RF-Diffusion is a novel generative model designed to create high-quality, time-series radio frequency (RF) signals. Inspired by the success of diffusion models in computer vision and natural language processing, the authors adapt these models to the RF domain, addressing the limitations of existing RF-oriented generative solutions. The key contributions of RF-Diffusion include: 1. **Time-Frequency Diffusion Theory**: This theory enhances the original diffusion model to capture information from both the time and frequency domains of RF signals, enabling more accurate and diverse signal generation. 2. **Hierarchical Diffusion Transformer (HDT)**: A restructured DNN model that aligns with the Time-Frequency Diffusion theory, incorporating a hierarchical architecture, attention-based diffusion blocks, and complex-valued operations to generate high-quality RF data. 3. **Performance Evaluation**: Extensive experiments demonstrate that RF-Diffusion outperforms existing generative models (DDPM, DCGAN, CVAE) in synthesizing Wi-Fi and FMCW signals, achieving an average structural similarity of 81% and a significant improvement in channel estimation tasks. 4. **Case Studies**: RF-Diffusion is applied to enhance Wi-Fi gesture recognition and 5G FDD channel estimation, showing substantial performance improvements in both areas. The paper provides a comprehensive overview of RF-Diffusion, including its theoretical foundations, architectural design, and practical implementation, making it a valuable contribution to the field of RF signal generation and wireless systems.RF-Diffusion is a novel generative model designed to create high-quality, time-series radio frequency (RF) signals. Inspired by the success of diffusion models in computer vision and natural language processing, the authors adapt these models to the RF domain, addressing the limitations of existing RF-oriented generative solutions. The key contributions of RF-Diffusion include: 1. **Time-Frequency Diffusion Theory**: This theory enhances the original diffusion model to capture information from both the time and frequency domains of RF signals, enabling more accurate and diverse signal generation. 2. **Hierarchical Diffusion Transformer (HDT)**: A restructured DNN model that aligns with the Time-Frequency Diffusion theory, incorporating a hierarchical architecture, attention-based diffusion blocks, and complex-valued operations to generate high-quality RF data. 3. **Performance Evaluation**: Extensive experiments demonstrate that RF-Diffusion outperforms existing generative models (DDPM, DCGAN, CVAE) in synthesizing Wi-Fi and FMCW signals, achieving an average structural similarity of 81% and a significant improvement in channel estimation tasks. 4. **Case Studies**: RF-Diffusion is applied to enhance Wi-Fi gesture recognition and 5G FDD channel estimation, showing substantial performance improvements in both areas. The paper provides a comprehensive overview of RF-Diffusion, including its theoretical foundations, architectural design, and practical implementation, making it a valuable contribution to the field of RF signal generation and wireless systems.