Abstract
Integrated sensing and communication (ISAC) is considered a key enabler for the future Internet of Things (IoT), as it enables wireless networks to simultaneously support high-capacity data transmission and precise environmental sensing. Indoor localization, as a representative sensing service in ISAC, has attracted considerable research attention. Nevertheless, its performance is largely constrained by the quality and granularity of the collected data. In this work, we propose an attention-based framework for cost-efficient indoor fingerprint localization that exploits extended trajectory map construction through a novel trajectory-based data augmentation (TDA) method. In particular, fingerprints at unmeasured locations are synthesized using a conditional Wasserstein generative adversarial network (CWGAN). A path generation algorithm is employed to produce diverse trajectories and construct the extended trajectory map. Based on this map, a multi-head attention model with direction-constrained auxiliary loss is then applied for accurate mobile device localization. Experiments in a real 5G indoor environment demonstrate the system's effectiveness, achieving an average localization error of 1.09 m and at least 34% higher accuracy than existing approaches.