Enhanced Transformer-Based Gated Recurrent Unit Predictive Channel Model for 6G Prediction

In recent scenarios, the rapid evolution of sixthgeneration (6 G) wireless communication has demanded accurate and adaptive channel prediction to support ultrareliable and high-mobility services. Traditional Gated Recurrent Unit (GRU)-based channel modelling approaches struggle to capture long-range spatial-temporal dependencies and generalize well in data-scarce environments. Hence, this research proposes an Enhanced Transformer-based GRU Predictive Channel Model (ETG-PCM) for robust space-time channel forecasting. Initially, Channel Impulse Response (CIR) data were collected from real-world environments and preprocessed through normalization and data augmentation. To mitigate data scarcity, a Space-Time Conditional Generative Adversarial Network (STGAN) was employed to generate synthetic CIRs based on channel characteristics. Subsequently, a transformer encoder with positional encoding was used to extract global spatial-temporal dependencies. Then, the data are classified into line-of-sight (LoS) and non-LoS (NLoS) paths using power spectral density, and temporal sequence modelling is performed using parallel GRU decoders. Finally, the decoded outputs were fused to predict future CIR values across space and time. The proposed ETG-PCM attained better results in terms of accuracy (85.9 %) and recall (87.0 %) when compared to the existing Convolutional Neural Networks and Customized Vision Transformer (CNN-ViT) model.