problems about using C64+ DSPlib realize input signal's fft

Hi ouyang,

I hope, you have referred TI resources for FFT.

Please refer below links which will be useful to you.

Wiki Link about FFT Implementation:

http://processors.wiki.ti.com/index.php/FFT_Implementation_With_No_Data_Scaling

TMS320C64x+ DSP Library Programmer’s Reference:

http://www.ti.com/lit/ug/sprueb8b/sprueb8b.pdf

Please get back to us if you have any question.

Thanks.

Hi,

     Thanks for you apply ,i have reviewed relevant pages，i don't want the output data is no data scaling, but  i have a question  the scaling of data to avoid overflow is only used by outputdata,is that ture?

     i have a another question is that the in the  DSP64+lib the example code that define  input data that is summation of Sine wave data of various freqs,now i change the Parameter "#define NUM_SIN_WAVES 1" so the input is a signal sine wave ,but i download the file to DM6437 and  Single-step run the program use the signal time to plot  the prameter of x_16x16 is not a signal sine and it is a Constant ?why it is not a signal sine wave data .

      Attached is a example code7776.fftdsplib.zip

 Thanks a lot.

ouyang

Hi Ouyang,

Thanks for your post.

I have reviewed the code and have addressed my answers below:

Yes, it is true, the scaling of data is to avoid overflow. Scale needs to be applied to input data for 16x32 and 32x32 FFT kernel for their output to match 16x16 FFT since scaling is already there in 16x16 kernel. 

 Please check the function generateInput (numSinWaves) in the fft example code, under which it is mentioned, based on the numSinWave information, the input data array shall be created. If you declare macro "#define NUM_SIN_WAVES 1", there is no point in calculating sinWaveIncFreq & sinWaveMag in the code. Please see the code snippet below:

      /* Calculate the incremental freq for each sin wave

                              sinWaveIncFreq =((float)1.0)*((float)3.142)/(numSinWaves*(float)1.0);

      /* Calculate the magnitude for each sin wave */

                             sinWaveMag = (float)1.0/(numSinWaves * (float)1.0*N);

                            /* Create the input array as sum of the various sin wave data */

                            for (j = 0; j < numSinWaves; j++) {

                                 for (i = 0; i < N; i++) {

                                             x_ref_float[2*i] += sinWaveMag * (float)cos(sinWaveIncFreq*j*i);

                                             x_ref_float[2*i + 1] = (float) 0.0;

                                  }

                           }

As per above code snippet, if your numSinWaves is 1, the "j" loop will run only once for j=0, and it is cos0=1. So, the input data array x_ref_float[ ] will be always constant when you multiply cos0=1 with minimum value of sinWaveMag. This is the reason, your signal is constant and not a sine wave.

According to fft example code, it should be atleast numSinWaves=2 inorder to generate a sine wave signal data.

Thanks & regards,

Sivaraj K

Hi Ouyang,

Thanks for your response.

if it answers your post, please close the thread.

Hi,

   now, i want rebuild the dsplib,but i  find the C64x+ DSPLIB i download  is not the same with the file 'TMS320C64x+ DSP Little-Endian DSP Library Programmer’s Reference' said ,could you send me a C64x+ DSPLIB  file to me  ?

  Thanks.

ouyang 

Hi,

          I generate a sine wave type float, then convert it to Q15 format, call the library function fft16x16,    then converted output data to a float, draw the output waveform I found on the  float isine wave data do fft by usingn MATLAB and  graphics the output  are not the same, why? i know the output is scaled 8 time ,is it ture that the output data y_real_16x16 and y_imag_16x16 Multiplied by 8 that we can make the data is right.

         The second  question is i changed the' .sa 'file and add into the project " include option ' i found the output data is not change before i don't  change it .

this is a sine wave float data as the input 

this is the output 's Amplitude after  fft

This is the outputAmplitude 'after  fft by MATALB

   3252.fftdsplib.zip

         Attatched is my program.

Thanks.

Hi Ouyang,

Thanks for your post.

From next time, if you place a new request, you shall raise it as a new thread sothat, it would be easy for us to track thread status as per E2E forum policies.

But any way, as per you request on the above post, please find the attached zip file for C64x+ DSP library installer.

Thanks & regards,

Sivaraj K

Hi Ouyang,

Could you please raise the above request as a new thread? As per E2E forum policies, we are not supposed to answer multiple request in one thread, since it would be more tedious to track the thread status.

To address your above query, basically fft16x16 DSP library function and fft computation on MATLAB will not match accurately because of the inaccurate float precision truncation. So, the output data from both the case will be obviously different.

I didn't clearly understand your second question, kindly provide more details for the same in the new thread.

Thanks for your understanding.

Regards,

Sivaraj K