RFBoost: Understanding and Boosting Deep WiFi Sensing via Physical Data Augmentation

TL;DR

RFBoost pioneers the study of RDA, an important yet currently underexplored building block for DWS, which is expected to become a standard DWS component of WiFi sensing and beyond.

Abstract

Deep learning shows promising performance in wireless sensing. However, deep wireless sensing (DWS) heavily relies on large datasets. Unfortunately, building comprehensive datasets for DWS is difficult and costly, because wireless data depends on environmental factors and cannot be labeled offline. Despite recent advances in few-shot/cross-domain learning, DWS is still facing data scarcity issues. In this paper, we investigate a distinct perspective of radio data augmentation (RDA) for WiFi sensing and present a data-space solution. Our key insight is that wireless signals inherently exhibit data diversity, contributing more information to be extracted for DWS. We present RFBoost, a simple and effective RDA framework encompassing novel physical data augmentation techniques. We implement RFBoost as a plug-and-play module integrated with existing deep models and evaluate it on multiple datasets. Experimental results demonstrate that RFBoost achieves remarkable average accuracy improvements of 5.4% on existing models without additional data collection or model modifications, and the best-boosted performance outperforms 11 state-of-the-art baseline models without RDA. RFBoost pioneers the study of RDA, an important yet currently underexplored building block for DWS, which we expect to become a standard DWS component of WiFi sensing and beyond. RFBoost is released at https://github.com/aiot-lab/RFBoost.

Figures (22)

• Figure 1: Image data augmentation for images does not apply well to RF spectrograms.
• Figure 2: Design choices for data preprocessing in DWS. Unprocessed or minimally processed data preserves the most information for learning, but suffers from severe noises; Highly processed data denoises the raw signals but also experiences information loss; Medium processed data seems to balance noise reduction and signal reserving for learning.
• Figure 3: Performance comparison using CSI, DFS, and BVPs as inputs on different models. DFS inputs outperform raw CSI for most models, and achieve comparable accuracy with BVPs. Note that the original Widar3 model (Widar3-M) only accepts BVPs, and is modified here to intake CSI and DFS. Model description can be found in Tab.\ref{['tab:models']}.
• Figure 4: Overview of RFBoost design. RFBoost is designed as a plug-and-play module that can be flexibly integrated with existing models (left). It performs RDA between the raw CSI data and the DNN inputs (right).
• Figure 5: Data diversity. For the same activity, the spectrograms using different windows ((b) and (c)), on different antennas ((d) and (e)), on different subcarriers ((e) and (f)), capture different DFS responses and background noise.