Skip to content

Overview

Introduction

mmWave FMCW radar is critical to many industries including obstacle sensing for cars and robots, person detection for security or opening doors, detecting different objects based on their reflection characteristics, or even recognizing targets based on their Radar signature using AI.

My goal is to build an advanced but affordable multi-purpose 77GHz mmWave Radar system which can be added as a sensor for various projects in the aforementioned fields. Some goals include:

  • Highly accurate distance and velocity measurements using mmWave (~77GHz) Radar.
  • Cost-effective design and circuit board material/fab
  • DSP on an FPGA (Zynq Ultrascale +). This includes mixers, etc. This will reduce the analog complexity of the design and make it significantly cheaper.

Ultimate goals

  • FPGA-accelerated machine vision AI model to detect objects. Custom model trained on external GPUs and then deployed to FPGA on Radar for edge computing application.
  • Implement phased array and beam sweeping capability

Design

Design Drawing

Component selection

K26 SoM

The K26 SoM by AMD/Xilinx hosting a Zynq Ultrascale+ SoC is the chosen SoC for this project. I already have this SoM on a KR260 robotics kit. I have therefore chosen to use the SoM for this Radar and simply remove it from the KR260. This lowers the cost of the project.

IWR1443 RFIC

I have chosen the IWR1443 RFIC. It contains the whole Analog Front End (AFE) and supports communication of raw ADC data straight to the K26 SOM using LVDS lanes.

It is chosen for it's low cost ($23) and integration of all the necessary analog systems which would be expensive to implement individually.

RFIC Block Diagram

Choices

I/O

This board should have plenty of I/O for both debug and deployment purposes. This includes:

  • JTAG for FPGA bitstream programming and for RFIC debug
  • UARTs for basic debug
  • QSPI flash (SD card) for storing default settings or FPGA bitstream and PetaLinux image
  • 1G RJ45 Ethernet and potentially even SFP/QSFP ultra-high-speed ports for live data transmission
  • USB-C data for easy connection to laptop for data transfer or setup
  • USB-C PD for power
  • 12V barrel jack for power
  • USB-A port for adding a WiFi adapter!

SSD

An SSD is added in order to store Radar data in standalone operation modes.

Power distribution

The power distribution for this project will be done using a number of discrete Power Managemet ICs (PMICs) such as Voltage Regulation Modules (VRMs), DC-DC converters, LDOs, etc.

Thermals

If needed, heasinks and fans will be added to the design to cool the RFIC. The KR260 has an attached fan and heatsink.

Antennas

The RFIC supports up to 3 TX and 4 RX antennas. These will be patch antennas for minimal footprint. These will be simulated using HFSS.

PCB

The stackup will likely be fully selected once the schematic is finalized. However, a few important issues are raised and are related to cost:

Backdrilling vias

Backdrilling vias is quite expensive and will therefore be avoided. Hence, sensitive impedance controlled signals such as RF traces and the LVDS signals will be routed on the external layers or layers close to the external layers. This way, any vias used go through the entire board and no stub is left.

Dielectric choice

The dielectric will need to be high quality and reduce losses. This is likely to be a Rogers dielectric or similar.

Manufacturing

The PCB manufacturer will be JLC PCB. They are cheap and cost is a concern. However, their capabilities are limmited and design decisions will need to be made in layout in order to meet these limitations and reduce the costs.

Digital Signal Processing (DSP)

The RFIC performs the mixing of the TX and RX signals in order to extract the beat frequency and related data. This is then fed into an ADC, the data of which can be sent directly to the FPGA for DSP.

The DSP on the FPGA will include:

  • Noise filtering
  • Target identification
  • Target highlighting
  • etc

The big advantage of using an FPGA for this is the ultra-high-speed parallel processing capacity, which will allow for high sampling rate.

Where does the data go?

The data can either be stored on the SSD or also sent via high speed ethernet interfaces via the RJ45 or possibly SFP/QSFP interfaces. This will allow for a program running on an external machine to plot the data in real time.

The device can also be connected to a machine on a robot as a sensor.

Linux on the SoM

A PetaLinux image will be generated for the SoM. This will be the Radar's OS and allow for easier setup of peripherals and user interfacing.