| 1 | /* ======================================================================== *\
|
|---|
| 2 | !
|
|---|
| 3 | ! *
|
|---|
| 4 | ! * This file is part of MARS, the MAGIC Analysis and Reconstruction
|
|---|
| 5 | ! * Software. It is distributed to you in the hope that it can be a useful
|
|---|
| 6 | ! * and timesaving tool in analysing Data of imaging Cerenkov telescopes.
|
|---|
| 7 | ! * It is distributed WITHOUT ANY WARRANTY.
|
|---|
| 8 | ! *
|
|---|
| 9 | ! * Permission to use, copy, modify and distribute this software and its
|
|---|
| 10 | ! * documentation for any purpose is hereby granted without fee,
|
|---|
| 11 | ! * provided that the above copyright notice appear in all copies and
|
|---|
| 12 | ! * that both that copyright notice and this permission notice appear
|
|---|
| 13 | ! * in supporting documentation. It is provided "as is" without express
|
|---|
| 14 | ! * or implied warranty.
|
|---|
| 15 | ! *
|
|---|
| 16 | !
|
|---|
| 17 | !
|
|---|
| 18 | ! Author(s): Christoph Kolodziejski, 11/2004 <mailto:>
|
|---|
| 19 | ! Author(s): Thomas Bretz, 11/2004 <mailto:tbretz@astro.uni-wuerzburg.de>
|
|---|
| 20 | !
|
|---|
| 21 | ! Copyright: MAGIC Software Development, 2004-2005
|
|---|
| 22 | !
|
|---|
| 23 | !
|
|---|
| 24 | \* ======================================================================== */
|
|---|
| 25 |
|
|---|
| 26 | //////////////////////////////////////////////////////////////////////////////
|
|---|
| 27 | //
|
|---|
| 28 | // MHexagonalFT
|
|---|
| 29 | //
|
|---|
| 30 | // This is a class representating a (fast) fourier transformation explicitly
|
|---|
| 31 | // for hexagonal geometries as described in astro-ph/0409388.
|
|---|
| 32 | //
|
|---|
| 33 | //
|
|---|
| 34 | // WARNING:
|
|---|
| 35 | // ========
|
|---|
| 36 | // Be carefull using the fast transformation (Prepare())! The precalculation
|
|---|
| 37 | // consumes a lot of memory. fPsi has the size of 2*n^4 (while n is the
|
|---|
| 38 | // number of rows in fourier space). For the enhanced MAGIC camery fPsi has
|
|---|
| 39 | // the size 27691682*sizeof(float) = 105.6MB (Std MAGIC: ~12MB)
|
|---|
| 40 | //
|
|---|
| 41 | // The runtime is more or less determined by the speed of accessing a
|
|---|
| 42 | // huge amount of memory (see above) sequentially.
|
|---|
| 43 | //
|
|---|
| 44 | //
|
|---|
| 45 | // Coordinate systems:
|
|---|
| 46 | // ===================
|
|---|
| 47 | //
|
|---|
| 48 | // original hexagonal structure enhanced hexagonal structure
|
|---|
| 49 | // ---------------------------- ----------------------------
|
|---|
| 50 | //
|
|---|
| 51 | // structure
|
|---|
| 52 | // ---------
|
|---|
| 53 | //
|
|---|
| 54 | // h h h f f h h h f f
|
|---|
| 55 | // h h h h f h h h h f
|
|---|
| 56 | // h h h h h -----> h h h h h
|
|---|
| 57 | // h h h h h h h h
|
|---|
| 58 | // h h h h h h
|
|---|
| 59 | // f f
|
|---|
| 60 | // f
|
|---|
| 61 | //
|
|---|
| 62 | // numbering
|
|---|
| 63 | // ---------
|
|---|
| 64 | // d c b m n o p q r s
|
|---|
| 65 | // e 4 3 a g h i j k l
|
|---|
| 66 | // f 5 1 2 9 -----> b c d e f
|
|---|
| 67 | // g 6 7 8 7 8 9 a
|
|---|
| 68 | // h i j 4 5 6
|
|---|
| 69 | // 2 3
|
|---|
| 70 | // 1
|
|---|
| 71 | //
|
|---|
| 72 | // In reality the fourier space looks like because of symmetries:
|
|---|
| 73 | //
|
|---|
| 74 | // real part imaginary part
|
|---|
| 75 | // --------- --------------
|
|---|
| 76 | // m n o p o n m m n o 0 -o -n -m
|
|---|
| 77 | // g h i i h g g h i -i -h -g
|
|---|
| 78 | // b c d c b b c 0 -c -b
|
|---|
| 79 | // 7 8 8 7 7 8 -8 -7
|
|---|
| 80 | // 4 5 4 4 0 -4
|
|---|
| 81 | // 2 2 2 -2
|
|---|
| 82 | // 1 0
|
|---|
| 83 | //
|
|---|
| 84 | // column: GetK() row: GetM()
|
|---|
| 85 | // -------------- -----------
|
|---|
| 86 | // 6 5 4 3 2 1 0 0 1 2 3 4 5 6
|
|---|
| 87 | // 5 4 3 2 1 0 0 1 2 3 4 5
|
|---|
| 88 | // 4 3 2 1 0 0 1 2 3 4
|
|---|
| 89 | // 3 2 1 0 0 1 2 3
|
|---|
| 90 | // 2 1 0 0 1 2
|
|---|
| 91 | // 1 0 0 1
|
|---|
| 92 | // 0 0
|
|---|
| 93 | //
|
|---|
| 94 | // row: GetRow() (m+k) column: GetCol() (m-k)
|
|---|
| 95 | // ------------------- ----------------------
|
|---|
| 96 | // 6 6 6 6 6 6 6 -6 -4 -2 0 2 4 6
|
|---|
| 97 | // 5 5 5 5 5 5 -5 -3 -1 1 3 5
|
|---|
| 98 | // 4 4 4 4 4 -4 -2 0 2 4
|
|---|
| 99 | // 3 3 3 3 -3 -1 1 3
|
|---|
| 100 | // 2 2 2 -2 0 2
|
|---|
| 101 | // 1 1 -1 1
|
|---|
| 102 | // 0 0
|
|---|
| 103 | //
|
|---|
| 104 | //
|
|---|
| 105 | // The coordinates of the pixels in the triangle are:
|
|---|
| 106 | //
|
|---|
| 107 | // Double_t dx; // Distance of too pixels in x
|
|---|
| 108 | // Double_t dy; // Distance of to pixel rows in y
|
|---|
| 109 | // Int_t idx; // Index of pixel in triangle (see above)
|
|---|
| 110 | //
|
|---|
| 111 | // const Float_t x = dx*GetCol(idx);
|
|---|
| 112 | // const Float_t y = dy*Int_t(GetRow(idx)-2*GetNumRows()/3);
|
|---|
| 113 | //
|
|---|
| 114 | // You can use MGeomCam::GetPixelIdxXY(x, y) to get the corresponding index
|
|---|
| 115 | // in space space.
|
|---|
| 116 | //
|
|---|
| 117 | //////////////////////////////////////////////////////////////////////////////
|
|---|
| 118 | #include "MHexagonalFT.h"
|
|---|
| 119 |
|
|---|
| 120 | #include <TMath.h>
|
|---|
| 121 |
|
|---|
| 122 | #include "MLog.h"
|
|---|
| 123 | #include "MLogManip.h"
|
|---|
| 124 |
|
|---|
| 125 | #include "MArrayD.h"
|
|---|
| 126 |
|
|---|
| 127 | ClassImp(MHexagonalFT);
|
|---|
| 128 |
|
|---|
| 129 | using namespace std;
|
|---|
| 130 |
|
|---|
| 131 | // ---------------------------------------------------------------------------
|
|---|
| 132 | //
|
|---|
| 133 | // Default Constructor - empty
|
|---|
| 134 | //
|
|---|
| 135 | MHexagonalFT::MHexagonalFT()
|
|---|
| 136 | {
|
|---|
| 137 | }
|
|---|
| 138 |
|
|---|
| 139 | // ---------------------------------------------------------------------------
|
|---|
| 140 | //
|
|---|
| 141 | // Default Constructor - num is the number of lines the fourier space has.
|
|---|
| 142 | // It calls Prepare to fill the arrays with the necessary coefficients.
|
|---|
| 143 | //
|
|---|
| 144 | // Here are some simple rules to calculate parameters in a hexagonal space:
|
|---|
| 145 | //
|
|---|
| 146 | // Number of Rings (r) ---> Number of Pixels (p)
|
|---|
| 147 | // p = 3*r*(r-1)+1
|
|---|
| 148 | //
|
|---|
| 149 | // Number of Pixels (p) ---> Number of Rings (r)
|
|---|
| 150 | // p = (sqrt(9+12*(p-1))+3)/6
|
|---|
| 151 | //
|
|---|
| 152 | // Number of pixels at one border == number of rings (r)
|
|---|
| 153 | // Row of border == number of rings (r)
|
|---|
| 154 | //
|
|---|
| 155 | // Number of rows to get a triangle: 3r-2
|
|---|
| 156 | //
|
|---|
| 157 | MHexagonalFT::MHexagonalFT(Int_t num)
|
|---|
| 158 | {
|
|---|
| 159 | Prepare(num);
|
|---|
| 160 | }
|
|---|
| 161 |
|
|---|
| 162 | // ---------------------------------------------------------------------------
|
|---|
| 163 | //
|
|---|
| 164 | // Calculate the contents of: fM, fK, fP, fIdx and fPsi.
|
|---|
| 165 | //
|
|---|
| 166 | // While fPsi are the fourier coefficients, fM and fK are the hexagonal x and
|
|---|
| 167 | // y coordinates of the pixel corresponding to the index i which is the
|
|---|
| 168 | // common index of all arrays. fP is P(i,j) for all pixels.
|
|---|
| 169 | //
|
|---|
| 170 | // fIdx is also filled and used for reverse mapping. Due to the geometry
|
|---|
| 171 | // the right and left side of the fourier space triangle has identical
|
|---|
| 172 | // values. fIdx 'maps' the indices from the right to the left side.
|
|---|
| 173 | //
|
|---|
| 174 | void MHexagonalFT::Prepare(Int_t num)
|
|---|
| 175 | {
|
|---|
| 176 | fNumRows = num;
|
|---|
| 177 |
|
|---|
| 178 | fPsi.Set(num*num*num*num*2);
|
|---|
| 179 |
|
|---|
| 180 | Int_t lim = num*(num+1)/2;
|
|---|
| 181 |
|
|---|
| 182 | fM.Set(lim);
|
|---|
| 183 | fK.Set(lim);
|
|---|
| 184 | fP.Set(lim);
|
|---|
| 185 | fIdx.Set(lim);
|
|---|
| 186 |
|
|---|
| 187 | for(int j=0; j<num; j++)
|
|---|
| 188 | {
|
|---|
| 189 | for(int n=0; n+j<num; n++)
|
|---|
| 190 | {
|
|---|
| 191 | int idx1 = (j+n)*(j+n+1)/2 + j;
|
|---|
| 192 |
|
|---|
| 193 | fM[idx1]=n;
|
|---|
| 194 | fK[idx1]=j;
|
|---|
| 195 |
|
|---|
| 196 | fP[idx1]=P(j,n);
|
|---|
| 197 |
|
|---|
| 198 | for(int k=0; k<num; k++)
|
|---|
| 199 | {
|
|---|
| 200 | for(int m=0; m+k<num; m++)
|
|---|
| 201 | {
|
|---|
| 202 | const Double_t dx = TMath::Pi()*(m-k)/(num-1)/3;
|
|---|
| 203 | const Double_t dy = TMath::Pi()*(m+k)/(num-1);
|
|---|
| 204 |
|
|---|
| 205 | const Double_t cos1 = TMath::Cos(dy*(j+n));
|
|---|
| 206 | const Double_t cos2 = TMath::Cos(dy*j);
|
|---|
| 207 | const Double_t cos3 = TMath::Cos(dy*n);
|
|---|
| 208 |
|
|---|
| 209 | const Double_t psire = 2*(
|
|---|
| 210 | +cos1*TMath::Cos(dx*(j-n))
|
|---|
| 211 | +cos2*TMath::Cos(dx*(j+2*n))
|
|---|
| 212 | +cos3*TMath::Cos(dx*(2*j+n)));
|
|---|
| 213 |
|
|---|
| 214 | const Double_t psiim = 2*(
|
|---|
| 215 | +cos1*TMath::Sin(dx*(j-n))
|
|---|
| 216 | +cos2*TMath::Sin(dx*(j+2*n))
|
|---|
| 217 | -cos3*TMath::Sin(dx*(2*j+n)));
|
|---|
| 218 |
|
|---|
| 219 | const Int_t idx3 = (k+m)*(k+m+1)/2 + k;
|
|---|
| 220 | const Int_t id1 = idx1*lim + idx3;
|
|---|
| 221 |
|
|---|
| 222 | fPsi[id1*2] = psire;
|
|---|
| 223 | fPsi[id1*2+1] = psiim;
|
|---|
| 224 | }
|
|---|
| 225 | }
|
|---|
| 226 | }
|
|---|
| 227 | }
|
|---|
| 228 |
|
|---|
| 229 | for (int idx1=0; idx1<lim; idx1++)
|
|---|
| 230 | {
|
|---|
| 231 | int n = fM[idx1];
|
|---|
| 232 | int j = fK[idx1];
|
|---|
| 233 |
|
|---|
| 234 | int idx0;
|
|---|
| 235 | for (idx0=0; idx0<lim; idx0++)
|
|---|
| 236 | if (fM[idx0]==j && fK[idx0]==n)
|
|---|
| 237 | break;
|
|---|
| 238 |
|
|---|
| 239 | fIdx[idx1]=idx0;
|
|---|
| 240 | }
|
|---|
| 241 |
|
|---|
| 242 | }
|
|---|
| 243 |
|
|---|
| 244 | // ---------------------------------------------------------------------------
|
|---|
| 245 | //
|
|---|
| 246 | // Do a fast forward tranformation. Because all coefficients are
|
|---|
| 247 | // precalculated, the tranformation is reduced to a simple pointer based
|
|---|
| 248 | // loop over the coeffiecients multiplied with the corresponding input
|
|---|
| 249 | // values.
|
|---|
| 250 | //
|
|---|
| 251 | // Parameters:
|
|---|
| 252 | // inre: array storing the real part of the input (eg. pixel contents)
|
|---|
| 253 | // outre: array storing the real part of the output
|
|---|
| 254 | // outim: array storing the imaginary part of the output
|
|---|
| 255 | //
|
|---|
| 256 | // inre must be of the size of the fourier space triangle. The pixel
|
|---|
| 257 | // contents must have been mapped into this new space with the proper
|
|---|
| 258 | // pixel indices. The size of outre and outim is set accordingly.
|
|---|
| 259 | //
|
|---|
| 260 | // After initialization (Prepare()) you can get the size of the arrays with
|
|---|
| 261 | // GetNumKnots()
|
|---|
| 262 | //
|
|---|
| 263 | // For the definition of the coordinate system see class description
|
|---|
| 264 | //
|
|---|
| 265 | void MHexagonalFT::TransformFastFWD(const MArrayD &inre, MArrayD &outre,
|
|---|
| 266 | MArrayD &outim) const
|
|---|
| 267 | {
|
|---|
| 268 | const UInt_t num = fP.GetSize();
|
|---|
| 269 |
|
|---|
| 270 | if (inre.GetSize()!=num)
|
|---|
| 271 | {
|
|---|
| 272 | cout << "ERROR - MHexagonalFT prepared for different size." << endl;
|
|---|
| 273 | return;
|
|---|
| 274 | }
|
|---|
| 275 |
|
|---|
| 276 | outre.Set(num);
|
|---|
| 277 | outim.Set(num);
|
|---|
| 278 |
|
|---|
| 279 | const Int_t cnt = 108*(fNumRows-1)*(fNumRows-1);
|
|---|
| 280 |
|
|---|
| 281 | const Float_t *endp = fP.GetArray()+num;
|
|---|
| 282 |
|
|---|
| 283 | for (UInt_t idx1=0; idx1<num; idx1++)
|
|---|
| 284 | {
|
|---|
| 285 | if (fK[idx1]>fM[idx1])
|
|---|
| 286 | continue;
|
|---|
| 287 |
|
|---|
| 288 | Double_t sumre=0;
|
|---|
| 289 | Double_t sumim=0;
|
|---|
| 290 |
|
|---|
| 291 | Float_t *psi = fPsi.GetArray() + idx1*num*2;
|
|---|
| 292 | Float_t *p = fP.GetArray();
|
|---|
| 293 | Double_t *re = inre.GetArray();
|
|---|
| 294 |
|
|---|
| 295 | // 1st access to psi: const Float_t psire = *psi++;
|
|---|
| 296 | // 2nd access to psi: const Float_t psiim = *psi++;
|
|---|
| 297 | // sumre += f * *psire;
|
|---|
| 298 | // sumim += f * *psiim;
|
|---|
| 299 | while (p<endp)
|
|---|
| 300 | {
|
|---|
| 301 | const Double_t f = *p++ * *re++;
|
|---|
| 302 |
|
|---|
| 303 | sumre += f * *psi++;
|
|---|
| 304 | sumim += f * *psi++;
|
|---|
| 305 | }
|
|---|
| 306 |
|
|---|
| 307 | const Double_t factor2 = fP[idx1]/cnt;
|
|---|
| 308 |
|
|---|
| 309 | outre[fIdx[idx1]] = (outre[idx1] = factor2 * sumre);
|
|---|
| 310 | outim[fIdx[idx1]] = -(outim[idx1] = -factor2 * sumim);
|
|---|
| 311 | }
|
|---|
| 312 | }
|
|---|
| 313 |
|
|---|
| 314 | // ---------------------------------------------------------------------------
|
|---|
| 315 | //
|
|---|
| 316 | // Do a fast backward tranformation. Because all coefficients are
|
|---|
| 317 | // precalculated, the tranformation is reduced to a simple pointer based
|
|---|
| 318 | // loop over the coeffiecients multiplied with the corresponding input
|
|---|
| 319 | // values.
|
|---|
| 320 | //
|
|---|
| 321 | // Parameters:
|
|---|
| 322 | // inre: outre of TransformFastBwd
|
|---|
| 323 | // inim: outim of TransformFastBwd
|
|---|
| 324 | // outre: backward tranformed real part of the resulting
|
|---|
| 325 | //
|
|---|
| 326 | // inre and inim must be of the size of the fourier space triangle. The
|
|---|
| 327 | // pixel contents must have been mapped into this new space with the proper
|
|---|
| 328 | // pixel indices. The size of outre is set accordingly.
|
|---|
| 329 | //
|
|---|
| 330 | // After initialization (Prepare()) you can get the size of the arrays with
|
|---|
| 331 | // GetNumKnots()
|
|---|
| 332 | //
|
|---|
| 333 | // For the definition of the coordinate system see class description
|
|---|
| 334 | //
|
|---|
| 335 | void MHexagonalFT::TransformFastBWD(const MArrayD &inre, const MArrayD &inim,
|
|---|
| 336 | MArrayD &outre) const
|
|---|
| 337 | {
|
|---|
| 338 | const UInt_t num = fP.GetSize();
|
|---|
| 339 |
|
|---|
| 340 | // Sanity check: check size of arrays
|
|---|
| 341 | if (inre.GetSize()!=num)
|
|---|
| 342 | {
|
|---|
| 343 | cout << "ERROR - MHexagonalFT prepared for different size." << endl;
|
|---|
| 344 | return;
|
|---|
| 345 | }
|
|---|
| 346 | if (inim.GetSize()!=num)
|
|---|
| 347 | {
|
|---|
| 348 | cout << "ERROR - MHexagonalFT prepared for different size." << endl;
|
|---|
| 349 | return;
|
|---|
| 350 | }
|
|---|
| 351 | outre.Set(num);
|
|---|
| 352 |
|
|---|
| 353 | const Double_t *endre = inre.GetArray()+num;
|
|---|
| 354 |
|
|---|
| 355 | for (UInt_t idx1=0; idx1<num; idx1++)
|
|---|
| 356 | {
|
|---|
| 357 | Float_t *psi = fPsi.GetArray() + idx1*num*2;
|
|---|
| 358 | Double_t *im = inim.GetArray();
|
|---|
| 359 | Double_t *re = inre.GetArray();
|
|---|
| 360 |
|
|---|
| 361 | Double_t sumre=0;
|
|---|
| 362 | while (re<endre)
|
|---|
| 363 | {
|
|---|
| 364 | const Float_t psire = *psi++;
|
|---|
| 365 | const Float_t psiim = *psi++;
|
|---|
| 366 |
|
|---|
| 367 | sumre += *re++ * psire - *im++ * psiim;
|
|---|
| 368 | }
|
|---|
| 369 |
|
|---|
| 370 | outre[idx1] = sumre;
|
|---|
| 371 | }
|
|---|
| 372 | }
|
|---|
| 373 |
|
|---|
| 374 | // ---------------------------------------------------------------------------
|
|---|
| 375 | //
|
|---|
| 376 | // This is a slow (direct) version of the tranformation. It is identical
|
|---|
| 377 | // for forward and backward tranformation.
|
|---|
| 378 | //
|
|---|
| 379 | // The whole calculation is done straight forward without any precalculation.
|
|---|
| 380 | //
|
|---|
| 381 | // Parameters:
|
|---|
| 382 | // inre: real part of input
|
|---|
| 383 | // inim: imaginary part of input
|
|---|
| 384 | // outre: real part of output
|
|---|
| 385 | // outim: imaginary part of output
|
|---|
| 386 | // fwd: kTRUE for forward, kFALSE for backward transformations
|
|---|
| 387 | //
|
|---|
| 388 | // After initialization (Prepare()) you can get the size of the arrays with
|
|---|
| 389 | // GetNumKnots()
|
|---|
| 390 | //
|
|---|
| 391 | // For the definition of the coordinate system see class description
|
|---|
| 392 | //
|
|---|
| 393 | // It is currently not tested!
|
|---|
| 394 | //
|
|---|
| 395 | void MHexagonalFT::TransformSlow(const MArrayD &inre, const MArrayD &inim,
|
|---|
| 396 | MArrayD &outre, MArrayD &outim,
|
|---|
| 397 | Bool_t fwd)
|
|---|
| 398 | {
|
|---|
| 399 | static const Double_t fgSqrt3 = TMath::Sqrt(3.);
|
|---|
| 400 | static const Double_t fgTan30 = TMath::Tan(30*TMath::DegToRad())*3;
|
|---|
| 401 |
|
|---|
| 402 | Int_t num = (Int_t)TMath::Sqrt((Float_t)inim.GetSize());
|
|---|
| 403 | Int_t cnt = 108*(num-1)*(num-1);
|
|---|
| 404 | Int_t inv = fwd?-1:1;
|
|---|
| 405 |
|
|---|
| 406 | // FIXME: For p(j,n)
|
|---|
| 407 | fNumRows = num;
|
|---|
| 408 |
|
|---|
| 409 | for(int j=0; j<num; j++)
|
|---|
| 410 | {
|
|---|
| 411 | for(int n=0; n+j<num; n++)
|
|---|
| 412 | {
|
|---|
| 413 | if (j-n>0 && fwd)
|
|---|
| 414 | continue;
|
|---|
| 415 |
|
|---|
| 416 | Double_t sumre=0;
|
|---|
| 417 | Double_t sumim=0;
|
|---|
| 418 |
|
|---|
| 419 | for(int k=0; k<num; k++)
|
|---|
| 420 | {
|
|---|
| 421 | for(int m=0; m+k<num; m++)
|
|---|
| 422 | {
|
|---|
| 423 | Double_t dx = 0.5*(m-k)/num;
|
|---|
| 424 | Double_t dy = 0.5*(m+k)/num*fgTan30;
|
|---|
| 425 |
|
|---|
| 426 | dx *= TMath::TwoPi()/3;
|
|---|
| 427 | dy *= TMath::TwoPi()/fgSqrt3;
|
|---|
| 428 |
|
|---|
| 429 | const Double_t cos1 = TMath::Cos(dy*(j+n));
|
|---|
| 430 | const Double_t cos2 = TMath::Cos(dy*j);
|
|---|
| 431 | const Double_t cos3 = TMath::Cos(dy*n);
|
|---|
| 432 |
|
|---|
| 433 | //Alternatie nach Paper:
|
|---|
| 434 | const Double_t psire = 2*(
|
|---|
| 435 | +cos1*TMath::Cos(dx*(j-n))
|
|---|
| 436 | +cos2*TMath::Cos(dx*(j+2*n))
|
|---|
| 437 | +cos3*TMath::Cos(dx*(2*j+n)));
|
|---|
| 438 |
|
|---|
| 439 | const Double_t psiim = 2*inv*(
|
|---|
| 440 | +cos1*TMath::Sin(dx*(j-n))
|
|---|
| 441 | +cos2*TMath::Sin(dx*(j+2*n))
|
|---|
| 442 | -cos3*TMath::Sin(dx*(2*j+n)));
|
|---|
| 443 |
|
|---|
| 444 | const Double_t factor = (fwd==1?P(k,m):1.);
|
|---|
| 445 |
|
|---|
| 446 | sumre += factor * (inre[k*num+m]*psire - inim[k*num+m]*psiim);
|
|---|
| 447 | sumim += factor * (inre[k*num+m]*psiim + inim[k*num+m]*psire);
|
|---|
| 448 | }
|
|---|
| 449 | }
|
|---|
| 450 |
|
|---|
| 451 | const Double_t factor = (fwd==1?P(j,n)/cnt:1.);
|
|---|
| 452 |
|
|---|
| 453 | outre[j*num+n] = factor * sumre;
|
|---|
| 454 | outim[j*num+n] = factor * sumim;
|
|---|
| 455 |
|
|---|
| 456 | if (fwd)
|
|---|
| 457 | {
|
|---|
| 458 | outre[n*num+j] = factor * sumre;
|
|---|
| 459 | outim[n*num+j] = -factor * sumim;
|
|---|
| 460 | }
|
|---|
| 461 | }
|
|---|
| 462 | }
|
|---|
| 463 | }
|
|---|
| 464 |
|
|---|
| 465 |
|
|---|