Fifth-generation (5G) networks have been massively deployed in commerce. In addition, the new features introduced by 5G networks have been recognized to be beneficial to wireless localization. In this study, we investigate the performance of indoor localization with commercial 5G new radio (NR) signals, and the channel state information (CSI) is used for localization, which is acquired from the synchronization signal block (SSB) of the downlink physical channel. A hybrid indoor localization system named Hi-Loc has been developed. Initially, a feature enhancement module is designed to generate new features from the SSB CSI, and a data construction module is applied to prepare the dataset for subsequent localization algorithms. By introducing the dual attention mechanism deep network with convolutional neural networks (CNN) and bidirectional long short-term memory (BiLSTM), a feature-based attention mechanism and a sample-based attention mechanism are designed to extract the implicit spatial and time information of CSI, respectively. Finally, in the online stage, by applying a fully connected neural network (FCNN), the 2D coordinates are estimated based on the corresponding CSI features. We carry out indoor field tests in the typical office and corridor scenario. The results show that Hi-Loc achieves mean absolute errors of 0.65 and 1.03 m in the internal test cases of the office scenario and corridor scenario, respectively, and 1.93 and 3.10 m in the external test cases of the office scenario and corridor scenario, respectively, indicating its superiority over the compared CSI fingerprint-based methods in localization accuracy.
Read full abstract