TIDUF89 September   2024

 

  1.   1
  2.   Description
  3.   Resources
  4.   Features
  5.   Applications
  6.   6
  7. 1System Description
    1. 1.1 Detection Theory
    2. 1.2 Multi-Pass Architecture
  8. 2System Overview
    1. 2.1 System Design Theory
      1. 2.1.1 Long Detection Range
        1. 2.1.1.1 Antenna Design for Long Detection Range
        2. 2.1.1.2 SNR Compensation for Long Detection Range
        3. 2.1.1.3 Smart Detection Logic
      2. 2.1.2 Low Power Consumption
        1. 2.1.2.1 Efficient Chirp Design
        2. 2.1.2.2 Deep Sleep Power Modes
        3. 2.1.2.3 Hardware Accelerator
      3. 2.1.3 Low False Alarm Rate
        1. 2.1.3.1 Typical Causes of False Alarms
        2. 2.1.3.2 False Alarms Outside the Detection Zone
        3. 2.1.3.3 False Alarms Within the Detection Zone
        4. 2.1.3.4 Adaptive State Machine
  9. 3Hardware, Software, Testing Requirements, and Test Results
    1. 3.1 Hardware Requirements
    2. 3.2 Software Requirements
    3. 3.3 Test Setup
      1. 3.3.1 Test 1 - Detection Range
      2. 3.3.2 Test 2 - False Alarm Rate
      3. 3.3.3 Test 3 - Power Consumption
    4. 3.4 Test Results
  10. 4Design Files
    1. 4.1 Schematics
    2. 4.2 Bill of Materials
  11. 5Tools and Software
  12. 6Document Support
  13. 7Support Resources
  14. 8Trademarks
  15. 9About the Authors

2.1.1.3 Smart Detection Logic

As the motion and presence documentation in the MMWAVE-L-SDK shows, the IWRL6432AOP detects points with the following steps:

  1. Compute range, Doppler, azimuth, and elevation FFTs on the incoming data
  2. Compute the sum or maximum over the Doppler and elevation dimensions
  3. Run CFAR on the range-azimuth heat map in the range dimension
  4. Filter detected points to reject side-lobes (peaks in the same range bin caused by power from a different azimuth bin spilling over into adjacent azimuth bins)
TIDEP-01035 Motion and Presence Detection Block Diagram Figure 2-1 Motion and Presence Detection Block Diagram

After points are detected, the video doorbell reference design passes them into a state machine to determine whether presence is detected in a certain zone. This state machine is based off the MPD State Machine, which is also described in the MMWAVE-L-SDK. The state machine examines all the points, clusters them together using the DBSCAN algorithm, counting the number of points and SNR statistics on the points in the cluster. Based off that information, the state machine determines whether a zone is occupied or unoccupied. The following sequence illustrates how a zone transitions from the unoccupied to the occupied state. For a full description of the state machine flow, see the tuning guide, found in the docs/ folder of the MMWAVE-L-SDK.

Generally speaking, when people are closer to the radar, the number of points detected is a better indicator of presence than the SNR of the points. However, when targets are further away from the radar, the SNR of the detected points becomes a better indicator than the number of points. The video doorbell reference design provides ways to achieve robust detection at short and long ranges by favoring the number of detected points at short ranges, but favoring the detection SNR at long ranges.