[1942] | 1 | #include "FSAppIrrSmpl.h"
|
---|
| 2 |
|
---|
| 3 |
|
---|
| 4 | FSApproximationIrregularSampling::FSApproximationIrregularSampling() : fftIntfPtr_(NULL) {;}
|
---|
| 5 |
|
---|
| 6 | FSApproximationIrregularSampling::FSApproximationIrregularSampling(TVector<double>& sampling, double offset, double range)
|
---|
| 7 | {
|
---|
| 8 | initFFT();
|
---|
| 9 | makeSamplingVector(sampling, offset, range);
|
---|
| 10 | M_ = 0;
|
---|
| 11 | }
|
---|
| 12 |
|
---|
| 13 |
|
---|
| 14 | FSApproximationIrregularSampling::~FSApproximationIrregularSampling()
|
---|
| 15 | {if (fftIntfPtr_!=NULL) delete fftIntfPtr_;}
|
---|
| 16 |
|
---|
| 17 | void FSApproximationIrregularSampling::makeRHS(TVector<complex<double> >& coefSolution)
|
---|
| 18 | {
|
---|
| 19 | int k;
|
---|
| 20 | if (exponFourier_.Size() == 0) calculeExponentiellesFourier();
|
---|
| 21 | coefSolution.ReSize(2*M_+1);
|
---|
| 22 | coefSolution = complex<double>(0.,0.);
|
---|
| 23 | int nbEchantillons = samplingValues_.Size();
|
---|
| 24 |
|
---|
| 25 | // initialisation d'un tableau de travail pour calcul des termes
|
---|
| 26 | // du second membre par recurrence
|
---|
| 27 | TVector<complex<double> > travail(nbEchantillons);
|
---|
| 28 | for (k=0; k < nbEchantillons; k++)
|
---|
| 29 | {
|
---|
| 30 | travail(k) = poids_(k)*signal_(k);
|
---|
| 31 | coefSolution(M_) += travail(k);
|
---|
| 32 | }
|
---|
| 33 |
|
---|
| 34 | // recurrence
|
---|
| 35 | for (k=1; k<=M_; k++)
|
---|
| 36 | {
|
---|
| 37 | int j;
|
---|
| 38 | for (j=0; j < nbEchantillons; j++)
|
---|
| 39 | {
|
---|
| 40 | travail(j) *= exponFourier_(j);
|
---|
| 41 | coefSolution(M_+k) += travail(j);
|
---|
| 42 | }
|
---|
| 43 | coefSolution(M_-k) = conj(coefSolution(M_+k));
|
---|
| 44 | }
|
---|
| 45 | }
|
---|
| 46 |
|
---|
| 47 | void FSApproximationIrregularSampling::makeToeplitzMatrix(int M)
|
---|
| 48 | {
|
---|
| 49 | int j,k;
|
---|
| 50 | M_ = M;
|
---|
| 51 | int nbEchantillons = samplingValues_.Size();
|
---|
| 52 | // matrice de Toeplitz
|
---|
| 53 | if (exponFourier_.Size() == 0) calculeExponentiellesFourier();
|
---|
| 54 |
|
---|
| 55 | TVector<complex<double> > gamma(2*M_+1);
|
---|
| 56 | gamma = complex<double>(0.,0.);
|
---|
| 57 | // initialisation d'un tableau de travail pour calcul des termes
|
---|
| 58 | // de la matrice par recurrence
|
---|
| 59 | TVector<complex<double> > travail(nbEchantillons);
|
---|
| 60 | travail = poids_;
|
---|
| 61 |
|
---|
| 62 | for (j=0; j<nbEchantillons; j++) gamma(0) += travail(j);
|
---|
| 63 |
|
---|
| 64 | // recurrence
|
---|
| 65 | for (k=1; k<=2*M_; k++)
|
---|
| 66 | {
|
---|
| 67 |
|
---|
| 68 | for (j=0; j<nbEchantillons; j++)
|
---|
| 69 | {
|
---|
| 70 | travail(j) *= exponFourier_(j);
|
---|
| 71 | gamma(k) += travail(j);
|
---|
| 72 | }
|
---|
| 73 |
|
---|
| 74 | }
|
---|
| 75 |
|
---|
| 76 | tptz_.setMatrix(gamma);
|
---|
| 77 | }
|
---|
| 78 | void FSApproximationIrregularSampling::approximateSignal(int M, const TVector<double>& signal)
|
---|
| 79 | {
|
---|
| 80 | int k;
|
---|
| 81 | if (delta_ <= 1./(2.*M) )
|
---|
| 82 | {
|
---|
| 83 | cout << " FSApproximationIrregularSampling : BON ECHANTILLONNAGE " << endl;
|
---|
| 84 | }
|
---|
| 85 | else
|
---|
| 86 | {
|
---|
| 87 | cout << " FSApproximationIrregularSampling : ATTENTION : SIGNAL SOUS-ECHANTILLONNE " << endl;
|
---|
| 88 | cout << " deltaMax (normalise) = " << delta_ << " devrait etre inferieur a " << 1./(2.*M) << endl;
|
---|
| 89 | cout << " ecart max intervient entre echantillon no " << nokdelta_ << " et le suivant, abscisse= " << samplingValues_(nokdelta_)*samplingRange_+samplingOffset_ << endl;
|
---|
| 90 |
|
---|
| 91 | }
|
---|
| 92 | // PrtTim(" avant toeplitz " );
|
---|
| 93 |
|
---|
| 94 | if ( M != M_ ) makeToeplitzMatrix(M);
|
---|
| 95 |
|
---|
| 96 | // PrtTim(" fin toeplitz " );
|
---|
| 97 |
|
---|
| 98 |
|
---|
| 99 | makeSignalVector(signal);
|
---|
| 100 | // PrtTim(" fin fabrication signal " );
|
---|
| 101 |
|
---|
| 102 | // second membre
|
---|
| 103 |
|
---|
| 104 | TVector<complex<double> > coefSolution;
|
---|
| 105 | makeRHS(coefSolution);
|
---|
| 106 | int j;
|
---|
| 107 | // PrtTim(" fin fabrication second membre " );
|
---|
| 108 |
|
---|
| 109 | int niter = tptz_.gradientToeplitz(coefSolution);
|
---|
| 110 | // int niter = tptz_.gradientToeplitzPreconTChang(coefSolution);
|
---|
| 111 | cout << " FSApproximationIrregularSampling::approximateSignal : converged in " << niter << " iterations " << endl;
|
---|
| 112 | coefFourier_.ReSize(M_+1);
|
---|
| 113 | coefFourier_ = coefSolution(Range(M_, 2*M_));
|
---|
| 114 | }
|
---|
| 115 |
|
---|
| 116 | // la periode normalisee 1 est divisee en nbInterv intervalles
|
---|
| 117 | // les valeurs de la solution sont donnes en 0, 1/n, ..... (n-1)/n
|
---|
| 118 | // le calcul est beaucoup plus rapide si nbInterv est pair (FFT)
|
---|
| 119 | void FSApproximationIrregularSampling::restaureRegularlySampledSignal(int nbInterv, TVector<double>& solution) const
|
---|
| 120 | {
|
---|
| 121 | if (nbInterv < 2*M_+1 || nbInterv%2 != 0)
|
---|
| 122 | {
|
---|
| 123 | int k;
|
---|
| 124 | solution.ReSize(nbInterv);
|
---|
| 125 | double delta = 1./nbInterv;
|
---|
| 126 | for (k=0; k<nbInterv; k++)
|
---|
| 127 | {
|
---|
| 128 | double u = k*delta;
|
---|
| 129 | solution(k) = valeursSerie(u);
|
---|
| 130 | }
|
---|
| 131 | }
|
---|
| 132 | else
|
---|
| 133 | {
|
---|
| 134 | int tailleTF = nbInterv/2+1;
|
---|
| 135 | TVector<complex<double> > TFf(tailleTF);
|
---|
| 136 | TFf = complex<double>(0.,0.);
|
---|
| 137 | TFf(Range(0,M_)) = coefFourier_;
|
---|
| 138 | fftIntfPtr_-> FFTBackward(TFf, solution);
|
---|
| 139 | }
|
---|
| 140 | reshapeSignalInUsersFrame(solution);
|
---|
| 141 | }
|
---|
| 142 |
|
---|
| 143 | void FSApproximationIrregularSampling::computeSignalOnASampling(const TVector<double>& abscisses, TVector<double>& solution ) const
|
---|
| 144 | {
|
---|
| 145 | int k;
|
---|
| 146 | int n= abscisses.Size();
|
---|
| 147 | if (n<=0)
|
---|
| 148 | {
|
---|
| 149 | cout << " restaurationEnPoints: vecteurs de points vide " << endl;
|
---|
| 150 | return;
|
---|
| 151 | }
|
---|
| 152 | TVector<double> abscissesLocales;
|
---|
| 153 | abscissesLocales = abscisses;
|
---|
| 154 | matchToSamplingReference(abscissesLocales);
|
---|
| 155 | solution.ReSize(n);
|
---|
| 156 | for (k=0; k<n; k++)
|
---|
| 157 | {
|
---|
| 158 | double u = abscissesLocales(k);
|
---|
| 159 | solution(k) = valeursSerie(u);
|
---|
| 160 | }
|
---|
| 161 | reshapeSignalInUsersFrame(abscisses, solution);
|
---|
| 162 |
|
---|
| 163 | }
|
---|
| 164 |
|
---|
| 165 | double FSApproximationIrregularSampling::estimationConditionnement() const
|
---|
| 166 | {
|
---|
| 167 | double deuxDeltaM = 2.*delta_*M_;
|
---|
| 168 | double cond = (1.+deuxDeltaM)/(1.-deuxDeltaM);
|
---|
| 169 | cond *= cond;
|
---|
| 170 | return cond;
|
---|
| 171 | }
|
---|
| 172 |
|
---|
| 173 | void FSApproximationIrregularSampling::samplingValues(TVector<double>& sv) const
|
---|
| 174 | {
|
---|
| 175 | int k;
|
---|
| 176 | int n = samplingValues_.Size();
|
---|
| 177 | sv.ReSize(n);
|
---|
| 178 | for (k=0; k<n;k++)
|
---|
| 179 | {
|
---|
| 180 | sv(k) = samplingOffset_+samplingRange_*samplingValues_(k);
|
---|
| 181 | }
|
---|
| 182 |
|
---|
| 183 | }
|
---|
| 184 |
|
---|
| 185 | // terme constant, puis cos, sin, cos, sin......
|
---|
| 186 | void FSApproximationIrregularSampling::coeffCosSin(TVector<double>& coef) const
|
---|
| 187 | {
|
---|
| 188 | int j;
|
---|
| 189 | coef.ReSize(2*M_+1);
|
---|
| 190 | coef(0) = coefFourier_(0).real();
|
---|
| 191 | for (j=1; j<M_; j++)
|
---|
| 192 | {
|
---|
| 193 | double aj = 2.*coefFourier_(j).real();
|
---|
| 194 | double bj = -2.*coefFourier_(j).imag();
|
---|
| 195 | coef(2*(j-1)+1) = aj;
|
---|
| 196 | coef(2*(j-1)+2) = bj;
|
---|
| 197 | }
|
---|
| 198 | coef *= normeSignal_;
|
---|
| 199 | }
|
---|
| 200 |
|
---|
| 201 | // exprime les valeurs d'abscisses, selon la reference locale
|
---|
| 202 | void FSApproximationIrregularSampling::matchToSamplingReference(TVector<double>& sampling) const
|
---|
| 203 |
|
---|
| 204 | {
|
---|
| 205 | int k;
|
---|
| 206 | int compteur = sampling.Size();
|
---|
| 207 | double fac = 1./samplingRange_;
|
---|
| 208 | for (k=0; k<compteur; k++)
|
---|
| 209 | {
|
---|
| 210 | sampling(k) = (sampling(k)-samplingOffset_)*fac;
|
---|
| 211 |
|
---|
| 212 | }
|
---|
| 213 | if ( sampling(0) <0. || sampling(compteur-1) >1. )
|
---|
| 214 | {
|
---|
| 215 | cout << " matchToSamplingReference: points hors [0.,1.] " << endl;
|
---|
| 216 | cout << " " << sampling(0) << " " << sampling(compteur-1) << endl;
|
---|
| 217 | }
|
---|
| 218 | }
|
---|
| 219 |
|
---|
| 220 | // exprime les valeurs d'echantillonnage entre 0 et 1
|
---|
| 221 | void FSApproximationIrregularSampling::resizeSamplingIn_0_1(double offset, double range)
|
---|
| 222 | {
|
---|
| 223 | int k;
|
---|
| 224 | int compteur = samplingValues_.Size();
|
---|
| 225 | samplingOffset_ = offset;
|
---|
| 226 | samplingRange_ = range;
|
---|
| 227 | double fac = 1./samplingRange_;
|
---|
| 228 | for (k=0; k<compteur ;k++)
|
---|
| 229 | {
|
---|
| 230 | samplingValues_(k) = (samplingValues_(k)-samplingOffset_)*fac;
|
---|
| 231 | }
|
---|
| 232 | }
|
---|
| 233 |
|
---|
| 234 | void FSApproximationIrregularSampling::computeWeights()
|
---|
| 235 | {
|
---|
| 236 | int k;
|
---|
| 237 | int nbEchantillons = samplingValues_.Size();
|
---|
| 238 | nokdelta_ = 0;
|
---|
| 239 |
|
---|
| 240 | // calcul de l'ecart maximum entre deux temps d'echantillonnage consecutifs
|
---|
| 241 | delta_ = samplingValues_(0)-samplingValues_(nbEchantillons-1)+1;
|
---|
| 242 | for (k=0; k< nbEchantillons-1; k++)
|
---|
| 243 | {
|
---|
| 244 | double diff = samplingValues_(k+1)-samplingValues_(k);
|
---|
| 245 | if ( diff > delta_ )
|
---|
| 246 | {
|
---|
| 247 | delta_ = diff;
|
---|
| 248 | nokdelta_ = k;
|
---|
| 249 | }
|
---|
| 250 | }
|
---|
| 251 | // calcul des poids (pour tenir compte de l'irregularite de
|
---|
| 252 | // l'echantillonnage)
|
---|
| 253 | poids_.ReSize(nbEchantillons);
|
---|
| 254 | poids_(0) = 0.5*(samplingValues_(1)-samplingValues_(nbEchantillons-1) + 1.);
|
---|
| 255 | for (k=1; k< nbEchantillons-1; k++)
|
---|
| 256 | {
|
---|
| 257 | poids_(k) = 0.5*(samplingValues_(k+1)-samplingValues_(k-1));
|
---|
| 258 | }
|
---|
| 259 | poids_(nbEchantillons-1) = 0.5*(samplingValues_(0) +1 - samplingValues_(nbEchantillons-2));
|
---|
| 260 |
|
---|
| 261 | }
|
---|
| 262 | void FSApproximationIrregularSampling::NormSignal()
|
---|
| 263 | {
|
---|
| 264 | int k;
|
---|
| 265 | int nbEchantillons = samplingValues_.Size();
|
---|
| 266 | normeSignal_=0;
|
---|
| 267 | for (k=0; k< nbEchantillons; k++)
|
---|
| 268 | {
|
---|
| 269 | double s = signal_(k);
|
---|
| 270 | normeSignal_ += s*s*poids_(k);
|
---|
| 271 | }
|
---|
| 272 | normeSignal_=sqrt(normeSignal_);
|
---|
| 273 | double fac = 1./normeSignal_;
|
---|
| 274 | signal_ *= fac;
|
---|
| 275 | }
|
---|
| 276 |
|
---|
| 277 |
|
---|
| 278 |
|
---|
| 279 | void FSApproximationIrregularSampling::reshapeSignalInUsersFrame(const TVector<double>& abscisses, TVector<double>& resultat) const
|
---|
| 280 | {
|
---|
| 281 | if (resultat.Size() <= 0)
|
---|
| 282 | {
|
---|
| 283 | cout << " reshapeSignalInUsersFrame: vecteur solution VIDE" << endl;
|
---|
| 284 | }
|
---|
| 285 | resultat *= normeSignal_;
|
---|
| 286 | }
|
---|
| 287 |
|
---|
| 288 | void FSApproximationIrregularSampling::reshapeSignalInUsersFrame(TVector<double>& resultat) const
|
---|
| 289 | {
|
---|
| 290 | if (resultat.Size() <= 0)
|
---|
| 291 | {
|
---|
| 292 | cout << " reshapeSignalInUsersFrame: vecteur solution VIDE" << endl;
|
---|
| 293 | }
|
---|
| 294 | resultat *= normeSignal_;
|
---|
| 295 | }
|
---|
| 296 |
|
---|
| 297 |
|
---|
| 298 | void FSApproximationIrregularSampling::makeSamplingVector(const TVector<double>& sampling, double offset, double range)
|
---|
| 299 | {
|
---|
| 300 | samplingValues_.ReSize(sampling.Size());
|
---|
| 301 | samplingValues_ = sampling;
|
---|
| 302 | resizeSamplingIn_0_1(offset, range);
|
---|
| 303 | computeWeights();
|
---|
| 304 | }
|
---|
| 305 |
|
---|
| 306 | void FSApproximationIrregularSampling::makeSignalVector(const TVector<double>& signal)
|
---|
| 307 | {
|
---|
| 308 | int n = samplingValues_.Size();
|
---|
| 309 | if (n != signal.Size() )
|
---|
| 310 | {
|
---|
| 311 | cout << " echantillonnage et signal n'ont pas les memes dimensions " << endl;
|
---|
| 312 | }
|
---|
| 313 | signal_ = signal;
|
---|
| 314 | NormSignal();
|
---|
| 315 | }
|
---|
| 316 |
|
---|
| 317 |
|
---|
| 318 |
|
---|
| 319 |
|
---|
| 320 |
|
---|
| 321 | void FSApproximationIrregularSampling::restaureSignal(TVector<double>& solution) const
|
---|
| 322 | {
|
---|
| 323 | int k;
|
---|
| 324 | int n= samplingValues_.Size();
|
---|
| 325 | if (n<=0)
|
---|
| 326 | {
|
---|
| 327 | cout << " restaurationEnPoints: vecteurs de points vide " << endl;
|
---|
| 328 | return;
|
---|
| 329 | }
|
---|
| 330 | solution.ReSize(n);
|
---|
| 331 | for (k=0; k<n; k++)
|
---|
| 332 | {
|
---|
| 333 | double u = samplingValues_(k);
|
---|
| 334 | solution(k) = valeursSerie(u);
|
---|
| 335 | }
|
---|
| 336 | reshapeSignalInUsersFrame(solution);
|
---|
| 337 | }
|
---|
| 338 |
|
---|
| 339 |
|
---|