My AGD analog detector work

[Auto-Mation-Assist]

Yesterday evening I started making some changes to my PCB layout for the AGD26.1 version and started thinking about the effects of circuit board capacitance which can be difficult to simulate with what I have to work with. With that in mind I took another look at my circuit board and my coil damping circuit. As a result I changed the active clamping circuit a bit and gained some performance improvements and also improved stability. For the simulation I also added the method I will use to be able to adjust damping for both mono and DD coils. Using active damping is like building a difference amplifier which requires many very precision parts to have the ability to reach even a -60db null point when both its inputs a equal in every way. When dealing with getting a nice null, or at least a reasonable correction of coil decay issues in a PI metal detector the it should be understood that there are to many variables beyond our control and that we can only reach a reasonable compromise at best.

The simulations that I have done using various methods over the past years all indicate that it is possible to reach beneficial control of the coils decay slope of the critical knee point before the signal ever reaches the first receive gain stage. This allows receive sampling to start much earlier without running into gain stage saturation. In order to get the maximum performance increase careful construction techniques and high quantity parts and must be used. Results can be disappointing even when small things deemed not to be important are overlooked.


The circuitry related to this post in in the blue box in the below diagram:



The output signal of the THS4631D wide band OpAmp drives the fixed resistive load R9 via R17 and thus modifies the decay curve at the proper time.

Adjustment of the decay is done via gain controls mono coil or DD coil via diodes D3A and D5B. Resistors R4 and R7 provide some positive feedback to help improve early response time. The fixed resistive load is R9, R2, R37, R39 and are selected like normal damping along with the series total value R13, R12 and R10. If resistor values need changing I suggest keeping the ratio of R12 to R10 the same. The design value of both of these together was 11K. R16 is used to equalize input offset currents and also to aid in controlling overshoot and is part of gain control for R5.

The below picture shows to output of the THS4631D wide band OpAmp in green, and in red the lower point of the control correction diodes D3A and D5B which do the work. The yellow marker is the point at which decay cay point reached the 2mv level on this and other pictures.



The below show the decayed curve generated by the wave forms of the picture above. It shows the blue marker,1 at a level of 2mv up from zero at 2.564us after the end of the 36us TX pulse. About 730V kickback. The yellow marker is at 3.8us after is is not possible to detect very small gold at 3/4 inch based on my prior standard tests. In looking at the time between the two markers it shows that there is 1.236us available for fast sampling, starting at 2.564us after the TX pulse ends. Now determine how long it takes the signal to reach the first sampling gate an how long it takes that gate to close and open to determine gate pulse timing. The AGD26.1 determines its start time from when the decay curve lowers to around 10mv and adds a delay amount of nano seconds to determine first sampling gate closing.



Test inject signal recovery.

The vertical scale in the picture below is 5uv per division. As is my prior posts I used the same 1Mhz test signal to get see if signal loss due to active damping is high ot not since it provides a load at the decay curves knee point. So far it has not show to be a issue. The peak to peak voltage level was 10.995uV which is fine. The red trace is the output of the THS4631D wide band OpAmp. Its scale is 2 volts per division and shows that its output signal is at or close to zero volts.



One thing to consider is that, this is all about cancelling out parts of a signal because it is not actually correct with another signal that was generated from another source and hope that they are equal but in opposing polarity at the right time. When forced to merge together at the correct amplitude and phase we can end up with a signal that is mostly correct down to very low signal levels. If done reasonably well it works fine, but it does not take a lot to make it go totally out of balance.

Things to consider:

PCB trace capacitance presents the biggest problem since this affects phasing and can cause a wavy wave shape. Next would likely be temperature tracking between the two sets of diodes.

I think its very difficult hard to get it 100% percent perfect especially at the tens of nano seconds that all this needs to happen at.

I will see what happens when I build up a actual AGD26.1 PCB-1 circuit board for testing.

End of post

[pito]

Hello pito,
For which project you ask - AGD or AMX? Here the discussion is for AGD project from AMA. The TX circuit is well described in the previous posts.

Auto-Mation-Assist
claims that the his coil voltage is 500V, I want to see that my self.

[Auto-Mation-Assist]

These are pictures of the prototype AGD26.0 version development circuit boards that I had ordered last month. PCB-1 is almost like a complete front end, while PCB-2 is mostly for demodulating the signal from PCB-1 and handle required signal handling. It could be replaced with a board that does that using digital technology and still use the same front end board.

Prototype AGD26.0-PCB-1





Prototype AGD26.0-PCB-2



AGD26.0-PCB-2 has been built up and the analog sections tested but it needs changes that handles routing of the channels into the ADC converters. Not sure on what I want to do there yet.

Also not sure if I really need the difference channel that is centered at 1.024 volts. This channel has much higher bandwidth of the fast and slow channels and was planned to be used for digital processing to help with ferrous and none-ferrous separation.

Both of these boards were developed to use my existing enclosures. The mounting holes were moved just a bit farther inward from the outside edge as compared to my older boards. This makes it possible to clear a ridge inside the enclosures that the end cover screw use and can eliminate the internal mounting shelf that I have used to mount the boards to. But not having the mounting self makes it more difficult to install it as a unit in the enclosure and also more difficult to service.

Both these circuit boards will see changes. I will likely not populate AGD26.0- PCB-1 on the left in the picture and wait to complete the update to AGD26.1-PCB1 once the board is reconfigured for the updated version of the coil decay control.

End of post

[AhmatAhmatov1959]

I have built a good number of these detectors and they all worked fine for detecting very small gold. I use the dig all method but usually disregard real loud return signals that are likely not what I'm looking for. One of the problems I ran into using the CMOS VCO and running it a idle frequency of 300Hz was that any substantial detection voltage of the incorrect polarity could drive the VCO frequency down so low that it could not be heard or stop outputting a tone. To eliminate this a resistor was added and installed that prevented this from occurring. This addition appears to have produced a issue when using the detector design in a way that I normally do not use.

If you look at any of the old circuit diagrams that I posted you will find the resistor to the far right of DSC connector (Discrimination Control), It was a 100K, 1206 size resistor in those early versions. The resistor could be R23, R24,or R97 depending on development version. This resistor needs to be removed to be able to properly use the DSC adjustment control on the front panel to null or reduce FE high level signals. I have verified this in my own AGD24.2 version.

Present work in progress:

I have been busy working getting 2026 version designed and have both its initial circuit boards lay out completed and verified but holding on in case I overlooked something. The 2026 version will offer a lot more flexibly due its build in STM32 processor. The STM32 C code that I worked on for the 2025.1 version can be used to allow for initial checkout after re-assigning some processor pins in the code. I hope to have the initial circuit boards here by start of December and assembly starting several weeks later if all parts are on hand. I have not yet decided on a PCB manufacturer.

Hello Auto-Mation-Assist. Could you please explain what the DSC is on page 5 of the AGD23.4 schematic? If it's a switch, it's not on the front panel. Thank you for your reply. Sorry for my poor English.

[Auto-Mation-Assist]

ahmatahmatov1959, The DSC connector routes to the front panel Ratio control and a Normal / Invert switch for channel 2. The Ratio control allows for mixing channel 1 and channel 2 signals for further processing. Channel 1 is the fast or leary channel while channel 2 is the following slower and wider channel. Pin 1 is the channel 1 front end output, pin 2 in the channel 2 front end output, pin three is the inverted channel 1 output, and pin 4 is the wiper output.

[Auto-Mation-Assist]

Last month I assembled my proposed circuit board for my AGD26 detector and ran into some issues.

1. Wrong foot print for 5V voltage regulator caused input and ground to be reversed due to a different letter code in the parts suffix.
2. A fast type D flip flop that would only switch its its Q not output and never the Q. Replacing the part did not cure that problem. 74LVCG74DP and works fine in simulations but not in circuit.
3. Missing pull down resistor on the TX pulse 3.3V that comes form the MPU. Not needed if the two circuit boards are plugged together.

Every thing else on the circuit board checked out fine.

I took a picture of the analog front end board for the AGD26.




This circuit board is designed to to function as the analog from end of the AGD26 and routes its two X channel outputs to 16 bit analog to digital converters controlled by the STM32G474VET6 100pin MPU. The MPU will also handle timing, additional analog to digital conversion and waveform generation. Timing will be divided between slow for the master period and TX pulse and a fast for gating and critical timing. Most of the required MPU code is on paper now and just needs to assembled and linked into a fully working package sometime after its processor board is layout is finalized from its preset 64 pin MPU to the 100 Pin version.

I'm thinking about not building a speaker into the enclosure for this version and instead using the broad space for wireless headset driver. I have not done any research on the best option for this yet.

[Auto-Mation-Assist]

One of the things I have been considering is what kind of coil damping I should use. I have the choice of more complex active damping or using just passive components that can also work fairly well. There are a number of ways to interface a coil that can reach a near a 1000volt level to a pre-amp that has sensitivity down to around a micro volt. Every method used to accomplish this has its own drawbacks but can be made to work and all will have some form of brute force clamping. Lets analyze what kind of parts have the least built in capacitance and have the most linear response do not require any power source when used as series elements in clamping networks?. Ok so we have our chosen our part and it has the absolute minimal capacitance feed through. This is good. Now we need out brute force clamping device that responds fast and has no slow decaying memory. Not a lot of real good choices here due to the requirements. We need fairly high pulse current ability, low clamping voltage and very high reliability. We really want clamping at a steady level of 50mV or less which is not obtainable with ready available parts and thus we are going to over drive most all front end pre-amps. and likely exceed its differential input limits. When this happens we need front ends that recover from overloads faster that a microsecond. A best thing to use for a fast clamp is likely shottky diodes that are designed for UHF use. Two in parallel with opposite polarity have approximately 1.845pf of capacitance which in many cases is lower that the capacitance of a PCB board traces used to wire them in.

In my AGD26 type detector I have the option all known methods for isolation the coils high voltage recoil from the RX pre-amp. But I like the simple resistive rework and brute force clamp the best since it uses mostly linear parts that require no power or switching during the critical receive time. In my prior post you may spot a good number of black resistors used for coil damping. Each one is rated at three watts and they warm up nicely with a driver Mosfet rated at 1200 volts and the isolation diode at 950 volts. This circuit board has active coil damping via the parts in the lower left corner just under the Mono - DD switch. This damping works but has to be carefully tuned to so that a proper ratio of fixed damping resistance and the adjustments range of the active part is proper. It can work well but could I replace it with something simpler?

I think I can but is a different way. The main function is to eliminate the large spike that typically occurs when the brute force clamping diodes stop conducting when the coil decays to a near zero volts point. Active damping attacks this problem by modification of the coil damping during that period of time, and which will also lower the apparent receive signal level slightly. The spike I'm talking about can vary from 100mv to 20mv depending on how the resistive coil damping load is adjusted. And if this spike is present it will usually cause the decay length of time to increase by about 1.5us more than it needs to be. I show this in the picture below.




This picture shows the voltage across the coils damping resistor on channel 1 (red) and the pre-amp output in green using my typical
clamping network. The test signal wavery lines is the 1Mhz signal that I use to verify that it will pass a receive signal with minimal loss
and at which point it becomes usable. This is at about 3.8us after my TX pulse ends and just barely meets my absolute maximum.
I want to get rid of that sharp dip without loosing to much if any of 1Mhz test signal and move my maximum time hopefully to the left.

Here is my picture for my proposed method of doing this with passive components and not changing the values of my fixed damping
resistors.


This picture relates to the supplied diagram later in this post - The value of C2 is 100pf.



As you can see the output for my pre-amp output has improved enough to have more time for signal processing while the coil in red,
remains as before. This change was made by using just a slight of automatic current following bias for the clamping diodes.
Circuit diagram follows.




In the picture that I showed before this diagram capacitor C2 was at 50% or 100pf. This capacitor does all the work based on the voltage present during the decay across R19. Next is a picture with C2 at Max (200pf).


With C2 at Max 200pf.



With C2 at Min 0pf.



I have not done to much work on this to see what the limits are by readjusting my fixed value damping resistors but the simulation does
show potential for reducing the decay time substantially without to much complexity. Perhaps the left shift can be enough to help some
obtain a bit more detect sensitivity and not being bothered so much with the large spike overloading intermediate gain stages.

[ivconic]

Sheeesh!
Your ADG still not working!?
I'll be back in 3-4 years to check the progress!

[Auto-Mation-Assist]

In completing the design of the AGD26 version of my detector I have found many ways to accomplish the same thing, some work well and some not so well. None have been what I would consider to be the very best option.Since my earlier versions work fine for me there is no need to finish my latest version to rapidly. Thus I take my time to do other things as well.

Modified circuitry using different components still has to do the same basic job. In a PI metal detector it is to detect a change in a self generated waveform which to some degree can be changed by a external source when the user supplies motion.

Lets all have a look at some recent simulations I have done to finalize the front end stage of the AGD26 after working out a number of other versions.


In the picture below each vertical division is one millivolt and the horizontal is one microsecond per division.

The peak coil voltage was 754 volts which decay cleanly to zero volts in 2.2us and into a flat line. The decay is thus fully controlled like a balancing act at its zero point and at the same time stays stable there. This makes the area around the zero volt point very sensitive to changes. Thus the coil that generated this wave form also becomes more sensitive as compared to sloppy decay control which causes a much more rounded profile in the knee area of the decay waveform..

The red trace is the output of the out put of the AGD26 17dB gain pre-amp which, at the same time is also used to control the decay waveform. The pre-amp thus serves two functions, provide enough gain to set the systems noise floor and also to controls the coils decay time to the minimum amount of time possible while at the same time assuring that the desired part of the wave form needed for detection maintains high fidelity. When compared no my very early versions of the AGD detector this allowed start of signal detection to occur 1.6us earlier.



One of the test that is of value is to inject a tests signal into the coil. Here is the result of a 2Mhz signal being added. It demonstrates that the zero cross over point point remains stable. Part of a test signal cycle can be seen modifying the decay waveform on the left in the green coil line just before it reaches the zero cross over point which is excellent. Also take notice that the tests waveform is nearly level from left to right.

In the picture below the vertical divisions are 1mv per division and the that the pre-amp output signal levels are a little more than seven times the size on the coils green input. Thus our 17dB expected signal gain is present and our system noise level is also set by the pre-amp. It looks good and thus we can expect excellent performance with maximum sensitivity when we are in the field.



One of the requirements for a good front end is that it be very fast and thus it will have extended bandwidth. Always remember that using non linear components that switch on and off can create lots of inter-mod.
If you have ever listed to a short wave radio and found it was full of whistles while you tune across the band? That is the intermod effect within the receiver since those signals are not really there but generated inside the receiver itself.
If you have a good properly designed receiver those signals will not be there and desired signals would just appear with almost no noise. The same goes for detectors, its worthwhile to pay very close attention to the analog section of any detector.

Just for general info here is an expected front end bandwidth for the AGD26 front end. Down about 2.5dB at 10 MHz. Marker at 450 KHz.



So far I have given some important background info for the AGD26 front end and ready for the important technical part. The basic circuit diagram used for developing the AGD26 front end.




Be aware that the op-amps mus be able to handle the high current that will be routed into their output stage them through R14 during the high voltage decay.

If you what to play with trying this using your own components.

To set this up as a starting point.
Set up the diode clamps with series resistors.
In parallel with this add the a variable resistor between coil and ground.
Adjust the variable resistor so that when pulsing the decay wave form tail appears near the 2us point after the TX pulse ends.
Make the tail 100-20mv below the zero cross over.
Measure variable resistor and for a string like R11, R12 and R13.
Add perhaps 100 ohms.
Adjust the value of R14 until desired suppression of decay tail reached. This will have no effect on pre-amp gain.

The op-amps are current feedback types with "combined" output current rating of over 550ma and high speed. Have a drawback with DC offsets due to high input currents.

Voltage feed back type of amps can be used if output current rating is high but there will be some peaking due to clamping diodes capacitance in feed back loop which is hard to compensate for.
Voltage feedback type amps will work but expect less gain in microseconds and curve less flatness on right side of curve knee.

[Auto-Mation-Assist]

I have completed the circuit board layout for PCB-1 of the AGD26.2 now that the the circuit design of this board was completed. PCB-1 handles most of the analog section and has a design goal of having the RX gain of 60dB or a voltage gain of about x137. The circuit is designed to maintain this gain for normal low levels signals and reduced for high signal levels.
The RX receive front end is designed to obtain its receive signal from either one or two coils. It has two separate preamp sections which can be routed to either or both of the analog channels. The preamp that handles the RX signal from the TX coil also provides for TX coil damping and has circuitry to allow for critical fine damping adjustment. Also provided is a level detector that looks at the TX coil decay waveform and provides a 3V trigger output that will be used to start the receive gating pulses coming from PCB-2 and also provide information on timing changes caused by different types of metal. The standard coil connector wiring used by many coils with a five pin plug is not suitable for a dual coil receive system and thus will be as per the drawing below for the transmit section of the detector.



Associated with the drawing above is how to get control of TX coil damping that gives the best results. For that important job I will use my TX coil receive preamp to get control over the critical TX coil damping point as I have described before. The actual adjustment range to accomplish this is actually very narrow and as such the resistor values to set the adjustment range have to be determined based on circuit board layout and the characteristics of the coil, cabling and other unknowns.Thus expect the values of the parts used to make the damping adjustments to change.

Here is the drawing of the TX coil preamp that is doing the important work.. The TH3092 is a dual high current output -current feedback amplifier. It does not have normal opamp inputs.



I'm waiting for the AGD26.2 circuit boards to arrive about the last week of April and this is what they will look like. Three connectors on the bottom of the circuit board plug into the DEMOD and digital PCB-2. I will be working on updating AGC26.1 PCB -2 to make it compatible with the new AGD26.2 PCB-1.

AGD26.2 PCB-1 layout



It is using the right angle GX16 5 pin connector that mounts directly onto the rear panel. The common mode chokes mount to the back of the circuit board. No wires.

[Auto-Mation-Assist]

The main part of the receive signal chain are the stages following the two front end preamps. This section provides about 43dB of gain for very small signals and reduces that gain for high level signals. For a 10 microvolt signal this section has a voltage gain of about x139 while at 1V volt input the voltage gain is reduced to just under x4. Normal front end gating will prevents such high input but 50mv may not be unusual.

500kHz sine wave voltage gain versus input level, rounded values:

10uV x139, 100uV x139, 1mV x139, 10mV x137, 20mV x122 , 30mV x97, 40mV 78, 50mV x65, 100mV x35, 500mV x8, 1V x4.

Here is my diagram for this circuitry. The output routes to AGD26.2 PCB-2.



As in all my prior designs it also allows for DC offset control from within PCB-1 it self or from the DEMOD section on PCB-2 to maintain a solid signal reference level with microvolt DC offsets. In the AGD26.1 the front end preamp output offsets are set with input current balancing resistors on the positive inputs of the THS3092DDA amplifiers. In the design of the AGD26.2 I decided to make the gain drop off below 1kHz via the addition of some capacitors in the signal and feedback paths. These capacitors can be replace with zero ohm resistors if the low frequency drop off is not desirable.

Bode plot of this two stage amplifier. Minus 1dB at 1.43kHz and 1.24MHz. Marker at 10kHz

[c-zc]

Hello Sir,

The following message was translated using AI.

I’ve carefully reviewed your schematics for version 23.3 and the 23.4 RX board. Both the overall design approach and the level of detail are truly impressive. The system appears to be very rigorously engineered, and it’s clear that you’ve chosen high-performance and advanced components.

I am currently preparing to replicate a similar system, and I’m particularly interested in the coil parameters you are using, as well as the achievable detection depth.

If possible, could you share some details about your coil specifications and the detection performance for larger targets?
For example, what kind of detection distance can be achieved for an object like a smartphone? (It would be very helpful if you could also mention the model, since size and weight vary significantly between devices.)

I also have a question regarding metal identification:
Does your system provide any form of metal discrimination? I noticed that the sampling window in your design is adjustable, which makes me think there may be some potential in that area.

Additionally, I’ve recently had a batch of coils manufactured (30 units), with the following specifications:
Inductance: approximately 700–730 µH
Outer diameter: ~90 mm
Inner diameter: ~70 mm

These coils are highly sensitive to small targets and are machine-wound, ensuring very consistent winding quality.

If you’re interested, feel free to leave your shipping address — I can send some samples from China for your testing or evaluation.

Looking forward to your reply.