Answers are grounded in this paper and cite their sources.
Accurate, low-latency, and infrastructure-light positioning is a core requirement for 6G integrated positioning and communications (IPAC), yet remains challenging when robustness to fast fading and applicability to both static and dynamic scenarios are required. This paper proposes a unified massive MIMO positioning framework built on Beam-Domain Statistical Channel Fingerprints (BDSCFs), which are long-term, location dependent angular–delay energy matrices that are temporally stable and naturally available at the base station. BDSCFs serve as a unified representation that bridges fingerprint-based localization and continuous tracking. For static localization, we reformulate positioning as a conditional distribution learning problem and develop a BDSCF-Feature Extraction Conditional GAN (BFGAN), in which the generator maps latent noise and BDSCFs to 2-D coordinates while the discriminator enforces data-consistent sampling. The training objective combines ad versarial loss with a geometric regularizer; we show theoreti cally that the adversarial term minimizes the Jensen-Shannon divergence between generated and true conditional distributions, while a Lipschitz-type bound on the generator preserves spatial consistency. For dynamic tracking, we design a Transformer based tracker that operates on short temporal sequences of BDSCF tensors. By applying patch embedding and multi-head self-attention, the tracker learns motion-aware spatio-temporal features and enables low-latency regression under an offline training/online-inference paradigm. Geometry-based simulations demonstrate that BFGAN surpasses deterministic regression baselines and benefits from denser reference grids and more antennas; the Transformer attains 95% of errors ≤ 2 m and shows the expected sensitivity to sequence length. The BDSCF (BFGAN + Transformer) pipeline bridges fingerprinting and continuous tracking, offering a practical path toward accurate positioning in 6G IPAC systems.
No takes yet. Share an insight, caveat, or question.
Gong et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: