My take on the HH2

[green]

We are doing the Same timing sequence- I do the EF sample just before the TX pulse and it is the beginning of the timing sequence.
A big difference is that I run an adjustable GB and EF sample time whereas you seem to have both fixed to 100us plus the target sample.
Whereas I have a Target sample and the No GB sample the same (15us) then as GB adjust is increased the GB sample and EF samples increase by the same amount.
So with GB = 0: EF sample = 0us, TX =100us, Delay1, Sample1 = 15us, delay 2 = 20us, GB sample = 15us.
With GB = 50us: EF = 50us, TX = 100us, delay 1, target sample = 15us, delay 2 = 20us, GB sample = 65us.
This meets the diagram is Post #72 of: http://www.geotech1.com/forums/showt...-Balance/page3

I do have most of the timing adjustable using either a trim pot or panel mounted pot int the processor's ADCs.
Either way, hardware timing or software timing we accomplish the same thing.

Have you been able to Test the GB on mineralized soil? I haven't as yet so do not know if this GB scheme actually works.
Have you tried slightly different timing, delays & samples, and tested on mineralized ground?

The one thing I have found is with the GB sampling there becomes a 'pivot point' where high conductors cause a negative output from the integrator and low conductors give a positive output.
Other variations of timing does not do this.

Looks like we have the same sequence with GB on. With GB off I have about 600usec decay time between target sample and EF sample and you have 20usec. I either operate with GB on with target sample time adjusted to cancel ground or GB off.

I have some ground from California and my back yard(in a 1qt zip lock bags) and a couple bricks that cancel at about the same GB settings that I use for bench testing. The California ground is a lot hotter. The detector does GB in the back yard.
Including a test I posted awhile back with some different first delay times and adjusting target sample time to cancel ground signal. Voltmeter readings of integrator out.

Zapped my bench circuit awhile back. My son has the first completed detector that I was using in the yard. Have a second detector over 90% complete. Read where someone could detect a quarter at 24inches and have been trying to figure out what I would need to change for mine to do it. Got side tracked building Target Response Tester with log amplifier. Never done any metal detecting, doing it mostly for a challenge and to learn something.

[waltr]

Those posts you linked to are what I used to do the timing.

My timing does this (from post #72 of link above): http://www.geotech1.com/forums/attac...0&d=1405340472
The TDI does four samples with the last two (positive and negative into integrator) being the Same time (width) but with DC gain after the integrator (GEB adjust).
The discussions in all the links both you and I posted change this adjustable DC gain to an adjustable Width. As per the diagram, the last two samples partly cancel, therefore only one of the samples, at a reduced width, is required to satisfy the equation:

x = A1(S1) + A3(S3) - A2(S2)
where A2 = A1 + A3

Therefore my samples times are:
A1 = 15us Int out Pos
A2 = 15us to 200us Int out neg
A3 = 0 to 185us Int out pos

I do understand that without GB sampling the target sample is a positive out of Integrator and the EF sample should be much later and is negative out of integrator so the the EF sample Cancels any EF signal. Both samples are the same time (width) and gain.
What I don't understand, except by the equation stated, is how to do an EF sample far from the target sample when the GB adjustment goes to Zero?
Also, do I need to adjust the GB sampling time (width) to cancel different ground mineralization? Or is there just ONE GB timing, width, that cancels All bad ground?

As I have setup timing the detector does not respond to Ferrite or to red bricks.
Playing with other timing schemes I have gotten the detector to respond to both the ferrite and red brick. These other timings also tend to False when moving the coil, EF not canceled?

In the chart you posted, what is the timing sequence when you have No GEB?

[green]

Those posts you linked to are what I used to do the timing.

My timing does this (from post #72 of link above): http://www.geotech1.com/forums/attac...0&d=1405340472
The TDI does four samples with the last two (positive and negative into integrator) being the Same time (width) but with DC gain after the integrator (GEB adjust).
The discussions in all the links both you and I posted change this adjustable DC gain to an adjustable Width. As per the diagram, the last two samples partly cancel, therefore only one of the samples, at a reduced width, is required to satisfy the equation:


Therefore my samples times are:
A1 = 15us Int out Pos
A2 = 15us to 200us Int out neg
A3 = 0 to 185us Int out pos

I do understand that without GB sampling the target sample is a positive out of Integrator and the EF sample should be much later and is negative out of integrator so the the EF sample Cancels any EF signal. Both samples are the same time (width) and gain.
What I don't understand, except by the equation stated, is how to do an EF sample far from the target sample when the GB adjustment goes to Zero?
Also, do I need to adjust the GB sampling time (width) to cancel different ground mineralization? Or is there just ONE GB timing, width, that cancels All bad ground?

As I have setup timing the detector does not respond to Ferrite or to red bricks.
Playing with other timing schemes I have gotten the detector to respond to both the ferrite and red brick. These other timings also tend to False when moving the coil, EF not canceled?

In the chart you posted, what is the timing sequence when you have No GEB?

What I don't understand, except by the equation stated, is how to do an EF sample far from the target sample when the GB adjustment goes to Zero?
In the chart you posted, what is the timing sequence when you have No GEB?

I don't use a micro because I never learned how to program one. I use a minor loop(first delay, target sample, second delay, ground sample). Rate oscillator starts minor loop(EF samples), minor loop ending starts Tx, Tx ending starts minor loop second time(target samples), waits for rate oscillator to start sequence again. Target sample and ground sample control opposite integrator switches the second time through. A switch enables or disables taking a ground sample. Didn't make EF sample, ground- target sample time because I didn't see a easy way with logic chips. Starting the minor loop before Tx makes for maximum time between target samples and EF samples. Would think it would be easy to program sequence in a micro. I've tried making different parts of the minor loop adjustable. Now approximate, first delay(6 to 20usec) and target sample time(9 to 25usec)are pot adjustable, second delay about 4usec, ground sample 100usec, Tx 160usec, rate 1000pps.
I use a switch to disable GB so minor loop is(first delay, target sample).

Rate oscillator starts minor loop(EF samples), minor loop ending starts Tx, Tx ending starts minor loop second time(target samples), waits for rate oscillator to start sequence again. Target sample and ground sample control opposite integrator switches the second time through. A switch enables or disables taking a ground sample.

Also, do I need to adjust the GB sampling time (width) to cancel different ground mineralization? Or is there just ONE GB timing, width, that cancels All bad ground?

Just my experience. The grounds I've tried and the bricks all cancel at or very near the same settings when sampling Tx off. When you say ferrite is that ferrite beads or ferrous material. For me ferrite beads have a very low signal or none at all unless I try to sample to soon since they have a large X signal.

What are your sample times to cancel red brick?

These other timings also tend to False when moving the coil, EF not canceled?

If you are talking bench testing, shifting my weight on the folding chair I'm sitting on is detectable. I find holding a fridge magnet near the stationary coil and moving it away quickly is a good way to test for EF if moving hand without magnet isn't detected.

[Qiaozhi]

What I don't understand, except by the equation stated, is how to do an EF sample far from the target sample when the GB adjustment goes to Zero?

Mathematically, with S1 = main sample, S2 = EF sample, and S3 = ground sample:

X = A1(S1-S3) - A2(S2-S3)

which subtracts the ground sample from the main and EF samples before finding the difference.
A2 is the ground balance control gain, Adjusting this will eliminate ground.

Alternatively, you can also use this equation:

X = A1(S1) + A3(S3) - A2(S2)

where A2 = A1 + A3

If you set GB to zero, then S3 = 0.
Therefore: X = A1(S1) - A2(S2), and A2 = A1.
Then: X = S1 - S2, which is the situation where GB is off.

If you're referring to the case, where the gain of each term is controlled by changing the sample pulse width, this becomes more difficult to implement. The effective gain is not a linear function of the sample width, so I suspect you will need two EFE samples to make this work properly. I believe this is the way that it's handled in the TDI.

[waltr]

Thanks and it is in those threads I posted a link where this idea comes from.

X = A1(S1) + A3(S3) - A2(S2)

where A2 = A1 + A3

If you set GB to zero, then S3 = 0.
Therefore: X = A1(S1) - A2(S2), and A2 = A1.
Then: X = S1 - S2, which is the situation where GB is off.

This is the equation I use but do sampling pulse width instead of a DC Gain.
I have read that the effective gain in not linear but do not know what the non-linear relationship is.
Any hints (posts, papers to read)?
There is a paper published by Minelab ( https://www.minelab.com/__files/f/11...S_&_THEORY.pdf) that goes into the math of Ground balance of VLF and PI detectors.
When plotting the decay curves for a PI detector on a log/log graph the Ground decay became a Straight line (figure 9B of Minelab's paper) whereas a metal target's decay is curved. This must be the Hint on the GB to EF sampling times.

[green]

Thanks and it is in those threads I posted a link where this idea comes from.

This is the equation I use but do sampling pulse width instead of a DC Gain.
I have read that the effective gain in not linear but do not know what the non-linear relationship is.
Any hints (posts, papers to read)?
There is a paper published by Minelab ( https://www.minelab.com/__files/f/11...S_&_THEORY.pdf) that goes into the math of Ground balance of VLF and PI detectors.
When plotting the decay curves for a PI detector on a log/log graph the Ground decay became a Straight line (figure 9B of Minelab's paper) whereas a metal target's decay is curved. This must be the Hint on the GB to EF sampling times.

I'm wondering why the effective gain isn't linear. Integrator gain=R feedback/Rin*sample time*sample rate(pps).

[Qiaozhi]

Thanks and it is in those threads I posted a link where this idea comes from.

This is the equation I use but do sampling pulse width instead of a DC Gain.
I have read that the effective gain in not linear but do not know what the non-linear relationship is.
Any hints (posts, papers to read)?
There is a paper published by Minelab ( https://www.minelab.com/__files/f/11...S_&_THEORY.pdf) that goes into the math of Ground balance of VLF and PI detectors.
When plotting the decay curves for a PI detector on a log/log graph the Ground decay became a Straight line (figure 9B of Minelab's paper) whereas a metal target's decay is curved. This must be the Hint on the GB to EF sampling times.

I think the problem is related to the fact that you're using one EF sample. If you use two separate EF samples, and associate one with the main sample and the other to the ground sample, then you won't have to be concerned with the non-linear relationship.

The equation will then be: X = A1(S1 - S3) - A2(S2 - S4)
where S4 is the second EF sample.
In your case, you'll need to make S1 width = S3 width, and S2 width = S4 width.

In conclusion, when you amplify the ground sample signal, this also contains the EF signal. Therefore you must amplify the associated EF sample by the same amount. In your current configuration the same amount of amplification is being used for both cases, but the amplification required for the ground sample will not be the same as that required to eliminate EF from the main sample. Hence the EF elimination is compromised.

[Qiaozhi]

I'm wondering why the effective gain isn't linear. Integrator gain=R feedback/Rin*sample time*sample rate(pps).

Just consider the main sample pulse on its own:
Let's say you have a sample width of 10us. If you put this signal through an amplifier and double the gain, you'll have twice the signal amplitude. However, if you simply double the sample width to 20us, you do not get twice the amplitude. The relationship is non-linear.

[waltr]

I think the problem is related to the fact that you're using one EF sample. If you use two separate EF samples, and associate one with the main sample and the other to the ground sample, then you won't have to be concerned with the non-linear relationship.

The equation will then be: X = A1(S1 - S3) - A2(S2 - S4)
where S4 is the second EF sample.
In your case, you'll need to make S1 width = S3 width, and S2 width = S4 width.

In conclusion, when you amplify the ground sample signal, this also contains the EF signal. Therefore you must amplify the associated EF sample by the same amount. In your current configuration the same amount of amplification is being used for both cases, but the amplification required for the ground sample will not be the same as that required to eliminate EF from the main sample. Hence the EF elimination is compromised.

I'll try a four sample method using the equation: X = A1(S1 - S3) - A2(S2 - S4) and making A2 (gain/width) variable with GEB adjust.

Just consider the main sample pulse on its own:
Let's say you have a sample width of 10us. If you put this signal through an amplifier and double the gain, you'll have twice the signal amplitude. However, if you simply double the sample width to 20us, you do not get twice the amplitude. The relationship is non-linear.

That sounds like the Integrator output as being non-linear with Sample time.
Verse the Ground signal being non-linear with time from TX pulse.

[Qiaozhi]

That sounds like the Integrator output as being non-linear with Sample time.
Verse the Ground signal being non-linear with time from TX pulse.

Whatever the reason, the fact is that adjusting the sample pulse width does not have the same effect as adjusting the gain.

[green]

What I don't understand, except by the equation stated, is how to do an EF sample far from the target sample when the GB adjustment goes to Zero?
In the chart you posted, what is the timing sequence when you have No GEB?

I don't use a micro because I never learned how to program one. I use a minor loop(first delay, target sample, second delay, ground sample). Rate oscillator starts minor loop(EF samples), minor loop ending starts Tx, Tx ending starts minor loop second time(target samples), waits for rate oscillator to start sequence again. Target sample and ground sample control opposite integrator switches the second time through. A switch enables or disables taking a ground sample. Didn't make EF sample, ground- target sample time because I didn't see a easy way with logic chips. Starting the minor loop before Tx makes for maximum time between target samples and EF samples. Would think it would be easy to program sequence in a micro. I've tried making different parts of the minor loop adjustable. Now approximate, first delay(6 to 20usec) and target sample time(9 to 25usec)are pot adjustable, second delay about 4usec, ground sample 100usec, Tx 160usec, rate 1000pps.
I use a switch to disable GB so minor loop is(first delay, target sample).

Rate oscillator starts minor loop(EF samples), minor loop ending starts Tx, Tx ending starts minor loop second time(target samples), waits for rate oscillator to start sequence again. Target sample and ground sample control opposite integrator switches the second time through. A switch enables or disables taking a ground sample.

Also, do I need to adjust the GB sampling time (width) to cancel different ground mineralization? Or is there just ONE GB timing, width, that cancels All bad ground?

Just my experience. The grounds I've tried and the bricks all cancel at or very near the same settings when sampling Tx off. When you say ferrite is that ferrite beads or ferrous material. For me ferrite beads have a very low signal or none at all unless I try to sample to soon since they have a large X signal.

What are your sample times to cancel red brick?

These other timings also tend to False when moving the coil, EF not canceled?

If you are talking bench testing, shifting my weight on the folding chair I'm sitting on is detectable. I find holding a fridge magnet near the stationary coil and moving it away quickly is a good way to test for EF if moving hand without magnet isn't detected.

I use a switch to disable GB so minor loop is(first delay, target sample).

Not correct, minor loop stays the same. GB switch off, disables ground sample command to integrator switch.

I've been thinking adjusting sample width to GB would be better than taking = length samples and adjusting gain. Is there an advantage to = length samples and adjusting gain?

[Carl-NC]

In the sampling integrator, different signals produce different effects. If the integrator is non-lossy and the input signal is "DC" then integrator gain is linear with pulse width. Earth field is the closest thing you have to a DC signal. Target signals are exponential and ground is 1/t, so integrator gain is non-linear with pulse width for those signals.

In most PI detectors EF cancelation is done with late subtractive samples that exactly match the early samples in pulse width, because EF is considered "static" over the course of 200us or so. Ground & (most) target signals have decayed away so they are unaffected. If the PI design uses static target & ground pulse widths and analog gain for adjusting GB, then the EF pulses are matched to the T & G pulses and subtraction takes place in the channel integrators, where the same gains are applied to T & G signals, as well as EFT and EFG signals.

If the PI design uses a variable pulse width instead of variable gain for GB, then the EFG pulse width should track the G pulse width. If you want to use only one EF pulse, then the math is simple: the EF pulse width should be equal to the difference in the ground-minus-target pulse width. That is, if the T pulse is 10us and the G pulse is 40us, you need an EF pulse of 30us, and added to the T signal. If adjusting GB changes the G pulse width, then the EF pulse width also has to change. All this assumes both signal channels have the same gain; if they don't, then the EF pulse width has to be adjusted to compensate for the gain difference. This is why most PI designs use separate EF samples, each one fed through the different T & G channels. It's easier.

[green]

In the sampling integrator, different signals produce different effects. If the integrator is non-lossy and the input signal is "DC" then integrator gain is linear with pulse width. Earth field is the closest thing you have to a DC signal. Target signals are exponential and ground is 1/t, so integrator gain is non-linear with pulse width for those signals.

In most PI detectors EF cancelation is done with late subtractive samples that exactly match the early samples in pulse width, because EF is considered "static" over the course of 200us or so. Ground & (most) target signals have decayed away so they are unaffected. If the PI design uses static target & ground pulse widths and analog gain for adjusting GB, then the EF pulses are matched to the T & G pulses and subtraction takes place in the channel integrators, where the same gains are applied to T & G signals, as well as EFT and EFG signals.

If the PI design uses a variable pulse width instead of variable gain for GB, then the EFG pulse width should track the G pulse width. If you want to use only one EF pulse, then the math is simple: the EF pulse width should be equal to the difference in the ground-minus-target pulse width. That is, if the T pulse is 10us and the G pulse is 40us, you need an EF pulse of 30us, and added to the T signal. If adjusting GB changes the G pulse width, then the EF pulse width also has to change. All this assumes both signal channels have the same gain; if they don't, then the EF pulse width has to be adjusted to compensate for the gain difference. This is why most PI designs use separate EF samples, each one fed through the different T & G channels. It's easier.

I guess it's how you look at it. I see it as integrator out=average input*integrator gain(R feed back/R in*sample time*sample rate pps). I see integrator being linear and input changing causing the output to not double when sample time doubled.

[waltr]

thanks for posting Carl,
Yep, that is exactly what I am doing.
This was discussed and you even made this same argument in the GB threads on this. I am trying to do GEB with a single differential integrator (original Hammer Head analog circuits) that each integrator input have the same gain, just inverted. I posted my schematic early in this thread.
TX pulse = 100us
D1 = 5-35us
S1 (target into inverting integrator) = 15us
D2 = 20us (but can be adjusted with a trim pot)
S2 (into non-inverting integrator) = 15 + GEB time
S3 (into inverting integrator) = GEB time (EFE sample)
S3 is just before the TX pulse.

Pre-amp is inverting so the inverting integrator out is a positive signal.
If GEB = 50us then S1 = 15us, S2 = 65us, S3 = 50us.
IF GEB = 0 then S1 = 15us, S2 = 15us, S3 = 0us.
This seems correct?
Question is: is S2 at only 20us after S1 effective for EFE?
Or: does the three Sample method really not work and four samples are needed with the last two (more than 200us from the first two) widths be varied as GEB adjust?

Since the HH2 timing is microprocessor controlled it is easy to program or change any timing we can think of verse hardware timers that are hard to make big changes to.
The ease of changing timing allows easy experimenting based on theory and discussions. If it does not works then just re-flash the processor with the code that does work. I can go back to the simple two sample (target-EFE) in less than a minute.

I had thought of using part of a second HH2 PCB to add a GEB channel but if GEB can be done with just software and pulse timing why add the additional hardware. If this proves to work then any of the processor controlled single differential integrator PI detectors can use this scheme to add GEB. This is the purpose of this thread and the discussions herein (and maybe added to the Third edition of the your book).

[Qiaozhi]

In the sampling integrator, different signals produce different effects. If the integrator is non-lossy and the input signal is "DC" then integrator gain is linear with pulse width. Earth field is the closest thing you have to a DC signal. Target signals are exponential and ground is 1/t, so integrator gain is non-linear with pulse width for those signals.

In most PI detectors EF cancelation is done with late subtractive samples that exactly match the early samples in pulse width, because EF is considered "static" over the course of 200us or so. Ground & (most) target signals have decayed away so they are unaffected. If the PI design uses static target & ground pulse widths and analog gain for adjusting GB, then the EF pulses are matched to the T & G pulses and subtraction takes place in the channel integrators, where the same gains are applied to T & G signals, as well as EFT and EFG signals.

If the PI design uses a variable pulse width instead of variable gain for GB, then the EFG pulse width should track the G pulse width. If you want to use only one EF pulse, then the math is simple: the EF pulse width should be equal to the difference in the ground-minus-target pulse width. That is, if the T pulse is 10us and the G pulse is 40us, you need an EF pulse of 30us, and added to the T signal. If adjusting GB changes the G pulse width, then the EF pulse width also has to change. All this assumes both signal channels have the same gain; if they don't, then the EF pulse width has to be adjusted to compensate for the gain difference. This is why most PI designs use separate EF samples, each one fed through the different T & G channels. It's easier.

Thanks for the improved explanation.

I did once upon a time fiddle about with GB using variable pulse widths and only one EF, but it was tricky to implement, and the results were not that good.