LCOV - code coverage report
Current view: top level - synthesis/MeasurementComponents - DelayRateFFT.cc (source / functions) Hit Total Coverage
Test: casacpp_coverage.info Lines: 0 840 0.0 %
Date: 2024-10-28 15:53:10 Functions: 0 34 0.0 %

          Line data    Source code
       1             : //# DelayRateFFT.cc: Part of the implementation of FringeJones
       2             : //# Copyright (C) 1996,1997,1998,1999,2000,2001,2002,2003,2011
       3             : //# Associated Universities, Inc. Washington DC, USA.
       4             : //#
       5             : //# This library is free software; you can redistribute it and/or modify it
       6             : //# under the terms of the GNU Library General Public License as published by
       7             : //# the Free Software Foundation; either version 2 of the License, or (at your
       8             : //# option) any later version.
       9             : //#
      10             : //# This library is distributed in the hope that it will be useful, but WITHOUT
      11             : //# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
      12             : //# FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Library General Public
      13             : //# License for more details.
      14             : //#
      15             : //# You should have received a copy of the GNU Library General Public License
      16             : //# along with this library; if not, write to the Free Software Foundation,
      17             : //# Inc., 675 Massachusetts Ave, Cambridge, MA 02139, USA.
      18             : //#
      19             : //# Correspondence concerning AIPS++ should be addressed as follows:
      20             : //#        Internet email: casa-feedback@nrao.edu.
      21             : //#        Postal address: AIPS++ Project Office
      22             : //#                        National Radio Astronomy Observatory
      23             : //#                        520 Edgemont Road
      24             : //#                        Charlottesville, VA 22903-2475 USA
      25             : //#
      26             : 
      27             : #include <synthesis/MeasurementComponents/FringeJones.h>
      28             : 
      29             : #include <synthesis/MeasurementComponents/SolveDataBuffer.h>
      30             : #include <casacore/lattices/Lattices/ArrayLattice.h>
      31             : #include <casacore/lattices/LatticeMath/LatticeFFT.h>
      32             : 
      33             : #include <casacore/casa/Exceptions/Error.h>
      34             : 
      35             : 
      36             : #include <gsl/gsl_rng.h>
      37             : #include <gsl/gsl_vector.h>
      38             : #include <gsl/gsl_multimin.h>
      39             : 
      40             : #include <iomanip>                // needed for setprecision
      41             : 
      42             : // DEVDEBUG gates the development debugging information to standard
      43             : // error; it should be set to 0 for production.
      44             : 
      45             : #define DEVDEBUG false
      46             : 
      47             : using namespace casa::vi;
      48             : using namespace casacore;
      49             :  
      50             : namespace casa {
      51             : 
      52             : Complex
      53             : dotMatrixWithModel(const Matrix<Complex>& data, Double k, Double l, Double offset);
      54             : 
      55             : Complex
      56             : dotMatrixWithModel2(const Matrix<Complex>& f, Double k, Double l, Double offset);
      57             : 
      58             : Complex
      59             : dotWithOffsets(const Cube<Complex>& ft, const Vector<Float>& offsets, double k, double l);
      60             : 
      61             : tuple<Double, Double, Double, Double>
      62             : multibandFFTs(const Array<Complex>& ffts, const Vector<Float>& offsets);
      63             : 
      64             : Double
      65             : multibandFFT(const Vector<Complex>& peaks, const Vector<Float>& offsets);
      66             : 
      67             : tuple<Double, Double, Double, Double>
      68             : refineSearch2(const Cube<Complex>& ft,  const Vector<Float>& offsets, Double ipkt0, Double ipkch0);
      69             : 
      70             : tuple<Double, Double, Double, Double>
      71             : bruteForceDelay(const Cube<Complex>& ft,  const Vector<Float>& offsets, Int ipkt, Int ipkch);
      72             : 
      73             : 
      74             : 
      75           0 : static void unitize(Array<Complex>& vC) 
      76             : {
      77           0 :     Array<Float> vCa(amplitude(vC));
      78             :     // Divide by non-zero amps
      79           0 :     vCa(vCa<FLT_EPSILON)=1.0;
      80           0 :     vC /= vCa;
      81           0 : }
      82             : 
      83           0 : SDBListGridManagerCombo::SDBListGridManagerCombo(SDBList& sdbs) :
      84             :     SDBListGridManager(sdbs),
      85           0 :     nchan_(0)
      86             : {
      87           0 :     std::set<Double> fmaxes_;
      88           0 :     std::set<Double> fmins_;
      89             : 
      90           0 :     if (sdbs.nSDB()==0) {
      91             :         // The for loop is fine with an empty list, but code below it
      92             :         // isn't and there's nothing to lose by bailing early!
      93             :         if (DEVDEBUG) {
      94             :             cerr << "No data I guess?";
      95             :         }
      96           0 :         return;
      97             :     }
      98             :     
      99           0 :     for (Int i=0; i != sdbs_.nSDB(); i++) {
     100           0 :         SolveDataBuffer& sdb = sdbs_(i);
     101           0 :         Int pspw = sdb.spectralWindow()(0);
     102           0 :         Double t = sdbs_(i).time()(0);
     103           0 :         times_.insert(t); 
     104           0 :         if (spwins_.find(pspw) == spwins_.end()) {
     105           0 :             spwins_.insert(pspw);
     106           0 :             const Vector<Double>& fs = sdb.freqs();
     107           0 :             df_ = fs[1] - fs[0];
     108           0 :             pspwIdToFreqMap_[pspw] = &(sdb.freqs());
     109           0 :             nchan_ = max(nchan_, sdb.nChannels());
     110           0 :             fmaxes_.insert(fs(nchan_-1));
     111           0 :             fmins_.insert(fs(0));
     112             :         } else {
     113           0 :             continue;
     114             :         }
     115             :     }
     116           0 :     nt_ = times_.size();
     117             :     // nt_ = sdbs_.nSDB()/spwins_.size();
     118           0 :     tmin_ = *(times_.begin());
     119           0 :     tmax_ = *(times_.rbegin());
     120           0 :     dt_ = (tmax_ - tmin_)/(nt_ - 1);
     121           0 :     if (nchan_ == 1) {
     122           0 :       df_ = 1;
     123           0 :       return;
     124             :     }
     125           0 :     size_t i = 0;
     126           0 :     for (auto p=spwins_.begin(); p!=spwins_.end(); p++) {
     127           0 :         spwPMap_[*p] = i;
     128           0 :         i++;
     129             :     }
     130           0 : }
     131             : 
     132             : 
     133             : 
     134             : 
     135             : Float
     136           0 : SDBListGridManagerCombo::getRefFreqFromLSPW(Int lspw) {
     137           0 :     auto p = spwins_.begin();
     138           0 :     std::advance(p, lspw);
     139           0 :     Int pspw = *p;
     140           0 :     const Vector<Double>& fs = *(pspwIdToFreqMap_[pspw]);
     141           0 :     return fs[0];
     142             : }
     143             :    
     144           0 : DelayRateFFT::DelayRateFFT(SDBList& sdbs, Int refant, Array<Double>& delayWindow, Array<Double>& rateWindow) :
     145           0 :     f0_(sdbs.centroidFreq() / 1.e9),      // GHz, for delayrate calc
     146           0 :     refant_(refant),
     147           0 :     delayWindow_(delayWindow),
     148           0 :     rateWindow_(rateWindow),
     149           0 :     Vall_(),
     150           0 :     xcount_(),
     151           0 :     sumw_(),
     152           0 :     sumww_(),
     153           0 :     activeAntennas_(),
     154           0 :     allActiveAntennas_()
     155           0 : {}
     156             : 
     157             : DelayRateFFT *
     158           0 : DelayRateFFT::makeAChild(int concat, SDBList& sdbs,
     159             :                          Int refant,
     160             :                          Array<Double>& delayWindow_,
     161             :                          Array<Double>& rateWindow_)
     162             : {
     163           0 :     if (concat) {
     164           0 :         return new DelayRateFFTConcat(sdbs, refant, delayWindow_, rateWindow_);
     165             :     } else {
     166           0 :         return new DelayRateFFTCombo(sdbs, refant, delayWindow_, rateWindow_);
     167             :     }
     168             : }
     169             : 
     170             : Matrix<Float>
     171           0 : DelayRateFFT::delay() const {
     172           0 :     IPosition start(2, 1, 0);
     173           0 :     IPosition stop(2, 3*nCorr_-1, nElem_-1);
     174           0 :     IPosition stride(2, 3, 1);
     175           0 :     Slicer sl(start,  stop, stride, Slicer::endIsLast);
     176           0 :     return param_(sl);
     177           0 : }
     178             : 
     179             : Matrix<Float>
     180           0 : DelayRateFFT::rate() const {
     181           0 :     IPosition start(2, 2, 0);
     182           0 :     IPosition stop(2, 3*nCorr_-1, nElem_-1);
     183           0 :     IPosition stride(2, 3, 1);
     184           0 :     Slicer sl(start,  stop, stride, Slicer::endIsLast);
     185           0 :     return param_(sl);
     186           0 : }
     187             : 
     188             : 
     189           0 : void DelayRateFFT::removeAntennasCorrelation(Int icor, std::set< Int > s) {
     190           0 :     std::set< Int > & as = activeAntennas_.find(icor)->second;
     191           0 :     for (std::set< Int >::iterator it=s.begin(); it!=s.end(); it++) {
     192           0 :         as.erase(*it);
     193             :     }
     194           0 : }
     195             : 
     196             : void
     197           0 : DelayRateFFT::printActive() {
     198           0 :     for (Int icorr=0; icorr != nCorr_; icorr++) {
     199           0 :         cerr << "Antennas found for correlation " << icorr << ": ";
     200           0 :         for (std::set<Int>::iterator it = activeAntennas_[icorr].begin();
     201           0 :              it != activeAntennas_[icorr].end(); it++) {
     202           0 :             cerr << *it << ", ";
     203             :         }
     204           0 :         cerr << endl;
     205             :     }
     206           0 :     cerr << endl;
     207           0 : }    
     208             : 
     209             : 
     210             : 
     211             : // DelayRateFFTCombo is modeled on DelayFFT in KJones.{cc|h}
     212           0 : DelayRateFFTCombo::DelayRateFFTCombo(SDBList& sdbs, Int refant, Array<Double>& delayWindow,
     213           0 :                                      Array<Double>& rateWindow) :
     214             :     DelayRateFFT(sdbs, refant, delayWindow, rateWindow),
     215           0 :     gm_(sdbs)
     216             : {
     217             :     if (DEVDEBUG) {
     218             :         gm_.describe();
     219             :     }
     220           0 :     nt_ = gm_.nt_;
     221           0 :     nChan_ = gm_.nchan_;
     222             :     
     223             :     // We might want to pad these too
     224           0 :     nPadFactor_ = 4;
     225           0 :     nPadT_ = nPadFactor_*nt_;
     226           0 :     nPadChan_ = nPadFactor_*nChan_;
     227           0 :     nspw_ = gm_.nSPW();
     228           0 :     dt_ = gm_.dt_;
     229           0 :     df_ = gm_.df_ / 1.e9;
     230             : 
     231           0 :     if (nt_ < 2) {
     232           0 :         throw(AipsError("Can't do a 2-dimensional FFT on a single timestep! Please consider changing solint to avoid orphan timesteps."));
     233             :     }
     234           0 :     IPosition ds = delayWindow_.shape();
     235           0 :     if (ds.size()!=1 || ds.nelements() != 1) {
     236           0 :         throw AipsError("delaywindow must be a list of length 2.");
     237             :     }
     238           0 :     IPosition rs = rateWindow_.shape();
     239           0 :     if (rs.size()!=1 || rs.nelements() != 1) {
     240           0 :         throw AipsError("ratewindow must be a list of length 2.");
     241             :     }
     242           0 :     Int nCorrOrig(sdbs(0).nCorrelations());
     243           0 :     nCorr_ = (nCorrOrig> 1 ? 2 : 1); // number of p-hands
     244           0 :     for (Int i=0; i<nCorr_; i++) {
     245           0 :         activeAntennas_[i].insert(refant_);
     246             :     }
     247             :     // when we get the visCubecorrected it is already
     248             :     // reduced to parallel hands, but there isn't a
     249             :     // corresponding method for flags.
     250           0 :     Int corrStep = (nCorrOrig > 2 ? 3 : 1); // step for p-hands
     251           0 :     for (Int ibuf=0; ibuf != sdbs.nSDB(); ibuf++) {
     252           0 :         SolveDataBuffer& s(sdbs(ibuf));
     253           0 :         for (Int irow=0; irow!=s.nRows(); irow++) {
     254           0 :             Int a1(s.antenna1()(irow)), a2(s.antenna2()(irow));
     255           0 :             allActiveAntennas_.insert(a1);
     256           0 :             allActiveAntennas_.insert(a2);
     257             :         }
     258             :     }
     259           0 :     nElem_ =  1 + *(allActiveAntennas_.rbegin()) ;
     260             : 
     261           0 :     IPosition aggregateDim(2, nCorr_, nElem_);
     262           0 :     xcount_.resize(aggregateDim);
     263           0 :     sumw_.resize(aggregateDim);
     264           0 :     sumww_.resize(aggregateDim);
     265           0 :     peak_.resize(aggregateDim);
     266             : 
     267           0 :     xcount_ = 0;
     268           0 :     sumw_ = 0.0;
     269           0 :     sumww_ = 0.0;
     270           0 :     IPosition dataSize(5, nCorr_, nElem_, nspw_, nPadT_, nPadChan_);
     271           0 :     Vall_.resize(dataSize);
     272           0 :     Int totalRows = 0;
     273           0 :     Int goodRows = 0;
     274             : 
     275             :     if (DEVDEBUG) {
     276             :       cerr << "DelayRateFFTCombo constructor: Here" << endl;
     277             :     }
     278           0 :     for (Int ibuf=0; ibuf != sdbs.nSDB(); ibuf++) {
     279           0 :         SolveDataBuffer& s(sdbs(ibuf));
     280           0 :         totalRows += s.nRows();
     281           0 :         if (!s.Ok())
     282           0 :             continue;
     283             : 
     284           0 :         Int nr = 0;
     285           0 :         for (Int irow=0; irow!=s.nRows(); irow++) {
     286           0 :             if (s.flagRow()(irow))
     287           0 :                 continue;
     288             :             Int iant;
     289           0 :             Int a1(s.antenna1()(irow)), a2(s.antenna2()(irow));
     290           0 :             if (a1 == a2) { continue; } // we don't do autocorrelations
     291           0 :             else if (a1 == refant_) { iant = a2; }
     292           0 :             else if (a2 == refant_) { iant = a1; }
     293           0 :             else { continue; } // not a baseline to reference antenna
     294             :             // v has shape (nelems, ?, nrows, nchannels); can't be a reference.
     295           0 :             Cube<Complex> v = s.visCubeCorrected();
     296           0 :             const Cube<Float>& w( s.weightSpectrum() );
     297           0 :             const Cube<Bool>& fl( s.flagCube() );
     298           0 :             Int pspw = s.spectralWindow()(0);
     299           0 :             Int ispw = gm_.getLSPW(pspw);
     300           0 :             Int t_index = gm_.getTimeIndex(s.time()(0));
     301             :             // FIXME! ispw is the Measurement Set spectral window; it
     302             :             // doesn't follow that that is addressible in our array
     303             :             // which is dimensioned for logical spectral
     304             :             // window. E.g. and i.e., if we are not joining spectral
     305             :             // windows, the spectral-window dimension of our target
     306             :             // array will always have size 1
     307           0 :             IPosition start (5,        0,   iant, ispw, t_index,      0);
     308           0 :             IPosition stop  (5,   nCorr_,      1,    1,       1, nChan_);
     309           0 :             IPosition stride(5,        1,      1,    1,       1,      1);
     310           0 :             Slicer target_slice(start, stop, stride, Slicer::endIsLength); 
     311             :             // Slicer::endIsLast is also possible
     312             :             if (DEVDEBUG && false) {
     313             :                 cerr << " nSDBs" << sdbs.nSDB()
     314             :                      << " nspwins " << gm_.spwins_.size() << endl;
     315             :                 cerr <<  "nCorr_ " << nCorr_
     316             :                      << " iant " << iant
     317             :                      << " ispw " << ispw
     318             :                      << " t_index " << t_index
     319             :                      << " s.time()(0) " << s.time()(0)
     320             :                      << " nt_ " << nt_
     321             :                      << " dt_ " << dt_
     322             :                      << " nChan_ " << nChan_
     323             :                      << " corrStep " << corrStep
     324             :                      << " distinct times " << gm_.times_.size()
     325             :                      << endl;
     326             :             }
     327           0 :             Slicer source_slice(IPosition(3, 0,         0, irow),
     328           0 :                                 IPosition(3, nCorr_,  nChan_, 1),
     329           0 :                                 IPosition(3, corrStep,      1,  1), Slicer::endIsLength);
     330             :                 
     331           0 :             Slicer flagSlice(IPosition(3, 0,             0, irow),
     332           0 :                              IPosition(3, nCorr_,    nChan_, 1),
     333           0 :                              IPosition(3, corrStep,       1, 1), Slicer::endIsLength);
     334           0 :             nr++;
     335           0 :             Array<Complex>rhs( v(source_slice).nonDegenerate(1) );
     336           0 :             const Array<Float>& weights( w(source_slice).nonDegenerate(1) );
     337           0 :             unitize(rhs); 
     338           0 :             Vall_(target_slice).nonDegenerate(1) = rhs * weights;
     339           0 :             const Array<Bool>& flagged( fl(flagSlice).nonDegenerate(1) );
     340             : 
     341             :             // Zero flagged entries.
     342           0 :             Vall_(target_slice).nonDegenerate(1)(flagged) = Complex(0.0);
     343             : 
     344           0 :             if (!allTrue(flagged)) {
     345           0 :                 for (Int icorr=0; icorr<nCorr_; ++icorr) {
     346             :                     // Book keeping now done per spw!
     347           0 :                     IPosition p(3, icorr, iant, ispw);
     348           0 :                     Bool actually = false;
     349           0 :                     activeAntennas_[icorr].insert(iant);
     350           0 :                     for (Int ichan=0; ichan != (Int) nChan_; ichan++) {
     351           0 :                         IPosition pchan(2, icorr, ichan);
     352           0 :                         if (!flagged(pchan)) {
     353           0 :                             Float wv = weights(pchan);
     354           0 :                             xcount_(p)++;
     355           0 :                             sumw_(p) += wv;
     356           0 :                             sumww_(p) += wv*wv;
     357           0 :                             actually = true;
     358             :                         }
     359           0 :                     }
     360           0 :                     if (actually) {
     361           0 :                         activeAntennas_[icorr].insert(iant);
     362           0 :                         goodRows++;
     363             :                     }
     364           0 :                 }
     365             :             }
     366           0 :         }
     367             :     }
     368           0 : }
     369             : 
     370             : 
     371             : // What even *is* this?
     372           0 : DelayRateFFTCombo::DelayRateFFTCombo(Array<Complex>& data, Float f0, Float df, Float dt, SDBList& s,
     373           0 :                                      Array<Double>& delayWindow, Array<Double>& rateWindow) :
     374             :     DelayRateFFT(s, -1, delayWindow, rateWindow),
     375           0 :     gm_(s)
     376             : {
     377           0 :     dt_ = dt;
     378           0 :     df_ = df;
     379           0 :     IPosition shape = data.shape();
     380           0 :     nCorr_ = shape(0);
     381           0 :     nElem_ = shape(1);
     382           0 :     nspw_ = shape(2);
     383           0 :     nt_ = shape(3);
     384           0 :     nChan_ = shape(4);
     385           0 :     IPosition dataSize(5, nCorr_, nElem_, nspw_, nt_, nChan_);
     386           0 :     Vall_.resize(dataSize);
     387             :     
     388           0 :     IPosition start(5, 0, 0, 0, 0, 0);
     389           0 :     IPosition stop(5, nCorr_,  nElem_, nspw_, nt_, nChan_);
     390           0 :     IPosition stride(5, 1, 1, 1, 1);
     391           0 :     Slicer target_slice(start, stop, stride, Slicer::endIsLength);
     392           0 :     Vall_(target_slice) = data;
     393             :     
     394           0 :     unitize(Vall_);
     395             : 
     396           0 : }
     397             : 
     398             : void
     399           0 : DelayRateFFTCombo::FFT() {
     400             :     if (DEVDEBUG) {
     401             :         cerr << "DelayRateFFTCombo::FFT()" << endl;
     402             :     }
     403             :     // Axes are 0: correlation (i.e., hand of polarization), 1: antenna, 2: spectral window, 3: time, 4: channel
     404             :     // the machinery is there to say that we only want to FFT the last two axes
     405           0 :     Vector<Bool> ax(5, false);
     406           0 :     ax(3) = true;
     407           0 :     ax(4) = true;
     408             :     // Also copied from DelayFFT in KJones.
     409             :     // we make a copy to FFT in place.
     410           0 :     ArrayLattice<Complex> c(Vall_);
     411           0 :     LatticeFFT::cfft0(c, ax, true);
     412             :     if (DEVDEBUG) {
     413             :         cerr << "FFT transformed" << endl;
     414             :     }
     415           0 : }
     416             : 
     417             : // In the new paradigm, this is where the new stuff happens. Instead of
     418             : // interpolating the peaks on a single big grid, we have to synthesise
     419             : // our estimate from separately FFT-ed spectral windows, using the new
     420             : // off-grid peak formalism
     421             : void
     422           0 : DelayRateFFTCombo::searchPeak() {
     423             :     if (DEVDEBUG) {
     424             :         cerr << "DelayRateFFTCombo::searchPeak()" << endl;
     425             :     }
     426             :     
     427             :     // Recall param_ -> [phase, delay, rate] for each correlation
     428           0 :     param_.resize(3*nCorr_, nElem_); // This might be better done elsewhere.
     429           0 :     param_.set(0.0);
     430           0 :     flag_.resize(3*nCorr_, nElem_);
     431           0 :     flag_.set(true);  // all flagged initially
     432             : 
     433           0 :     Double bw = Float(nChan_)*df_;
     434           0 :     for (Int icorr=0; icorr<nCorr_; ++icorr) {
     435           0 :         flag_(icorr*3 + 0, refant()) = false; 
     436           0 :         flag_(icorr*3 + 1, refant()) = false;
     437           0 :         flag_(icorr*3 + 2, refant()) = false;
     438           0 :         for (Int ielem=0; ielem<nElem_; ++ielem) {
     439           0 :             if (ielem==refant()) {
     440           0 :                 continue;
     441             :             }
     442           0 :             if (activeAntennas_[icorr].find(ielem)==activeAntennas_[icorr].end()) {
     443           0 :                 continue;
     444             :             }
     445             : 
     446             :             // Below is the gory details for turning delay window into index range
     447           0 :             Int sgn = (ielem < refant()) ? 1 : -1;
     448           0 :             Double d0 = sgn*delayWindow_(IPosition(1, 0));
     449           0 :             Double d1 = sgn*delayWindow_(IPosition(1, 1));
     450           0 :             if (d0 > d1) std::swap(d0, d1);
     451           0 :             d0 = max(d0, -0.5/df_);
     452           0 :             d1 = min(d1, (0.5-1/nChan_)/df_);
     453             :     
     454             :             // It's simpler to keep the ranges as signed integers and
     455             :             // handle the wrapping of the FFT in the loop over
     456             :             // indices. Recall that the FFT result returned has indices
     457             :             // that run from 0 to nChan_/2 -1 and then from
     458             :             // -nChan/2 to -1, so far as our delay is concerned.
     459           0 :             Int i0 = bw*d0;
     460           0 :             Int i1 = bw*d1;
     461           0 :             i0 *= nPadFactor_;
     462           0 :             i1 *= nPadFactor_;
     463           0 :             if (i1==i0) i1++;
     464             :             // Now for the gory details for turning rate window into index range
     465           0 :             Double width = nt_*dt_*1e9*f0_;
     466           0 :             Double r0 = sgn*rateWindow_(IPosition(1,0));
     467           0 :             Double r1 = sgn*rateWindow_(IPosition(1,1));
     468           0 :             if (r0 > r1) std::swap(r0, r1);
     469           0 :             r0 = max(r0, -0.5/(dt_*1e9*f0_));
     470           0 :             r1 = min(r1, (0.5 - 1/nt_)/(dt_*1e9*f0_));
     471             :     
     472           0 :             Int j0 = width*r0;
     473           0 :             Int j1 = width*r1;
     474           0 :             j0 *= nPadFactor_;
     475           0 :             j1 *= nPadFactor_;
     476           0 :             if (j1==j0) j1++;
     477             :             Array<Complex> blVis(
     478           0 :                 Vall_(Slicer(
     479           0 :                           IPosition(5, icorr, ielem,     0,   0,      0),
     480           0 :                           IPosition(5,     1,     1, nspw_, nPadT_, nPadChan_),
     481           0 :                           IPosition(5,     1,     1,     1,   1,      1),
     482           0 :                           Slicer::endIsLength)).nonDegenerate(IPosition(1,2)));
     483             : 
     484           0 :             Vector<Float> offsets(nspw_);
     485           0 :             Double bw = gm_.nchan_ * gm_.df_;
     486           0 :             for (size_t lspw=0; lspw!=size_t(nspw_); lspw++) {
     487           0 :                 Double f = gm_.getRefFreqFromLSPW(lspw);
     488           0 :                 Double f0 = gm_.getRefFreqFromLSPW(0);
     489           0 :                 offsets(lspw) = (f - f0)/bw;
     490             :             }
     491             : 
     492             :             // Get the DFT estimates of peak location instead
     493           0 :             tuple<Double, Double, Double, Double> mp = multibandFFTs(blVis , offsets);
     494           0 :             Double mpkt  = std::get<0>(mp);
     495           0 :             Double mpkch = std::get<1>(mp);
     496           0 :             Double mpeak = std::get<2>(mp);
     497           0 :             Double marg  = std::get<3>(mp);
     498             : 
     499           0 :             Complex c0 = dotWithOffsets(blVis, offsets, mpkt, mpkch);
     500             :             if (DEVDEBUG) {
     501             :                 cerr << "DelayRateFFTCombo: abs(c1) " << abs(c0) << endl;
     502             :             }
     503           0 :             tuple<Double, Double, Double, Double> p = refineSearch(blVis, offsets, mpkt, mpkch);
     504           0 :             Double pkt   = std::get<0>(p);
     505           0 :             Double pkch  = std::get<1>(p);
     506           0 :             Double peak  = std::get<2>(p);
     507           0 :             Double phase = std::get<3>(p);
     508             : 
     509             :             if (DEVDEBUG) {
     510             :                 cerr << "[DelayRateFFTCombo::SearchPeak] " << "icorr " << icorr << " ielem " << ielem
     511             :                      << " from (" << mpkt << ", " << mpkch << ", peak "  << abs(c0)
     512             :                      <<  ", angle " << arg(c0) << ")" 
     513             :                      << " to (" << pkt << ", " << pkch << ", peak " << peak << ", angle " << phase <<  ")"
     514             :                      << endl;
     515             :             }
     516           0 :             peak_(IPosition(2, icorr, ielem)) = peak;
     517           0 :             param_(icorr*3 + 0, ielem) = sgn*phase;
     518           0 :             Double delay = (pkch)/Double(nChan_);
     519           0 :             if (delay > 0.5) delay -= 1.0;           // fold
     520           0 :             delay /= df_;                            // nsec
     521           0 :             param_(icorr*3 + 1, ielem) = sgn*delay; 
     522           0 :             Double rate = (pkt)/Double(nt_);
     523           0 :             if (rate > 0.5) rate -= 1.0;
     524           0 :             Double rate0 = rate/dt_;
     525           0 :             Double rate1 = rate0/(1e9 * f0_); 
     526           0 :             param_(icorr*3 + 2, ielem) = Double(sgn*rate1);
     527             :             if (DEVDEBUG) {
     528             :                 cerr << "delay " << sgn*delay << " rate1 " << sgn*rate1 << endl;
     529             :             }
     530             :             // Set 3 flags.
     531           0 :             flag_(icorr*3 + 0, ielem)=false; 
     532           0 :             flag_(icorr*3 + 1, ielem)=false;
     533           0 :             flag_(icorr*3 + 2, ielem)=false;
     534             :             if (DEVDEBUG) {
     535             :                 cerr << "End of this element/correlation combination" << endl;
     536             :                 //cerr << "Set everything " << endl;
     537             :             }
     538           0 :         }
     539             :     }
     540           0 : }
     541             : 
     542             : 
     543             : tuple<Double, Double, Double, Double>
     544           0 : multibandFFTs(const Array<Complex>& ffts, const Vector<Float>& offsets) {
     545             :     if (DEVDEBUG) {
     546             :         cerr << "multibandFFTs() " << endl;
     547             :     }
     548             :     // We take the individual band phases from the peaks of the SPWs
     549             :     // and assume they are the peculiar phases for their SPW; then we
     550             :     // calculate the Direct Discrete FT of those phases on their
     551             :     // offsets and read the peak off of that
     552             : 
     553             :     // This code is based on multibandFFT below, which we were using to
     554             :     // combine the peak values of individual SPW ffts; this version
     555             :     // combines *all* the values, because it turns out we can't
     556             :     // reliably get the peaks first.
     557             :     
     558           0 :     IPosition shp = ffts.shape();
     559           0 :     Int nspw = shp(0);
     560           0 :     Int nt = shp(1);
     561           0 :     Int nchan = shp(2);
     562             : 
     563           0 :     Int binScale = 1;
     564           0 :     size_t nbins= size_t(binScale*max(offsets)+1+0.5);
     565           0 :     Vector<Double> freqs(nbins);
     566           0 :     freqs = 0;
     567           0 :     for (size_t k=0; k!=nbins; k++) {
     568           0 :         freqs[k] = (Double(k) - nbins/2)/float(nbins);
     569             :     }
     570             :     // We need to store out results in an (nt, nchan, nbins) array this time
     571             :     // We do the loop manually because array arithmetic in Casacore is beyond me
     572           0 :     Array<Complex> X(IPosition(3, nbins, nt, nchan));
     573           0 :     X = 0;
     574             : 
     575           0 :     for (size_t ispw=0; ispw!=size_t(nspw); ispw++) {
     576           0 :         for (size_t it=0; it != nt; it++) {
     577           0 :             for (size_t ichan=0; ichan != nchan; ichan++) {
     578           0 :                 Double x = abs(ffts(IPosition(3, ispw, 0, 0)));
     579           0 :                 if (std::isnan(x)) {cerr << "Skipping FFT " << ispw << endl;
     580           0 :                     continue;
     581             :                 }
     582           0 :                 for (size_t ibin=0; ibin!=nbins; ibin++) {
     583           0 :                     Complex rotation = exp(-Complex(0,1)*Complex(C::_2pi*offsets(ispw)*freqs(ibin)));
     584           0 :                     X(IPosition(3, ibin, it, ichan)) += ffts(IPosition(3, ispw, it, ichan))*rotation;
     585             :                 }
     586             :             }
     587             :         }
     588             :     }
     589             :     /* 
     590             :     // New!
     591             :     Array<Complex> rateAveraged(IPosition(2, nbins, nchan));
     592             :     rateAveraged = 0;
     593             :     for (size_t ibin; ibin!=nbins; ibin++) {
     594             :         for (size_t ichan=0; ichan != nchan; ichan++) {
     595             :             for (size_t it=0; it != nt; it++) {
     596             :                 rateAveraged(IPosition(2, ibin, ichan)) += abs(ffts(IPosition(3, ibin, it, ichan)));
     597             :             }
     598             :         }
     599             :     }
     600             :     
     601             :     // Search for a peak with delay in the average first
     602             :     Int ichanmax1 = -1;
     603             :     Int ibinmax1 = -1;
     604             :     Double zmax1 = -1.0;
     605             :     for (Int ichan=0; ichan != nchan; ichan++) {
     606             :         for (Int ibin=0; ibin != nbins; ibin++) {
     607             :             Complex c = rateAveraged(IPosition(2, ibin, ichan));
     608             :             Double a = abs(c);
     609             :             if (a>zmax1) {
     610             :                 ichanmax1 = ichan;
     611             :                 ibinmax1 = ibin;
     612             :                 zmax1 = a;
     613             :             }
     614             :         }
     615             :     }
     616             :     // Then we search the time dimension
     617             :     zmax1 = -1.0;
     618             :     Int itmax1 = -1;
     619             :     for (Int it = 0; it != nt; it++) {
     620             :         Complex c = X(IPosition(3, ibinmax1, it, ichanmax1));
     621             :         Double a = abs(c);
     622             :         if (a>zmax1) {
     623             :             itmax1 = it;
     624             :         }
     625             :     }
     626             :     cerr << "Averaged stacked DFT peaks at " << itmax1 << "/" << nt << ", "
     627             :          << ichanmax1 << "/" << nchan << "; Bin " << ibinmax1
     628             :          << " Peak " << abs(X(IPosition(3, ibinmax1, itmax1, ichanmax1)))
     629             :          << endl;
     630             :  */
     631             :     // We search for the maximum ourself, by brute force.
     632           0 :     Int itmax = -1;
     633           0 :     Int ichanmax = -1;
     634           0 :     Int ibinmax = -1;
     635           0 :     Double zmax = -1.0;
     636           0 :     for (Int ibin=0; ibin != nbins; ibin++) {
     637           0 :         for (Int it = 0; it != nt; it++) {
     638           0 :             for (Int ichan=0; ichan != nchan; ichan++) {
     639           0 :                 Complex c = X(IPosition(3, ibin, it, ichan));
     640           0 :                 Double a = abs(c);
     641           0 :                 if (a>zmax) {
     642           0 :                     itmax = it;
     643           0 :                     ichanmax = ichan;
     644           0 :                     ibinmax = ibin;
     645           0 :                     zmax = a;
     646             :                 }
     647             :             }
     648             :         }
     649             :     }
     650             :     //cerr << "Unaveraged stacked DFT peaks at " << itmax << "/" << nt << ", "
     651             :     //     << ichanmax << "/" << nchan << "; Bin " << ibinmax
     652             :     //     << " Peak " << abs(X(IPosition(3, ibinmax, itmax, ichanmax))) 
     653             :     //     << endl;
     654             :     //for (Int i=0; i < nspw; i++) {
     655             :     //    cerr << "Peak for bin " << i << " = " << abs(ffts(IPosition(3, i, itmax, ichanmax))) << endl;
     656             :     //}
     657             :     
     658           0 :     Complex c = X(IPosition(3, ibinmax, itmax, ichanmax));
     659           0 :     Double dpkch = freqs(ibinmax);
     660             :     
     661           0 :     tuple<Double, Double, Double, Double> t;
     662           0 :     t = std::make_tuple(Double(itmax), Double(ichanmax) + dpkch, abs(c), arg(c));
     663             :     // t = std::make_tuple(Double(itmax), Double(ichanmax), abs(c), arg(c));
     664             :     if (DEVDEBUG) {
     665             :         cerr << "itmax " << itmax << " ichanmax " << ichanmax << " ibinmax " << ibinmax
     666             :              << " peak " << abs(c) << endl;
     667             :         cerr << "MultibandFFTs dpkch=" << dpkch << endl;
     668             :         cerr << "multibandFFTs() finished" << endl;
     669             :     }
     670           0 :     return t;
     671           0 : }
     672             :     
     673             : Double
     674           0 : multibandFFT(const Vector<Complex>& peaks, const Vector<Float>& offsets) {
     675             :     // We take the individual band phases from the peaks of the SPWs
     676             :     // and assume they are the peculiar phases for their SPW; then we
     677             :     // calculate the Direct Discrete FT of those phases on their
     678             :     // offsets and read the peak off of that
     679           0 :     Int nspw = peaks.size();
     680             : 
     681           0 :     Vector<Float> amps( amplitude(peaks) );
     682             :     Int len;
     683           0 :     amps.shape(len);
     684             : 
     685           0 :     Float eps = 1e-10;
     686           0 :     std::set<size_t> badInds;
     687           0 :     for (size_t i=0; i!=size_t(len); i++) {
     688           0 :         if (amps[i] < eps) {
     689             :             if (DEVDEBUG) {
     690             :                 cerr << " blacklisting " << i << endl;
     691             :             }
     692           0 :             badInds.insert(i);
     693             :         }
     694             :     }
     695           0 :     Vector<Complex> phasors( peaks/amplitude(peaks) );
     696             : 
     697             :     // I want to print angles but I can't get the API to do that
     698             :     // in a reasonable way; arg(phasors) doesn't work
     699             :     // Vector<Float> angles ( arg(phasors) );
     700             :     if (DEVDEBUG) {
     701             :         cerr << "phasors " << phasors << endl;
     702             :     }
     703             :     // FIXME: the last offset is at the *beginning* of the subband!
     704             :     // size_t nbins=2*size_t(max(offsets)+1+0.5);
     705           0 :     size_t nbins= size_t(max(offsets)+1+0.5);
     706             : 
     707           0 :     Vector<Float> freqs(nbins);
     708           0 :     freqs = 0;
     709           0 :     for (size_t k=0; k!=nbins; k++) {
     710           0 :         freqs[k] = (Float(k) - nbins/2)/float(nbins);
     711             :     }
     712           0 :     Vector<Complex> X(nbins);
     713           0 :     X = 0;
     714           0 :     for (size_t n=0; n!=size_t(nspw); n++) {
     715           0 :         if (badInds.find(n) != badInds.end()) {
     716             :             if (DEVDEBUG) {
     717             :                 cerr << " skipping " << n << endl;
     718             :             }
     719           0 :             continue;
     720             :         }
     721           0 :         for (size_t k=0; k!=nbins; k++) {
     722             :             // FIXME: also swap signs here!
     723           0 :             X[k] += peaks[n]*exp(-Complex(0,1)*Complex(C::_2pi*offsets(n)*freqs(k)));
     724             :             // X[k] += phasors[n]*exp(-Complex(0,1)*Complex(C::_2pi*offsets(n)*freqs(k)));
     725             :         }
     726             :     }
     727           0 :     Int k_max = -1;
     728           0 :     Float zmax = -1.0;
     729           0 :     for (size_t k=0; k!=nbins; k++) {
     730           0 :         Complex z = X[k];
     731           0 :         Float a = abs(z);
     732             :         if (DEVDEBUG) {
     733             :             cerr << "DFT Bin " << k << " freq " << freqs(k) << " coeff " << a << endl;
     734             :         }
     735           0 :         if (a>zmax) {
     736           0 :             k_max = k;
     737           0 :             zmax = a;
     738             :         }
     739             :     }
     740             :     if (DEVDEBUG) {
     741             :         cerr << "k_max = " << k_max 
     742             :              << "; peak at k_max= " << abs(X[k_max]) 
     743             :              << "; freqs(k_max) = " << freqs(k_max) << endl;
     744             :     }
     745           0 :     return freqs(k_max);
     746           0 : }
     747             : 
     748             : // Auxilliary function
     749             : Complex
     750           0 : dotWithOffsets(const Cube<Complex>& ft, const Vector<Float>& offsets, double k, double l) {
     751           0 :     Int nspw = ft.nrow();
     752           0 :     Int ni = ft.ncolumn();
     753           0 :     Int nj = ft.nplane();
     754             : 
     755           0 :     if (k<0) k += (ni);
     756           0 :     if (l<0) l += (nj);
     757           0 :     Complex p(0.0, 0.0);
     758           0 :     for (Int s=0; s!=nspw; s++) {
     759           0 :         const Matrix<Complex>& ft_s = ft.yzPlane(s);
     760           0 :         Double f_off = offsets(s);
     761           0 :         Complex r = dotMatrixWithModel(ft_s, k, l, f_off);
     762           0 :         p += r;
     763           0 :     }    
     764           0 :     return p;
     765             : }
     766             : 
     767             : // We need a function to minimize, which has to return a real value,
     768             : // but when we're done we need to find the peak and also its argument,
     769             : // so we factor out the complex version...
     770             : Complex
     771           0 : c_peak_fn(const gsl_vector *x, void *vparams) {
     772           0 :     std::pair< Cube<Complex> const *, Vector<Float> const * > *params =
     773             :         (std::pair< Cube<Complex> const *, Vector<Float> const * > *)(vparams );
     774             : 
     775           0 :     double k = gsl_vector_get(x, 0);
     776           0 :     double l = gsl_vector_get(x, 1);
     777             : 
     778           0 :     const Cube<Complex>& ft ( *(params->first ));
     779           0 :     const Vector<Float>& offsets ( *(params->second));
     780           0 :     return dotWithOffsets(ft, offsets, k, l);
     781             : }
     782             : 
     783             : // ... and then call it from a real version that we can give to the
     784             : // optimizer (with a sign change; by tradition these are minimizer
     785             : // routines and we want a maximum)
     786             : double
     787           0 : my_peak_fn(const gsl_vector *x, void *vparams) {
     788           0 :     Complex p = c_peak_fn(x, vparams);
     789           0 :     return -abs(p);
     790             : }
     791             : 
     792             : tuple<Double, Double, Double, Double>
     793           0 : bruteForceDelay(const Cube<Complex>& ft,  const Vector<Float>& offsets, Int ipkt, Int ipkch) {
     794           0 :     Double k(ipkt);
     795           0 :     Double l(ipkch);
     796           0 :     Int n = 10;
     797             : 
     798           0 :     Double l_p = l;
     799           0 :     Complex peak = Complex(0, 0);
     800           0 :     for (Int i=-n; i!=n+1; i++) {
     801           0 :         Double dch = double(i)/(2*n);
     802           0 :         Complex p = dotWithOffsets(ft, offsets, k, l+dch);
     803           0 :         if (abs(p) > abs(peak)) {
     804             :             if (DEVDEBUG) {
     805             :                 cerr << "[bruteForceDelay] l " << l+dch << " peak " << abs(p) << endl;
     806             :             }
     807           0 :             peak = p;
     808           0 :             l_p = l+dch;
     809             :             
     810             :         }
     811             :     }            
     812           0 :     tuple<Double, Double, Double, Double> t;
     813           0 :     t = std::make_tuple(k, l_p, abs(peak), arg(peak));
     814           0 :     return t;
     815             : }
     816             : 
     817             : tuple<Double, Double, Double, Double>
     818           0 : DelayRateFFTCombo::refineSearch(const Cube<Complex>& ft,  const Vector<Float>& offsets, Double pkt, Double pkch) {
     819             :     // small@jive.eu: I borrowed most of this code from the GSL documentation of multimin:
     820             :     // <https://www.gnu.org/software/gsl/doc/html/multimin.html>
     821             :     if (DEVDEBUG) {
     822             :         cerr << "refineSearch" << endl;
     823             :     }
     824           0 :     const gsl_multimin_fminimizer_type *T = gsl_multimin_fminimizer_nmsimplex2;
     825             :     /* Starting point */
     826           0 :     gsl_vector *x = gsl_vector_alloc (2);
     827           0 :     gsl_vector_set(x, 0, pkt);
     828           0 :     gsl_vector_set(x, 1, pkch);
     829             :     if (DEVDEBUG) {
     830             :         cerr << "pkch starts at " << pkch << endl;
     831             :     }
     832             :     /* Set initial step sizes */
     833             :     /* Fixme! I need to think harder about this value! */
     834           0 :     gsl_vector* steps = gsl_vector_alloc (2);
     835           0 :     gsl_vector_set_all(steps, 0.01);
     836             :     
     837             :     /* Initialize method and iterate */
     838             :     
     839             :     gsl_multimin_function minex_func;
     840           0 :     std::pair< Cube<Complex> const * , Vector<Float> const * > par = make_pair(&ft, &offsets);
     841             :      
     842           0 :     minex_func.n = 2;
     843           0 :     minex_func.f = my_peak_fn;
     844           0 :     minex_func.params = &par;
     845             :      
     846           0 :     gsl_multimin_fminimizer *s = gsl_multimin_fminimizer_alloc(T, 2);
     847           0 :     gsl_multimin_fminimizer_set (s, &minex_func, x, steps);
     848             : 
     849           0 :     float minstep = 1e-5;
     850             :     int status;
     851           0 :     size_t iter = 0;
     852             :     do {
     853             :         if (DEVDEBUG) {
     854             :             cerr << "Starting iteration" << endl;
     855             :         }
     856             :         
     857           0 :         iter++;
     858           0 :         status = gsl_multimin_fminimizer_iterate(s);
     859           0 :         if (status) break;
     860           0 :         double size = gsl_multimin_fminimizer_size(s);
     861           0 :         status = gsl_multimin_test_size(size, minstep);
     862           0 :         if (status == GSL_SUCCESS) {
     863           0 :             printf ("converged to minimum at\n");
     864             :         }
     865             :         if (DEVDEBUG) {
     866             :             printf("%5zd ipkt %12.5e ipkch %12.5e f() = %12.5g size = %12.5f\n",
     867             :                    iter,
     868             :                    gsl_vector_get(s->x, 0),
     869             :                    gsl_vector_get(s->x, 1),
     870             :                    s->fval,
     871             :                    size);
     872             :         }
     873           0 :     } while (status == GSL_CONTINUE && iter < 5);
     874             :     // while (status == GSL_CONTINUE && iter < 100);
     875           0 :     tuple<Double, Double, Double, Double> p;
     876             :     // if (status == GSL_SUCCESS) {
     877             :     if (1) {
     878             :         // Double peak = gsl_multimin_fminimizer_minimum(s);
     879           0 :         Double ipkt  = gsl_vector_get(s->x, 0);
     880           0 :         Double ipkch = gsl_vector_get(s->x, 1);
     881           0 :         Complex c = c_peak_fn(s->x, &par);
     882           0 :         p = std::make_tuple(ipkt, ipkch, abs(c), arg(c));
     883             :     } else {
     884             :         p = std::make_tuple(pkt, pkch, 0.0, 0.0);
     885             :     }
     886           0 :     gsl_vector_free(steps);
     887           0 :     gsl_vector_free(x);
     888           0 :     gsl_multimin_fminimizer_free(s);
     889             :      
     890           0 :     return p;
     891             : }
     892             : 
     893             : Float
     894           0 : DelayRateFFTCombo::snr(Int icorr, Int ielem, Float delay, Float rate) {
     895             :     // We calculate a signal-to-noise ration for the 2D FFT fringefit
     896             :     // using a formula transcribed from AIPS FRING.
     897             :     //
     898             :     // Have to convert delay and rate back into indices on the padded 2D grid.
     899             : 
     900           0 :     IPosition p(2, icorr, ielem);
     901           0 :     Float peak = peak_(p);
     902           0 :     if (peak > 0.999*sumw_(p)) {
     903             :       //cerr << "Clipping peak for element " << ielem << " from " << peak << " to " << 0.999*sumw_(p) << endl;
     904           0 :         peak=0.999*sumw_(p);
     905             :     }
     906             :     // xcount is number of data points for baseline to ielem
     907             :     // sumw is sum of weights,
     908             :     // sumww is sum of squares of weights
     909             :     Float cwt;
     910           0 :     if (fabs(sumw_(p))<FLT_EPSILON) {
     911           0 :         cwt = 0;
     912             :         if (DEVDEBUG) {
     913             :             cerr << "Correlation " << icorr << " antenna " << ielem << ": sum of weights is zero." << endl;
     914             :         }
     915             :     } else {
     916           0 :         Float x = C::pi/2*peak/sumw_(p);
     917             :         // The magic numbers in the following formula are from AIPS FRING
     918           0 :         cwt = (pow(tan(x), 1.163) * sqrt(sumw_(p)/sqrt(sumww_(p)/xcount_(p))));
     919             :         if (DEVDEBUG) {
     920             :             cerr << "Correlation " << icorr << " antenna " << ielem 
     921             :                  << " peak=" << peak << "; xang=" << x << "; xcount=" << xcount_(p) << "; sumw=" << sumw_(p) << "; sumww=" << sumww_(p)
     922             :                  << " snr " << cwt << endl;
     923             :         }
     924             :     }
     925           0 :     return cwt;
     926           0 : }
     927             :     
     928             : 
     929             : // There isn't a usable sinc lying around that I can see
     930           0 : Double sinc(Double x) {
     931           0 :     Double p = C::pi*x;
     932           0 :     return sin(p)/p;
     933             : }
     934             : 
     935             : // This implements a Fancy Sinc strategy for finding the peak of a
     936             : // single SPW. I no longer know if this can be extended to multiple
     937             : // SPWs.
     938             : tuple<Double, Double, Double, Double>
     939           0 : refineSearch2(const Cube<Complex>& ft, const Vector<Float>& offsets, Int ipkt0, Int ipkch0) {
     940             :     Double imax, jmax;
     941             :     // FIXME! We need to do a single SPW first
     942           0 :     if (ft.nrow() > 1) {
     943           0 :         throw AipsError("Only one spw for now");
     944             :     }
     945           0 :     Int s = 0;
     946           0 :     const Matrix<Complex>& ft_s(ft.yzPlane(s));
     947           0 :     size_t ni = ft_s.nrow();
     948           0 :     size_t nj = ft_s.ncolumn();
     949             :     // we need to do a roll left/down and sum but Casacore doesn't have a matrix roll so we may as well do it by hand
     950           0 :     Float pmax = -1;
     951           0 :     size_t ipkt(0);
     952           0 :     size_t ipkch(0);
     953           0 :     Matrix<Float> sumAbs(ni, nj);
     954           0 :     for (size_t i=0; i!=ni; i++) {
     955             :         // FIXME: The variable i1 is unused!
     956             :         // Either use it or lose it!
     957             :         // size_t i1 = (i==ni-1) ? 0 : i+1;
     958           0 :         for (size_t j=0; j!=nj; j++) {
     959           0 :             size_t j1 = (j==nj-1) ? 0 : j+1;
     960           0 :             Float sumAbs = (abs(ft_s(i, j)) + abs(ft_s(i, j1)));
     961           0 :             if ((sumAbs) > pmax) {
     962           0 :                 ipkt = i;
     963           0 :                 ipkch = j;
     964           0 :                 pmax = sumAbs;
     965             :             }
     966             :         }
     967             :     }
     968           0 :     Double y0 = abs(ft_s(ipkt, ipkch));
     969           0 :     Double y1 = abs(ft_s(ipkt, (ipkch==nj-1)? 0 : ipkch+1));
     970           0 :     Double ym1 = abs(ft_s(ipkt, (ipkch==0)? nj-1 : ipkch-1));
     971             : 
     972           0 :     Double z1 =  abs(ft_s((ipkt==ni-1) ? 0 : ipkt+1, ipkch));
     973           0 :     Double zm1 =  abs(ft_s((ipkt==0) ? ni-1 : ipkt-1, ipkch));
     974             :     
     975             :     
     976           0 :     Double d0 = y1/(y0+y1);
     977           0 :     Double peak( y0/sinc(d0) );
     978             :     if (DEVDEBUG) {
     979             :         cerr << "y0 " << y0 << " y1 " << y1 << " ym1 " << ym1 << " d0 " << d0 << " sinc(d0) "
     980             :              << sinc(d0) << " peak " << peak << endl;
     981             :         cerr << "y0 " << y0 << " z1 " << z1 << " zm1 " << zm1 << endl;
     982             :         cerr << real(abs(ft_s.row(ipkt)))/y0 << endl;
     983             :     }
     984           0 :     Double phase( 0.0 );
     985           0 :     imax = Double(ipkt);
     986           0 :     jmax = Double(ipkch) + d0;
     987           0 :     tuple<Double, Double, Double, Double> p = std::make_tuple(imax, jmax, peak, phase);
     988           0 :     return p;
     989           0 : }
     990             : 
     991             : Complex
     992           0 : dotMatrixWithModel(const Matrix<Complex>& data, Double k, Double l, Double offset)
     993             : {
     994           0 :     size_t ni = data.nrow();
     995           0 :     size_t nj = data.ncolumn();
     996           0 :     Matrix<Complex> model(ni, nj);
     997             : 
     998             :     // Note that k is the time index of the array, l is the frequency index.
     999             :     // Offsets correspond to frequencies
    1000           0 :     Double eps = 1e-8;
    1001           0 :     Int k_int = floor(k);
    1002           0 :     Double k_del = k - k_int;
    1003           0 :     Bool k_flag =  (fabs(k_del) < eps);
    1004           0 :     Int l_int = floor(l);
    1005           0 :     Double l_del = l - l_int;
    1006           0 :     Bool l_flag = (fabs(l_del) < eps);
    1007           0 :     Complex t0;
    1008           0 :     Complex t1;
    1009             :     // FIXME! We're messing with reversing signs
    1010           0 :     for (size_t i=0; i!=ni; i++) {
    1011           0 :         if (k_flag) {
    1012           0 :             t0 = Complex(i==k);
    1013             :         } else { // if k isn't an integer!
    1014           0 :             t0 = ( (1-exp(Complex(0, -1.0*C::_2pi*(k-i))))/
    1015           0 :                    (1-exp(Complex(0, -1.0*C::_2pi*(k-i)/Double(ni)))) );
    1016           0 :             t0 /= ni;
    1017             :         }
    1018           0 :         for (size_t j=0; j!=nj; j++) {
    1019           0 :             if (l_flag) {
    1020           0 :                 t1 = Complex(j==l);
    1021             :             } else { // if l isn't an integer!
    1022           0 :                 t1 = ( (1-exp(Complex(0, -1.0*C::_2pi*(l-j))))/
    1023           0 :                        (1-exp(Complex(0, -1.0*C::_2pi*(l-j)/Double(nj)))) );
    1024           0 :                 t1 /= nj;
    1025             :             }
    1026           0 :             model(i, j) = exp(Complex(0, -1.0*C::_2pi*offset*l_del))*t0*t1;
    1027             :         }
    1028             :     }
    1029           0 :     Complex t2 = sum(data*model);
    1030           0 :     return t2;
    1031           0 : }
    1032             : 
    1033             : 
    1034             : Complex
    1035           0 : dotMatrixWithModel2(const Matrix<Complex>& data, Double k0, Double l0, Double offset)
    1036             : {
    1037           0 :     size_t ni = data.nrow();
    1038           0 :     size_t nj = data.ncolumn();
    1039           0 :     Matrix<Complex> model(ni, nj);
    1040           0 :     model = 0.0;
    1041             :     
    1042           0 :     Double eps = 1e-8;
    1043           0 :     Int k_int = Int(ceil(k0));
    1044           0 :     Int l_int = Int(ceil(l0));
    1045             :     
    1046           0 :     Double d0 = k_int - k0;
    1047           0 :     Double d1 = l_int - l0;
    1048             : 
    1049           0 :     Bool k_flag = (fabs(d0) < eps);
    1050           0 :     Bool l_flag = (fabs(d1) < eps);
    1051             : 
    1052           0 :     Complex c0, c1;
    1053             : 
    1054           0 :     Complex s (0.0);
    1055           0 :     Complex t0 = Complex(sin(C::pi*d0)/C::pi)*exp(Complex(0, C::pi*d0));
    1056           0 :     Complex t1 = Complex(sin(C::pi*d1)/C::pi)*exp(Complex(0, C::pi*d1));
    1057           0 :     for (Int dk=-2; dk!=2; dk++) {
    1058           0 :         size_t k = (k_int + dk + ni) % ni;
    1059           0 :         if (k_flag) {
    1060           0 :             c0 = Complex(dk==0);
    1061             :         } else { // if k isn't an integer!
    1062           0 :             c0 = Complex(1/(d0+dk))*t0;
    1063             :         }
    1064           0 :         for (Int dl=-2; dl!=2; dl++) {
    1065           0 :             size_t l = (l_int+dl+nj) % nj;
    1066           0 :             if (l_flag) {
    1067           0 :                 c1 = Complex(dl==0);
    1068             :             } else { // if l isn't an integer!
    1069           0 :                 c1 = Complex(1/(d1+dl))*t1;
    1070             :             }
    1071           0 :             Complex d = data(k, l);
    1072           0 :             Complex m = c0*c1;
    1073           0 :             s += d*m;
    1074             :         }
    1075             :     }
    1076           0 :     return s;
    1077           0 : }
    1078             : 
    1079             : 
    1080           0 : SDBListGridManagerConcat::SDBListGridManagerConcat(SDBList& sdbs) :
    1081           0 :     SDBListGridManager(sdbs)
    1082             : {
    1083           0 :     std::set<Double> fmaxes;
    1084           0 :     std::set<Double> fmins;
    1085           0 :     Float dfn(0.0);
    1086           0 :     Int totalChans0(0) ;
    1087           0 :     Int nchan(0);
    1088             : 
    1089           0 :     if (sdbs_.nSDB()==0) {
    1090             :         // The for loop is fine with an empty list, but code below it
    1091             :         // isn't and there's nothing to lose by bailing early!
    1092           0 :         return;
    1093             :     }
    1094             :         
    1095           0 :     for (Int i=0; i != sdbs_.nSDB(); i++) {
    1096           0 :         SolveDataBuffer& sdb = sdbs_(i);
    1097           0 :         Int spw = sdb.spectralWindow()(0);
    1098           0 :         Double t = sdbs_(i).time()(0);
    1099           0 :         times_.insert(t); 
    1100           0 :         if (spwins_.find(spw) == spwins_.end()) {
    1101           0 :             spwins_.insert(spw);
    1102           0 :             const Vector<Double>& fs = sdb.freqs();
    1103           0 :             spwIdToFreqMap_[spw] = &(sdb.freqs());
    1104           0 :             nchan = sdb.nChannels();
    1105           0 :             fmaxes.insert(fs(nchan-1));
    1106           0 :             fmins.insert(fs(0));
    1107             :             // We assume they're all at the same time.
    1108             : 
    1109           0 :             totalChans0 += nchan;
    1110           0 :             Float df0 = fs(1) - fs(0);
    1111           0 :             dfn = (fs(nchan-1) - fs(0))/(nchan-1);
    1112             :             if (DEVDEBUG) {
    1113             :                 cerr << "Spectral window " << spw << " has " << nchan << " channels" << endl;
    1114             :                 cerr << "df0 "<< df0 << "; " << "dfn " << dfn << endl;
    1115             :             }
    1116             :         } else {
    1117           0 :             continue;
    1118             :         }
    1119             :     }
    1120             :     // nt_ = sdbs_.nSDB()/spwins_.size();
    1121           0 :     nt_ = times_.size();
    1122           0 :     tmin_ = *(times_.begin());
    1123           0 :     tmax_ = *(times_.rbegin());
    1124           0 :     dt_ = (tmax_ - tmin_)/(nt_ - 1);
    1125           0 :     nSPWChan_ = nchan;
    1126           0 :     if (nchan == 1) {
    1127           0 :       nSPWChan_ = nchan;
    1128           0 :       fmin_ = *(fmins.begin());
    1129           0 :       fmax_ = fmin_;
    1130           0 :       totalChans_ = 1;
    1131           0 :       df_ = 1;
    1132           0 :       return;
    1133             :     }
    1134             :       
    1135           0 :     fmin_ = *(fmins.begin());
    1136           0 :     fmax_ = *(fmaxes.rbegin());
    1137           0 :     totalChans_ = round((fmax_ - fmin_)/dfn + 1);
    1138           0 :     df_ = (fmax_ - fmin_)/(totalChans_-1);
    1139             :     if (DEVDEBUG) {
    1140             :         cerr << "Global fmin " << fmin_ << " global max " << fmax_ << endl;
    1141             :         cerr << "nt " << nt_ << " dt " << dt_ << endl;
    1142             :         cerr << "tmin " << tmin_ << " tmax " << tmax_ << endl;
    1143             :         cerr << "Global dt " << tmax_ - tmin_ << endl;
    1144             :         cerr << "Global df " << df_ << endl;
    1145             :         cerr << "I guess we'll need " << totalChans_ << " freq points in total." << endl;
    1146             :         cerr << "Compared to " << totalChans0 << " with simple-minded concatenation." << endl;
    1147             :         cerr << "spwins_.size() " << spwins_.size() << endl;
    1148             :         cerr << "nSPW() " << nSPW() << endl;
    1149             :     }
    1150           0 : }
    1151             : 
    1152             : // checkAllGridPoints is a diagnostic funtion that should not be called
    1153             : // in production releases, but it doesn't do any harm to have it
    1154             : // latent.
    1155             : void
    1156           0 : SDBListGridManagerConcat::checkAllGridpoints() {
    1157           0 :     map<Int, Vector<Double> const *>::iterator it;
    1158           0 :     for (it = spwIdToFreqMap_.begin(); it != spwIdToFreqMap_.end(); it++) {
    1159           0 :         Int spwid = it->first;
    1160           0 :         Vector<Double> const* fs = it->second;
    1161             :         Int length;
    1162           0 :         fs->shape(length);
    1163           0 :         for (Int i=0; i!=length; i++) {
    1164           0 :             Double f = (*fs)(i);
    1165           0 :             Int j = bigFreqGridIndex(f);
    1166             :             if (DEVDEBUG) {
    1167             :                 cerr << "spwid, i = (" << spwid << ", " << i << ") => " << j << " (" << f << ")" << endl;
    1168             :             }
    1169             :         }
    1170             :     }
    1171             :     if (DEVDEBUG) {
    1172             :         cerr << "[1] spwins.size() " << nSPW() << endl;
    1173             :     }
    1174           0 : }
    1175             : 
    1176             : Int
    1177           0 : SDBListGridManagerConcat::swStartIndex(Int spw) {
    1178           0 :     Vector<Double> const* fs = spwIdToFreqMap_[spw];
    1179           0 :     Double f0 = (*fs)(0);
    1180           0 :     return bigFreqGridIndex(f0);
    1181             : }
    1182             :    
    1183           0 : DelayRateFFTConcat::DelayRateFFTConcat(SDBList& sdbs, Int refant, Array<Double>& delayWindow,
    1184           0 :                                        Array<Double>& rateWindow) :
    1185             :     DelayRateFFT(sdbs, refant, delayWindow, rateWindow),
    1186           0 :     gm_(sdbs)
    1187             : {
    1188             :     if (DEVDEBUG) {
    1189             :         gm_.describe();
    1190             :         cerr << "gm_.nSPW() " << gm_.nSPW() << endl;
    1191             :     }
    1192           0 :     nPadFactor_= max(2, 8  / gm_.nSPW());
    1193           0 :     nt_ = gm_.nt_;
    1194           0 :     nPadT_ = nPadFactor_ * nt_;
    1195           0 :     nChan_ = gm_.nChannels();
    1196           0 :     nPadChan_ = nPadFactor_*nChan_;
    1197           0 :     dt_ = gm_.dt_;
    1198           0 :     f0_ = sdbs.centroidFreq() / 1.e9;      // GHz, for delayrate calc
    1199           0 :     df_ = gm_.df_ / 1.e9;
    1200           0 :     df_all_ = gm_.fmax_ - gm_.fmin_;
    1201             :     // This check should be commented out in production:
    1202             :     // gm_.checkAllGridpoints();
    1203           0 :     if (nt_ < 2) {
    1204           0 :         throw(AipsError("Can't do a 2-dimensional FFT on a single timestep! Please consider changing solint to avoid orphan timesteps."));
    1205             :     }
    1206           0 :     IPosition ds = delayWindow_.shape();
    1207           0 :     if (ds.size()!=1 || ds.nelements() != 1) {
    1208           0 :         throw AipsError("delaywindow must be a list of length 2.");
    1209             :     }
    1210           0 :     IPosition rs = rateWindow_.shape();
    1211           0 :     if (rs.size()!=1 || rs.nelements() != 1) {
    1212           0 :         throw AipsError("ratewindow must be a list of length 2.");
    1213             :     }
    1214           0 :     Int nCorrOrig(sdbs(0).nCorrelations());
    1215           0 :     nCorr_ = (nCorrOrig> 1 ? 2 : 1); // number of p-hands
    1216           0 :     for (Int i=0; i<nCorr_; i++) {
    1217           0 :         activeAntennas_[i].insert(refant_);
    1218             :     }
    1219             :     
    1220             :     // when we get the visCubecorrected it is already
    1221             :     // reduced to parallel hands, but there isn't a
    1222             :     // corresponding method for flags.
    1223           0 :     Int corrStep = (nCorrOrig > 2 ? 3 : 1); // step for p-hands
    1224           0 :     for (Int ibuf=0; ibuf != sdbs.nSDB(); ibuf++) {
    1225           0 :         SolveDataBuffer& s(sdbs(ibuf));
    1226           0 :         for (Int irow=0; irow!=s.nRows(); irow++) {
    1227           0 :             Int a1(s.antenna1()(irow)), a2(s.antenna2()(irow));
    1228           0 :             allActiveAntennas_.insert(a1);
    1229           0 :             allActiveAntennas_.insert(a2);
    1230             :         }
    1231             :     }
    1232           0 :     nElem_ =  1 + *(allActiveAntennas_.rbegin()) ;
    1233           0 :     IPosition aggregateDim(2, nCorr_, nElem_);
    1234           0 :     xcount_.resize(aggregateDim);
    1235           0 :     sumw_.resize(aggregateDim);
    1236           0 :     sumww_.resize(aggregateDim);
    1237             : 
    1238           0 :     xcount_ = 0;
    1239           0 :     sumw_ = 0.0;
    1240           0 :     sumww_ = 0.0;
    1241             :     
    1242             :     if (DEVDEBUG) {
    1243             :         cerr << "Filling FFT grid with " << sdbs.nSDB() << " data buffers." << endl;
    1244             :     }
    1245           0 :     IPosition paddedDataSize(4, nCorr_, nElem_, nPadT_, nPadChan_);
    1246           0 :     Vpad_.resize(paddedDataSize);
    1247           0 :     Int totalRows = 0;
    1248           0 :     Int goodRows = 0;
    1249             : 
    1250             :     
    1251           0 :     for (Int ibuf=0; ibuf != sdbs.nSDB(); ibuf++) {
    1252           0 :         SolveDataBuffer& s(sdbs(ibuf));
    1253           0 :         totalRows += s.nRows();
    1254           0 :         if (!s.Ok())
    1255           0 :             continue;
    1256             : 
    1257           0 :         Int nr = 0;
    1258           0 :         for (Int irow=0; irow!=s.nRows(); irow++) {
    1259           0 :             if (s.flagRow()(irow))
    1260           0 :                 continue;
    1261             :             Int iant;
    1262           0 :             Int a1(s.antenna1()(irow)), a2(s.antenna2()(irow));
    1263           0 :             if (a1 == a2) {
    1264           0 :                 continue;
    1265             :             }
    1266           0 :             else if (a1 == refant_) {
    1267           0 :                 iant = a2;
    1268             :             }
    1269           0 :             else if (a2 == refant_) {
    1270           0 :                 iant = a1;
    1271             :             }
    1272             :             else {
    1273           0 :                 continue;
    1274             :             }
    1275             :             // OK, we're not skipping this one so we have to do something.
    1276             : 
    1277             :             // v has shape (nelems, ?, nrows, nchannels)
    1278           0 :             Cube<Complex> v = s.visCubeCorrected();
    1279           0 :             Cube<Float> w = s.weightSpectrum();
    1280           0 :             Cube<Bool> fl = s.flagCube();
    1281           0 :             Int spw = s.spectralWindow()(0);
    1282           0 :             Int f_index = gm_.swStartIndex(spw);    // ditto!
    1283           0 :             Int t_index = gm_.getTimeIndex(s.time()(0));
    1284           0 :             Int spwchans = gm_.nSPWChan_;
    1285           0 :             IPosition start(4,      0,      iant, t_index, f_index);
    1286           0 :             IPosition stop(4,      nCorr_,    1,       1, spwchans);
    1287           0 :             IPosition stride(4,      1,         1,       1, 1);
    1288           0 :             Slicer sl1(start,     stop, stride, Slicer::endIsLength);
    1289           0 :             Slicer sl2(IPosition(3, 0,         0, irow),
    1290           0 :                        IPosition(3, nCorr_, spwchans, 1),
    1291           0 :                        IPosition(3, corrStep,        1,  1), Slicer::endIsLength);
    1292             :                 
    1293           0 :             Slicer flagSlice(IPosition(3, 0,         0, irow),
    1294           0 :                              IPosition(3, nCorr_, spwchans, 1),
    1295           0 :                              IPosition(3, corrStep,        1, 1), Slicer::endIsLength);
    1296           0 :             nr++;
    1297             :             if (DEVDEBUG && false) {
    1298             :                 cerr << "nr " << nr
    1299             :                      << " irow " << endl
    1300             :                      << "Vpad shape " << Vpad_.shape() << endl
    1301             :                      << "v shape " << v.shape() << endl
    1302             :                      << "sl2 " << sl2 << endl
    1303             :                      << "sl1 " << sl1 << endl
    1304             :                      << "flagSlice " << flagSlice << endl;
    1305             :             }
    1306           0 :             Array<Complex> rhs = v(sl2).nonDegenerate(1);
    1307           0 :             Array<Float> weights = w(sl2).nonDegenerate(1);
    1308             :                 
    1309           0 :             unitize(rhs);
    1310           0 :             Vpad_(sl1).nonDegenerate(1) = rhs * weights;
    1311             : 
    1312           0 :             Array<Bool> flagged(fl(flagSlice).nonDegenerate(1));
    1313             :             // Zero flagged entries.
    1314           0 :             Vpad_(sl1).nonDegenerate(1)(flagged) = Complex(0.0);
    1315             : 
    1316           0 :             if (!allTrue(flagged)) {
    1317           0 :                 for (Int icorr=0; icorr<nCorr_; ++icorr) {
    1318           0 :                     IPosition p(2, icorr, iant);
    1319           0 :                     Bool actually = false;
    1320           0 :                     activeAntennas_[icorr].insert(iant);
    1321           0 :                     for (Int ichan=0; ichan != (Int) spwchans; ichan++) {
    1322           0 :                         IPosition pchan(2, icorr, ichan);
    1323           0 :                         if (!flagged(pchan)) {
    1324           0 :                             Float wv = weights(pchan);
    1325           0 :                             if (wv < 0) {
    1326           0 :                                 cerr << "spwchans " << spwchans << endl;
    1327           0 :                                 cerr << "Negative weight << (" << wv << ") on row "
    1328           0 :                                      << irow << " baseline (" << a1 << ", " << a2 << ") "
    1329           0 :                                      << " channel " << ichan << endl;
    1330           0 :                                 cerr << "pchan " << pchan << endl;
    1331           0 :                                 cerr << "Weights " << weights << endl;
    1332             :                             }
    1333           0 :                             xcount_(p)++;
    1334           0 :                             sumw_(p) += wv;
    1335           0 :                             sumww_(p) += wv*wv;
    1336           0 :                             actually = true;
    1337             :                         }
    1338           0 :                     }
    1339           0 :                     if (actually) {
    1340           0 :                         activeAntennas_[icorr].insert(iant);
    1341           0 :                         goodRows++;
    1342             :                     }
    1343           0 :                 }
    1344             :             }                
    1345             :             if (DEVDEBUG && 0) {
    1346             :                 cerr << "flagged " << flagged << endl;
    1347             :                 cerr << "flagSlice " << flagSlice << endl
    1348             :                      << "fl.shape() " << fl.shape() << endl
    1349             :                      << "Vpad_.shape() " << Vpad_.shape() << endl
    1350             :                      << "flagged.shape() " << flagged.shape() << endl
    1351             :                      << "sl1 " << sl1 << endl;
    1352             :             }
    1353           0 :         }
    1354             :     }
    1355             :     if (DEVDEBUG) {
    1356             :         cerr << "In DelayRateFFTConcat::DelayRateFFTConcat " << endl;
    1357             :         printActive();
    1358             :         cerr << "sumw_ " << sumw_ << endl;
    1359             :         cerr << "Constructed a DelayRateFFTConcat object." << endl;
    1360             :         cerr << "totalRows " << totalRows << endl;
    1361             :         cerr << "goodRows " << goodRows << endl;
    1362             :     }
    1363             :     
    1364           0 : }
    1365             : 
    1366           0 : DelayRateFFTConcat::DelayRateFFTConcat(Array<Complex>& data, Int nPadFactor, Float f0, Float df, Float dt, SDBList& s,
    1367           0 :                            Array<Double>& delayWindow, Array<Double>& rateWindow) :
    1368             :     DelayRateFFT(s, 0, delayWindow, rateWindow),
    1369           0 :     gm_(s),
    1370           0 :     nPadFactor_(nPadFactor),
    1371           0 :      Vpad_() {
    1372           0 :     dt_ = dt;
    1373           0 :     f0_ = f0;
    1374           0 :     df_ = df;
    1375           0 :     IPosition shape = data.shape();
    1376           0 :     nCorr_ = shape(0);
    1377           0 :     nElem_ = shape(1);
    1378           0 :     nt_ = shape(2);
    1379           0 :     nChan_ = shape(3);
    1380           0 :     nPadT_ = nPadFactor_*nt_;
    1381           0 :     nPadChan_ = nPadFactor_*nChan_;
    1382           0 :     IPosition paddedDataSize(4, nCorr_, nElem_, nPadT_, nPadChan_);
    1383           0 :     Vpad_.resize(paddedDataSize);
    1384             :     
    1385           0 :     IPosition start(4, 0, 0, 0, 0);
    1386           0 :     IPosition stop(4, nCorr_,  nElem_, nt_, nChan_);
    1387           0 :     IPosition stride(4, 1, 1, 1, 1);
    1388           0 :     Slicer sl1(start, stop, stride, Slicer::endIsLength);
    1389           0 :     Vpad_(sl1) = data;
    1390             : 
    1391           0 :     unitize(Vpad_);
    1392             : 
    1393           0 : }
    1394             : 
    1395             : void
    1396           0 : DelayRateFFTConcat::FFT() {
    1397             :     if (DEVDEBUG) {
    1398             :         cerr << "DelayRateFFTConcat::FFT()" << endl;
    1399             :     }
    1400             :     // Axes are 0: correlation (i.e., hand of polarization), 1: antenna, 2: time, 3: channel
    1401           0 :     Vector<Bool> ax(4, false);
    1402           0 :     ax(2) = true;
    1403           0 :     ax(3) = true;
    1404             :     // Also copied from DelayFFT in KJones.
    1405             :     // we make a copy to FFT in place.
    1406             :     if (DEVDEBUG) {
    1407             :         cerr << "Vpad_.shape() " << Vpad_.shape() << endl;
    1408             :     }
    1409           0 :     ArrayLattice<Complex> c(Vpad_);
    1410           0 :     LatticeFFT::cfft0(c, ax, true);
    1411             :     if (DEVDEBUG) {
    1412             :         cerr << "FFT transformed" << endl;
    1413             :     }
    1414           0 : }
    1415             : 
    1416             : std::pair<Bool, Float>
    1417           0 : DelayRateFFTConcat::xinterp(Float alo, Float amax, Float ahi) {
    1418           0 :     if (amax > 0.0 && alo == amax && amax == ahi)
    1419           0 :         return std::make_pair(true, 0.5);
    1420             : 
    1421           0 :     Float denom(alo-2.0*amax+ahi);
    1422           0 :     Bool cond = amax>0.0 && abs(denom)>0.0 ;
    1423           0 :     Float fpk = cond ? 0.5-(ahi-amax)/denom : 0.0;
    1424           0 :     return std::make_pair(cond, fpk);
    1425             : }
    1426             :     
    1427             : void
    1428           0 : DelayRateFFTConcat::searchPeak() {
    1429             :     if (DEVDEBUG) {
    1430             :         cerr << "DelayRateFFTConcat::searchPeak()" << endl;
    1431             :     }
    1432             : 
    1433             :     // Recall param_ -> [phase, delay, rate] for each correlation
    1434           0 :     param_.resize(3*nCorr_, nElem_); // This might be better done elsewhere.
    1435           0 :     param_.set(0.0);
    1436           0 :     flag_.resize(3*nCorr_, nElem_);
    1437           0 :     flag_.set(true);  // all flagged initially
    1438             :     if (DEVDEBUG) {
    1439             :         cerr << "nt_ " << nt_ << " nPadChan_ " << nPadChan_ << endl;
    1440             :         cerr << "Vpad_.shape() " << Vpad_.shape() << endl;
    1441             :         cerr << "delayWindow_ " << delayWindow_ << endl;
    1442             : 
    1443             :     }
    1444             :     
    1445           0 :     for (Int icorr=0; icorr<nCorr_; ++icorr) {
    1446           0 :         flag_(icorr*3 + 0, refant()) = false; 
    1447           0 :         flag_(icorr*3 + 1, refant()) = false;
    1448           0 :         flag_(icorr*3 + 2, refant()) = false;
    1449           0 :         for (Int ielem=0; ielem<nElem_; ++ielem) {
    1450           0 :             if (ielem==refant()) {
    1451           0 :                 continue;
    1452             :             }
    1453             :             // NB: Time, Channel
    1454             :             // And once again we fail at slicing
    1455           0 :             IPosition start(4, icorr, ielem,      0,         0);
    1456           0 :             IPosition stop(4,     1,     1, nPadT_, nPadChan_);
    1457           0 :             IPosition step(4,     1,     1,       1,        1);
    1458           0 :             Slicer sl(start, stop, step, Slicer::endIsLength);
    1459           0 :             Matrix<Complex> aS = Vpad_(sl).nonDegenerate();
    1460           0 :             Int sgn = (ielem < refant()) ? 1 : -1;
    1461             : 
    1462             :             // Below is the gory details for turning delay window into index range
    1463           0 :             Double bw = Float(nPadChan_)*df_;
    1464           0 :             Double d0 = sgn*delayWindow_(IPosition(1, 0));
    1465           0 :             Double d1 = sgn*delayWindow_(IPosition(1, 1));
    1466           0 :             if (d0 > d1) std::swap(d0, d1);
    1467           0 :             d0 = max(d0, -0.5/df_);
    1468           0 :             d1 = min(d1, (0.5-1/nPadChan_)/df_);
    1469             : 
    1470             :             // It's simpler to keep the ranges as signed integers and
    1471             :             // handle the wrapping of the FFT in the loop over
    1472             :             // indices. Recall that the FFT result returned has indices
    1473             :             // that run from 0 to nPadChan_/2 -1 and then from
    1474             :             // -nPadChan/2 to -1, so far as our delay is concerned.
    1475           0 :             Int i0 = bw*d0;
    1476           0 :             Int i1 = bw*d1;
    1477           0 :             if (i1==i0) i1++;
    1478             :             // Now for the gory details for turning rate window into index range
    1479           0 :             Double width = nPadT_*dt_*1e9*f0_;
    1480           0 :             Double r0 = sgn*rateWindow_(IPosition(1,0));
    1481           0 :             Double r1 = sgn*rateWindow_(IPosition(1,1));
    1482           0 :             if (r0 > r1) std::swap(r0, r1);
    1483           0 :             r0 = max(r0, -0.5/(dt_*1e9*f0_));
    1484           0 :             r1 = min(r1, (0.5 - 1/nPadT_)/(dt_*1e9*f0_));
    1485             :             
    1486           0 :             Int j0 = width*r0;
    1487           0 :             Int j1 = width*r1;
    1488           0 :             if (j1==j0) j1++;
    1489             :             if (DEVDEBUG) {
    1490             :                 cerr << "Checking the windows for delay and rate search." << endl;
    1491             :                 cerr << "bw " << bw << endl;
    1492             :                 cerr << "d0 " << d0 << " d1 " << d1 << endl;
    1493             :                 cerr << "i0 " << i0 << " i1 " << i1 << endl; 
    1494             :                 cerr << "r0 " << r0 << " r1 " << r1 << endl;
    1495             :                 cerr << "j0 " << j0 << " j1 " << j1 << endl; 
    1496             :             }
    1497           0 :             Matrix<Float> amp(amplitude(aS));
    1498           0 :             Int ipkch(0);
    1499           0 :             Int ipkt(0);
    1500           0 :             Float amax(-1.0);
    1501             :             // Unlike KJones we have to iterate in time too
    1502           0 :             for (Int itime0=j0; itime0 != j1; itime0++) {
    1503           0 :                 Int itime = (itime0 < 0) ? itime0 + nPadT_ : itime0;
    1504           0 :                 for (Int ich0=i0; ich0 != i1; ich0++) {
    1505           0 :                     Int ich = (ich0 < 0) ? ich0 + nPadChan_ : ich0;
    1506             :                     // cerr << "Gridpoint " << itime << ", " << ich << "->" << amp(itime, ich) << endl;
    1507           0 :                     if (amp(itime, ich) > amax) {
    1508           0 :                         ipkch = ich;
    1509           0 :                         ipkt  = itime;
    1510           0 :                         amax=amp(itime, ich);
    1511             :                     }
    1512             :                 }
    1513             :             }
    1514             :             // Finished grovelling. Now we have the location of the
    1515             :             // maximum amplitude.
    1516           0 :             Float alo_ch = amp(ipkt, (ipkch > 0) ? ipkch-1 : nPadChan_-1);
    1517           0 :             Float ahi_ch = amp(ipkt, ipkch<(nPadChan_-1) ? ipkch+1 : 0);
    1518           0 :             std::pair<Bool, Float> maybeFpkch = xinterp(alo_ch, amax, ahi_ch);
    1519             :             // We handle wrapping while looking for neighbours
    1520           0 :             Float alo_t = amp(ipkt > 0 ? ipkt-1 : nPadT_ -1,     ipkch);
    1521           0 :             Float ahi_t = amp(ipkt < (nPadT_ -1) ? ipkt+1 : 0,   ipkch);
    1522             :             if (DEVDEBUG) {
    1523             :                 cerr << "For element " << ielem << endl;
    1524             :                 cerr << "In channel dimension ipkch " << ipkch << " alo " << alo_ch
    1525             :                      << " amax " << amax << " ahi " << ahi_ch << endl;
    1526             :                 cerr << "In time dimension ipkt " << ipkt << " alo " << alo_t
    1527             :                      << " amax " << amax << " ahi " << ahi_t << endl;
    1528             :             }
    1529           0 :             std::pair<Bool, Float> maybeFpkt = xinterp(alo_t, amax, ahi_t);
    1530             : 
    1531           0 :             if (maybeFpkch.first and maybeFpkt.first) {
    1532             :                 // Phase
    1533           0 :                 Complex c = aS(ipkt, ipkch);
    1534           0 :                 Float phase = arg(c);
    1535           0 :                 param_(icorr*3 + 0, ielem) = sgn*phase;
    1536           0 :                 Float delay = (ipkch)/Float(nPadChan_);
    1537           0 :                 if (delay > 0.5) delay -= 1.0;           // fold
    1538           0 :                 delay /= df_;                           // nsec
    1539           0 :                 param_(icorr*3 + 1, ielem) = sgn*delay; //
    1540           0 :                 Double rate = (ipkt)/Float(nPadT_);
    1541           0 :                 if (rate > 0.5) rate -= 1.0;
    1542           0 :                 Double rate0 = rate/dt_;
    1543           0 :                 Double rate1 = rate0/(1e9 * f0_); 
    1544             : 
    1545           0 :                 param_(icorr*3 + 2, ielem) = Float(sgn*rate1); 
    1546             :                 if (DEVDEBUG) {
    1547             :                     cerr << "maybeFpkch.second=" << maybeFpkch.second
    1548             :                          << ", df_= " << df_ 
    1549             :                          << " fpkch=" << (ipkch + maybeFpkch.second) << endl;
    1550             :                     cerr << " maybeFpkt.second=" << maybeFpkt.second
    1551             :                          << " rate0=" << rate
    1552             :                          << " 1e9 * f0_=" << 1e9 * f0_ 
    1553             :                          << ", dt_=" << dt_
    1554             :                          << " fpkt=" << (ipkt + maybeFpkt.second) << endl;
    1555             :                         
    1556             :                 }
    1557             :                 if (DEVDEBUG) {
    1558             :                     cerr << "Found peak for element " << ielem << " correlation " << icorr
    1559             :                          << " ipkt=" << ipkt << "/" << nPadT_ << ", ipkch=" << ipkch << "/" << nPadChan_
    1560             :                          << " peak=" << amax 
    1561             :                          << "; delay " << delay << ", rate " << rate
    1562             :                          << ", phase " << arg(c) << " sign= " << sgn << endl;
    1563             :                 }
    1564             :                 // Set 3 flags.
    1565           0 :                 flag_(icorr*3 + 0, ielem)=false; 
    1566           0 :                 flag_(icorr*3 + 1, ielem)=false;
    1567           0 :                 flag_(icorr*3 + 2, ielem)=false;
    1568           0 :             }
    1569             :             else {
    1570             :                 if (DEVDEBUG) {
    1571             :                     cerr << "No peak in 2D FFT for element " << ielem << " correlation " << icorr << endl;
    1572             :                 }
    1573             :             }
    1574           0 :         }
    1575             :     }
    1576           0 : }
    1577             : 
    1578             : 
    1579             : Float
    1580           0 : DelayRateFFTConcat::snr(Int icorr, Int ielem, Float delay, Float rate) {
    1581             :     if (DEVDEBUG) {
    1582             :         cerr << "DelayRateFFTConcat::snr"<< endl;
    1583             :     }
    1584             :     // We calculate a signal-to-noise ration for the 2D FFT fringefit
    1585             :     // using a formula transcribed from AIPS FRING.
    1586             :     //
    1587             :     // Have to convert delay and rate back into indices on the padded 2D grid.
    1588           0 :     Int sgn = (ielem < refant()) ? 1 : -1;
    1589           0 :     delay *= sgn*df_;
    1590           0 :     if (delay < 0.0) delay += 1;
    1591           0 :     Int ichan = Int(delay*nPadChan_ + 0.5); 
    1592           0 :     if (ichan == nPadChan_) ichan = 0;
    1593             :         
    1594           0 :     rate *= sgn*1e9 * f0_;
    1595           0 :     rate *= dt_;
    1596           0 :     if (rate < 0.0) rate += 1;
    1597           0 :     Int itime = Int(rate*nPadT_ + 0.5);
    1598           0 :     if (itime == nPadT_) itime = 0;
    1599             :     // What about flags? If the datapoint closest to the computed
    1600             :     // delay and rate values is flagged we probably shouldn't use
    1601             :     // it, but what *should* we use?
    1602           0 :     IPosition ipos(4, icorr, ielem, itime, ichan);
    1603           0 :     IPosition p(2, icorr, ielem);
    1604           0 :     Complex v = Vpad_(ipos);
    1605           0 :     Float peak = abs(v);
    1606           0 :     if (peak > 0.999*sumw_(p)) peak=0.999*sumw_(p);
    1607             :     // xcount is number of data points for baseline to ielem
    1608             :     // sumw is sum of weights,
    1609             :     // sumww is sum of squares of weights
    1610             : 
    1611             :     Float cwt;
    1612           0 :     if (fabs(sumw_(p))<FLT_EPSILON) {
    1613           0 :         cwt = 0;
    1614             :         if (DEVDEBUG) {
    1615             :             cerr << "Correlation " << icorr << " antenna " << ielem << ": sum of weights is zero." << endl;
    1616             :         }
    1617             :     } else {
    1618           0 :         Float x = C::pi/2*peak/sumw_(p);
    1619             :         // The magic numbers in the following formula are from AIPS FRING
    1620           0 :         cwt = (pow(tan(x), 1.163) * sqrt(sumw_(p)/sqrt(sumww_(p)/xcount_(p))));
    1621             :         if (DEVDEBUG) {
    1622             :             cerr << "Correlation " << icorr << " antenna " << ielem << " ipos " << ipos
    1623             :                  << " peak=" << peak << "; xang=" << x << "; xcount=" << xcount_(p) << "; sumw=" << sumw_(p) << "; sumww=" << sumww_(p)
    1624             :                  << " snr " << cwt << endl;
    1625             :         }
    1626             :     }
    1627           0 :     return cwt;
    1628           0 : }
    1629             :     
    1630             : 
    1631             : } // end namespace

Generated by: LCOV version 1.16