Multi-RIS-Assisted Fingerprint-Based 3-D Localization via Attention Enhanced Joint Feature Extraction Network

This letter investigates three-dimensional (3D) user equipment (UE) localization assisted by multiple reconfigurable intelligent surfaces (RISs) in a scattering-rich environment. To effectively exploit the cascaded channels introduced by RISs, a novel fingerprint named joint angle delay power matrix (JADPM) is proposed, which is extracted by decoupling rich angle and delay information of the cascaded channel. To fully extract the diverse features contained in JADPM, we further propose an attention-enhanced joint feature extraction network (AE-JFEN), which consists of a joint feature foundation module (JFFM), an attention-enhanced hierarchical module (AEHM), and a regression module (RM). Simulation results demonstrate that the proposed JADPM fingerprint with the AE-JFEN method achieves high-accuracy localization.