AWRL6844 Sbr Users Guide

Table of Contents

Overview

This lab demonstrates the use of TI 60GHz mmWave sensors AWRL6844 for in-cabin sensing applications including Vehicle Intruder Detection (ID), Child Presence Detection (CPD) and Seat Belt Reminder (SBR). CPD and SBR will share the same signal processing chain for point cloud generation. Note that this lab is tested and tuned with AWRL6844EVM board and top front console mounting. Optimization effort will be needed to achieve best performance for other antenna design and other mounting position.

The same demo binary and visualizer will be used for all three applications, i.e., ID, CPD and SBR. Different configuration files can be loaded which enable the three different applications.

In this users guide, we will focus on SBR application. Real-time demo recording clips are shown below to demonstrate the SBR performance in driving. In these video clips, front row is always occupied with a driver and a passenger holding a PC, which is not obvious due to the limitation of the camera field of view. On the second row, different people combinations are used. A 24-bottle water pack is added in several test cases to demonstrate the robustness to the non-living object.

Requirements

Hardware Requirements

Software Requirements

Calibration

In order to get the best detection performance, users should run the Out of Box Demo Range Bias and Rx Channel Gain/Phase Measurement and Compensation procedure to get the calibration coefficients for their Antenna Module and replace the default coefficients in the desired configuration file which can be found in <Radar_Toolbox_LOCATION>\tools\visualizers\AWRL6844_Incabin_GUI\src\chirpConfigs6844 with the values returned by the calibration procedure.

The following command in the chirp configuration file needs to be updated with the calibrated values.

compRangeBiasAndRxChanPhase 0 1 0 1 0 1 0 1 0 -1 0 -1 0 -1 0 -1 0 1 0 1 0 1 0 1 0 -1 0 -1 0 -1 0 -1 0 (This is the default command)

The updated command will look similar to the following (but with different values)

compRangeBiasAndRxChanPhase 0 1.00  0  0.90 0.44 0.98 0.21  1.00 -0.09 -0.65 -0.76 -0.13 -0.99  -0.47 -0.88  -0.73 -0.68  0.54 0.84 -0.03 1.00 0.31 0.95 0.61 0.79 -0.98 -0.21 -0.68 -0.79 -0.90 -0.42 -0.98 -0.19 

The instructions to run the calibration procedure is not yet available. For now, please follow the instructions provided for AWR6843 device in Out of Box demo HTML documentation given in the calibration section of the file listed below. <MMWAVE_SDK3_INSTALL_DIR>\packages\ti\demo\xwr68xx\mmw\docs\doxygen\html\index.html We will update this section soon.

Quickstart

1. Configure the EVM for Flashing Mode

Instructions for EVM setup for Flashing mode can be found in the EVM Setup Guide

2. Flash the EVM using Uniflash

Flash the binary listed below using Uniflash. Follow the instructions found in the Using Uniflash Guide
<Radar_Toolbox>\source\ti\examples\Automotive_InCabin_Security_and_Safety\AWRL6844_InCabin_Demos\prebuilt_binaries\demo_in_cabin_sensing_6844_system.release.appimage

3. Configure the EVM for Functional Mode

Instructions for EVM setup for Functional mode can be found in the EVM Setup Guide

4. Mount the EVM and Create Test Environment

Mount the sensor at the desired location securely so that the sensor is stable during runtime. Update the desired configuration file so that it reflects the device’s location. More information on mounting can be found in the section below.

5. Launch the Visualizer

⚠️ MATLAB Runtime Version R2024b (24.2)
Exact version R2024b (24.2) required.

1. Launch the visualizer executable located at <Radar_Toolbox>\tools\visualizers\AWRL6844_Incabin_GUI\src\
2. Double click to launch occupancy_demo_gui.exe.
3. A black console window will appear. After 30-60sec, a configuration window will appear as shown below.

   

6. Config Visualizer

1. The COM ports will be auto-populated. Users should confirm or change as needed.

2. (Optional) Select and enable the webcam if it is connected to the PC

3. You should see the chirp configuration file automatically selected in the Configuration dialogue. Always browse and choose the configuration file when running for the first time. If the configuration file is not the correct one, browse to select a file provided in the config_file directory: <Radar_Toolbox_LOCATION>\tools\visualizers\AWRL6844_Incabin_GUI\src\chirpConfigs6844

   📝 NOTE
   The same demo binary supports Intruder Detection, Child Presence Detection and Seat Belt Reminder mode by loading different configuration file.

4. Press Done to start the Visualizer.

5. The Visualizer window will appear as shown below.
   

   

6. Press OK to start the running.

7. The Visualizer window will start output the point cloud. After 20 frames (as a default setting), the demo will start output the occupancy decisions. The zone will be shown as green if detected as occupied.

7. Understand the Visualizer

The Visualizer window is divided into six panels as shown in the annotated picture here.

1. Statistics: This section prints real-time information from the target: Frame number, total number of detected points and occupancy status. Processing time is also provided.

2. Chirp Configuration: This panels shows the static chirp configuration loaded at startup time.

3. Control: This panel provides an Exit button and a Pause button.

4. Point Cloud and Decision: This panel display the point cloud as well as the occupancy decisions. The point clouds are not per frame based, instead, accumulated point cloud from the most recent 20 fraems are showed. The occupancy display assumes 2 row car with 5 seat. Users can modify the zone definition as needed. Users can rotate the figure to see 2D and 3D figure.

5. ML Probability: This panel display the occupancy probability (between 0 and one) for all the zones.

6. Ground Truth Video: A camera video can be included to the visualizer to serve as the groud truth video. However, we have observed some delays with this feature enabled.

8. Rotate the Point cloud and decision plot

Users can rotate the point cloud and decision figure to see a 3D view, a useful tool when tuning the zone definition and understand the false detection. Note that currently visualizer only displays the point within the zone definition and different color is used for different zone.

Note that by default, the TLV output data for each frame will be saved automatrically in the subfolder \fHistFiles as fHistRT_0000.mat, fHistRT_0001.mat and etc. Each file will contain TLV results for 1000 frames. These files will be overwritten at the next run. Users need to move them to avoid data files from being overwritten. Users also need to copy this files to other directory after each run to avoid files being combined non-intentionally.

Algorithm Overview

This demo can be viewed as two pieces: the point cloud detection and the high-level processing.

• The point cloud detection is based on the AWR6843 CPD demo with classification, and additional information related to that part of the algorithm can be viewed in the AWR6843 CPD With Classification Users Guide

• High-level processing is ML-based to make a decision whether a seat zone is occupied or not.

In order to increase performance of detection of nearly-static objects such as people sitting still in the vehicle, this lab uses multiple frames to generate range-angle heatmap, rather than a single frame. By using a sliding window of multiple frames, there is a finer granularity of detection with respect to time. The number of frames in this window is configured through CLI command sigProcChainCommonCfg (set to 4 by default).

The high level flow for this lab is described here:

1. This demo uses the standard range processing used in most mmWave Radar demos, and is controlled by the HWA (Hardware Accelerator) to produce the radar cube.
   
2. When the range processing is done for every chirp in the frame, RadarCube is ready. ARM core will send a IPC (inter-processor communication) message DPC_MSS_TO_DSS_RADAR_CUBE_READY.
3. Upon receiving IPC message DPC_MSS_TO_DSS_RADAR_CUBE_READY, the DSP core will call DPC_ObjectDetection_execute function to start the inter-frame process for point cloud detection. The point cloud generation contains 3 steps:
   • 3D range-azimuth-elevation heatmap is generated using Capon beamforming algorithm on a coarse angle grid.
   • 2-pass CFAR is applied on both range and angle direction.
     
   • For every detected peak, zoom-in azimuth-elevation angle spectrum will be calculated to get accurate angle estimation.
4. After the point cloud generation, DSP core will send out a IPC message DPC_DSS_TO_MSS_POINT_CLOUD_READY.
5. Upon receiving this IPC message, the ARM core will post a Semaphore classifierTaskSemHandle.
6. MmwDemo_ClassifierTask processes the feature extraction and classification (upon receiving Semaphore classifierTaskSemHandle) and posts Samaphore tlvSemHandle.
7. MmwDemo_transmitProcessedOutputTask outputs the point cloud, features and classifier information through UART (upon receiving Semaphore tlvSemHandle).

EVM Mounting and Coordinate Transforms

The detected point cloud is all relative to the sensor. To make the occupancy decision for each seat in the car, we need to transform the point cloud from sensor coordinates to the car coordinates. The car coordinate are plotted in the figure below.

Since the point cloud will be transformed to the car’s coordinates, the zone definition for each seat need to follow the car’s coordinates as well.

• For x direction, 0 represents the center of the car, the driver side X values are positive, the passenger side are negative;
• For y direction, 0 represents front window.
• For z direction, 0 represents the car floor.

There are different positions and different mounting angles that can be considered for mounting the device. To give the full flexibility, “sensorPosition” CLI is used to indicate the mounting offset in (x, y, z) and mounting rotation angle in x-y plane and y-z plane. Based on these offsets and tilting angles, the visualizer will transform the point cloud from the sensor coordinate to the car coordinate.

Next we give one example:

• when the sensor is mounted around the center close to the front overhead console tilting down by 60 degrees, then sensor position should be programmed as: sensorPosition 0 0.8 1.15 0 -60

This indicates the sensor is mounted at (x = 0, y = 0.8m, z = 1.15m) and is tilted down by 60 degree in y-z domain to facing the floor. Please referred to the figure below to help understanding.

Antenna Configuration

In this package, we only provided an example configurations with TI xWRL6844 EVM modules. The AWRL6844 Antenna is shown in the following image along with the virtual antenna array. There are some build-in 180 phase offset between different channels.

antGeometry0 -2 -2 -3 -3 0 0 -1 -1 0 0 -1 -1 -2 -2 -3 -3
antGeometry1 0 -1 -1 0 0 -1 -1 0 -2 -3 -3 -2 -2 -3 -3 -2
antPhaseRot 1 1 1 1 -1 -1 -1 -1 1 1 1 1 -1 -1 -1 -1

Note that it is suggested to using compRangeBiasAndRxChanPhase to compensate the phase (including this 180 degree built-in phase offset), and program antPhaseRot to be all 1.

Zone Definition

In this demo, we may define 1, 2, or 3 cuboids per zone. A detected point in the point cloud residing inside any of these cuboids will be included in the calculations for determining occupancy. If your zone is a cargo area, you may need to define only one cuboid to approximate the space. An example of three cuboids can be used as follows

• Cuboid 1: Head and chest area
• Cuboid 2: Lap area
• Cuboid 3: leg and Foot well

which is illustrated here:

Zones may overlap or be disjoint. Cuboids are defined by the cuboidDef CLI command in a chirp configuration file. All the cuboid definition follows car coordination

Example of 5 Zoness CLI Definitions

% zone 0 (driver) cuboids
cuboidDef <zoneInd> <coboidInd> <xMIn> <xMax> <yMin> <yMax> <zMin> <zMax>
cuboidDef 0 0   0.15 0.75    0.6 1.1   0.5  1.1
cuboidDef 0 1   0.15 0.75    0.3 0.9   0.5  0.85
cuboidDef 0 2   0.15 0.75   -0.4 0.7  -0.1  0.5
% zone 1 (front passenger) cuboids
cuboidDef 1 0  -0.75 -0.15   0.6 1.1    0.5  1.1
cuboidDef 1 1  -0.75 -0.15   0.3 0.9    0.5  0.85
cuboidDef 1 2  -0.75 -0.15  -0.4 0.7   -0.1  0.5
% zone 2 (2nd row driver side) cuboids
cuboidDef 2 0   0.28 0.75    1.5 2.0    0.8  1.1
cuboidDef 2 1   0.28 0.75    1.25 2.0    0.3  0.8
cuboidDef 2 2   0.28 0.75    0.95 1.4   -0.1  0.4
% zone 3 (2nd row middle) cuboids
cuboidDef 3 0  -0.2 0.2     1.5 2.0     0.8  1.1
cuboidDef 3 1  -0.2 0.2     1.4 2.0     0.35  0.8
cuboidDef 3 2  -0.2 0.2     1.25 1.4     0  0.4
% zone 4 (2nd row passenger side) cuboids
cuboidDef 4 0  -0.75 -0.28   1.5 2.0    0.8  1.1
cuboidDef 4 1  -0.75 -0.28   1.25 2.0   0.3  0.8
cuboidDef 4 2  -0.75 -0.28   0.95 1.4  -0.1  0.4

Limitations in Zone Definition

• The first row supports two seats
• The second row supports three seats. Users can use zone definition to disable some of these seats
• New zones can be defined, such as the footwell, intruder area, third row, etc. The limitations are
  • Maimum number of zones: 8
  • Maximum number of cuboids per zone: 3

Example NULL Zone Definition To define a NULL zone, simply define a zone with a single cuboid, and set all parameters to zero. This makes it impossible for the zone assignment function to map any points to the zone, and the zone will stay silent.

% zone 3 (NULL zone)
cuboidDef 3 0   0.0 0.0  0.0 0.0  0.0 0.0

Demo Configuration File Parameters

The configuration file contains three type of CLI commands. We will focus on the ones that are specific to this demo.

span>

Tuning

The key items to focus on for tuning are as follows:

• Proper mounting - The classifer model has been trained with the sensor mounted at the front overhead console position.
       % Sensor position and tilt angles
     % <xOffset> <yOffset> <zOffset> <azimuthTilt> <elevationTilt>
     sensorPosition 0 0.8 1.15 0 -60
  The performance will be similar to our findings when the sensor being mounted is in a similar position and a similar tilting angle even in different cars. For other positions like center mount, side mount, and back mount, the performance has not been tested at this time.

• Accurate zone definition - Ensuring the zones align with the physical space of the vehicle. Please try to exclude the seat itself in the zone definition. For simplicity, we would suggest to start with the default zone definition in the example configuration file,

  • x-direction: use center of the car in width as zero reference.
  • z-direction: Use car floor as zero reference.
  • y-direction: use the ending zone depth of the first row (zone 0/1, cuboid 1) as reference (yRef = 0.9m) as shown in the figure below

    
    cuboidDef 0 1 0.15 0.75 0.3 0.9 0.5 0.85

  Following the steps below to adjust the zone definition for a different car.

  1. Measure the sensor depth relative to the yRef to calculate yOffset = yRef - y1. In our example, we have y1 = 0.15m
     

         sensorPosition 0 0.75 1.15 0 -60
     

  2. Measure the height of the EVM antenna relative to the car floor. In our example, we have zOffset = 1.15m
     

         sensorPosition 0 0.8 1.15 0 -60
     

  3. Measure the distance between the yRef and the end of 2nd-row seat. In our example, we have y2 = 1.1m
     Measure the distance between the yRef and the begining of the 2nd-row seat. In our example, we have y3 = 0.45m
     Then adjust the depth of 2nd row zone accordingly.

          % zone 2 (2nd row driver side) cuboids
         cuboidDef 2 0 0.2 0.75 1.5 2.0 0.8 1.1
         cuboidDef 2 1 0.2 0.75 1.25 2.0 0.3 0.8
         cuboidDef 2 2 0.2 0.75 0.95 1.35 -0.1 0.4
         % zone 3 (2nd row middle) cuboids
         cuboidDef 3 0 -0.12 0.12 1.5 2.0 0.8 1.1
         cuboidDef 3 1 -0.12 0.12 1.25 2.0 0.35 0.8
         cuboidDef 3 2 -0.12 0.12 0.95 1.35 0 0.4
         % zone 4 (2nd row passenger side) cuboids`
         cuboidDef 4 0 -0.75 -0.2 1.5 2.0 0.8 1.1
         cuboidDef 4 1 -0.75 -0.2 1.25 2.0 0.3 0.8
         cuboidDef 4 2 -0.75 -0.2 0.95 1.35 -0.1 0.4
     

• Classifier is trained with the following feature extraction parameters, and modification of these paramters can result in non-optimal solutions.
      % SBR feature extraction parameters
    featExtrCfg 200 20 1 1 0.4 10

• Matching Antenna Pattern - Make sure antGeometry0, antGeometry1 and antPhaseRot settings in the configuration file matches with the antenna pattern on the device/EVM.

• Point Cloud Tuning - Point cloud quality is important to the classifier. Users can adjust CFAR detection threshold “K0 cross range” and “K0 cross angle” in dynamicRACfarCfg to achieve richer point cloud. Note that Lower CFAR detection threshold “K0” results in richer point cloud but higher false detection ratio. Users can also lower RefRangeBinIdx in dynamicRACfarCfg to get richer point cloud in closer range bins. Users can also lower “Linear SNR threshold” in dynamic2DAngleCfg to lower the point cloud extension threshold to include more neighboring points next to the detected peak.

Developer’s Guide

1. Import Lab Project

To import the source code into your CCS 20 workspace, a CCS project is provided in the lab at the path given below.

• Start CCS and configure your workspace as desired.

• Import the projects specified in the following folder to CCS. <Radar_Toolbox_LOCATION>\source\ti\examples\Automotive_InCabin_Security_and_Safety\AWRL6844_InCabin_Demos\src\

• Verify that the import occurred without error by checking that demo_in_cabin_sensing_6844_dss, demo_in_cabin_sensing_6844_mss and demo_in_cabin_sensing_6844_system show up in the project explorer

⚠️ Error during Import to IDE
If an error occurs, check that the software dependencies listed above have been installed. Errors will occur if necessary files are not installed in the correct location for importing.

2. Build the Lab through CCS 20

1. Select the demo_in_cabin_sensing_6844_dss project in the Project Explorer so that it is highlighted. Right click on the project and select Rebuild Project. The project will then build.

2. Select the demo_in_cabin_sensing_6844_mss project in the Project Explorer so that it is highlighted. Right click on the project and select Rebuild Project. The project will then build.

3. Select the demo_in_cabin_sensing_6844_system project in the Project Explorer so that it is highlighted. Right click on the project and select Rebuild Project. The project will then build.

Note
As mentioned in the Quickstart section, pre-built binary is provided in the folder <Radar_Toolbox_LOCATION>\source\ti\examples\Automotive_InCabin_Security_and_Safety\AWRL6844_InCabin_Demos\prebuilt_binaries

Note
Images can be built in either debug mode or release mode. Debug mode will face some performance limitations due to the un-optimized DSP implementation in debug mode

Known Issues and Limitations

1. When the number of range bins used for Dynamic CFAR is less than 29, false detections begin to occur.
   • To avoid this, keep the CFAR leftSkipSize and rightSkipSize small enough to ensure that at least 29 range bins are used.

Need More Help?

• Search for your issue or post a new question on the mmWave E2E forums
• See the SDK for more documentation on various algorithms used in this demo. Start at <MMWAVE_LSDK6_INSTALL_DIR>\docs\mmwave_sdk_module_documentation.html