How does ground balance work in GPX 4500?

[kt315]

direct link http://www.geotech1.com/forums/showt...ersal-PI-Micro

[Carl-NC]

waltr is right, read the patents. The older ML designs used simple multiple pulse widths. The newer designs (all the GPX series, AFAIK) use "DVT" which means the short TX pulses are driven off a boosted supply. This matches their turn-on exponentials with the long pulse width exponential. I've not looked into the details of how the ground signal is extracted, though.

[Mechanic]

waltr is right, read the patents. The older ML designs used simple multiple pulse widths. The newer designs (all the GPX series, AFAIK) use "DVT" which means the short TX pulses are driven off a boosted supply. This matches their turn-on exponentials with the long pulse width exponential. I've not looked into the details of how the ground signal is extracted, though.

It just comes down to the sample timings.

Each of the target channels is near ground balanced with it's receive timings, and the ground channel is added/subtracted at the relevant % from the target channels.

The GB channel samples are not very complicated and are quite similar to the 2200 in operation.

Probably one of the biggest differences with the gpx series is the removal of the dc blocking stage from the signal chain.
Although the gpx is a motion detector, the "dc blocking" is done in the processor.

This allows for much higher dynamic range in the back end. The older detectors suffered from the dc blocking stage hitting the rails before the processor causing the gb to fail in bad conditions. Now that stage is digital and allows for working much more severe ground.

Cheers Mick

[Carl-NC]

Good to know, thanks.

[belax2018]

I have got it.
KT315 Thank you for the files.

[belax2018]

Dear Green.

Did three tests with different timing last weekend.
Will post it later today. Will be interesting.

[belax2018]

Could you please explain more on "dc blocking"?

As I understand almost all dc and ("dc-like") components are subtracted based on data received in a very late sample - we call EF sample.

Thank you

[Carl-NC]

Could you please explain more on "dc blocking"?

As I understand almost all dc and ("dc-like") components are subtracted based on data received in a very late sample - we call EF sample.

Thank you

No, that just subtracts out common-mode signals. The early sample usually has a DC component just from sampling the static flyback decay, whereas the late sample does not. So the output of the demod has a DC offset. Normally an RC high-pass filter follows the demod and knocks this out.

[waltr]

No, that just subtracts out common-mode signals. The early sample usually has a DC component just from sampling the static flyback decay, whereas the late sample does not. So the output of the demod has a DC offset. Normally an RC high-pass filter follows the demod and knocks this out.

In the simpler PI detectors (Hammer Head, etc) the SAT stage is DC blocking. The Integrator stage output goes through a Cap (DC Block) to the Sat stage which then has a Zero offset output when No change happens under to coil.

Did I get this correct Carl?

[belax2018]

Couple charts I posted in other threads. TRT_21, example of ground slope changing with Tx on time with constant current Tx vs constant rate Tx, each would require a different balance setting to cancel ground signal. target delay_1, some data taken awhile back showing detection distance and integrator out volts with some different targets. GEB on causes distance loss because noise increases when the 100usec GEB sample is added. I use a(Rx, two 8inch round connected figure eight)(Tx surrounds Rx)reduces noise and ground signal. Don't think it would have different timings vs mono coil. I did bury a target in back yard 10inches deep to see if integrator out voltage was the same as air with GEB on adjusted to cancel ground signal. Little or no difference.

Ok. I have tested some ground balance with two red bricks tight together with insulation tape.

TX is 100 us. Peak current 7A. Coil resistance 1 Ohm. Ampere-turns 15*7=105 A*Turns
Ground balance sample is adjustable.

1) D1=10us S1=10us D2=5us S2=86.7us.
In this case I have full ground balance.
But I can not detect any targets - they just give negative value - small size (half gramm blob of solder, 1 gram of pure gold) and big size (coins). I am loosing almost 90 percent in detection distance.
But Green had no difference with detecting targets (GB_off vs GB_on). Strange. I think this is because your current-turns is many times smaller.


2) Later I have decided to move ground balance sample further away from the target sample - so to make sure all target currents are died away for sure.
So timings are, D1=8us S1=8us D1=90us S2=165.2us. But here I multiply ground sample S2 by a multiplier of 3.
(Signal is equal to X=S1-S2*3+static_fields.)
In this case detection distance is just inch of half less than in GB_off case.

Anyway I don't understand why math prediction doesn't work.
I mean
Integral of ~1/t from t1=10 to t2=20 (= ln(20/10)) is definitely not equal to the integral of 1/t from t1=25 to t2=111.7 (= ln (111.7/25)).


Ok.

Thanks to everybody for all information you gave to me. I need few days to digest all of it.
"Feedback" will be soon.

[green]

Ok. I have tested some ground balance with two red bricks tight together with insulation tape.

TX is 100 us. Peak current 7A. Coil resistance 1 Ohm. Ampere-turns 15*7=105 A*Turns
Ground balance sample is adjustable.

1) D1=10us S1=10us D2=5us S2=86.7us.
In this case I have full ground balance.
But I can not detect any targets - they just give negative value - small size (half gramm blob of solder, 1 gram of pure gold) and big size (coins). I am loosing almost 90 percent in detection distance.
But Green had no difference with detecting targets (GB_off vs GB_on). Strange. I think this is because your current-turns is many times smaller.


2) Later I have decided to move ground balance sample further away from the target sample - so to make sure all target currents are died away for sure.
So timings are, D1=8us S1=8us D1=90us S2=165.2us. But here I multiply ground sample S2 by a multiplier of 3.
(Signal is equal to X=S1-S2*3+static_fields.)
In this case detection distance is just inch of half less than in GB_off case.

Anyway I don't understand why math prediction doesn't work.
I mean
Integral of ~1/t from t1=10 to t2=20 (= ln(20/10)) is definitely not equal to the integral of 1/t from t1=25 to t2=111.7 (= ln (111.7/25)).


Ok.

Thanks to everybody for all information you gave to me. I need few days to digest all of it.
"Feedback" will be soon.

Do you read a target signal with above targets when not taking ground sample? With no ground sample all targets signal should be +. With ground sample, small targets + large coins probably -. I would be surprised you can sample at 10usec delay with Tx 7A peak. Try 15usec delay and ground balance, might lose small targets.

[waltr]

I use similar method to GB.
My timings are: D1 = 12 to 16us, S1 = 15us, D2 = 15us, S2 = 98us to GB conductive soil.

Still have very good sensitivity except to target that match the GB hole, like some iron.
Also this gives an increase of the integrator output Voltage for low conductive targets (nickles, pull tabs, brass shells), then the decrease in integrator Voltage for high conductive targets (silver and clad coins, copper).
This actually improves the sensitivity to high conductors versus with no GB sample (S2 = S1).

I have been able to simulate this in an Excel spreadsheet.
Target signal is calculated for time increments from: V = exp(-t/Tau)
Where:
Tau = target TC
t = time after TX Stops
V = received Voltage
Ground response simply calculated as -1/t
Fill columns in t, V & Gnd
Then SUM the V's during each sampling period. S1 is positive, S2 is negative (integrator inputs).
Gnd sum 0 zero.
For short target TCs (low conductive) the sum is positive.
For long TCs (high conductive) the sum is negative.

[Carl-NC]

In the simpler PI detectors (Hammer Head, etc) the SAT stage is DC blocking. The Integrator stage output goes through a Cap (DC Block) to the Sat stage which then has a Zero offset output when No change happens under to coil.

Did I get this correct Carl?

Yes, the RC high-pass filter used for DC blocking is also part of the SAT circuit.

[green]

I use similar method to GB.
My timings are: D1 = 12 to 16us, S1 = 15us, D2 = 15us, S2 = 98us to GB conductive soil.

Still have very good sensitivity except to target that match the GB hole, like some iron.
Also this gives an increase of the integrator output Voltage for low conductive targets (nickles, pull tabs, brass shells), then the decrease in integrator Voltage for high conductive targets (silver and clad coins, copper).
This actually improves the sensitivity to high conductors versus with no GB sample (S2 = S1).

I have been able to simulate this in an Excel spreadsheet.
Target signal is calculated for time increments from: V = exp(-t/Tau)
Where:
Tau = target TC
t = time after TX Stops
V = received Voltage
Ground response simply calculated as -1/t
Fill columns in t, V & Gnd
Then SUM the V's during each sampling period. S1 is positive, S2 is negative (integrator inputs).
Gnd sum 0 zero.
For short target TCs (low conductive) the sum is positive.
For long TCs (high conductive) the sum is negative.

I've been using an Excel program also. Doesn't seem to balance ground with -1 slope with your settings. Wondering if I'm missing something. Not as easy to use as I would like. Do you know how to have Excel fill in the blanks after entering the timings or maybe your Excel program is easier to use.

[belax2018]

Do you read a target signal with above targets when not taking ground sample? With no ground sample all targets signal should be +. With ground sample, small targets + large coins probably -. I would be surprised you can sample at 10usec delay with Tx 7A peak. Try 15usec delay and ground balance, might lose small targets.

Yes I confirm. That with GB_off all targets are plus and nice detectable.
At your and walter settings all targets loose 90% depth.
Only if I use huge D2 > 80 us and then take ground sample (whith multiplication coefficient of 3) I can detect all targets (up to a D=25mm coin) as before.
Also I confirm that GB works and timing tuned correctly - I simulated with my hands movments (up and down) of the bricks in front of the coil no noticeable change in the final signal.
I can sample targets after 6.5 us - 7 us. Even at Tx=200us I can sample at 7.5 us which ends up at around 18 Amps.
I use two AD8139 amplifiers.
First has gain 2400/600 ~4. Second 4700/150 = 31. Overall gain is around 120. Also After each AD8139 is a low pass filter network. But I don't think it can change a math behind GB very much.
After second AD8139 I have ADS5560 - 40msps ADC. After ADC I have Cyclone 4 FPGA. All integrations are inside FPGA. MCU reads only the results of integrations.
I will send you soon oscilloscope waveforms after second AD8139.


To Carl:
DC blocking worth a try. Thank you.

Also according to what you say GPX 4500 ground balance works as follows.

GPX-4500 has two power supplies. One is Vh (high voltage) and second is Vl (low voltage).
Waveform is: 7 short TX pulses with Vh where they take Target sample = Th. And one very long TX pulse with Vl where GPX takes only ground sample GBl after some big delay (to make sure all target response died away) . Then they do the math X = Th-GBl*Coeff + static_fields_account.



Also it has some type of DC blocking scheme in front of the amplifiers - simple CR high pass filter

Sorry for my English, but do I understand it right?
Thank you.