File indexing completed on 2024-05-19 15:01:44

 /***************************************************************************
     File                 : nsl_conv.c
     Project              : LabPlot
     Description          : NSL discrete convolution functions
     --------------------------------------------------------------------
     Copyright            : (C) 2018 by Stefan Gerlach (stefan.gerlach@uni.kn)
 
  ***************************************************************************/
 
 /***************************************************************************
  *                                                                         *
  *  This program is free software; you can redistribute it and/or modify   *
  *  it under the terms of the GNU General Public License as published by   *
  *  the Free Software Foundation; either version 2 of the License, or      *
  *  (at your option) any later version.                                    *
  *                                                                         *
  *  This program is distributed in the hope that it will be useful,        *
  *  but WITHOUT ANY WARRANTY; without even the implied warranty of         *
  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the          *
  *  GNU General Public License for more details.                           *
  *                                                                         *
  *   You should have received a copy of the GNU General Public License     *
  *   along with this program; if not, write to the Free Software           *
  *   Foundation, Inc., 51 Franklin Street, Fifth Floor,                    *
  *   Boston, MA  02110-1301  USA                                           *
  *                                                                         *
  ***************************************************************************/
 
 #include "nsl_conv.h"
 #include "nsl_common.h"
 #include <gsl/gsl_cblas.h>
 #include <gsl/gsl_fft_halfcomplex.h>
 #ifdef HAVE_FFTW3
 #include <fftw3.h>
 #endif
 #include "backend/nsl/nsl_stats.h"
 
 const char* nsl_conv_direction_name[] = {i18n("forward (convolution)"), i18n("backward (deconvolution)")};
 const char* nsl_conv_type_name[] = {i18n("linear (zero-padded)"), i18n("circular")};
 const char* nsl_conv_method_name[] = {i18n("auto"), i18n("direct"), i18n("FFT")};
 const char* nsl_conv_norm_name[] = {i18n("none"), i18n("sum"), i18n("Euclidean")};
 const char* nsl_conv_wrap_name[] = {i18n("none"), i18n("maximum"), i18n("center (acausal)")};
 const char* nsl_conv_kernel_name[] = {i18n("sliding average"), i18n("triangular smooth"), i18n("pseudo-Gaussian smooth"), i18n("first derivative"), i18n("smooth first derivative"),
     i18n("second derivative"), i18n("third derivative"), i18n("fourth derivative"), i18n("Gaussian"), i18n("Lorentzian") };
 
 int nsl_conv_standard_kernel(double k[], size_t n, nsl_conv_kernel_type type) {
     size_t i;
 
     int validn = 1;
     switch (type) {
     case nsl_conv_kernel_avg:
         for (i = 0; i < n; i++)
             k[i] = 1.;
         break;
     case nsl_conv_kernel_smooth_triangle:
         for (i = 0; i < n/2; i++)
             k[i] = i + 1.;
         for (i = n/2; i < n; i++)
             k[i] = (double)(n - i);
         break;
     case nsl_conv_kernel_smooth_gaussian:
         switch (n) {
         case 5:
             k[0]=1;k[1]=4;k[2]=6;k[3]=4;k[4]=1;
             break;
         case 7:
             k[0]=1;k[1]=4;k[2]=8;k[3]=10;k[4]=8;k[5]=4;k[6]=1;
             break;
         case 9:
             k[0]=1;k[1]=4;k[2]=9;k[3]=14;k[4]=17;k[5]=14;k[6]=9;k[7]=4;k[8]=1;
             break;
         default:
             validn = 0;
         }
         break;
     case nsl_conv_kernel_first_derivative:
         if (n == 2) {
             k[0]=-1;k[1]=1;
         } else
             validn = 0;
         break;
     case nsl_conv_kernel_smooth_first_derivative:
         if (n%2) {
             for (i = 0; i < n; i++)
                 k[i] = (int)(i - n/2);
         } else
             validn = 0;
         break;
     case nsl_conv_kernel_second_derivative:
         if (n == 3) {
             k[0]=1;k[1]=-2;k[2]=1;
         } else
             validn = 0;
         break;
     case nsl_conv_kernel_third_derivative:
         if (n == 4) {
             k[0]=1;k[1]=-3;k[2]=3;k[3]=-1;
         } else
             validn = 0;
         break;
     case nsl_conv_kernel_fourth_derivative:
         if (n == 5) {
             k[0]=1;k[1]=-4;k[2]=6;k[3]=-4;k[4]=1;
         } else
             validn = 0;
         break;
     case nsl_conv_kernel_gaussian: {
         double s = n/5.;    /* relative width */
         for (i = 0;i < n; i++) {
             double x = i - (n-1.)/2.;
             k[i] = M_SQRT1_2/M_SQRTPI/s * exp(-x*x/2./s/s);
         }
         break;
     }
     case nsl_conv_kernel_lorentzian: {
         double s = n/5.;    /* relative width */
         for (i = 0;i < n; i++) {
             double x = i - (n-1.)/2.;
             k[i] = s/M_PI/(s*s + x*x);
         }
         break;
     }
     }
 
     if (!validn) {
         printf("ERROR: kernel size %lu not supported for kernel %s\n", (unsigned long)n, nsl_conv_kernel_name[type]);
         return -1;
     }
 
     /* debug */
     printf("[");
     for (i = 0; i < n; i++)
         printf("%g ", k[i]);
     puts("]");
 
     return 0;
 }
 
 int nsl_conv_convolution_direction(double s[], size_t n, double r[], size_t m, nsl_conv_direction_type dir, nsl_conv_type_type type, nsl_conv_method_type method,
         nsl_conv_norm_type normalize, nsl_conv_wrap_type wrap, double out[]) {
     if (dir == nsl_conv_direction_forward)
         return nsl_conv_convolution(s, n, r, m, type, method, normalize, wrap, out);
     else
         return nsl_conv_deconvolution(s, n, r, m, type, normalize, wrap, out);
 }
 
 int nsl_conv_convolution(double s[], size_t n, double r[], size_t m, nsl_conv_type_type type, nsl_conv_method_type method, nsl_conv_norm_type normalize, nsl_conv_wrap_type wrap, double out[]) {
     if (method == nsl_conv_method_direct || (method == nsl_conv_method_auto && GSL_MAX_INT(n, m) <= NSL_CONV_METHOD_BORDER)) {
         if (type == nsl_conv_type_linear)
             return nsl_conv_linear_direct(s, n, r, m, normalize, wrap, out);
         else if (type == nsl_conv_type_circular)
             return nsl_conv_circular_direct(s, n, r, m, normalize, wrap, out);
     } else {
         return nsl_conv_fft_type(s, n, r, m, nsl_conv_direction_forward, type, normalize, wrap, out);
     }
 
     return 0;
 }
 
 int nsl_conv_deconvolution(double s[], size_t n, double r[], size_t m, nsl_conv_type_type type, nsl_conv_norm_type normalize, nsl_conv_wrap_type wrap, double out[]) {
     /* only supported by FFT method */
     return nsl_conv_fft_type(s, n, r, m, nsl_conv_direction_backward, type, normalize, wrap, out);
 }
 
 int nsl_conv_linear_direct(double s[], size_t n, double r[], size_t m, nsl_conv_norm_type normalize, nsl_conv_wrap_type wrap, double out[]) {
     size_t i, j, size = n + m - 1, wi = 0;
     double norm = 1;
     if (normalize == nsl_conv_norm_euclidean) {
         if ((norm = cblas_dnrm2((int)m, r, 1)) == 0)
             norm = 1.;
     } else if (normalize == nsl_conv_norm_sum) {
         if ((norm = cblas_dasum((int)m, r, 1)) == 0)
             norm = 1.;
     }
 
     if (wrap == nsl_conv_wrap_max)
         nsl_stats_maximum(r, m, &wi);
     else if (wrap == nsl_conv_wrap_center)
         wi = m/2;
 
     for (j = 0; j < size; j++) {
         int index;  // can be negative
         double res = 0;
         for (i = 0; i < n; i++) {
             index = (int)(j - i);
             if (index >= 0 && index < (int)m)
                 res += s[i] * r[index]/norm;
         }
         index = (int)(j - wi);
         if (index < 0)
             index += (int)size;
         out[index] = res;
     }
 
     return 0;
 }
 
 int nsl_conv_circular_direct(double s[], size_t n, double r[], size_t m, nsl_conv_norm_type normalize, nsl_conv_wrap_type wrap, double out[]) {
     size_t i, j, size = GSL_MAX(n,m), wi = 0;
     double norm = 1;
     if (normalize == nsl_conv_norm_euclidean) {
         if ((norm = cblas_dnrm2((int)m, r, 1)) == 0)
             norm = 1.;
     } else if (normalize == nsl_conv_norm_sum) {
         if ((norm = cblas_dasum((int)m, r, 1)) == 0)
             norm = 1.;
     }
 
     if (wrap == nsl_conv_wrap_max)
         nsl_stats_maximum(r, m, &wi);
     else if (wrap == nsl_conv_wrap_center)
         wi = m/2;
 
     for (j = 0; j < size; j++) {
         int index;  // can be negative
         double res = 0;
         for (i = 0; i < n; i++) {
             index = (int)(j - i);
             if (index < 0)
                 index += (int)size;
             if (index < (int)m)
                 res += s[i]*r[index]/norm;
         }
         index = (int)(j - wi);
         if (index < 0)
             index += (int)size;
         out[index] = res;
     }
 
     return 0;
 }
 
 int nsl_conv_fft_type(double s[], size_t n, double r[], size_t m, nsl_conv_direction_type dir, nsl_conv_type_type type, nsl_conv_norm_type normalize, nsl_conv_wrap_type wrap, double out[]) {
     size_t i, size, wi = 0;
     if (type == nsl_conv_type_linear)
         size = n + m - 1;
     else    // circular
         size = GSL_MAX(n, m);
 
     double norm = 1.;
     if (normalize == nsl_conv_norm_euclidean) {
         if ((norm = cblas_dnrm2((int)m, r, 1)) == 0)
             norm = 1.;
     } else if (normalize == nsl_conv_norm_sum) {
         if ((norm = cblas_dasum((int)m, r, 1)) == 0)
             norm = 1.;
     }
 
     if (wrap == nsl_conv_wrap_max)
         nsl_stats_maximum(r, m, &wi);
     else if (wrap == nsl_conv_wrap_center)
         wi = m/2;
 
 #ifdef HAVE_FFTW3
     // already zero-pad here for FFT method and FFTW r2c
     size_t oldsize = size;
     size = 2*(size/2+1);
 #endif
 
     // zero-padded arrays
     double *stmp = (double*)malloc(size*sizeof(double));
     if (stmp == NULL) {
         printf("nsl_conv_fft_type(): ERROR allocating memory for 'stmp'!\n");
         return -1;
     }
 
     double *rtmp = (double*)malloc(size*sizeof(double));
     if (rtmp == NULL) {
         free(stmp);
         printf("nsl_corr_fft_type(): ERROR allocating memory for 'rtmp'!\n");
         return -1;
     }
 
     for (i = 0; i < n; i++)
         stmp[i] = s[i];
     for (i = n; i < size; i++)
         stmp[i] = 0;
     for (i = 0; i < m; i++) {
         rtmp[i] = r[i]/norm;    /* norm response */
     }
     for (i = m; i < size; i++)
         rtmp[i] = 0;
 
     int status;
 #ifdef HAVE_FFTW3   // already wraps output
     status = nsl_conv_fft_FFTW(stmp, rtmp, oldsize, dir, wi, out);
 #else   // GSL
     status = nsl_conv_fft_GSL(stmp, rtmp, size, dir, out);
     for (i = 0; i < size; i++) {    // wrap output
         size_t index = (i + wi) % size;
         stmp[i] = out[index];
     }
     memcpy(out, stmp, size * sizeof(double));
 #endif
 
     free(stmp);
     free(rtmp);
 
     return status;
 }
 
 #ifdef HAVE_FFTW3
 int nsl_conv_fft_FFTW(double s[], double r[], size_t n, nsl_conv_direction_type dir, size_t wi, double out[]) {
     size_t i;
     const size_t size = 2*(n/2+1);
     double* in = (double*)malloc(size*sizeof(double));
     fftw_plan rpf = fftw_plan_dft_r2c_1d((int)n, in, (fftw_complex*)in, FFTW_ESTIMATE);
 
     fftw_execute_dft_r2c(rpf, s, (fftw_complex*)s);
     fftw_execute_dft_r2c(rpf, r, (fftw_complex*)r);
     fftw_destroy_plan(rpf);
     free(in);
 
     // multiply/divide
     if (dir == nsl_conv_direction_forward) {
         for (i = 0; i < size; i += 2) {
             double re = s[i]*r[i] - s[i+1]*r[i+1];
             double im = s[i]*r[i+1] + s[i+1]*r[i];
 
             s[i] = re;
             s[i+1] = im;
         }
     } else {
         for (i = 0; i < size; i += 2) {
             double norm = r[i]*r[i] + r[i+1]*r[i+1];
             if (norm < DBL_MIN)
                 norm = 1.;
             double re = (s[i]*r[i] + s[i+1]*r[i+1])/norm;
             double im = (s[i+1]*r[i] - s[i]*r[i+1])/norm;
 
             s[i] = re;
             s[i+1] = im;
         }
     }
 
     // back transform
     double* o = (double*)malloc(size*sizeof(double));
     fftw_plan rpb = fftw_plan_dft_c2r_1d((int)n, (fftw_complex*)o, o, FFTW_ESTIMATE);
 
     fftw_execute_dft_c2r(rpb, (fftw_complex*)s, s);
     fftw_destroy_plan(rpb);
 
     for (i = 0; i < n; i++) {
         size_t index = (i + wi) % n;
         out[i] = s[index]/n;
     }
     free(o);
 
     return 0;
 }
 #endif
 
 int nsl_conv_fft_GSL(double s[], double r[], size_t n, nsl_conv_direction_type dir, double out[]) {
     gsl_fft_real_workspace *work = gsl_fft_real_workspace_alloc(n);
     gsl_fft_real_wavetable *real = gsl_fft_real_wavetable_alloc(n);
 
     /* FFT s and r */
     gsl_fft_real_transform(s, 1, n, real, work);
     gsl_fft_real_transform(r, 1, n, real, work);
     gsl_fft_real_wavetable_free(real);
 
     size_t i;
     /* calculate halfcomplex product/quotient depending on direction */
     if (dir == nsl_conv_direction_forward) {
         out[0] = s[0]*r[0];
         for (i = 1; i < n; i++) {
             if (i%2) { /* Re */
                 out[i] = s[i]*r[i];
                 if (i < n-1) /* when n is even last value is real */
                     out[i] -= s[i+1]*r[i+1];
             } else  /* Im */
                 out[i] = s[i-1]*r[i] + s[i]*r[i-1];
         }
     } else {
         out[0] = s[0]/r[0];
         for (i = 1; i < n; i++) {
             if (i%2) { /* Re */
                 if (i == n-1) /* when n is even last value is real */
                     out[i] = s[i]/r[i];
                 else {
                     double norm = r[i]*r[i] + r[i+1]*r[i+1];
                     if (norm < DBL_MIN)
                         norm = 1.;
                     out[i] = (s[i]*r[i] + s[i+1]*r[i+1])/norm;
                 }
             } else {    /* Im */
                 double norm = r[i-1]*r[i-1] + r[i]*r[i];
                 if (norm < DBL_MIN)
                     norm = 1.;
                 out[i] = (s[i]*r[i-1] - s[i-1]*r[i])/norm;
             }
         }
     }
 
     /* back transform */
     gsl_fft_halfcomplex_wavetable *hc = gsl_fft_halfcomplex_wavetable_alloc(n);
     gsl_fft_halfcomplex_inverse(out, 1, n, hc, work);
     gsl_fft_halfcomplex_wavetable_free(hc);
     gsl_fft_real_workspace_free(work);
 
     return 0;
 }