• State Resolved
  • Replies 3 replies
  • Subscribers 32 subscribers
  • Views 43 views
  • Users 0 members are here
Support feedback
Options
Options
Related

Original question:

IWRL6432WMOD: EVK Board Uniflash Firmware Download Stopping Issue

IWRL6432WMOD: Flashing an appimage

Marvn Arcangel
Genius 16640 points

Part Number: IWRL6432WMOD
Other Parts Discussed in Thread: IWRL6432, IWRL6432W, IWRL6432AOP, UNIFLASH

Hi Experts,

Can you please clarify if the IWRL6432WMOD is programable? if yes, how can I flash my own firmware?

For example, in the radar toolbox, I saw some examples regarding human vs. non-human classification and motion and presence detection, but they don't have images to upload as firmware.

Regards,

MArvin

  • +1
    TI__Intellectual 920 points

    Hi Marvin,

    The IWRL6432WMOD is only configurable; not programmable. You can adjust chirp parameters, data output methods, etc., but you cannot write custom firmware. This allows the module to be certified.

    The Radar ToolBox does not have any demos relating to the IWRL6432WMOD - the examples there apply to IWRL6432, IWRL6432AOP, and IWRL6432W (this is the WCSP package, not WMOD module).

    Best,

    Josh

  • 0 in reply to Josh Deardorff
    TI__Genius 16640 points

    Hi Josh,

    I have a follow up question.

    The customer tried flashing the Motion and Presence firmware onto the WMOD. Uniflash reports "SUCCESS!! META_IMAGE1 downloaded successfully to SFLASH", but after switching to functional mode the device outputs nothing at all on the UART.

    I checked the mmWaveULink ICD (1.1.0.5). The Sensor Information Get API returns frame count, up to 64 detected points with X/Y/Z/velocity/SNR/noise, and a 7-bit zone occupancy byte. I could not find any API that returns micro-Doppler spectrum, micro-Doppler features, tracker output, or classifier output.

    1. Is there any way to get micro-Doppler data, tracker output, or classifier output from the IWRL6432WMOD through uDFP? If not today, is it planned for a future uDFP release?
    2. If no, is human vs non-human classification possible at all on the WMOD, using only the point cloud and zone occupancy data that the ICD exposes? Is there any application note or reference for doing this on the host side?
    3. Is the trained model or the training dataset used by the Human vs Non-Human demo available in any form (weights, ONNX, dataset)
    4. If custom firmware on the WMOD is not an option, what is the recommended migration path? The uDFP release notes describe the WMOD as an "IWRL6432 WCSP MODULE", and the Radar Toolbox has a WCSP project and a WCSP prebuilt binary. If I move to the bare IWRL6432W (WCSP) on my own PCB:
      1. would the Motion and Presence demo and its classifier run?
      2. does the human/non-human classifier accuracy depend on the antenna array it was trained against, and would I need to re-derive antGeometryCfg for my own antenna layout?
      3. would the existing trained model still be valid on a different antenna design, or would it have to be retrained?

    Thanks!

    Marvin

  • 0 in reply to Marvn Arcangel
    TI__Intellectual 920 points

    Hi Marvin,

    I believe you may have an earlier testing release of the IWRL6432WMOD that can be flashed. The production device has no capability to be flashed with any firmware - the application is stored in ROM.

    1. No, there is no planned support for micro Doppler data on the WMOD.

    2. Classification is not possible on-chip, since the firmware is closed. A host controller could do extended processing on the point cloud data the device outputs, but I would look into a different IWRL6432 device, since they can do that processing on-chip.

    3. I'll send you a follow up email on the training of the human/non-human classifier.

    4. Yes, the WMOD is built upon the IWRL6432WSCP. The module contains many other components around the WSCP to create a drop in sensing device. The good news is yes, the discrete WCSP can run all demos developed for the IWRL6432 family. It's a fully programmable device. You would need to describe your antenna layout in antGeometryCfg. The other good news is we've seen promising results with the human/non-human classifier on non-AOP boards, so the model could work, but we can't guarantee any specific performance.

    Best,

    Josh