Mity CPU Platforms » MityDSP-L138 (ARM9 Based Platforms)

Hello,
I am trying to compute the FFT from 2 signals at once. So far I managed to compute the FFTs of my signals basing on the SigProcFFT.cpp file with the commented example. However, for efficienty purposes(I am dealing with real-time system and computation time of FFT is critical in my project), I would like to compute the FFT using the Efficient Computation of the DFT of Two Real Sequences, just like it is done in commented example in SigProcTIDspSupport.cpp. As a test, I was trying to compute the FFT for tho sine signals, defined in the following lines of code:

//signals eclarations
   static float sin21[2048];
   static float sin42[2048];
   float omega21 = 2.0*( 3.14159265358979323)/64;
   float omega42 = 2.0*omega21;
      for(int i = 0; i < 2048; i++)
    {
         sin21[i] = sin(omega21*i);
          sin42[i] = sin(omega42*i);
       }

Subsequently, I implemented the FFT computation basing on the example code in SigProcTIDspSupport.cpp, except I did not want to use the Hanning window and I already had float-type signals, so I ommited the convervion to float and the multiplication of signal. The computation in my code looked like this:

float* mpTwiddles    = new float[FFT_SIZE*2+8];   // TI FFT twiddle factors
       float* mpWorkBuff    = new float[FFT_SIZE*2+8];   // working buffer
       float* mpFFTA        = new float[FFT_SIZE*2];     // FFT output for A
       float* mpFFTB        = new float[FFT_SIZE*2];     // FFT output for B
       short* mpBitReverseIndex = new short[tcSigProcTIDspSupport::bitrev_index_length(FFT_SIZE)];

//allignment of data structures for twiddle factor and working buffer
       if ((unsigned int)mpTwiddles & 7)
{
          mpTwiddles = (float*)(((unsigned int)mpTwiddles+8)&~7);
       }

if ((unsigned int)mpWorkBuff & 7)
{
          mpWorkBuff = (float*)(((unsigned int)mpWorkBuff+8)&~7);
       }
// I am not quite sure about second parameter of the following method- In the example there was the LOW_RES_FFT_SIZE parameter, but I did not find its definition
//so I set the value to the size of mpBitReverseIndex array
       tcSigProcTIDspSupport::compute_bitrev_index(mpBitReverseIndex, tcSigProcTIDspSupport::bitrev_index_length(FFT_SIZE));

// Generating the twiddle factors
       tcSigProcTIDspSupport::gen_w_r2(mpTwiddles,FFT_SIZE);
       tcSigProcTIDspSupport::bit_rev(mpTwiddles,FFT_SIZE>>1);

// move the signals into working buffer- data signals are put in odd and even indexes of working buffer
       for(int i=0;i!=FFT_SIZE;i+=2)
{
           mpWorkBuff[i]=sin21[i];
           mpWorkBuff[i+1]=sin42[i];
       }

// do the TI FFT
       DSPF_sp_cfftr2_dit(mpWorkBuff,mpTwiddles,FFT_SIZE);

// fix the output order to be sorted normally
       DSPF_sp_bitrev_cplx((double*)mpWorkBuff, mpBitReverseIndex, FFT_SIZE);

// extract the two sets of data
       tcSigProcTIDspSupport::ExtractDualRealFFT(mpWorkBuff, mpFFTA, mpFFTB,
                                                 FFT_SIZE, FFT_SIZE);

I anticipated to get as a results the FFT spectra in mpFFTA and mpFFTB with single positive "spike" (one value in output array much larger than others- quite visible if you view the array as an expression in Debug perspective in CCSv5 and then right-click view it as a graph), just like it was when I implemented the transform using the SigProcFFT.cpp example. However, the resulting spectra were incorrect. Do you know, what could be the reason for this ?If you have any suggestions, any help is invariably welcome.
Thank you in advance
Rafal Krawczyk