Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

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

TL;DR

The paper tackles the limited transferability of deep learning for CSI-based positioning across different environments. It introduces Multi Environment meta-Learning (MEML), a two-part model with a shared environment-independent feature extractor and environment-specific heads, trained over multiple source environments to improve adaptation to unseen targets, including uncertainty estimation via NLL and CRPS. Empirical results on real Dichasus data (LOS and NLOS) show MEML outperforms direct transfer and scratch training, and that gradual unfreezing further boosts data efficiency and accuracy, reducing target data requirements by about a third without performance loss. Overall, the approach significantly enhances the practicality of CSI fingerprinting for robust, uncertainty-aware UE positioning in diverse environments.

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.

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

TL;DR

The paper tackles the limited transferability of deep learning for CSI-based positioning across different environments. It introduces Multi Environment meta-Learning (MEML), a two-part model with a shared environment-independent feature extractor and environment-specific heads, trained over multiple source environments to improve adaptation to unseen targets, including uncertainty estimation via NLL and CRPS. Empirical results on real Dichasus data (LOS and NLOS) show MEML outperforms direct transfer and scratch training, and that gradual unfreezing further boosts data efficiency and accuracy, reducing target data requirements by about a third without performance loss. Overall, the approach significantly enhances the practicality of CSI fingerprinting for robust, uncertainty-aware UE positioning in diverse environments.

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.
Paper Structure (8 sections, 9 equations, 14 figures, 7 tables)

This paper contains 8 sections, 9 equations, 14 figures, 7 tables.

Table of Contents

  1. Introduction
  2. System Model
  3. Simulation Setup
  4. Database Description
  5. Neural Network and training
  6. Simulation Results
  7. Conclusion
  8. CRPS Toy Example

Figures (14)

  • Figure 1: Multi-environment training over $N_B$ environments
  • Figure 2: Training of target environment
  • Figure 3: Layout of considered source environment. DICHASUS Industrial environment dichasus2021
  • Figure 4: Layout of considered LOS target environment. DICHASUS Indoor LOS lab room dichasus2021
  • Figure 5: Layout of considered NLOS target environment. DICHASUS Indoor NLOS lab room dichasus2021
  • ...and 9 more figures