A Deep Neural Network-Based DOA Estimation Method for Sound-Guided Relative Localization of Micro-UAVs

Relative measurement technology for cooperative objectives is a crucial foundation for autonomous flight and collaborative task execution in Unmanned Aerial Vehicle (UAV) swarms. To achieve onboard relative positioning of micro-UAV swarms in complex environments, this work proposes a lightweight Deep Learning (DL)-based Direction of Arrival (DOA) estimation method for miniaturized UAV-embedded microphone arrays. A Deep Neural Network (DNN) model consisting of one convolution block and four residual blocks is developed. The spatial covariance matrix of sound signals is reshaped as input feature vector to improve the generalization ability of the DNN model. We have conducted real-world experiments on a 2×2 cm UAV-embedded microphone array. Experimental results indicate that the proposed model achieves an average error of less than 10° in low signal-noise ratio (SNR) environment, demonstrating significant improvements over traditional methods and Convolutional Neural Network (CNN).