Scalable Fingerprint-based Localization Framework for 3D Cell-Free Radio Access Networks

In this letter, to address the high-precision localization issue in wide-area three-dimensional (3D) scenes, we investigate a scalable fingerprint-based localization scheme for cell-free radio access network (CF-RAN). Based on the dynamic correlation mechanism, we first propose a scalable angle-space channel power matrix (SASCPM) fingerprint, which only consumes finite overhead and is closely related to 3D location. Then, based on the partially distributed collaboration mechanism, we design a scalable fingerprint-based localization (FedSLoc) framework for 3D CF-RAN. In FedSLoc, the centralized localization pressure in the cloud is sunk to multiple edge distributed units, which are equipped with local models to enable localization with limited resource consumption. Moreover, in FedSLoc, we use a modified federated block coordinate descent algorithm to solve the feature-partitioned collaborative learning issue with low communication overheads. By extensive simulations, the proposed framework is proven to achieve scalable high-precision localization.