RHS-based Robust Hybrid Cooperative Beamforming in mmWave IoV : A Lifelong Graph Learning-based Method

Meng, Chanyuan and Xiong, Ke and Gao, Bo and Ni, Qiang and Fan, Pingyi and Ng, Derrick Wing Kwan and Ai, Bo and Letaief, Khaled Ben (2026) RHS-based Robust Hybrid Cooperative Beamforming in mmWave IoV : A Lifelong Graph Learning-based Method. IEEE Transactions on Mobile Computing. ISSN 1536-1233

Text (Author final)
Author_final.pdf - Accepted Version
Available under License Creative Commons Attribution-NonCommercial.
Download (9MB)

Abstract

As a promising candidate for next-generation reconfigurable antennas (NGRAs), the reconfigurable holographic surface (RHS) offers significant advantages in reducing hardware costs and power consumption. To enable green millimeter-wave (mmWave) Internet of Vehicles (IoV) operations in dynamic environments, this paper proposes an RHS-based robust hybrid cooperative beamforming (HCBF) transmission scheme, integrated with a machine learning (ML)-based optimization framework, to minimize the total transmit power at the base station (BS) while taking into account imperfect channel state information (CSI). First, by leveraging the unique advantages of RHS, the proposed HCBF fully exploits multi-antenna collaborative gain, thereby effectively reducing overall power consumption of the IoV system. To overcome the limitations of traditional ML-based optimization methods, which require retraining the neural network (NN) from scratch whenever the IoV network topology varies due to the dynamic entry or exit of mobile vehicles (MVs), a lifelong graph learning-based HCBF optimization method (LGL-HCBF) is developed. Specifically, in LGL-HCBF, an incremental graph convolutional network (IGCN) is employed to facilitate parameter sharing and local structural information exchange across different graph structures, thus providing stronger adaptability to dynamic variations in IoV network topology. To further address the potential stability-plasticity dilemma of LGL-HCBF, two complementary mechanisms, namely, knowledge replay and weight consolidation, are introduced to preserve the knowledge learned from previous scenarios. Simulation results demonstrate that the proposed RHS-based HCBF achieves much lower power consumption compared with traditional phased-array beamforming schemes under the same hardware-cost budget, deployment scales, and rate constraints. Moreover, the proposed LGL-HCBF exhibits faster adaptation, and superior approximation ratio, outperforming several baseline methods.