not mine.

[Tinkerer]

I'm wondering if a statistical approach to comparing waveforms is a better way to tackle the problem rather than their absolute difference at any instance in time.
ie. the difference in their means, std deviation, in other words using the correlation coefficient and regression to determine significant difference between the reference (no target but with ground included) and a potential target signal.

Absolute/normalised values would be needed throughout.
The regression (R) value, mean or std deviation or all three could be used as the threshold value for discrimination, identification etc.

I'm sure these basic statistical algorithms/libraries must exist already for most MCU compilers.

The offset from mean would always be different as the return signal contains the target signature. Spread (std deviation) will increase for all targets but significantly so for large or long TC targets.

Thank you for the feedback.
You bring up interesting points. How would you implement it? What MCU would you use? Do you write the software?
I like the idea of sdt. deviation. We could probably eliminate a lot of noise if we eliminate unwanted signals with sdt. deviation.
One of the PI problems is that we need a preamp with a bandwidth of 0 to 1MHz. This allows a lot of EMI to be picked up by the coil.
Frequencies above 1MHz are easily filtered out with high pass filters. However, high pass filters based on integration reduce high frequency noise by lowering its frequency, integrating or spreading the noise.

Low frequency noise is much more difficult to filter out, because our target acquisition involves low frequency sweep.

Digital signal processing allows for many processing possibilities that would not be possible or to complicated in analog.

[pito]

It is hard to help if info is limited. What for you ask for coil nulling if you not using it ?

[Tinkerer]

It is hard to help if info is limited. What for you ask for coil nulling if you not using it ?

pito, sorry we seem to have a language problem. I work on a PI Time Domain detector. You are offering a circuit for a VLF detector?

[pito]

my guess (based on info you provided) , you are using 2 wire audio cable with the shield, for Rx one wire and ground for Tx second wire and ground,
You are offering a circuit for a VLF detector = for any balancing

[Carl-NC]

You are offering a circuit for a VLF detector = for any balancing

The nulling circuit in post #63 will only work for a narrowband single frequency. Won't work for MF or pulse.