HERCULES-DSPLIB: FFT of real valued data in time domain

Dhruva Krishnadas

Part Number: HERCULES-DSPLIB
Other Parts Discussed in Thread: AM6546

We are using the Hercules DSPLIB v1.0.0 to perform some signal processing in our application. To simply touch base with the Library we simply tried to perform an 128-point FFT of a DC signal to expect an impulse response.

Code excerpt below:

float32_t test[128] = {1.0};

arm_rfft_instance_f32 rfft_instance;
arm_rfft_instance_f32* rfft_instance_ptr = (arm_rfft_instance_f32*) &rfft_instance;

arm_cfft_radix4_instance_f32 cfft_instance;
arm_cfft_radix4_instance_f32* cfft_instance_ptr = (arm_cfft_radix4_instance_f32*) &cfft_instance;

arm_rfft_init_f32(rfft_instance_ptr,cfft_instance_ptr,128,0,1);

float32_t out[256] = { 0.0 };
arm_rfft_f32(rfft_instance_ptr,test,out);

However, when I printed the result of "out" I got a stream of 1, 0, 1, 0 .... instead of an impulse of magnitude of 128.
real[0]=1, real[1]=1, real[2]=1 ....

complex[0]=0,complex[1]=0,complex[2]=0 ...

Am I using this function incorrectly ?

over 4 years ago

0 QJ Wang over 4 years ago

TI__Guru**** 191946 points

Hello Dhruva,

BTW, which device are you using?

The format of the output of arm_rfft_f32(...) is: { real[0], imag[0], real[1], imag[1], real[2], imag[2] ... real[(N/2)-1], imag[(N/2)-1 }

N/2 are the real parts and N/2 are imaginary parts interleaved, in total N/2 complex pairs.

0 QJ Wang over 4 years ago in reply to QJ Wang

TI__Guru**** 191946 points

Hello Dhruva,

It should be fine to use the code on AM65x device.

0 Dhruva Krishnadas over 4 years ago in reply to QJ Wang

Intellectual 410 points

Hi Wang,

Yes. It's just that I got a stream of alternate 1s and 0s as output. The FFT of a DC signal should have given me a purely-real impulse response.
I was expecting only real[0], to have a value and everything else to be zero.

Expected output was [128,0,0,0,0,.....]

but what I got was [1,0,1,0,1,0..........]

0 Dhruva Krishnadas over 4 years ago in reply to QJ Wang

Intellectual 410 points

I am using it on an AM6546. We compiled the HerculesLIB for the R5F and got things going.

0 Brad Griffis over 4 years ago

TI__Guru*** 125430 points

Dhruva Krishnadas said:
float32_t test[128] = {1.0};

I believe this only initializes the first element to 1.0. All others will be 0.0. Have you double checked in the debugger?

0 Dhruva Krishnadas over 4 years ago in reply to Brad Griffis

Intellectual 410 points

Thanks Brad. That did the trick.

Arm-based microcontrollers

Arm-based microcontrollers forum

HERCULES-DSPLIB: FFT of real valued data in time domain