MMWiLoc: A Multi-Sensor Dataset and Robust Device-Free Localization Method Using Commercial Off-The-Shelf Millimeter Wave Wi-Fi Devices

The DATA

The Data is available at the following link:

https://pan.baidu.com/s/1iYMAeTR7FSUjaYhVwq9ZQg?pwd=pjmv

or

https://drive.google.com/drive/folders/1ao1aB2wNAWePW--nphxJXJFen__ukPFO?usp=sharing

More data and documentation will be added to the repository in the future, but the above link contains the core measurement assets used in the experiments.

The CODE

This repository contains research code for millimeter-wave WiFi localization, with millimeter-wave radar processing code included as a reference baseline.

The project is organized around two main scripts:

• mmwifi_process.py for millimeter-wave WiFi localization
• radar_process.py for radar-based reference processing

The PAPER

Arxiv: https://arxiv.org/pdf/2506.11540

Acknowledgements

Thanks to the following open-source projects that informed parts of this repository:

• PreSenseRadar/OpenRadar for the radar processing pipeline reference
• seemoo-lab/talon-sector-patterns for the measured mmWiFi sector patterns

Project overview

The main goal of the repository is to study how localization can be performed from millimeter-wave wireless measurements and compared against radar-style sensing pipelines.

At a high level, the codebase includes:

• a WiFi localization pipeline built around sector sweep measurements and sparse reconstruction
• a radar reference pipeline built around range, angle, Doppler, and tracking
• measurement assets used to model directional behavior
• tracking and evaluation tools for analyzing estimated trajectories
• CSI conversion and parsing helpers for data preparation

Core components

WiFi localization

The WiFi side of the repository focuses on converting directional millimeter-wave WiFi measurements into angle and position estimates. It uses measured sector patterns from the repository as a sensing dictionary and applies sparse estimation methods to infer likely target directions.

The main implementation lives in mmwifi_process.py.

Radar reference code

The radar side provides a conventional mmWave radar processing pipeline for comparison. It includes ADC loading, range processing, beamforming, detection, Doppler estimation, clustering, and target tracking.

The main implementation lives in radar_process.py.

Evaluation and analysis

The repository also contains scripts for evaluating localization outputs against ground-truth trajectories or shapes, and for combining metrics across multiple experiments.

Relevant files include:

• precision.py
• combined_summary.py

Data conversion and utilities

Several helper scripts support data preparation and experimentation, including CSI conversion and parsing tools.

Relevant files include:

• save_to_h5.py
• subwifi_process.py
• util.py
• omp.py

Repository layout

• mmwifi_process.py — WiFi localization entry point
• estimate_meas_matrix.py — measurement matrix estimation
• radar_process.py — radar reference entry point
• precision.py — evaluation utilities
• combined_summary.py — combined experiment summaries
• save_to_h5.py — CSI to HDF5 conversion
• subwifi_process.py — processed CSI reading helpers
• util.py — shared utility functions
• stone_tracker.py — Stone Soup-based tracking utilities
• dsp/ — radar DSP modules
• dataloader/ — radar data loading helpers
• tracking/ — radar tracking implementation
• clustering/ — clustering helpers
• precise_measurements/ — measured WiFi sector pattern data
• array_factor/ — antenna/array factor reference data
• legacy_measurements/ — legacy calibration and plotting materials
• parser_scripts/ — parser examples and support scripts

Entry points

If starting from the code, begin with:

• mmwifi_process.py
• radar_process.py

These two scripts define the primary workflows for the repository.