[FFmpeg-devel] ac3 imdct question

[Arpi]

Hi,

Could someone DSP guru please explain me, what ff_fft_calc exactly
calculates in ff_imdct_half_c() of mdct.c ?
I've tried to replace it (for experimenting with pitch correction...)
with the 'standard' inverse discrete fourier transform but it does
not work, the sound (ac3 music sample file) gets heavily distorted :(
here's my modification:

--- mdct.c.ORIG	2008-09-08 15:37:12.000000000 +0200
+++ mdct.c	2008-10-13 15:10:45.000000000 +0200
@@ -122,6 +122,7 @@
     const FFTSample *tcos = s->tcos;
     const FFTSample *tsin = s->tsin;
     const FFTSample *in1, *in2;
+    FFTComplex buf[1024];
     FFTComplex *z = (FFTComplex *)output;
 
     n = 1 << s->nbits;
@@ -134,11 +135,29 @@
     in2 = input + n2 - 1;
     for(k = 0; k < n4; k++) {
         j=revtab[k];
-        CMUL(z[j].re, z[j].im, *in2, *in1, tcos[k], tsin[k]);
+//        CMUL(z[j].re, zi[j].im, *in2, *in1, tcos[k], tsin[k]);
+        CMUL(buf[k].re, buf[k].im, *in2, *in1, tcos[k], tsin[k]);
         in1 += 2;
         in2 -= 2;
     }
-    ff_fft_calc(&s->fft, z);
+
+#if 1
+#undef fprintf
+    fprintf(stderr,"ff_imdct_half_c: n4=%d\n",n4);
+    // n4==128
+    for(k=0;k<n4;k++){
+	float s=k*(2*M_PI/n4);
+	double x=0,y=0;
+	for(j=0;j<n4;j++){
+	    x+=buf[j].re*cos(j*s);
+	    y+=buf[j].im*sin(j*s);
+	}
+	z[k].re=x;
+	z[k].im=y;
+    }
+//    ff_fft_calc(&s->fft, buf);
+#endif
+
 
     /* post rotation + reordering */
     output += n4;


thanks,

A'rpi