Ground Balance Theory

[dougAEGPF]

Another interesting paper
UXO TARGET DETECTION AND DISCRIMINATION WITH EM DIFFERENTIAL ILLUMINATION
http://australianelectronicgoldprosp...-illumination/
dougAEGPF

[Aziz]

Where are all the math gurus?

Here is a clue for everyone. You can make assumptions and make a numerical & fourier analysis.
1. Assumption: Ground magnetisation relaxation follows the rule 1/t^a
Implement this wave form.
2. Assumption: Target response is exponential decay.
Implement this. Add more target time constant response as well (optional).
3. Look at the FFT response for 1 alone.
4. Look at the FFT response for 1+2 together and look at the frequency response change (FFT spectrum change).

You will see, that both functions 1/t^a and exponential decay are a wideband response type. Exponential decay reaches zero line faster (compared to the 1/t^a response). And you will see, that 1/t^a response is dominating in the low frequency region.

Anyone want to make the numerical simulations (with graphical output)?
Aziz

[Davor]

I have an alphabetical solution instead: LF.
Susceptibility meets viscosity there, and there is also more headroom for small stuff.

[Aziz]

I have an alphabetical solution instead: LF.
Susceptibility meets viscosity there, and there is also more headroom for small stuff.

That's a different ball game (the CW IB).

Let's focus to the PI first: Off-time sampling (TX = switched off).
We have the magnetization relaxation effect, magnetic viscosity effect (time-delayed, frequency dependent susceptibility) and the target eddy current response or ground conductivity response. And maybe the neglectable TX circuit off-time TC effect (time/frequency dependent damping, changing of TX inductivity, etc.).

I could show you all some interesting FFT responses.
Do you deserve it? Davor would deserve it. But the others? Convince me!!!

Aziz

[Davor]

IMHO there are better matches for such samples than FFT. Say DCT. In case of log weighted samples it would extract slope and offset directly.

There is also a chirplet transform that could work with chirp-spaced samples.

This is all kinda sexy, but I also lost liking for PI. To me PI is fine for a simple and cheerful beach combing rig where all it's strengths are at optimum. Everything beyond that becomes increasingly odd and exotic, with only marginal benefits.

[Aziz]

Hi all,

if you want to implement the numerical simulation, here is more clue:
1. 1/t^a (the magnetisation relaxation decay):
Implement the following:
f(t) = b*((t+phi)^a), where
b: amplitude of the response
phi: phase offset (you might want to implement the delay time)
a: the exponent: -0.95, -1.0, -1.05, -1.1, note: negative exponent this time, implementing the 1/((t+phi)^a)
t: time variable

2. Same with the exponential decay response:
f(t) = a*e^(-(t+phi)/TC), where
a: amplitude at t+phi=0
phi: phase offset (you might want to implement the delay time)
TC: time constant of the target
t: time variable
e^: exp()-function

Look at the damn FFT responses and expose the thief!!!

Aziz

[Ferric Toes]

IMHO there are better matches for such samples than FFT. Say DCT. In case of log weighted samples it would extract slope and offset directly.

There is also a chirplet transform that could work with chirp-spaced samples.

This is all kinda sexy, but I also lost liking for PI. To me PI is fine for a simple and cheerful beach combing rig where all it's strengths are at optimum. Everything beyond that becomes increasingly odd and exotic, with only marginal benefits.

Depends which side of the fence you are on, and how long for. FD seems odd and exotic to me having been in TD since 1966. The one thing that does seem better in FD is discrimination, but I am sure that will change. I had a very good PI discriminator in the 1980's but no good on mineralised ground. Now I have good PI for mineralised ground but no discrimination. Combine the two and voila! Dave Emery almost did it, and others are working on it. I don't now have the time, but I am pursuing magnetic soil experiments and TD GB. Don't see a reason to change.

As I have said before, particularly to Aziz. Make a TD/FFT detector with these exotic features and let it be tested in the real world. Then decide whether it is better than a full TD PI delays and sampling rig. Minelab have both technologies and their GB gold detectors are still full TD as far as I am aware. Is it because TD with no reactive component handles bad ground better?

Eric.

[Davor]

To be honest I wouldn't know much about extremely bad ground as it is in a completely different hemisphere from my current location. What I've seen so far is that Pareto distribution of R&D time and money worked exclusively for PI development in the last ~20years, and the discriminating machine with good GB is as near as it was 20 years ago. Practically all the nowadays machines are but copies of your design with a few bells or whistles here and there, and it is unlikely that the whole PI idea is going to make a next step unless you do it first.

Some say that SD2000 made a big step forward, yet I wonder - to what direction? Point is that besides the silly play with timing, it is still a battery hog and a copy of your design. UXO applications of that exact design did not go as deep as other detectors could, so where exactly that step forward landed? IMHO it was a MD marketing lalaland of broken promises. Please don't get me wrong, I'm in no dispute with a certain manufacturer, except that it trod on my hobbyist bubble of comfort by bully brute force, while in fact straining your design and the Pooles. My whole point is that, whether you like it or not, It is you who hold the keys to PI Nirvana and no one else. If you don't have a single solution that does it all - I'm sure it is out of other people wits, and especially those that have to steal other people's IP to go on.

I'm not in a mood to buy a story that just because some company sells it's toys as gold extracting monsters that they are any good for a marketed purpose, especially after a story of a big nugget find by a particular detector was confirmed to be a fake. Maybe it's because in my country there is no gold rush at the moment so I'm not in a right state of mind to be that gullible.

PI surely has it's advantages, and it's kryptonite. Same goes with CW. Perhaps Carl and his PI CW frankendetector is a solution that does it all.

[Ferric Toes]

What I would like to see is full analysis of the whole decay curve, not just a few bits early on. There is a lot of information there which can be extracted, even whether a target is ferrous or non-ferrous. Where I would start is to scrap the linear front end amplifier and have a wide band log amplifier instead, or maybe as well as. I have used a log amp experimentally, pre 1970, but there must be much better devices on the market now. One difficulty back then was noise level when the signal was low and the log amp runs at max gain. You need to set a threshold level below which the log amp is muted. In conjunction with that I would have logarithmically spaced time gates right to the start of the next TX pulse, which is bipolar of course, with energy recovery. I would stick with mono coils for simplicity, but use the centre tapped bifilar type so that you get a fast TX and a better optimised RX. A micro could handle the processing, i.e. integration, subtraction, division, etc.

Anybody know of a good low noise log amp? I suspect that a bit of linear low noise preamplification ahead of it would be a good idea.

Eric.

[Tepco]

Very interesting proposal, actually i tried something similar time ago, but with not much success. Analog Device AD8307 log amp chip, whit and without front-end is tested. Naturally, 500MHz BW and over 90dB dynamic range is overkill, but this particular chip was at hand. As a result, i just get long linear ramp down during waveform decay, exponential decay observed thru the log amp. This waveform changes very little in presence of any metal, i don’t believe any practical information can be extracted from it. Maybe exactly opposite approach may be better, some sort of anti-log amp, or squaring amp ( built using analog multiplier or something) to enhance initial decay part of waveform, instead of linear front-end. I just looked at waveforms, never tried to build something this way, but just not looked promising with log amp. Actually i still have this hardware somewhere in my junkbox, maybe i can find and revive it and post some results.

[dougAEGPF]

Anybody know of a good low noise log amp? I suspect that a bit of linear low noise preamplification ahead of it would be a good idea.

Eric.

?
https://docs.google.com/viewer?a=v&q...xhXt-b2CAnvuDg
dougAEGPF

[Davor]

@ Ferric Toes, perhaps a small deviation towards IB would get you out of noise troubles. The closest to such description is the Tinkerer's design, only lacking a log amp. Maybe I'd regain my confidence in PI technology if two of you could join forces.

Regarding log amplifiers, things have not changed much regarding the noise. There are log amp chips produced mainly for RSSI measurements, but I'm not sure you'd get far with these and very late samples. Carl worked on GSM chips, so he could illuminate the subject better than the rest of us.

A better question is - how many samples you really want? In case of a limited number of samples it might be a better idea to consider taps with log-weighted attenuation instead. Whatever the complexity of soil/target, you'll always have only a limited number of ranges to observe, say 3, and only a limited number of samples per range, say 2. Due to diminishing frequency content of the later samples, those could benefit from the improved S/N. BW did that by integration, yet his samples were not log weighted, but the principle stands anyway.

Frequency analysis of such samples also require late samples, otherwise the targets' frequency content is too similar to the ground's. Provided you are into learning lots about the ground. Otherwise you could eliminate charging period (by discharging the cored coil or something) and use fast bipolar repetition to improve target to ground S/N simply and directly. True, it could mask the large targets, but some kind of iambic pulsing could do the trick. Anticipating large targets at every pulse is just a waste of time.

[deemon]

Maybe exactly opposite approach may be better, some sort of anti-log amp, or squaring amp ( built using analog multiplier or something) to enhance initial decay part of waveform,

This initial decay must be not enhanced but completely suppressed - it's the only way to good discrimination . I mean that we must have a zero signal without a metal target presence . So when we see the signal - it must be the target or ground signal , and it must be easier to separate them ...

[Davor]

Actually i still have this hardware somewhere in my junkbox, maybe i can find and revive it and post some results.

Did you try bricks with that? Point is that in log domain targets of different tau will present themselves with unique slopes, shifted up or down due to proximity to the coil but at constant angle, so a mere derivation of these would give you taus directly. Magnetically relaxing materials would present themselves as ski-like curves, and in log time they straighten up. This could lead to their exclusion, hence GB. Otherwise it is irrelevant what kind of signal presentation you have, as long as it is not grazing on the noise floor.

[Tepco]

Never tried actual bricks, i have some very large ferrite for tests (E100 core) but i considered using one brick to disassemble test setup. On log scale presentation, notoriously small difference will be even smaller.