Diffusion Model Based Resource Allocation Strategy in Ultra-Reliable Wireless Networked Control Systems

Diffusion models offer a promising alternative to Deep Reinforcement Learning (DRL) for resource allocation in wireless networks due to their capability to model complex data distributions with greater accuracy, yet their potential remains largely unexplored. This letter proposes a diffusion model-based approach for Wireless Networked Control Systems (WNCSs) to minimize power consumption by optimizing the sampling period, blocklength, and packet error probability within the finite blocklength regime. The problem is simplified to optimizing blocklength through optimality conditions, and a dataset of channel gains and optimal blocklengths is generated via an optimization theory-based solution. The Denoising Diffusion Probabilistic Model (DDPM) is employed to generate optimal blocklength values, conditioned on channel state information (CSI). The core idea is to train the diffusion model to generate blocklength values from noise, essentially replicating the process by which the optimization solution is derived. Extensive simulations reveal that the proposed approach surpasses existing DRL-based methods, achieving near-optimal performance in terms of total power consumption. Additionally, the proposed method reduces critical constraint violations by up to eighteen times, further highlighting the enhanced accuracy of the solution.