SPICE Simulation Challenge

[Teleno]

You may also want to consider Carl's reply in post #36, where he said "In PI, the same mechanisms cause off-time decay responses. Eddy responses are pretty close to exponential. Magnetic responses are not, they are still controlled by the magnetic lag dictated by the B-H curve. Ferrite again has no gap and therefore no magnetic lag; when the TX field collapses, the ferrite response collapses along with it. Targets with a B-H gap are not instantaneous, they have a time lag in their collapse. Then there are targets with magnetic remanence, which is yet another mechanism."

Sampling along the curve may reveal a difference between ferrous and non-ferrous. That's your mission, if you choose to accept it.

Target geometry usually has several axes of symmetry causing responses which are a sum of exponentials. Additionally ferrous targets, like magnetic soils, may have magnetic particles wth a size distribution causing an additional sum of exponential decays. Discrimination would then involve obtaining a spectrum of time constants. The difficulty being that fitting a sum of exponentials is a very well known ill-posed problem. The manufacturer whose name must not be said has a patent for best fitting two time constants as an approximate discrimination technique.

[waltr]

Sampling along the curve may reveal a difference between ferrous and non-ferrous. That's your mission, if you choose to accept it.

That is my plan. Here is a plot, log/lin, of some targets sampled at 2usec intervals.
I am calculating the TC over two intervals, 24-64usec & 44-84usec, with an exp curve fit.
Note the curve of the Nail verse the Cent, at first glace they look the same but there is a difference at the very beginning, the Nail has a slight lag then the curves diverge later giving a different 2nd TC.
I am exploring all possibilities that I have read about & can think of.

[Davor]

Could you do the averaging of multiple readouts in linear domain, and then present it in a log scale? I think I already see what I expected.

[waltr]

Could you do the averaging of multiple readouts in linear domain, and then present it in a log scale? I think I already see what I expected.

Is this question to me?
If so then the above plots are sums (average without the divide) of 8 sets of ADC samples (after 8 TC pulses).
Then I put the data into a code function that emulates an Op-amp integrator which is a type of filter.
The data was sampled a 2usec intervals in linear then put in Excel. The Excel plot has a log y-axis to see the difference amoung targets.
I could in Excel average more sets of data which should help smooth the curves (remove noise).
Which targets would you like to see?

[waltr]

Target geometry usually has several axes of symmetry causing responses which are a sum of exponentials. Additionally ferrous targets, like magnetic soils, may have magnetic particles wth a size distribution causing an additional sum of exponential decays. Discrimination would then involve obtaining a spectrum of time constants. The difficulty being that fitting a sum of exponentials is a very well known ill-posed problem. The manufacturer whose name must not be said has a patent for best fitting two time constants as an approximate discrimination technique.

Yes, I have played a little with summing exponentials and curve fitting to explain data I have measured (see the Plot I posted).
The TC of each target in the plot fitted to portions 24-64usec & 44-84usec of the curves are:
None: 25/35
Brass: 20/20 Note this is a large disk that produces a very strong signal
nail: 28/38
US nickle: 17/25
US cent: 29/32
US Quarter: 33/39
gold ring: 16/19

I originally expected the TC to be the same regardless of where on the curve the fit was done but quickly realized that the curve is a sum of exponentials with at least the coil itself having a decay TC. Also, when the target signal is strong then its sum becomes dominate (Brass disk) whereas if the signal is weak then the coil, etc exp sums dominate.

[green]

Thinking US nickel should have a TC close to 10us anywhere in the decay it's tested.

[dbanner]

An exponential curve implies varying tc along the curve, however small or large.

[green]

An exponential curve implies varying tc along the curve, however small or large.

TC is the same if I did the chart correctly.

[waltr]

Thinking US nickel should have a TC close to 10us anywhere in the decay it's tested.

My data/calc shows that the TC drops closer to 10us when a Nickle is closer to the coil - gives a stronger signal and become a larger part of the exp sums.

[Davor]

Which targets would you like to see?

I'd like to see the rightmost part with less noise. But if you summed the samples already, I'm afraid it is nearly the best possible.
I'm seeking the knees in ferrous response.

[green]

I'd like to see the rightmost part with less noise. But if you summed the samples already, I'm afraid it is nearly the best possible.
I'm seeking the knees in ferrous response.

What do you mean by a knee. Some scope pictures of a screw and washer. Rx_(2)200mm round fig8 IB, Tx surrounds Rx. Labeled centered on coil, should be centered over one of the Rx coils.

[Davor]

I expect to see some bending of the ferrous responses in log amplitude scale.

[green]

I expect to see some bending of the ferrous responses in log amplitude scale.

Ferrous targets don't have straight line decay, log amplitude-linear time. Thicker non ferrous targets don't have straight line decay, log amplitude-linear time.

[Davor]

Well, having a few different specimens' responses in log amplitude should do. There is a growing number of Fe responses in frequency domain in trade/scientific papers, but not so much in time domain PI.

[green]

Well, having a few different specimens' responses in log amplitude should do. There is a growing number of Fe responses in frequency domain in trade/scientific papers, but not so much in time domain PI.

log amplitude-linear time or log amplitude-log time?