Ultra-wideband digital channel modeling based on generative adversarial network

Abstract

In ultra-wideband communication technology, high-quality channel impulse response data is crucial for system design and performance optimization. A least squares generative adversarial network (LSGAN) and an improved loss function were introduced, which significantly enhanced the ability to capture and reproduce channel data. By combining feature matching techniques with conditional generative adversarial networks (CGAN), it was able to improve the detail accuracy and diversity of the generated data. The model was allowed to generate data according to different communication environments and signal scenarios. During the model training phase, reconstructed channel data representing global features were used, while actual channel data experiencing wireless fading were employed during the testing phase. Experimental results demonstrate that the model outperforms the WGAN-GP in small sample datasets and complex fading channel environments, with a 4.8% increase in recognition accuracy and a 5% reduction in mode collapse issues.

Keywords

• digital twin
• channel modeling
• generative adversarial network
• intelligent communication network