1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
|
/***********************************************/
/**
* @file gnssProcessingStep.cpp
*
* @brief Processing steps for GNSS normal equations.
*
* @author Torsten Mayer-Guerr
* @date 2021-09-05
*
*/
/***********************************************/
#define DOCSTRING_GnssProcessingStep
#include "base/import.h"
#include "parallel/matrixDistributed.h"
#include "config/configRegister.h"
#include "config/config.h"
#include "inputOutput/logging.h"
#include "misc/varianceComponentEstimation.h"
#include "gnss/gnss.h"
#include "gnss/gnssProcessingStep/gnssProcessingStep.h"
#include "gnss/gnssProcessingStep/gnssProcessingStepEstimate.h"
#include "gnss/gnssProcessingStep/gnssProcessingStepResolveAmbiguities.h"
#include "gnss/gnssProcessingStep/gnssProcessingStepComputeCovarianceMatrix.h"
#include "gnss/gnssProcessingStep/gnssProcessingStepWriteResults.h"
#include "gnss/gnssProcessingStep/gnssProcessingStepWriteNormalEquations.h"
#include "gnss/gnssProcessingStep/gnssProcessingStepWriteAprioriSolution.h"
#include "gnss/gnssProcessingStep/gnssProcessingStepWriteResiduals.h"
#include "gnss/gnssProcessingStep/gnssProcessingStepWriteUsedStationList.h"
#include "gnss/gnssProcessingStep/gnssProcessingStepWriteUsedTransmitterList.h"
#include "gnss/gnssProcessingStep/gnssProcessingStepPrintResidualStatistics.h"
#include "gnss/gnssProcessingStep/gnssProcessingStepSelectParametrizations.h"
#include "gnss/gnssProcessingStep/gnssProcessingStepSelectEpochs.h"
#include "gnss/gnssProcessingStep/gnssProcessingStepSelectNormalsBlockStructure.h"
#include "gnss/gnssProcessingStep/gnssProcessingStepSelectReceivers.h"
#include "gnss/gnssProcessingStep/gnssProcessingStepForEachReceiverSeparately.h"
#include "gnss/gnssProcessingStep/gnssProcessingStepGroup.h"
#include "gnss/gnssProcessingStep/gnssProcessingStepDisableTransmitterShadowEpochs.h"
/***********************************************/
GROOPS_REGISTER_CLASS(GnssProcessingStep, "gnssProcessingStepType",
GnssProcessingStepEstimate,
GnssProcessingStepResolveAmbiguities,
GnssProcessingStepComputeCovarianceMatrix,
GnssProcessingStepWriteResults,
GnssProcessingStepWriteNormalEquations,
GnssProcessingStepWriteAprioriSolution,
GnssProcessingStepWriteResiduals,
GnssProcessingStepWriteUsedStationList,
GnssProcessingStepWriteUsedTransmitterList,
GnssProcessingStepPrintResidualStatistics,
GnssProcessingStepSelectParametrizations,
GnssProcessingStepSelectEpochs,
GnssProcessingStepSelectNormalsBlockStructure,
GnssProcessingStepSelectReceivers,
GnssProcessingStepForEachReceiverSeparately,
GnssProcessingStepGroup,
GnssProcessingStepDisableTransmitterShadowEpochs)
GROOPS_READCONFIG_UNBOUNDED_CLASS(GnssProcessingStep, "gnssProcessingStepType")
/***********************************************/
GnssProcessingStep::GnssProcessingStep(Config &config, const std::string &name)
{
try
{
std::string type;
while(readConfigChoice(config, name, type, Config::OPTIONAL, "", ""))
{
if(readConfigChoiceElement(config, "estimate", type, "least squares adjustment"))
bases.push_back(new GnssProcessingStepEstimate(config));
if(readConfigChoiceElement(config, "resolveAmbiguities", type, "resolve integer ambiguities"))
bases.push_back(new GnssProcessingStepResolveAmbiguities(config));
if(readConfigChoiceElement(config, "computeCovarianceMatrix", type, "compute covariance matrix"))
bases.push_back(new GnssProcessingStepComputeCovarianceMatrix(config));
if(readConfigChoiceElement(config, "writeResults", type, "write all estimated parameters"))
bases.push_back(new GnssProcessingStepWriteResults(config));
if(readConfigChoiceElement(config, "writeNormalEquations", type, "write unconstrained and constraint normal equations"))
bases.push_back(new GnssProcessingStepWriteNormalEquations(config));
if(readConfigChoiceElement(config, "writeAprioriSolution", type, "write apriori solution vector"))
bases.push_back(new GnssProcessingStepWriteAprioriSolution(config));
if(readConfigChoiceElement(config, "writeResiduals", type, "write observation residuals"))
bases.push_back(new GnssProcessingStepWriteResiduals(config));
if(readConfigChoiceElement(config, "writeUsedStationList", type, "write used stations"))
bases.push_back(new GnssProcessingStepWriteUsedStationList(config));
if(readConfigChoiceElement(config, "writeUsedTransmitterList", type, "write used transmitters"))
bases.push_back(new GnssProcessingStepWriteUsedTransmitterList(config));
if(readConfigChoiceElement(config, "printResidualStatistics", type, "print residual statistics"))
bases.push_back(new GnssProcessingStepPrintResidualStatistics(config));
if(readConfigChoiceElement(config, "selectParametrizations", type, "select parametrizations for all subsequent processing steps"))
bases.push_back(new GnssProcessingStepSelectParametrizations(config));
if(readConfigChoiceElement(config, "selectEpochs", type, "select epochs to be used in all subsequent processing steps"))
bases.push_back(new GnssProcessingStepSelectEpochs(config));
if(readConfigChoiceElement(config, "selectNormalsBlockStructure", type, "select block structure of the distributed normal equation matrix for all subsequent processing steps"))
bases.push_back(new GnssProcessingStepSelectNormalsBlockStructure(config));
if(readConfigChoiceElement(config, "selectReceivers", type, "use this subset of stations in all subsequent processing steps"))
bases.push_back(new GnssProcessingStepSelectReceivers(config));
if(readConfigChoiceElement(config, "forEachReceiverSeparately", type, "process receiver by receiver, transmitter parameters disabled"))
bases.push_back(new GnssProcessingStepForEachReceiverSeparately(config));
if(readConfigChoiceElement(config, "group", type, "group processing steps"))
bases.push_back(new GnssProcessingStepGroup(config));
if(readConfigChoiceElement(config, "disableTransmitterShadowEpochs", type, "disable transmitter epochs in eclipse"))
bases.push_back(new GnssProcessingStepDisableTransmitterShadowEpochs(config));
endChoice(config);
if(isCreateSchema(config))
return;
}
}
catch(std::exception &e)
{
GROOPS_RETHROW(e)
}
}
/***********************************************/
GnssProcessingStep::~GnssProcessingStep()
{
for(auto base : bases)
delete base;
}
/***********************************************/
void GnssProcessingStep::process(GnssProcessingStep::State &state)
{
try
{
for(auto base : bases)
{
Parallel::peek(state.normalEquationInfo.comm);
if(base->expectInitializedParameters() && state.changedNormalEquationInfo)
{
state.gnss->initParameter(state.normalEquationInfo);
state.changedNormalEquationInfo = FALSE;
}
base->process(state);
}
}
catch(std::exception &e)
{
GROOPS_RETHROW(e)
}
}
/***********************************************/
/***********************************************/
GnssProcessingStep::State::State(GnssPtr gnss, Parallel::CommunicatorPtr comm) :
gnss(gnss), normalEquationInfo(gnss->times.size(), gnss->receivers.size(), gnss->transmitters.size(), comm),
changedNormalEquationInfo(TRUE), sigmaType(gnss->receivers.size()), sigmaFactor(gnss->receivers.size())
{
try
{
}
catch(std::exception &e)
{
GROOPS_RETHROW(e)
}
}
/***********************************************/
void GnssProcessingStep::State::regularizeNotUsedParameters(UInt blockStart, UInt blockCount, const std::vector<ParameterName> ¶meterNames)
{
try
{
UInt countZeros = 0;
for(UInt i=blockStart; i<blockStart+blockCount; i++)
if(normals.isMyRank(i,i))
{
Matrix &N = normals.N(i,i);
for(UInt k=0; k<N.rows(); k++)
if(N(k,k)==0.)
{
N(k,k) += 1.0;
countZeros++;
logWarning<<" "<<parameterNames.at(normals.blockIndex(i)+k).str()<<" has zero diagonal element"<<Log::endl;
}
}
Parallel::reduceSum(countZeros, 0, normals.communicator());
if(countZeros && Parallel::isMaster(normals.communicator()))
{
logWarning<<" "<<countZeros<<" parameters have zero diagonal elements -> set to one"<<Log::endl;
obsCount -= countZeros;
}
}
catch(std::exception &e)
{
GROOPS_RETHROW(e)
}
}
/***********************************************/
void GnssProcessingStep::State::collectNormalsBlocks(UInt blockStart, UInt blockCount)
{
try
{
if(Parallel::size(normals.communicator()) <= 1)
return;
auto countBlocks = [&]()
{
UInt countBlocks = 0;
for(UInt idBlock=blockStart; idBlock<blockStart+blockCount; idBlock++)
normals.loopBlockRow(idBlock, {idBlock, normals.blockCount()}, [&](UInt /*k*/, UInt /*ik*/) {countBlocks++;});
return countBlocks;
};
// synchronize used blocks
for(UInt idProcess=0; idProcess<Parallel::size(normals.communicator()); idProcess++)
{
Matrix rowColRank(countBlocks(), 3);
UInt i = 0;
for(UInt idBlock=blockStart; idBlock<blockStart+blockCount; idBlock++)
normals.loopBlockRow(idBlock, {idBlock, normals.blockCount()}, [&](UInt k, UInt ik)
{
rowColRank(i, 0) = idBlock;
rowColRank(i, 1) = k;
rowColRank(i, 2) = normals._rank[ik];
i++;
});
Parallel::broadCast(rowColRank, idProcess, normals.communicator());
for(UInt i=0; i<rowColRank.rows(); i++)
normals.setBlock(static_cast<UInt>(rowColRank(i,0)), static_cast<UInt>(rowColRank(i,1)), static_cast<UInt>(rowColRank(i,2)));
}
// for each block x which process contains
Matrix usedRanks(countBlocks(), Parallel::size(normals.communicator()));
UInt i = 0;
for(UInt idBlock=blockStart; idBlock<blockStart+blockCount; idBlock++)
normals.loopBlockRow(idBlock, {idBlock, normals.blockCount()}, [&](UInt /*k*/, UInt ik)
{
if(normals._N[ik].size())
usedRanks(i, Parallel::myRank(normals.communicator())) = 1.;
i++;
});
Parallel::reduceSum(usedRanks, 0, normals.communicator());
Parallel::broadCast(usedRanks, 0, normals.communicator());
// reduce sum normals
i = 0;
for(UInt idBlock=blockStart; idBlock<blockStart+blockCount; idBlock++)
normals.loopBlockRow(idBlock, {idBlock, normals.blockCount()}, [&](UInt /*k*/, UInt ik)
{
std::vector<Bool> usedRank(usedRanks.columns());
for(UInt idProcess=0; idProcess<usedRank.size(); idProcess++)
usedRank[idProcess] = usedRanks(i, idProcess);
normals.reduceSum(normals._N[ik], ik, usedRank, TRUE/*free*/);
i++;
});
// right hand side
for(UInt idBlock=blockStart; idBlock<blockStart+blockCount; idBlock++)
{
Parallel::reduceSum(n.at(idBlock), 0, normals.comm);
if(!Parallel::isMaster(normals.comm))
n.at(idBlock).setNull();
}
}
catch(std::exception &e)
{
GROOPS_RETHROW(e)
}
}
/***********************************************/
void GnssProcessingStep::State::buildNormals(Bool constraintsOnly, Bool solveEpochParameters)
{
try
{
normals.initEmpty(normalEquationInfo.blockIndices(), normalEquationInfo.comm,
std::bind(&GnssNormalEquationInfo::normalsBlockRank, &normalEquationInfo, std::placeholders::_1, std::placeholders::_2, std::placeholders::_3));
n.resize(normals.blockCount());
for(UInt i=0; i<normals.blockCount(); i++)
n.at(i) = Vector(normals.blockSize(i));
lPl = Vector(1);
obsCount = 0;
// Loop over all epochs
// --------------------
Parallel::barrier(normalEquationInfo.comm);
logStatus<<"accumulate normals"<<Log::endl;
GnssDesignMatrix A(normalEquationInfo);
std::vector<GnssObservationEquation> eqns(gnss->transmitters.size());
UInt blockStart = 0; // first block, which is not regularized and reduced
UInt blockCount = 0;
UInt idLoop = 0;
Log::Timer timer(normalEquationInfo.idEpochs.size());
for(UInt idEpoch : normalEquationInfo.idEpochs)
{
timer.loopStep(idLoop++);
gnss->constraintsEpoch(normalEquationInfo, idEpoch, normals, n, lPl(0), obsCount);
// loop over all receivers
if(!constraintsOnly)
for(UInt idRecv=0; idRecv<gnss->receivers.size(); idRecv++)
if(normalEquationInfo.estimateReceiver.at(idRecv) && gnss->receivers.at(idRecv)->isMyRank())
{
// all observation equations for this epoch
UInt countEqn = 0;
for(UInt idTrans=0; idTrans<gnss->receivers.at(idRecv)->idTransmitterSize(idEpoch); idTrans++)
if(gnss->basicObservationEquations(normalEquationInfo, idRecv, idTrans, idEpoch, eqns.at(countEqn)))
{
eqns.at(countEqn).eliminateGroupParameters();
if(!normalEquationInfo.accumulateEpochObservations)
{
A.init(eqns.at(countEqn).l);
gnss->designMatrix(normalEquationInfo, eqns.at(countEqn), A);
A.accumulateNormals(normals, n, lPl(0), obsCount);
obsCount -= eqns.at(countEqn).rankDeficit;
}
countEqn++;
}
if(normalEquationInfo.accumulateEpochObservations)
{
// copy all observations to a single vector
A.init(Vector(std::accumulate(eqns.begin(), eqns.begin()+countEqn, UInt(0), [](UInt count, auto &eqn) {return count+eqn.l.rows();})));
UInt idx=0;
for(UInt i=0; i<countEqn; i++)
{
copy(eqns.at(i).l, A.l.row(idx, eqns.at(i).l.rows()));
gnss->designMatrix(normalEquationInfo, eqns.at(i), A.selectRows(idx, eqns.at(i).l.rows()));
idx += eqns.at(i).l.rows();
}
A.selectRows(0, 0); // select all
A.accumulateNormals(normals, n, lPl(0), obsCount);
obsCount -= std::accumulate(eqns.begin(), eqns.begin()+countEqn, UInt(0), [](UInt count, auto &eqn) {return count+eqn.rankDeficit;});
}
} // for(idRecv)
// perform following steps not every epoch
blockCount += normalEquationInfo.blockCountEpoch(idEpoch);
if((blockCount < normalEquationInfo.defaultBlockCountReduction) && (idEpoch != normalEquationInfo.idEpochs.back()))
continue;
collectNormalsBlocks(blockStart, blockCount);
if(solveEpochParameters && !constraintsOnly)
{
regularizeNotUsedParameters(blockStart, blockCount, normalEquationInfo.parameterNames());
normals.cholesky(FALSE, blockStart, blockCount, FALSE);
normals.triangularTransSolve(n, blockStart, blockCount, FALSE);
if(Parallel::isMaster(normalEquationInfo.comm))
for(UInt idBlock=blockStart; idBlock<blockStart+blockCount; idBlock++)
{
lPl(0) -= quadsum(n.at(idBlock)); // lPl = lPl - n1' N1^(-1) n1
obsCount -= normals.blockSize(idBlock);
}
}
if(solveEpochParameters)
{
// remove N12 (epoch <-> other parameters)
for(UInt idBlock=blockStart; idBlock<blockStart+blockCount; idBlock++)
normals.loopBlockRow(idBlock, {normalEquationInfo.blockInterval(), normals.blockCount()}, [&](UInt /*k*/, UInt ik)
{
if(normals._N[ik].size())
normals._N[ik] = Matrix();
});
}
blockStart += blockCount;
blockCount = 0;
} // for(idEpoch)
Parallel::barrier(normalEquationInfo.comm);
timer.loopEnd();
// other observations and constraints
// ----------------------------------
gnss->constraints(normalEquationInfo, normals, n, lPl(0), obsCount);
// collect normal equations
// ------------------------
if(Parallel::size(normals.comm)>1)
{
logStatus<<"collect normal equations"<<Log::endl;
collectNormalsBlocks(blockStart, normals.blockCount()-blockStart);
Parallel::reduceSum(lPl, 0, normalEquationInfo.comm);
Parallel::reduceSum(obsCount, 0, normalEquationInfo.comm);
}
}
catch(std::exception &e)
{
GROOPS_RETHROW(e)
}
}
/***********************************************/
Double GnssProcessingStep::State::estimateSolution(const std::function<Vector(const_MatrixSliceRef xFloat, MatrixSliceRef W, const_MatrixSliceRef d, Vector &xInt, Double &sigma)> &searchInteger,
const std::vector<Byte> &ambiguityTransmitters, const std::vector<Byte> &ambiguityReceivers,
Bool computeResiduals, Bool computeWeights, Bool adjustSigma0, Double huber, Double huberPower)
{
try
{
// setup normal equations
const Bool solveEpochParameters = !normalEquationInfo.keepEpochNormalsInMemory && (normalEquationInfo.blockInterval() < normalEquationInfo.blockCount());
buildNormals(FALSE/*constraintsOnly*/, solveEpochParameters);
// eliminate all other parameters from the ambiguity normals
// ---------------------------------------------------------
Parallel::barrier(normalEquationInfo.comm);
logStatus<<"solve"<<Log::endl;
const UInt blockStart = (solveEpochParameters) ? normalEquationInfo.blockInterval() : 0; // epoch parameters already eliminated?
const UInt blockCount = ((searchInteger) ? normalEquationInfo.blockAmbiguity() : normals.blockCount()) - blockStart;
regularizeNotUsedParameters(blockStart, normals.blockCount()-blockStart, normalEquationInfo.parameterNames());
normals.cholesky(TRUE/*timing*/, blockStart, blockCount, TRUE/*collect*/);
normals.triangularTransSolve(n, blockStart, blockCount, TRUE/*collect*/); // forward step
if(Parallel::isMaster(normalEquationInfo.comm))
for(UInt z=blockStart; z<blockStart+blockCount; z++)
{
lPl(0) -= quadsum(n.at(z)); // lPl = lPl - n1' N1^(-1) n1
obsCount -= normals.blockSize(z);
}
// resolve integer ambiguities
// ---------------------------
if(searchInteger)
{
logStatus<<"Resolve integer ambiguities (may take a while)"<<Log::endl;
Double sigmaFloat = gnss->ambiguityResolve(normalEquationInfo, normals, n, lPl(0), obsCount,
ambiguityTransmitters, ambiguityReceivers, searchInteger);
logInfo<<" sigma(float) = "<<sigmaFloat%"%.2f"s<<Log::endl;
logInfo<<" sigma(fixed) = "<<std::sqrt(lPl(0)/obsCount)%"%.2f"s<<Log::endl;
Parallel::barrier(normalEquationInfo.comm);
} // if(parameterCountAmbiguities)
// Compute sigma
// -------------
if(Parallel::isMaster(normalEquationInfo.comm))
{
const Double sigma = Vce::standardDeviation(lPl(0), obsCount, huber, huberPower);
logInfo<<" sigma = "<<sigma%"%.2f"s<<Log::endl;
if((sigma!=sigma) || (sigma<=0))
logWarning<<" Cannot compute sigma = sqrt("<<lPl(0)<<"/"<<obsCount<<")"<<Log::endl;
}
constexpr UInt monteCarloColumns = 100;
std::vector<Matrix> monteCarlo; // Monte Carlo Vector for redundancy computation
// ========================================================================================
// solve (Backward step)
// ---------------------
normals.triangularSolve(n, blockStart, normals.blockCount()-blockStart);
if(computeResiduals)
{
monteCarlo.resize(normals.blockCount());
for(UInt i=blockStart; i<normals.blockCount(); i++)
monteCarlo.at(i) = Matrix(normals.blockSize(i), monteCarloColumns);
if(Parallel::isMaster(normalEquationInfo.comm))
for(UInt i=blockStart; i<blockStart+blockCount; i++) // possible without resolved ambiguity parameters
monteCarlo.at(i) = Vce::monteCarlo(monteCarlo.at(i).rows(), monteCarlo.at(i).columns());
normals.triangularSolve(monteCarlo, blockStart, blockCount);
}
// ========================================================================================
// Reconstruct epoch parameters
// ----------------------------
if(solveEpochParameters)
{
logStatus<<"Reconstruct epoch parameters"<<Log::endl;
normals.eraseBlocks(blockStart, normals.blockCount()-blockStart); // free N12, N22
// broadcast n, Wz
// ---------------
for(UInt i=0; i<blockStart; i++)
n.at(i).setNull();
for(UInt i=blockStart; i<normalEquationInfo.blockCount(); i++)
Parallel::broadCast(n.at(i), 0, normalEquationInfo.comm);
if(computeResiduals)
{
for(UInt i=0; i<blockStart; i++)
monteCarlo.at(i) = Matrix(normalEquationInfo.blockSize(i), monteCarloColumns);
for(UInt i=blockStart; i<blockStart+blockCount; i++)
Parallel::broadCast(monteCarlo.at(i), 0, normalEquationInfo.comm);
}
// reconstruct N12
// ---------------
GnssDesignMatrix A(normalEquationInfo);
UInt idLoop = 0;
Log::Timer timer(normalEquationInfo.idEpochs.size());;
for(UInt idEpoch : normalEquationInfo.idEpochs)
{
timer.loopStep(idLoop++);
// loop over all receivers
GnssObservationEquation eqn;
for(UInt idRecv=0; idRecv<gnss->receivers.size(); idRecv++)
if(normalEquationInfo.estimateReceiver.at(idRecv) && gnss->receivers.at(idRecv)->isMyRank())
for(UInt idTrans=0; idTrans<gnss->receivers.at(idRecv)->idTransmitterSize(idEpoch); idTrans++)
if(gnss->basicObservationEquations(normalEquationInfo, idRecv, idTrans, idEpoch, eqn))
{
eqn.eliminateGroupParameters();
A.init(eqn.l);
gnss->designMatrix(normalEquationInfo, eqn, A);
A.transMult(eqn.l-A.mult(n, blockStart, normalEquationInfo.blockCount()-blockStart), n, 0, blockStart);
A.transMult(-A.mult(monteCarlo, blockStart, blockCount), monteCarlo, 0, blockStart);
}
} // for(idEpoch)
Parallel::barrier(normalEquationInfo.comm);
timer.loopEnd();
// collect
// -------
auto collect = [&](std::vector<Matrix> &n, UInt columns)
{
Matrix tmp(normalEquationInfo.blockIndex(blockStart), columns);
for(UInt i=0; i<blockStart; i++) // copy to one block (better performance of reduceSum)
if(normalEquationInfo.blockSize(i))
copy(n.at(i), tmp.row(normalEquationInfo.blockIndex(i), normalEquationInfo.blockSize(i)));
Parallel::reduceSum(tmp, 0, normalEquationInfo.comm);
if(Parallel::isMaster(normalEquationInfo.comm))
for(UInt i=0; i<blockStart; i++)
if(normalEquationInfo.blockSize(i))
copy(tmp.row(normalEquationInfo.blockIndex(i), normalEquationInfo.blockSize(i)), n.at(i));
if(!Parallel::isMaster(normalEquationInfo.comm))
for(UInt i=0; i<blockStart; i++)
n.at(i).setNull();
};
collect(n, 1);
if(computeResiduals)
collect(monteCarlo, monteCarloColumns);
// solve epoch (Forward step)
// --------------------------
normals.triangularTransSolve(n, 0, blockStart, FALSE);
if(computeResiduals)
{
normals.triangularTransSolve(monteCarlo, 0, blockStart, FALSE);
if(Parallel::isMaster(normalEquationInfo.comm))
for(UInt i=0; i<blockStart; i++)
axpy(1, Vce::monteCarlo(monteCarlo.at(i).rows(), monteCarlo.at(i).columns()), monteCarlo.at(i));
}
// solve epoch (Backward step)
// ---------------------------
normals.triangularSolve(n, 0, blockStart);
if(computeResiduals)
normals.triangularSolve(monteCarlo, 0, blockStart);
}
// ========================================================================================
// free memory
normals.initEmpty(normals.blockIndex(), normals.communicator());
// copy solution to one vector
// ---------------------------
Vector x;
if(Parallel::isMaster(normalEquationInfo.comm))
{
x = Vector(normalEquationInfo.parameterCount());
for(UInt i=0; i<normalEquationInfo.blockCount(); i++)
if(normalEquationInfo.blockSize(i))
copy(n.at(i), x.row(normalEquationInfo.blockIndex(i), normalEquationInfo.blockSize(i)));
}
n.clear();
n.shrink_to_fit();
Parallel::broadCast(x, 0, normalEquationInfo.comm);
if(!computeResiduals)
{
logInfo<<"Parameter changes"<<Log::endl;
return gnss->updateParameter(normalEquationInfo, x, Matrix());
}
// copy monte carlo to one vector
// ------------------------------
Matrix Wz;
if(Parallel::isMaster(normalEquationInfo.comm))
{
Wz = Matrix(normalEquationInfo.parameterCount(), monteCarloColumns);
for(UInt i=0; i<normalEquationInfo.blockCount(); i++)
if(normalEquationInfo.blockSize(i))
copy(monteCarlo.at(i), Wz.row(normalEquationInfo.blockIndex(i), normalEquationInfo.blockSize(i)));
}
monteCarlo.clear();
monteCarlo.shrink_to_fit();
Parallel::broadCast(Wz, 0, normalEquationInfo.comm);
// Residual tracking
// -----------------
Gnss::InfoParameterChange infoTec("tec");
std::vector<GnssType> typesResiduals = gnss->types(~(GnssType::PRN + GnssType::FREQ_NO));
std::vector<Gnss::InfoParameterChange> infosResiduals(typesResiduals.size(), Gnss::InfoParameterChange("mm"));
Double minSTEC = std::numeric_limits<Double>::infinity();
Double maxSTEC = -std::numeric_limits<Double>::infinity();
Double meanSTEC = 0;
Double stdSTEC = 0;
UInt countSTEC = 0;
Parallel::barrier(normalEquationInfo.comm);
logStatus<<"Compute residuals"<<Log::endl;
GnssDesignMatrix A(normalEquationInfo);
UInt idLoop = 0;
Log::Timer timer(normalEquationInfo.idEpochs.size());
for(UInt idEpoch : normalEquationInfo.idEpochs)
{
timer.loopStep(idLoop++);
// loop over all receivers
for(UInt idRecv=0; idRecv<gnss->receivers.size(); idRecv++)
if(normalEquationInfo.estimateReceiver.at(idRecv) && gnss->receivers.at(idRecv)->isMyRank())
{
GnssObservationEquation eqn;
for(UInt idTrans=0; idTrans<gnss->receivers.at(idRecv)->idTransmitterSize(idEpoch); idTrans++)
if(gnss->basicObservationEquations(normalEquationInfo, idRecv, idTrans, idEpoch, eqn))
{
// setup observation equations
A.init(eqn.l);
gnss->designMatrix(normalEquationInfo, eqn, A);
Vector We = eqn.l - A.mult(x); // decorrelated residuals
Matrix AWz = A.mult(Wz); // redundancies
// estimate & reduce B parameters (ionosphere)
// -------------------------------------------
if(eqn.B.size())
{
// estimate and eliminate STEC parameter
Matrix B = eqn.B;
Vector tau = QR_decomposition(B);
QTransMult(B, tau, We);
triangularSolve(1., B.row(0, B.columns()), We.row(0, B.columns()));
eqn.receiver->observation(eqn.transmitter->idTrans(), eqn.idEpoch)->updateParameter(We.row(0,B.columns())); // ionosphere parameter
if((norm(eqn.sigma-eqn.sigma0) < 1e-8) && infoTec.update(We(0)))
infoTec.info = "STEC ("+eqn.receiver->name()+", "+eqn.transmitter->name()+", "+eqn.timeRecv.dateTimeStr()+")";
// STEC statistics
const Double STEC = gnss->receivers.at(idRecv)->observation(idTrans, idEpoch)->dSTEC;
minSTEC = std::min(STEC, minSTEC);
maxSTEC = std::max(STEC, maxSTEC);
meanSTEC += STEC;
stdSTEC += STEC*STEC;
countSTEC++;
// remove STEC from residuals
We.row(0, B.columns()).setNull();
QMult(B, tau, We);
// influence of B parameters = B(B'B)^(-1)B'
QTransMult(B, tau, AWz);
AWz.row(0, B.columns()).setNull();
for(UInt i=0; i<B.columns(); i++)
AWz(i,i) = 1.0;
QMult(B, tau, AWz);
}
// redundancies
// ------------
Vector r(We.rows());
for(UInt i=0; i<We.rows(); i++)
r(i) = 1. - quadsum(AWz.row(i));
// find max. residual (for statistics)
// -----------------------------------
if(norm(eqn.sigma-eqn.sigma0) < 1e-8) // without outlier
for(UInt k=0; k<eqn.types.size(); k++)
{
const UInt idType = GnssType::index(typesResiduals, eqn.types.at(k));
if(infosResiduals.at(idType).update(1e3*(We(k)*eqn.sigma(k) - gnss->receivers.at(idRecv)->observation(idTrans, idEpoch)->at(eqn.types.at(k)).residuals)))
infosResiduals.at(idType).info = (typesResiduals.at(idType)+eqn.transmitter->PRN()).str()+", ("+ eqn.receiver->name()+", "+gnss->times.at(idEpoch).dateTimeStr()+")";
} // for(k)
gnss->receivers.at(idRecv)->observation(idTrans, idEpoch)->setDecorrelatedResiduals(eqn.types, We, r);
} // for(idTrans)
} // for(idRecv)
} // for(idEpoch)
Parallel::barrier(normalEquationInfo.comm);
timer.loopEnd();
// new weights
// -----------
if(computeWeights || adjustSigma0)
{
if(computeWeights) logStatus<<"Downweight outliers"<<Log::endl;
if(adjustSigma0) logStatus<<"Estimate variance factors"<<Log::endl;
for(auto recv : gnss->receivers)
if(normalEquationInfo.estimateReceiver.at(recv->idRecv()) && recv->isMyRank())
for(UInt iter=0; iter<10; iter++)
{
// --- lambda -------------------------------------
auto getResiduals = [&](UInt idEpoch, UInt idTrans, Bool useSigma0, std::vector<GnssType> &types, std::vector<Double> &ePe, std::vector<Double> &redundancy)
{
GnssObservation &obs = *(recv->observation(idTrans, idEpoch));
for(UInt idType=0; idType<obs.size(); idType++)
if((obs.at(idType).sigma0 > 0) && (obs.at(idType).redundancy > 0))
{
UInt idx = GnssType::index(types, obs.at(idType).type);
if(idx == NULLINDEX) // new type?
{
idx = types.size();
if(obs.at(idType).type == GnssType::PHASE)
types.push_back(obs.at(idType).type & (GnssType::TYPE + GnssType::SYSTEM)); // only PHASE and System
else
types.push_back(obs.at(idType).type & (GnssType::TYPE + GnssType::FREQUENCY + GnssType::ATTRIBUTE + GnssType::SYSTEM));
ePe.push_back(0);
redundancy.push_back(0);
}
ePe.at(idx) += std::pow(obs.at(idType).residuals/(useSigma0 ? obs.at(idType).sigma0 : obs.at(idType).sigma), 2);
redundancy.at(idx) += obs.at(idType).redundancy;
}
};
// ------------------------------------------------
if(adjustSigma0)
{
std::vector<GnssType> types;
std::vector<Double> ePe, redundancy;
for(UInt idEpoch : normalEquationInfo.idEpochs)
if(recv->useable(idEpoch))
for(UInt idTrans=0; idTrans<recv->idTransmitterSize(idEpoch); idTrans++)
if(recv->observation(idTrans, idEpoch))
getResiduals(idEpoch, idTrans, FALSE/*useSigma0*/, types, ePe, redundancy);
std::vector<Double> factors(types.size(), 1.);
for(UInt idType=0; idType<types.size(); idType++)
if(redundancy.at(idType) > 3)
factors.at(idType) = Vce::standardDeviation(ePe.at(idType), redundancy.at(idType), huber, huberPower);
// adjust sigma and sigma0
UInt idx;
for(UInt idEpoch=0; idEpoch<recv->idEpochSize(); idEpoch++)
for(UInt idTrans=0; idTrans<recv->idTransmitterSize(idEpoch); idTrans++)
if(recv->observation(idTrans, idEpoch))
{
GnssObservation &obs = *(recv->observation(idTrans, idEpoch));
for(UInt idType=0; idType<obs.size(); idType++)
if((obs.at(idType).sigma0 > 0) && obs.at(idType).type.isInList(types, idx))
{
obs.at(idType).sigma0 *= factors.at(idx);
obs.at(idType).sigma *= factors.at(idx);
} // for(idType)
} // for(idTrans, idEpoch)
// apply old factors to store total factor
for(UInt idType=0; idType<types.size(); idType++)
if(types.at(idType).isInList(sigmaType.at(recv->idRecv()), idx))
factors.at(idType) *= sigmaFactor.at(recv->idRecv()).at(idx);
sigmaType.at(recv->idRecv()) = types;
sigmaFactor.at(recv->idRecv()) = factors;
} // if(adjustSigma0)
if(computeWeights)
{
for(UInt idEpoch : normalEquationInfo.idEpochs)
if(recv->useable(idEpoch))
for(UInt idTrans=0; idTrans<recv->idTransmitterSize(idEpoch); idTrans++)
if(recv->observation(idTrans, idEpoch))
{
std::vector<GnssType> types;
std::vector<Double> ePe, redundancy;
getResiduals(idEpoch, idTrans, TRUE/*useSigma0*/, types, ePe, redundancy);
Vector factors(types.size(), 1.);
for(UInt i=0; i<types.size(); i++)
if(redundancy.at(i) > 0.1)
{
const Double s = std::sqrt(ePe.at(i)/redundancy.at(i));
factors(i) = ((s > huber) ? std::pow(s/huber, huberPower) : 1.);
}
GnssObservation &obs = *(recv->observation(idTrans, idEpoch));
UInt idx;
for(UInt idType=0; idType<obs.size(); idType++)
if(obs.at(idType).type.isInList(types, idx))
obs.at(idType).sigma = obs.at(idType).sigma0 * factors(idx);
}
} // if(computeWeights)
if(!(computeWeights && adjustSigma0))
break; // iteration not needed
} // for(recv, iter)
}
// residual analysis
// -----------------
std::vector<GnssType> types = typesResiduals;
std::vector<Double> ePe(types.size(), 0), redundancy(types.size(), 0);
std::vector<UInt> obsCount(types.size(), 0), outlierCount(types.size(), 0);
residualsStatistics(NULLINDEX/*idRecv*/, NULLINDEX/*idTrans*/, types, ePe, redundancy, obsCount, outlierCount);
if(Parallel::isMaster(normalEquationInfo.comm))
for(UInt i=0; i<types.size(); i++)
if(obsCount.at(i))
{
logInfo<<" "<<types.at(i).str()
<<": sigma0 = " <<Vce::standardDeviation(ePe.at(i), redundancy.at(i), huber, huberPower)%"%4.2f"s
<<", redundancy = "<<(redundancy.at(i)/obsCount.at(i))%"%4.2f"s
<<", count = " <<obsCount.at(i)%"%7i"s
<<", outliers = " <<outlierCount.at(i)%"%6i"s<<" ("<<(100.*outlierCount.at(i)/obsCount.at(i))%"%4.2f"s<<" %)"
<<Log::endl;
}
logInfo<<"Residuals changes (without outliers)"<<Log::endl;
Double maxChange = 0;
for(auto &info : infosResiduals)
info.synchronizeAndPrint(normalEquationInfo.comm, 1e-3, maxChange);
logInfo<<"Parameter changes"<<Log::endl;
// STEC analysis
// -------------
Parallel::reduceSum(countSTEC, 0, normalEquationInfo.comm);
Parallel::broadCast(countSTEC, 0, normalEquationInfo.comm);
if(countSTEC)
{
Parallel::reduceMin(minSTEC, 0, normalEquationInfo.comm);
Parallel::reduceMax(maxSTEC, 0, normalEquationInfo.comm);
Parallel::reduceSum(meanSTEC, 0, normalEquationInfo.comm);
Parallel::reduceSum(stdSTEC, 0, normalEquationInfo.comm);
stdSTEC = std::sqrt((stdSTEC-meanSTEC*meanSTEC/countSTEC)/(countSTEC-1));
meanSTEC /= countSTEC;
std::string infoTecStr = " (total: "+meanSTEC%"%.2f +- "s+stdSTEC%"%.2f ["s+minSTEC%"%.2f -- "s+maxSTEC%"%.2f])"s;
Parallel::broadCast(infoTecStr, 0, normalEquationInfo.comm);
infoTec.info += infoTecStr;
infoTec.synchronizeAndPrint(normalEquationInfo.comm, 0, maxChange);
}
return gnss->updateParameter(normalEquationInfo, x, Wz);
}
catch(std::exception &e)
{
GROOPS_RETHROW(e)
}
}
/***********************************************/
void GnssProcessingStep::State::residualsStatistics(UInt idRecv, UInt idTrans,
std::vector<GnssType> &types, std::vector<Double> &ePe, std::vector<Double> &redundancy,
std::vector<UInt> &obsCount, std::vector<UInt> &outlierCount)
{
try
{
const UInt idTransStart = ((idTrans!=NULLINDEX) ? idTrans : 0);
const UInt idTransEnd = ((idTrans!=NULLINDEX) ? idTrans : MAX_UINT);
for(auto recv : gnss->receivers)
if(normalEquationInfo.estimateReceiver.at(recv->idRecv()) && recv->isMyRank() && ((idRecv == NULLINDEX) || (idRecv == recv->idRecv())))
for(UInt idEpoch : normalEquationInfo.idEpochs)
if(recv->useable(idEpoch))
for(UInt idTrans=idTransStart; (idTrans<=idTransEnd) && (idTrans<recv->idTransmitterSize(idEpoch)); idTrans++)
if(recv->observation(idTrans, idEpoch))
{
const GnssObservation &obs = *recv->observation(idTrans, idEpoch);
UInt idx;
for(UInt idType=0; idType<obs.size(); idType++)
if((obs.at(idType).sigma0 > 0) && (obs.at(idType).type.isInList(types, idx)))
{
ePe.at(idx) += std::pow(obs.at(idType).residuals/obs.at(idType).sigma, 2);
redundancy.at(idx) += obs.at(idType).redundancy;
obsCount.at(idx)++;
if(obs.at(idType).sigma > obs.at(idType).sigma0)
outlierCount.at(idx)++;
}
} // for(idTrans, idEpoch)
for(UInt i=0; i<types.size(); i++)
{
Parallel::reduceSum(ePe.at(i), 0, normalEquationInfo.comm);
Parallel::reduceSum(redundancy.at(i), 0, normalEquationInfo.comm);
Parallel::reduceSum(obsCount.at(i), 0, normalEquationInfo.comm);
Parallel::reduceSum(outlierCount.at(i), 0, normalEquationInfo.comm);
}
}
catch(std::exception &e)
{
GROOPS_RETHROW(e)
}
}
/***********************************************/
|
