A Hybrid Gnn-lstm Model For Feedback-free Adaptive Modulation And Coding In Massive Mimo

Abstract

Adaptive Modulation And Coding (AMC) Is A Key Enabler For Achieving High Spectral Efficiency And Reliability In Massive Multiple-Input Multiple-Output (MIMO) Systems. Conventional AMC Techniques, However, Rely Heavily On Explicit Channel State Information (CSI) Feedback, Which Introduces Latency, Signaling Overhead, And Performance Degradation Under Fast-varying Channels. To Overcome These Limitations, This Work Proposes A Hybrid Graph Neural Network–Long Short-Term Memory (GNN-LSTM) Model For Feedback-free AMC In Massive MIMO. The Proposed Framework Exploits The Graph-based Representation Of User–antenna Connectivity To Capture Spatial Correlations, While The LSTM Layers Effectively Learn Temporal Dependencies In Channel Variations. Simulation Results Demonstrate That The GNN-LSTM Outperforms CNN, LSTM, And CNN-LSTM Architectures In Terms Of Testing Accuracy, Achieving More Than 95% Prediction Accuracy For Modulation And Coding Scheme (MCS) Classification. The Training And Testing Curves Confirm Strong Generalization Capability With Stable Convergence Across 300 Epochs, While Histogram Analysis Validates The Correct Distribution Of Predicted MCS Classes. Compared To Conventional Learning Models, The Hybrid GNN-LSTM Significantly Reduces Feedback Dependency, Enabling Robust And Intelligent AMC Decision-making For Next-generation Wireless Networks. This Work Highlights The Potential Of Combining Spatial–temporal Deep Learning Techniques To Enhance The Efficiency Of Massive MIMO Systems In 5G And Beyond..

Keywords