DGSense: A Domain Generalization Framework for Wireless Sensing

TL;DR

DGSense tackles the domain shift problem in wireless sensing by learning domain-independent representations from multiple source domains and applying them to unseen target domains without any data from those domains. The approach combines a cross-modal virtual data generator (based on a VAE) to enrich training diversity, a spatial-temporal feature extractor (ResNet CBAM for spatial, 1D-CNN for temporal) to capture modality-specific patterns, and episodic training between a main network and domain-specific networks to enforce domain-invariant features. Empirical evaluations on WiFi gesture recognition, mmWave activity recognition, and acoustic fall detection demonstrate strong generalization to new users, new locations, and new environments without target-domain data or retraining. The work provides a general DG framework for wireless sensing, showing that diversity via virtual data alongside episodic domain-extraction yields robust cross-domain performance across multiple sensing modalities. Keywords: domain generalization, cross-modal virtual data generation, episodic training, wireless sensing, range-Doppler maps, Doppler spectrogram.

Abstract

Wireless sensing is of great benefits to our daily lives. However, wireless signals are sensitive to the surroundings. Various factors, e.g. environments, locations, and individuals, may induce extra impact on wireless propagation. Such a change can be regarded as a domain, in which the data distribution shifts. A vast majority of the sensing schemes are learning-based. They are dependent on the training domains, resulting in performance degradation in unseen domains. Researchers have proposed various solutions to address this issue. But these solutions leverage either semi-supervised or unsupervised domain adaptation techniques. They still require some data in the target domains and do not perform well in unseen domains. In this paper, we propose a domain generalization framework DGSense, to eliminate the domain dependence problem in wireless sensing. The framework is a general solution working across diverse sensing tasks and wireless technologies. Once the sensing model is built, it can generalize to unseen domains without any data from the target domain. To achieve the goal, we first increase the diversity of the training set by a virtual data generator, and then extract the domain independent features via episodic training between the main feature extractor and the domain feature extractors. The feature extractors employ a pre-trained Residual Network (ResNet) with an attention mechanism for spatial features, and a 1D Convolutional Neural Network (1DCNN) for temporal features. To demonstrate the effectiveness and generality of DGSense, we evaluated on WiFi gesture recognition, Millimeter Wave (mmWave) activity recognition, and acoustic fall detection. All the systems exhibited high generalization capability to unseen domains, including new users, locations, and environments, free of new data and retraining.

Figures (14)

• Figure 1: The domain generalization framework (DGSense).
• Figure 2: The virtual data generator.
• Figure 3: Episodic training for domain independent feature extraction.
• Figure 4: Process and effect of episodic training on WiFi gesture data (5 training users, 1 new user).
• Figure 5: The feature extractor and classifier of WiFi gesture recognition.