Abstract:Artificial Intelligence (AI)-based radio fingerprinting (FP) outperforms classic localization methods in propagation environments with strong multipath effects. However, the model and data orchestration of FP are time-consuming and costly, as it requires many reference positions and extensive measurement campaigns for each environment. Instead, modern unsupervised and self-supervised learning schemes require less reference data for localization, but either their accuracy is low or they require additional sensor information, rendering them impractical. In this paper we propose a self-supervised learning framework that pre-trains a general transformer (TF) neural network on 5G channel measurements that we collect on-the-fly without expensive equipment. Our novel pretext task randomly masks and drops input information to learn to reconstruct it. So, it implicitly learns the spatiotemporal patterns and information of the propagation environment that enable FP-based localization. Most interestingly, when we optimize this pre-trained model for localization in a given environment, it achieves the accuracy of state-of-the-art methods but requires ten times less reference data and significantly reduces the time from training to operation.

What problem does this paper attempt to address?

The paper attempts to address the problem of achieving high-precision and robust positioning using 5G signals in complex indoor environments. Traditional methods, such as those based on Time of Arrival (TOA) or Angle of Arrival (AOA), experience performance degradation in environments with strong multipath effects. While AI-based radio fingerprinting (FP) can overcome these issues, it requires a large amount of reference data for training, which is both time-consuming and expensive in practical applications. Additionally, existing self-supervised and semi-supervised learning methods either have low accuracy or require additional sensor information, making them impractical for real-world applications. To tackle these challenges, the paper proposes a self-supervised learning framework that reduces the need for reference data by pre-training a general Transformer neural network, thereby enabling indoor positioning based on 5G signals. The core of this framework lies in the design of a novel pre-training task, which involves randomly masking and discarding input information to learn to reconstruct it. This process not only allows the model to implicitly learn the spatiotemporal patterns in the propagation environment but also achieves positioning accuracy comparable to existing state-of-the-art methods when fine-tuning this pre-trained model in a given environment, with only one-tenth of the reference data required, significantly reducing the time from training to actual operation. Experiments demonstrate that this framework outperforms existing supervised pre-training and fingerprinting methods on two 5G datasets while requiring the least amount of reference data. This achievement provides a new direction for realizing cost-effective, accurate, and robust 5G indoor positioning.

Radio Foundation Models: Pre-training Transformers for 5G-based Indoor Localization

Supreme: Fine-grained Radio Map Reconstruction Via Spatial-Temporal Fusion Network

Model and Transfer Spatial-Temporal Knowledge for Fine-Grained Radio Map Reconstruction

Neural 5G Indoor Localization with IMU Supervision

A FL-Based Radio Map Reconstruction Approach for UAV-Aided Wireless Networks

Improving RF-based device-free passive localization in cluttered indoor environments through probabilistic classification methods

Rf-based fingerprinting for indoor localization: deep transfer learning approach

Building 6G Radio Foundation Models with Transformer Architectures

NeRF2: Neural Radio-Frequency Radiance Fields

Indoor Localization with Multi-beam of 5G New Radio Signals

Radio Environment Map Construction Based on Privacy-Centric Federated Learning

The Case for Efficient and Robust RF-Based Device-Free Localization.

Localization in Digital Twin MIMO Networks: A Case for Massive Fingerprinting

Radio Map Prediction from Aerial Images and Application to Coverage Optimization

Scalable Fingerprint-Based Localization Framework for 3D Cell-Free Radio Access Networks

Mean Teacher based SSL Framework for Indoor Localization Using Wi-Fi RSSI Fingerprinting

Enhancing Wi-Fi RSS-Based Indoor Positioning Under Dynamic AP Availability: Leveraging Virtual Feature Maps and Contrastive Learning

Diffraction and Scattering Aware Radio Map and Environment Reconstruction using Geometry Model-Assisted Deep Learning

DeepRSSI: Generative Model for Fingerprint-Based Localization

A Multi-Scale Spatial Model for RSS-based Device-Free Localization

Train Once, Locate Anytime for Anyone: Adversarial Learning Based Wireless Localization.