NLOS Signal Identification for Acoustic Indoor Localization Using a Smartphone’s Dual Microphones and Trainable FrFT

Acoustic indoor positioning technologies are gaining attention for their low cost and high accuracy. However, most ranging-based localization technologies suffer from significant distance measurement errors caused by Non-Line-of-Sight (NLOS) signal propagation, making the accurate identification of NLOS signals crucial for achieving precise localization. Existing NLOS identification methods either rely on carefully crafted handcrafted features or perform poorly in cross-scenario environments. This paper presents a novel NLOS identification algorithm that automatically extracts key features from dynamic acoustic signals for end-to-end recognition. Dual-channel data is collected using a smartphone’s microphones, and a trainable fractional Fourier transform (FrFT) is applied for feature extraction, followed by NLOS identification using a convolutional neural network (CNN) based on the ResNet architecture. Experimental results show that the proposed method outperforms existing algorithms, achieving the best performance metrics in cross-scenario experiments with an average accuracy of 99.34%.