Dear MMWAVE-SDK experts,
In AoA DPU, is the 2D FFT the same thing as the 2D FFT in Doppler DPU?
If it is, why not reuse it instead of calculating it again?
Thank you.
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.
Dear MMWAVE-SDK experts,
In AoA DPU, is the 2D FFT the same thing as the 2D FFT in Doppler DPU?
If it is, why not reuse it instead of calculating it again?
Thank you.
Hi,
Which document does this graphic come from? I want to make sure I have the right context for providing a correct answer here.
Best,
Nate
Hi Yuan,
It looks to me like the 2D FFT there is the calculation of the input antenna symbols to use for the 3D azimuth FFT afterwards.
Best,
Nate
Not the "2D angle FFT" at the middle of the signal processing chain, I meant the "2D FFT" at the very beginning of the signal processing chain.
And in the document it says:
"2D FFT is computed, the output of the FFT has numDopplerBins samples of type cmplx32ReIm_t. The CFAR detection list determines which Doppler bin should be stored, the other bins are discarded."
And also the input data is only "1D FFT radar cube" which is before Doppler FFT processing.
Hi Yuan,
The doppler DPU does not store the entire 2D FFT in memory. We only store the 1D FFT in the radar data cube. We need to recompute the 2D doppler FFT for the points detected by the CFAR. That is what the 2D FFT refers to here.
Bets,
Nate
Hi Yuan,
There is not enough memory on the device to hold both the 1D Range FFT and the 2D Range/Velocity FFT. The 1D Range FFT is needed for subsequent angle processing, so it is saved, and the 2D FFT is recomputed for the indices indicated by the CFAR (which will be substantially fewer than the entire radar cube).
Best,
Nate
To my understanding, information flow is 1D-FFT => 2D-FFT => Angle, and Angle processing only needs information from 2D-FFT. And, I am planning on storing 2D-FFT instead of 1D-FFT to reduce computation time. (I understand 2D-FFT is float type and 1D-FFT is int16, so memory size will double, and this is fine.)
Please advise, in particular, what is the information needed for angle processing that is not captured in 2D-FFT and can only be found in 1D-FFT? I will need to exact that information before 1D-FFT get replaced by 2D-FFT.
This may be a hard question, but the answer is very important to me.
Thank you.
Hello Yuan.
Nathan is currently out so Nathan or an expert will get back to you tomorrow.
Sincerely,
Santosh
Hi Yuan,
Let me investigate. I will get back to you by EOD tomorrow.
Best,
Nate
Hi Yuan,
If the range-FFT is 16bit I, 16bit Q, then the Doppler FFT is typically 32bit I and 32 bit Q. So storing back to the radar cube is not possible (memory expansion). That is why we cannot store it in L3.
Best,
Nate
Hi Yuan,
Depending on your implementation it may be advantageous.
Best,
Nate