Context-Aware Predictive Coding: A Representation Learning Framework for WiFi Sensing

TL;DR

This paper introduces a novel SSL framework called Context-Aware Predictive Coding (CAPC), which effectively learns from unlabelled data and adapts to diverse environments and achieves superior generalization capabilities.

Abstract

WiFi sensing is an emerging technology that utilizes wireless signals for various sensing applications. However, the reliance on supervised learning, the scarcity of labelled data, and the incomprehensible channel state information (CSI) pose significant challenges. These issues affect deep learning models' performance and generalization across different environments. Consequently, self-supervised learning (SSL) is emerging as a promising strategy to extract meaningful data representations with minimal reliance on labelled samples. In this paper, we introduce a novel SSL framework called Context-Aware Predictive Coding (CAPC), which effectively learns from unlabelled data and adapts to diverse environments. CAPC integrates elements of Contrastive Predictive Coding (CPC) and the augmentation-based SSL method, Barlow Twins, promoting temporal and contextual consistency in data representations. This hybrid approach captures essential temporal information in CSI, crucial for tasks like human activity recognition (HAR), and ensures robustness against data distortions. Additionally, we propose a unique augmentation, employing both uplink and downlink CSI to isolate free space propagation effects and minimize the impact of electronic distortions of the transceiver. Our evaluations demonstrate that CAPC not only outperforms other SSL methods and supervised approaches, but also achieves superior generalization capabilities. Specifically, CAPC requires fewer labelled samples while significantly outperforming supervised learning and surpassing SSL baselines. Furthermore, our transfer learning studies on an unseen dataset with a different HAR task and environment showcase an accuracy improvement of 1.8 percent over other SSL baselines and 24.7 percent over supervised learning, emphasizing its exceptional cross-domain adaptability.

Figures (9)

• Figure 1: Illustration of the proposed CAPC framework. Initially, unlabelled uplink and downlink CSI are utilized to pre-train an encoder through an unsupervised approach. Subsequently, the model undergoes fine-tuning using a limited set of labelled CSI for the HAR task in the unseen environment.
• Figure 2: Overview of the CAPC's architecture. Here, $\boldsymbol{w_t}$ denotes a window of sample $\boldsymbol{u}$. The symbols $\boldsymbol{x_t}$, $\boldsymbol{z_t}$, and $\boldsymbol{c_t}$ represent the augmented CSI for window $t$, the latent representation of this window, and the accumulated context embedding up to window $t$, respectively. Different colours signify distinction in the windows, their representations, and model parameters between branches A and B.
• Figure 3: Supervised evaluation: A linear classifier $C_{\boldsymbol{\phi}}$ is fine-tuned with labelled CSI based on the concatenated representations from all windows generated by the pre-trained encoder $E_{\boldsymbol{\theta^A}}$. The pre-trained encoder's weights $\boldsymbol{\theta^A}$ are frozen in linear classification but fine-tuned in the semi-supervised evaluation.
• Figure 4: Linear evaluation of individual and compositional data augmentations. Each diagonal element represents the effect of a single transformation, while off-diagonal elements illustrate the combined impact of two sequentially applied transformations. We report the accuracies (in %) with 6 shots in the labelled dataset. Red circles indicate the best combination of augmentations.
• Figure 5: A comparative study of the proposed CAPC method. The CAPC w/ SimCLR and AutoFi mean that we have replaced the Barlow Twins loss in our design with SimCLR and AutoFi, respectively. Showcasing that Barlow Twins has superior performance for enforcing context embedding consistency. We report the experiments under linear evaluation of SignFi Home dataset with 2, 4, and 6 shots.