Knowledge-Infused Diffusion Model for Effective Radio Spectrum Map Construction

Radio Spectrum Map (RSM) is crucial for enabling generative communications and dynamic spectrum management in 6G. However, existing RSM construction methods face a persistent trade-off: model-driven methods exhibit low computational complexity but may be limited in complex environments. Conversely, data-driven methods suffer from high model complexity, which significantly reduces inference speed. To address these issues, we propose a Knowledge-Infused Diffusion Model (KIDM) for constructing RSM. Specifically, we design a conditional module that integrates inverse distance weighting (IDW) priors via cross-attention mechanisms, guiding the model to capture fine-grained spatial features. Moreover, we employ a lightweight acceleration strategy to optimize the denoising schedule, significantly expediting the reverse generation process. Extensive experiments on two open-source datasets demonstrate that KIDM outperforms the compared baseline methods in average RMSE across different sampling patterns and rates, and achieves a single inference latency of less than 5 ms, meeting the latency requirements of typical real-time 6G applications.