Adaptive Amplitude–Phase Fusion Fingerprint Localization Based on Self-Paced Broad Learning System

Traditional Channel State Information (CSI)-based indoor fingerprint localization methods with amplitudephase fusion usually treat consecutively collected CSI samples equally, ignoring sample reliability differences induced by short-term temporal variations of CSI in complex indoor environments, thereby affecting localization accuracy. To deal with this problem, we firstly propose an adaptive amplitude-phase fusion network based on the Self-Paced Broad Learning System (SPBLS). Through the self-paced learning mechanism, the method adaptively learns sample weights to reflect their reliability, dynamically assigns weights to amplitude and phase data, allows reliable samples to dominate the fusion process, thereby effectively suppressing noise interference.Secondly, we present a feature extraction model that simultaneously extracts both local and global features of CSI, which leverages the complementarity between the two to fully exploit the rich multi-level information of CSI, enhancing the distinguishability of location features and further improving the localization accuracy of the system. Experimental results show that, compared with multi-antenna fusion fingerprint localizationbased on Weighted Broad Learning System(MFFL), ensemble broad learning localization(EnsemLoca), (parallel AdaBoost indoor localization(PAIL), broad learning system(BLS), AdaBoost positioning system(ABPS), and so on, the proposed approach in this paper improves localization accuracy by 5.92% to 45.85%.