VLF MD with digital signal processing : Bee-Buzz 1

[Detectorist#1]

You now design real working VLF metal detector with real tests or just ultra low noise amplifier?

[Aziz]

You now design real working VLF metal detector with real tests or just ultra low noise amplifier?

Real tests. Digital decoding: FFT, Goertzel and Lockin-Amplifier. Just for fun and quriosity.
Without preamp and with preamp.

Forget Rocket-Science, this is Oreshnik-Technology.

Aziz

[moodz]

Just reading the data sheets for a "cheap" ne5534 shows the source resistance will have more effect on noise than your wallet will. LOL
( and it is random noise anyway .. non synchronous )

zz0.ntdhu6imfqzz

[Aziz]

Hi Moodz,

it would be interesting to know, why Fisher made such an effort on the input stage. I do not have any idea.
Cheers
Aziz

[moodz]

Hi Aziz ... there may or may not be a reason. Maybe it was a "bluebird" .. some designers choose to overdesign to maybe counteract a percieved deficiency somewhere else. VLF machines are fundamentally frequency domain and use ( or should use ) synchronous demodulation. If the mixers used are "leaky" ... ie the mixers are not balanced and port to port isolation is poor then low frequency noise at the input will impact the target frequency bandwidth. So if you lower the noise at the frontend you can offset this effect ... however you should really look at getting a better mixer as the demod.

Most schemes I see use chopper switches like the 4066 or 4053 in single ended mixer configuration ... these are not really ideal if high performance mixing is required.

Look up Gilbert Cell for ( you can get a chip NE602 NE612 I think ) ... they can achieve balanced mixing ... theres lots of schemes though even using 4066 in some circuits.

Of course with high spec ADCs this is all a moot point as nowadays the mixing can be done in the DSP and achieve near perfect results ( depending on the bit resolution and noise floor of your ADC ).

[Aziz]

Thanks Paul.
If I find my other preamp bread board with the NE5534, the comparison would be interesting too.
I'm sure, there isn't any remarkable difference.

I tend to upgrade my experimental detector software, just to show the simulated mixer outputs (time domain and frequency domain).
One can see, how the synchronous demodulator works in digital wourld.
Aziz

[Carl-NC]

The Gold Bug was designed by David Johnson. Most likely he thought he could design a better discrete opamp than what was available. The same thing was used in the CZ designs. David was insistent that the preamp needs to have ultra-low flicker noise even though a balanced demodulator cancels flicker noise. I suspect you can replace the discrete amp with any decent opamp and not see any difference.

Look up Gilbert Cell for ( you can get a chip NE602 NE612 I think ) ... they can achieve balanced mixing ...

The Bounty Hunter Red Baron (George Payne, 1977) used LM1496 analog mixers (Gilbert cells). Barry Gilbert worked at Analog Devices when I was there, chatted with him a few times. He was a master of bipolar design.

[Aziz]

Hi all,

I have purchased a new external USB sound card just to test new possibilities. Its the Creative Sound BlasterX G6.
I am going to test the 192 kHz SR. So we are leaving the VLF and entering the LF region (>30 kHz). I hope, they haven't band limitted the output and input lines to the standard audio range (up to 20 kHz). Bad companies and bad sound cards do this just to hide their bad performance.

The sound card has a discreate headphone amplifiers in it and is capable to drive upto 600 Ohm impedances. This gives the TX coil enough bang.
Cheers
Aziz

[idhafa]

[moodz]

Hi all,

I have purchased a new external USB sound card just to test new possibilities. Its the Creative Sound BlasterX G6.
I am going to test the 192 kHz SR. So we are leaving the VLF and entering the LF region (>30 kHz). I hope, they haven't band limitted the output and input lines to the standard audio range (up to 20 kHz). Bad companies and bad sound cards do this just to hide their bad performance.

The sound card has a discreate headphone amplifiers in it and is capable to drive upto 600 Ohm impedances. This gives the TX coil enough bang.
Cheers
Aziz

I think they may have limited the input BW to 20 khz .... this would be as for anti aliasing purposes at the lowest sample rate 44.1 Khz with 20 Khz being just below the nyquist or 22.05 khz.
I use 20 Khz in the FPGA detector ( it is VLF not PI ) and it detects .05 gram gold no problemo.
Users have reported some problems with the ASIO driver for this unit ... but the specs are excellent in the audio performance department.

[Aziz]

I think they may have limited the input BW to 20 khz .... this would be as for anti aliasing purposes at the lowest sample rate 44.1 Khz with 20 Khz being just below the nyquist or 22.05 khz.
I use 20 Khz in the FPGA detector ( it is VLF not PI ) and it detects .05 gram gold no problemo.
Users have reported some problems with the ASIO driver for this unit ... but the specs are excellent in the audio performance department.

Hi Moodz,

I have seen a frequency response of the Creative Sound BlasterX G6 somehere in the internet, which is obviously capable to detect signals close up to the nyquist frequency of 96 kHz (@192 kHz SR). It implements a wall-brick anti-aliasing filter. So I should be able to sample up to 80 - 90 kHz true analog signals. Right now, my Creative Sound Blaster X-Fi Surround can go up to 48 kHz bandwith with wall-brick anti-aliasing filter. So I'm able to operate up to 45-47 kHz signals with ease.

I don't use the ASIO interface. ASIO, WDM, DirectSound, DirectPlay is quite dead. WASAPI is the new cash cow. WASAPI is complex to use. M$ developers don't know the KISS-Principle. (M$: Keep It Super Strange! Keep It Super Struggle!)

My detector software is at least 20 years old and I don't really need low latency interfaces as I use two dedicated sound cards: One for signal processing (external USB sound card) and another one for beeping using the internal sound card with low sampling rates and shorter buffers for low latency. It isn't much time critical and I can use the high level API (MME API, WaveIn/WaveOut..). It works so far from XP to Win11. The use of external USB sound cards benefit from high SNR as they don't pickup much noise from the main board.

Your FPGA detector is really very sensitive.
It will be a challenge to outperform your's.

Aziz

[Atul Asthana]

I think, the task scheduler for the Bee-Buzz 1 is working
I will test its logic next week, when I have the time.
It will then be tested with the coils and opamps

This amounts to solving major portion of the process complexity.
Building rest of the hardware and software is much less complicated and more straight forward, with lower chances of errors.

[Aziz]

Hi all,

just want to tell you, if you are going to make the best digital VLF detector with an embedded project, be prepared to transmit and decode at least 3 frequencies on a driven LC-tank (note: not free running LC-oscillator!). If the resonant frequency of the LC-tank is fr and you have a quite high Q coil, define a short narrow band frequency step df. df could be in range 50 Hz - 300 Hz for instance depending on your Q of your LC-tank.

fr = resonant frequency of the LC-tank

Transmit:
Transmit frequency 1: f1 = fr-df
Transmit frequency 2: f2 = fr
Transmit frequency 3: f3 = fr+df
The transmit frequencies can be driven via narrow band chirp wave form going from f1 to f3 or you could transmit them sequentially in a burst f1, f2, f3.

Decode:
Decoding frequency 1: f1 = fr-df
Decoding frequency 2: f2 = fr
Decoding frequency 3: f3 = fr+df

Then you have 3 magnitude and 3 phase information for single frequency LC resonant tank. Enough for good ground balancing and discrimination (resistive R and reactive X response extraction). A simplified version would process the magnitudes only.

Now, let's make it more:
If you make your LC-tank to resonate on two different frequencies (for instance: 12 kHz & 40 kHz), then you have 6 magnitudes and 6 phase informations. This should give even better results. But you need definitelly more processing power on your embedded project.

Oh well, I have really enough number crunching power on my Tablet PC. I could implement much much more.

Cheers,
Aziz

[Atul Asthana]

Hi all,

just want to tell you, if you are going to make the best digital VLF detector with an embedded project, be prepared to transmit and decode at least 3 frequencies on a driven LC-tank (note: not free running LC-oscillator!). If the resonant frequency of the LC-tank is fr and you have a quite high Q coil, define a short narrow band frequency step df. df could be in range 50 Hz - 300 Hz for instance depending on your Q of your LC-tank.

fr = resonant frequency of the LC-tank

Transmit:
Transmit frequency 1: f1 = fr-df
Transmit frequency 2: f2 = fr
Transmit frequency 3: f3 = fr+df
The transmit frequencies can be driven via narrow band chirp wave form going from f1 to f3 or you could transmit them sequentially in a burst f1, f2, f3.

Decode:
Decoding frequency 1: f1 = fr-df
Decoding frequency 2: f2 = fr
Decoding frequency 3: f3 = fr+df

Then you have 3 magnitude and 3 phase information for single frequency LC resonant tank. Enough for good ground balancing and discrimination (resistive R and reactive X response extraction). A simplified version would process the magnitudes only.

Now, let's make it more:
If you make your LC-tank to resonate on two different frequencies (for instance: 12 kHz & 40 kHz), then you have 6 magnitudes and 6 phase informations. This should give even better results. But you need definitelly more processing power on your embedded project.

Oh well, I have really enough number crunching power on my Tablet PC. I could implement much much more.

Cheers,
Aziz

great,
even narrowband / high Q tank, excited by 3 or more frequencies, should be able to provide enough info, for you to calculate about more of target's properties.

my design has considered this point, despite the lower computational power stm32f103c8t6, which should be able to handle the three closely spaced frequency issue with moderate Q of the coil.

the resultant data can be processed with material science / electrical / magnetic properties to guess the material/metal.

[Aziz]

Hi all,

in my special USB high latency case (Windows isn't an RTOS),
I have to decode even twice more to get the correct absolute phase lag between TX and RX. So I have to process the TX signal as well (besides the RX signal). Btw, any TX energy loss and frequency shift will be detected too. This gives more info for further processing.
For a true dual frequency VLF/LF detector with three narrow frequencies around the resonant frequencies, I need 12 decoders. This is no problem with CPU power on Tablet PC. But can be critical on embedded systems with micro controllers.

BTW, it is important, that the narrow band width should not be large, as we don't want to to operate the coils in the high Z region (Z is impedance of the LC-tank). We need some current flow through the TX coil of course. We are operating the TX coil in the low Z region around the resonant frequency fr.

Heavy mineralisation will increase the TX/RX inductance (and thus lower the resonant frequency) and targets will lower the inductance slightly (hence increases the resonant frequency). Magnetic field conduction TX -> RX occurs on ferro magnetic materials nearby the coil. Eddy current induction on metal targets too. All possible effects can be processed at the same time.

Aziz