Cellular Network Optimization Using Unfolding-Based Graph Neural Networks

The rapid development of wireless communication networks has driven the need for careful optimization of network parameters to achieve optimal performance. Traditional optimization approaches, however, face challenges in managing a vast number of tunable parameters and scalability across different scenarios. In this paper, we propose a novel bipartite graph neural network (Bi-GNN) approach for scalable cellular network optimization, focusing on network parameters including the azimuth angle and tilt angle of the antenna array, as well as the transmission power of the base station (BS). Our approach leverages the bipartite structure of cellular network topology and incorporates a message-passing mechanism by unfolding the Zeroth-Order Block Coordinate Projected Gradient Descent (ZO-BCPGD) algorithm, offering performance guarantees while maintaining a low computational overhead. Numerical results demonstrate that the Bi-GNN approach not only achieves a higher coverage ratio, but also significantly reduces the computational cost compared to existing methods. Furthermore, our approach exhibits robust scalability across various network scenarios, making it a versatile tool for optimizing a wide range of wireless networks. The proposed Bi-GNN approach has the potential to revolutionize cellular network optimization, paving the way for more efficient and scalable network solutions.