source: trunk/Mars/mcore/DrsCalib.h@ 18293

Last change on this file since 18293 was 18156, checked in by tbretz, 10 years ago
Added Reset to DrsCalibrateTime and removed the reading and writing of fits files. It is now in MDrsCalibrationTime. If needed, it must be made conistent, e.g. diffferent table name.
File size: 49.1 KB
Line 
1#ifndef MARS_DrsCalib
2#define MARS_DrsCalib
3
4#include <math.h> // fabs
5#include <errno.h> // errno
6
7#ifndef MARS_fits
8#include "fits.h"
9#endif
10
11#ifndef MARS_ofits
12#include "ofits.h"
13#endif
14
15class DrsCalibrate
16{
17protected:
18 int64_t fNumEntries;
19
20 size_t fNumSamples;
21 size_t fNumChannels;
22
23 std::vector<int64_t> fSum;
24 std::vector<int64_t> fSum2;
25
26public:
27 DrsCalibrate() : fNumEntries(0), fNumSamples(0), fNumChannels(0)
28 {
29 fSum.reserve(1024*1440);
30 fSum2.reserve(1024*1440);
31 }
32
33 void Reset()
34 {
35 fNumEntries = 0;
36 fNumSamples = 0;
37 fNumChannels = 0;
38
39 fSum.clear();
40 fSum2.clear();
41 }
42
43 void InitSize(uint16_t channels, uint16_t samples)
44 {
45 fNumChannels = channels;
46 fNumSamples = samples;
47
48 fSum.assign(samples*channels, 0);
49 fSum2.assign(samples*channels, 0);
50 }
51
52 void AddRel(const int16_t *val, const int16_t *start)
53 {
54 /*
55 for (size_t ch=0; ch<fNumChannels; ch++)
56 {
57 const int16_t &spos = start[ch];
58 if (spos<0)
59 continue;
60
61 const size_t pos = ch*1024;
62 for (size_t i=0; i<1024; i++)
63 {
64 // Value is relative to trigger
65 // Abs is corresponding index relative to DRS pipeline
66 const size_t rel = pos + i;
67 const size_t abs = pos + (spos+i)%1024;
68
69 const int64_t v = val[rel];
70
71 fSum[abs] += v;
72 fSum2[abs] += v*v;
73 }
74 }*/
75
76 // This version is 2.5 times faster because the compilers optimization
77 // is not biased by the evaluation of %1024
78 for (size_t ch=0; ch<fNumChannels; ch++)
79 {
80 const int16_t &spos = start[ch];
81 if (spos<0)
82 continue;
83
84 const size_t pos = ch*1024;
85
86 const int16_t *beg_val = val + pos;
87 int64_t *beg_sum = fSum.data() + pos;
88 int64_t *beg_sum2 = fSum2.data() + pos;
89
90 const int16_t *pval = beg_val; // val[rel]
91 int64_t *psum = beg_sum + spos; // fSum[abs]
92 int64_t *psum2 = beg_sum2 + spos; // fSum2[abs]
93
94 while (psum<beg_sum+1024)
95 {
96 const int64_t v = *pval++;
97
98 *psum++ += v;
99 *psum2++ += v*v;
100 }
101
102 psum = beg_sum;
103 psum2 = beg_sum2;
104
105 while (pval<beg_val+1024)
106 {
107 const int64_t v = *pval++;
108
109 *psum++ += v;
110 *psum2++ += v*v;
111 }
112 }
113
114 fNumEntries++;
115 }
116
117 void AddRel(const int16_t *val, const int16_t *start,
118 const int32_t *offset, const int64_t scale)
119 {
120 /*
121 for (size_t ch=0; ch<fNumChannels; ch++)
122 {
123 const int16_t spos = start[ch];
124 if (spos<0)
125 continue;
126
127 const size_t pos = ch*1024;
128
129 for (size_t i=0; i<fNumSamples; i++)
130 {
131 // Value is relative to trigger
132 // Offset is relative to DRS pipeline
133 // Abs is corresponding index relative to DRS pipeline
134 const size_t rel = pos + i;
135 const size_t abs = pos + (spos+i)%1024;
136
137 const int64_t v = int64_t(val[rel])*scale-offset[abs];
138
139 fSum[abs] += v;
140 fSum2[abs] += v*v;
141 }
142 }*/
143
144 // This version is 2.5 times faster because the compilers optimization
145 // is not biased by the evaluation of %1024
146 for (size_t ch=0; ch<fNumChannels; ch++)
147 {
148 const int16_t &spos = start[ch];
149 if (spos<0)
150 continue;
151
152 const size_t pos = ch*1024;
153
154 const int16_t *beg_val = val + pos;
155 const int32_t *beg_offset = offset + pos;
156 int64_t *beg_sum = fSum.data() + pos;
157 int64_t *beg_sum2 = fSum2.data() + pos;
158
159
160 const int16_t *pval = beg_val; // val[rel]
161 const int32_t *poffset = beg_offset + spos; // offset[abs]
162 int64_t *psum = beg_sum + spos; // fSum[abs]
163 int64_t *psum2 = beg_sum2 + spos; // fSum2[abs]
164
165 while (psum<beg_sum+1024)
166 {
167 const int64_t v = int64_t(*pval++)*scale - *poffset++;
168
169 *psum++ += v;
170 *psum2++ += v*v;
171 }
172
173 psum = beg_sum;
174 psum2 = beg_sum2;
175 poffset = beg_offset;
176
177 while (pval<beg_val+1024)
178 {
179 const int64_t v = int64_t(*pval++)*scale - *poffset++;
180
181 *psum++ += v;
182 *psum2++ += v*v;
183 }
184 }
185
186 fNumEntries++;
187 }
188 /*
189 void AddAbs(const int16_t *val, const int16_t *start,
190 const int32_t *offset, const uint32_t scale)
191 {
192 for (size_t ch=0; ch<fNumChannels; ch++)
193 {
194 const int16_t spos = start[ch];
195 if (spos<0)
196 continue;
197
198 const size_t pos = ch*fNumSamples;
199
200 for (size_t i=0; i<fNumSamples; i++)
201 {
202 // Value is relative to trigger
203 // Offset is relative to DRS pipeline
204 // Abs is corresponding index relative to DRS pipeline
205 const size_t rel = pos + i;
206 const size_t abs = pos + (spos+i)%1024;
207
208 const int64_t v = int64_t(val[rel])*scale-offset[abs];
209
210 fSum[rel] += v;
211 fSum2[rel] += v*v;
212 }
213 }
214
215 fNumEntries++;
216 }*/
217
218 void AddAbs(const int16_t *val, const int16_t *start,
219 const int32_t *offset, const int64_t scale)
220 {
221 /*
222 // 1440 without tm, 1600 with tm
223 for (size_t ch=0; ch<fNumChannels; ch++)
224 {
225 const int16_t spos = start[ch];
226 if (spos<0)
227 continue;
228
229 const size_t pos = ch*fNumSamples;
230 const size_t drs = ch>1439 ? ((ch-1440)*9+8)*1024 : ch*1024;
231
232 for (size_t i=0; i<fNumSamples; i++)
233 {
234 // Value is relative to trigger
235 // Offset is relative to DRS pipeline
236 // Abs is corresponding index relative to DRS pipeline
237 const size_t rel = pos + i;
238 const size_t abs = drs + (spos+i)%1024;
239
240 const int64_t v = int64_t(val[rel])*scale-offset[abs];
241
242 fSum[rel] += v;
243 fSum2[rel] += v*v;
244 }
245 }*/
246
247 // This version is 1.5 times faster because the compilers optimization
248 // is not biased by the evaluation of %1024
249 for (size_t ch=0; ch<fNumChannels; ch++)
250 {
251 const int16_t &spos = start[ch];
252 if (spos<0)
253 continue;
254
255 const size_t pos = ch*fNumSamples;
256
257 const int32_t *beg_offset = offset + ch*1024;
258 const int16_t *beg_val = val + pos;
259 int64_t *beg_sum = fSum.data() + pos;
260 int64_t *beg_sum2 = fSum2.data() + pos;
261
262
263 const int16_t *pval = beg_val; // val[rel]
264 const int32_t *poffset = beg_offset + spos; // offset[abs]
265 int64_t *psum = beg_sum; // fSum[rel]
266 int64_t *psum2 = beg_sum2; // fSum2[rel]
267
268 if (spos+fNumSamples>1024)
269 {
270 while (poffset<beg_offset+1024)
271 {
272 const int64_t v = int64_t(*pval++)*scale - *poffset++;
273
274 *psum++ += v;
275 *psum2++ += v*v;
276 }
277
278 poffset = beg_offset;
279 }
280
281 while (psum<beg_sum+fNumSamples)
282 {
283 const int64_t v = int64_t(*pval++)*scale - *poffset++;
284
285 *psum++ += v;
286 *psum2++ += v*v;
287 }
288 }
289
290 fNumEntries++;
291 }
292
293
294 static void ApplyCh(float *vec, const int16_t *val, int16_t start, uint32_t roi,
295 const int32_t *offset, const int64_t scaleabs,
296 const int64_t *gain, const int64_t scalegain)
297 {
298 if (start<0)
299 {
300 memset(vec, 0, roi);
301 return;
302 }
303 /*
304 for (size_t i=0; i<roi; i++)
305 {
306 // Value is relative to trigger
307 // Offset is relative to DRS pipeline
308 // Abs is corresponding index relative to DRS pipeline
309 const size_t abs = (start+i)%1024;
310
311 const int64_t v =
312 + int64_t(val[i])*scaleabs-offset[abs]
313 ;
314
315 const int64_t div = gain[abs];
316 vec[i] = div==0 ? 0 : double(v)*scalegain/div;
317 }
318 */
319
320 // This version is faster because the compilers optimization
321 // is not biased by the evaluation of %1024
322 // (Here we are dominated by numerics... improvement ~10%)
323 const int32_t *poffset = offset + start; // offset[abs]
324 const int64_t *pgain = gain + start; // gain[abs]
325 const int16_t *pval = val; // val[rel]
326 float *pvec = vec; // vec[rel]
327
328 if (start+roi>1024)
329 {
330 while (poffset<offset+1024)
331 {
332 const int64_t v =
333 + int64_t(*pval++)*scaleabs - *poffset++
334 ;
335
336 *pvec++ = *pgain==0 ? 0 : double(v)*scalegain / *pgain;
337
338 pgain++;
339 }
340
341 poffset = offset;
342 pgain = gain;
343 }
344
345 while (pvec<vec+roi)
346 {
347 const int64_t v =
348 + int64_t(*pval++)*scaleabs - *poffset++
349 ;
350
351 *pvec++ = *pgain==0 ? 0 : double(v)*scalegain / *pgain;
352
353 pgain++;
354 }
355 }
356
357 static void ApplyCh(float *vec, const int16_t *val, int16_t start, uint32_t roi,
358 const int32_t *offset, const int64_t scaleabs,
359 const int64_t *gain, const int64_t scalegain,
360 const int64_t *trgoff, const int64_t scalerel)
361 {
362 if (start<0)
363 {
364 memset(vec, 0, roi);
365 return;
366 }
367 /*
368 for (size_t i=0; i<roi; i++)
369 {
370 // Value is relative to trigger
371 // Offset is relative to DRS pipeline
372 // Abs is corresponding index relative to DRS pipeline
373 const size_t abs = (start+i)%1024;
374
375 const int64_t v =
376 + (int64_t(val[i])*scaleabs-offset[abs])*scalerel
377 - trgoff[i]
378 ;
379
380 const int64_t div = gain[abs]*scalerel;
381 vec[i] = div==0 ? 0 : double(v)*scalegain/div;
382 }
383 */
384 // (Here we are dominated by numerics... improvement ~10%)
385 const int32_t *poffset = offset + start; // offset[abs]
386 const int64_t *pgain = gain + start; // gain[abs]
387 const int16_t *pval = val; // val[rel]
388 const int64_t *ptrgoff = trgoff; // trgoff[rel]
389 float *pvec = vec; // vec[rel]
390
391 if (start+roi>1024)
392 {
393 while (poffset<offset+1024)
394 {
395 const int64_t v =
396 + (int64_t(*pval++)*scaleabs - *poffset++)*scalerel
397 - *ptrgoff++;
398 ;
399
400 const int64_t div = *pgain * scalerel;
401 *pvec++ = div==0 ? 0 : double(v)*scalegain / div;
402
403 pgain++;
404 }
405
406 poffset = offset;
407 pgain = gain;
408 }
409
410 while (pvec<vec+roi)
411 {
412 const int64_t v =
413 + (int64_t(*pval++)*scaleabs - *poffset++)*scalerel
414 - *ptrgoff++;
415 ;
416
417 const int64_t div = *pgain * scalerel;
418 *pvec++ = div==0 ? 0 : double(v)*scalegain / div;
419
420 pgain++;
421 }
422 }
423
424 static double FindStep(const size_t ch0, const float *vec, int16_t roi, const int16_t pos, const uint16_t *map=NULL)
425 {
426 // We have about 1% of all cases which are not ahndled here,
427 // because the baseline jumps up just before the readout window
428 // and down just after it. In this cases we could determine the jump
429 // from the board time difference or throw the event away.
430 if (pos==0 || pos>=roi)
431 return 0;
432
433 double step = 0; // before
434 double rms = 0; // before
435 int cnt = 0;
436
437 // Exclude TM channel
438 for (int p=0; p<8; p++)
439 {
440 const size_t hw = ch0+p;
441 const size_t sw = (map?map[hw]:hw)*roi + pos;
442
443 const double diff = vec[sw]-vec[sw-1];
444
445 step += diff;
446 rms += (vec[sw]-vec[sw-1])*(vec[sw]-vec[sw-1]);
447
448 cnt++;
449 }
450
451 return cnt==0 ? 0 : step/cnt;
452 }
453
454 static void SubtractStep(const size_t ch0, const double avg, float *vec, int16_t roi, int32_t pos, const uint16_t *map=NULL)
455 {
456 if (pos==0 || pos>=roi)
457 return;
458
459 const int begin = avg>0 ? pos : 0;
460 const int end = avg>0 ? roi : pos;
461
462 const double sub = fabs(avg);
463
464 for (int p=0; p<9; p++)
465 {
466 for (int j=begin; j<end; j++)
467 {
468 const size_t hw = ch0+p;
469 const size_t sw = (map?map[hw]:hw)*roi + j;
470
471 vec[sw] -= sub;
472 }
473 }
474 }
475
476 struct Step
477 {
478 Step() : avg(0), rms(0), pos(0), cnt(0) { }
479 double avg;
480 double rms;
481 double pos;
482 uint16_t cnt;
483
484 static bool sort(const Step &s, const Step &r) { return s.avg<r.avg; }
485 };
486
487 static Step AverageSteps(const std::vector<Step>::iterator beg, const std::vector<Step>::iterator end)
488 {
489 Step rc;
490 for (auto it=beg; it!=end; it++)
491 {
492 rc.pos += it->pos;
493 rc.avg += it->avg;
494 rc.rms += it->avg*it->avg;
495 }
496
497 rc.cnt = end-beg;
498
499 rc.pos /= rc.cnt;
500 rc.avg /= rc.cnt;
501 rc.rms /= rc.cnt;
502 rc.rms -= rc.avg*rc.avg;
503 rc.rms = rc.rms<0 ? 0 : sqrt(rc.rms);
504
505 return rc;
506 }
507
508
509 static Step CorrectStep(float *vec, uint16_t nch, uint16_t roi,
510 const int16_t *prev, const int16_t *start,
511 const int16_t offset, const uint16_t *map=NULL)
512 {
513
514 std::vector<Step> list;
515 list.reserve(nch);
516
517 // Fill steps into array
518 // Exclude broken pixels?
519 // Remove maximum and minimum patches (4max and 4min)?
520 for (size_t ch=0; ch<nch; ch += 9)
521 {
522 if (prev[ch]<0 || start[ch]<0)
523 continue;
524
525 const int16_t dist = (prev[ch]-start[ch]+1024+offset)%1024;
526 const double step = FindStep(ch, vec, roi, dist, map);
527 if (step==0)
528 continue;
529
530 Step rc;
531 rc.pos = dist;
532 rc.avg = step;
533 list.push_back(rc);
534 }
535
536 if (list.empty())
537 return Step();
538
539 Step rc = AverageSteps(list.begin(), list.begin()+list.size());;
540
541 if (rc.avg==0)
542 return Step();
543
544 // std::cout << " A0=" << rc.avg << " rms=" << rc.rms << std::endl;
545 if (rc.rms>5)
546 {
547 sort(list.begin(), list.end(), Step::sort);
548
549 //for (auto it=list.begin(); it!=list.end(); it++)
550 // std::cout << " " << it->avg << std::endl;
551
552 const size_t skip = list.size()/10;
553 rc = AverageSteps(list.begin()+skip, list.begin()+list.size()-skip);
554
555 // std::cout << " A1=" << rc.avg << " rms=" << rc.rms << std::endl;
556 }
557
558 for (size_t ch=0; ch<nch; ch += 9)
559 {
560 const int16_t dist = (prev[ch]-start[ch]+1024+offset)%1024;
561 SubtractStep(ch, rc.avg, vec, roi, dist, map);
562 }
563
564 return rc;
565 }
566
567 static void RemoveSpikes(float *p, uint32_t roi)
568 {
569 if (roi<4)
570 return;
571
572 for (size_t i=1; i<roi-2; i++)
573 {
574 if (p[i]-p[i-1]>25 && p[i]-p[i+1]>25)
575 {
576 p[i] = (p[i-1]+p[i+1])/2;
577 }
578
579 if (p[i]-p[i-1]>22 && fabs(p[i]-p[i+1])<4 && p[i+1]-p[i+2]>22)
580 {
581 p[i] = (p[i-1]+p[i+2])/2;
582 p[i+1] = p[i];
583 }
584 }
585 }
586
587 static void RemoveSpikes2(float *p, uint32_t roi)//from Werner
588 {
589 if (roi<4)
590 return;
591
592 std::vector<float> Ameas(p, p+roi);
593
594 std::vector<float> diff(roi);
595 for (size_t i=1; i<roi-1; i++)
596 diff[i] = (p[i-1] + p[i+1])/2 - p[i];
597
598 //std::vector<float> N1mean(roi);
599 //for (size_t i=1; i<roi-1; i++)
600 // N1mean[i] = (p[i-1] + p[i+1])/2;
601
602 const float fract = 0.8;
603
604 for (size_t i=0; i<roi-3; i++)
605 {
606 if (diff[i]<5)
607 continue;
608
609 if (Ameas[i+2] - (Ameas[i] + Ameas[i+3])/2 > 10)
610 {
611 p[i+1]= (Ameas[i+3] - Ameas[i])/3 + Ameas[i];
612 p[i+2]= 2*(Ameas[i+3] - Ameas[i])/3 + Ameas[i];
613
614 i += 3;
615
616 continue;
617 }
618
619 if ( (diff[i+1]<-diff[i]*fract*2) && (diff[i+2]>10) )
620 {
621 p[i+1] = (Ameas[i]+Ameas[i+2])/2;
622 diff[i+2] = (p[i+1] + Ameas[i+3])/2 - Ameas[i+2];
623
624 i += 2;
625 }
626
627 // const float x = Ameas[i] - N1mean[i];
628 // if (x > -5.)
629 // continue;
630
631 // if (Ameas[i+2] - (Ameas[i] + Ameas[i+3])/2. > 10.)
632 // {
633 // p[i+1]= (Ameas[i+3] - Ameas[i])/3 + Ameas[i];
634 // p[i+2]= 2*(Ameas[i+3] - Ameas[i])/3 + Ameas[i];
635 // i += 3;
636 // continue;
637 // }
638
639 // const float xp = Ameas[i+1] - N1mean[i+1];
640 // const float xpp = Ameas[i+2] - N1mean[i+2];
641
642 // if ( (xp > -2.*x*fract) && (xpp < -10.) )
643 // {
644 // p[i+1] = N1mean[i+1];
645 // N1mean[i+2] = Ameas[i+1] - Ameas[i+3]/2;
646 //
647 // i += 2;
648 // }
649 }
650 }
651
652 static void RemoveSpikes3(float *vec, uint32_t roi)//from Werner
653 {
654 if (roi<4)
655 return;
656
657 const float SingleCandidateTHR = -10.;
658 const float DoubleCandidateTHR = -5.;
659
660 const std::vector<float> src(vec, vec+roi);
661
662 std::vector<float> diff(roi);
663 for (size_t i=1; i<roi-1; i++)
664 diff[i] = src[i] - (src[i-1] + src[i+1])/2;
665
666 // find the spike and replace it by mean value of neighbors
667 for (unsigned int i=1; i<roi-3; i++)
668 {
669 // Speed up (no leading edge)
670 if (diff[i]>=DoubleCandidateTHR)
671 continue;
672
673 //bool checkDouble = false;
674
675 // a single spike candidate
676 if (diff[i]<SingleCandidateTHR)
677 {
678 // check consistency with a single channel spike
679 if (diff[i+1] > -1.6*diff[i])
680 {
681 vec[i+1] = (src[i] + src[i+2]) / 2;
682
683 i += 2;
684
685 /*** NEW ***/
686 continue;
687 /*** NEW ***/
688 }
689 /*
690 else
691 {
692 // do nothing - not really a single spike,
693 // but check if it is a double
694 checkDouble = true;
695 }*/
696 }
697
698 // a double spike candidate
699 //if (diff[i]>DoubleCandidateTHR || checkDouble == 1)
700 {
701 // check the consistency with a double spike
702 if ((diff[i+1] > -DoubleCandidateTHR) &&
703 (diff[i+2] > -DoubleCandidateTHR))
704 {
705 vec[i+1] = (src[i+3] - src[i])/3 + src[i];
706 vec[i+2] = 2*(src[i+3] - src[i])/3 + src[i];
707
708 //vec[i] = (src[i-1] + src[i+2]) / 2.;
709 //vec[i+1] = (src[i-1] + src[i+2]) / 2.;
710
711 //do not care about the next sample it was the spike
712 i += 3;
713 }
714 }
715 }
716 }
717
718 static void RemoveSpikes4(float *ptr, uint32_t roi)
719 {
720 if (roi<7)
721 return;
722
723 for (uint32_t i=0; i<roi-6; i++)
724 {
725 double d10, d21, d32, d43, d54;
726
727 // ============================================
728 d43 = ptr[i+4]-ptr[i+3];
729 d54 = ptr[i+5]-ptr[i+4];
730
731 if ((d43>35 && -d54>35) || (d43<-35 && -d54<-35))
732 {
733 ptr[i+4] = (ptr[i+3]+ptr[i+5])/2;
734 }
735
736 // ============================================
737 d32 = ptr[i+3]-ptr[i+2];
738 d54 = ptr[i+5]-ptr[i+4];
739
740 if ((d32>9 && -d54>13 && d32-d54>31)/* || (d32<-13 && -d54<-13 && d32+d54<-63)*/)
741 {
742 double avg0 = (ptr[i+2]+ptr[i+5])/2;
743 double avg1 = (ptr[i+3]+ptr[i+4])/2;
744
745 ptr[i+3] = ptr[i+3] - avg1+avg0;
746 ptr[i+4] = ptr[i+4] - avg1+avg0;
747 }
748
749 // ============================================
750 d21 = ptr[i+2]-ptr[i+1];
751 d54 = ptr[i+5]-ptr[i+4];
752
753 if (d21>15 && -d54>17)
754 {
755 double avg0 = (ptr[i+1]+ptr[i+5])/2;
756 double avg1 = (ptr[i+2]+ptr[i+3]+ptr[i+4])/3;
757
758 ptr[i+2] = ptr[i+2] - avg1+avg0;
759 ptr[i+3] = ptr[i+3] - avg1+avg0;
760 ptr[i+4] = ptr[i+4] - avg1+avg0;
761 }
762
763 // ============================================
764 d10 = ptr[i+1]-ptr[i];
765 d54 = ptr[i+5]-ptr[i+4];
766
767 if (d10>18 && -d54>20)
768 {
769 double avg0 = (ptr[i]+ptr[i+5])/2;
770 double avg1 = (ptr[i+1]+ptr[i+2]+ptr[i+3]+ptr[i+4])/4;
771
772 ptr[i+1] = ptr[i+1] - avg1+avg0;
773 ptr[i+2] = ptr[i+2] - avg1+avg0;
774 ptr[i+3] = ptr[i+3] - avg1+avg0;
775 ptr[i+4] = ptr[i+4] - avg1+avg0;
776 }
777 }
778 }
779
780 static void SlidingAverage(float *const vec, const uint32_t roi, const uint16_t w)
781 {
782 if (w==0 || w>roi)
783 return;
784
785 for (float *pix=vec; pix<vec+1440*roi; pix += roi)
786 {
787 for (float *ptr=pix; ptr<pix+roi-w; ptr++)
788 {
789 for (float *p=ptr+1; p<ptr+w; p++)
790 *ptr += *p;
791 *ptr /= w;
792 }
793 }
794 }
795
796 std::pair<std::vector<double>,std::vector<double> > GetSampleStats() const
797 {
798 if (fNumEntries==0)
799 return make_pair(std::vector<double>(),std::vector<double>());
800
801 std::vector<double> mean(fSum.size());
802 std::vector<double> error(fSum.size());
803
804 std::vector<int64_t>::const_iterator it = fSum.begin();
805 std::vector<int64_t>::const_iterator i2 = fSum2.begin();
806 std::vector<double>::iterator im = mean.begin();
807 std::vector<double>::iterator ie = error.begin();
808
809 while (it!=fSum.end())
810 {
811 *im = /*cnt<fResult.size() ? fResult[cnt] :*/ double(*it)/fNumEntries;
812 *ie = sqrt(double(*i2*int64_t(fNumEntries) - *it * *it))/fNumEntries;
813
814 im++;
815 ie++;
816 it++;
817 i2++;
818 }
819
820
821 /*
822 valarray<double> ...
823
824 mean /= fNumEntries;
825 error = sqrt(error/fNumEntries - mean*mean);
826 */
827
828 return make_pair(mean, error);
829 }
830
831 void GetSampleStats(float *ptr, float scale) const
832 {
833 const size_t sz = fNumSamples*fNumChannels;
834
835 if (fNumEntries==0)
836 {
837 memset(ptr, 0, sizeof(float)*sz*2);
838 return;
839 }
840
841 std::vector<int64_t>::const_iterator it = fSum.begin();
842 std::vector<int64_t>::const_iterator i2 = fSum2.begin();
843
844 while (it!=fSum.end())
845 {
846 *ptr = scale*double(*it)/fNumEntries;
847 *(ptr+sz) = scale*sqrt(double(*i2*fNumEntries - *it * *it))/fNumEntries;
848
849 ptr++;
850 it++;
851 i2++;
852 }
853 }
854
855 static double GetPixelStats(float *ptr, const float *data, uint16_t roi, uint16_t begskip=0, uint16_t endskip=0)
856 {
857 if (roi==0)
858 return -1;
859
860 // Skip first 10 samples
861 const uint beg = roi>begskip ? begskip : 0;
862 const uint end = roi-beg>endskip ? roi-endskip : roi;
863 const uint len = end-beg;
864
865 double max = 0;
866 double patch = 0;
867 for (uint i=0; i<1440; i++)
868 {
869 const float *vec = data+i*roi;
870
871 uint pos = beg;
872 double sum = vec[beg];
873 double sum2 = vec[beg]*vec[beg];
874
875 for (uint j=beg+1; j<end; j++)
876 {
877 sum += vec[j];
878 sum2 += vec[j]*vec[j];
879
880 if (vec[j]>vec[pos])
881 pos = j;
882 }
883 sum /= len;
884 sum2 /= len;
885 sum2 -= sum*sum;
886
887 if (i%9!=8)
888 patch += vec[pos];
889 else
890 {
891 if (patch > max)
892 max = patch;
893 patch = 0;
894 }
895
896 *(ptr+0*1440+i) = sum;
897 *(ptr+1*1440+i) = sum2<0 ? 0 : sqrt(sum2);
898 *(ptr+2*1440+i) = vec[pos];
899 *(ptr+3*1440+i) = pos;
900 }
901
902 return max/8;
903 }
904
905 static void GetPixelMax(float *max, const float *data, uint16_t roi, int32_t first, int32_t last)
906 {
907 if (roi==0 || first<0 || last<0 || first>=roi || last>=roi || last<first)
908 return;
909
910 for (int i=0; i<1440; i++)
911 {
912 const float *beg = data+i*roi+first;
913 const float *end = data+i*roi+last;
914
915 const float *pmax = beg;
916
917 for (const float *ptr=beg+1; ptr<=end; ptr++)
918 if (*ptr>*pmax)
919 pmax = ptr;
920
921 max[i] = *pmax;
922 }
923 }
924
925 const std::vector<int64_t> &GetSum() const { return fSum; }
926
927 int64_t GetNumEntries() const { return fNumEntries; }
928};
929
930class DrsCalibrateTime
931{
932public:
933 int64_t fNumEntries;
934
935 size_t fNumSamples;
936 size_t fNumChannels;
937
938 std::vector<std::pair<double, double>> fStat;
939
940public:
941 DrsCalibrateTime() : fNumEntries(0), fNumSamples(0), fNumChannels(0)
942 {
943 InitSize(160, 1024);
944 }
945
946 DrsCalibrateTime(const DrsCalibrateTime &p) : fNumEntries(p.fNumEntries), fNumSamples(p.fNumSamples), fNumChannels(p.fNumChannels), fStat(p.fStat)
947 {
948 }
949 virtual ~DrsCalibrateTime()
950 {
951 }
952
953 double Sum(uint32_t i) const { return fStat[i].first; }
954 double W(uint32_t i) const { return fStat[i].second; }
955
956 virtual void InitSize(uint16_t channels, uint16_t samples)
957 {
958 fNumChannels = channels;
959 fNumSamples = samples;
960
961 fNumEntries = 0;
962
963 fStat.clear();
964
965 fStat.resize(samples*channels);
966 }
967
968 void Reset()
969 {
970 for (auto it=fStat.begin(); it!=fStat.end(); it++)
971 {
972 it->first = 0;
973 it->second = 0;
974 }
975 }
976
977 void AddT(const float *val, const int16_t *start, signed char edge=0)
978 {
979 if (fNumSamples!=1024 || fNumChannels!=160)
980 return;
981
982 // Rising or falling edge detection has the advantage that
983 // we are much less sensitive to baseline shifts
984
985 for (size_t ch=0; ch<160; ch++)
986 {
987 const size_t tm = ch*9+8;
988
989 const int16_t spos = start[tm];
990 if (spos<0)
991 continue;
992
993 const size_t pos = ch*1024;
994
995 double p_prev = 0;
996 int32_t i_prev = -1;
997
998 for (size_t i=0; i<1024-1; i++)
999 {
1000 const size_t rel = tm*1024 + i;
1001
1002 const float &v0 = val[rel]; //-avg;
1003 const float &v1 = val[rel+1];//-avg;
1004
1005 // If edge is positive ignore all falling edges
1006 if (edge>0 && v0>0)
1007 continue;
1008
1009 // If edge is negative ignore all falling edges
1010 if (edge<0 && v0<0)
1011 continue;
1012
1013 // Check if there is a zero crossing
1014 if ((v0<0 && v1<0) || (v0>0 && v1>0))
1015 continue;
1016
1017 // Calculate the position p of the zero-crossing
1018 // within the interval [rel, rel+1] relative to rel
1019 // by linear interpolation.
1020 const double p = v0==v1 ? 0.5 : v0/(v0-v1);
1021
1022 // If this was at least the second zero-crossing detected
1023 if (i_prev>=0)
1024 {
1025 // Calculate the distance l between the
1026 // current and the last zero-crossing
1027 const double l = i+p - (i_prev+p_prev);
1028
1029 // By summation, the average length of each
1030 // cell is calculated. For the first and last
1031 // fraction of a cell, the fraction is applied
1032 // as a weight.
1033 const double w0 = 1-p_prev;
1034 fStat[pos+(spos+i_prev)%1024].first += w0*l;
1035 fStat[pos+(spos+i_prev)%1024].second += w0;
1036
1037 for (size_t k=i_prev+1; k<i; k++)
1038 {
1039 fStat[pos+(spos+k)%1024].first += l;
1040 fStat[pos+(spos+k)%1024].second += 1;
1041 }
1042
1043 const double w1 = p;
1044 fStat[pos+(spos+i)%1024].first += w1*l;
1045 fStat[pos+(spos+i)%1024].second += w1;
1046 }
1047
1048 // Remember this zero-crossing position
1049 p_prev = p;
1050 i_prev = i;
1051 }
1052 }
1053 fNumEntries++;
1054 }
1055
1056 void FillEmptyBins()
1057 {
1058 for (int ch=0; ch<160; ch++)
1059 {
1060 const auto beg = fStat.begin() + ch*1024;
1061 const auto end = beg + 1024;
1062
1063 double avg = 0;
1064 uint32_t num = 0;
1065 for (auto it=beg; it!=end; it++)
1066 {
1067 if (it->second<fNumEntries-0.5)
1068 continue;
1069
1070 avg += it->first / it->second;
1071 num++;
1072 }
1073 avg /= num;
1074
1075 for (auto it=beg; it!=end; it++)
1076 {
1077 if (it->second>=fNumEntries-0.5)
1078 continue;
1079
1080 // {
1081 // result[i+1].first = *is2;
1082 // result[i+1].second = *iw2;
1083 // }
1084 // else
1085 // {
1086 it->first = avg*fNumEntries;
1087 it->second = fNumEntries;
1088 // }
1089 }
1090 }
1091 }
1092
1093 DrsCalibrateTime GetComplete() const
1094 {
1095 DrsCalibrateTime rc(*this);
1096 rc.FillEmptyBins();
1097 return rc;
1098 }
1099
1100 void CalcResult()
1101 {
1102 for (int ch=0; ch<160; ch++)
1103 {
1104 const auto beg = fStat.begin() + ch*1024;
1105 const auto end = beg + 1024;
1106
1107 // First calculate the average length s of a single
1108 // zero-crossing interval in the whole range [0;1023]
1109 // (which is identical to the/ wavelength of the
1110 // calibration signal)
1111 double s = 0;
1112 double w = 0;
1113 for (auto it=beg; it!=end; it++)
1114 {
1115 s += it->first;
1116 w += it->second;
1117 }
1118 s /= w;
1119
1120 // Dividing the average length s of the zero-crossing
1121 // interval in the range [0;1023] by the average length
1122 // in the interval [0;n] yields the relative size of
1123 // the interval in the range [0;n].
1124 //
1125 // Example:
1126 // Average [0;1023]: 10.00 (global interval size in samples)
1127 // Average [0;512]: 8.00 (local interval size in samples)
1128 //
1129 // Globally, on average one interval is sampled by 10 samples.
1130 // In the sub-range [0;512] one interval is sampled on average
1131 // by 8 samples.
1132 // That means that the interval contains 64 periods, while
1133 // in the ideal case (each sample has the same length), it
1134 // should contain 51.2 periods.
1135 // So, the sampling position 512 corresponds to a time 640,
1136 // while in the ideal case with equally spaces samples,
1137 // it would correspond to a time 512.
1138 //
1139 // The offset (defined as 'ideal - real') is then calculated
1140 // as 512*(1-10/8) = -128, so that the time is calculated as
1141 // 'sampling position minus offset'
1142 //
1143 double sumw = 0;
1144 double sumv = 0;
1145 int n = 0;
1146
1147 // Sums about many values are numerically less stable than
1148 // just sums over less. So we do the exercise from both sides.
1149 // First from the left
1150 for (auto it=beg; it!=end-512; it++, n++)
1151 {
1152 const double valv = it->first;
1153 const double valw = it->second;
1154
1155 it->first = sumv>0 ? n*(1-s*sumw/sumv) : 0;
1156
1157 sumv += valv;
1158 sumw += valw;
1159 }
1160
1161 sumw = 0;
1162 sumv = 0;
1163 n = 1;
1164
1165 // Second from the right
1166 for (auto it=end-1; it!=beg-1+512; it--, n++)
1167 {
1168 const double valv = it->first;
1169 const double valw = it->second;
1170
1171 sumv += valv;
1172 sumw += valw;
1173
1174 it->first = sumv>0 ? n*(s*sumw/sumv-1) : 0;
1175 }
1176
1177 // A crosscheck has shown, that the values from the left
1178 // and right perfectly agree over the whole range. This means
1179 // the a calculation from just one side would be enough, but
1180 // doing it from both sides might still make the numerics
1181 // a bit more stable.
1182 }
1183 }
1184
1185 DrsCalibrateTime GetResult() const
1186 {
1187 DrsCalibrateTime rc(*this);
1188 rc.CalcResult();
1189 return rc;
1190 }
1191
1192 double Offset(uint32_t ch, double pos) const
1193 {
1194 const auto p = fStat.begin() + ch*1024;
1195
1196 const uint32_t f = floor(pos);
1197
1198 const double v0 = p[f].first;
1199 const double v1 = p[(f+1)%1024].first;
1200
1201 return v0 + fmod(pos, 1)*(v1-v0);
1202 }
1203
1204 double Calib(uint32_t ch, double pos) const
1205 {
1206 return pos-Offset(ch, pos);
1207 }
1208
1209 /*
1210 // See MDrsCalibrationTime
1211 std::string ReadFitsImp(const std::string &str)
1212 {
1213 fits file(str);
1214 if (!file)
1215 {
1216 std::ostringstream msg;
1217 msg << "Could not open file '" << str << "': " << strerror(errno);
1218 return msg.str();
1219 }
1220
1221 if (file.GetStr("TELESCOP")!="FACT")
1222 {
1223 std::ostringstream msg;
1224 msg << "Reading '" << str << "' failed: Not a valid FACT file (TELESCOP not FACT in header)";
1225 return msg.str();
1226 }
1227
1228 if (file.GetNumRows()!=1)
1229 {
1230 std::ostringstream msg;
1231 msg << "Reading '" << str << "' failed: Number of rows in table is not 1.";
1232 return msg.str();
1233 }
1234
1235 fNumSamples = file.GetUInt("NROI");
1236 fNumChannels = file.GetUInt("NCH");
1237 fNumEntries = file.GetUInt("NBTIME");
1238
1239 fStat.resize(fNumSamples*fNumChannels);
1240
1241 double *f = reinterpret_cast<double*>(file.SetPtrAddress("CellWidthMean"));
1242 double *s = reinterpret_cast<double*>(file.SetPtrAddress("CellWidthRms"));
1243
1244 if (!file.GetNextRow())
1245 {
1246 std::ostringstream msg;
1247 msg << "Reading data from " << str << " failed.";
1248 return msg.str();
1249 }
1250
1251 for (uint32_t i=0; i<fNumSamples*fNumChannels; i++)
1252 {
1253 fStat[i].first = f[i];
1254 fStat[i].second = s[i];
1255 }
1256
1257 return std::string();
1258 }
1259
1260 std::string WriteFitsImp(const std::string &filename, uint32_t night=0) const
1261 {
1262 ofits file(filename.c_str());
1263 if (!file)
1264 {
1265 std::ostringstream msg;
1266 msg << "Could not open file '" << filename << "': " << strerror(errno);
1267 return msg.str();
1268 }
1269
1270 file.SetDefaultKeys();
1271 file.AddColumnDouble(fStat.size(), "CellWidthMean", "ratio", "Relative cell width mean");
1272 file.AddColumnDouble(fStat.size(), "CellWidthRms", "ratio", "Relative cell width rms");
1273
1274 file.SetInt("ADCRANGE", 2000, "Dynamic range of the ADC in mV");
1275 file.SetInt("DACRANGE", 2500, "Dynamic range of the DAC in mV");
1276 file.SetInt("ADC", 12, "Resolution of ADC in bits");
1277 file.SetInt("DAC", 16, "Resolution of DAC in bits");
1278 file.SetInt("NPIX", 1440, "Number of channels in the camera");
1279 file.SetInt("NTM", 0, "Number of time marker channels");
1280 file.SetInt("NROI", fNumSamples, "Region of interest");
1281 file.SetInt("NCH", fNumChannels, "Number of chips");
1282 file.SetInt("NBTIME", fNumEntries, "Num of entries for time calibration");
1283
1284 file.WriteTableHeader("DrsCellTimes");
1285 if (night>0)
1286 file.SetInt("NIGHT", night, "Night as int");
1287
1288 std::vector<double> data(fNumSamples*fNumChannels*2);
1289
1290 for (uint32_t i=0; i<fNumSamples*fNumChannels; i++)
1291 {
1292 data[i] = fStat[i].first;
1293 data[fNumSamples*fNumChannels+i] = fStat[i].second;
1294 }
1295
1296 if (!file.WriteRow(data.data(), data.size()*sizeof(double)))
1297 {
1298 std::ostringstream msg;
1299 msg << "Writing data to " << filename << " failed.";
1300 return msg.str();
1301 }
1302
1303 return std::string();
1304 }*/
1305};
1306
1307struct DrsCalibration
1308{
1309 std::vector<int32_t> fOffset;
1310 std::vector<int64_t> fGain;
1311 std::vector<int64_t> fTrgOff;
1312
1313 int64_t fNumOffset;
1314 int64_t fNumGain;
1315 int64_t fNumTrgOff;
1316
1317 uint32_t fStep;
1318 uint16_t fRoi; // Region of interest for trgoff
1319 uint16_t fNumTm; // Number of time marker channels in trgoff
1320
1321 std::string fDateObs;
1322 std::string fDateRunBeg[3];
1323 std::string fDateRunEnd[3];
1324 std::string fDateEnd;
1325
1326// uint16_t fDAC[8];
1327
1328 DrsCalibration() :
1329 fOffset(1440*1024, 0),
1330 fGain(1440*1024, 4096),
1331 fTrgOff (1600*1024, 0),
1332 fNumOffset(1),
1333 fNumGain(2000),
1334 fNumTrgOff(1),
1335 fStep(0),
1336 fDateObs("1970-01-01T00:00:00"),
1337 fDateEnd("1970-01-01T00:00:00")
1338 {
1339 for (int i=0; i<3; i++)
1340 {
1341 fDateRunBeg[i] = "1970-01-01T00:00:00";
1342 fDateRunEnd[i] = "1970-01-01T00:00:00";
1343 }
1344 }
1345
1346 DrsCalibration(const DrsCalibration &cpy) :
1347 fOffset(cpy.fOffset),
1348 fGain(cpy.fGain),
1349 fTrgOff(cpy.fTrgOff),
1350 fNumOffset(cpy.fNumOffset),
1351 fNumGain(cpy.fNumGain),
1352 fNumTrgOff(cpy.fNumTrgOff),
1353 fStep(cpy.fStep),
1354 fRoi(cpy.fRoi),
1355 fNumTm(cpy.fNumTm),
1356 fDateObs(cpy.fDateObs),
1357 fDateRunBeg(cpy.fDateRunBeg),
1358 fDateRunEnd(cpy.fDateRunEnd),
1359 fDateEnd(cpy.fDateEnd)
1360 {
1361 }
1362
1363 void Clear()
1364 {
1365 // Default gain:
1366 // 0.575*[45590]*2.5V / 2^16 = 0.99999 V
1367 fOffset.assign(1440*1024, 0);
1368 fGain.assign (1440*1024, 4096);
1369 fTrgOff.assign(1600*1024, 0);
1370
1371 fNumOffset = 1;
1372 fNumGain = 2000;
1373 fNumTrgOff = 1;
1374
1375 fStep = 0;
1376
1377 fDateObs = "1970-01-01T00:00:00";
1378 fDateEnd = "1970-01-01T00:00:00";
1379
1380 for (int i=0; i<3; i++)
1381 {
1382 fDateRunBeg[i] = "1970-01-01T00:00:00";
1383 fDateRunEnd[i] = "1970-01-01T00:00:00";
1384 }
1385 }
1386
1387 std::string ReadFitsImp(const std::string &str, std::vector<float> &vec)
1388 {
1389 fits file(str);
1390 if (!file)
1391 {
1392 std::ostringstream msg;
1393 msg << "Could not open file '" << str << "': " << strerror(errno);
1394 return msg.str();
1395 }
1396
1397 if (file.GetStr("TELESCOP")!="FACT")
1398 {
1399 std::ostringstream msg;
1400 msg << "Reading '" << str << "' failed: Not a valid FACT file (TELESCOP not FACT in header)";
1401 return msg.str();
1402 }
1403
1404 if (!file.HasKey("STEP"))
1405 {
1406 std::ostringstream msg;
1407 msg << "Reading '" << str << "' failed: Is not a DRS calib file (STEP not found in header)";
1408 return msg.str();
1409 }
1410
1411 if (file.GetNumRows()!=1)
1412 {
1413 std::ostringstream msg;
1414 msg << "Reading '" << str << "' failed: Number of rows in table is not 1.";
1415 return msg.str();
1416 }
1417
1418 fStep = file.GetUInt("STEP");
1419 fNumOffset = file.GetInt("NBOFFSET");
1420 fNumGain = file.GetInt("NBGAIN");
1421 fNumTrgOff = file.GetInt("NBTRGOFF");
1422 fRoi = file.GetUInt("NROI");
1423 fNumTm = file.HasKey("NTM") ? file.GetUInt("NTM") : 0;
1424
1425 if (file.HasKey("DATE-OBS"))
1426 fDateObs = file.GetStr("DATE-OBS");
1427 if (file.HasKey("DATE-END"))
1428 fDateEnd = file.GetStr("DATE-END");
1429
1430 if (file.HasKey("RUN0-BEG"))
1431 fDateRunBeg[0]= file.GetStr("RUN0-BEG");
1432 if (file.HasKey("RUN1-BEG"))
1433 fDateRunBeg[1]= file.GetStr("RUN1-BEG");
1434 if (file.HasKey("RUN2-BEG"))
1435 fDateRunBeg[2]= file.GetStr("RUN2-BEG");
1436 if (file.HasKey("RUN0-END"))
1437 fDateRunEnd[0]= file.GetStr("RUN0-END");
1438 if (file.HasKey("RUN1-END"))
1439 fDateRunEnd[1]= file.GetStr("RUN1-END");
1440 if (file.HasKey("RUN2-END"))
1441 fDateRunEnd[2]= file.GetStr("RUN2-END");
1442/*
1443 fDAC[0] = file.GetUInt("DAC_A");
1444 fDAC[1] = file.GetUInt("DAC_B");
1445 fDAC[4] = file.GetUInt("DAC_C");
1446*/
1447 vec.resize(1440*1024*4 + (1440+fNumTm)*fRoi*2 + 4);
1448
1449 float *base = vec.data();
1450
1451 reinterpret_cast<uint32_t*>(base)[0] = fRoi;
1452
1453 file.SetPtrAddress("RunNumberBaseline", base+1, 1);
1454 file.SetPtrAddress("RunNumberGain", base+2, 1);
1455 file.SetPtrAddress("RunNumberTriggerOffset", base+3, 1);
1456 file.SetPtrAddress("BaselineMean", base+4+0*1024*1440, 1024*1440);
1457 file.SetPtrAddress("BaselineRms", base+4+1*1024*1440, 1024*1440);
1458 file.SetPtrAddress("GainMean", base+4+2*1024*1440, 1024*1440);
1459 file.SetPtrAddress("GainRms", base+4+3*1024*1440, 1024*1440);
1460 file.SetPtrAddress("TriggerOffsetMean", base+4+4*1024*1440, fRoi*1440);
1461 file.SetPtrAddress("TriggerOffsetRms", base+4+4*1024*1440+fRoi*1440, fRoi*1440);
1462 if (fNumTm>0)
1463 {
1464 file.SetPtrAddress("TriggerOffsetTMMean", base+4+4*1024*1440+ 2*fRoi*1440, fRoi*fNumTm);
1465 file.SetPtrAddress("TriggerOffsetTMRms", base+4+4*1024*1440+ 2*fRoi*1440+ fRoi*fNumTm, fRoi*fNumTm);
1466 }
1467
1468 if (!file.GetNextRow())
1469 {
1470 std::ostringstream msg;
1471 msg << "Reading data from " << str << " failed.";
1472 return msg.str();
1473 }
1474/*
1475 fDAC[2] = fDAC[1];
1476 fDAC[4] = fDAC[1];
1477
1478 fDAC[5] = fDAC[4];
1479 fDAC[6] = fDAC[4];
1480 fDAC[7] = fDAC[4];
1481*/
1482 fOffset.resize(1024*1440);
1483 fGain.resize(1024*1440);
1484
1485 fTrgOff.resize(fRoi*(1440+fNumTm));
1486
1487 // Convert back to ADC counts: 256/125 = 4096/2000
1488 // Convert back to sum (mean * num_entries)
1489 for (int i=0; i<1024*1440; i++)
1490 {
1491 fOffset[i] = fNumOffset *256*base[i+1024*1440*0+4]/125;
1492 fGain[i] = fNumOffset*fNumGain*256*base[i+1024*1440*2+4]/125;
1493 }
1494
1495 for (int i=0; i<fRoi*1440; i++)
1496 fTrgOff[i] = fNumOffset*fNumTrgOff*256*base[i+1024*1440*4+4]/125;
1497
1498 for (int i=0; i<fRoi*fNumTm; i++)
1499 fTrgOff[i+1440*fRoi] = fNumOffset*fNumTrgOff*256*base[i+1024*1440*4+2*fRoi*1440+4]/125;
1500
1501
1502 // DAC: 0..2.5V == 0..65535
1503 // V-mV: 1000
1504 //fNumGain *= 2500*50000;
1505 //for (int i=0; i<1024*1440; i++)
1506 // fGain[i] *= 65536;
1507 if (fStep==0)
1508 {
1509 for (int i=0; i<1024*1440; i++)
1510 fGain[i] = fNumOffset*4096;
1511 }
1512 else
1513 {
1514 fNumGain *= 1953125;
1515 for (int i=0; i<1024*1440; i++)
1516 fGain[i] *= 1024;
1517 }
1518
1519 // Now mark the stored DRS data as "officially valid"
1520 // However, this is not thread safe. It only ensures that
1521 // this data is not used before it is completely and correctly
1522 // read.
1523 fStep++;
1524
1525 return std::string();
1526 }
1527
1528 std::string WriteFitsImp(const std::string &filename, const std::vector<float> &vec, uint32_t night=0) const
1529 {
1530 const size_t n = 1440*1024*4 + 1440*fRoi*2 + fNumTm*fRoi*2 + 3;
1531
1532 ofits file(filename.c_str());
1533 if (!file)
1534 {
1535 std::ostringstream msg;
1536 msg << "Could not open file '" << filename << "': " << strerror(errno);
1537 return msg.str();
1538 }
1539
1540 file.AddColumnInt("RunNumberBaseline");
1541 file.AddColumnInt("RunNumberGain");
1542 file.AddColumnInt("RunNumberTriggerOffset");
1543
1544 file.AddColumnFloat(1024*1440, "BaselineMean", "mV");
1545 file.AddColumnFloat(1024*1440, "BaselineRms", "mV");
1546 file.AddColumnFloat(1024*1440, "GainMean", "mV");
1547 file.AddColumnFloat(1024*1440, "GainRms", "mV");
1548 file.AddColumnFloat(fRoi*1440, "TriggerOffsetMean", "mV");
1549 file.AddColumnFloat(fRoi*1440, "TriggerOffsetRms", "mV");
1550 file.AddColumnFloat(fRoi*fNumTm, "TriggerOffsetTMMean", "mV");
1551 file.AddColumnFloat(fRoi*fNumTm, "TriggerOffsetTMRms", "mV");
1552
1553 file.SetDefaultKeys();
1554 if (night>0)
1555 file.SetInt("NIGHT", night, "Night as int");
1556
1557 file.SetStr("DATE-OBS", fDateObs, "First event of whole DRS calibration");
1558 file.SetStr("DATE-END", fDateEnd, "Last event of whole DRS calibration");
1559 file.SetStr("RUN0-BEG", fDateRunBeg[0], "First event of run 0");
1560 file.SetStr("RUN1-BEG", fDateRunBeg[1], "First event of run 1");
1561 file.SetStr("RUN2-BEG", fDateRunBeg[2], "First event of run 2");
1562 file.SetStr("RUN0-END", fDateRunEnd[0], "Last event of run 0");
1563 file.SetStr("RUN1-END", fDateRunEnd[1], "Last event of run 1");
1564 file.SetStr("RUN2-END", fDateRunEnd[2], "Last event of run 2");
1565
1566 file.SetInt("STEP", fStep, "");
1567
1568 file.SetInt("ADCRANGE", 2000, "Dynamic range of the ADC in mV");
1569 file.SetInt("DACRANGE", 2500, "Dynamic range of the DAC in mV");
1570 file.SetInt("ADC", 12, "Resolution of ADC in bits");
1571 file.SetInt("DAC", 16, "Resolution of DAC in bits");
1572 file.SetInt("NPIX", 1440, "Number of channels in the camera");
1573 file.SetInt("NTM", fNumTm, "Number of time marker channels");
1574 file.SetInt("NROI", fRoi, "Region of interest");
1575
1576 file.SetInt("NBOFFSET", fNumOffset, "Num of entries for offset calibration");
1577 file.SetInt("NBGAIN", fNumGain/1953125, "Num of entries for gain calibration");
1578 file.SetInt("NBTRGOFF", fNumTrgOff, "Num of entries for trigger offset calibration");
1579
1580 // file.WriteKeyNT("DAC_A", fData.fDAC[0], "Level of DAC 0 in DAC counts") ||
1581 // file.WriteKeyNT("DAC_B", fData.fDAC[1], "Leval of DAC 1-3 in DAC counts") ||
1582 // file.WriteKeyNT("DAC_C", fData.fDAC[4], "Leval of DAC 4-7 in DAC counts") ||
1583
1584 file.WriteTableHeader("DrsCalibration");
1585
1586 if (!file.WriteRow(vec.data()+1, n*sizeof(float)))
1587 {
1588 std::ostringstream msg;
1589 msg << "Writing data to " << filename << " failed.";
1590 return msg.str();
1591 }
1592
1593 return std::string();
1594 }
1595
1596
1597 std::string ReadFitsImp(const std::string &str)
1598 {
1599 std::vector<float> vec;
1600 return ReadFitsImp(str, vec);
1601 }
1602
1603 bool IsValid() { return fStep>2; }
1604
1605 bool Apply(float *vec, const int16_t *val, const int16_t *start, uint32_t roi)
1606 {
1607 if (roi!=fRoi)
1608 {
1609 for (size_t ch=0; ch<1440; ch++)
1610 {
1611 const size_t pos = ch*roi;
1612 const size_t drs = ch*1024;
1613
1614 DrsCalibrate::ApplyCh(vec+pos, val+pos, start[ch], roi,
1615 fOffset.data()+drs, fNumOffset,
1616 fGain.data() +drs, fNumGain);
1617 }
1618
1619 return false;
1620 }
1621
1622 for (size_t ch=0; ch<1440; ch++)
1623 {
1624 const size_t pos = ch*fRoi;
1625 const size_t drs = ch*1024;
1626
1627 DrsCalibrate::ApplyCh(vec+pos, val+pos, start[ch], roi,
1628 fOffset.data()+drs, fNumOffset,
1629 fGain.data() +drs, fNumGain,
1630 fTrgOff.data()+pos, fNumTrgOff);
1631 }
1632
1633 for (size_t ch=0; ch<fNumTm; ch++)
1634 {
1635 const size_t pos = (ch+1440)*fRoi;
1636 const size_t drs = (ch*9+8)*1024;
1637
1638 DrsCalibrate::ApplyCh(vec+pos, val+pos, start[ch], roi,
1639 fOffset.data()+drs, fNumOffset,
1640 fGain.data() +drs, fNumGain,
1641 fTrgOff.data()+pos, fNumTrgOff);
1642 }
1643
1644 return true;
1645 }
1646};
1647
1648#endif
Note: See TracBrowser for help on using the repository browser.