CASCADE project

4D radar with a compact architecture for object detection and tracking

Name: AWR2944P — Compact Single-Chip 4D Radar (4TX/4RX, on-chip DSP + HWA)
Brand: CASCADE

A prototype built on TI's single-chip AWR2944P radar and the DCA1000EVM raw-data capture module. The radar "sees" through darkness, fog and smoke where optics and lidar fail — making it suitable for autonomous platforms, perimeter security and search-and-rescue missions.

Technical specifications Join the team

~220 m

detection range

4 TX / 4 RX

transmitters / receivers

virtual MIMO channels

drag to rotate

Why radar, not a camera or lidar

Microsecond-long 76–81 GHz chirps pass through rain, dust and darkness, measuring range, angle and velocity simultaneously. The AWR2944P combines a C66x DSP, an Arm Cortex-R5F and a hardware accelerator (HWA) on a single chip, so signal processing happens right on the board, and its compact form factor suits wide-angle coverage on mobile platforms.

Choose the radar variant

Prototype hardware

Hardware platform

1 × AWR2944P, 76–81 GHz

Radar module

compact wide-angle surveillance radar

Boards AWR2944PEVM + DCA1000EVM (raw data capture)

Transmitters / receivers 4 TX / 4 RX

Virtual MIMO array 16 channels

Processing on-chip DSP C66x, Arm Cortex-R5F, hardware accelerator (HWA)

Detection range and cost

Based on TI data

Estimated detection range for different object types and the approximate cost of the prototype boards.

Detection range · FPV drone ~40–60 m

Detection range · car ~220 m

Board price $768.90 + $691.85

How it works

From chirp to object coordinates

Chirp transmission

Antennas on the board form a frequency-modulated 76–81 GHz signal aimed at the object within a 120° azimuth sector.

Signal reception

The AWR2944P chip, with 4 TX/4 RX, receives the reflected signal across a wide field of view.

DSP capture

The DCA1000EVM captures raw ADC data while the on-chip DSP C66x and HWA process the signal on the board itself.

Object data output

Range, angle and velocity of objects are sent over CAN-FD, Ethernet or to the host system.

Prototype in hardware

From evaluation board to a working prototype

An FMCW Image Radar assembled into a working rig, controlling a rotating platform with two optical channels based on AI object recognition via a Hailo hardware accelerator for real-time operation.

AWR2944PEVM board in its handled enclosure, front view, cut out from background — 01 · AWR2944PEVM, front view

DCA1000EVM board in its enclosure, cut out from background — 02 · DCA1000EVM data capture

Hailo accelerator board on a HAT+ for Raspberry Pi 5, cut out from background — 03 · Hailo accelerator on a Raspberry Pi 5

Dual-camera module on a servo gimbal, cut out from background — 04 · Dual-camera module on a servo gimbal

AWR2944P rig with a camera on a rotating platform, cut out from background — 05 · Working rig with a rotating platform

Live demonstration

Camera and radar — in sync

The first video is a synchronized capture of the camera feed and the radar's point cloud (range-azimuth) in nighttime city traffic: the radar keeps a solid lock on objects where oncoming headlights blind the camera. The second is a recording of the live 3D point-cloud viewer, showing the same data in real time and space.

REC · Live capture

02:50 RANGE-AZIMUTH