基于神经网络的毫米波跨场景路径损耗预测研究

Research on Neural Network-Based mm Wave Cross-Scenario Path Loss Prediction

毫米波通信因其可以提供更高的传输速率和系统容量受到了广泛关注。无人机空地通信技术是未来通信网络的重要组成部分,路径损耗预测对于无人机的节点布局、轨迹优化和功率分配具有重要意义。针对无人机通信多场景PL预测问题,结合参数化的几何场景和多输入反向传播神经网络,提出了一种具有跨场景能力的PL迁移预测模型。该模型利用有限场景下信道数据进行网络训练,可以预测未知新场景下的PL。最后,利用射线跟踪仿真数据进行模型有效性验证,仿真结果表明,所提模型的神经网络训练收敛效果较好,在新场景下预测PL结果与RT仿真结果基本吻合,验证了该模型的跨场景迁移预测能力。

Millimeter wave communication has gained extensive attention due to its capability of higher transmission rate and system capacity. UAV air-to-ground communication technology is an important part in future communication network, where path loss(PL) prediction is of great significance for UAV placement, trajectory optimization, and power allocation. Aiming at the multi-scenario PL prediction problem of UAV communication, a path loss migration prediction model with cross-scenario capability is proposed by combining parameterized geometric scenarios and multi-input backpropagation neural network. The proposed model utilizes the channel data in limited scenarios for network training and can predict the PL in unknown new scenarios. Finally, the effectiveness of the proposed model is validated by ray tracing simulation data. The simulation results show that the neural network of the proposed model is well trained, and the PL prediction results are well consistent with those of RT simulation in a new scenario, which verifies the cross-scenario prediction ability.