一种基于深度学习的室内波束选择方法  

An Indoor Beam Selection Method Based on Deep Learning

作　　者：王俊智 仲伟志[1] 王鑫 肖丽君 WANG Junzhi;ZHONG Weizhi;WANG Xin;XIAO Lijun(College of Astronautics,NUAA,Nanjing 210016,China)

机构地区：[1]南京航空航天大学航天学院,江苏南京210016

出　　处：《移动通信》2022年第12期46-51,共6页Mobile Communications

基　　金：国家重大科研仪器研制项目(61827801)。

摘　　要：针对以往利用搜索匹配进行波束对准的方法不适用于具有随机方向的室内手持用户终端这一问题,引入一种基于深度神经网络的室内波束选择方法。设定该室内场景中发射机位置和方向固定,接收机的位置和方向任意以模拟手持用户终端。该方法首先通过测量每个波束对的传输效率以及信噪比等特征,获得在设定场景中的波束对匹配列表数据库;而后将接收机的位置和方向作为输入,每个波束对的最佳概率作为输出,采用机器学习的分类方法对测试的室内场景进行训练,获得学习模型;最后,结合已训练好的学习模型,根据实际接收机状态信息匹配最佳波束对。仿真结果表明,与广义逆指纹算法相比,该方法具有更高的波束匹配概率和搜索效率。Aiming at the problem that the previous beam alignment method using search matching is not suitable for indoor handheld user terminals with random directions, an indoor beam selection method is introduced based on deep neural network. In the indoor scenario, the position and direction of the transmitter are fixed while the ones of the receiver are arbitrary to simulate the handheld user terminal. The proposed method firstly obtains the matching list database of beam pairs in the predefined scenario by measuring the characteristics of each beam pair, such as transmission efficiency and signal-to-noise ratio. Then the position and direction of the receiver are used as the input, and the best probability of each beam pair is used as the output. The machine learning classification method is used to train the indoor scenario to obtain the learning model. Finally, combined with the trained learning model, the best beam pair is matched according to the actual receiver state information. The simulation results show that the proposed method has higher beam matching probability and searching efficiency than the generalized inverse fingerprint algorithm.

关 键 词：室内通信 毫米波 波束选择 深度学习 

分 类 号：TN929.5[电子电信—通信与信息系统]