Acoustic NLOS Signal Recognition Based on 1-D Convolutional Neural Network With Channel Attention Mechanism

The precise recognition of acoustic non-line-of-sight (NLOS) signal is essential for precise and robust position measurement in acoustic indoor localization instruments. Conventional approaches, which depend on manual feature extraction and threshold-based methods, often prove to be labor-intensive and lack flexibility. Furthermore, the inherent complexity and variability of indoor settings pose significant challenges in achieving high precision and generalizability in acoustic NLOS signal recognition. This article presents a novel 1-D convolutional neural network (1D-CNN) that integrates a channel attention mechanism (CAM) and a long short-term memory (LSTM) network to enhance acoustic NLOS signal classification. The CAM enhances prominent features by modulating the feature weights at the channel level and the LSTM network is used to capture dynamic temporal dependencies, thus ensuring precise signal recognition. The proposed model enables automated feature extraction and adapts to various indoor environments. The experimental results show that the proposed model achieves an NLOS recognition accuracy of 98.7% when using the identical dataset for both training and testing. Furthermore, the NLOS recognition accuracy attains 97.7% when evaluated with new test data acquired from various indoor environments, demonstrating superior performance compared to established benchmark classification algorithms.