Tweet
Following the "simple" detectors that are popping up on the forums, here's a simple low power method of (yet another) BFO detector, with a bit of "twist" which I believe hasn't been shown previously. It uses a digital method of frequency difference multiplication to take advantage of a lower search oscillator frequency, a hint of discrimination, low power draw and tuning by a pushbutton. The last two features are necessary to reduce frustration as I designed it to be a kids toy 
From what I've measured, freq shift based discrimination works less well at higher frequencies (300-800kHz) where some conductive targets start to behave like ferrites and vice versa. Ground penetration is also horrible at 800..400kHz compared to 100..50kHz. The disadvantage of running a lower frequency is a lower frequency difference, but the digital gain will help restore that lost sensitivity. It's also non-motion, it would be relatively simple to make it a motion detector with some software changes.
There are some discrete digital implementations of harmonic mixers for multiplying the frequency difference, but they usually result in a less "clicky" triangle wave output, lack self tuning, and modern programmable complex devices are already available for about the price of separate discretes and board area.
It works in the following way:
-A simple colpitts search oscillator is hooked to a comparator input of a 20MHz clocked AVR microcontroller. The comparator input gives the oscillator's cycle time in microcontroller clocks (ie. 20MHz xtal clock, 100kHz oscillator = 200 mcu clocks).
-To tune the detector, 65536 cycle time samples are taken and summed together, then divided by the same number (shifted 16 bits toward lsb) to make a "fixed point" tuning value - effectively the fractional average of all 65536 samples. There will be a slight error due to fractional phase difference between the start and end of tuning measurement.
-A 24 or 32 bit phase accu variable is then set up, fixed point - meaning the less significant 16 bits are the fractional part, and the more significant bits are the integer. For each following oscillator cycle, the cycle time is added into the phase accumulator at the integer point, and the tuning value is subtracted from the phase accumulator.
-The phase accumulator will start growing an error that's either positive or negative depending on the frequency shift. If we sound a click whenever it grows as large as the tuning value, we will output the frequency difference in clicks. We can also get digital gain from the system by sounding a click at lower differences - we can get 16x the frequency difference if we sound a click every 1/16th of the difference value.
-If we also make note of positive or negative frequency shift, we can give some crude target identification for conductive (rising freq) and ferrous (falling freq) targets.
I imagine this method to be applicable to just about any simple cheap microcontroller. The micro could also give battery voltage indication, add a degree of continuous auto-tuning (currently experimenting with this one), and maybe even give an indication of the search osc frequency to help in setting up the search coil for those building the thing as a project with just a pre-programmed microcontroller.
Here's a video from bench-testing. 8" kitchen foil shielded coil, 33 turns of blasting wire, 100kHz jfet colpitts-oscillator. Pardon the low volume and poor lighting, but the response should be noticeable and the distances possible to gauge by eye. Absolute detection distance is hard to gauge, since the output is a multiplied frequency shift, and it hasn't been ground tested yet because of frozen earth.
http://www.youtube.com/watch?v=ymLzpCrZyuk
First test object is a beer can (emptied), second is a key bundle, third is 1euro coin, and fourth is a key rotated to show changes in response compared to target angle.
One good point is that the current draw for this apparatus is around 15..20mA using a piezo speaker and bicolor power-led/target-identification, and a 9V battery would last a long time. Further power reduction could be had by sleeping the microcontroller between calculations. I haven't had the chance to do ground testing since it's frozen, but I will once spring arrives over here.
When I have the time, I will clean up and post my schematic and asm source over here. It would be interesting to hear if that principle has been used previously or if someone can make use of it. This would probably work as a pinpointer as well, like other simple low power detector projects.
From what I've measured, freq shift based discrimination works less well at higher frequencies (300-800kHz) where some conductive targets start to behave like ferrites and vice versa. Ground penetration is also horrible at 800..400kHz compared to 100..50kHz. The disadvantage of running a lower frequency is a lower frequency difference, but the digital gain will help restore that lost sensitivity. It's also non-motion, it would be relatively simple to make it a motion detector with some software changes.
There are some discrete digital implementations of harmonic mixers for multiplying the frequency difference, but they usually result in a less "clicky" triangle wave output, lack self tuning, and modern programmable complex devices are already available for about the price of separate discretes and board area.
It works in the following way:
-A simple colpitts search oscillator is hooked to a comparator input of a 20MHz clocked AVR microcontroller. The comparator input gives the oscillator's cycle time in microcontroller clocks (ie. 20MHz xtal clock, 100kHz oscillator = 200 mcu clocks).
-To tune the detector, 65536 cycle time samples are taken and summed together, then divided by the same number (shifted 16 bits toward lsb) to make a "fixed point" tuning value - effectively the fractional average of all 65536 samples. There will be a slight error due to fractional phase difference between the start and end of tuning measurement.
-A 24 or 32 bit phase accu variable is then set up, fixed point - meaning the less significant 16 bits are the fractional part, and the more significant bits are the integer. For each following oscillator cycle, the cycle time is added into the phase accumulator at the integer point, and the tuning value is subtracted from the phase accumulator.
-The phase accumulator will start growing an error that's either positive or negative depending on the frequency shift. If we sound a click whenever it grows as large as the tuning value, we will output the frequency difference in clicks. We can also get digital gain from the system by sounding a click at lower differences - we can get 16x the frequency difference if we sound a click every 1/16th of the difference value.
-If we also make note of positive or negative frequency shift, we can give some crude target identification for conductive (rising freq) and ferrous (falling freq) targets.
I imagine this method to be applicable to just about any simple cheap microcontroller. The micro could also give battery voltage indication, add a degree of continuous auto-tuning (currently experimenting with this one), and maybe even give an indication of the search osc frequency to help in setting up the search coil for those building the thing as a project with just a pre-programmed microcontroller.
Here's a video from bench-testing. 8" kitchen foil shielded coil, 33 turns of blasting wire, 100kHz jfet colpitts-oscillator. Pardon the low volume and poor lighting, but the response should be noticeable and the distances possible to gauge by eye. Absolute detection distance is hard to gauge, since the output is a multiplied frequency shift, and it hasn't been ground tested yet because of frozen earth.
http://www.youtube.com/watch?v=ymLzpCrZyuk
First test object is a beer can (emptied), second is a key bundle, third is 1euro coin, and fourth is a key rotated to show changes in response compared to target angle.
One good point is that the current draw for this apparatus is around 15..20mA using a piezo speaker and bicolor power-led/target-identification, and a 9V battery would last a long time. Further power reduction could be had by sleeping the microcontroller between calculations. I haven't had the chance to do ground testing since it's frozen, but I will once spring arrives over here.
When I have the time, I will clean up and post my schematic and asm source over here. It would be interesting to hear if that principle has been used previously or if someone can make use of it. This would probably work as a pinpointer as well, like other simple low power detector projects.
I bet any schematic from the CMOS cookbook has a handful of separate digital and analog ICs.
Comment