Close

Possible Improvements

A project log for Junk Box Metal Detector

Can I build a working metal detector from what's in my junk box?

zaphodzaphod 09/01/2021 at 22:040 Comments

As mentioned in the last post, I have some ideas about improvements to the circuit. I haven't tried any of these out (and won't be able to in the immediate future since school's starting again), I wanted to write them down anyway.


I think that there are three areas for possible improvement:

  1. better architecture
  2. better electromagnetics
  3. better electronics

Better Architecture:

I'm not going to go to deep here, basically BFO detectors aren't a super good way of making a metal detector, and if I was going to build a metal detector 'for real' I'd order parts so that I could build a different architecture. @mircemk has a number of metal detector projects posted that use non-BFO architectures, and seem to achieve better results than what I was able to get. If you need a better metal detector start looking there.

Better Electromagnetics:

I didn't try to optimize the coil design[1] at all while I was building. I think if one knew more about coil design one could probably come up with a better design, but I don't so there's not much for me to comment on here.

([1]  I did try to get the master oscillator frequency as low as possible, since it felt like a lower frequency should 'go farther into the ground' but I wasn't able to get the colpitts oscillator working at frequencies below ~120 kHz with the parts that I had on hand, and I don't actually know if lower freq is better.)

Better Electronics:

that leaves only one area for improvement on which I can comment. As a refresher; here's the circuit I went with:

The sense oscillator and BFO designs actually both worked reasonably well in isolation. The only issue is that when the 555 timer shares power rails with the colpitts oscillator the two oscillators tend to couple when the sense oscillator drifts to a frequency near the beat frequency. This is a problem since it means that minimum difference frequency generated by the mixer is limited to how close you can bring the two oscillators without them coupling. In practice this means that the operator hears a pretty high pitched whistle, small variations in which are hard to discern. I think if you could get the oscillator frequencies closer (thereby making the whistle pitch lower) it would be easier to discern small variation, essentially making the detector 'more sensitive'.

I'm not entirely sure how one would go about decoupling the two circuits however. Decoupling caps on the 555 certainly help, but the circuits still couple a bit. I think that moving the circuits farther apart, and possibly trying to separate ground connections/planes might improve coupling a little as well.

Barring better layout I think you could perhaps just run the two oscillators on different batteries, but then getting the sense signal across to the other power domain could be a bit difficult...

The other obvious area for improvement is the FET mixer. I didn't really design this circuit, so much as just stick stuff into a breadboard until things worked enough. Eventually I did end up spending an afternoon messing around in LTspice, and came up with the following as an improvement for the mixer:

the improved mixer design in on the left, and the original mixer design is on the right. (original design uses some guesstimated values for the biasing arrangement)

A quick FFT gives us an idea of how poorly my initial design works:

a shot of the full spectrum. green is the new mixer, and blue is the old mixer. Both suck so there are significant components at 135.5kHz and 136kHz (BFO and sense respectively), but that's what filters are for. we only really care about the low end
here's the low end. At ~500Hz we see the peak that we care about (136kHz - 135.5kHz = 500Hz). As you can see the green line is higher than the blue line. LTspice thinks that its like 15dB higher, but I don't know how far I'd trust the exact number

Basically it seems like the newer mixer design could provide a higher level output, which would likely result in the operator being more able to discern a small change in sense frequency.

(Note: I guessed at the biasing values for the old FET mixer (the voltage divider is a trimpot in the real thing, and I didn't measure it to see what the final position was for the simulation), so its possible that the old circuit might actually perform better than what's shown here, but I don't think that its doing that much better)

Finally I think that a better amplification stage is in order. Specifically the signal from the sense oscillator was attenuated before feeding into the mixer in order to stop the amp from making a horrific screeching when hooked up to the earphones. If I'd been less lazy I think I could have redone the amp circuit to have a gain lower than the default (20V/V), and then done away with the attenuation, which would probably also help with overall sensitivity.

Its also possible that the better mixer stage might not need an amplifier at all, but that would require some real world testing. If that were the case though you could see a pretty spectacular reduction in the BoM magnitude, which is always welcome.

OK, that's the end of my ramblings, and I think I can declare this project completed now.

Discussions