basis.hpp

Go to the documentation of this file.

/** \file basis.hpp
 * \author Jared R. Males (jaredmales@gmail.com)
 * \brief Utilities for working with a modal basis.
 * \ingroup mxAO_files
 *
 */
 
#ifndef __basis_hpp__
#define __basis_hpp__
 
#include "../ioutils/fits/fitsFile.hpp"
using namespace mx::fits;
 
#include "../sigproc/fourierModes.hpp"
#include "../sigproc/gramSchmidt.hpp"
// #include "../improc/eigenUtils.hpp"
#include "../improc/imageFilters.hpp"
#include "../improc/imagePads.hpp"
#include "../improc/eigenCube.hpp"
 
#include "../sigproc/signalWindows.hpp"
using namespace mx::improc;
using namespace mx::sigproc;
 
#include "../math/eigenLapack.hpp"
using namespace mx::math;
 
#include "aoPaths.hpp"
 
namespace mx
{
 
namespace AO
{
 
/// Multiply a raw modal basis by a pupil mask.
template <typename realT>
void applyPupil2Basis( eigenCube<realT> &modes,
                       const std::string &basisName,
                       const std::string &pupilName,
                       realT fwhm = 0 )
{
    fitsFile<realT> ff;
 
    std::string rawFName = mx::AO::path::basis::modes( basisName );
    ff.read( modes, rawFName );
 
    std::string pupilFName = mx::AO::path::pupil::pupilFile( pupilName );
    Eigen::Array<realT, -1, -1> pupil;
    ff.read( pupil, pupilFName );
 
    Eigen::Array<realT, -1, -1> im, pim, fim;
 
    for( int i = 0; i < modes.planes(); ++i )
    {
 
        modes.image( i ) = modes.image( i ) * pupil;
 
        if( fwhm > 0 )
        {
            im = modes.image( i );
 
            padImage( pim, im, pupil, 3 );
 
            filterImage( fim, pim, gaussKernel<Eigen::Array<realT, -1, -1>>( fwhm ) );
 
            modes.image( i ) += fim * ( pupil - 1 ).abs();
        }
    }
 
    // std::string procFName = mx::AO::path::basis::pupilModes(basisName, pupilName, true);
 
    // mx::fitsHeader head;
    // head.append( "FWHM", fwhm, "FWHM of smoothing kernel used for slaving");
    // ff.write(procFName, rawModes, head);
}
 
/** \def MXAO_ORTHO_METHOD_SGS
 * \brief Constant to specify using the stabilized Gramm Schmidt (SGS) orthogonalization procedure.
 */
#define MXAO_ORTHO_METHOD_SGS 0
 
/** \def MXAO_ORTHO_METHOD_SVD
 * \brief Constant to specify using the singular value decomposition (SVD) for orthogonalizing a modal basis.
 */
#define MXAO_ORTHO_METHOD_SVD 1
 
template <typename realT, typename spectRealT>
int orthogonalizeBasis( eigenCube<realT> &ortho,
                        Eigen::Array<spectRealT, -1, -1> &spect,
                        eigenCube<realT> &modes,
                        int method,
                        realT psum = 1.0 )
{
 
    if( method == MXAO_ORTHO_METHOD_SGS )
    {
        ortho.resize( modes.rows(), modes.cols(), modes.planes() );
        Eigen::Map<Eigen::Array<realT, -1, -1>> gsout( ortho.data(), ortho.rows() * ortho.cols(), ortho.planes() );
 
        gramSchmidtSpectrum<1>( gsout, spect, modes.asVectors(), psum );
    }
    else if( method == MXAO_ORTHO_METHOD_SVD )
    {
        Eigen::Array<realT, -1, -1> U, S, VT, A;
        int info;
        A = modes.asVectors();
        info = eigenGESDD( U, S, VT, A );
 
        if( info != 0 )
        {
            std::cerr << "Some error occurred in SVD\n";
            return -1;
        }
 
        // This maps to a cube, but does not copy or take ownership.
 
        eigenCube<realT> omodes( U.data(), modes.rows(), modes.cols(), modes.planes() );
 
        ortho.resize( modes.rows(), modes.cols(), modes.planes() );
        ortho = omodes;
    }
    else
    {
        std::cerr << "Invalid orthogonalization method.\n";
        return -1;
    }
 
    return 0;
}
 
/// Calculate an orthogonal projection of a basis on a pupil.
/** Can use either the stabilized Gramm Schmidt (SGS) procedure, or Singular Value Decomposition (SVD).
 * This calls \ref mx::AO::applyPupil2Basis as a first step.
 *
 * \param [in] basisName the name of the basis to orthogonalize
 * \param [in] pupilName the name of the pupil on which to orthogonalize.
 * \param [in] method either `MXAO_ORTHO_METHOD_SGS` (for SGS) or `MXAO_ORTHO_METHOD_SVD` (for SVD)
 *
 * \tparam realT
 */
template <typename realT>
void orthogonalizeBasis( const std::string &orthoName,
                         const std::string &basisName,
                         const std::string &pupilName,
                         int method )
{
    fitsFile<realT> ff;
    fitsHeader head;
 
    eigenCube<realT> modes;
    applyPupil2Basis( modes, basisName, pupilName );
 
    eigenCube<realT> ortho;
    Eigen::Array<realT, -1, -1> spect;
 
    orthogonalizeBasis( ortho, spect, modes, method );
 
    modes.resize( 1, 1, 1 );
 
    std::string pupilFName = mx::AO::path::pupil::pupilFile( pupilName );
    Eigen::Array<realT, -1, -1> pupil;
    ff.read( pupil, pupilFName );
 
    realT psum = pupil.sum();
    realT norm;
 
    for( int i = 0; i < ortho.planes(); ++i )
    {
        norm = ortho.image( i ).square().sum() / psum;
        ortho.image( i ) /= sqrt( norm );
 
        if( method == MXAO_ORTHO_METHOD_SGS )
        {
            // if(i == 0) std::cout << sqrt(norm) << "\n";
 
            // spect.row(i) /= sqrt(norm);
        }
    }
 
    std::string orthoFName = mx::AO::path::basis::modes( orthoName, true );
 
    std::string orthm = "UNK";
    if( method == MXAO_ORTHO_METHOD_SGS )
        orthm = "SGS";
    if( method == MXAO_ORTHO_METHOD_SVD )
        orthm = "SVD";
 
    head.append( "ORTHMETH", orthm, "Orthogonalization method." );
    head.append( "ORTHPUP", pupilName, "Pupil used for orthogonalization" );
 
    ff.write( orthoFName, ortho, head );
 
    if( method == MXAO_ORTHO_METHOD_SGS )
    {
        std::string orthoFName = mx::AO::path::basis::spectrum( orthoName, true );
        ff.write( orthoFName, spect, head );
    }
}
 
template <typename spectRealT, typename realT>
void normalizeSpectrum( mx::improc::eigenImage<spectRealT> &spectrum,
                        mx::improc::eigenCube<realT> &modes,
                        mx::improc::eigenCube<realT> &ortho,
                        mx::improc::eigenImage<realT> &pupil )
{
    int nModes = modes.planes();
 
    std::vector<realT> w( nModes );
 
    size_t psum = pupil.sum();
 
#pragma omp parallel
    {
        std::vector<realT> amps( nModes, 0.0 );
        std::vector<realT> uwAmps( nModes );
        std::vector<realT> rat( psum );
 
#pragma omp for
        for( int k = 0; k < nModes; ++k )
        {
            amps[k] = 1.0;
 
            for( int j = 0; j < nModes; ++j )
            {
                uwAmps[j] = 0; // amps(j, k)*spectrum(j,j);
 
                for( int i = j; i < nModes; ++i )
                {
                    uwAmps[j] += amps[i] * spectrum( i, j );
                }
            }
 
            mx::improc::eigenImage<realT> uwp( modes.rows(), modes.cols() );
            uwp.setZero();
 
            for( int i = 0; i < nModes; ++i )
                uwp += uwAmps[i] * modes.image( i );
 
            int idx = 0;
            for( int r = 0; r < pupil.rows(); ++r )
            {
                for( int c = 0; c < pupil.cols(); ++c )
                {
                    if( pupil( r, c ) == 1 )
                        rat[idx++] = ( uwp( r, c ) / ortho.image( k )( r, c ) );
                }
            }
 
            w[k] = mx::math::vectorMedianInPlace( rat );
 
            if( k == 1200 )
                std::cout << w[k] << "\n";
 
            amps[k] = 0.0;
        }
    }
 
    for( int i = 0; i < nModes; ++i )
        spectrum.row( i ) /= w[i];
 
    return;
}
 
template <typename realT>
int slaveBasis( const std::string &outputBasisN,
                const std::string &inputBasisN,
                const std::string &pupilN,
                const std::string &dmN,
                realT fwhm,
                realT fsmooth,
                int firstMode = 0 )
{
    fitsFile<realT> ff;
    fitsHeader head;
 
    // get DM size and clear memory first
    std::string dmFName = mx::AO::path::dm::influenceFunctions( dmN );
    eigenCube<realT> inf;
    ff.read( dmFName, inf );
 
    int dmSz = inf.rows();
 
    inf.resize( 0, 0 );
 
    // Now load basis
    std::string basisFName = mx::AO::path::basis::modes( inputBasisN ); //, pupilName);
    eigenCube<realT> modes;
    ff.read( basisFName, modes );
 
    // Now load pupil
    std::string pupilFName = mx::AO::path::pupil::pupilFile( pupilN );
    Eigen::Array<realT, -1, -1> pupil;
    ff.read( pupilFName, pupil );
 
    Eigen::Array<realT, -1, -1> im, ppim, pim, fim, ppupil;
 
    int padw = 2 * fwhm;
 
    if( fwhm == 0 )
        padw = 2 * fsmooth;
 
    int cutw = 0.5 * ( dmSz - modes.rows() );
 
    if( padw < cutw )
    {
        padw = cutw;
    }
 
    eigenCube<realT> oModes;
 
    oModes.resize( modes.rows() + 2 * cutw, modes.cols() + 2 * cutw, modes.planes() );
 
    if( modes.rows() > pupil.rows() )
    {
        padImage( ppupil, pupil, 0.5 * ( modes.rows() - pupil.rows() ) );
        pupil = ppupil;
    }
 
    padImage( ppupil, pupil, padw );
 
    for( int i = 0; i < firstMode; ++i )
    {
        im = modes.image( i );
        padImage( ppim, im, padw );
        cutPaddedImage( im, ppim, 0.5 * ppim.rows() - ( 0.5 * modes.rows() + cutw ) );
 
        oModes.image( i ) = im;
    }
 
    for( int i = firstMode; i < modes.planes(); ++i )
    {
 
        if( fwhm > 0 || fsmooth > 0 )
        {
 
            im = modes.image( i ) * pupil;
            padImage( ppim, im, padw );
            padImage( pim, im, ppupil, padw ); // ppupil,4);
 
            if( fwhm > 0 )
            {
                filterImage( fim, pim, gaussKernel<Eigen::Array<realT, -1, -1>>( fwhm ) );
                fim = ppim + fim * ( ppupil - 1 ).abs();
            }
            else
            {
                fim = ppim;
            }
 
            if( fsmooth > 0 )
            {
                pim = fim;
 
                filterImage( fim, pim, gaussKernel<Eigen::Array<realT, -1, -1>>( fsmooth ) );
            }
 
            cutPaddedImage( im, fim, 0.5 * fim.rows() - ( 0.5 * modes.rows() + cutw ) );
 
            oModes.image( i ) = im;
        }
    }
 
    std::string oFName = mx::AO::path::basis::modes( outputBasisN, true );
    ff.write( oFName, oModes );
 
    return 0;
}
 
template <typename realT>
int apodizeBasis( const std::string &outputBasisN,
                  const std::string &inputBasisN,
                  realT tukeyAlpha,
                  realT centralObs,
                  realT overScan,
                  int firstMode )
{
    fitsFile<realT> ff;
    fitsHeader head;
 
    std::string basisFName = mx::AO::path::basis::modes( inputBasisN ); //, pupilName);
    eigenCube<realT> modes;
    ff.read( basisFName, modes );
 
    realT cen = 0.5 * ( modes.rows() - 1.0 );
 
    Eigen::Array<realT, -1, -1> pupil;
    pupil.resize( modes.rows(), modes.cols() );
 
    if( centralObs == 0 )
    {
        window::tukey2d<realT>( pupil.data(), modes.rows(), modes.rows() + overScan, tukeyAlpha, cen, cen );
    }
    else
    {
        window::tukey2dAnnulus<realT>(
            pupil.data(), modes.rows(), modes.rows() + overScan, centralObs, tukeyAlpha, cen, cen );
    }
 
    for( int i = firstMode; i < modes.planes(); ++i )
    {
        modes.image( i ) = modes.image( i ) * pupil;
    }
 
    std::string oFName = mx::AO::path::basis::modes( outputBasisN, true );
    ff.write( oFName, modes );
 
    return 0;
}
 
template <typename realT>
int subtractBasis( const std::string &basisName,
                   const std::string &basisName1,
                   const std::string &basisName2,
                   const std::string &pupilName,
                   int method )
{
    fitsFile<realT> ff;
    fitsHeader head;
 
    std::string basisFName1 = mx::AO::path::basis::modes( basisName1 ); //, pupilName);
    eigenCube<realT> modes1;
    ff.read( basisFName1, modes1 );
 
    std::string basisFName2 = mx::AO::path::basis::modes( basisName2 ); //, pupilName);
    eigenCube<realT> modes2;
    ff.read( basisFName2, modes2, head );
    realT fwhm = head["FWHM"].value<realT>();
 
    eigenCube<realT> modes;
    modes.resize( modes1.rows(), modes1.cols(), modes1.planes() + modes2.planes() );
 
    std::string pupilFName = mx::AO::path::pupil::pupilFile( pupilName );
    Eigen::Array<realT, -1, -1> pupil;
    ff.read( pupilFName, pupil );
 
    for( int i = 0; i < modes1.planes(); ++i )
    {
        modes.image( i ) = modes1.image( i ) * pupil;
    }
 
    for( int i = 0; i < modes2.planes(); ++i )
    {
        modes.image( modes1.planes() + i ) = modes2.image( i ) * pupil;
    }
 
    eigenCube<realT> ortho;
 
    orthogonalizeBasis( ortho, modes, method );
 
    // Now copy out just the modes corresponding to the 2nd basis set.
    // Apply the gaussian smooth FWHM again.
    Eigen::Array<realT, -1, -1> im, ppim, pim, fim, ppupil;
 
    modes.resize( ortho.rows(), ortho.cols(), modes2.planes() );
 
    realT fsmooth = 6.0;
 
    padImage( ppupil, pupil, 10 );
 
    for( int i = 0; i < modes2.planes(); ++i )
    {
        modes.image( i ) = ortho.image( modes1.planes() + i ) * pupil;
 
        if( fwhm > 0 )
        {
            im = modes.image( i );
            padImage( ppim, im, 10 );
            padImage( pim, im, 10 ); // ppupil,4);
 
            filterImage( fim, pim, gaussKernel<Eigen::Array<realT, -1, -1>>( fwhm ) );
            fim = ppim + fim * ( ppupil - 1 ).abs();
 
            if( fsmooth > 0 )
            {
                pim = fim;
                filterImage( fim, pim, gaussKernel<Eigen::Array<realT, -1, -1>>( fsmooth ) );
            }
 
            cutPaddedImage( im, fim, 10 );
            modes.image( i ) = im;
        }
    }
 
    std::string orthoFName = mx::AO::path::basis::modes( basisName, true );
    ff.write( orthoFName, modes );
 
    return 0;
}
 
} // namespace AO
} // namespace mx
 
#endif //__basis_hpp__