CSI-4CAST: A Hybrid Deep Learning Model for CSI Prediction with Comprehensive Robustness and Generalization Testing

TL;DR

CSI-4CAST introduces a lightweight, hybrid DL architecture for DL CSI prediction that fuses CNN-based residuals, Adaptive Correction Layers, ShuffleNet blocks, and Transformer encoders to capture local and long-range dependencies with high robustness and efficiency. It is evaluated on CSI-RRG, a large-scale benchmark with 3,060 realistic scenarios spanning TDD and FDD, multiple channel models, delay spreads, velocities, and noise types, and outperforms baselines in most Regular and Robustness settings while significantly reducing FLOPs. The work provides comprehensive insights into how channel factors like velocity, delay spread, and SNR affect prediction and generalization, and it releases both the dataset and evaluation protocols to promote standardized benchmarking. Overall, CSI-4CAST advances robust and efficient CSI forecasting for next-generation wireless systems, with strong practical implications for real-time precoding and resource management.

Abstract

Channel state information (CSI) prediction is a promising strategy for ensuring reliable and efficient operation of massive multiple-input multiple-output (mMIMO) systems by providing timely downlink (DL) CSI. While deep learning-based methods have advanced beyond conventional model-driven and statistical approaches, they remain limited in robustness to practical non-Gaussian noise, generalization across diverse channel conditions, and computational efficiency. This paper introduces CSI-4CAST, a hybrid deep learning architecture that integrates 4 key components, i.e., Convolutional neural network residuals, Adaptive correction layers, ShuffleNet blocks, and Transformers, to efficiently capture both local and long-range dependencies in CSI prediction. To enable rigorous evaluation, this work further presents a comprehensive benchmark, CSI-RRG for Regular, Robustness and Generalization testing, which includes more than 300,000 samples across 3,060 realistic scenarios for both TDD and FDD systems. The dataset spans multiple channel models, a wide range of delay spreads and user velocities, and diverse noise types and intensity degrees. Experimental results show that CSI-4CAST achieves superior prediction accuracy with substantially lower computational cost, outperforming baselines in 88.9% of TDD scenarios and 43.8% of FDD scenario, the best performance among all evaluated models, while reducing FLOPs by 5x and 3x compared to LLM4CP, the strongest baseline. In addition, evaluation over CSI-RRG provides valuable insights into how different channel factors affect the performance and generalization capability of deep learning models. Both the dataset (https://huggingface.co/CSI-4CAST) and evaluation protocols (https://github.com/AI4OPT/CSI-4CAST) are publicly released to establish a standardized benchmark and to encourage further research on robust and efficient CSI prediction.

Figures (22)

• Figure 1: The channel aging problem.
• Figure 2: An illustration of the DL CSI acquisition schemas.
• Figure 3: An illustration of the CSI Prediction in the time-frequency domain.
• Figure 4: The proposed CSI-4CAST. Historical CSI is first processed by a CNN residual block, followed by an inverse DFT (IDFT) to obtain the delay-domain representation. Both frequency- and delay-domain features are then refined by ACL layers and passed through a ShuffleNet block. Finally, the Transformer block maps the embedded features to predict future CSI. The ACL2 layer (in gray) applies only to the FDD.
• Figure 5: TDD: NMSE rank distribution of Regular, Robustness, and Generalization. Within each panel, models are ordered left to right by their mean rank, $\mathrm{MeanRank}$ in \ref{['eq:mean-rank']} (lower is better). Rank distributions are shown as violin plots, while top-1 percentages, $\mathbf{P}_{\mathrm{rank1}}$ in \ref{['eq:rank-1-percentage']}, are plotted as a red line graph.