The simplest bipolar PI metal detector

[boilcoil]

Hello guys, since the bipolar mode of the pulse induction metal detector is being discussed very lively here, I decided to see how a simple bipolar detector could be constructed. Since I had three completely free days (my wife is on vacation, along with the granddaughters), I sat down at the computer to see what I could do. It really took me a long time to draw the schematic (the easier part) and make the basic firmware for the job. The detector turned out really simple, I don't think a simpler circuit could be made. I use direct sampling after the amplifiers and everything after is processed virtually.

The first job was to make the TX module. I tried to solve the main problem related to the slow discharge of the input capacitances of the MOSFET, so I did a "hybrid circuit". From the MCU I control powerful bipolar transistors that can be turned off instantly (they are controlled by current). They, in turn, turn on the diagonal MOSFET, which allows me to use very low gate-source resistances. By this way every MOSFET can be turned off for a really very short time. Bipolar transistors are protected from the high voltage of the coil by high-voltage diodes, but if any MOSFET enters in avalanche mode, a high voltage could occur at the bipolar switches. For this, I have put protective zener diodes, they are really necessary (I attach all the circuits).

CHARGE PUMP is a classic circuit, controlled by the MCU on each half-cycle. Since I read somewhere (perhaps Pechkata wrote somewhere) that this method is not successful, on a PCB I left the door open - if it does not work well, due to the relatively low frequency, I can easily "graft" a generator based on 555. True, it will generate higher noise, but it would work.
All parameters are adjustable via the user menu. The current program and all of the corresponding settings can be saved in EEPROM. So when starting the detector, you can choose whether to start it from a basic program or from a saved user program (there are 8 user programs).

The detector uses 7 samples:
Three samples are taken during ON_TIME and four samples during OFF_TIME, respectively:
Atg1 = A1positive + A1negative; ON_TIME
Atg2 = A2positive + A2negative; ON_TIME
CoilCurr = (CoilCurr positive + CoilCurr negative)/2; ON_TIME
Tg1 = Tg1positive + Tg1negative; OFF_TIME
Tg2 = Tg2positive + Tg2negative; OFF_TIME
GB1 = GB1positive + GB1 negative; OFF_TIME
GB2 = GB2positive + GB2 negative; OFF_TIME
For odd cycles (positive+negative) the samples with odd number are taken, and for even cycles the samples with even number. This allows much more flexibility in the settings (e.g. TA1 can have a delay from Tmin to (Timpulse - width of ADC CoilCurr), and not from Tmin to TA2, if they were taken every time).

I'm using a 12-bit MCU, but after summ the positive and negative samples it becomes 13 bits, and after adding the even/odd samples ( or the results of GB channels) it becomes 14 bits. I could further increase the sample resolution during the averaging process, but at least for now I don't see a good reason. The averaging of the samples is performed by a digital LOWPASS FILTER, whose "integration constant" can be set from the menu.

The detector has two modes: Monocoil Mode and Induction Balance Mode.
FERRO/COLOUR discrimination, for the moment, is only done in IB MODE.

There are two independent GB channels, the outputs of which are normalized in such a way, that all unbalanced signals from them are in one direction and in one quadrant, which allows me to simply sum them at the end, and eliminate (I hope) the "functional hole". The ground balance can be turned off (for very light soils) or worked with.
The ground balance is adjusted manually for now, (channel by channel), but it's not difficult to automate it.

The CAT has three adjustable ramps that allow different speeds to be achieved when a target signal is detected and when the target zone is left. Of course, "common ramp" can be used for both events, as in the classic detectors. The CAT can be switched on or off, and by changing the settings the following modes can be implemented:
( NO_MOTION MODE);
(LOW_SPEEDMOTION MODE) or
(NORMAL_SPEEDMOTION MODE).

In general, this detector can be used for searching for coins and larger targets, but it is not suitable for GOLD PROSPECTING. The minimum ADC duration to obtain a result is about 13.5us (75 ksps), and together with the ADC request times it reaches 15us, so it is not possible to take a sample too early.



All simulations were done on PROTEUS, so I don't expect any conflicts with the real detector (moreover, almost all things were implemented on my previous monopolar PI detector).


I have attached the files.
It is true that there is much to be desired in the PCB (some links and elements need to be placed on the side of the foil), but these are my capabilities.

Regards, guys

PS:The sound is monotone and is generated by the PWM module of the MCU, with the strength being proportional to the signal from the target (the other PWM module adjusts the display backlight).

I Schems.rar

[Detectorist#1]

Hi boilcoil,
I admire your attempt to make really simple bipolar pulse detector with GB and Fe/non-Fe discrimination in our country! The proposed schematic is far from final solution but is good for the starting point of the thinking for the problem. The used bipolar power transistors in final TX stage are typical bipolar transistors with 8A max collector current. Unfortunately, they have very low current gain at big collector current ( Bmin is =24 at 3A collector current). This nulling all advantages of the use of bipolar transistors in upper nodes ( Ib = 12V/330R=36mA. 36mA x 24 = 0.864A max collector current). More real solution is to use P-ch MOSFETs in this place with TC4427CPA drivers. These drivers is possible to be controlled direct from 5V signals from the micro. But the big question rests - is this micro with relatively slow ADC (75 KSPS) and with only 12bits base resolution will be enough to made the work behind the direct sampling solution? We will see this in your next posts.

[boilcoil]

Hi Detectorist#1,
I agree 100%. Everything is exactly as you say. Let's go further, when I have at least 3 free days again (most likely in the spring or summer) I will assemble it and try it out.

PS: I forgot to mention that slow sampling, in itself, is not a big problem in cases where less frequent samples are taken. In practice, the actual sampling time is the time during which the internal switch to the ADC block is turned on, (before analog to digital conversion), which is in the range of less than 5 us. The problem comes when several fast consecutive samples are needed. In the old detector, this was solved by temporarily storing the signal in buffer capacitors, and the MCU took the samples when it was free. But this does not fit into the framework of "the simplest metal detector"

[Detectorist#1]

The end solution is far from only 3 days efforts but this is your right - to work when you want.

[Detectorist#1]

The simplest bipolar PI metal detector haves to use DD search coil for VLF detectors with center-tapped TX coil (many VLF detectors use this type) controlled via TC4427 MOSFET driver and the idea of Don (dfbowers) for Fe discriminating PI detector. Using of center-tapped TX coil will allow one-polar supply solution of 14.8V from 4x LiPo accus with simple +8V, +4V stabilizers (or +3.3V voltage stabilizer for the micro). Plus 4V will be "analog ground" and the "munus" of the battery will be "digital ground" Separate analog amplifiers for "Target" and "Disc" channels and micro with relatively fast 16-20-24 bits ADC. All rest is software (with limited number of revisions - I hope).

[Detectorist#1]

Also, the second possible solution is to use two LiPo cells as power supply and more modern OpAmps with +/- 2.5V stabilized power voltages for the analog part of the detector.

[boilcoil]

Thanks to Detectorist#1 for trying to keep the topic alive, but unfortunately, there's nothing new for me, so I haven't written anything on it.
Still, he was the reason I reread what I had written and thought about what I had written. I wrote: "...and eliminate (I hope) the "functional hole..." - but how realistic is that?
What is actually a "functional hole" in the ground balance of the detector and how to eliminate it? In practice, ground balance is a target discrimination, arranged in such a way that the metal detector rejects/misses a target, i.e. it reject a target with a certain time constant. For this reason, I think it is better to talk not about a "functional hole", but about a "target hole".
In this line of thought, there must be a target hole for one object - "earth/hot rock/mineralization, etc.", and the detector will reject any target for which Atg - A0 = k*(GBtg - GB0) where:
- A0 and GB0 are the signals from the target sample and the ground balance sample made without an object/target in the coil range (only ground, hot rock, etc..., for which the detector is balanced),
-Atg and GBtg are the corresponding samples with an object/target present in the coil range.
More precisely, if | (Atg - A0) - k*(GBtg - GB0) | < Threshold, the detector will not register anything. In other words, if the change in the decay chart, introduced by the target is such that the change of the target signal is close to the change in (ground signal)/k , the detector will not respond.

What would I gain from having more than one ground balance channel?
Let us consider two ground balance channels. Let the unbalanced signals be normalized in such a way that any change in the signal from one channel, regardless of the time constant introduced by the target, summed with the change in the signal from the other channel, increases the total unbalanced signal. Then, since the two ground balance samples are at different time locations, the two ground signals will have different changes and their sum will maybe greater than the detector sensitivity threshold. But for an object (ground + target) with a time constant equal to the time constant for which the detector is balanced, the total change will remain 0. In reality, we have not eliminated the target hole, but only narrowed it around the time constant of the balance. But we could achieve this with one ground balance channel - by moving the ground balance probe further away and increasing the gain of the ground channel.
Yes, but no - moving the ground balance probe too far away will lead to the fact that the used/ground signal will be commensurate with all interfering signals - high-frequency radio signals, low-frequency mains power supplies, and even more - the signal from the earth's magnetic field may turn out to be larger than ground signal and at certain coil swings, when the EMF signal is subtracted from the total signal, the ground signal could change its polarity, which could completely compromise the work. There are no free lunches.
In other words - it is impossible to eliminate the target hole, we can only reduce it. When I came to this result I was very surprised and decided that I was mistaken somewhere in my thinking. After all, there must be a detector without a target hole.
That's why I ran a search engine for a detector without a target hole. All I got was FISHER IMPULSE AQ.
Here, I thought, so I must be mistaken. But after reading more carefully what FISHER wrote, I noticed something strange - they define the scope of use of the detector as: "Specialized design for gold jewelry hunting" (I attach a photo). In other words - they know that there is a target hole, but they "masked" it by determining the scope of use of the detector and moving the hole out of range of the sought targets.

However, how could the problem be solved? I think the best way was proposed by Detectorist#1 (a brilliant idea, for which I take my hat off to him).
He suggested taking ground balance samples during ON-TIME and using them during OFF-TIME. What is different - during OFF-TIME all additional/imported time constants from the target raise the decay chart, respectively increase the value of the samples (let's assume that we take samples only in the first quadrant), while during ON-TIME things are radically different. During ON-TIME the signal from the search coil depends on the ratio/effect between the eddy currents and the magnetic permeability of the target, so when the coil passes over a real target the signal from it/the sample (we are talking about only one sample, with more than one sample, things can be even more detailed) can:
-increase (a target with high magnetic permeability, e.g. iron),
-decrease (a highly conductive target)
-or remain almost the same (high-resistance targets or targets with equal of both effects).
This would eliminate the need to take a ground balance sample during OFF-TIME and the associated problems mentioned above. The only downside to this is that it is only applicable to IB systems, not of my favorite Monocoil systems.
I repeat that this is a wonderful suggestion and I will try it out when I make the detector (in the spring, summer or beyond, I will put it together at some point).
Until then, I will not be able to write anything new on the topic, for which I apologize to the forum members.

[Carl-NC]

The Impulse AQ doesn't have a target hole because it does not do a complete ground balance. There are 2 channels but they overlap without a null.

Otherwise, yes, you can eliminate the target hole with another pair of GB'd channels where the GB time constant is different from the first GB pair. Combine the two, and you end up with 2 non-overlapping target holes:



Edit: In a single GB receiver, as you push the target hole farther out it gets noisier, so there is a limit to how far you can go.

[boilcoil]

Look, I would never have thought of combining an incomplete ground balance. Thanks, Karl

[Detectorist#1]

Hi boilcoil,
The idea to take the signal for ground balancing during TX-ON time for eliminating of "target hole" is not mine. This idea is mentioned in the patent US8629677B2 from John Earle and further improved in the patent US9285496B1 by Carl Moreland.

[boilcoil]

I apologize to everyone - my mistake.

[boilcoil]

Thanks to Karl for the idea of incomplete/complementary ground balance. In reality, all ground balance probes increase the signal in the presence of ground or a target by values depending on their position and the additional effect of the object. After more careful consideration, I came to the idea of changing the working procedure of the old unipolar pulse-induction detector. Until now I have two independent GB channels, which I summed after normalizing the signal ( to not use logical OR summation).
Instead, I think to use only one GB channel, which will operate with all signals - targeted and ground-balanced (more precisely, with the sum of all GB samples). But, in order not to lose sensitivity, especially of long-constant targets, after making GB balance (according to the classic GB procedure), I will "limit" the values obtained from the GB samples as maximum. So that during normal working operation with the GB turned on, the values (which are actually subtracts from the useful target signals) do not increase above this level under any conditions (up to this level they must exist, because they are obtained from the balance of real earth/hot stone, etc.) with no limit of target samples. Thus, I think, I will solve the problem with the targets that are after the time position of the ground balance samples.
On the other hand, this will greatly simplify the process of setting the ground balance (I will only have to work with one channel and several target and ground balance samples).
Am I wrong somewhere?

EDIT: This should improve the sensitivity of the detector overall

[boilcoil]

Hello guys.
I changed the code, but with a slightly softer version - I "locked" only the two earliest ground balance samples (I use four ground balance samples and four target samples) and left the other two free (so that slight fluctuations above the ground balance levels can be accommodated). I made several "table" experiments with different targets.
As results:
- there is no change at all for targets with a small time constant - foil, gold earrings.
- there seems to be a slight improvement in detecting different types of coins (but it is not what I expected)
- there is an improvement for massive conductive objects - a silver-plated aluminum sugar bowl and a copper plate.

I don't think the benefits are what I expected, so I might go back to the old ground balance routine, or continue to think about how to improve it.

Regards