VLF MD with digital signal processing : Bee-Buzz 1

[Atul Asthana]

please comment and identify faults or suggest improvements

A Conceptual Design for a Fully Digital VLF Metal Detector Using STM32F103C8T6

This paper outlines the concept of a fully digital Very Low Frequency (VLF) metal detector based on the STM32F103C8T6 microcontroller. The design prioritizes simplicity, low cost, and ease of construction for electronics hobbyists while leveraging the processing power of a 32-bit microcontroller running at 72 MHz. STM32F103C8T6 is about 20 times faster than popular 8 bit microcontrollers.

Design Objectives:

The design adheres to the following objectives:

Fully digital signal processing except for the analog front-end of the receiver.
Mostly automatic adjustment to environmental conditions.
Single, fixed frequency selectable at the design stage.
Stable transmit frequency.
Variable transmit power.
Low cost and readily available components.
Minimal component count.
Suitable for construction by hobbyists.

System Architecture:

The system revolves around the STM32F103C8T6 microcontroller running at 72 MHz.

Transmitter:
Frequency Generation: A PWM channel of the STM32 generates the transmit frequency. A lookup table containing 64 points representing a sine wave drives the PWM output. This reduces harmonic content, with the first significant harmonic (3rd) attenuated by at least 40 dBc.

Output Stage: An H-bridge, driven by the PWM output, powers the transmitting coil. The coil is tuned to the desired transmit frequency using a capacitor.

Power Control: PWM duty cycle controls the power delivered to the transmitting coil, allowing for adjustable transmit power.

Receiver:

Front-End: The receiving coil, tuned to the transmit frequency, picks up the signal. This signal is amplified and bandpass filtered to reduce noise and interference.

Analog-to-Digital Conversion: The filtered signal is sampled by the STM32's 12 bits ADC at a rate of 8 samples per transmit cycle. This oversampling technique reduces sampling noise by approximately 60 dB.

Digital Signal Processing: The sampled data is processed using the Goertzel algorithm. This algorithm efficiently calculates the amplitude and phase of the received signal at the transmit frequency, requiring significantly fewer computations than a full FFT, especially for single-frequency detection.

Signal Processing and Target Identification:

The output of the Goertzel algorithm (amplitude and phase) is further processed:

Low-Pass Filtering: Both amplitude and phase data are passed through two low-pass filters with different cutoff frequencies (e.g., 100 Hz and 1 Hz).

Environmental Compensation: The output of the 1 Hz filter represents the slowly changing ground conditions (ambient response).

Target Response: The output of the 100 Hz filter represents the more rapidly changing target response.

Target Detection: A target is detected if the target amplitude response exceeds a selectable threshold above or below the ambient response.

Target Identification: Upon detection, the processed phase information is used in conjunction with the amplitude to categorize the detected metal. A VDI (Visual Discrimination Indicator) number is calculated based on these parameters.

Output: The VDI number and the identified metal type are displayed on an LCD connected to the STM32. An audible tone is also generated.

Rest of the features can easily be added in the software, including transmitting some data to a remote unit, or including GPS data, or plotting the values on a map etc.

Design Parameters and Flexibility:

The design allows for several adjustable parameters:

Transmit Power: Controlled by the PWM duty cycle.
Detection Threshold: Adjustable to fine-tune sensitivity.
Transmit Frequency: Selectable during the design phase by changing the tuning components of the coils and adjusting the PWM frequency.
Signal Processing Methodology: The choice of filters, thresholds, and VDI calculation methods can be adapted to specific needs.

Perceived Advantages:

Digital Signal Processing: Enables precise control and flexible signal analysis.
Goertzel Algorithm: Efficient single-frequency detection.
Low Component Count: Simplifies construction and reduces cost.
STM32F103C8T6: Powerful, low cost and readily available microcontroller.

Conclusion:

This design provides an experimental concept for a fully digital VLF metal detector using readily available components and a powerful microcontroller. The use of digital signal processing techniques like the Goertzel algorithm and flexible parameter adjustments allows for a high-performance, low-cost solution suitable for hobbyist construction. Further development and testing would be required to optimize performance and refine the target identification algorithms.

[ArchibaldSTM]

Is there a circuit?

[Atul Asthana]

Is there a circuit?

not yet. though, there isnt much in the circuit, and description explains almost all of it except the analogue part.

but, its not the circuit, its the embedded software on the microconyroller, thats of prime importance.

[ArchibaldSTM]

That's why I asked. Andrey Fedorov already made a similar device. I wanted to see yours too.

[Carl-NC]

That's why I asked. Andrey Fedorov already made a similar device. I wanted to see yours too.

This is a PI design, although it is really just a capacitor away from being a VLF design.

[ArchibaldSTM]

Oops! I posted the wrong picture))) I'm sorry

[Detectorist#1]

Hi Atul Asthana,
The general question is - what really new will be the results of this project comparing with so much known VLF serious projects in last 20 years. Using of uP spends 3-4-5-6 integral circuits and what will be the different in the end representations of this MD. Yes, the possibility of automatic adaptation to different coils (maybe will be implemented in this design) is convenient, but every type of coils (size and configuration) is optimal for a specific conditions. The idea - now we will use uP and after that we will think how to achieve a new different results is not new.

[boilcoil]

Hi Atul Asthana,
The general question is - what really new .........

The benefit is, that with direct sampling and digital signal processing can be made as many virtual channels as you want - something impossible for an analog detector.

[Atul Asthana]

Hi Atul Asthana,
The general question is - what really new will be the results of this project comparing with so much known VLF serious projects in last 20 years. Using of uP spends 3-4-5-6 integral circuits and what will be the different in the end representations of this MD. Yes, the possibility of automatic adaptation to different coils (maybe will be implemented in this design) is convenient, but every type of coils (size and configuration) is optimal for a specific conditions. The idea - now we will use uP and after that we will think how to achieve a new different results is not new.

My apologies sir, for not clearly bringing out the purported benefits of 'conjuring' this project / device / idea. Allow me to add some comments.

This project aims to bridge the gap between simple hobbyist detectors and more advanced commercial detectors by leveraging the power of Digital Signal Processing on a cost-effective and open-source platform. It offers the potential for significantly improved performance and features compared to typical hobbyist designs while retaining the educational and modifiable nature of open-source projects.

This microcontroller-based VLF metal detector project differentiates itself from existing open-source/hobbyist designs through its focus on improving performance, features and extendability by using full Digital Signal Processing (DSP). While most hobbyist detectors rely on analog circuitry or basic microcontroller functions, this project utilizes the STM32F103C8T6 (or may be other new microcontrollers) for complete signal analysis. This enables advanced digital filtering, precise phase/amplitude extraction via algorithms like Goertzel, and sophisticated VDI calculation for improved target identification.

Unlike simpler designs with limited features, this project has the potential for advanced capabilities like automatic ground tracking, notch filtering, customisable audio tones, and data logging, all implemented in software for greater flexibility. Like other hobbyist projects, it intends to be open-source and modifiable, encouraging experimentation. This approach aims to bridge the gap between basic hobbyist detectors and more advanced commercial models by offering significantly enhanced performance and features within an accessible, open-source platform.

This project also offers significant expansion potential. Integrating GPS data could enable location-tagged target mapping, even tracking shifting ambient conditions for subsurface archaeological surveys using GIS mapping. The open-source nature empowers diverse users to contribute their expertise, extending functionality through software modifications and integration with other tools, far beyond the initial design of this project.

[Carl-NC]

All of the latest VLF/MF metal detector models from the leading companies are direct sampling designs. RX coil followed by a preamp, ADC, and DSP. Incredibly simple & cheap. Pretty much all these designs use a 24b ADC (usually an audio CODEC such as a PCM3060) so a 12-bit ADC may prove to be a little light for this task. With direct sampling & DSP, you can do things that are difficult and maybe impossible in analog circuitry. But don't get your hopes up on precision target phase; there is no such thing. Target phase is always roughly-approximate as it can vary with coil position, sweep speed, and ground mineralization. Trying to achieve precision here is pretty meaningless and 1-2° is about as good as it gets.

If I ever get around to doing ITMD4, it will be with a greatly expanded focus on digital detectors and MF. So I encourage you to continue.

[Atul Asthana]

yes, a 24-bit ADC offers superior performance, however the increased hardware interfacing and software development complexity is undesirable for a hobbyist-level project. Since commercial VLF detectors achieve approximately 80 dB of dynamic range, a 12-bit ADC, providing over 70 dB, is considered sufficient for this design's objectives.

The goal is to create a simple yet effective VLF metal detector, maximizing the STM32 Blue Pill's capabilities to minimize hardware and software complexity. This approach avoids the need for external drivers, additional chips, multiple power supplies, and complex tuning procedures.

Similarly, existing software libraries will be used to reduce the burden of writing and testing custom signal processing code.

Essentially, it's an exercise in fully utilizing the Blue Pill for this application.
The primary objective is a "good enough" metal detector built with simplicity in mind.

[ArchibaldSTM]

Many devices are built on a similar principle of signal processing. One of them is Stalker-IB. However, this is a commercial project. If yours is free, then I only support your desire to do this!

[JoyJo]

All of the latest VLF/MF metal detector models from the leading companies are direct sampling designs. RX coil followed by a preamp, ADC, and DSP. Incredibly simple & cheap. Pretty much all these designs use a 24b ADC (usually an audio CODEC such as a PCM3060) ...

The Minelab Equinox board also has an audio codec chip with a dual-channel ADC. The only thing is that the multiplexer chip is also placed on the board. I still haven't figured out what it's for. Of course, it somehow has to be involved in the multi-frequency mode, but in which part of the circuit: in TX or RХ?

[Atul Asthana]

Many devices are built on a similar principle of signal processing. One of them is Stalker-IB. However, this is a commercial project. If yours is free, then I only support your desire to do this!

This project embraces a fully open-source and free philosophy, mirroring the nature of the Blue Pill hardware and its associated software ecosystem. Leveraging existing open-source components minimizes development effort. The core objective is collaborative development, harnessing collective expertise and knowledge to create a metal detector accessible to all.

[Atul Asthana]

So, please comment on the following :

Let's discuss the software and hardware design of this VLF metal detector, aiming for simplicity and efficient use of the STM32 Blue Pill.

Software Considerations:

We need to ensure the microcontroller can process data quickly enough. The processing involves:

1. Generating the transmit signal (PWM).
2. Sampling the received signal 8 times per cycle.
3. Calculating signal strength and phase using the Goertzel algorithm.
4. Filtering the signal strength and phase.
5. Checking and responding to user input (switches).
6. Updating the display and generating audio tones.
7. Sending data via serial communication.
8. (Optional) Reading GPS data.

I was reading that its possible to apply goertzel algorithm on just one single cycle and get good enough results, since we are using a filtered single frequency. and single cycle processing can be done within 2-4 microseconds, giving us a lerway of about 1.9 milliseconds for rrst of the tasks.Since we're working with a filtered, single frequency (6 kHz), the Goertzel algorithm can be efficiently applied to just one cycle of the received signal. This single-cycle processing is estimated to take only 2-4 microseconds, leaving approximately 1.9 milliseconds for other tasks within each sampling period (assuming a sampling rate around 512 Hz). This significantly reduces the computational burden.

We'll sample the target signal around 512 times per second. For each sample, we'll use the Goertzel algorithm to get signal strength and phase, then smooth these values using digital filters. These smoothed values will be used to determine the target type (VDI) and generate output.

This approach offers several advantages:

1. Lower power consumption.
2. Sufficient processing time. Using a simplified (fixed-point) Goertzel algorithm avoids complex calculations.
3. Time for other tasks.
4. Reduced stress on the H-bridge transistors.
5. Allows use of single-precision floating-point math for better accuracy.

Hardware Considerations:

We'll use DD coils:

• The transmit coil (25 turns, 22 SWG, 10" diameter) has approximately 223 μH inductance, 845 ohms impedance at 6 kHz, and requires a 315 nF capacitor for resonance.
• The receive coil (40 turns, 28 SWG, 10" diameter) has approximately 678 μH inductance, 1357 ohms impedance at 6 kHz, and requires a 104 nF capacitor for resonance. This means the receive amplifier must match this impedance.

A 1 cm steel coin 5" away, with 100 mW transmit power, is expected to generate a very weak signal (microvolts) in the receive coil. Since the STM32's ADC has a least count of about 0.8 mV (with a 3.3V full scale), we'll need an amplifier with a gain of around 1000.

Additionally, we could increase the transmitting power, though this may increase the noise or increase false targets/wrong identification.

To simplify coil tuning, we'll implement a frequency counter mode on the STM32 using a Colpitts oscillator. This will measure the resonant frequency of each coil. Ideally, the transmit frequency will be automatically adjusted to match the measured resonant frequency (within a few Hz of 6 kHz to avoid affecting filter and impedance calculations).