Spatial Sound Field Modeling and Noise Cancellation for Drones

Abstract

Abstract This thesis advances the modeling and reduction of drone noise through the use of on-board microphones and directional active noise control (ANC) systems. In recent years, small-scale unmanned aerial vehicles (UAVs), commonly known as drones, have become increasingly prevalent across a wide array of applications, including parcel delivery, mapping, and emergency response. Despite their growing utility, drone operations contribute significantly to environmental noise pollution, posing risks to both daily comfort and public health. Consequently, the development of effective drone noise mitigation techniques has become a pressing concern. To address this challenge, the thesis first introduces novel methods for drone noise modeling based solely on on-board microphone measurements. These methods generalise the drone noise modeling task as a sound field extrapolation problem. A sound field extrapolation framework based on spherical sector harmonics is proposed, and a comprehensive theoretical analysis is conducted to support its application to sectorial sound field reconstruction. Furthermore, the method is enhanced through the derivation of an explicit mapping between spherical and spherical sector harmonics, improving both accuracy and flexibility. To overcome geometric limitations in microphone placement, we further develop a physics-informed neural network (PINN) architecture called the point neuron net- work. This model embeds the fundamental solution of the wave equation into its structure, ensuring physical consistency. The network handles complex-valued sig- nals directly, improves interpretability, and generalizes effectively across various acoustic conditions. Unlike existing approaches, it does not rely on pre-collected datasets and remains robust to sensor noise and irregular sensor configurations. Building upon the proposed modeling approaches, we develop two directional drone ANC systems that target noise propagation in specific far-field regions. The first sys- tem is based on the spherical sector harmonics extrapolation framework and incorporates constraints on payload and power consumption. We analyze the secondary- to-primary source power ratio for different secondary source configurations by simulations. Additionally, we enhance the ANC algorithm to relax the geometric requirements of the monitoring microphone array and validate the system's performance through hardware experiments. Complementing these systems, we introduce a propeller phase control method aimed at reducing tonal drone noise in the downward far-field region. By adjusting the relative phase of multiple propellers, this technique achieves destructive interference in targeted directions, thereby reducing noise radiation without additional hardware. Overall, the contributions presented in this thesis constitute a comprehensive framework for drone noise modeling and directional noise reduction, offering physically grounded, practically feasible, and experimentally validated solutions to a critical problem in drone operations. Furthermore, the point-neuron learning framework and the spherical sector harmonic-based extrapolation method presented in this thesis are applicable to a broad range of sound field-related problems, including sound field reproduction, interpolation, and extrapolation. Show more