SDP: A Unified Protocol and Benchmarking Framework for Reproducible Wireless Sensing

TL;DR

The paper tackles reproducibility in learning-based wireless sensing by addressing hardware-induced variability that hampers fair cross-device evaluation. It introduces the Sensing Data Protocol (SDP), a protocol-level framework that enforces deterministic signal sanitization, canonical frequency projection, and canonical tensor construction to map heterogeneous CSI into a device-agnostic representation, paired with a fixed Transformer backbone for benchmarking. Across multiple sensing tasks and datasets, SDP delivers competitive accuracy while dramatically reducing inter-seed variance and improving reproducibility, including a real-world deployment with commercial Wi-Fi hardware. By decoupling learning performance from hardware artifacts and providing a standardized evaluation pipeline, SDP enables fair, cross-task, cross-device comparisons and paves the way for more reliable engineering practice in wireless sensing.

Abstract

Learning-based wireless sensing has made rapid progress, yet the field still lacks a unified and reproducible experimental foundation. Unlike computer vision, wireless sensing relies on hardware-dependent channel measurements whose representations, preprocessing pipelines, and evaluation protocols vary significantly across devices and datasets, hindering fair comparison and reproducibility. This paper proposes the Sensing Data Protocol (SDP), a protocol-level abstraction and unified benchmark for scalable wireless sensing. SDP acts as a standardization layer that decouples learning tasks from hardware heterogeneity. To this end, SDP enforces deterministic physical-layer sanitization, canonical tensor construction, and standardized training and evaluation procedures, decoupling learning performance from hardware-specific artifacts. Rather than introducing task-specific models, SDP establishes a principled protocol foundation for fair evaluation across diverse sensing tasks and platforms. Extensive experiments demonstrate that SDP achieves competitive accuracy while substantially improving stability, reducing inter-seed performance variance by orders of magnitude on complex activity recognition tasks. A real-world experiment using commercial off-the-shelf Wi-Fi hardware further illustrating the protocol's interoperability across heterogeneous hardware. By providing a unified protocol and benchmark, SDP enables reproducible and comparable wireless sensing research and supports the transition from ad hoc experimentation toward reliable engineering practice.

Figures (8)

• Figure 2: The SDP unified data processing pipeline. This figure details the SDP's deterministic workflow: Stage I removes hardware impairments like STO and CFO; Stage II projects diverse CSI into a uniform Canonical Tensor; Stage III flattens the tensor into a token sequence, ready for the Transformer backbone.
• Figure 3: Performance of the SDP benchmark across heterogeneous sensing tasks. (a) Mean Top-1 accuracy with 95% confidence intervals over five runs. (b--e) Normalized confusion matrices on Widar3.0, GaitID, XRF55, and ElderAL-CSI, showing consistent diagonal dominance across datasets.
• Figure 4: Performance stability comparison between the baseline and SDP across five random seeds. Boxplots show the distribution of Top-1 accuracy, with scattered dots indicating individual runs.
• Figure 5: Representation-level ablation via DFS spectrogram visualizations across heterogeneous tasks. Rows correspond to datasets (Widar3.0, GaitID, XRF55). In each subfigure, the left panel shows the raw spectrogram, while the right panel shows the SDP-processed representation. SDP consistently suppresses hardware-induced noise and enhances motion-related structures across tasks.
• Figure 6: Ablation study results analyzing performance stability across datasets (a) and rank consistency on ElderAL-CSI (b).