MMWCAS-DSP-EVM: eve FFT

Hi Sara,

    I am assuming that when you say signals you mean signals from different antennas. EVE applet for FFT can work with single antenna also. From where did you got the information that it cannot work with less than 4 antennas?

Regards,

Anshu

Hi Anshu, 

I mean signals from same antennas, but different chirps...

if I run the usecase with CHAINS_CASCADE_RADAR_RADAR_HEIGHT  = 4 (defined in chains_common_cascade_ar12xx_config_mimo.c) everything is ok, but if I set it to 1 I get some compilation error. I took a look at the source code of the EVE-fft (fft_alg.c) and I found FFT_TI_MIN_NUM_LINES = 4 ... So I thought that it means that each fft processing must elaborate 4 signals corresponding to 4 different chirps...

Sara

HI Sara,

   Do you mean to say that your Doppler dimension is 4?

Regards,

Anshu

Yes, with Doppler dimension = 4 everything is ok, but I would like to have Doppler dimension = 1

Sara

HI Sara,

   Your intention is to process just one line for range FFT? We process multiple lines together to utilize the SIMD processing power of EVE. Mostly thats the reason there is a restriction. But I would like to understand about how much data your are planning to process to make sure that I am not misunderstanding you.Can you tell the following :

• Range dimension for FFT
• Doppler dimension as understand is 1
• Number of antennas in the system

Regards,

Anshu

Hi, 

yes I want to process only one line. 

Range FFT on 1024 points (a lot of zero-padding)

Doppler dimension = 1

Number of antennas : 4 rx for each sensor, 10 transmitters in total (but I might reduce it to 7)

The "problem" is that my frame rate should be 5 ms, and with Doppler dimension = 4 the processing time is about 6 ms ...

Sara

Hi Sara,

   Sorry for delay in response. If I understand correctly your data for range FFT processing is of dimension 1024 x 4 x 10. As mentioned in our data sheet for 1024 point FFT we need ~2.5 cycles per point. As per this total cycles required for this processing should be 1024 * 4 * 10 * 2.5. Assuming EVE is running at 500MHz, it will take only 0.2 ms to process the above amount of data. I am not sure why you are not able to fit this in 5ms? Am I missing something here?

Regards,
Anshu

Hi, 

The range FFT processing is 1024(numer of samples) x 4 (number of doppler bins) x 4 (receivers) x 10 (transmitters) , which makes around 0.8 ms. In your calculation of 0.2 ms you are considering only 1 receiver but in my understanding every EVE processes the signals coming from all the receivers of a sensor, is that right?

I measured with a timer the time elapsed between the start and stop of the 1D-fft process (i.e. beginning and end of function "AlgorithmFxn_RadarFftProcess")  and I visualized the measure on the command window. What I see is that it takes 1.2 ms more or less to process all the signals, and then the synch link takes almost the same time to synchronize all the buffers. I guess that by reducing the number of doppler bins also the synch link will become faster....

Sara 

Hi Sara,

   The calculation which i did was for 1 doppler bin, so it was 1024 ( number of samples) x 1 ( doppler bin) x 4 ( receivers) x 10 ( transmitters). From this thread I got the impression that with 4 doppler you are not able to fit in budget of 5 ms and thats why you want to use 1 doppler bin. I am referring to the following comment : "The "problem" is that my frame rate should be 5 ms, and with Doppler dimension = 4 the processing time". But as you mentioned even with 4 doppler bins you are taking 1.2ms which is well within 5ms budget. Am I missing something here?

Regards,

Anshu 

Hi, 

no you've understood correctly. But in my 5ms budget I do a lot of other stuff, not only FFTs...so I am trying to reduce wasted time wherever I can.

Sara

Hi Sara,

   Once range FFT is done what is the next stage of processing? 

Regards,

Anshu

Sara,

   I was looking at the code and I think for range FFT minimum doppler dimension restriction probably can be removed. Are you using FFT's standalone test bench for your benchmarking?

Regards,

Anshu

Hi, 

no I am using the object-detection usecase

Sara

Like in the object detection usecase, I calculate the energy sum and then apply the cfar. I DO NOT calculated the Doppler FFT.

Hi Sara,

   Can you tell me the EVE release version which you are using. I want to suggest a change which probably should solve the problem.

Regards,

ANshu

Hi,

is  "eve_sw_01_20_00_00" (I am working with PROCESSOR_SDK_RADAR_03_07_00_00)

Sara

Hi Sara,

    Can you make the value of following macro to 1 and rebuild EVE FFT Library to see if  it solves your problem. 

fft_alg.c file :

#define FFT_TI_MIN_NUM_LINES (1U)

   Regards,

Anshu