CGAN-Based Data Augmentation for Enhanced Channel Prediction in Massive MIMO Under Subway Tunnels

Massive multiple-input multiple-output (MIMO) channel modeling poses a crucial yet challenging task for wireless communication systems under subway tunnel scenarios. Due to the high cost associated with on-site measurement campaigns, the amount of channel data is often limited and insufficient. This paper proposes a conditional generative adversarial network (cGAN)-based channel model that can learn from the limited measurement data and generate realistic samples of channel impulse response (CIR) matrices specific to given conditions. The proposed channel model provides a data augmentation approach to enhance the training of neural networks for channel prediction, overcoming the prediction performance loss caused by insufficient number of training samples. Simulation results validate that the generated data exhibits high consistency with the actual channel and accurately captures the statistical properties in both delay and spatial domains. By expanding the available training dataset effectively, it is shown that the accuracy of neural network-based channel prediction has improved significantly with the proposed channel data augmentation.