基于多层感知器神经网络的路径损耗预测研究

为了更好地服务于5G及未来无线通信系统的网络规划与优化，开展了基于多层感知器(multi-layer perceptron, MLP)神经网络的路径损耗预测研究. 利用有限的地物类型，提出一种表征传播环境的简易方法，避免了繁琐的三维场景建模. 结合测量数据和由环境表征方法提取的环境特征，基于MLP神经网络建立了路径损耗模型. 数据实验的对比分析表明MLP神经网络能够实现路径损耗的准确预测，且环境特征的引入有助于提升模型性能. 为解决干扰地物影响路径损耗模型的准确性以及模型对环境变化的敏感性问题，根据视距(line-of-sight, LoS)和非视距(non-line-of-sight, NLoS)标签改进环境表征方法，进一步提升了模型的稳定性和泛化能力. 所做工作有助于了解无线电波传播特性，为无线网络优化和通信系统设计提供了理论依据.

Path loss prediction based on multi-layer perceptron artificial neural network

In order to better serve the network planning and optimization of the 5th generation and future communication systems, the path loss prediction based on multi-layer perceptron (MLP) neural network is carried out in this paper. A simple method to characterize the propagation environment is proposed by the limited clutter type information, avoiding the cumbersome three-dimensional (3D) scenario modeling. Combining with the measurement data and environmental features extracted by the environmental characterization method, the path loss model based on the MLP neural network is established. The comparative analysis of data experiments shows that the MLP neural network can achieve accurate prediction of path loss, and the introduction of environmental features can help improve the performance of the MLP-based path loss model. In order to solve the problem that interference clutters reduce the accuracy of the MLP-based path loss model and this model is sensitive to environmental changes, the environmental characterization method is improved based on the label that can judge whether it is line-of-sight (LoS) or non-line-of-sight (NLoS), which further enhances the stability and generalization ability of the MLP-based path loss model. This paper is designed to understand the propagation characteristics of radio wave, which can provide a theoretical basis for wireless network optimization and communication system design.