Chirp

Using mmWave radar nodes to track a moving target's trajectory through multiple FOVs

📡 Multi-Node Radar Fusion

Project Overview

As artificial intelligence is embodied into daily, real-world applications, the need for low-cost and low-compute methods of determining the kinematics of moving objects in the application environment becomes more prevalent to ensure correct operation and safety. Mmwave Radar, offering a small form factor and high range resolution could be a suitable alternative to a more mainstream approach like computer vision, which is computationally intensive in comparison.

Additionally, research has demonstrated that measurements from multiple mmWave radar nodes may be fused to accurately determine the relative pose of nodes and resolve the absolute trajectories of objects passing through the overlapping field of views of the nodes with higher accuracy.

01

Spatial Calibration

Accurately determine the relative poses of multiple radars (up to 4) by moving a single object through their FOVs

02

Trajectory Stitching

Once spatially synchronized, stitch together trajectories of objects moving through the FOVs

03

Real-Time Fusion

Perform fusion in a real-time field environment with millisecond-level time synchronization

How It Works

🔄

Time Synchronization

Each Jetson communicates with a base station for time synchronization over Wi-Fi

📡

Radar Data Acquisition

Base station commands each Jetson to start frame collection. The AWR board transmits chirps and receives reflections via CSI2 to DCA1000, then to Jetson over Ethernet

âš¡

Signal Processing Pipeline

ADC samples undergo sequential processing:

  • Range FFT: ADC samples → range profile
  • Doppler FFT: Chirps per frame → velocity profile
  • Static Clutter Removal: Filter out stationary objects
  • 2D OSCFAR: Detect likely target coordinates
  • Angle FFT: Generate range-doppler-angle point clouds
🎯

Data Fusion

Point clouds sent to base station for DBSCAN clustering. Extended Kalman Filter estimates relative radar positions and fuses trajectories across FOVs

System Architecture

Base Station Wi-Fi Sync Wi-Fi Sync Jetson Node 1 Jetson Node 2 DCA1000 DCA1000 AWR Radar AWR Radar Point Clouds Point Clouds 🎯

Hardware Layer

AWR Radar
DCA1000
Jetson Orin/Xavier

Processing Layer

Range FFT
Doppler FFT
Angle FFT
OSCFAR

Fusion Layer

DBSCAN Clustering
Extended Kalman Filter
Trajectory Estimation

Technical Highlights

Synchronization Millisecond precision
Processing Real-time pipeline
Nodes Up to 4 radar units
Communication Wi-Fi + Ethernet

Team

OS

Oviya Seeniraj

NTP & Time Synchronization

AO

Andrey Otvagin

Jetson/Node Development

VJ

Vihan Jayaraman

Wireless Infrastructure

WN

William Ni

Visualization Development

JW

Jason Wang

Auto-Calibration Algorithms