A NAS Method for Cross-room NLOS Acoustic Signal Identification via Domain Adaptation

Acoustic signals offer distinct advantages for indoor positioning systems (IPS), combining compatibility and cost-effectiveness. However, Non-Line-of-Sight (NLOS) conditions caused by obstructed signal paths introduce critical ranging errors that degrade positioning accuracy. Effective LOS/NLOS identification is therefore essential for reliable positioning. While deep learning models like CNNs and LSTMs excel at NLOS recognition, their environment-dependent nature requires costly retraining for new spaces. A semi-supervised cross-room identification method integrating domain adaptation with Neural Architecture Search (NAS)-optimized networks is proposed, achieving 99% (source) and 95% (target domain) accuracy with reduced environmental dependence.