• State TI Thinks Resolved
  • Locked Locked
  • Replies 3 replies
  • Answers 2 answers
  • Subscribers 30 subscribers
  • Views 689 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.

AWR1642: question on number of Doppler bins

Chris Meng
Mastermind 38096 points
Part Number: AWR1642
Other Parts Discussed in Thread: AWR1843

Hi,

In SDK 2.1 for xWR1642, I found there is a limitation for number of Doppler bin. That is ,the number of Doppler bin must be power of 2. MmwDemo_dataPathComputeDerivedConfig() API will check it and if it is not power of 2, the code will exit with assertion.

But in SDK3.1, I found I can set non-power of 2 for number of chirps of one frame for AWR1843 in mmw demo, such as 60 (download the cfg in visualizer). I know current mmw demo for AWR1843 is using HWA for FFT/window, while AWR1642 is using DSP.

What does cause the limitation of number of Doppler bin on AWR1642?

I only found below info in SDK2.1. They don't require power of 2.

/* Multiplicity of 4 due to windowing library function requirement.
               Minimum size of 16 due to DSPLib restriction - FFT size must be bigger than 16.*/

  • 0
    TI__Expert 4720 points
    Hi Chris,

    The DSP (and the HWA) implements the 'radix-2 FFT', which requires that the FFT input (and output) be 'a power of 2'. However, it is possible to 'zero-pad' an FFT input, whereby an input whose length is not 'a power of 2' is made to 'a power of 2' by adding zeros.

    However, in the DSP's doppler dimension processing, additional code needs to be written to do 'zero-padding' of the doppler input. Also, because of 'zero padding' the sizes of the 'doppler FFT output' wouldn't be equal to the 'doppler FFT input', which would require some changes to buffer management.

    In mmwaveSDK 2.1, the above code was not added.

    Regards
    Anil
  • 0 in reply to Anil Mani
    TI__Mastermind 38096 points
    Anil,

    Thanks for your reply.

    Would you pls kindly advise if the zero padding is done in L1/L2? The reason I asked is that I would like to know if it will cost L3 memory as customer is limited on L3.
  • 0 in reply to Chris Meng
    TI__Expert 4720 points

    Hi Meng,

    To clarify, we do not do 'zero-padding' for the doppler FFT in the 2.1 SDK. However, the customer can implement it. If the customer is implementing it, then they can implement it in the L1/L2. L3 wouldn't be affected (except for the detection matrix that is going to be stored there).

    Regards

    Anil