Latest Minelab Patent - 11th Feb 2016

[crane]

Well, ML didn't come up with anything new either, since modeling the signal by two time constants - and doing so by crude curve fitting!!! see [0052], [0052] and 0071] - for classification purposes is known ages ago. Whether you classify a given pair of taus as ground, saline, target X or smurf droppings is immaterial because the method is exactly the same.

Quote:

[0084] Although the brute-force search performs the best...

wow!

Toleno, your response is predictable!!!!!!!!!
The TDI and early ML PI GB methods leave curve fitting for dead and yet they still suffer hot rock and ground noise at high gains so imagine how useless curve fitting would be at say a gain of 2,000,000 or more!!! The TDI and early ML PI GB methods rely on the ground following a certain rule but there is a tiny variation from this rule that pokes its head up at high gains and causes large signals from the ground and hot rocks that completely mask target signals. Errors in curve fitting to remove ground noise are huge compared to this. The log uniform, log linear, saline soil and any remaining reactive signal must be removed first and fiddling with the surf Pi preamp isn't going to ever tell you how this might be done.

[clancy]

Teleno, I think you not understand fundamental difference between your UXO paper, and patent.
Paper you site, for page 50 (6.3.1), this fit log-uniform + saline soil model to Rx time data points, and then research vary Tx pulse length to see if Rx signal be just soil signal, or, UXO + soil. Paper do NOT transform any data to target TC spectrum and also do not actually “ground balance.”
Signals in patent that process to give TC spectrum be from demodulate and then LPF for good signal-to-noise and cancel vary static magnetic field; NB: Process “signals” to get TC spectrum in patent NOT = time data signals. Patent completely different to UXO paper.
This patent be first publish to say how get TC spectrum from ground balance signals (from demodulate then LPF) I think, but maybe someone find real prior art, not a paper that actually do something different? (If they do, I be very surprise.)
OK Teleno, you say art transform unground balance signal (demodulate and LPF) to TC spectrum same as art transform ground balance signal to TC spectrum:
First please you tell how many minimum unground balance signals are need to make 3 point TC spectrum? Easy answer of course.

Now I ask you how many conductive ½ space + log-linear + log-uniform simultaneous null demodulate and LPF minimum signals be need to make 3 point TC spectrum?

[Carl-NC]

We're going around in circles here. I'm not saying BBS/FBS is PI, I'm saying it is more closely related to PI than it is to VLF.

Yes, multiperiod pulse is for Minelab's PI but they apply the same to BBS/FBS. The patent description even says so, ie, Discriminating Time Domain Conducting Metal Detector Utilising Multi-Period Rectangular Transmitted Pulses. There are no two frequencies just as there are no two frequencies in their MPS PI detectors. You can't possibly come away thinking VLF after reading the patent!! The processing is entirely different!

Crane, I think there's pretty much unanimous agreement that things are not always as Minelab describes. 17 frequencies? 28 frequencies? So I wouldn't take their patent wording at face value. I will say that the way the do signal sampling in BBS/FBS appears to be more time-domainish than frequency-domainish, but the reality is it's not as far from traditional FD methods as it appears. And as I said, all detectors -- whether VLF, PI, FD, or TD -- do sampling/demodulation pretty much the same way, all in the time domain. All that differs is the timing.

In any case, consider the following transmit waveform:



What would you call it? Multifrequency? multiperiod? How many frequencies/periods/whatever? Would you process it in the FD or TD? Is it even a practical or realizable approach?

[moodz]

Bi level amplitude hyper abrupt frequency shift keying or BLAHAFSK ..

[moodz]

...theres no such thing as a tc spectrum. In a tech discussion we can make an assumption you are referring to a freq spectrum ...however not in a patent. Legally spectrum means a range from the latin specere to look. The patent attempts to define a tc spectrum in 63 but uses etc ...not in a patent.!

[Qiaozhi]

In any case, consider the following transmit waveform:

[ATTACH]35281[/ATTACH]

What would you call it? Multifrequency? multiperiod? How many frequencies/periods/whatever? Would you process it in the FD or TD? Is it even a practical or realizable approach?

This signal is an example of "Analog Time Domain Multiplexing".

[WM6]

Probably better than push Time Domain technologies under the Pulse Induction ranking, is to done vice versa.

Then we have primary three classifications:

1. Frequency domain detectors (all sort of continuous wave -CW- detectors)
2. Time domain detectors (PI and other time domain solutions)
3. Hybrid technologies (combination of 1. and 2. and even some 3th technology as p.e. combinations with magnetic field measurement)

Maybe such start classification will lead to less controversial views on the matter.

[Teleno]

Toleno, your response is predictable!!!!!!!!!
The TDI and early ML PI GB methods leave curve fitting for dead and yet they still suffer hot rock and ground noise at high gains so imagine how useless curve fitting would be at say a gain of 2,000,000 or more!!!

ML is doing curve fitting too, so what are you telling me?

The log uniform, log linear, saline soil and any remaining reactive signal must be removed first and fiddling with the surf Pi preamp isn't going to ever tell you how this might be done.

Certainly not, it's the publications I linked on modelling signals based on two time constants what tells the skilled person (and taught ML) how it's done.

Any more straws to cling to?

[Teleno]

Teleno, I think you not understand fundamental difference between your UXO paper, and patent.
Paper you site, for page 50 (6.3.1), this fit log-uniform + saline soil model to Rx time data points, and then research vary Tx pulse length to see if Rx signal be just soil signal, or, UXO + soil. Paper do NOT transform any data to target TC spectrum and also do not actually “ground balance.”

TC spectrum is taught in this paper: http://geosensors.com/global/Hollada...SAGEEP2004.pdf on page 8, "Estimation of Decay Parameters" for unexploded ordinance (UXO) discrimination.

This paper from 2011 discloses ground balance using the same technique: LANDMINE DETECTION USING THE DISCRETE SPECTRUM OF RELAXATION
FREQUENCIES

We propose to detect landmines using an EMI model
based on the discrete relaxations of the target [5]. This model
has several attractive features, including a sound theoretical
treatment, physical significance of model parameters, and
orientation invariance.

2.2. Soil Prescreener

A soil model and a prescreener based on this model is presented
here. The prescreener filters out responses that are like
those due to the magnetic properties of the soil. The prescreening
process is very efficient.
The frequency dependence of the soil responses share a
similar trend. The real part has a linear trend with respect
to the log-frequency, and the imaginary tends to be a constant
[11, 12]. From these observations, we propose a model:
...
where p1 and p2 are model parameters. Given a response
measured at N frequencies ω1, . . . , ωN, the response can be
fitted to the model (4) via a least-squares minimization, which
can be performed very efficiently.
...
Because the model describes a behavior very specific to
the soil and is not usually observed in metallic objects, the
model can be used as a prescreener to determine whether a
target is present based on the fitting residual.

As I said, ML has plagiarized the technique (fitting of two time constants) to model targets, ground, saline and also the testes of the Pharao, but it remains the same modeling technique known from the prior art and therefore obvious and not inventive.

I repeat for clarity: the ML patent models ground the same way it models targets: as a TC spectrum of two time constants that are fitted to the signal by brute force. No invention!

Edit: A bonus paper on the same subject (see 3.2.1 and 3.2.2): ESTIMATION OF THE DISCRETE SPECTRUM OF RELAXATIONS FOR ELECTROMAGNETIC INDUCTION RESPONSES

[moodz]

...hmm the word spectrum does not even appear on p.8

[Teleno]

...hmm the word spectrum does not even appear on p.8

The definition of discrete SPECTRUM it is "decay parameters (time constants and weights)" (page 8 )
The signal is "decomposed as a weighted sum of pole responses", that is a discrete SPECTRUM of time constants:

On page 9:

"...two time constants were estimated ... for this example .. approximating the two strongest eigencurrent modes in the plate model..."

Again, the two time constant approximation of a target in ML's "invention"

[moodz]

The definition of discrete SPECTRUM it is "decay parameters (time constants and weights)" (page 8 )
The signal is "decomposed as a weighted sum of pole responses", that is a discrete SPECTRUM of time constants:

On page 9:

"...two time constants were estimated ... for this example .. approximating the two strongest eigencurrent modes in the plate model..."

Again, the two time constant approximation of a target in ML's "invention"

...then that is the wording that should have been used in the clear and unambiguous language in a well written patent. :-)

[Teleno]

...then that is the wording that should have been used in the clear and unambiguous language in a well written patent. :-)

The language is OK, the skilled person knows what the mathematical representation of a discrete spectrum is and the terminology.

What's not OK is trying to stop others from using the known state of the art.

[kt315]

Crane, I think there's pretty much unanimous agreement that things are not always as Minelab describes. 17 frequencies? 28 frequencies? So I wouldn't take their patent wording at face value. I will say that the way the do signal sampling in BBS/FBS appears to be more time-domainish than frequency-domainish, but the reality is it's not as far from traditional FD methods as it appears. And as I said, all detectors -- whether VLF, PI, FD, or TD -- do sampling/demodulation pretty much the same way, all in the time domain. All that differs is the timing.

In any case, consider the following transmit waveform:

[ATTACH]35281[/ATTACH]

What would you call it? Multifrequency? multiperiod? How many frequencies/periods/whatever? Would you process it in the FD or TD? Is it even a practical or realizable approach?

not same signal Carl.
http://www.geotech1.com/forums/showt...&highlight=bbs
Ok, this is the TX image for a Sovereign Elite and similar MD by Minelab.
It's composed by a large pulse and a sequence of many small pulse, each
one have the same time period. I'm not shure but it seem to be a train of
17 pulse (and this is the number of freq. from 1.5 to 25KHz inside-analizer).

In this MD the power is from one battery-pack of 12V but the pulse go from
-10 to +10V (my image have an error...), this is obtained by a bridge over
the TX coil. On each pulse my Tecktronics show a pure-ring with
exponentiation decay, like a PulseInduction Coil signal without the dumping
resistor.

[Carl-NC]

not same signal Carl.
http://www.geotech1.com/forums/showt...&highlight=bbs
Ok, this is the TX image for a Sovereign Elite and similar MD by Minelab.
It's composed by a large pulse and a sequence of many small pulse, each
one have the same time period. I'm not shure but it seem to be a train of
17 pulse (and this is the number of freq. from 1.5 to 25KHz inside-analizer).

In this MD the power is from one battery-pack of 12V but the pulse go from
-10 to +10V (my image have an error...), this is obtained by a bridge over
the TX coil. On each pulse my Tecktronics show a pure-ring with
exponentiation decay, like a PulseInduction Coil signal without the dumping
resistor.

KT, you continue to look at the BBS/FBS TX drive voltage, not the drive current. The voltage is irrelevant other than the mathematical relationship it has to current, due to the fact that an inductor is in play. What matters to the transmitted magnetic field is the coil current. Look at that, and you'll see a different picture.

The ringing seen by Cossaro in the other thread was not real, it was probably a residual effect of a poorly grounded probe.