Beamforming-Based Multicast Routing Protocol in Underlay Cognitive MANETs With STAR-RIS: Deep Learning Design
In this article, we propose a novel deep learning (DL) and beamforming-based multicast routing (BMR) protocol in underlay cognitive mobile ad hoc networks (MANETs) with simultaneously transmitting and reflecting-reconfigurable intelligent surfaces (STAR-RISs). The proposed BMR protocol employs a residual deep neural network (ResDNN)-based clustering (RDC) protocol as the underlying structure to improve stability and minimize control overhead, while jointly integrating beamforming and cross-layer design to improve energy efficiency (EE) during secondary user (SU) routing. Specifically, the beamforming strategy dynamically partitions the transmission area into regions and zones according to the receiver positions. In addition, a cross-layer design leverages contextual information from both physical and network layers to enhance quality-of-service (QoS) performance in terms of packet delivery ratio (PDR), routing delay, control overhead, spectral efficiency (SE), and EE. We formulated the min–min problem to select the optimal zone and region, and the weight-sum problem for clustering and routing, and solved them using ResDNN. Furthermore, STAR-RIS is employed to improve connectivity between nodes. In addition, we design a DL framework to address three key problems: 1) infer the optimal weight values for cluster formation; 2) infer the optimal weight values for multicast route selection; and 3) determine the optimal region and zone for the beamforming strategy. The numerical results show that the combination of the proposed BMR and RDC protocols enables more robust multicast routing, outperforming the benchmark protocols in terms of connectivity, scalability, SE, and EE.