1 | /* ======================================================================== *\
|
---|
2 | !
|
---|
3 | ! *
|
---|
4 | ! * This file is part of MARS, the MAGIC Analysis and Reconstruction
|
---|
5 | ! * Software. It is distributed to you in the hope that it can be a useful
|
---|
6 | ! * and timesaving tool in analysing Data of imaging Cerenkov telescopes.
|
---|
7 | ! * It is distributed WITHOUT ANY WARRANTY.
|
---|
8 | ! *
|
---|
9 | ! * Permission to use, copy, modify and distribute this software and its
|
---|
10 | ! * documentation for any purpose is hereby granted without fee,
|
---|
11 | ! * provided that the above copyright notice appear in all copies and
|
---|
12 | ! * that both that copyright notice and this permission notice appear
|
---|
13 | ! * in supporting documentation. It is provided "as is" without express
|
---|
14 | ! * or implied warranty.
|
---|
15 | ! *
|
---|
16 | !
|
---|
17 | ! Author(s): Wolfgang Wittek 07/2004 <mailto:wittek@mppmu.mpg.de>
|
---|
18 | !
|
---|
19 | ! Copyright: MAGIC Software Development, 2000-2004
|
---|
20 | !
|
---|
21 | !
|
---|
22 | \* ======================================================================== */
|
---|
23 |
|
---|
24 | /////////////////////////////////////////////////////////////////////////////
|
---|
25 | //
|
---|
26 | // MTelAxisFromStars
|
---|
27 | //
|
---|
28 | // This task
|
---|
29 | // - determines the transformation from expected positions of stars
|
---|
30 | // in the camera to measured positions of these stars in the camera
|
---|
31 | // - applies this transformation to expected positions of other objects
|
---|
32 | // to obtain the estimated positions of these objects in the camera
|
---|
33 | // - puts the estimated positions into the relevant containers
|
---|
34 | //
|
---|
35 | // Input Containers :
|
---|
36 | // MStarCam[MStarCam], MStarCamSource[MStarCam]
|
---|
37 | //
|
---|
38 | // Output Containers :
|
---|
39 | // MSkyCamTrans, MSrcPosCam
|
---|
40 | //
|
---|
41 | /////////////////////////////////////////////////////////////////////////////
|
---|
42 | #include <TMath.h>
|
---|
43 | #include <TList.h>
|
---|
44 | #include <TSystem.h>
|
---|
45 |
|
---|
46 | #include <fstream>
|
---|
47 |
|
---|
48 | #include "MTelAxisFromStars.h"
|
---|
49 |
|
---|
50 | #include "MParList.h"
|
---|
51 |
|
---|
52 |
|
---|
53 | #include "MLog.h"
|
---|
54 | #include "MLogManip.h"
|
---|
55 |
|
---|
56 | #include "MReportDrive.h"
|
---|
57 | #include "MPointingPos.h"
|
---|
58 | #include "MSrcPosCam.h"
|
---|
59 |
|
---|
60 | #include "MStarCam.h"
|
---|
61 | #include "MStarPos.h"
|
---|
62 | #include "MSkyCamTrans.h"
|
---|
63 | #include "MStarCamTrans.h"
|
---|
64 |
|
---|
65 | ClassImp(MTelAxisFromStars);
|
---|
66 |
|
---|
67 | using namespace std;
|
---|
68 |
|
---|
69 | // --------------------------------------------------------------------------
|
---|
70 | //
|
---|
71 | // Constructor
|
---|
72 | //
|
---|
73 | MTelAxisFromStars::MTelAxisFromStars(const char *name, const char *title)
|
---|
74 | {
|
---|
75 | fName = name ? name : "MTelAxisFromStars";
|
---|
76 | fTitle = title ? title : "Calculate source position from star positions";
|
---|
77 |
|
---|
78 | // if scale factor fLambda should NOT be fixed set fFixdScaleFactor to
|
---|
79 | // -1.0; otherwise set it to the value requested
|
---|
80 | fFixedScaleFactor = 1.0;
|
---|
81 |
|
---|
82 | // if rotation angle fAlfa should NOT be fixed set fFixdRotationAngle to
|
---|
83 | // -1.0; otherwise set it to the requested value
|
---|
84 | fFixedRotationAngle = 0.0;
|
---|
85 |
|
---|
86 | // default type of input is : the result of the Gauss fit
|
---|
87 | // type 0 : result from the weighted average
|
---|
88 | // type 1 : result from the Gauss fit
|
---|
89 | fInputType = 1;
|
---|
90 |
|
---|
91 | // default value of fAberr
|
---|
92 | // the value 1.07 is valid if the expected position (with aberration)
|
---|
93 | // in the camera is calculated as the average of the positions obtained
|
---|
94 | // from the reflections at the individual mirrors
|
---|
95 | fAberr = 1.07;
|
---|
96 |
|
---|
97 | }
|
---|
98 |
|
---|
99 | // --------------------------------------------------------------------------
|
---|
100 | //
|
---|
101 | // Destructor
|
---|
102 | //
|
---|
103 | MTelAxisFromStars::~MTelAxisFromStars()
|
---|
104 | {
|
---|
105 | delete fStarCamTrans;
|
---|
106 | }
|
---|
107 |
|
---|
108 | // --------------------------------------------------------------------------
|
---|
109 | //
|
---|
110 | // Set links to containers
|
---|
111 | //
|
---|
112 |
|
---|
113 | Int_t MTelAxisFromStars::PreProcess(MParList *pList)
|
---|
114 | {
|
---|
115 | fDrive = (MReportDrive*)pList->FindObject(AddSerialNumber("MReportDrive"));
|
---|
116 | if (!fDrive)
|
---|
117 | {
|
---|
118 | *fLog << err << AddSerialNumber("MReportDrive")
|
---|
119 | << " not found... aborting." << endl;
|
---|
120 | return kFALSE;
|
---|
121 | }
|
---|
122 |
|
---|
123 |
|
---|
124 | fStarCam = (MStarCam*)pList->FindObject("MStarCam", "MStarCam");
|
---|
125 | if (!fStarCam)
|
---|
126 | {
|
---|
127 | *fLog << err << "MStarCam not found... aborting." << endl;
|
---|
128 | return kFALSE;
|
---|
129 | }
|
---|
130 |
|
---|
131 |
|
---|
132 | fSourceCam = (MStarCam*)pList->FindObject("MSourceCam", "MStarCam");
|
---|
133 | if (!fSourceCam)
|
---|
134 | {
|
---|
135 | *fLog << warn << "MSourceCam[MStarCam] not found... continue " << endl;
|
---|
136 | }
|
---|
137 |
|
---|
138 |
|
---|
139 | //-----------------------------------------------------------------
|
---|
140 | fSrcPos = (MSrcPosCam*)pList->FindCreateObj(AddSerialNumber("MSrcPosCam"), "MSrcPosCam");
|
---|
141 | if (!fSrcPos)
|
---|
142 | return kFALSE;
|
---|
143 |
|
---|
144 | fPntPos = (MSrcPosCam*)pList->FindCreateObj(AddSerialNumber("MSrcPosCam"), "MPntPosCam");
|
---|
145 | if (!fPntPos)
|
---|
146 | return kFALSE;
|
---|
147 |
|
---|
148 | fPointPos = (MPointingPos*)pList->FindCreateObj(AddSerialNumber("MPointingPos"), "MPointingPos");
|
---|
149 | if (!fPointPos)
|
---|
150 | return kFALSE;
|
---|
151 |
|
---|
152 | fSourcePos = (MPointingPos*)pList->FindCreateObj(AddSerialNumber("MPointingPos"), "MSourcePos");
|
---|
153 | if (!fSourcePos)
|
---|
154 | return kFALSE;
|
---|
155 |
|
---|
156 | fSkyCamTrans = (MSkyCamTrans*)pList->FindCreateObj(AddSerialNumber("MSkyCamTrans"));
|
---|
157 | if (!fSkyCamTrans)
|
---|
158 | return kFALSE;
|
---|
159 |
|
---|
160 |
|
---|
161 | //-----------------------------------------------------------------
|
---|
162 | // book an MStarCamTrans object
|
---|
163 | // this is needed when calling one of the member functions of MStarCamTrans
|
---|
164 |
|
---|
165 | MGeomCam *geom = (MGeomCam*)pList->FindObject(AddSerialNumber("MGeomCam"));
|
---|
166 | if (!geom)
|
---|
167 | {
|
---|
168 | *fLog << err << AddSerialNumber("MGeomCam")
|
---|
169 | << " not found... aborting." << endl;
|
---|
170 | return kFALSE;
|
---|
171 | }
|
---|
172 |
|
---|
173 | MObservatory *obs = (MObservatory*)pList->FindObject(AddSerialNumber("MObservatory"));
|
---|
174 | if (!obs)
|
---|
175 | {
|
---|
176 | *fLog << err << AddSerialNumber("MObservatory")
|
---|
177 | << " not found... aborting." << endl;
|
---|
178 | return kFALSE;
|
---|
179 | }
|
---|
180 |
|
---|
181 | //-----------------------------------------------------------------
|
---|
182 | fStarCamTrans = new MStarCamTrans(*geom, *obs);
|
---|
183 |
|
---|
184 | *fLog << all << "MTelAxisFromStars::Preprocess; the optical aberration factor is set equal to : "
|
---|
185 | << fAberr ;
|
---|
186 |
|
---|
187 | *fLog << all << "MTelAxisFromStars::Preprocess; input type is set equal to : "
|
---|
188 | << fInputType ;
|
---|
189 | if (fInputType == 0)
|
---|
190 | *fLog << " (calculated star positions)" << endl;
|
---|
191 | else
|
---|
192 | *fLog << " (fitted star positions)" << endl;
|
---|
193 |
|
---|
194 | *fLog << all << "MTelAxisFromStars::Preprocess; scale factor will be fixed at : "
|
---|
195 | << fFixedScaleFactor << endl;
|
---|
196 |
|
---|
197 | *fLog << all << "MTelAxisFromStars::Preprocess; rotation angle will be fixed at : "
|
---|
198 | << fFixedRotationAngle << endl;
|
---|
199 |
|
---|
200 | return kTRUE;
|
---|
201 | }
|
---|
202 |
|
---|
203 | // --------------------------------------------------------------------------
|
---|
204 | //
|
---|
205 | // Set optical aberration factor
|
---|
206 | //
|
---|
207 | // fAberr is the ratio between
|
---|
208 | // the distance from the camera center with optical aberration and
|
---|
209 | // the distance from the camera center with an ideal imaging
|
---|
210 | //
|
---|
211 | // fAberr = r_real/r_ideal
|
---|
212 | //
|
---|
213 | void MTelAxisFromStars::SetOpticalAberr(Double_t aberr)
|
---|
214 | {
|
---|
215 | fAberr = aberr;
|
---|
216 | }
|
---|
217 |
|
---|
218 | // --------------------------------------------------------------------------
|
---|
219 | //
|
---|
220 | // Set the type of the input
|
---|
221 | //
|
---|
222 | // type = 0 calculated star positions (by averaging)
|
---|
223 | // type = 1 fitted star positions (by Gauss fit)
|
---|
224 | //
|
---|
225 | void MTelAxisFromStars::SetInputType(Int_t type)
|
---|
226 | {
|
---|
227 | fInputType = type;
|
---|
228 | }
|
---|
229 |
|
---|
230 | // --------------------------------------------------------------------------
|
---|
231 | //
|
---|
232 | // Fix the scale factor fLambda
|
---|
233 | //
|
---|
234 | //
|
---|
235 | void MTelAxisFromStars::FixScaleFactorAt(Double_t lambda)
|
---|
236 | {
|
---|
237 | fFixedScaleFactor = lambda;
|
---|
238 | }
|
---|
239 |
|
---|
240 |
|
---|
241 | // --------------------------------------------------------------------------
|
---|
242 | //
|
---|
243 | // Fix rotation angle fAlfa
|
---|
244 | //
|
---|
245 | //
|
---|
246 | void MTelAxisFromStars::FixRotationAngleAt(Double_t alfa)
|
---|
247 | {
|
---|
248 | fFixedRotationAngle = alfa; // [degrees]
|
---|
249 | }
|
---|
250 |
|
---|
251 |
|
---|
252 | // --------------------------------------------------------------------------
|
---|
253 | //
|
---|
254 | // Process
|
---|
255 | //
|
---|
256 | // call FindSkyCamTrans to find the Sky-Camera transformation
|
---|
257 | // call TransSkyCam to transform some sky directions
|
---|
258 | // into the camera system
|
---|
259 | // put the estimated source position into MSrcPosCam
|
---|
260 | //
|
---|
261 |
|
---|
262 | Int_t MTelAxisFromStars::Process()
|
---|
263 | {
|
---|
264 | //Int_t run = fRun->GetRunNumber();
|
---|
265 | //*fLog << "MTelAxisFromStars::Process; run = " << run << endl;
|
---|
266 |
|
---|
267 | //--------------------------------------
|
---|
268 | // Define the input for FindSkyCamTrans
|
---|
269 | //
|
---|
270 |
|
---|
271 | // get the expected (axy[0], axy[1]) and the measured positions
|
---|
272 | // (bxy[0], bxy[1]) of stars in the camera from MStarCam
|
---|
273 | Int_t fNumStars = fStarCam->GetNumStars();
|
---|
274 |
|
---|
275 | if (fNumStars <= 0)
|
---|
276 | return kTRUE;
|
---|
277 |
|
---|
278 | TArrayD axy[2];
|
---|
279 | axy[0].Set(fNumStars);
|
---|
280 | axy[1].Set(fNumStars);
|
---|
281 |
|
---|
282 | TArrayD bxy[2];
|
---|
283 | bxy[0].Set(fNumStars);
|
---|
284 | bxy[1].Set(fNumStars);
|
---|
285 |
|
---|
286 | // error matrix of bxy
|
---|
287 | TArrayD exy[2][2];
|
---|
288 | exy[0][0].Set(fNumStars);
|
---|
289 | exy[0][1].Set(fNumStars);
|
---|
290 | exy[1][0].Set(fNumStars);
|
---|
291 | exy[1][1].Set(fNumStars);
|
---|
292 |
|
---|
293 | // transformation parameters
|
---|
294 | Double_t fLambda;
|
---|
295 | Double_t fAlfa;
|
---|
296 | Double_t fA[2][2];
|
---|
297 | Double_t fD[2];
|
---|
298 | Double_t fErrD[2][2];
|
---|
299 | Int_t fNumIter;
|
---|
300 | Int_t fNdof;
|
---|
301 | Double_t fChi2;
|
---|
302 | Double_t fChi2Prob;
|
---|
303 |
|
---|
304 | MStarPos *star = 0;
|
---|
305 | TIter next(fStarCam->GetList());
|
---|
306 | Int_t ix = 0;
|
---|
307 |
|
---|
308 | // loop over all stars
|
---|
309 | while ( (star = (MStarPos*)next()) )
|
---|
310 | {
|
---|
311 | axy[0][ix] = star->GetXExp();
|
---|
312 | axy[1][ix] = star->GetYExp();
|
---|
313 |
|
---|
314 | if (fInputType == 0)
|
---|
315 | {
|
---|
316 | // values from averaging
|
---|
317 | bxy[0][ix] = star->GetMeanXCalc();
|
---|
318 | bxy[1][ix] = star->GetMeanYCalc();
|
---|
319 |
|
---|
320 | // this is the error matrix for (MeanXCalc, MeanYCalc);
|
---|
321 | // this is the error matrix which should be used
|
---|
322 | exy[0][0][ix] = star->GetXXErrCalc();
|
---|
323 | exy[0][1][ix] = star->GetXYErrCalc();
|
---|
324 | exy[1][0][ix] = star->GetXYErrCalc();
|
---|
325 | exy[1][1][ix] = star->GetYYErrCalc();
|
---|
326 |
|
---|
327 | //exy[0][0][ix] = star->GetSigmaXCalc()*star->GetSigmaXCalc();
|
---|
328 | //exy[0][1][ix] = 0.0;
|
---|
329 | //exy[1][0][ix] = 0.0;
|
---|
330 | //exy[1][1][ix] = star->GetSigmaYCalc()*star->GetSigmaYCalc();
|
---|
331 | }
|
---|
332 |
|
---|
333 | else if (fInputType == 1)
|
---|
334 | {
|
---|
335 | // values from Gauss fit
|
---|
336 | bxy[0][ix] = star->GetMeanXFit();
|
---|
337 | bxy[1][ix] = star->GetMeanYFit();
|
---|
338 |
|
---|
339 | // this is the error matrix for (MeanXFit, MeanYFit);
|
---|
340 | // this is the error matrix which should be used
|
---|
341 | exy[0][0][ix] = star->GetXXErrFit();
|
---|
342 | exy[0][1][ix] = star->GetXYErrFit();
|
---|
343 | exy[1][0][ix] = star->GetXYErrFit();
|
---|
344 | exy[1][1][ix] = star->GetYYErrFit();
|
---|
345 |
|
---|
346 | // this is the error matrix constructed from SigmaXFit and SigmaYFit;
|
---|
347 | // it is used because the errors above are too small, at present
|
---|
348 | //exy[0][0][ix] = star->GetSigmaXFit() * star->GetSigmaXFit();
|
---|
349 | //exy[0][1][ix] = star->GetCorrXYFit() *
|
---|
350 | // star->GetSigmaXFit() * star->GetSigmaYFit();
|
---|
351 | //exy[1][0][ix] = exy[0][1][ix];
|
---|
352 | //exy[1][1][ix] = star->GetSigmaYFit() * star->GetSigmaYFit();
|
---|
353 | }
|
---|
354 |
|
---|
355 | else
|
---|
356 | {
|
---|
357 | *fLog << err << "MTelAxisFromStars::Process; type of input is not defined"
|
---|
358 | << endl;
|
---|
359 | return kFALSE;
|
---|
360 | }
|
---|
361 |
|
---|
362 | // don't include stars with undefined error
|
---|
363 | Double_t deter = exy[0][0][ix]*exy[1][1][ix]
|
---|
364 | - exy[0][1][ix]*exy[1][0][ix];
|
---|
365 |
|
---|
366 | //*fLog << "ix ,deter, xx, xy, yy = " << ix << ": "
|
---|
367 | // << deter << ", " << exy[0][0][ix] << ", "
|
---|
368 | // << exy[0][1][ix] << ", " << exy[1][1][ix] << endl;
|
---|
369 | if (deter <= 0.0)
|
---|
370 | continue;
|
---|
371 |
|
---|
372 | //*fLog << "MTelAxisFromStars : " << endl;
|
---|
373 | //*fLog << " ix, XExp, YExp, XFit, YFit, SigmaX2, SigmaXY, SigmaY2 = "
|
---|
374 | // << ix << " : "
|
---|
375 | // << axy[0][ix] << ", " << axy[1][ix] << ", "
|
---|
376 | // << bxy[0][ix] << ", " << bxy[1][ix] << ", "
|
---|
377 | // << exy[0][0][ix] << ", " << exy[0][1][ix] << ", "
|
---|
378 | // << exy[1][1][ix] << endl;
|
---|
379 |
|
---|
380 | ix++;
|
---|
381 | }
|
---|
382 |
|
---|
383 | //--------------------------------------
|
---|
384 | // Find the transformation from expected positions (axy[1], axy[2])
|
---|
385 | // to measured positions (bxy[1], bxy[2]) in the camera
|
---|
386 |
|
---|
387 | Int_t fNStars = ix;
|
---|
388 |
|
---|
389 | if (ix < fNumStars)
|
---|
390 | {
|
---|
391 | // reset the sizes of the arrays
|
---|
392 | Int_t fNStars = ix;
|
---|
393 | axy[0].Set(fNStars);
|
---|
394 | axy[1].Set(fNStars);
|
---|
395 |
|
---|
396 | bxy[0].Set(fNStars);
|
---|
397 | bxy[1].Set(fNStars);
|
---|
398 |
|
---|
399 | exy[0][0].Set(fNStars);
|
---|
400 | exy[0][1].Set(fNStars);
|
---|
401 | exy[1][0].Set(fNStars);
|
---|
402 | exy[1][1].Set(fNStars);
|
---|
403 | }
|
---|
404 |
|
---|
405 | Bool_t fitOK;
|
---|
406 | if (fNStars < 1)
|
---|
407 | {
|
---|
408 | //*fLog << "MTelAxisFromStars::Process; no star for MTelAxisFromStars"
|
---|
409 | // << endl;
|
---|
410 | fitOK = kFALSE;
|
---|
411 | }
|
---|
412 | else
|
---|
413 | {
|
---|
414 | fitOK = FindSkyCamTrans(axy, bxy, exy,
|
---|
415 | fFixedRotationAngle, fFixedScaleFactor, fLambda,
|
---|
416 | fAlfa , fA, fD, fErrD,
|
---|
417 | fNumIter, fNdof, fChi2, fChi2Prob);
|
---|
418 | }
|
---|
419 |
|
---|
420 | if (!fitOK && fNStars >= 1)
|
---|
421 | {
|
---|
422 | *fLog << err
|
---|
423 | << "MTelAxisFromStars::Process; Fit to find transformation from star to camera system failed"
|
---|
424 | << endl;
|
---|
425 |
|
---|
426 | //if (fNStars > 0)
|
---|
427 | //{
|
---|
428 | // *fLog << err
|
---|
429 | // << " fNumIter, fNdof, fChi2, fChi2Prob = " << fNumIter
|
---|
430 | // << ", " << fNdof << ", " << fChi2 << ", " << fChi2Prob << endl;
|
---|
431 | //}
|
---|
432 |
|
---|
433 | return kTRUE;
|
---|
434 | }
|
---|
435 |
|
---|
436 |
|
---|
437 | //--------------------------------------
|
---|
438 | // Put the transformation parameters into the MSkyCamTrans container
|
---|
439 |
|
---|
440 | fSkyCamTrans->SetParameters(fLambda, fAlfa, fA, fD, fErrD,
|
---|
441 | fNumStars, fNumIter, fNdof, fChi2, fChi2Prob);
|
---|
442 | fSkyCamTrans->SetReadyToSave();
|
---|
443 |
|
---|
444 | //--------------------------------------
|
---|
445 | // Put the camera position (X, Y)
|
---|
446 | // obtained by transforming the camera center (0, 0)
|
---|
447 | // into MPntPosCam[MSrcPosCam]
|
---|
448 |
|
---|
449 | fPntPos->SetXY(fD[0], fD[1]);
|
---|
450 | fPntPos->SetReadyToSave();
|
---|
451 |
|
---|
452 |
|
---|
453 | //--------------------------------------
|
---|
454 | // Put the corrected pointing position into MPointingPos
|
---|
455 | //
|
---|
456 | SetPointingPosition(fStarCamTrans, fDrive, fSkyCamTrans, fPointPos);
|
---|
457 |
|
---|
458 |
|
---|
459 | //--------------------------------------
|
---|
460 | // Put the estimated position of the source into SrcPosCam
|
---|
461 | //
|
---|
462 | // get the source direction from MReportDrive
|
---|
463 | // Note : this has to be changed for the wobble mode, where the source
|
---|
464 | // isn't in the center of the camera
|
---|
465 | Double_t decsource = fDrive->GetDec();
|
---|
466 | Double_t rasource = fDrive->GetRa();
|
---|
467 | //
|
---|
468 | Double_t radrive = fDrive->GetRa();
|
---|
469 | Double_t hdrive = fDrive->GetHa();
|
---|
470 | Double_t hsource = hdrive + radrive - rasource;
|
---|
471 | fSourcePos->SetSkyPosition(rasource, decsource, hsource);
|
---|
472 |
|
---|
473 | SetSourcePosition(fStarCamTrans, fPointPos, fSourcePos, fSrcPos);
|
---|
474 |
|
---|
475 | //*fLog << "after calling SetSourcePosition : , X, Y ,fD = "
|
---|
476 | // << fSrcPos->GetX() << ", " << fSrcPos->GetY() << ", "
|
---|
477 | // << fD[0] << ", " << fD[1] << endl;
|
---|
478 |
|
---|
479 | //--------------------------------------
|
---|
480 | // Apply the transformation to some expected positions (asxy[1], asxy[2])
|
---|
481 | // to obtain estimated positions (bsxy[1], bsxy[2]) in the camera
|
---|
482 | // and their error matrices esxy[2][2]
|
---|
483 |
|
---|
484 | // get the expected positions (asxy[1], asxy[2]) from another MStarCam
|
---|
485 | // container (with the name "MSourceCam")
|
---|
486 | Int_t fNumStarsSource = 0;
|
---|
487 |
|
---|
488 | if (fSourceCam)
|
---|
489 | fNumStarsSource = fSourceCam->GetNumStars();
|
---|
490 |
|
---|
491 | //*fLog << "MTelAxisFromStars::Process; fNumStarsSource = "
|
---|
492 | // << fNumStarsSource << endl;
|
---|
493 |
|
---|
494 | if (fNumStarsSource > 0)
|
---|
495 | {
|
---|
496 | TArrayD asxy[2];
|
---|
497 | asxy[0].Set(fNumStarsSource);
|
---|
498 | asxy[1].Set(fNumStarsSource);
|
---|
499 |
|
---|
500 | TArrayD bsxy[2];
|
---|
501 | bsxy[0].Set(fNumStarsSource);
|
---|
502 | bsxy[1].Set(fNumStarsSource);
|
---|
503 |
|
---|
504 | TArrayD esxy[2][2];
|
---|
505 | esxy[0][0].Set(fNumStarsSource);
|
---|
506 | esxy[0][1].Set(fNumStarsSource);
|
---|
507 | esxy[1][0].Set(fNumStarsSource);
|
---|
508 | esxy[1][1].Set(fNumStarsSource);
|
---|
509 |
|
---|
510 | MStarPos *starSource = 0;
|
---|
511 | TIter nextSource(fSourceCam->GetList());
|
---|
512 | ix = 0;
|
---|
513 | while ( (starSource = (MStarPos*)nextSource()) )
|
---|
514 | {
|
---|
515 | asxy[0][ix] = starSource->GetXExp();
|
---|
516 | asxy[1][ix] = starSource->GetYExp();
|
---|
517 |
|
---|
518 | ix++;
|
---|
519 | }
|
---|
520 |
|
---|
521 | TransSkyCam(fLambda, fA, fD, fErrD, asxy, bsxy, esxy);
|
---|
522 |
|
---|
523 | // put the estimated positions into the MStarCam container
|
---|
524 | // with name "MSourceCam"
|
---|
525 | TIter setnextSource(fSourceCam->GetList());
|
---|
526 | ix = 0;
|
---|
527 | while ( (starSource = (MStarPos*)setnextSource()) )
|
---|
528 | {
|
---|
529 | Double_t corr = esxy[0][1][ix] /
|
---|
530 | TMath::Sqrt( esxy[0][0][ix] * esxy[1][1][ix] );
|
---|
531 | starSource->SetFitValues(100.0, 100.0, bsxy[0][ix], bsxy[1][ix],
|
---|
532 | TMath::Sqrt(esxy[0][0][ix]), TMath::Sqrt(esxy[1][1][ix]), corr,
|
---|
533 | esxy[0][0][ix], esxy[0][1][ix], esxy[1][1][ix],
|
---|
534 | fChi2, fNdof);
|
---|
535 |
|
---|
536 | ix++;
|
---|
537 | }
|
---|
538 |
|
---|
539 | }
|
---|
540 |
|
---|
541 | //--------------------------------------
|
---|
542 |
|
---|
543 | return kTRUE;
|
---|
544 | }
|
---|
545 |
|
---|
546 |
|
---|
547 |
|
---|
548 | //---------------------------------------------------------------------------
|
---|
549 | //
|
---|
550 | // FindSkyCamTrans
|
---|
551 | //
|
---|
552 | // This routine determines the transformation
|
---|
553 | //
|
---|
554 | // ( cos(alfa) -sin(alfa) )
|
---|
555 | // b = lambda * A * a + d A = ( )
|
---|
556 | // ^ ^ ^ ( sin(alfa) cos(alfa) )
|
---|
557 | // | | |
|
---|
558 | // scale rotation shift
|
---|
559 | // factor matrix
|
---|
560 | //
|
---|
561 | // from sky coordinates 'a' (projected onto the camera) to camera
|
---|
562 | // coordinates 'b', using the positions of known stars in the camera.
|
---|
563 | // The latter positions may have been determined by analysing the
|
---|
564 | // DC currents in the different pixels.
|
---|
565 | //
|
---|
566 | // Input : a[2] x and y coordinates of stars projected onto the camera;
|
---|
567 | // they were obtained from (RA, dec) of the stars and
|
---|
568 | // (ThetaTel, PhiTel) and the time of observation;
|
---|
569 | // these are the 'expected positions' of stars in the camera
|
---|
570 | // b[2] 'measured positions' of these stars in the camera;
|
---|
571 | // they may have been obtained from the DC currents
|
---|
572 | // e[2][2] error matrix of b[2]
|
---|
573 | // fixedrotationangle value [in degrees] at which rotation angle
|
---|
574 | // alfa should be fixed; -1 means don't fix
|
---|
575 | // fixedscalefactor value at which scale factor lambda
|
---|
576 | // should be fixed; -1 means don't fix
|
---|
577 | //
|
---|
578 | // Output : lambda, alfadeg, A[2][2], d[2] fit results;
|
---|
579 | // parameters describing the transformation
|
---|
580 | // from 'expected positions' to the 'measured
|
---|
581 | // positions' in the camera
|
---|
582 | // errd[2][2] error matrix of d[2]
|
---|
583 | // fNumIter number of iterations
|
---|
584 | // fNdoF number of degrees of freedom
|
---|
585 | // fChi2 chi-square value
|
---|
586 | // fChi2Prob chi-square probability
|
---|
587 | //
|
---|
588 | // The units are assumed to be
|
---|
589 | // [degrees] for alfadeg
|
---|
590 | // [mm] for a, b, d
|
---|
591 | // [1] for lambda
|
---|
592 |
|
---|
593 | Bool_t MTelAxisFromStars::FindSkyCamTrans(
|
---|
594 | TArrayD a[2], TArrayD b[2], TArrayD e[2][2],
|
---|
595 | Double_t &fixedrotationang, Double_t &fixedscalefac, Double_t &lambda,
|
---|
596 | Double_t &alfadeg, Double_t A[2][2], Double_t d[2], Double_t errd[2][2],
|
---|
597 | Int_t &fNumIter, Int_t &fNdof, Double_t &fChi2, Double_t &fChi2Prob)
|
---|
598 | {
|
---|
599 | Int_t fNumStars = a[0].GetSize();
|
---|
600 |
|
---|
601 | //*fLog << "MTelAxisFromStars::FindSkyCamTrans; expected and measured positions :"
|
---|
602 | // << endl;
|
---|
603 | for (Int_t ix=0; ix<fNumStars; ix++)
|
---|
604 | {
|
---|
605 | //*fLog << " ix, a[0], a[1], b[0], b[1], errxx, errxy, erryy = "
|
---|
606 | // << ix << " : "
|
---|
607 | // << a[0][ix] << ", " << a[1][ix] << "; "
|
---|
608 | // << b[0][ix] << ", " << b[1][ix] << "; "
|
---|
609 | // << e[0][0][ix] << ", " << e[0][1][ix] << ", "
|
---|
610 | // << e[1][1][ix] << endl;
|
---|
611 | }
|
---|
612 |
|
---|
613 |
|
---|
614 | //-------------------------------------------
|
---|
615 | // fix some parameters if the number of degrees of freedom is too low
|
---|
616 | // (<= 0.0)
|
---|
617 |
|
---|
618 | Double_t fixedscalefactor = fixedscalefac;
|
---|
619 | Double_t fixedrotationangle = fixedrotationang;
|
---|
620 |
|
---|
621 | // calculate number of degrees of freedom
|
---|
622 | fNdof = 2 * fNumStars - 4;
|
---|
623 | if (fixedscalefactor != -1.0)
|
---|
624 | fNdof += 1;
|
---|
625 | if (fixedrotationangle != -1.0)
|
---|
626 | fNdof += 1;
|
---|
627 |
|
---|
628 | // if there is only 1 star fix both rotation angle and scale factor
|
---|
629 | if (fNumStars == 1)
|
---|
630 | {
|
---|
631 | if (fixedscalefactor == -1.0)
|
---|
632 | {
|
---|
633 | fixedscalefactor = 1.0;
|
---|
634 | *fLog << warn << "MTelAxisFromStars::FindSkyCamTrans; scale factor is fixed at "
|
---|
635 | << fixedscalefactor << endl;
|
---|
636 | }
|
---|
637 | if (fixedrotationangle == -1.0)
|
---|
638 | {
|
---|
639 | fixedrotationangle = 0.0;
|
---|
640 | *fLog << warn << "MTelAxisFromStars::FindSkyCamTrans; rotation angle is fixed at "
|
---|
641 | << fixedrotationangle << endl;
|
---|
642 | }
|
---|
643 | }
|
---|
644 | // otherwise fix only 1 parameter if possible
|
---|
645 | else if (fNdof < 0)
|
---|
646 | {
|
---|
647 | if (fNdof < -2)
|
---|
648 | {
|
---|
649 | *fLog << warn << "MTelAxisFromStars::FindSkyCamTrans; number of degrees of freedom is too low : "
|
---|
650 | << fNdof << "; fNumStars = " << fNumStars << endl;
|
---|
651 | return kFALSE;
|
---|
652 | }
|
---|
653 | else if (fNdof == -2)
|
---|
654 | {
|
---|
655 | if (fixedscalefactor == -1.0 && fixedrotationangle == -1.0)
|
---|
656 | {
|
---|
657 | fixedscalefactor = 1.0;
|
---|
658 | fixedrotationangle = 0.0;
|
---|
659 | *fLog << warn << "MTelAxisFromStars::FindSkyCamTrans; scale factor and rotation angle are fixed at "
|
---|
660 | << fixedscalefactor << " and " << fixedrotationangle
|
---|
661 | << " respectively" << endl;
|
---|
662 | }
|
---|
663 | else
|
---|
664 | {
|
---|
665 | *fLog << warn << "MTelAxisFromStars::FindSkyCamTrans; number of degrees of freedom is too low : "
|
---|
666 | << fNdof << "; fNumStars = " << fNumStars << endl;
|
---|
667 | return kFALSE;
|
---|
668 | }
|
---|
669 | }
|
---|
670 | else if (fNdof == -1)
|
---|
671 | {
|
---|
672 | if (fixedrotationangle == -1.0)
|
---|
673 | {
|
---|
674 | fixedrotationangle = 0.0;
|
---|
675 | *fLog << warn << "MTelAxisFromStars::FindSkyCamTrans; rotation angle is fixed at "
|
---|
676 | << fixedrotationangle << endl;
|
---|
677 | }
|
---|
678 | else if (fixedscalefactor == -1.0)
|
---|
679 | {
|
---|
680 | fixedscalefactor = 1.0;
|
---|
681 | *fLog << warn << "MTelAxisFromStars::FindSkyCamTrans; scale factor is fixed at "
|
---|
682 | << fixedscalefactor << endl;
|
---|
683 | }
|
---|
684 | else
|
---|
685 | {
|
---|
686 | *fLog << warn << "MTelAxisFromStars::FindSkyCamTrans; number of degrees of freedom is too low : "
|
---|
687 | << fNdof << "; fNumStars = " << fNumStars<< endl;
|
---|
688 | return kFALSE;
|
---|
689 | }
|
---|
690 | }
|
---|
691 | }
|
---|
692 |
|
---|
693 | // recalculate number of degrees of freedom
|
---|
694 | fNdof = 2 * fNumStars - 4;
|
---|
695 | if (fixedscalefactor != -1.0)
|
---|
696 | fNdof += 1;
|
---|
697 | if (fixedrotationangle != -1.0)
|
---|
698 | fNdof += 1;
|
---|
699 |
|
---|
700 | if (fNdof < 0)
|
---|
701 | return kFALSE;
|
---|
702 | //-------------------------------------------
|
---|
703 |
|
---|
704 |
|
---|
705 | // get first approximation of scaling factor
|
---|
706 | if (fixedscalefactor != -1.0)
|
---|
707 | lambda = fixedscalefactor;
|
---|
708 | else
|
---|
709 | lambda = 1.0;
|
---|
710 |
|
---|
711 | Double_t lambdaold = lambda;
|
---|
712 | Double_t dlambda = 0.0;
|
---|
713 |
|
---|
714 | // get first approximation of rotation angle
|
---|
715 | Double_t alfa = 0.0;
|
---|
716 | if (fixedrotationangle != -1.0)
|
---|
717 | alfa = fixedrotationangle / TMath::RadToDeg();
|
---|
718 |
|
---|
719 | Double_t alfaold = alfa;
|
---|
720 | // maximum allowed change of alfa in 1 iteration step (5 degrees)
|
---|
721 | Double_t dalfamax = 5.0 / TMath::RadToDeg();
|
---|
722 | Double_t dalfa = 0.0;
|
---|
723 |
|
---|
724 | Double_t cosal = TMath::Cos(alfa);
|
---|
725 | Double_t sinal = TMath::Sin(alfa);
|
---|
726 |
|
---|
727 | A[0][0] = cosal;
|
---|
728 | A[0][1] = -sinal;
|
---|
729 | A[1][0] = sinal;
|
---|
730 | A[1][1] = cosal;
|
---|
731 |
|
---|
732 |
|
---|
733 | Double_t absdold2 = 10000.0;
|
---|
734 | Double_t fChangeofd2 = 10000.0;
|
---|
735 |
|
---|
736 |
|
---|
737 | TArrayD Aa[2];
|
---|
738 | Aa[0].Set(fNumStars);
|
---|
739 | Aa[1].Set(fNumStars);
|
---|
740 |
|
---|
741 |
|
---|
742 | Double_t sumEbminlamAa[2];
|
---|
743 |
|
---|
744 | TArrayD Ebminlambracd[2];
|
---|
745 | Ebminlambracd[0].Set(fNumStars);
|
---|
746 | Ebminlambracd[1].Set(fNumStars);
|
---|
747 |
|
---|
748 | TArrayD EAa[2];
|
---|
749 | EAa[0].Set(fNumStars);
|
---|
750 | EAa[1].Set(fNumStars);
|
---|
751 |
|
---|
752 | // invert the error matrices
|
---|
753 | TArrayD c[2][2];
|
---|
754 | c[0][0].Set(fNumStars);
|
---|
755 | c[0][1].Set(fNumStars);
|
---|
756 | c[1][0].Set(fNumStars);
|
---|
757 | c[1][1].Set(fNumStars);
|
---|
758 |
|
---|
759 | for (Int_t ix=0; ix<fNumStars; ix++)
|
---|
760 | {
|
---|
761 | Double_t XX = e[0][0][ix];
|
---|
762 | Double_t XY = e[0][1][ix];
|
---|
763 | Double_t YY = e[1][1][ix];
|
---|
764 |
|
---|
765 | // get inverse of error matrix
|
---|
766 | Double_t determ = XX*YY - XY*XY;
|
---|
767 | c[0][0][ix] = YY / determ;
|
---|
768 | c[0][1][ix] = -XY / determ;
|
---|
769 | c[1][0][ix] = -XY / determ;
|
---|
770 | c[1][1][ix] = XX / determ;
|
---|
771 | }
|
---|
772 |
|
---|
773 |
|
---|
774 |
|
---|
775 | // calculate sum of inverted error matrices
|
---|
776 | Double_t determsumc;
|
---|
777 | Double_t sumc[2][2];
|
---|
778 | sumc[0][0] = 0.0;
|
---|
779 | sumc[0][1] = 0.0;
|
---|
780 | sumc[1][0] = 0.0;
|
---|
781 | sumc[1][1] = 0.0;
|
---|
782 |
|
---|
783 | for (Int_t ix=0; ix<fNumStars; ix++)
|
---|
784 | {
|
---|
785 | sumc[0][0] += c[0][0][ix];
|
---|
786 | sumc[0][1] += c[0][1][ix];
|
---|
787 | sumc[1][0] += c[1][0][ix];
|
---|
788 | sumc[1][1] += c[1][1][ix];
|
---|
789 | }
|
---|
790 | determsumc = sumc[0][0]*sumc[1][1] - sumc[0][1]*sumc[1][0];
|
---|
791 |
|
---|
792 | // calculate inverse of sum of inverted error matrices
|
---|
793 | Double_t sumcinv[2][2];
|
---|
794 | sumcinv[0][0] = sumc[1][1] / determsumc;
|
---|
795 | sumcinv[0][1] = -sumc[0][1] / determsumc;
|
---|
796 | sumcinv[1][0] = -sumc[1][0] / determsumc;
|
---|
797 | sumcinv[1][1] = sumc[0][0] / determsumc;
|
---|
798 |
|
---|
799 | //*fLog << "sumcinv = " << sumcinv[0][0] << ", " << sumcinv[0][1]
|
---|
800 | // << ", " << sumcinv[1][1] << endl;
|
---|
801 |
|
---|
802 |
|
---|
803 | // minimize chi2 by iteration ***** start **********************
|
---|
804 |
|
---|
805 | // stop iteration when change in |d|*|d| is less than 'told2'
|
---|
806 | // and change in alfa is less than 'toldalfa'
|
---|
807 | // and change in lambda is less than 'toldlambda'
|
---|
808 | // or chi2 is less than 'tolchi2'
|
---|
809 | Double_t told2 = 0.3*0.3; // [mm*mm]; 1/100 of an inner pixel diameter
|
---|
810 | Double_t toldalfa = 0.01 / TMath::RadToDeg(); // 0.01 degrees
|
---|
811 | Double_t toldlambda = 0.00006; // uncertainty of 1 mm of distance
|
---|
812 | // between camera and reflector
|
---|
813 | Double_t tolchi2 = 1.e-5;
|
---|
814 |
|
---|
815 | Int_t fNumIterMax = 100;
|
---|
816 | fNumIter = 0;
|
---|
817 |
|
---|
818 | for (Int_t i=0; i<fNumIterMax; i++)
|
---|
819 | {
|
---|
820 | fNumIter++;
|
---|
821 |
|
---|
822 | // get next approximation of d ------------------
|
---|
823 | for (Int_t ix=0; ix<fNumStars; ix++)
|
---|
824 | {
|
---|
825 | Aa[0][ix] = A[0][0] * a[0][ix] + A[0][1]*a[1][ix];
|
---|
826 | Aa[1][ix] = A[1][0] * a[0][ix] + A[1][1]*a[1][ix];
|
---|
827 |
|
---|
828 | //*fLog << "ix, Aa = " << ix << " : " << Aa[0][ix] << ", "
|
---|
829 | // << Aa[1][ix] << endl;
|
---|
830 | }
|
---|
831 |
|
---|
832 | sumEbminlamAa[0] = 0.0;
|
---|
833 | sumEbminlamAa[1] = 0.0;
|
---|
834 |
|
---|
835 | for (Int_t ix=0; ix<fNumStars; ix++)
|
---|
836 | {
|
---|
837 | sumEbminlamAa[0] += c[0][0][ix] * (b[0][ix] - lambda*Aa[0][ix])
|
---|
838 | + c[0][1][ix] * (b[1][ix] - lambda*Aa[1][ix]);
|
---|
839 |
|
---|
840 | sumEbminlamAa[1] += c[1][0][ix] * (b[0][ix] - lambda*Aa[0][ix])
|
---|
841 | + c[1][1][ix] * (b[1][ix] - lambda*Aa[1][ix]);
|
---|
842 | }
|
---|
843 |
|
---|
844 | //*fLog << "sumEbminlamAa = " << sumEbminlamAa[0] << ", "
|
---|
845 | // << sumEbminlamAa[1] << endl;
|
---|
846 |
|
---|
847 | d[0] = sumcinv[0][0] * sumEbminlamAa[0]
|
---|
848 | + sumcinv[0][1] * sumEbminlamAa[1] ;
|
---|
849 |
|
---|
850 | d[1] = sumcinv[1][0] * sumEbminlamAa[0]
|
---|
851 | + sumcinv[1][1] * sumEbminlamAa[1] ;
|
---|
852 |
|
---|
853 | Double_t absdnew2 = d[0]*d[0] + d[1]*d[1];
|
---|
854 | fChangeofd2 = absdnew2 - absdold2;
|
---|
855 |
|
---|
856 | //*fLog << "fNumIter : " << fNumIter
|
---|
857 | // << "; alfa, lambda, d[0], d[1], absdold2, absdnew2 = " << endl;
|
---|
858 | //*fLog << alfa << ", " << lambda << ", " << d[0] << ", " << d[1]
|
---|
859 | // << ", " << absdold2 << ", " << absdnew2 << endl;
|
---|
860 |
|
---|
861 |
|
---|
862 | if ( fabs(fChangeofd2) < told2 && fabs(dalfa) < toldalfa &&
|
---|
863 | fabs(dlambda) < toldlambda )
|
---|
864 | {
|
---|
865 | //*fLog << "Iteration stopped because of small changes : fChangeofd2, dalfa, dlambda = "
|
---|
866 | // << fChangeofd2 << ", " << dalfa << ", " << dlambda << endl;
|
---|
867 | break;
|
---|
868 | }
|
---|
869 | absdold2 = absdnew2;
|
---|
870 |
|
---|
871 | // get next approximation of matrix A ----------------
|
---|
872 | if (fFixedRotationAngle == -1.0)
|
---|
873 | {
|
---|
874 | for (Int_t ix=0; ix<fNumStars; ix++)
|
---|
875 | {
|
---|
876 | Ebminlambracd[0][ix] =
|
---|
877 | c[0][0][ix] * ( b[0][ix] - lambda*Aa[0][ix] - d[0] )
|
---|
878 | + c[0][1][ix] * ( b[1][ix] - lambda*Aa[1][ix] - d[1] );
|
---|
879 |
|
---|
880 | Ebminlambracd[1][ix] =
|
---|
881 | c[1][0][ix] * ( b[0][ix] - lambda*Aa[0][ix] - d[0] )
|
---|
882 | + c[1][1][ix] * ( b[1][ix] - lambda*Aa[1][ix] - d[1] );
|
---|
883 |
|
---|
884 | //*fLog << "ix, Ebminlambracd = " << ix << " : "
|
---|
885 | // << Ebminlambracd[0][ix] << ", "
|
---|
886 | // << Ebminlambracd[1][ix] << endl;
|
---|
887 | }
|
---|
888 |
|
---|
889 | // stop iteration if fChi2 is small enough
|
---|
890 | fChi2 = 0.0;
|
---|
891 | for (Int_t ix=0; ix<fNumStars; ix++)
|
---|
892 | {
|
---|
893 | fChi2 += (b[0][ix]-lambda*Aa[0][ix]-d[0] ) * Ebminlambracd[0][ix]
|
---|
894 | + (b[1][ix]-lambda*Aa[1][ix]-d[1] ) * Ebminlambracd[1][ix];
|
---|
895 | }
|
---|
896 | if ( fChi2 < tolchi2 )
|
---|
897 | {
|
---|
898 | //*fLog << "iteration stopped because of small fChi2 : "
|
---|
899 | // << fChi2 << endl;
|
---|
900 | break;
|
---|
901 | }
|
---|
902 |
|
---|
903 |
|
---|
904 | Double_t dchi2dA[2][2];
|
---|
905 | dchi2dA[0][0] = 0.0;
|
---|
906 | dchi2dA[0][1] = 0.0;
|
---|
907 | dchi2dA[1][0] = 0.0;
|
---|
908 | dchi2dA[1][1] = 0.0;
|
---|
909 |
|
---|
910 | for (Int_t ix=0; ix<fNumStars; ix++)
|
---|
911 | {
|
---|
912 | dchi2dA[0][0] += Ebminlambracd[0][ix] * a[0][ix];
|
---|
913 | dchi2dA[0][1] += Ebminlambracd[0][ix] * a[1][ix];
|
---|
914 | dchi2dA[1][0] += Ebminlambracd[1][ix] * a[0][ix];
|
---|
915 | dchi2dA[1][1] += Ebminlambracd[1][ix] * a[1][ix];
|
---|
916 | }
|
---|
917 |
|
---|
918 | //*fLog << "dchi2dA = " << dchi2dA[0][0] << ", " << dchi2dA[0][1]
|
---|
919 | // << ", " << dchi2dA[1][0] << ", " << dchi2dA[1][1] << endl;
|
---|
920 |
|
---|
921 | // ********* 1st derivative (d chi2) / (d alfa) ************
|
---|
922 | Double_t dchi2dalfa = -2.0*lambda *
|
---|
923 | ( - sinal*(dchi2dA[0][0]+dchi2dA[1][1])
|
---|
924 | + cosal*(dchi2dA[1][0]-dchi2dA[0][1]) );
|
---|
925 |
|
---|
926 |
|
---|
927 | //Double_t dalfa1st = - fChi2 / dchi2dalfa;
|
---|
928 |
|
---|
929 | //*fLog << "fChi2, dchi2dalfa = " << fChi2 << ", "
|
---|
930 | // << dchi2dalfa << endl;
|
---|
931 | //*fLog << "proposed change of alfa using 1st derivative = "
|
---|
932 | // << dalfa1st << endl;
|
---|
933 |
|
---|
934 | // ********* 2nd derivative (d2 chi2) / (d alfa2) ******
|
---|
935 | Double_t term1 = 0.0;
|
---|
936 | Double_t term2 = 0.0;
|
---|
937 | Double_t term3 = 0.0;
|
---|
938 | Double_t term4 = 0.0;
|
---|
939 |
|
---|
940 | for (Int_t ix=0; ix<fNumStars; ix++)
|
---|
941 | {
|
---|
942 | term1 += a[0][ix]*c[0][0][ix]*a[0][ix] + a[1][ix]*c[1][0][ix]*a[0][ix]
|
---|
943 | + a[0][ix]*c[0][1][ix]*a[1][ix] + a[1][ix]*c[1][1][ix]*a[1][ix];
|
---|
944 |
|
---|
945 | term2 += a[0][ix]*c[1][0][ix]*a[0][ix] - a[1][ix]*c[0][0][ix]*a[0][ix]
|
---|
946 | + a[0][ix]*c[1][1][ix]*a[1][ix] - a[1][ix]*c[0][1][ix]*a[1][ix];
|
---|
947 |
|
---|
948 | term3 = a[0][ix]*c[0][0][ix]*a[1][ix] + a[1][ix]*c[1][0][ix]*a[1][ix]
|
---|
949 | - a[0][ix]*c[0][1][ix]*a[0][ix] - a[1][ix]*c[1][1][ix]*a[0][ix];
|
---|
950 |
|
---|
951 | term4 += a[0][ix]*c[1][0][ix]*a[1][ix] - a[1][ix]*c[0][0][ix]*a[1][ix]
|
---|
952 | - a[0][ix]*c[1][1][ix]*a[0][ix] + a[1][ix]*c[0][1][ix]*a[0][ix];
|
---|
953 | }
|
---|
954 |
|
---|
955 | Double_t d2chi2dalfa2 =
|
---|
956 | - 2.0*lambda * ( - cosal*(dchi2dA[0][0]+dchi2dA[1][1])
|
---|
957 | - sinal*(dchi2dA[1][0]-dchi2dA[0][1]) )
|
---|
958 | + 2.0*lambda*lambda * ( sinal*sinal * term1 - sinal*cosal * term2
|
---|
959 | + sinal*cosal * term3 - cosal*cosal * term4);
|
---|
960 |
|
---|
961 | // Gauss-Newton step
|
---|
962 | Double_t dalfa2nd = - dchi2dalfa / d2chi2dalfa2;
|
---|
963 |
|
---|
964 | //*fLog << "proposed change of alfa using 2st derivative = "
|
---|
965 | // << dalfa2nd << endl;
|
---|
966 |
|
---|
967 | //dalfa = dalfa1st;
|
---|
968 | dalfa = dalfa2nd;
|
---|
969 |
|
---|
970 | // ******************************************
|
---|
971 |
|
---|
972 |
|
---|
973 | // restrict change of alfa
|
---|
974 | if ( fabs(dalfa) > dalfamax )
|
---|
975 | {
|
---|
976 | dalfa = TMath::Sign( dalfamax, dalfa );
|
---|
977 | }
|
---|
978 | alfa = alfaold + dalfa;
|
---|
979 |
|
---|
980 | if ( alfa < -5.0/TMath::RadToDeg() )
|
---|
981 | alfa = -5.0/TMath::RadToDeg();
|
---|
982 | else if ( alfa > 5.0/TMath::RadToDeg() )
|
---|
983 | alfa = 5.0/TMath::RadToDeg();
|
---|
984 |
|
---|
985 | dalfa = alfa - alfaold;
|
---|
986 |
|
---|
987 | alfaold = alfa;
|
---|
988 |
|
---|
989 | sinal = TMath::Sin(alfa);
|
---|
990 | cosal = TMath::Cos(alfa);
|
---|
991 |
|
---|
992 | A[0][0] = cosal;
|
---|
993 | A[0][1] = -sinal;
|
---|
994 | A[1][0] = sinal;
|
---|
995 | A[1][1] = cosal;
|
---|
996 |
|
---|
997 | //*fLog << "alfa-alfaold = " << dalfa << endl;
|
---|
998 | //*fLog << "new alfa = " << alfa << endl;
|
---|
999 | }
|
---|
1000 |
|
---|
1001 |
|
---|
1002 | // get next approximation of lambda ----------------
|
---|
1003 | if (fFixedScaleFactor == -1.0)
|
---|
1004 | {
|
---|
1005 | for (Int_t ix=0; ix<fNumStars; ix++)
|
---|
1006 | {
|
---|
1007 | Aa[0][ix] = A[0][0]*a[0][ix] + A[0][1]*a[1][ix];
|
---|
1008 | Aa[1][ix] = A[1][0]*a[0][ix] + A[1][1]*a[1][ix];
|
---|
1009 |
|
---|
1010 | EAa[0][ix] =
|
---|
1011 | c[0][0][ix] * Aa[0][ix] + c[0][1][ix] * Aa[1][ix];
|
---|
1012 | EAa[1][ix] =
|
---|
1013 | c[1][0][ix] * Aa[0][ix] + c[1][1][ix] * Aa[1][ix];
|
---|
1014 |
|
---|
1015 | //*fLog << "ix, Aa = " << ix << " : " << Aa[0][ix] << ", "
|
---|
1016 | // << Aa[1][ix] << endl;
|
---|
1017 |
|
---|
1018 | //*fLog << "ix, EAa = " << ix << " : " << EAa[0][ix] << ", "
|
---|
1019 | // << EAa[1][ix] << endl;
|
---|
1020 | }
|
---|
1021 |
|
---|
1022 | Double_t num = 0.0;
|
---|
1023 | Double_t denom = 0.0;
|
---|
1024 | for (Int_t ix=0; ix<fNumStars; ix++)
|
---|
1025 | {
|
---|
1026 | num += (b[0][ix]-d[0]) * EAa[0][ix]
|
---|
1027 | + (b[1][ix]-d[1]) * EAa[1][ix];
|
---|
1028 |
|
---|
1029 | denom += Aa[0][ix] * EAa[0][ix]
|
---|
1030 | + Aa[1][ix] * EAa[1][ix];
|
---|
1031 |
|
---|
1032 | //*fLog << "ix : b-d = " << ix << " : " << b[0][ix]-d[0]
|
---|
1033 | // << ", " << b[1][ix]-d[1] << endl;
|
---|
1034 |
|
---|
1035 | //*fLog << "ix : Aa = " << ix << " : " << Aa[0][ix]
|
---|
1036 | // << ", " << Aa[1][ix] << endl;
|
---|
1037 | }
|
---|
1038 |
|
---|
1039 | lambda = num / denom;
|
---|
1040 |
|
---|
1041 | if ( lambda < 0.9 )
|
---|
1042 | lambda = 0.9;
|
---|
1043 | else if ( lambda > 1.1 )
|
---|
1044 | lambda = 1.1;
|
---|
1045 |
|
---|
1046 | dlambda = lambda - lambdaold;
|
---|
1047 | lambdaold = lambda;
|
---|
1048 |
|
---|
1049 | //*fLog << "num, denom, lambda, dlambda = " << num
|
---|
1050 | // << ", " << denom << ", " << lambda << ", "
|
---|
1051 | // << dlambda << endl;
|
---|
1052 | }
|
---|
1053 |
|
---|
1054 | }
|
---|
1055 | //------- end of iteration *****************************************
|
---|
1056 |
|
---|
1057 | alfadeg = alfa * TMath::RadToDeg();
|
---|
1058 |
|
---|
1059 | // calculate error matrix of d[2]
|
---|
1060 | errd[0][0] = sumcinv[0][0];
|
---|
1061 | errd[0][1] = sumcinv[0][1];
|
---|
1062 | errd[1][0] = sumcinv[1][0];
|
---|
1063 | errd[1][1] = sumcinv[1][1];
|
---|
1064 |
|
---|
1065 | // evaluate quality of fit
|
---|
1066 |
|
---|
1067 | // calculate chi2
|
---|
1068 | for (Int_t ix=0; ix<fNumStars; ix++)
|
---|
1069 | {
|
---|
1070 | Ebminlambracd[0][ix] =
|
---|
1071 | c[0][0][ix] * ( b[0][ix] - lambda*Aa[0][ix] - d[0] )
|
---|
1072 | + c[0][1][ix] * ( b[1][ix] - lambda*Aa[1][ix] - d[1] );
|
---|
1073 |
|
---|
1074 | Ebminlambracd[1][ix] =
|
---|
1075 | c[1][0][ix] * (b[0][ix] - lambda*Aa[0][ix] - d[0] )
|
---|
1076 | + c[1][1][ix] * (b[1][ix] - lambda*Aa[1][ix] - d[1] );
|
---|
1077 | }
|
---|
1078 |
|
---|
1079 | fChi2 = 0.0;
|
---|
1080 | for (Int_t ix=0; ix<fNumStars; ix++)
|
---|
1081 | {
|
---|
1082 | fChi2 += (b[0][ix] - lambda*Aa[0][ix] - d[0] ) * Ebminlambracd[0][ix]
|
---|
1083 | + (b[1][ix] - lambda*Aa[1][ix] - d[1] ) * Ebminlambracd[1][ix];
|
---|
1084 | }
|
---|
1085 |
|
---|
1086 | fChi2Prob = TMath::Prob(fChi2, fNdof);
|
---|
1087 |
|
---|
1088 | //*fLog << "MTelAxisFromStars::FindSkyCamTrans :" << endl;
|
---|
1089 | //*fLog << " fNumStars, fChi2, fNdof, fChi2Prob, fNumIter, fChangeofd2, dalfa, dlambda = "
|
---|
1090 | // << fNumStars << ", " << fChi2 << ", " << fNdof << ", "
|
---|
1091 | // << fChi2Prob << ", "
|
---|
1092 | // << fNumIter << ", " << fChangeofd2 << ", " << dalfa << ", "
|
---|
1093 | // << dlambda << endl;
|
---|
1094 | //*fLog << " lambda, alfadeg, d[0], d[1] = " << lambda << ", "
|
---|
1095 | // << alfadeg << ", " << d[0] << ", " << d[1] << endl;
|
---|
1096 |
|
---|
1097 | return kTRUE;
|
---|
1098 | }
|
---|
1099 |
|
---|
1100 | // --------------------------------------------------------------------------
|
---|
1101 | //
|
---|
1102 | // Apply transformation (lambda, A, d)
|
---|
1103 | // to the expected positions (a[1], a[2])
|
---|
1104 | // to obtain the estimated positions (b[1], b[2])
|
---|
1105 | //
|
---|
1106 | // e[2][2] is the error matrix of b[2]
|
---|
1107 |
|
---|
1108 | void MTelAxisFromStars::TransSkyCam(
|
---|
1109 | Double_t &lambda, Double_t A[2][2], Double_t d[2], Double_t errd[2][2],
|
---|
1110 | TArrayD a[2], TArrayD b[2], TArrayD e[2][2])
|
---|
1111 | {
|
---|
1112 | Int_t numpos = a[0].GetSize();
|
---|
1113 | if (numpos <= 0)
|
---|
1114 | return;
|
---|
1115 |
|
---|
1116 | //*fLog << "MTelAxisFromStars::TransSkyCam; expected and estimated positions :"
|
---|
1117 | // << endl;
|
---|
1118 |
|
---|
1119 | for (Int_t ix=0; ix<numpos; ix++)
|
---|
1120 | {
|
---|
1121 | //*fLog << "MTelAxisFromStars; ix = " << ix << endl;
|
---|
1122 |
|
---|
1123 | b[0][ix] = lambda * (A[0][0]*a[0][ix] + A[0][1]*a[1][ix]) + d[0];
|
---|
1124 | b[1][ix] = lambda * (A[1][0]*a[0][ix] + A[1][1]*a[1][ix]) + d[1];
|
---|
1125 |
|
---|
1126 | e[0][0][ix] = errd[0][0];
|
---|
1127 | e[0][1][ix] = errd[0][1];
|
---|
1128 | e[1][0][ix] = errd[1][0];
|
---|
1129 | e[1][1][ix] = errd[1][1];
|
---|
1130 |
|
---|
1131 | //*fLog << " ix, a[0], a[1], b[0], b[1], errxx, errxy, erryy = "
|
---|
1132 | // << ix << " : "
|
---|
1133 | // << a[0][ix] << ", " << a[1][ix] << "; "
|
---|
1134 | // << b[0][ix] << ", " << b[1][ix] << "; "
|
---|
1135 | // << e[0][0][ix] << ", " << e[0][1][ix] << ", "
|
---|
1136 | // << e[1][1][ix] << endl;
|
---|
1137 | }
|
---|
1138 | }
|
---|
1139 |
|
---|
1140 | // --------------------------------------------------------------------------
|
---|
1141 | //
|
---|
1142 | // SetPointingPosition
|
---|
1143 | //
|
---|
1144 | // put the corrected pointing direction into MPointingPos[MPointingPos];
|
---|
1145 | // this direction corresponds to the position (0,0) in the camera
|
---|
1146 | //
|
---|
1147 |
|
---|
1148 | Bool_t MTelAxisFromStars::SetPointingPosition(MStarCamTrans *fstarcamtrans,
|
---|
1149 | MReportDrive *fdrive, MSkyCamTrans *ftrans, MPointingPos *fpointpos)
|
---|
1150 | {
|
---|
1151 | Double_t decdrive = fdrive->GetDec();
|
---|
1152 | Double_t hdrive = fdrive->GetHa();
|
---|
1153 | Double_t radrive = fdrive->GetRa();
|
---|
1154 |
|
---|
1155 | // this is the estimated position (with optical aberration) in the camera
|
---|
1156 | // corresponding to the direction in MReportDrive
|
---|
1157 | Double_t Xpoint = (ftrans->GetShiftD())[0];
|
---|
1158 | Double_t Ypoint = (ftrans->GetShiftD())[1];
|
---|
1159 |
|
---|
1160 | // get the sky direction corresponding to the position (0,0) in the camera
|
---|
1161 | Double_t decpoint = 0.0;
|
---|
1162 | Double_t hpoint = 0.0;
|
---|
1163 | fstarcamtrans->CelCamToCel0(decdrive, hdrive,
|
---|
1164 | Xpoint/fAberr, Ypoint/fAberr, decpoint, hpoint);
|
---|
1165 | Double_t rapoint = radrive - hpoint + hdrive;
|
---|
1166 | fpointpos->SetSkyPosition(rapoint, decpoint, hpoint);
|
---|
1167 |
|
---|
1168 | // get the local direction corresponding to the position (0,0) in the camera
|
---|
1169 | Double_t thetadrive = fdrive->GetNominalZd();
|
---|
1170 | Double_t phidrive = fdrive->GetNominalAz();
|
---|
1171 | Double_t thetapoint = 0.0;
|
---|
1172 | Double_t phipoint = 0.0;
|
---|
1173 | fstarcamtrans->LocCamToLoc0(thetadrive, phidrive,
|
---|
1174 | Xpoint/fAberr, Ypoint/fAberr, thetapoint, phipoint);
|
---|
1175 | fpointpos->SetLocalPosition(thetapoint, phipoint);
|
---|
1176 | fpointpos->SetReadyToSave();
|
---|
1177 |
|
---|
1178 | //*fLog << "SetPointingPosition : decdrive, hdrive, radrive Xpoint, Ypoint = "
|
---|
1179 | // << decdrive << ", " << hdrive << ", " << radrive << ", "
|
---|
1180 | // << Xpoint << ", " << Ypoint << endl;
|
---|
1181 |
|
---|
1182 | //*fLog << "SetPointingPosition : thetadrive, phidrive, thetapoint, phipoint = "
|
---|
1183 | // << thetadrive << ", " << phidrive << ", " << thetapoint << ", "
|
---|
1184 | // << phipoint << endl;
|
---|
1185 |
|
---|
1186 | return kTRUE;
|
---|
1187 | }
|
---|
1188 |
|
---|
1189 | // --------------------------------------------------------------------------
|
---|
1190 | //
|
---|
1191 | // SetSourcePosition
|
---|
1192 | //
|
---|
1193 | // put the estimated position of the source in the camera into
|
---|
1194 | // MSrcPosCam[MSrcPosCam]
|
---|
1195 | //
|
---|
1196 | // and the estimated local direction of the source into
|
---|
1197 | // MSourcePos[MPointingPos]
|
---|
1198 | //
|
---|
1199 |
|
---|
1200 | Bool_t MTelAxisFromStars::SetSourcePosition(MStarCamTrans *fstarcamtrans,
|
---|
1201 | MPointingPos *fpointpos, MPointingPos *fsourcepos, MSrcPosCam *fsrcpos)
|
---|
1202 | {
|
---|
1203 | // get the corrected pointing direction
|
---|
1204 | // corresponding to the position (0,0) in the camera
|
---|
1205 | Double_t decpoint = fpointpos->GetDec();
|
---|
1206 | Double_t hpoint = fpointpos->GetHa();
|
---|
1207 |
|
---|
1208 | // get the sky direction of the source
|
---|
1209 | Double_t decsource = fsourcepos->GetDec();
|
---|
1210 | Double_t hsource = fsourcepos->GetHa();
|
---|
1211 |
|
---|
1212 | // get the estimated position (Xsource, Ysource) of the source in the camera;
|
---|
1213 | // this is a position for an ideal imaging, without optical aberration
|
---|
1214 | Double_t Xsource = 0.0;
|
---|
1215 | Double_t Ysource = 0.0;
|
---|
1216 | fstarcamtrans->Cel0CelToCam(decpoint, hpoint,
|
---|
1217 | decsource, hsource, Xsource, Ysource);
|
---|
1218 | fsrcpos->SetXY(Xsource*fAberr, Ysource*fAberr);
|
---|
1219 | fsrcpos->SetReadyToSave();
|
---|
1220 |
|
---|
1221 | // get the estimated local direction of the source
|
---|
1222 | Double_t thetapoint = fpointpos->GetZd();
|
---|
1223 | Double_t phipoint = fpointpos->GetAz();
|
---|
1224 | Double_t thetasource = 0.0;
|
---|
1225 | Double_t phisource = 0.0;
|
---|
1226 | fstarcamtrans->Loc0CamToLoc(thetapoint, phipoint,
|
---|
1227 | Xsource, Ysource, thetasource, phisource);
|
---|
1228 | fsourcepos->SetLocalPosition(thetasource, phisource);
|
---|
1229 | fsourcepos->SetReadyToSave();
|
---|
1230 |
|
---|
1231 | //*fLog << "SetSourcePosition : decpoint, hpoint, decsource, hsource, Xsource, Ysource = "
|
---|
1232 | // << decpoint << ", " << hpoint << ", " << decsource << ", "
|
---|
1233 | // << hsource << ", " << Xsource << ", " << Ysource << endl;
|
---|
1234 | //*fLog << "SetSourcePosition : thetapoint, phipoint, thetasource, phisource = "
|
---|
1235 | // << thetapoint << ", " << phipoint << ", " << thetasource << ", "
|
---|
1236 | // << phisource << endl;
|
---|
1237 |
|
---|
1238 | return kTRUE;
|
---|
1239 | }
|
---|
1240 |
|
---|
1241 | // --------------------------------------------------------------------------
|
---|
1242 |
|
---|
1243 |
|
---|
1244 |
|
---|
1245 |
|
---|
1246 |
|
---|
1247 |
|
---|
1248 |
|
---|
1249 |
|
---|
1250 |
|
---|
1251 |
|
---|
1252 |
|
---|
1253 |
|
---|