Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Finding the Missing Data: A BERT-inspired Approach Against Package Loss in Wireless Sensing

Zijian Zhao, Tingwei Chen, Fanyi Meng, Hang Li, Xiaoyang Li, Guangxu Zhu

TL;DR

This work tackles the problem of packet loss in Wi-Fi Channel State Information (CSI) time series, which disrupts continuous input for learning-based sensing. It introduces CSI-BERT, a BERT-inspired Transformer model that processes continuous CSI with a specialized embedding, self-supervised pre-training, adversarial regularization, and two recovery strategies to restore missing data across all subcarriers. The method demonstrates lower recovery errors and faster recovery than traditional interpolation, and the recovered CSI significantly boosts downstream sensing models by about 15% on tasks such as gesture recognition and person identification. The approach, datasets, and code are publicly available, highlighting a practical pathway to robust, real-time CSI sensing in lossy wireless environments and showcasing the transfer of NLP-style models to continuous CSI data.

Abstract

Despite the development of various deep learning methods for Wi-Fi sensing, package loss often results in noncontinuous estimation of the Channel State Information (CSI), which negatively impacts the performance of the learning models. To overcome this challenge, we propose a deep learning model based on Bidirectional Encoder Representations from Transformers (BERT) for CSI recovery, named CSI-BERT. CSI-BERT can be trained in an self-supervised manner on the target dataset without the need for additional data. Furthermore, unlike traditional interpolation methods that focus on one subcarrier at a time, CSI-BERT captures the sequential relationships across different subcarriers. Experimental results demonstrate that CSI-BERT achieves lower error rates and faster speed compared to traditional interpolation methods, even when facing with high loss rates. Moreover, by harnessing the recovered CSI obtained from CSI-BERT, other deep learning models like Residual Network and Recurrent Neural Network can achieve an average increase in accuracy of approximately 15\% in Wi-Fi sensing tasks. The collected dataset WiGesture and code for our model are publicly available at https://github.com/RS2002/CSI-BERT.

Finding the Missing Data: A BERT-inspired Approach Against Package Loss in Wireless Sensing

TL;DR

This work tackles the problem of packet loss in Wi-Fi Channel State Information (CSI) time series, which disrupts continuous input for learning-based sensing. It introduces CSI-BERT, a BERT-inspired Transformer model that processes continuous CSI with a specialized embedding, self-supervised pre-training, adversarial regularization, and two recovery strategies to restore missing data across all subcarriers. The method demonstrates lower recovery errors and faster recovery than traditional interpolation, and the recovered CSI significantly boosts downstream sensing models by about 15% on tasks such as gesture recognition and person identification. The approach, datasets, and code are publicly available, highlighting a practical pathway to robust, real-time CSI sensing in lossy wireless environments and showcasing the transfer of NLP-style models to continuous CSI data.

Abstract

Despite the development of various deep learning methods for Wi-Fi sensing, package loss often results in noncontinuous estimation of the Channel State Information (CSI), which negatively impacts the performance of the learning models. To overcome this challenge, we propose a deep learning model based on Bidirectional Encoder Representations from Transformers (BERT) for CSI recovery, named CSI-BERT. CSI-BERT can be trained in an self-supervised manner on the target dataset without the need for additional data. Furthermore, unlike traditional interpolation methods that focus on one subcarrier at a time, CSI-BERT captures the sequential relationships across different subcarriers. Experimental results demonstrate that CSI-BERT achieves lower error rates and faster speed compared to traditional interpolation methods, even when facing with high loss rates. Moreover, by harnessing the recovered CSI obtained from CSI-BERT, other deep learning models like Residual Network and Recurrent Neural Network can achieve an average increase in accuracy of approximately 15\% in Wi-Fi sensing tasks. The collected dataset WiGesture and code for our model are publicly available at https://github.com/RS2002/CSI-BERT.
Paper Structure (15 sections, 7 equations, 5 figures, 3 tables)

This paper contains 15 sections, 7 equations, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Overview
  5. Model Structure
  6. Key Phases
  7. Pre-training Phase
  8. Recovering Phase
  9. Fine-tuning Phase
  10. Dataset and Numerical Analysis
  11. Dataset
  12. Experiment Setup
  13. Experiment Result
  14. Ablation Study
  15. Conclusion

Figures (5)

  • Figure 1: Workflow
  • Figure 2: Overview of the Proposed CSI-BERT Framework
  • Figure 3: Process of Pre-training and Recovering
  • Figure 4: CSI Recovery Performance at Different Loss Rates: The x-axis represents the CSI loss rate (excluding the real lost CSI), and the y-axis represents the different metrics between the real CSI and CSI recovered by various methods. The line representing Ordinary Kriging is shorter than others, indicating its failure at higher loss rates.
  • Figure 5: Amplitude Spectrum of Original CSI and Output of BERT and CSI-BERT: The blank in the original CSI represents the lost CSI.