Collaborative SA-BiLSTM With Multi-SSB Oriented Data Augmentation for Smart Indoor Localization in Cyber-Physical-Social Systems

As 5G networks are widely deployed in smart buildings and public safety scenarios, integrating indoor localization into the cyber-physical-social systems (CPSS) framework enables unified perception and intelligent decision making across the physical, cyber, and social domains. In this context, we propose a CPSS-oriented collaborative self-attention bidirectional long short-term memory (SA-BiLSTM) localization system. In each synchronization signal block (SSB) cycle defined in 5G new radio (NR), the user equipment (UE) captures reference signal received power (RSRP) and signal to interference plus noise ratio (SINR) measurements from multiple beams, forming spatially sensitive fingerprints closely correlated with the UE’s location. For data augmentation, we apply a Hampel filter and propose an improved Kriging interpolation method. Uniform manifold approximation and projection (UMAP) is then employed to reduce dimensionality while preserving the intrinsic relationship between signal features and spatial positions, facilitating efficient fingerprint database construction. Finally, a collaborative SA-BiLSTM combines a bidirectional LSTM for temporal modeling with a dual self-attention mechanism that jointly attends to time and beam dimensions, fusing spatiotemporal features from multiple beams to accurately predict the UE position. In a typical indoor office environment, 90th-percentile errors fall within 2.55 m, with a median error of 0.97 m. By extracting spatiotemporal features from 5G multibeam SSB signals for high precision UE localization, this method achieves interactive position awareness capabilities within CPSS.