Year: 2024
Type: conference-paper
Abstract: This paper presents an innovative approach for Channel State Information (CSI) prediction using Long-Short Term Memory (LSTM) networks in the frequency domain, addressing selective fading in mobile scenarios. Unlike models that assume perfect CSI, our method relies on realistic, estimated CSI and introduces strategies to mitigate estimation errors. We propose an adaptive prediction technique leveraging LSTM for precise 3D Downlink (DL) transmission control, alternating between previously predicted and high-accuracy estimates. Our system supports both open-loop and closed-loop predictions, enhancing resource allocation and signal reliability in Non-terrestrial Network (NTN). Experimental results demonstrate significant improvements in prediction accuracy, meeting the demands of complex Geometry-based Stochastic Channel Modeling (GSCM) over simpler 2D stochastic channel models with predefined temporal correlation, in compliance with the 3GPP 38.901 standard.
Authors: Mervat Zarour, Qiuheng Zhou, Sergiy Melnyk, Hans D. Schotten
Institution: German Research Centre for Artificial Intelligence
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W4406266762