MobiSys 2026
Multi-static COTS radar implementation built on TI AWR2243/1243BOOST and the DCA1000EVM, enabling distributed multi-view sensing and coherent processing across multiple radar nodes.
- Python 3.9+
- uv (recommended) or pip
- CUDA-compatible GPU (required for CuPy)
uv syncRequires NVIDIA Container Toolkit for GPU access.
docker build -t muldar .
docker run --gpus all -v $(pwd):/app -it muldar bashThis creates a virtual environment and installs all dependencies in one step. (Note: cupy may take longer time for installation)
For best flexibility and most control over the radars, each radar are connected (USB cables) to a separate PC the runs TI mmWaveStudio (version 03.00.00.14) since it can't run multiple instants on a single machine. Copy the hardware/matlab and hardware/mmWaveStudio folders to PCs, then run connect_and_config.lua in mmWaveStudio software.
On those PCs, run MATLAB script studio_server.m for receiving radar commands (config radar, start frame, end frame) from the host computer, and it controls the TI mmWaveStudio following the official radar SDK.
A Raspberry Pi 4B is used for simultanously triggering the hardware trigger of radars. It use this repo: https://github.com/xsun2445/WiringPi-Python-MultiPin for simultanously triggering GPIOs which are connected to the hardware triggers of AWR2243BOOST, which is pin 9 SYNC_IN on J5 connector, doc.
Note: R62 need to be removed for enabling SYNC_IN on AWR2243/1243BOOST, detailed SYNC_IN signal requirements are in 5.5.3 of mmwave_dfp_02_02_04_00 mmWave-Radar-Interface-Control.pdf that can be downloaded from TI.
Each radar has a distinct ip for data and config port which can be configured using scrips/config_dca_eeprom.py. All radars, PCs, Raspberry Pi are connected to a single network switch used for communication and radar data transfering. Radar data from all 3 boards are streamed to the host computer and the host PC simultanously renders the scene for visualization.
Measure the [x,y,yaw] coordinates of those radars and write in the configs.yml, no need to be super precise.
Then on host PC, run scripts/radar_config.py. It will config the starting frequency and transmitting/receiving for each of the radar through the MATLAB scripts on slave PCs. Run play.py with visualize_2d_fft(mgr), it will show the beamforming images of all radar channels. Each colomn represents a transmitting antenna, each radar has 2 TX antennas so there will be 6 colomns in total. Each row represents a receiving radar, it shows a beamforming image using 4 RX antennas. Hence the diagonal beamforming images will always be stable since they are monostatic channels, but other images are bi-static channels which shows frequency offsets (the range profile will goes up and downs).
Adjust the starting frequency of bi-static channels, then config and visualize again untill it shows solid peaks in all channels.
initial config after calibration
Finally, run play.py with start_visualization(mgr) or start_combined_visualization(mgr).
python play.py| Option | Type | Description |
|---|---|---|
--config-path |
str | Path to YAML config file (default: configs/config_new.yml) |
--flag-save / --no-flag-save |
bool | Enable/disable data saving |
--flag-visualize / --no-flag-visualize |
bool | Enable/disable real-time visualization |
--wait-for-threads / --no-wait-for-threads |
bool | Wait for threads to finish |
--num-trigger |
int | Number of triggers |
--period-frame |
int | Frame period |
--radar-timeout |
int | Radar timeout |
--saving-root-dir |
str | Root directory for saved data |
--warmup / --no-warmup |
bool | Run warmup before acquisition |
By configuring the config.yml, radar stream can be played from either real radar or a recorded file.
python scripts/single_frame_imaging.pyFor more evaluations please refer to evaluations/README.md.
If you find MulDar useful, please consider citing our MobiSys paper:
@inproceedings{sun2026muldar,
author = {Sun, Xinghua and Li, Qiancheng and Gadre, Akshay},
title = {MulDar: Unleashing the Potential of Distributed COTS mmWave Radar by Exploiting Cross-Device Channels},
booktitle = {Proceedings of the 24th Annual International Conference on Mobile Systems, Applications and Services (MobiSys '26)},
year = {2026},
doi = {10.1145/3745756.3809206},
publisher = {ACM},
address = {New York, NY, USA},
month = {June},
}MulDar/
├── play.py # Main entry point for data acquisition
├── configs.yml # Default configuration file
├── pyproject.toml # Python project & dependency config
├── uv.lock # Dependency lock file
├── Dockerfile # Docker build file
├── LICENSE
│
├── src/muldar/ # Core library package
│ ├── configs.py # Configuration loading & dataclasses
│ ├── utils.py # Utility functions
│ ├── devices/ # Hardware device interfaces
│ │ ├── muldar.py # Multi-radar system controller
│ │ ├── radar.py # Single radar (DCA1000EVM) interface
│ │ └── motor.py # Motor control for SAR scanning
│ ├── dsp/ # Digital signal processing
│ │ ├── dsp.py # Core DSP pipeline (range/Doppler FFT)
│ │ ├── bistatic.py # Bistatic radar processing
│ │ ├── sar.py # SAR imaging algorithms
│ │ ├── cfar.py # CFAR detection
│ │ └── music.py # MUSIC algorithm
│ ├── vis/ # Visualization
│ │ ├── vis.py # Plotting utilities
│ │ └── network_vis.py # Radar network visualization
│ └── eval/ # Evaluation metrics
│ └── chamfer.py # Chamfer distance metric
│
├── scripts/ # Standalone utility scripts
│ ├── calibration.py # Radar calibration
│ ├── config_dca_eeprom.py # DCA1000 EEPROM configuration
│ ├── radar_config.py # Radar parameter configuration
│ └── single_frame_imaging.py # Single-frame imaging script
│
├── evaluations/ # Evaluation experiments
│ ├── download_dataset.py # Dataset download script
│ ├── car/ # Car imaging evaluation
│ ├── chamfer/ # Chamfer distance evaluation
│ └── common_objects/ # Common object imaging evaluation
│
├── hardware/ # Hardware resources
│ ├── antenna.ipynb # Antenna pattern analysis
│ ├── listen_trigger.py # Trigger listener for sync
│ ├── cadmodels/ # 3D-printable CAD models (.stl)
│ ├── matlab/ # MATLAB scripts for mmWaveStudio
│ │ ├── mmWaveStudio/ # Lua configs for AWR1243/2243
│ │ ├── studio_server.m # MATLAB-mmWaveStudio bridge
│ │ └── *.m # RSTD connection scripts
│ └── radar/ # Radar config templates
│
├── adcData/ # Raw ADC data (not tracked in git)
│ ├── car/ # Car scene captures
│ ├── curve/ # Curved surface captures
│ ├── deformable/ # Deformable object captures
│ ├── drywall/ # Drywall captures
│ ├── fabrics/ # Fabric captures
│ ├── metal/ # Metal surface captures
│ ├── plastic/ # Plastic surface captures
│ ├── real_object/ # Real object captures
│ └── wood/ # Wood surface captures
│
└── assets/ # README images