随着智能反射表面(Reconfigurable Intelligent Surface,RIS)反射单元数量的增加以及定位范围的扩大,数据维度和计算复杂度也逐渐增大.普通的RIS辅助定位算法已经无法满足高维度和高强度计算的需求.随着深度学习等人工智能技术的发展,众多学者关注用深度学习进行定位.深度学习具有学习能力强、覆盖范围广且高度依赖数据量等优点,可以有效解决数据维度大以及计算量大等问题.考虑视距(Line of Sight,LoS)链路和非视距(Non-Line of Sight,NLoS)链路都存在和仅存在NLoS的定位场景下,引入深度学习技术,采用指纹定位的方法采集位置信息,将其输入到基于多头注意力机制(Multi-Head Attention,MHA)的Transformer网络模型中进行训练,实现RIS辅助定位,挖掘信道状态信息(Channel State Information,CSI)与用户位置之间的映射关系,研究三维场景下RIS辅助定位的定位精度.
Research on Single Station Intelligent Location Algorithm Based on Intelligent Reflective Surface
With the increase of the number of Reconfigurable Intelligent Surface(RIS)reflection units and the expansion of the positioning range,the data dimension and computational complexity are gradually increasing.The common RIS assisted positioning algorithm can no longer meet the needs of high-dimensional and high-intensity computing.However,with the development of artificial intelligence technologies such as deep learning,many scholars have focused on using deep learning for positioning.Deep learning,known for its strong learning capability,wide coverage,and high dependence on data volume,can effectively solve problems of large data dimensionality and computational complexity.Positioning scenarios where both Line of Sight(LoS)and Non-Line of sight(NLoS)exist,as well as scenarios with only NLoS are considered.By incorporating deep learning technology and using the fingerprinting method to collect location information,the data is fed into a Transformer network model based on Multi-Head Attention(MHA)for training.This approach realizes RIS-assisted positioning,explores the mapping relationship between Channel State Information(CSI)and user location,and studies the positioning accuracy of RIS-assisted positioning in three-dimensional scenarios.
万方数据
维普
