High-Precision Indoor Localization via Dual-Modal AOA/TOA Fusion with Deep Learning and Particle Filters

Abstract

As the era of IoT and artificial intelligence advances, the demand for high-precision indoor positioning systems continues to grow. Achieving accurate positioning in indoor environments remains challenging due to the presence of obstacles and signal interference, especially in Non-Line-of-Sight (NLOS) conditions. To address these challenges, this paper proposes a novel indoor positioning algorithm based on the fusion of Angle of Arrival (AOA) and Time of Arrival (TOA) data. A hybrid model combining Asymptotic Gradient Boosted Regression Trees (GBRT) and Elastic Net (EN) is used to reduce AOA measurement errors in NLOS environments, followed by the application of the Levenberg-Marquardt (LM) optimization algorithm to enhance localization accuracy. Experimental results show a significant reduction in positioning error, with an average error of 0.47 meters, representing a 41.25% improvement compared to the KF+WLS algorithm. Meanwhile, to improve TOA positioning, a deep learning-based TOA fingerprinting algorithm is proposed, this algorithm captures complex spatiotemporal features in TOA data, leading to a 25.00% and 15.22% reduction in root mean square error (RMSE) compared to the WKNN and WLS algorithms, respectively. Finally, a fusion strategy based on Particle Filtering (PF) is introduced to combine AOA and TOA data, achieving further RMSE reductions of 35.42% and 20.51%, compared to individual AOA and TOA methods.