Figure 2-5 shows the timing of the chirps and subsequent processing in the system.

Figure 2-5 Chirp Timing Sequence

Chirp acquisition happens in the radar front end, using a BPM-MIMO scheme. Due to the implementation of the maximum velocity extension feature in the BSD demo, which relies on varying chirp durations from frame to frame, the chirp duration ("short" and "long") is reconfigured with each frame.

The core of the data path processing, from chirp acquisition to point cloud and tracker output, is divided into the following data processing units (DPUs):

• Range DPU

• Doppler DPU

• CFAR DPU

• 2D Angle-of-Arrival (AoA2D) DPU

• Clutter Removal DPU

• Group Tracker DPU

As the acquisition occurs, the Range DPU performs 1D FFT for each antenna and chirp in parallel to the acquisition, compresses the output, and stores the output in memory as the compressed radar cube.

Next, the Doppler DPU decompresses the radar cube one range bin at a time and for each bin calculates velocity information by performing Doppler FFTs. The Doppler FFT magnitudes are sum across all virtual antennas for each range bin to create a Range-Doppler detection matrix. This is done in the hardware accelerator (HWA).

After this, the CFAR DPU then computes and crosschecks detected points in both range and Doppler dimensions. The AoA2D DPU then performs the maximum velocity extension algorithm to extend the velocities of detected points beyond Nyquist limits and generates a point cloud list in Cartesian format. These DPUs utilize both the HWA and M4F to achieve this.

Finally, the Clutter Removal DPU removes detected points identified as stationary roadside clutter and the Group Tracker DPU performs object tracking based on the point cloud data. Both the final point cloud list and tracker list are transmitted over UART.

For more details on the application flow and processing, see the mmWave low-power software development kit (MMWAVE-L-SDK).