Enhanced synthetic generation of channel state information for millimeter-wave networks in 5G communication systems

Abstract

In 5G communication systems, millimeter-wave networks are pivotal, relying heavily on Channel State Information (CSI) for effective user-to-base station (BS) transmission. However, the acquisition of genuine CSI data remains a hurdle, often due to the expenses associated with simulations or physical experiments. This paper introduces an innovative method for generating artificial CSI data from real datasets, aiming to closely replicate authentic CSI samples. The procedure begins with an initial clustering analysis, followed using Principal Component Analysis and Uniform Manifold Approximation and Projection to reduce dimensionality. Then, the data distributions are transformed into multivariate normal distributions using Probability Integral Transformations (PIT). For data synthesis, Multilayer Perceptron based regression models are utilized, followed by inverse PIT transformations to return the data to its original space. Our method is compared against KDE-based algorithms, demonstrating superior fidelity in reproducing real CSI samples. Additionally, we stress the importance of capturing CSI correlations among different BSs to refine data synthesis. This research propels forward data synthesis techniques, offering potential solutions for mitigating interference challenges in dense MMW networks and advancing 5G communication systems.