detection depth

[green]

Sorry, my last post was in error, I was tired.
I think my assessment of the vertical scale of Eric's curves is correct: 1" represents 2-fold gain (or drop).
The 1967 penny drops 2" in 140 usec, and hence 1" in 70 usec. So Time-constant = 70 / (0.693) = 100 usec. This compares to the 84 usec I got on my VLF, which seems to under-read TC (probably due to skin effect).
Like-wise, trace 5, the Cu-Ni half-crown takes 35 usec to fall by 3", = 11.7 usec for 1" drop. So TC = 11.7 / (0.693) = 16.8 usec. This roughly matches the 18 usec my VLF indicated.

I get different numbers. 2inches/decade. I've been dividing the time to decay 1 decade by 2.3 or 2 decades by 4.6 to get the TC.

For times when target doesn't decay 1 decade on the chart it gets difficult without a readable Y scale. Think I'm doing it right, maybe not.

The 1967 penny drops 2" in 140 usec 140/2.3=61usec not 100usec.

the Cu-Ni half-crown takes 35 usec to fall by 3" 2/3*35/2.3=10.1usec not 16.8usec.

.

[green]

I get different numbers. 2inches/decade. I've been dividing the time to decay 1 decade by 2.3 or 2 decades by 4.6 to get the TC.

For times when target doesn't decay 1 decade on the chart it gets difficult without a readable Y scale. Think I'm doing it right, maybe not.

The 1967 penny drops 2" in 140 usec 140/2.3=61usec not 100usec.

the Cu-Ni half-crown takes 35 usec to fall by 3" 2/3*35/2.3=10.1usec not 16.8usec.

.

After thinking about it, charting the log decay on a linear Y scale might make it easier to read TC. Example: 10 minor divisions/decade, TC=time for number of minor divisions crossed*10/number of minor divisions crossed/2.3.

X=time for number of minor of divisions crossed,(Y)=number of minor divisions/decade, Y=number of minor divisions crossed. TC=X*(Y)/Y/2.3. Examples: TC=23usec*10/10/2.3=10usec, TC=11.5usec*10/5/2.3=10usec, TC=46usec*10/20/2.3=10usec.

[Skippy]

I'm not convinced it's 2" / decade.... the numbers are too far out. But 3" / decade is a possibility (though a bit awkward).
The Penny is not too far from a US quarter in ID.
From my notes:
1967 Penny: TID = 80; F_c = 1.9KHz; TC = 84 usec
US 25c: TID = 83; F_c = 1.6kHz; TC = 99 usec

I'm going to have to rummage around, I've got a New Zealand CuNi half-crown, I think they're identical to the UK ones, but the only actual UK ones I have are dug-ups, and likely to be a bit 'off' thanks to corrosion. Pennies/ha'pennies aren't a problem.

At least Eric's 'soil decay' chart has labelled scales, so no guessing there.

[green]

Still thinking of purchasing a log114 amplifier. I could calculate the TC with linear time scale, log decay linear amplitude scale if reply #122 is correct. Is there a formula for calculating ground slope, -exponent with a log time scale, log decay linear amplitude scale? Don't know if it's worth trying a log amp if I can't get target TC and ground slope. I can get them now but I'm wondering if using the log amp the trace could be smoother with less noise. Haven't used a log amp except maybe once with sound measurement, something else to learn.

Any suggestions appreciated. different log amp? anything else to think about if using a log amplifier?

[Skippy]

Suggestions?
Well you could build your own log amp. I'm not familiar with the latest IC's, but they are intended for photodiode sensing in the optical comms business. They're cheap (5 dollars?) and their limitations don't seem to be a problem. Examples from AD:
http://www.analog.com/en/products/am...mplifiers.html
Getting one in a package bigger than a pinhead could be a problem... though there's always the Evaluation Boards, thankfully.

You could make one from opamps & matched transistors & special resistors with particular tempco characteristics, but I'm not sure it's worth it these days.. though it would be satisfying to do. The problem is they may be too slow for your needs.

Edit: having Googled it, I now see the log amp from TI is actually an IC, not some instrument/module, so all of the above is a bit redundant, you should've posted a link then things would've been clearer.

Have you considered taking multiple samples (eg. 4 ) and averaging them ( in the spreadsheet) to reduce noise? I suspect timing variations would make this hard.

[Davor]

You could try to employ a simple log amp as in NE604, or you can build one using a few dozen transistors.

@Eric, thanks immensely. The ferrous curves are a bit ski-like, as expected, except for the plated steel, where non-ferrous plating apparently dominates. I'd expect curvature to change with varying charge period. I think I have a candidate for Spice model of a ferrous target, that has a potential of enabling experimenting with discrimination strategies. It is not ready yet.

[green]

You could try to employ a simple log amp as in NE604, or you can build one using a few dozen transistors.

@Eric, thanks immensely. The ferrous curves are a bit ski-like, as expected, except for the plated steel, where non-ferrous plating apparently dominates. I'd expect curvature to change with varying charge period. I think I have a candidate for Spice model of a ferrous target, that has a potential of enabling experimenting with discrimination strategies. It is not ready yet.

http://www.geotech1.com/forums/showt...043#post200043

Some curves with a DD coil I did awhile back with some nails. More ferrous curves in earlier replies, probably more in latter replies.

[Davor]

Thanks. You captured them as linear in log-log scale, which requires fitting to achieve.

[green]

Thanks. You captured them as linear in log-log scale, which requires fitting to achieve.

I recorded them linear and can chart X or Y linear or log. If you have a chart you would like to change I could chart it with different scales.

[Davor]

I have an analogue scope, so no fingers - no chocolate.
But fitting a simulation is just a few clicks.

[green]

Charted some California ground and a steel washer at different Tx widths to compare with the US quarter and 1oz copper coin. The 125usec Tx curve is steeper than the longer Tx widths with the California ground. Tx width didn't effect the curves for the washer, 125usec is 5 times longer than the 23usec TC line on the chart. Looks like increasing Tx width over 4 times the target TC doesn't effect the Rx signal, stated more than a few times in other threads. The copper coin chart line wasn't straight on the linear log chart until after 300usec with Tx width 2000usec(4 times the TC). Looks like about 600usec for Tx width 500usec(coin TC) and 800usec for Tx width 125usec(1/4 coin TC). Looks like Tx width should be at least 4 times the target TC to get a good TC number. Maybe nothing new, but something I haven't always done for long TC targets. Charted the washer log-log because that is what I had done for the ferrous targets I referenced for Davor reply#127. The ferrous charts I referenced were straighter on the log-log chart than the washer I did today.

[scrungy_doolittle]

This leads me to a thought. IF you can digitize the curve for a PI detector, why couldn't you then run a curve fitting algorithm on the data points to get a good idea of the target? Idea is in the public domain. Given the speed of modern processors, it should be dead simple, to do a curve fit and compare the results against a table.

[Davor]

Perhaps. But there is a snag. In log-log scale you get a linear response only if you fit the beginning of a signal (x axis) properly.

[Ferric Toes]

Have a look at the patent US6326791 by Bosnar. 'Discrimination of Metallic targets in Magnetically susceptible soil'.

Eric.

[Davor]

There is a slight difference between having a ready time reference for to and fitting it from a picture. In detectors you have to at hand, and in a pictures you don't.
There are a few softwares for digitising graphs from pictures, and these work well enough for curves in linear scale. I used some in technical writing, a real help, especially for resurrecting some old graphs. However, fitting a graph in log-log scale, without knowing to is not as satisfying.