wsdp 0.0.9

A pipeline for large scale deep-learning based wireless sensing tasks, including preprocess, training and eval

SDP: Sensing Data Protocol for Scalable Wireless Sensing

SDP (Sensing Data Protocol) is a protocol-level abstraction framework and unified benchmark for scalable wireless sensing and perception based on wireless signals such as Channel State Information(CSI). The protocol is designed to decouple learning performance from hardware-specific artifacts, enabling fair, reproducible, and scalable evaluation of deep learning models for wireless sensing tasks.

SDP enforces deterministic physical-layer sanitization, canonical tensor construction, and standardized training and evaluation procedures, making it particularly suitable for wireless sensing research, activity recognition, device-free sensing, and cross-dataset benchmarking.

Our main result can be illustrated by the following pictures.

Mean Top-1 accuracy with 95% confidence intervals over five runs

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.

Rank consistency heatmap across five random seeds on the ElderAL-CSI dataset. Colors indicate per-seed performance rank (1 = best), with overlaid Top-1 accuracy values. Full SDP exhibits stable top-ranked performance, while ablated variants show higher ranking variability.

More details are illustrated in our paper A Sensing Dataset Protocol for Benchmarking and Multi-Task Wireless Sensing.

@misc{zhang2026sdpunifiedprotocolbenchmarking,
      title={SDP: A Unified Protocol and Benchmarking Framework for Reproducible Wireless Sensing}, 
      author={Di Zhang and Jiawei Huang and Yuanhao Cui and Xiaowen Cao and Tony Xiao Han and Xiaojun Jing and Christos Masouros},
      year={2026},
      eprint={2601.08463},
      archivePrefix={arXiv},
      primaryClass={eess.SP},
      url={https://arxiv.org/abs/2601.08463}, 
}

---

🔍 Why SDP?

Wireless sensing research often suffers from:

• Inconsistent hardware configurations
• Dataset-specific preprocessing pipelines
• Non-reproducible training and evaluation protocols

SDP addresses these challenges at the protocol level, rather than the model level. The SDP unified data processing pipeline, including sanitation and transformation, transform raw data into uniform canonical tensors ready for deep learning.

Core Design Principles

• Protocol-level abstraction
• Deterministic PHY-layer sanitization to eliminate randomness
• Canonical tensor representation for deep learning compatibility
• Unified benchmark pipeline across datasets and tasks
• Extensible architecture for new datasets, processors, and models

---

📦 Key Features

• Unified CSI abstraction across heterogeneous datasets
• Hardware-agnostic signal representation
• Modular reader–processor–model pipeline
• Deterministic preprocessing for reproducibility
• Plug-and-play extensibility
• Benchmark-ready training and evaluation flow

---

📚 Target Use Cases

SDP is optimized for:

• CSI-based Human Activity Recognition (HAR)
• Gait recognition and biometric identification
• Wireless sensing + deep learning research
• Cross-domain / cross-hardware generalization
• scalable sensing systems

Typical downstream models include CNNs, Transformers, BiLSTMs, GNNs, and hybrid architectures.

Supported Dataset:

Widar3.0

• Dataset Link: Widar3.0: Wi-Fi-based Hand Gesture Recognition Dataset
• CSI Shape: (Time, 30, 1, 3)
• num of classes: 6
• total num of used samples: 12,000

GaitID

• Dataset Link: GaitID: Wi-Fi-based Human Gait Recognition Dataset
• CSI Shape: (Time, 30, 1, 3)
• num of classes: 11
• total num of used samples: 22,500

XRF55

• Dataset Link: XRF55: A Radio Frequency Dataset for Human Indoor Action Analysis
• CSI Shape: (270, 1000)
• num of classes: 55
• total num of used samples: 9,900

ElderAL-CSI

• Dataset Link: ElderAL-CSI
• CSI Shape: (Time, 512, 3, 3)
• num of classes: 6
• total num of used samples: 2,400

---

📁 Project Structure Overview

algorithms/

Store various functions for implementing different signal processing algorithms.

• ./denoising.py
  Store functions for signal denoising
• ./phase_calibration.py
  Store functions for phase calibration

---

readers/

• Store Dataset-specific readers
• Converts raw files into List of CSIData

---

structure/

• Definition of CSIData and all kinds of CSIFrame

---

processors/

• Definition of processor for signal processing and sanitization

---

datasets/

• Definition of classes extend torch.utils.data.Dataset

---

models/

• Definition of deep learning models

---

🚀 Quick Start

Install Dependencies

Create a venv for dependencies, then run:

pip install wsdp

Download Data

Please download needed datasets from Our SDP Website or via command:

wsdp download [dataset_name] [dir]

dataset_name must be widar, gait, xrf55 and elderAL

After downloading, in the folder of your project, please organize elderAL datasets in the structure below for extracting labels:

├── data
    ├── elderAL
    │   ├── action0_static_new
    │   │   ├── user0_position1_activity0
    │   │   ├── ...
    │   │
    │   ├── action1_walk_new
    │   ├── ...
    │
    ├── widar
    ├── gait
    ├── xrf55

Run

This project supports both functional call and command-line call in the shell. The calling methods are as follows respectively:

---

• For input_path: /data/[widar, gait, xrf55, elderAL] are recommended
• For dataset_name: widar, gait, xrf55, elderAL are available

---

function call:

Create a script, say script.py, then copy the code below and paste into the script:

from wsdp import pipeline

pipeline(input_path, output_folder, dataset_name)

Considering that training process will generate numerous lines about information like acc and loss, in function-call, it is recommended to run via the command below:

nohup python script.py >> output.log 2>&1 &

command:

no need to create scripts, just run in command:

wsdp run [input_path] [output_folder] [dataset_name]