Core Concepts
The author presents a novel approach using complex-valued neural networks in federated learning to optimize indoor positioning performance, addressing challenges faced by traditional methods.
Abstract
The article explores the use of complex-valued neural networks in federated learning for indoor positioning optimization. It compares the proposed algorithm with real-valued methods and demonstrates significant improvements in accuracy. The study focuses on reducing mean positioning errors and enhancing user privacy.
The research introduces a new framework for indoor positioning using channel state information (CSI) and complex-valued neural networks. It highlights the advantages of distributed machine learning without compromising data privacy. By directly processing complex-valued CSI data, the proposed algorithm achieves better accuracy compared to traditional methods that require data transformation.
The study evaluates the performance of the designed algorithm through simulation results, showcasing its ability to reduce positioning errors by up to 36%. The approach offers a promising solution for accurate indoor positioning without compromising user data privacy.
Stats
Simulation results demonstrate up to 36% reduction in mean positioning error.
47600 CSI samples used for training.
25201 data samples from Cellular Ultra Dense CSI Dataset.
Parameters: T=85, η=1×10^-4, OI=4, S1=2, P I=5, SI=1, OII=8, S2=2, P II=9, SII=2, NI=64, NII=32.