Intelligent Beamforming for UAV-Assisted IIoT Based on Hypergraph Inspired Explainable Deep Learning

The development of industry is inseparable from the development of Industrial Internet of Things (IIoT), with the continuous development of artificial intelligence (AI), deep learning enabled intelligent communication has been widely studied. While, most existing neural networks for the current research often have poor performances on scalability and universality issues. In our previous work [1], a wireless communication scenario for UAV-assisted industrial IoT was considered, the beamforming problem in the above scenario was investigated and an unsupervised learning framework is introduced with mixed interference based graph neural network (MIGNN) to deal with the problem. However, it did not take into account the fact that different power constraints may have an impact on the performance of the model. This paper enriches the theoretical derivation and analysis, adds simulation experiments based on different power constraints and performance comparison analysis. Specifically, we conduct indepth analysis of the model's performance by adding two sets of comparative experiments, which can verify the stability and reliability of the model. Firstly, we constructed a system model for UAV assisted IIoT communication and modeled the beamforming problem. Secondly, we constructed heterogeneous graphs based on link state information. Then, a MIGNN model combining hypergraphs was proposed to obtain the optimal beamforming results. At last, from the experimental results, it can be verified that comparisons with traditional deep learning methods, the proposed one has stronger scalability and the results of the comparative experiments can successfully verifies the stability as well as the reliability of the model. Compared with our previous work, the simulation scenarios considered in this paper are more comprehensive and can be better applied to real wireless communication scenarios.

I. Introduction

The Industrial Internet of Things (IIoT) is the use of connected machines, devices and sensors in industrial applications. As IIoT continues to grow, the need for digital technology as well as artificial intelligence rises. In this context, Industry 5.0 emerges to promote a cooperative rather than a competitive relationship between humans and artificial intelligence [2]. Especially, the resource allocation of IIoT towards Industry 5.0 is indispensable and plays a vital role, to improve the spectrum efficiency, energy efficiency, capacity, robustness and security through beamforming [3], power control [4], channel assignment [5] and so on. Unfortunately, most resource allocation problems like maximizing weighted sum rate (WSR) and energy efficiency (EE) are non-convex, proved to be Non-deterministic Polynomial (NP) hard for achieving optimal or suboptimal solutions. Thus, it would bring out high costs on computation and time by using traditional optimization algorithms. Especially, considering the decreasing delay requirements for machine to machine (industrial M2M) communications within IIoT, the computation efficiency of the resource optimization problems need to be intensively considered in a real-time manner [6]. In addition, in IIoT environment, the exchange of information is always present and inevitable. In order to ensure the accuracy and real-time information transmission in the communication process, we must meet the reliability, security, effectiveness and intelligence requirements of IIoT, and only when the four indicators are met, the information transmission process can be guaranteed to a certain extent [7], [8], [9]. However, the existing communication technology is difficult to meet these indicators requirements. Therefore, it is necessary to apply advanced B5G and 6G technologies, such as cell free, millimeter wave, RIS and NOMA, to improve the system performances of IIoT towards Industry 5.0 [10], [11].