VLF MD with digital signal processing : Bee-Buzz 1

[Atul Asthana]

I'd be interested in an example. In general, the power consumption of the TX circuit has little to do with the transmitted energy except in a total-loss PI design. And the amount of power consumed by ground mineralization depends heavily on the loss angle and susceptibility of the soil. Most mundane soils have a near-0° loss angle so the power loss probably isn't measurable. What is the loss angle of the local soil? In the absence of ground losses or in a sim, how do you measure TX power?

Carl,

I think, I'll ask the profs of the students to verify, however, the figeures here are for various kinds of soils, including high iron and other metals content soils, and the measurements are the differences in current consumption before and after immersion.

importantly, in my case, power is controllable and any errors will be corrected by saturation analysis.

I do expect that core loss in the soil is both reactive and resistive to the extent that it is measurable, and for a free running oscillator, this will cause frequency shift/detuning/change in inductance. (thats how bfo mds work).

[moodz]

Guys ... the noise contributed from the preamp is the least of your concerns ... the environmental noise from the coil etc is orders of magnitude higher.
Below is a real pic of the noise off the coil ( on the CRO ) and the raw sampled signal from the ADC on a VLF frontend. The actual signal in this case is 19 khz and the magnitude is about 10 mV peak.
The noisy input is around 600 mV peak.( so the 10 mV signal is buried in there )

Yes you can see some noise on the ADC sampled signal o/p ( red trace VGA screen ) but let me spell it out ..
Its S-Y-N-C-H-R-O-N-O-U-S noise ... which is completely different from asynchronous noise ( AKA white / pink / shot / etc etc whatever noise etc ) which is the type of R-A-N-D-O-M noise you get in an opamp.

Previous threads have bogged down in who can find the lowest noise opamp .... move on!

[Carl-NC]

I would not think so. Post 83 has an example calculation; if you instead use -120dBfs noise floor and NBW = 100Hz then you need a preamp with 0.75nV/rtHz. You can get close to this with an off-the-shelf opamp but, personally, I've never seen any metal detector with such an aggressively low preamp noise.

[Detectorist#1]

Thanks Carl,

I understand that your answer supports my opinion that 0.6nV/sqrt(Hz) noise at front-end stage is not useful in VLF MD.

[Carl-NC]

I asked some of the engineering students to build some of the oscillators of vlf schematics, power them in the air and close to the ground, also analyse the circuits through simulators on the maximum possible power consumptions.

I'd be interested in an example. In general, the power consumption of the TX circuit has little to do with the transmitted energy except in a total-loss PI design. And the amount of power consumed by ground mineralization depends heavily on the loss angle and susceptibility of the soil. Most mundane soils have a near-0° loss angle so the power loss probably isn't measurable. What is the loss angle of the local soil? In the absence of ground losses or in a sim, how do you measure TX power?

[Aziz]

Thanks Carl,

yes uV. Not mV.

No, we are not going to average over 1024 samples of course. This was an example and comes from my sound card FFT parameters.
N <= 256 is more reasonable. On the other hand, more noise will be decoded.
Aziz

[Carl-NC]

Voltage noise =
Vrms = en * sqrt(BW) * G
Vrms = 5 nV/sqrt(Hz) * sqrt(93.8 Hz) * 100
Vrms = 4.843 mVrms

Lets take the 0.6 nV/sqrt(Hz) version
Vrms = 0.6 nV/sqrt(Hz) * sqrt(93.8 Hz) * 100
Vrms = 0.581 mVrms

These should both be uv, not mv. Also, averaging by 1024 is too much. This results in ~94 samples per second. In other words, if you are sweeping the coil at 1m/s you will get 94 samples per meter, or ~1 sample every 1cm. For a typical coin target at depth you will be lucky to get 2 data points. That's not much to work with for demodulation or post-demod processing. A 64x average seems far more reasonable.

The main my question is: Is the level of the input signal in VLF IB solution is in range of microvolts? In VLF detector Tesoro Bandido RX coil is connected to first OpAmp thru two resistors of 5.1 KoHms. This tell me that the level of the noise in front end OpAmp in this type of MD is not problem - or I'm not right?

From an analog perspective, 5.1k has 9.2nV/rtHz of noise. Eventually you will be looking at roughly 100Hz of noise bandwidth (post demod filters) so the input-referred noise is 91.9nv rms. It's actually worse than this because there are two 5.1k resistors (->13nv rms). This determines the minimum target signal you can see, assuming no EMI. But there is always some EMI and it will often dominate so beating down the thermal noise usually doesn't buy a whole lot.

[Aziz]

The main my question is: Is the level of the input signal in VLF IB solution is in range of microvolts? In VLF detector Tesoro Bandido RX coil is connected to first OpAmp thru two resistors of 5.1 KoHms. This tell me that the level of the noise in front end OpAmp in this type of MD is not problem - or I'm not right?

Yes,

it is in the range and well beyond and due to the high 24-bit ADC resolution realised in the USB sound card. Four least significant bits are pure noise in my case. The effective number of bits reduces to max 20 bit. Some bad sound cards will loose more bits. But my old usb sound card is fine. If we average the decoded signals, we can increase the accuracy and can achieve µV resolution easily too. Averaged over 2 samples gets us 3 dB (factor sqrt(2) ) improvement of SNR. This is 0.5 bit additional accuracy. 4 samples gets us 6 dB improvemen of SNR and this is one bit more accuracy. This is the reason, why we wouldn't really need an preamp for the RX side.

BTW, the RX-Coil will pickup a lot of EMI noise. This is well beyound the µV level.

Anyway, one wouldn't really need an ultra low noise preamp. A simple NE5534 preamp should be ok.
Let's make some comparison calculation about the expected noise voltage for single frequency decoding (DFT) to be sure:

Sample rate 96 kHz
Window size N: 1024 samples
Gain G = 100
Bandwidth BW: 93.8 Hz, for 1 dft bin
en = 5nV/sqrt(Hz), Voltage noise density for NE5534

Voltage noise =
Vrms = en * sqrt(BW) * G
Vrms = 5 nV/sqrt(Hz) * sqrt(93.8 Hz) * 100
Vrms = 4.843 mVrms

Lets take the 0.6 nV/sqrt(Hz) version
Vrms = 0.6 nV/sqrt(Hz) * sqrt(93.8 Hz) * 100
Vrms = 0.581 mVrms

Well, ok. There is a difference. But this is not much compared to the EMI noise, which will be amplified by G=100 too.

Why I have taken the 93.8 Hz BW? This is, when you apply a FFT at 96 kHz SR for N=1024 samples. BW of one dtf bin is SR/N = 96000/1024 = 93.75 Hz. You would have the same bandwidth for Goertzel, Lock-in amp decoding using SR=96 kHz and window size N=1024. We only look a the single frequency decoded noise. This is what is relevant in this case.

Aziz

[Detectorist#1]

The main my question is: Is the level of the input signal in VLF IB solution is in range of microvolts? In VLF detector Tesoro Bandido RX coil is connected to first OpAmp thru two resistors of 5.1 KoHms. This tell me that the level of the noise in front end OpAmp in this type of MD is not problem - or I'm not right?

[Aziz]

Hi Aziz,

The idea with paralleling of power transistors plus very low source resistor for low noise amplifiers is very old (not so old as you and me). In VLF MD the level of the signal on RX coil is far from microvolt and using of this idea doesn't make sense. In PI MD the value of the safety resistor (in mono coil configuration) is 300-1000-3000 oHms. This is too far from 5-10 oHms and again using of this idea doesn't make sense. In PI MD with DD coil maybe will be achieved some progress if soil's conditions allows this.

Hi,

no, this preamp is only designed for IB-Coils with AC coupling (not DC). Not for PI detectors. Preferably for Sound Card line inputs. Never use the microphone input line (bandwidth limitted, very noisy and high impedance).

The RX coil does typically have very low impedance (<1 Ohm if you can make it or not more than a few Ohms).

Take care of the power supply filter. Any voltage regulator produces more noise and would degrade the preamp. So the power suply filter is necessary to get ultra low noise specs.

I have testet many NPN and PNP transistors. The proposed transistor is the best.

This is not old but state of the art preamp. No op-amp chip can get the performance of the nice solution.

Aziz

[Detectorist#1]

Hi Aziz,

The idea with paralleling of power transistors plus very low source resistor for low noise amplifiers is very old (not so old as you and me). In VLF MD the level of the signal on RX coil is far from microvolt and using of this idea doesn't make sense. In PI MD the value of the safety resistor (in mono coil configuration) is 300-1000-3000 oHms. This is too far from 5-10 oHms and again using of this idea doesn't make sense. In PI MD with DD coil maybe will be achieved some progress if soil's conditions allows this.

[Atul Asthana]

I'm curious how you come up with these transmit power numbers.

I asked some of the engineering students to build some of the oscillators of vlf schematics, power them in the air and close to the ground, also analyse the circuits through simulators on the maximum possible power consumptions.

[Carl-NC]

Regarding VLF designs, many available circuits show oscillators / transmitters outputting around 10-30 mW.

I'm curious how you come up with these transmit power numbers.

[Atul Asthana]

My opinion is definitely to use an STM32 with an integrated 12bit or 16bit ADC, it is the cheapest solution and a simple programming environment.

I agree with this perspective

I’ve conducted extensive research, including technical analysis, cost-benefit evaluation, target audience considerations, and market/component availability assessments, before choosing the STM32.

My primary focus is on students, the individuals who will actually build this metal detector. Our role as mentors and guides is to support their learning journey and help them succeed.

Regarding VLF designs, many available circuits show oscillators / transmitters outputting around 10-30 mW. Given the ground’s resistive and lossy nature, much of this power is dissipated as heat, with some reflecting as noise or target response. I believe that 50 mW should be the upper limit in a quiet front end, particularly on dry, sandy soil. However, adjusting power to manage noise is crucial, which is why I’ve included provisions for varying transmission power.

I’m still seeking clarity on my earlier questions about corner frequencies and other fundamental processing parameters. These are critical for effective design, and not having a clear understanding is as detrimental as not designing a robust front end.

[Marchel]

Marchel and Aziz,

Let's talk about new and more good possibilities if they are guaranteed by new ideas, not for tools. For end users is more important the end result at reasonable price.

My opinion is definitely to use an STM32 with an integrated 12bit or 16bit ADC, it is the cheapest solution and a simple programming environment.