Hi All,

I have started a personal project to try to design a 21st-century metal detector for myself and I am hoping to get ideas, guidance and suggestions from knowledgeable members of this forum. Looking at high-end MDs on the market, you are looking to spend north of $7000 to $8000 on one!

I have never used an MD or designed one in the past but, I know the operating principles and for the last month or so I have dug up a lot of stuff, read docs and some patents and watched many videos, particularly useful videos from https://www.youtube.com/@DetectorMods.

I have also built the HammerHead version C2 MD by Carl Moreland on a breadboard. Nocked up a double D coil and found out that using just one of the D coils works best and unexpectedly found that a 2p coin can be detected at about 25cm, 20cm seems to be a reference performance, so I was pleasantly surprised. I have now made the PCB for rev D1 and am about to populate that so that this can be used as a reference to experiment with and make more advanced modifications. This design is from early to mid-2000, a long time ago, we now have much better opamps, instrumentation amps and microcontrollers with DSP capability, if you look at the circuit diagram of HH MD, you will see split power supplies which are not very efficient, I would like to possibly go with a single voltage.

HH MD circuit: hammerhead/HHd1.pdf

So I am looking for ideas and guidance to try the following:

• The receiver preamp currently uses an old opamp
• an MCU can replace many of the components and control timing, and various functions
• Looking to embed the preamp and possibly an A/D convertor in the search head as pro MD manufacturers are now doing
• I am surfing for low noise single rail opamp, with low drift, opamp
• Wondering if I can configure and use an instrumentation amplifier instead of an opamp, this may help to better deal with the dreaded ground mineralization effects

For now, this is my starting point, I am sure other things will pop up and I will update as I go along.
Massive amplification is not the way to go, noise and interference are an issue, and good discrimination/signal processing is a must.

Looking forward to your input, suggestions and guidance.

"This time next year we will be millionaires" This statement will only make sense to the UK members who know the comedy show

 

The question to consider is "what properties of metal do you want to detect?" Ferrous metals are simpler because of ferromagnetic properties, but they are usually less valuable. Non-ferrous metals would be detected by their different conductivity relative to the surrounding material. That is quite a bit different.
So understanding the mechanism of the different detectors will be the first order of business. Hopefully others know more about that.

 

Of course, for a true 21st century instrument you'll have to look at computational methods for processing the signals you receive. Possibly using Machine Learning. There are a lot of edge AI offerings now, even things like TensorFlow for MCUs.

As in other cases, a lot of the analog front end work of the detector could probably be replaced computationally which offers incremental improvement opportuntiies without hardware revisions.

It seems as though SDR would be a possible enhancement but I'm not certain where it fits in. Perhaps the frequency agility of SDR would offer new opportunities for discrimination of targets and localization.

 

The question to consider is "what properties of metal do you want to detect?" Ferrous metals are simpler because of ferromagnetic properties, but they are usually less valuable. Non-ferrous metals would be detected by their different conductivity relative to the surrounding material. That is quite a bit different.
So understanding the mechanism of the different detectors will be the first order of business. Hopefully others know more about that.

Generally, MD have a discriminator setting amongst many other controls to discriminate or ignore various metals, so its part of the settings and configuration of the MD. Most half-decent (above kids toys level) can do that.

Of course, for a true 21st century instrument you'll have to look at computational methods for processing the signals you receive. Possibly using Machine Learning. There are a lot of edge AI offerings now, even things like TensorFlow for MCUs.

As in other cases, a lot of the analog front end work of the detector could probably be replaced computationally which offers incremental improvement opportuntiies without hardware revisions.

It seems as though SDR would be a possible enhancement but I'm not certain where it fits in. Perhaps the frequency agility of SDR would offer new opportunities for discrimination of targets and localization.

Machine learning sounds interesting, though I have not had any experience with it. As long as it is quick and can process the data during a single swing of the MD (from side to side), then it could be very useful.

Not quite sure what you mean with the analogue frontend, it still needs to be an analogue amplifier to bring the signal to a usable level where it can be A/D and processed. Pass that, everything can be digital I guess.

As for the SDR (love them) I would imagine a bit too much digital noise, general external and internal noise as well as the ground effect seems to be a major issue/concern. Some new detectors have started to use multi-frequency scans, about 4 or 5 discreet frequencies which I would like to also experiment with but changing freq. is simple when its only a handful of spots. The more I read and watch, the most important parts of a detector are the search coil and its construction and then the analogue frontend and everyone is on a mission to reduce all digital noise from MCUs and PSUs.

 

Detecting metal is the easy part. Discriminating as to which metal is the hard part. Most high-end metal detectors discriminate between iron, nickel, silver, gold, and aluminum. Which requires quite a bit of processing and if I remember correctly comb filtering and have the ability to eliminate unwanted signals as well as "tuning and nulling" the coil for soil-based metals. And even the high-end ones do not correctly discriminate all the time...

 

Detecting metal is the easy part. Discriminating as to which metal is the hard part. Most high-end metal detectors discriminate between iron, nickel, silver, gold, and aluminum. Which requires quite a bit of processing and if I remember correctly comb filtering and have the ability to eliminate unwanted signals as well as "tuning and nulling" the coil for soil-based metals. And even the high-end ones do not correctly discriminate all the time...

I see the mention of "discriminator" circuits without even the smallest hint of what it is that they are doing, what variable they are sensing. So without any hint as to what variation the discriminator acts on, I reject the comment as useless.

 

I did mention comb filtering of frequencies. Exactly how they do it is their proprietary trade secret. Of the several major manufacturers, each apparently does it a bit differently and claims their method is the best.

 

See from 7:30.

 

I tried that yoo toob link explaining discriminator operation. Not sure what language the guy was speaking, I could not understand much of it at all. Fortunately closed captioning worked very well. So now it is clear that it is the other properties, paramagnetic versus diamagnetic that produce different phase shifts. And the conductivity has a big effect as well. Also learned a fair amount about search coils.
So now it is clear that a processor with the right code can provide a great deal more information. So there is the path towards a better scheme in the future.

 

Detecting metal is the easy part. Discriminating as to which metal is the hard part. Most high-end metal detectors discriminate between iron, nickel, silver, gold, and aluminum. Which requires quite a bit of processing and if I remember correctly comb filtering and have the ability to eliminate unwanted signals as well as "tuning and nulling" the coil for soil-based metals. And even the high-end ones do not correctly discriminate all the time...

Yes, @SamR thats essentially what I have found after reading and watching videos. Discriminating is the hardest part but not that hard, MCUs can deal with that. Besides, none of the devices out there discriminate correctly 100% and I am not after 100%, as long as it is done well. Discrimination starts with some kid's detectors all the way to the pros, the level and quality is what varies. There are lots of other factors that affect discrimination and false readings, so its the overall quality that matters and the man behind the detector needs to decide for himself.

@MisterBill2 I am glad you have got the idea of how the phase shift/discrimination works now, I had no idea and had to find out. The current proto circuit uses 3-4 chips to do that, but for an MCU its a piece of cake (almost). Lets put it this way, non MCU have many variable pots to set and they interact with each other, MCU can make that much easier to deal with.

Ground mineralisation is an issue that is generally dealt with by the double D coil, they seem to be able to deal with that much better than other configurations, and there are many. The ground effect generates noise and false signals. My starting point is double D search head for now.

Preamp (search coil amp) noise, offset, etc needs to be controlled with a good preamp design and I thought, as an SDR, just A/D the signal at the coil and feed it up to the processor further up in the handle of the detector to pull anything I want out of it (signal processing). Search head design is also a major key, graphite paint and quadrant grounding of the paint on the search head seem to deal with a lot of the ground noise - Detector Mods videos are full of hidden information and knowledge gained by Woody over many years which he is kindly making available, very useful.

 

I am thinking that using multiple frequencies can also provide useful information, the challenge being howto do it while areas are being scanned. Clearly the system must be constantly watching, and so sequential searching is not available.
UNLESS the MCU has the option of saying "Hey, wait, go back and scan that again, I think I saw something." An attentive operator may already do that based on some response, many would miss it.

 

I am thinking that using multiple frequencies can also provide useful information, the challenge being howto do it while areas are being scanned. Clearly the system must be constantly watching, and so sequential searching is not available.
UNLESS the MCU has the option of saying "Hey, wait, go back and scan that again, I think I saw something." An attentive operator may already do that based on some response, many would miss it.

One of the docs I read stated that some new detectors use multi-frequency however, there are something like 5 discreet spot frequencies, so not a massive task. I cant see why sweeping from side to side cant be done whilst say 5 spot frequencies are sequentially scanned. Discrimination is spread in maybe say 1ms, after every pulse. My current test bed the HH has a 2.8KHz top frequency, pulse width also is controllable via one of the pots.

So, an MCU can do spot Freq x number of times, followed by discriminate, then change freq. and repeat the whole cycle that's doable whilst sweeping side to side isn't it? Unless I am missing why you think we cant scan?

 

Ground mineralisation is an issue that is generally dealt with

Typically handled with a pot adjustment to null the response out. In extreme cases an extra filter is switched in to further null the response. Frequency sweeping has merit but each frequency would have to be nulled for mineralization. They have come a long way from the simple BFO designs of long ago. Coil size is also an issue. Big coils for large area coverage and small coils for more precise pinpointing in target rich environments. Think in terms of ground penetrating sonar and I'm bit surprised they haven't come up with waterfall displays but then most operators... There are some circuits available on the net for very old White designs from long ago. Most will never publish their schematics due to overseas copyright infringement. Same with most US depth finder manufacturers for the same reasons.

 

Yes pots and interactions etc. They have come a long way, even my proto from 2006 I think has an automatic mode that sets itself up for cancelling the ground effect, although I havent tested it yet as I am about to construct it on its PCB so I can do proper testing.
Big coils penetrate deeper but can miss smaller targets. Smaller coils for better resolution. My plan is to optimise both the hardware and the search coil design.
Its all for fun though, its a challenge when I see detectors upwards of $8000, I wonder why is that and then comes the challenge! Yes, you pay for R&D etc I know...

 

Hi.
If of any contribution value; I tried to come up with a wide row of sensing kits dragged trough the soil, perhaps a couple of metres, each one has a led and buzzer as indicator of metal presence. I have not continued nor finished the project. The intention is to cover a wide strip of terrain at once, stopping when chirps and checking which led is on. Got a few kits and was planning to make it like 30+ of them.
Perhaps not for the 21st century, but hey, just an idea... The main purpose was going after meteorites, not precious metals.

 

What comes to mind now is mine detectors, given how many thousands have been planted by the invading Russian army . But probably they are all plastic and mineral so as to be harder to detect. (Off topic I suppose)
Quite possibly some frequency would be affected by changes in density in mineralized soil, and that might be useful in treasure hunting.. So it might be that the new design could support a number of schemes.

 

@Externet Thanks, sounds like fun. Yes have seen draggers behind quad bikes large and small, different game as you say. Some of those meteorites are more precious than the usual stuff. Did you ever test the penetration?

@MisterBill2 Funny you should say that, I came across this a few weeks ago, and skimmed over it, it is an excellent informative document, with lots of information and theory in its 243 pages!

 

I read a portion of that paper, 243 pages being quite a lot. One very interesting thing is that the detectors referenced in the paper seemed to be quite different from the "Double D" arrangement in that U-tube video. So perhaps a combination system could be even more effective.
Or else, as one suggested to me a whole lot of years ago: First chase a large herd of wildebeast across the area, and then follow up with a herd of water buffalo.
Aside from the obvious logistics issues is the very real concern that even in a stampede, animals will tend to avoid some natural obstacles, leaving those areas untested.

 

Yes, a lot of pages. The idea is to have hardware that can take different configurations of search coils. That's already being done by the pro detectors out there.
Any guesses on whether an instrumentation amplifier would have an advantage over opamp?