A Novel Acoustic Source Localization Technique for Edge AI Applications: A Lightweight Framework and Implementation for IoT and Smart Sensing Devices

Abstract

This paper presents a novel and computationally efficient three-point signal estimation method for acoustic direction finding, designed specifically for low-cost embedded platforms. The proposed approach offers a lightweight alternative to traditional cross-correlation techniques by minimizing computational complexity while preserving high angular resolution. The method was implemented and tested on an STM32F429 microcontroller using a pair of MAX4466 electret microphones arranged on a fixed baseline. The system architecture leverages bare-metal signal processing routines optimized with Acorn RISC Machine Cortex. Microcontroller Software Interface Standard (ARM CMSIS-DSP) libraries, enabling real-time performance on resource-constrained hardware. Extensive experiments were conducted to evaluate the angular estimation accuracy under varying signal-tonoise ratios and source orientations. Results show that the system maintains sub-degree mean square error for source angles up to 30°, with noticeable performance degradation observed at 40° due to the directional response characteristics of the microphone elements. A comprehensive explanation is provided linking this degradation to reduced microphone sensitivity at wider angles of incidence. The proposed solution is ideal for applications requiring embedded acoustic localization, including smart interfaces, vehicular monitoring, and surveillance systems. In addition, the paper discusses the implications of deploying such systems in artificial intelligence (AI)-enabled and security-critical environments, highlighting emerging threats such as adversarial acoustic interference and spoofing attacks. These challenges underscore the importance of resilient and efficient DF methods that can operate reliably within the constraints of embedded systems. The presented work lays the foundation for future research in secure, scalable, and AI-compatible acoustic sensing platforms. © 2025 Elsevier B.V., All rights reserved.