High Dimensional User-Specific Channel Generation for CSI Compression Using Diffusion Models

The compression of channel state information (CSI) using deep neural networks (DNNs) is emerging as an important approach for future wireless communication systems. However, training such models typically requires a large dataset of high-dimensional channel measurements, which are very difficult and expensive to obtain. This paper introduces a novel method for generating synthetic wireless channel data using diffusion models to produce user-specific channels that accurately reflect real-world wireless environments. Our approach employs a conditional denoising diffusion implicit models (cDDIM) framework, effectively capturing the relationship between user location and multi-antenna channel characteristics. We generate synthetic high-fidelity channel samples using user positions as conditional inputs, creating larger augmented datasets to overcome measurement scarcity. Our approach significantly improves over prior methods, such as adding noise or using generative adversarial networks (GANs), especially in scenarios with limited initial measurements. For example, we can reduce the required amount of training data by 100× while maintaining nearly the same normalized mean squared error (NMSE) for CSI compression.