Sound Locator M33

Sound Locator M33

This project represents an ambitious attempt to create a real-time sound source localization system using an embedded platform. The objective was to accurately detect and locate a stationary sound source on a two-dimensional plane using multiple microphones and advanced signal processing techniques. This project was developed in collaboration with James Schutte and Julian Bruin as part of the 5LIU0: Premaster Linear Systems course at Eindhoven University of Technology.

Content

The system demonstrates core functionality for sound source localization, though with some components implemented as post-processing solutions due to time constraints.

Figure: Localization accuracy results showing better precision within the triangular sensor array

System Abstract

This project aimed to create a real-time sound localization system to address the challenge of accurately determining the position of acoustic sources in 2D space. The system uses a triangular microphone array with time difference of arrival (TDoA) algorithms and embedded signal processing.

System block diagram showing the complete signal processing pipeline

Features

• Triangular Microphone Array: Three omnidirectional MEMS microphones positioned 50cm apart for optimal spatial resolution
• Real-time Sampling: 166.67 kHz effective sampling rate per microphone using DMA-based batch processing
• Signal Amplification: Custom non-inverting operational amplifier with 8 kHz low-pass anti-aliasing filter
• Advanced Signal Processing: Cross-correlation based time difference of arrival determination
• Embedded Localization: Newton-Raphson method for real-time 2D position estimation
• Frequency Detection: FFT-based target signal validation using CMSIS-DSP library
• FreeRTOS Integration: Real-time operating system for concurrent signal processing tasks
• Precision Targeting: Theoretical millimeter-level accuracy with 2.06mm distance resolution

Technologies Used

• Hardware: Raspberry Pi Pico 2 (RP2350 Cortex-M33) with Floating-Point Unit
• Programming Languages: C++ Standard 17 with embedded signal processing libraries
• Real-time OS: FreeRTOS for concurrent task management
• Signal Processing: CMSIS-DSP and Eigen3 libraries for mathematical operations
• Simulation Tools: MATLAB for algorithm validation and LTspice for circuit design
• Analysis Tools: Python with signal processing libraries for post-processing validation
• Development Environment: CMake build system with GitHub version control
• Debugging: J-Link Edu with VSCode integration for real-time debugging

Implementation Details

The system architecture consists of several key components working together:

Hardware Signal Processing:

• Custom PCB design with non-inverting amplifiers providing optimized gain
• 8 kHz low-pass filters serving as anti-aliasing protection
• Variable amplifier stages with stable reference voltage signals

Embedded Software:

• DMA-based ADC sampling for maximum throughput (500 kHz total, 166.67 kHz per channel)
• Cross-correlation algorithms for time difference determination
• Newton-Raphson iterative solver for 2D position estimation
• FFT-based frequency detection for target signal validation

Technical Challenges Solved:

• Signal sensitivity optimization through calculated amplifier gain
• Noise reduction through digital filtering and frequency domain analysis
• Real-time processing constraints addressed with optimized algorithms
• Electromagnetic interference mitigation through proper PCB design considerations

Performance Results

The system achieved mixed results during evaluation:

Precision Metrics:

• Average X-coordinate error: 5.18 cm
• Average Y-coordinate error: 4.34 cm
• Success rate: 50% of measurements within 5.0 cm accuracy target

Key Findings:

• Best accuracy achieved when sound source is positioned within the triangular sensor array
• Performance degrades significantly for sources outside the sensor triangle
• Lower frequency signals (250 Hz) provide more reliable results than higher frequencies (4 kHz)
• Cross-correlation method sensitive to signal period relative to sensor spacing

System Limitations:

• Maximum effective distance limited by frequency: d = 1/(f × 340.29 × 100)
• Sensor casing may obstruct audio from certain angles
• Batch processing introduces timing artifacts affecting analysis

Future Improvements

Several enhancements have been identified for system optimization:

Hardware Enhancements:

• CAD-based PCB design with improved ground plane and signal integrity
• Active bandpass filtering instead of simple low-pass filtering
• Optimized sensor housing to reduce directional sensitivity

Software Optimizations:

• Complete integration of TDoA algorithms on embedded platform
• Real-time performance optimization for sub-5-second inference
• Enhanced noise filtering and signal validation algorithms

System Integration:

• Unified real-time processing pipeline eliminating post-processing dependencies
• Improved calibration procedures for sensor array geometry
• Advanced correlation techniques like GCC-PHAT for enhanced accuracy

Lessons Learned

This project provided valuable insights into embedded signal processing:

• Peak-based time difference methods proved unreliable for consistent localization
• Square wave signal sources (buzzers) create processing artifacts; sinusoidal sources preferred
• Batch sampling introduces timing challenges that must be carefully managed
• Cross-correlation requires careful consideration of signal period relative to sensor spacing
• Real-time embedded constraints demand algorithmic optimization and approximation techniques

The Sound Locator M33 project demonstrates the complexity and challenges inherent in real-time embedded signal processing while providing a solid foundation for future development in acoustic localization systems.

---

This project was completed as part of the Linear Systems - Project (5LIU0) course at Technical University of Eindhoven (TUe), Embedded Systems pre-master program, 2024-2025.