Transfer Learning for CSI-based Positioning with Multi-environment Meta-learning

Anastasios Foliadis,Mario H. Castañeda,Richard A. Stirling-Gallacher,Reiner S. Thomä

2024-05-20

Abstract:Utilizing deep learning (DL) techniques for radio-based positioning of user equipment (UE) through channel state information (CSI) fingerprints has demonstrated significant potential. DL models can extract complex characteristics from the CSI fingerprints of a particular environment and accurately predict the position of a UE. Nonetheless, the effectiveness of the DL model trained on CSI fingerprints is highly dependent on the particular training environment, limiting the trained model's applicability across different environments. This paper proposes a novel DL model structure consisting of two parts, where the first part aims at identifying features that are independent from any specific environment, while the second part combines those features in an environment specific way with the goal of positioning. To train such a two-part model, we propose the multi-environment meta-learning (MEML) approach for the first part to facilitate training across various environments, while the second part of the model is trained solely on data from a specific environment. Our findings indicate that employing the MEML approach for initializing the weights of the DL model for a new unseen environment significantly boosts the accuracy of UE positioning in the new target environment as well the reliability of its uncertainty estimation. This method outperforms traditional transfer learning methods, whether direct transfer learning (DTL) between environments or completely training from scratch with data from a new environment. The proposed approach is verified with real measurements for both line-of-sight (LOS) and non-LOS (NLOS) environments.

Signal Processing,Artificial Intelligence

What problem does this paper attempt to address?

This paper discusses the problem of radio device localization using deep learning (DL) technology through Channel State Information (CSI) fingerprints. Although existing DL models can extract complex features from CSI fingerprints of specific environments to accurately predict device locations, their effectiveness highly depends on the training environment, limiting the applicability of the models in different environments. The paper proposes a new DL model structure, divided into two parts. The first part aims to identify environment-independent features, while the second part combines these features with specific environments for localization. To train this two-part model, the paper introduces the Multi-Environment Meta-Learning (MEML) approach to initialize the weights of the first part, while the second part is trained only with data from specific environments. The research found that using MEML to initialize the weights of a DL model for a new unseen environment significantly improves the accuracy of UE localization and enhances the reliability of uncertainty estimation. This approach outperforms traditional direct transfer learning (DTL) and training from scratch using only new environment data. The method is validated based on real measurement data, including line-of-sight (LOS) and non-line-of-sight (NLOS) environments. The paper also introduces uncertainty estimation to enhance the model's transferability and proposes the use of progressive unfreezing technique for more effective training of the model, especially when the CSI sample size in the target environment is small. In conclusion, the paper aims to address the problem of developing a deep learning model that can effectively work across different environments, improve the accuracy of radio device localization, adapt to new environments, and provide reliable uncertainty estimation.

Transfer Learning for CSI-based Positioning with Multi-environment Meta-learning

Multi-Environment based Meta-Learning with CSI Fingerprints for Radio Based Positioning

Deep Learning based Positioning with Multi-task Learning and Uncertainty-based Fusion

Deep Learning-Based Multi-User Positioning in Wireless FDMA Cellular Networks

CSI of Each Subcarrier is a Fingerprint: Multi-Carrier Cumulative Learning Based Positioning in Massive MIMO Systems

Feature Learning for Neural-Network-Based Positioning with Channel State Information

A Meta-learning based Generalizable Indoor Localization Model using Channel State Information

MetaLoc: Learning to Learn Wireless Localization

Dynamic WiFi indoor positioning based on the multi-scale metric learning

Knowledge-driven Meta-learning for CSI Feedback

Complex-Valued Neural Network based Federated Learning for Multi-user Indoor Positioning Performance Optimization

Multi-modal Image and Radio Frequency Fusion for Optimizing Vehicle Positioning

Deep Learning for Efficient CSI Feedback in Massive MIMO: Adapting to New Environments and Small Datasets

Multi-Objective Deep Reinforcement Learning for 5G Base Station Placement to Support Localisation for Future Sustainable Traffic

Communication-efficient Personalized Federated Edge Learning for Massive MIMO CSI Feedback

Influence of Dataset Parameters on the Performance of Direct UE Positioning via Deep Learning

Deep Transfer Learning Based Downlink Channel Prediction for FDD Massive MIMO Systems

Neural RF SLAM for unsupervised positioning and mapping with channel state information

Deep Learning for UL/DL Channel Calibration in Generic Massive MIMO Systems

Transfer Learning Capabilities of Untrained Neural Networks for MIMO CSI Recreation

Channel State Information Based Localization with Deep Learning