Multi-Agent Reinforcement Learning for Power Control in Wireless Networks via Adaptive Graphs

The ever-increasing demand for high-quality and heterogeneous wireless communication services has driven extensive research on dynamic optimization strategies in wireless networks. Among several possible approaches, multi-agent deep reinforcement learning (MADRL) has emerged as a promising method to address a wide range of complex optimization problems like power control. However, the seamless application of MADRL to a variety of network optimization problems faces several challenges related to convergence. In this paper, we present the use of graphs as communication-inducing structures among distributed agents as an effective means to mitigate these challenges. Specifically, we harness graph neural networks (GNNs) as neural architectures for policy parameterization to introduce a relational inductive bias in the collective decision-making process. Most importantly, we focus on modeling the dynamic interactions among sets of neighboring agents through the introduction of innovative methods for defining a graph-induced framework for integrated communication and learning. Finally, the superior generalization capabilities of the proposed methodology to larger networks and to networks with different user categories is verified through simulations.

An End-To-End Analysis of Deep Learning-Based Remaining Useful Life Algorithms for Satefy-Critical 5G-Enabled IIoT Networks

Remaining Useful Life (RUL) prediction is a critical task that aims to estimate the amount of time until a system fails, where the latter is formed by three main components, that is, the application, communication network, and RUL logic. In this paper, we provide an end-to-end analysis of an entire RUL-based chain. Specifically, we consider a factory floor where Automated Guided Vehicles (AGVs) transport dangerous liquids whose fall may cause injuries to workers. Regarding the communication infrastructure, the AGVs are equipped with 5G User Equipments (UEs) that collect real-time data of their movements and send them to an application server. The RUL logic consists of a Deep Learning (DL)-based pipeline that assesses if there will be liquid falls by analyzing the collected data, and, eventually, sending commands to the AGVs to avoid such a danger. According to this scenario, we performed End-to-End 5G NR-compliant network simulations to study the Round-Trip Time (RTT) as a function of the overall system bandwidth, subcarrier spacing, and modulation order. Then, via real-world experiments, we collect data to train, test and compare 7 DL models and 1 baseline threshold-based algorithm in terms of cost and average advance. Finally, we assess whether or not the RTT provided by four different 5G NR network architectures is compatible with the average advance provided by the best-performing one-Dimensional Convolutional Neural Network (1D-CNN). Numerical results show under which conditions the DL-based approach for RUL estimation matches with the RTT performance provided by different 5G NR network architectures.