• State Resolved
  • Locked Locked
  • Replies 5 replies
  • Subscribers 30 subscribers
  • Views 533 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

AWR1843AOP: Multi object beamforming

Shoko Sakaguchi
Intellectual 560 points
Part Number: AWR1843AOP

Tool/software:

Hello, 

I have a question about Multi object beamforming.

I'm using the mmWave demo of mmwave_sdk_03_06_02_00-LTS on AWR1843AOP. 

I can choose between two settings for "peak selection in 2D azimuth/elevation FFT output"; 

 - Maximum peak search using HWA statistics block

 - Maximum peak search using HWA CFAR

How should I use the above two settings differently? 

Please let me know the examples of suitable applications for each setting? 

Please advise. 

Thanks, 

Shoko 

  • 0
    Intellectual 560 points

    Hello, 

    When can you give me an answer? 

    Thanks, 

    Shoko 

  • 0 in reply to Shoko Sakaguchi
    TI__Mastermind 20605 points

    Hi Shoko-san,

    Please give me a week's time to take a look at this and get back to you.

    Regards,

    Kaushik

  • 0 in reply to Kaushik Gowda
    TI__Mastermind 20605 points

    Hi Shoko-san,

    Thank you for your patience. I have looked at the code here and my comments are as follows:

    1. The configurable options I found for the MultiObjBeamforming operation are as follows:
      2. Here, I'm not sure if I am able to find the peak search method selection as you mentioned between HWA stats block or CFAR block.
    2. In both he aoaprocdsp.c and aoaprochwa.c, we see that the second peak search algorithm is executed on the DSP itself (one uses a helper function from mmWavelib). Again, both don't utilize the HWA for this second peak search and neither of these have a config along the peak detection method.
    3. If you are just asking for difference in peak estimation between stats block vs CFAR block on the HWA, there should be no difference in the actual peak estimated.
      1. For the CFAR based peak estimation using OS-CFAR, there might be limitations on window size or threshold configurability. Identifying the Kth largest can ideally be done in only one paramset.
      2. With the stats block, you should be able to operate on any size of data. For Nth largest, you might need to run it N-times through the stats block to identify those peaks. Given these considerations, it is best to look at the use case to decide for the adoption of either of these methods.

    Kindly let me know if I have missed anything on my end,

    Regards,

    Kaushik

  • 0 in reply to Kaushik Gowda
    Intellectual 560 points

    Hello Kaushik, 

    Thank you for your information. Could I ask you some more questions? 

    > Here, I'm not sure if I am able to find the peak search method selection as you mentioned between HWA stats block or CFAR block.

    --> I have found your document about multi-object peak search of 2D(azimuth/elevation) FFT caluculation. (Please see the following image.) 

      ti/mmwave_sdk_03_06_02_00-LTS/packages/ti/datapath/dpc/dpu/aoa2dproc/docs/doxygen/html/index.html

     The MultiObjBeamforming in mmWave_sdk_user_guide (mmWave demo) can switch HWA statistics block or HWA CFAR, right? 

    > Again, both don't utilize the HWA for this second peak search and neither of these have a config along the peak detection method.

    --> I’m not sure what you mean... Please let me know what's your point...? 

    Please advise. 

    Thanks, 

    Shoko 

  • +1 in reply to Shoko Sakaguchi
    TI__Mastermind 20605 points

    Hi Shoko-san,

    Thank you for the clarification. I was looking at the aoaproc DPU as opposed to the aoa2dproc DPU, but the mechanism should hold.

    From what I understand, it looks like after the azimuth and elevation FFT are performed in the aoa2dproc, you can either choose the CFAR based peak estimation, or the stats block based peak estimation as part of the 'multiobjbeamforming' operation. Here are some of the key differences:

    • When you use the CFAR based peak detection from the azimuth FFT, the number of peaks you obtain will be greater than 1 per FFT sequence. In the case of stats block, it will always be 1.
    • This method yields a greater number of detections in the case of multi object beamforming being enabled.
    • It can increase the total number of detections stemming from say an object with a very large area and instead of it being represented as a single point object in the point cloud, there will be multiple points mapped to the same object.

    Based on these details, it should be able to choose the right method for your use case.

    Regards,

    Kaushik