AWRL6432: CFAR DPU is showing detections but AOA2DPROC is having zero detections

Girish Veerendra Medisetty

Part Number: AWRL6432

Hi team,

We are using 5.3L SDK. As per customer requirement, changed aoa2proc code from BPM mode to TDM. We verified point cloud with reflector and it is showing proper angle. But for some testing case like adult still, we are seeing detection points after CFAR, but while doing AOA, AOACFAR is giving zero detections for those points.

We debugged, futher by extracting Doppler FFT and Azimuth and elevation heat map. It looks like below:

We have not changed AOACFAR HWA configuration.

Please let us know, why AOACFAR is always giving zero point? also we believe, doppler FFT has abrupt peaks, why so?

Regards

M Girish Veerendra

over 2 years ago

0 Clinton Rodgers over 2 years ago

TI__Genius 13235 points

Hi Girish,

Please allow a day or so to provide more detailed feedback. In the meantime, could you provide any additional information on how you are validating that what you are observing is not accurate?

Thanks,

Clinton

0 Girish Veerendra Medisetty over 2 years ago in reply to Clinton Rodgers

Prodigy 230 points

Hi Clinton
We extracted radar-cube and performed doppler fft and aoa for range index = 6 and doppler index = 0 in matlab. these are the respective plots:

Why there are are uprupt peaks in doppler FFT from board and then why there graph is not smooth from board.
Please let me know if you require any additional information?

0 Clinton Rodgers over 2 years ago in reply to Girish Veerendra Medisetty

TI__Genius 13235 points

Hi Girish,

Are you applying rectangular windowing to the doppler FFT? Do you have an idea why the MALTAB processed FFT goes up to 600, but the same data processed on the device only goes up to 25? Gain may need to be applied to make it less noisy.

Thanks,

Clinton

0 Girish Veerendra Medisetty over 2 years ago in reply to Clinton Rodgers

Prodigy 230 points

Hi Clinton,

Clinton Rodgers said:
Are you applying rectangular windowing to the doppler FFT?

No we are performing Hanning windowing for doppler FFT as per default 5.3L SDK.

Clinton Rodgers said:
Do you have an idea why the MALTAB processed FFT goes up to 600, but the same data processed on the device only goes up to 25?

We just applied windowing and performed FFT in MATLAB. we did not did any thing more than this. MATLAB is performing floating point FFT and HWA is performing fixed point FFT. we cannot expect the values should be same, but graph should be similar.

Clinton Rodgers said:
Gain may need to be applied to make it less noisy.

Do you mean RXgain??

Regards
M Girish Veerendra

0 Clinton Rodgers over 2 years ago in reply to Girish Veerendra Medisetty

TI__Genius 13235 points

Hi Girish,

If you had been using the rectangular windowing, the way it is defined it scales the output, so there may have been a way to change the code to reduce the impact of noise on the results. Since you are using hanning, which should perform better than the rectangular, this same change does not apply.

As I mentioned in the other thread (linked below for reference), could you provide the radar cube and MATLAB script for us to analyze? If you could also try running on another 6432 EVM and see how the results vary, that would be good. It would also be interesting to see how this compares to results from the 6443 EVM when tested in the same way.

https://e2e.ti.com/support/sensors-group/sensors/f/sensors-forum/1269002/awrl6432-detmatrix-has-all-doppler-spread-in-5-3l-sdk

Thanks,

Clinton