| PhD Seminar

Name of the Speaker: Mr. Ravi Kumar (EE20D004)
Guide: Dr. Manivasakan R
Venue: ESB-244 (Seminar Hall)
Date/Time: 8th October 2025 (Wednesday), 3:30 PM
Title: An ERAN-Based Dynamic Graph Neural Network for CSI Prediction in Massive MIMO Systems

Abstract :

Accurate channel state information (CSI) prediction is critical for enabling reliable and high-capacity communication in massive MIMO systems, especially under high user mobility. This seminar presents the Evolving Relational Attention Network (ERAN)—a dynamic spatio-temporal graph neural network—for CSI prediction. Unlike static graph models such as the Spectral-Temporal Graph Neural Network (STEM GNN), ERAN leverages a GRU-driven evolving graph learning module to capture real-time topological dependencies induced by mobility, fading, and scattering.

Experimental results on high-mobility scenarios (μ = 120 km/h) demonstrate that ERAN achieves a 34.93% reduction in RMSE compared to STEM, while limiting spectral efficiency degradation to only 5.64%, nearly three times more robust than STEM under similar conditions. These findings highlight ERAN’s ability to generalize over time by effectively modeling fast-varying channel dynamics caused by Doppler shifts and environmental variability.

By bridging advances in spatio-temporal graph learning with wireless channel modeling, this work demonstrates the potential of dynamic GNNs to push the frontier of intelligent CSI prediction in next-generation wireless systems.