RadioDiff: An Effective Generative Diffusion Model for Sampling-Free Dynamic Radio Map Construction

TL;DR

The sampling-free RM construction is modeled as a conditional generative problem, where a denoised diffusion-based method, named RadioDiff, is proposed to achieve high-quality RM construction and a comprehensive theoretical analysis of why the RM construction is a generative problem is provided for the first time.

Abstract

Radio map (RM) is a promising technology that can obtain pathloss based on only location, which is significant for 6G network applications to reduce the communication costs for pathloss estimation. However, the construction of RM in traditional is either computationally intensive or depends on costly sampling-based pathloss measurements. Although the neural network (NN)-based method can efficiently construct the RM without sampling, its performance is still suboptimal. This is primarily due to the misalignment between the generative characteristics of the RM construction problem and the discrimination modeling exploited by existing NN-based methods. Thus, to enhance RM construction performance, in this paper, the sampling-free RM construction is modeled as a conditional generative problem, where a denoised diffusion-based method, named RadioDiff, is proposed to achieve high-quality RM construction. In addition, to enhance the diffusion model's capability of extracting features from dynamic environments, an attention U-Net with an adaptive fast Fourier transform module is employed as the backbone network to improve the dynamic environmental features extracting capability. Meanwhile, the decoupled diffusion model is utilized to further enhance the construction performance of RMs. Moreover, a comprehensive theoretical analysis of why the RM construction is a generative problem is provided for the first time, from both perspectives of data features and NN training methods. Experimental results show that the proposed RadioDiff achieves state-of-the-art performance in all three metrics of accuracy, structural similarity, and peak signal-to-noise ratio. The code is available at https://github.com/UNIC-Lab/RadioDiff.

Figures (6)

• Figure 1: Illustration of the RM, where the yellow elements represent the heatmap of pathloss; the brighter the yellow, the higher the pathloss. The red elements denote vehicles, and the blue elements are static buildings. Since, static buildings can completely block electromagnetic signals from entering their interiors, resulting in an internal pathloss of zero, to intuitively represent the impact of static buildings on the RM, we have colored them blue.
• Figure 2: The diffusion procedure of RM, where in diffusion procedure the RM is diffused into noise, and in the denoising procedure the RM is revealed from the noise and prompt.
• Figure 3: The illustration of the proposed RadioDiff framework. The VAE is employed to encode the RM into a latent vector, thereby reducing the dimension of the input/output space for the denoise diffusion model. The framework incorporates a U-Net architecture, consisting of an encoder and decoder, to facilitate the denoising process. The prompt is represented as a grayscale diagram with three channels, each channel depicting the features of buildings, vehicles, and AP. After encoding the prompt, it is concatenated into the U-Net network, enabling the model to generate RMs under environmental conditions.
• Figure 4: The comparisons of constructed SRM on different methods.
• Figure 5: The comparisons of constructed DRM on different methods.