Large Wireless Localization Model (LWLM): A Foundation Model for Positioning in 6G Networks

TL;DR

This work introduces LWLM, a transformer-based wireless foundation model pretrained with a hybrid self-supervised framework (SF-MCM, DTI, PICL) to learn rich, transferable channel semantics from unlabeled CSI. Grounded in information bottleneck theory, the approach balances general-purpose representations with task-specific cues, enabling robust performance across ToA, AoA, single-BS, and multi-BS localization with limited labeled data. Experimental results on DeepMIMO data demonstrate substantial gains over model-based and supervised baselines, including strong generalization to unseen BS configurations and data-scarce regimes. The proposed architecture provides a scalable, flexible pathway to deploy high-accuracy localization in 6G networks, leveraging unlabeled CSI to reduce labeling costs while maintaining strong cross-scenario robustness.

Abstract

Accurate and robust localization is a critical enabler for emerging 5G and 6G applications, including autonomous driving, extended reality (XR), and smart manufacturing. While data-driven approaches have shown promise, most existing models require large amounts of labeled data and struggle to generalize across deployment scenarios and wireless configurations. To address these limitations, we propose a foundation-model-based solution tailored for wireless localization. We first analyze how different self-supervised learning (SSL) tasks acquire general-purpose and task-specific semantic features based on information bottleneck (IB) theory. Building on this foundation, we design a pretraining methodology for the proposed Large Wireless Localization Model (LWLM). Specifically, we propose an SSL framework that jointly optimizes three complementary objectives: (i) spatial-frequency masked channel modeling (SF-MCM), (ii) domain-transformation invariance (DTI), and (iii) position-invariant contrastive learning (PICL). These objectives jointly capture the underlying semantics of wireless channel from multiple perspectives. We further design lightweight decoders for key downstream tasks, including time-of-arrival (ToA) estimation, angle-of-arrival (AoA) estimation, single base station (BS) localization, and multiple BS localization. Comprehensive experimental results confirm that LWLM consistently surpasses both model-based and supervised learning baselines across all localization tasks. In particular, LWLM achieves 26.0%--87.5% improvement over transformer models without pretraining, and exhibits strong generalization under label-limited fine-tuning and unseen BS configurations, confirming its potential as a foundation model for wireless localization.

Figures (11)

• Figure 1: Schematic diagram of the LWLM algorithm framework. First, the LWLM foundation model is pretrained using the pretraining task, and then fine-tuned on specific localization tasks.
• Figure 2: Schematic diagram of the pretraining algorithm. The channel semantic representation $\bm{o}_s$ based on the hybrid SSL method is the concatenation of the three representations $\bm{o}^{\text{SF-MCM}_s}$, $\bm{o}^{\text{DTI}}_s$, $\bm{o}^{\text{PICL}}_s$. At each training step, we compute a weighted sum of three separate losses as the final pretraining loss.
• Figure 3: Schematic diagram of LWLM encoder architecture.
• Figure 4: Schematic diagram of pretraining decoder architecture.
• Figure 5: Schematic diagram of the encoder-decoder architecture for ToA estimation, AoA estimation, and single-BS localization. Blue boxes indicate the decoder components.

Theorems & Definitions (2)

• Theorem 1
• Theorem 2