Skip to main content

Radar Research

This page outlines initial understanding of the FMCW radars used by Texas Instruments

How does radar work?​

  • mmWave radar devices send out pulses of EM waves (called chirps) which will reflect off of any targets in the area and be received by the radar giving it information on range, velocity and direction.
  • TI uses FMCW tech (frequency modulated continuous wave) which means their radars are constantly sending out chirps rather than sending out chirps in frequent pulses.
  • Chirps are sent out by the TX (transmit) antenna and received by the RX (receive) antenna.
  • Chirps are first generated by a synth
  • The TX and RX chirps are combined by a signal mixer to create an intermediate frequency signal (IF signal)
  • The intermediate signal is then processed by analog filters and an ADC (analogue to digital converter)
  • The digital data is then either processed on the radar or on an alternate processor to generate the map of the targets location
  • When a chirp is plotted on amp vs time it is a sine wave which increases in frequency
  • Range estimations are calculated using the difference in frequency from the RX and TX chirps
  • Velocity estimations are calculated using a fourier transformation on the IF signal

What does radar data look like?​

  • Most TI radar sensors have an on board processor that turns complex raw data into a more manageable format before ever leaving the device
  • Data is sent in frames which are like snapshots in time
  • Frame rate is configurable (normally 1-20Hz)
  • The sensor sends data using standard interfaces such as - UART, SPI or I2C
  • The data in each frame is usually organized into packets called TLVs (Type, Length, Value)
  • The TLVs clearly define what kind of information is being transmitted

Types of radar data​

  • ADC data (raw data) : The most basic data from the sensor - useful for developing new algorithms
  • Range profile : A simple plot showing how far away objects are, but provides no angle information
  • Point cloud : The most common output. A list of detected points, each with an individual 3D position, velocity and signal strength
  • Occupancy detection : A boolean output that indicates if there is an object in a pre-defined 3D area
  • Tracks/targets : A more advanced processing step that groups points together to track real world objects (think a person moving - would have many different chirps being received by the RX, this groups it to show it as a person moving)
  • Classifiers : USes on-chip ML to classify detected objects (determining if the target is a human or a chair for example)