Abstract: In order to effectively control the co-channel interference in a Device-to-Device (D2D) communication system while reducing the implementation complexity, a graph convolutional network (GCN)-based distributed power allocation algorithm was proposed to maximize the sum weighted bit rate of all D2D links. To achieve the above goals, the system topology was first built into a graph model and the characteristics of nodes and edges as well as the message-passing manners were further defined. Then, the unsupervised learning model was leveraged to train the model parameters of the formulated GCN. After the offline training, each D2D link was able to obtain the optimal power allocation strategy in a distributed manner based on local channel state information and the interaction with neighboring nodes. The experimental results showed that compared with the optimization theory-based algorithm, the proposed algorithm substantially cut down the operation time by 97.41% while suffering only 3.409% performance loss, and compared with the deep reinforcement learning theory-based algorithm, the proposed algorithm exhibited better generalization and stability under different settings of parameters.

Key words: Device-to-Device (D2D) communication, interference coordination, Graph Convolutional Network (GCN), power allocation, distributed algorithms

摘要： 为有效控制终端直通(D2D)通信系统中同频干扰并降低实现复杂度，提出一种基于图卷积网络(GCN)的分布式功率控制算法，旨在最大化所有D2D链路的加权和速率。为实现上述目标，首先将系统拓扑结构建模为图模型并定义节点和边的特征以及消息传递方式，随后借助无监督学习模型训练GCN中模型参数。离线训练后，每条D2D链路可根据局部信道状态信息以及与相邻节点的交互过程分布式地得到最佳功率控制策略。实验结果表明，相较于基于优化理论的算法，所提算法大幅度降低了97.41%的运算时间且仅损失3.409%的性能；相较于基于深度强化学习理论的算法，所提算法具有良好的泛化能力，在不同参数设置下表现更加稳定。

关键词: 终端直通通信, 干扰协调, 图卷积网络, 功率控制, 分布式算法