delay.c

this the source code of addio compression standard CELP. Also, it is optimizied for the execution sp

#define NFRAC 5
#define TRUE 1
#define FALSE 0
#define M1 -4
#define M2  3

/* five fractional delays calculated over an 8 point interpolation 	*/
/* (-4 to 3)								*/

static float frac[NFRAC] = {0.25, 0.33333333, 0.5, 0.66666667, 0.75};
static int twelfths[NFRAC] = {3, 4, 6, 8, 9};

/**************************************************************************
*
* NAME
*		sinc
*
* FUNCTION
*	
*
* SYNOPSIS
*		subroutine sinc(arg)
*
*   formal 
*                       data    I/O
*       name            type    type    function
*       -------------------------------------------------------------------
*	arg		float	i	
*	y(n)		float	func	
*
***************************************************************************
*
* DESCRIPTION
*
*	This interpolating (sinc) function is Hamming windowed to bandlimit
*	the signal to reduce aliasing.
*
***************************************************************************
*
* CALLED BY
*
*	pitchvq  psearch
*
* CALLS
*
*
***************************************************************************/
#include <math.h>
#include <stdio.h>
float
sinc(arg)
float arg;
{
  float pi;

  pi = 4.0 * atan(1.0);
  if (arg == 0.0)
    return(1.0);
  else
    return(sin(pi * arg) / (pi * arg));
} 

/**************************************************************************
*
* NAME
*		qd
*
* FUNCTION
*	
*
* SYNOPSIS
*		subroutine qd(d)
*
*   formal 
*                       data    I/O
*       name            type    type    function
*       -------------------------------------------------------------------
*	d		float	i
*	qd(d)		int	o	quantize d function	
*
***************************************************************************
*
* DESCRIPTION
*
*	Quantize d function
*
***************************************************************************
*
* CALLED BY
*
*	delay
*
* CALLS
*
*
***************************************************************************/
int 
qd(d)
float d;
{

  int i, index, ok;

/*  ok = FALSE;   */
  for (i = 0; i < NFRAC; i++)
  {
    if (fabs(d - frac[i]) < 1.e-2)
    {
      index = i;
/*      ok = TRUE;         */
    }
  }
/*
  if (!ok)
  {
    fprintf(stderr, "delay: Invalid pitch delay = %f\n", d);
    exit(1);
  }
*/
  return(index);
}
  
/**************************************************************************
*
* NAME
*		delay
*
* FUNCTION
*		Time delay a bandlimited signal
*		using point-by-point recursion
*
* SYNOPSIS
*		subroutine delay(x, start, n, d, m, y)
*
*   formal 
*                       data    I/O
*       name            type    type    function
*       -------------------------------------------------------------------
*	x[n]		float	i	signal input (output in last 60)
*	start		int	i	beginning of output sequence
*	n		int	i	length of input sequence
*	d		float	i	fractional pitch
*	m		int	i	integer pitch
*	y[n]		float	o	delayed input signal
*
***************************************************************************
*
* DESCRIPTION
*
*	Subroutine to time delay a bandlimited signal by resampling the
*	reconstructed data (aka sinc interpolation).  The well known
*	reconstruction formula is:
*
*              |    M2      sin[pi(t-nT)/T]    M2
*   y(n) = X(t)| = SUM x(n) --------------- = SUM x(n) sinc[(t-nT)/T]
*              |   n=M1         pi(t-nT)/T    n=M1
*              |t=n+d
*
*	The interpolating (sinc) function is Hamming windowed to bandlimit
*	the signal to reduce aliasing.
*
*	Multiple simultaneous time delays may be efficiently calculated
*	by polyphase filtering.  Polyphase filters decimate the unused
*	filter coefficients.  See Chapter 6 in C&R for details. 
*
***************************************************************************
*
* CALLED BY
*
*	pitchvq	  psearch
*
* CALLS
*
*	ham
*
***************************************************************************
*
* REFERENCES
*
*	Crochiere & Rabiner, Multirate Digital Signal Processing,
*	P-H 1983, Chapters 2 and 6.
*
*
*	Kroon & Atal, "Pitch Predictors with High Temporal Resolution,"
*	ICASSP '90, S12.6
*
**************************************************************************/

#define SIZE (M2 - M1 + 1)
delay(x, start, n, d, m, y)
float x[], d, y[];
int m, n, start;
{
  static float wsinc[SIZE][NFRAC], hwin[12*SIZE+1];
  static int first = TRUE;
  register int i, j, k, index;
  register int count1;
  register float fcount1;

  /* Generate Hamming windowed sinc interpolating function for each	*/
  /* allowable fraction.  The interpolating functions are stored in  	*/
  /* time reverse order (i.e., delay appears as advance) to align    	*/
  /* with the adaptive code book v0 array.  The interp filters are:  	*/
  /* 		wsinc[.,0]	frac = 1/4 (3/12)		     	*/
  /* 		wsinc[.,1]	frac = 1/3 (4/12)		     	*/
  /* 		.		.				     	*/
  /* 		wsinc[.,4]	frac = 3/4 (9/12)		     	*/


  if (first)
  {
    ham(hwin, 12*SIZE+1);
    for (i = 0; i < NFRAC; i++)
      for (fcount1 = M1+frac[i], count1 = twelfths[i], j = 0; j < SIZE; j++)
      {
          wsinc[j][i] = sinc(fcount1) * hwin[count1];
          fcount1++;
          count1 += 12;
      }

    first = FALSE;
  }

  index = qd(d);

  /* *Resample:							 */
   for (count1 = start-1, i = 0; i < n; i++)
   {
     x[count1] = 0.0;
     for (k = M1-m+count1, j = 0; j < SIZE; j++)
     {
         x[count1] += x[k] * wsinc[j][index];
         k++;
     }
     count1++;
   }


 for (count1 = start - 1, i = 0; i < n; i++)
 {
     y[i] = x[count1];
     x[count1] = 0.0;
     count1++;

 }

}