🧲 High-Precision Fluxgate Magnetic Field Sensor Design

An open-source, high-sensitivity, single-axis Fluxgate magnetometer and development platform designed for aerospace, geophysics, and navigation applications. This project implements a fully integrated system containing hardware (KiCad PCBs), firmware (MSP430 24-bit SD-ADC & RP2040 Controller), simulations (LTSpice), and software (Python PyQt real-time GUI).

Developed as a senior capstone engineering project at Karadeniz Technical University (KTU) by Erdem Metin and Mücahit Ata, under the supervision of Prof. Dr. İsmail Kaya.

---

📸 System Overview

The Fluxgate sensor system is designed around a dual-MCU architecture consisting of an RP2040 (master controller and driver) and an MSP430AFE253 (24-bit Sigma-Delta ADC data acquisition).

Fig 1: Full system schematic diagram detailing all MCU pinouts, power distribution, and analog filters.

*Fig 2: KiCad PCB 3D layout render (left) and the fully assembled physical testing board (right).*

graph TD
    PC[Computer PyQt GUI / Matlab] <-->|USB virtual COM / UDP Wifi| Pico[Raspberry Pi Pico RP2040]
    Pico -->|PWM Excitation Control f| FET[H-Bridge MOSFET Driver]
    FET -->|AC Excitation staircase Wave| Core[Magnetic Core & Coils]
    Core -->|Induced Sensing Signal| Demod[Analog Processing Circuit]
    Demod -->|DC Level / AC Signal| MSP[MSP430AFE253 24-bit ADC]
    MSP -->|1.5 MBaud UART| Pico

Loading

---

📖 Theory of Operation

A Fluxgate Magnetometer measures weak static or low-frequency magnetic fields (in the nT to µT range) by exploiting the non-linear magnetic saturation characteristics of a ferromagnetic core.

1. Excitation (Drive): An AC current at frequency $f$ drives the core deep into saturation in both positive and negative directions.
2. Symmetric Core State: In the absence of an external magnetic field, the magnetic flux changes symmetrically. The voltage induced in the sensing (pick-up) coil contains only odd harmonics ($3f$, $5f$, etc.).
3. Asymmetric Core State (External Field Present): When an external magnetic field (like the Earth's field $H_{ext}$) is present, it biases the core, causing it to saturate earlier in one direction and later in the other.
4. Second Harmonic (2f) Detection: This asymmetry generates even harmonics (predominantly the $2f$ component) in the sensing coil. The amplitude of this second harmonic is directly proportional to the magnitude of the external magnetic field, and its phase indicates the field's direction.

---

🛠️ Project Structure

The repository is structured into logically separated folders to make implementation and replication straightforward:

• 📁 Hardware/: KiCad schematic and PCB layout designs, Gerber files, and Bill of Materials (BOM).
• 📁 Firmware/:
  • MSP430: IAR and CCS projects for 24-bit Sigma-Delta ADC sampling at 16.5 kSps and 1.5 MBaud USART transmission.
  • RP2040 (Raspberry Pi Pico): C/C++ SDK projects generating precise excitation PWM and streaming data to PC via USB Serial or UDP Wi-Fi.
• 📁 Software/: Python-based GUI utilizing pyqtgraph for real-time serial and UDP plotting, and Matlab analytics.
• 📁 Simulations/: LTSpice circuits simulating the H-Bridge excitation driver and the analog synchronous demodulation path.

---

🚀 Step-by-Step Build & Replication Guide

To replicate and build this high-precision Fluxgate sensor, follow these five steps:

1. Hardware Fabrication & Assembly

1. Go to Hardware/ and select either the Excitation Circuit V1, the Excitation Circuit Final, or the JLCPCB Version which is optimized for SMT assembly.
2. Export Gerbers or use the provided .zip files under the directories to order PCBs from your manufacturer (e.g., JLCPCB).
3. Assemble the PCB using the BOM. Key components include:
   • Primary Controller: Raspberry Pi Pico (RP2040)
   • Precision ADC: MSP430AFE253IPW (24-bit Sigma-Delta ADC)
   • H-Bridge Driver: HIP4082xB half-bridge gate driver
   • Excitation MOSFETs: AP2300 or IRFZ44N
   • Ultra Low-Noise Opamps: TP2311 or LT149x series (12nV/√Hz input noise)
   • Voltage References: AN431 shunt regulators (2.5V, 20ppm/°C stability)

2. Fluxgate Core & Coil Preparation

1. Core Selection: Use a high-permeability green iron-powder ferrite toroid core with an inner diameter of 7 mm, thickness of 2 mm, and height of 4.5 mm. (While a 6-81 Permalloy racetrack core with $100\text{ µm}$ laminations provides lower noise, a toroidal ferrite core is robust and hand-windable).
2. Drive Winding: Wind the excitation (drive) coil uniformly around the toroid using 32 AWG enameled copper wire. In the experimental prototype, the completed drive winding exhibited a measured series resistance of $17.945\text{ }\Omega$ and a series inductance of $19.686\text{ mH}$.
3. Sense Winding: Wind the sensing (pick-up) coil on top of the drive winding, orienting the windings to capture the fluxgate effect while canceling direct coupling from the excitation drive.

Fig 3: The custom hand-wound toroidal fluxgate core integrated onto the testing board.

3. Simulation & Validation

1. Open Simulations/LTSpice/.
2. Run fluxgate.asc to simulate the H-Bridge excitation path. Verify the staircase voltage waveform across the drive bobbin.
3. Run fluxgateACAnaliz.asc or Draft7.asc to analyze the frequency response of the active integrator and synchronous demodulator. Ensure the integration region lies between 159 Hz and 1591 Hz and the low-pass filter kesim frekansı is configured near 28 Hz.

4. Firmware Deployment

1. MSP430 Firmware:
   • Open IAR Embedded Workbench or Code Composer Studio.
   • Load the project in Firmware/MSP430_AFE_ADC/ or Firmware/MSP430_ADC_to_UART/.
   • Compile and flash the firmware using an eZ-FET debugger (or the Spy-Bi-Wire pins on an MSP430 Launchpad).
   • Function: Configures the 12MHz MCLK, initializes the SD24 Module in group conversion mode, and streams 24-bit raw ADC readings via USART0 at 1.5M Baud.
2. RP2040 Firmware:
   • Navigate to Firmware/Pico_Firmwares/.
   • Choose either pico-custom-firmware-usb (USB Virtual Serial) or pico-custom-firmware-udp (Wi-Fi streaming).
   • Build the project using the Pico SDK (CMake) and flash the resulting .uf2 file onto your Pi Pico.
   • Function: Emits precise complementary PWM drive signals (80µs duty, 40µs dead-time) at frequency $f$ to drive the H-Bridge. It uses a PIO state machine to reliably capture the 1.5MBaud UART stream from the MSP430 and forwards it to the PC.

5. GUI & Visualization

1. Install Python 3 and requirements:

   pip install pyqtgraph PyQt5 pyserial numpy

2. Run the visualization app in Software/Python_GUI_Demo/:

   python compass.py
   # or
   python pyqtgraph_float_v2x2_yesim.py

3. Connect your Pi Pico to the PC via USB. You will see a real-time rolling graph of the 24-bit sampled magnetic field data representing the external field.

---

📡 Digital Signal Processing (DSP) & Digital Lock-In Mode

During the final experimental testing phase of this project, the analog op-amps in the physical synchronous demodulation circuit were found to be defective or counterfeit, preventing analog phase detection from working out of the box. To solve this, the project utilized a highly robust Digital Signal Processing (DSP) pipeline to bypass the analog demodulator entirely.

Bypassing the Analog Demodulator

1. The analog SPDT switch (SN74LVC1G3157) was disabled (tied to a static ground state via the MCU).
2. The active low-noise op-amps were configured as a simple differential preamplifier.
3. The raw, high-speed AC signal (which contains the even and odd harmonics) was fed directly into the MSP430's 24-bit Sigma-Delta ADC.
4. The raw, un-demodulated samples were digitized at ~16.5 kSps and streamed to the PC via the RP2040 UART/USB bridge.

Two Host-Side Digital Processing Implementations

Your software repository implements two distinct mathematical approaches on the host PC to extract the external magnetic field from this raw stream:

1. Frequency-Domain Demodulation (Real-Time FFT):

   • File: pyqtgraph_float_v2x2_yesim.py
   • Method: Computes a high-performance Fast Fourier Transform (FFT) on blocks of 8192 raw samples using Scipy:

     Yf0 = fft(ch0)

   • Result: The GUI displays the real-time frequency spectrum. By measuring the amplitude peak of the second harmonic ($2f$ at 25 kHz) in the frequency domain, the software calculates the precise external magnetic field intensity, completely replacing the physical analog lock-in filters!

2. Time-Domain Smoothing (Moving Average Convolution):

   • Files: udp_plot_pyqt.py and udp_plot_pyqt_2.py
   • Method: Unpacks the high-speed telemetry and applies a real-time Moving Average Filter using NumPy convolution across a 1000-sample window:

     filtered_data = np.convolve(data, weights, mode='valid')

   • Result: Smooths out the excitation switching ripples and transient spikes, feeding a clean time-domain signal to the graphical compass needle interface to dynamically show the core's magnetic orientation.

---

🛠️ Hardware Debugging & Engineering Solutions

During the assembly and testing phase, two critical engineering challenges arose and were systematically resolved:

1. The H-Bridge Bootstrap Capacitor & Duty Cycle Adaptation

• The Issue: High-side gate drivers require a bootstrap capacitor ($C_{Bst}$) to maintain the gate-to-source voltage ($V_{GS}$) when the high-side MOSFET turns on. By design guidelines, the minimum bootstrap capacitance is: $$C_{Bst} = \frac{Q_{gate}}{\Delta V_{Bst}}$$ Using the AP2300 gate charge $Q_{gate} = 11\text{ nC}$ and an acceptable voltage ripple $\Delta V_{Bst} = 0.1\text{ V}$, the minimum capacitor required is $110\text{ nF}$. Due to a design oversight, a $1\text{ nF}$ capacitor was initially populated. This capacitor discharged instantly, causing the high-side gate drivers to drop out immediately.
• The Fix: Populating a $1\text{ µF}$ capacitor resolved the gate-dropout issue. However, experimental tests under a $330\text{ }\Omega$ load revealed that the high-side gate could remain continuously open for a maximum of $1.125\text{ ms}$ before the bootstrap charge depleted.
• PWM Adaptation: To prevent the high-side gate from collapsing during lower frequencies, the RP2040 firmware was adapted to emit a highly optimized PWM excitation pattern. Instead of a basic symmetrical square wave, the drive cycle was structured as:
  • $0.8\text{ ms}$ Positive Drive Phase (containing high-speed $80\text{ µs}$ pulses with $40\text{ µs}$ dead-times).
  • $0.4\text{ ms}$ Neutral Phase (H-bridge disabled, allowing the bootstrap capacitors to fully recharge).
  • $0.8\text{ ms}$ Negative Drive Phase (H-bridge reversed).

2. Defective Analog Op-Amps & DSP Bypass

• The Issue: During board validation, short-circuiting the sensing coil to ground resulted in the low-noise differential preamplifier and active integrator output locks stuck at a solid $2.5\text{ V}$ (the virtual ground analog reference voltage). The op-amps (TP2311 or LT1492) failed to respond to any AC or DC input signal changes, leading to the conclusion that the physical components received were defective or counterfeit.
• The Solution: The analog synchronous SPDT demodulator switch (SN74LVC1G3157) was set to ground mode via the RP2040, transforming the first stage into a simple, highly linear $1:2$ differential preamplifier (total differential gain of 20). The raw, high-speed AC pick-up signal was routed directly into the MSP430's 24-bit Sigma-Delta ADC, and the entire lock-in and filtering cascade was bypassed in hardware and reconstructed in real-time software on the host PC.

---

⚡ Specifications

Parameter	Value	Description
Sensor Type	Parallel Fluxgate	Single Axis Magnetometer
Excitation Waveform	AC Staircase Wave	Minimizes transient EMF spikes
Excitation Frequency	12.5 kHz	Configurable via Pico PWM registers
ADC Resolution	24-Bit	Sigma-Delta Converter (MSP430AFE253)
Sampling Rate	~16,500 samples/sec	High-speed oversampled stream
Baud Rate	1,500,000 Baud	High-speed MSP430 to RP2040 UART link
Demodulator Gain	1:10 (or 20 Differential)	Synchronous analog SPDT demodulation
Low-Pass Filter (LPF)	$f_c = 28.42\text{ Hz}$	Eliminates grid noise (50Hz) and ripples
Integrator Bandwidth	159 Hz to 1591 Hz	Limits noise and captures second harmonics

---

⚖️ License

This project is open-source and released under the MIT License. Feel free to use, modify, and distribute for personal, academic, or commercial purposes.