MORIC: CSI Delay-Doppler Decomposition for Robust Wi-Fi-based Human Activity Recognition

Abstract

The newly established IEEE 802.11bf Task Group aims to amend the WLAN standard to support advanced sensing applications such as human activity recognition (HAR). Although studies have demonstrated the potential of sub-7 GHz Wi-Fi Channel State Information (CSI) for HAR, existing methods often degrade substantially under realistic variations across users, environments, and sensing configurations. This work addresses the poor generalization of Wi-Fi-based HAR by extracting motion-centered representations that reduce dependence on static, environment-specific, and non-activity-related CSI magnitude and phase patterns. CSI signals are transformed into the delay-profile space and decomposed into multiple Doppler velocity projections, which are modeled as observations of a moving point's velocity from different unknown directions, analogous to virtual cameras observing the same motion with varying degrees of clarity. This yields a richer activity representation than either a single aggregated Doppler estimate or the spurious, environment-dependent CSI patterns used in prior works. Since these projections are unordered and may recur due to random multipath propagation, we introduce MORIC, a novel order- and repetition-invariant time-series classification model for robust Wi-Fi-based HAR. Experimental results on the collected dataset show that the proposed method outperforms state-of-the-art approaches in cross-user hand motion recognition, especially for challenging gestures. Incorporating only a few calibration samples further improves accuracy, demonstrating MORIC's adaptability and highlighting the potential of the proposed methodology for practical Wi-Fi sensing in real-world scenarios.