Channel-Aware Conditional Diffusion Model for Secure MU-MISO Communications

TL;DR

The work targets physical-layer security in MU-MISO by maximizing the sum secrecy rate $R_{sum}^{sec} = \sum_k R_k^{sec}$ with $R_k^{sec} = (R_k - \max_l R_{l,k}^{e})^+$, where $R_k = \log_2\left(1 + \dfrac{|h_k^H w_k|^2}{\sum_{i \neq k} |h_k^H w_i|^2 + \sum_j |h_k^H v_j|^2 + \sigma_0^2 / P}\right)$ and $R_{l,k}^{e} = \log_2\left(1 + \dfrac{|h_{E,l}^H w_k|^2}{\sum_{i \neq k} |h_{E,l}^H w_i|^2 + \sum_j |h_{E,l}^H v_j|^2 + \sigma_0^2 / P}\right)$. The proposed channel-aware conditional diffusion model learns $P(\mathbf{u}|\mathbf{h})$ using a denoising U‑Net with CSI-embedded conditioning, enabling fast sampling of multiple high-quality beamforming and AN vectors via DDIM. Training combines a diffusion-learning stage with a mean-squared-error loss on predicted noise and a subsequent fine-tuning stage that optimizes the sum secrecy rate with a power-constraint penalty, producing the CDM-F variant. Empirical results show that CDM and CDM-F outperform traditional optimization (OPT) and fixed-AN baselines (RZF-NS) across varying user/eavesdropper configurations and noise levels, demonstrating robustness and potential for real-time physical-layer security in dynamic networks.

Abstract

While information securityis a fundamental requirement for wireless communications, conventional optimization based approaches often struggle with real-time implementation, and deep models, typically discriminative in nature, may lack the ability to cope with unforeseen scenarios. To address this challenge, this paper investigates the design of legitimate beamforming and artificial noise (AN) to achieve physical layer security by exploiting the conditional diffusion model. Specifically, we reformulate the security optimization as a conditional generative process, using a diffusion model to learn the inherent distribution of near-optimal joint beamforming and AN strategies. We employ a U-Net architecture with cross-attention to integrate channel state information, as the basis for the generative process. Moreover, we fine-tune the trained model using an objective incorporating the sum secrecy rate such that the security performance is further enhanced. Finally, simulation results validate the learning process convergence and demonstrate that the proposed generative method achieves superior secrecy performance across various scenarios as compared with the baselines.

Figures (4)

• Figure 1: Downlink MU-MISO wiretap system with joint beamforming and AN
• Figure 2: Channel-conditional diffusion model training and fine-tuning architecture for secure MU-MISO systems.
• Figure 3: Convergence of model training and fine-tuning.
• Figure 4: Performance under different algorithms.