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 analyzing 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): Markus Gaug 05/2004 <mailto:markus@ifae.es>
|
---|
18 | !
|
---|
19 | ! Copyright: MAGIC Software Development, 2002-2004
|
---|
20 | !
|
---|
21 | !
|
---|
22 | \* ======================================================================== */
|
---|
23 |
|
---|
24 | //////////////////////////////////////////////////////////////////////////////
|
---|
25 | //
|
---|
26 | // MExtractTimeFastSpline
|
---|
27 | //
|
---|
28 | // Fast arrival Time extractor using a cubic spline algorithm of Numerical Recipes.
|
---|
29 | // It returns the position of the half maximum between absolute maximum
|
---|
30 | // and pedestal of the spline that interpolates the FADC slices.
|
---|
31 | //
|
---|
32 | // The precision of the half-maximum searches can be chosen by:
|
---|
33 | // SetPrecision().
|
---|
34 | //
|
---|
35 | // The precision of the maximum-finder is fixed to 0.025 FADC units.
|
---|
36 | //
|
---|
37 | //////////////////////////////////////////////////////////////////////////////
|
---|
38 | #include "MExtractTimeFastSpline.h"
|
---|
39 |
|
---|
40 | #include "MPedestalPix.h"
|
---|
41 |
|
---|
42 | #include "MLog.h"
|
---|
43 | #include "MLogManip.h"
|
---|
44 |
|
---|
45 |
|
---|
46 | ClassImp(MExtractTimeFastSpline);
|
---|
47 |
|
---|
48 | using namespace std;
|
---|
49 |
|
---|
50 | const Byte_t MExtractTimeFastSpline::fgHiGainFirst = 2;
|
---|
51 | const Byte_t MExtractTimeFastSpline::fgHiGainLast = 14;
|
---|
52 | const Byte_t MExtractTimeFastSpline::fgLoGainFirst = 3;
|
---|
53 | const Byte_t MExtractTimeFastSpline::fgLoGainLast = 14;
|
---|
54 | const Float_t MExtractTimeFastSpline::fgResolution = 0.003;
|
---|
55 | const Float_t MExtractTimeFastSpline::fgRiseTime = 2.;
|
---|
56 |
|
---|
57 | // --------------------------------------------------------------------------
|
---|
58 | //
|
---|
59 | // Default constructor.
|
---|
60 | //
|
---|
61 | // Calls:
|
---|
62 | // - SetRange(fgHiGainFirst, fgHiGainLast, fgLoGainFirst, fgLoGainLast)
|
---|
63 | //
|
---|
64 | // Initializes:
|
---|
65 | // - fResolution to fgResolution
|
---|
66 | //
|
---|
67 | MExtractTimeFastSpline::MExtractTimeFastSpline(const char *name, const char *title)
|
---|
68 | : fHiGainFirstDeriv(NULL), fLoGainFirstDeriv(NULL),
|
---|
69 | fHiGainSecondDeriv(NULL), fLoGainSecondDeriv(NULL)
|
---|
70 | {
|
---|
71 |
|
---|
72 | fName = name ? name : "MExtractTimeFastSpline";
|
---|
73 | fTitle = title ? title : "Calculate photons arrival time using a fast spline";
|
---|
74 |
|
---|
75 | SetResolution();
|
---|
76 | SetRiseTime ();
|
---|
77 | SetRange(fgHiGainFirst, fgHiGainLast, fgLoGainFirst, fgLoGainLast);
|
---|
78 |
|
---|
79 | }
|
---|
80 |
|
---|
81 | MExtractTimeFastSpline::~MExtractTimeFastSpline()
|
---|
82 | {
|
---|
83 |
|
---|
84 | if (fHiGainFirstDeriv)
|
---|
85 | delete [] fHiGainFirstDeriv;
|
---|
86 | if (fLoGainFirstDeriv)
|
---|
87 | delete [] fLoGainFirstDeriv;
|
---|
88 | if (fHiGainSecondDeriv)
|
---|
89 | delete [] fHiGainSecondDeriv;
|
---|
90 | if (fLoGainSecondDeriv)
|
---|
91 | delete [] fLoGainSecondDeriv;
|
---|
92 |
|
---|
93 | }
|
---|
94 |
|
---|
95 |
|
---|
96 | // --------------------------------------------------------------------------
|
---|
97 | //
|
---|
98 | // SetRange:
|
---|
99 | //
|
---|
100 | // Calls:
|
---|
101 | // - MExtractor::SetRange(hifirst,hilast,lofirst,lolast);
|
---|
102 | // - Deletes x, if not NULL
|
---|
103 | // - Creates x according to the range
|
---|
104 | //
|
---|
105 | void MExtractTimeFastSpline::SetRange(Byte_t hifirst, Byte_t hilast, Byte_t lofirst, Byte_t lolast)
|
---|
106 | {
|
---|
107 |
|
---|
108 | MExtractor::SetRange(hifirst,hilast,lofirst,lolast);
|
---|
109 |
|
---|
110 | if (fHiGainFirstDeriv)
|
---|
111 | delete fHiGainFirstDeriv;
|
---|
112 | if (fLoGainFirstDeriv)
|
---|
113 | delete fLoGainFirstDeriv;
|
---|
114 | if (fHiGainSecondDeriv)
|
---|
115 | delete fHiGainSecondDeriv;
|
---|
116 | if (fLoGainSecondDeriv)
|
---|
117 | delete fLoGainSecondDeriv;
|
---|
118 |
|
---|
119 | Int_t range = fHiGainLast - fHiGainFirst + 1;
|
---|
120 |
|
---|
121 | if (range < 2)
|
---|
122 | {
|
---|
123 | *fLog << warn << GetDescriptor()
|
---|
124 | << Form("%s%2i%s%2i%s",": Hi-Gain Extraction range [",(int)fHiGainFirst,","
|
---|
125 | ,fHiGainLast,"] too small, ") << endl;
|
---|
126 | *fLog << warn << GetDescriptor()
|
---|
127 | << " will move higher limit to obtain 4 slices " << endl;
|
---|
128 | SetRange(fHiGainFirst, fHiGainLast+4-range,fLoGainFirst,fLoGainLast);
|
---|
129 | range = fHiGainLast - fHiGainFirst + 1;
|
---|
130 | }
|
---|
131 |
|
---|
132 |
|
---|
133 | fHiGainFirstDeriv = new Float_t[range];
|
---|
134 | memset(fHiGainFirstDeriv,0,range*sizeof(Float_t));
|
---|
135 | fHiGainSecondDeriv = new Float_t[range];
|
---|
136 | memset(fHiGainSecondDeriv,0,range*sizeof(Float_t));
|
---|
137 |
|
---|
138 | range = fLoGainLast - fLoGainFirst + 1;
|
---|
139 |
|
---|
140 | if (range >= 2)
|
---|
141 | {
|
---|
142 |
|
---|
143 | fLoGainFirstDeriv = new Float_t[range];
|
---|
144 | memset(fLoGainFirstDeriv,0,range*sizeof(Float_t));
|
---|
145 | fLoGainSecondDeriv = new Float_t[range];
|
---|
146 | memset(fLoGainSecondDeriv,0,range*sizeof(Float_t));
|
---|
147 |
|
---|
148 | }
|
---|
149 |
|
---|
150 | }
|
---|
151 |
|
---|
152 |
|
---|
153 | // --------------------------------------------------------------------------
|
---|
154 | //
|
---|
155 | // Calculates the arrival time for each pixel
|
---|
156 | //
|
---|
157 | void MExtractTimeFastSpline::FindTimeHiGain(Byte_t *first, Float_t &time, Float_t &dtime,
|
---|
158 | Byte_t &sat, const MPedestalPix &ped) const
|
---|
159 | {
|
---|
160 |
|
---|
161 | const Int_t range = fHiGainLast - fHiGainFirst + 1;
|
---|
162 | const Byte_t *end = first + range;
|
---|
163 | Byte_t *p = first;
|
---|
164 | Byte_t max = 0;
|
---|
165 | Byte_t maxpos = 0;
|
---|
166 |
|
---|
167 | //
|
---|
168 | // Check for saturation in all other slices
|
---|
169 | //
|
---|
170 | while (p<end)
|
---|
171 | {
|
---|
172 | if (*p > max)
|
---|
173 | {
|
---|
174 | max = *p;
|
---|
175 | maxpos = p-first;
|
---|
176 | }
|
---|
177 |
|
---|
178 | if (*p++ >= fSaturationLimit)
|
---|
179 | {
|
---|
180 | sat++;
|
---|
181 | break;
|
---|
182 | }
|
---|
183 | }
|
---|
184 |
|
---|
185 | //
|
---|
186 | // allow one saturated slice
|
---|
187 | //
|
---|
188 | if (sat > 1)
|
---|
189 | return;
|
---|
190 |
|
---|
191 | if (maxpos < 1)
|
---|
192 | {
|
---|
193 | time = -999.;
|
---|
194 | return;
|
---|
195 | }
|
---|
196 |
|
---|
197 | Float_t pp;
|
---|
198 |
|
---|
199 | p = first;
|
---|
200 | fHiGainSecondDeriv[0] = 0.;
|
---|
201 | fHiGainFirstDeriv[0] = 0.;
|
---|
202 |
|
---|
203 | for (Int_t i=1;i<range-1;i++)
|
---|
204 | {
|
---|
205 | p++;
|
---|
206 | pp = fHiGainSecondDeriv[i-1] + 4.;
|
---|
207 | fHiGainSecondDeriv[i] = -1.0/pp;
|
---|
208 | fHiGainFirstDeriv [i] = *(p+1) - 2.* *(p) + *(p-1);
|
---|
209 | fHiGainFirstDeriv [i] = (6.0*fHiGainFirstDeriv[i]-fHiGainFirstDeriv[i-1])/pp;
|
---|
210 | }
|
---|
211 |
|
---|
212 | fHiGainSecondDeriv[range-1] = 0.;
|
---|
213 |
|
---|
214 | for (Int_t k=range-2;k>0;k--)
|
---|
215 | fHiGainSecondDeriv[k] = fHiGainSecondDeriv[k]*fHiGainSecondDeriv[k+1] + fHiGainFirstDeriv[k];
|
---|
216 | for (Int_t k=range-2;k>0;k--)
|
---|
217 | fHiGainSecondDeriv[k] /= 6.;
|
---|
218 |
|
---|
219 | //
|
---|
220 | // Now find the maximum
|
---|
221 | //
|
---|
222 | Float_t step = 0.2; // start with step size of 1ns and loop again with the smaller one
|
---|
223 | Float_t lower = (Float_t)maxpos-1.;
|
---|
224 | Float_t upper = (Float_t)maxpos;
|
---|
225 | Float_t x = lower;
|
---|
226 | Float_t y = 0.;
|
---|
227 | Float_t a = 1.;
|
---|
228 | Float_t b = 0.;
|
---|
229 | Int_t klo = maxpos-1;
|
---|
230 | Int_t khi = maxpos;
|
---|
231 | Float_t klocont = (Float_t)*(first+klo);
|
---|
232 | Float_t khicont = (Float_t)*(first+khi);
|
---|
233 | time = upper;
|
---|
234 | Float_t abmax = khicont;
|
---|
235 |
|
---|
236 | //
|
---|
237 | // Search for the maximum, starting in interval maxpos-1 in steps of 0.2 till maxpos-0.2.
|
---|
238 | // If no maximum is found, go to interval maxpos+1.
|
---|
239 | //
|
---|
240 | Float_t higainklo = fHiGainSecondDeriv[klo];
|
---|
241 | Float_t higainkhi = fHiGainSecondDeriv[khi];
|
---|
242 |
|
---|
243 | while ( x < upper - 0.3 )
|
---|
244 | {
|
---|
245 |
|
---|
246 | x += step;
|
---|
247 | a -= step;
|
---|
248 | b += step;
|
---|
249 |
|
---|
250 | y = a*klocont
|
---|
251 | + b*khicont
|
---|
252 | + (a*a*a-a)*higainklo
|
---|
253 | + (b*b*b-b)*higainkhi;
|
---|
254 |
|
---|
255 | if (y > abmax)
|
---|
256 | {
|
---|
257 | abmax = y;
|
---|
258 | time = x;
|
---|
259 | }
|
---|
260 | }
|
---|
261 |
|
---|
262 | //
|
---|
263 | // Search for the absolute maximum from maxpos to maxpos+1 in steps of 0.2
|
---|
264 | //
|
---|
265 | if (time > upper-0.1)
|
---|
266 | {
|
---|
267 |
|
---|
268 | upper = (Float_t)maxpos+1.;
|
---|
269 | lower = (Float_t)maxpos;
|
---|
270 | x = lower;
|
---|
271 | a = 1.;
|
---|
272 | b = 0.;
|
---|
273 | khi = maxpos+1;
|
---|
274 | klo = maxpos;
|
---|
275 | klocont = (Float_t)*(first+klo);
|
---|
276 | khicont = (Float_t)*(first+khi);
|
---|
277 |
|
---|
278 | higainklo = fHiGainSecondDeriv[klo];
|
---|
279 | higainkhi = fHiGainSecondDeriv[khi];
|
---|
280 |
|
---|
281 | while (x<upper-0.3)
|
---|
282 | {
|
---|
283 |
|
---|
284 | x += step;
|
---|
285 | a -= step;
|
---|
286 | b += step;
|
---|
287 |
|
---|
288 | y = a* klocont
|
---|
289 | + b* khicont
|
---|
290 | + (a*a*a-a)*higainklo
|
---|
291 | + (b*b*b-b)*higainkhi;
|
---|
292 |
|
---|
293 | if (y > abmax)
|
---|
294 | {
|
---|
295 | abmax = y;
|
---|
296 | time = x;
|
---|
297 | }
|
---|
298 | }
|
---|
299 | }
|
---|
300 |
|
---|
301 | //
|
---|
302 | // Now, the time, abmax and khicont and klocont are set correctly within the previous precision.
|
---|
303 | // Try a better precision.
|
---|
304 | //
|
---|
305 | const Float_t up = time+step-0.035;
|
---|
306 | const Float_t lo = time-step+0.035;
|
---|
307 | const Float_t maxpossave = time;
|
---|
308 |
|
---|
309 | x = time;
|
---|
310 | a = upper - x;
|
---|
311 | b = x - lower;
|
---|
312 |
|
---|
313 | step = 0.025; // step size of 83 ps
|
---|
314 |
|
---|
315 | higainklo = fHiGainSecondDeriv[klo];
|
---|
316 | higainkhi = fHiGainSecondDeriv[khi];
|
---|
317 |
|
---|
318 | //
|
---|
319 | // First, try from time up to time+0.2 in steps of 83ps.
|
---|
320 | //
|
---|
321 | while ( x < up )
|
---|
322 | {
|
---|
323 |
|
---|
324 | x += step;
|
---|
325 | a -= step;
|
---|
326 | b += step;
|
---|
327 |
|
---|
328 | y = a* klocont
|
---|
329 | + b* khicont
|
---|
330 | + (a*a*a-a)*higainklo
|
---|
331 | + (b*b*b-b)*higainkhi;
|
---|
332 |
|
---|
333 | if (y > abmax)
|
---|
334 | {
|
---|
335 | abmax = y;
|
---|
336 | time = x;
|
---|
337 | }
|
---|
338 |
|
---|
339 | }
|
---|
340 |
|
---|
341 |
|
---|
342 | //
|
---|
343 | // Second, try from time down to time-0.2 in steps of 0.04.
|
---|
344 | //
|
---|
345 | x = maxpossave;
|
---|
346 | //
|
---|
347 | // Test the possibility that the absolute maximum has not been found between
|
---|
348 | // maxpos and maxpos+0.02, then we have to look between maxpos-0.02 and maxpos
|
---|
349 | // which requires new setting of klocont and khicont
|
---|
350 | //
|
---|
351 | if (x < klo + 0.02)
|
---|
352 | {
|
---|
353 | klo--;
|
---|
354 | khi--;
|
---|
355 | klocont = (Float_t)*(first+klo);
|
---|
356 | khicont = (Float_t)*(first+khi);
|
---|
357 | upper--;
|
---|
358 | lower--;
|
---|
359 | }
|
---|
360 |
|
---|
361 | a = upper - x;
|
---|
362 | b = x - lower;
|
---|
363 |
|
---|
364 | higainklo = fHiGainSecondDeriv[klo];
|
---|
365 | higainkhi = fHiGainSecondDeriv[khi];
|
---|
366 |
|
---|
367 | while ( x > lo )
|
---|
368 | {
|
---|
369 |
|
---|
370 | x -= step;
|
---|
371 | a += step;
|
---|
372 | b -= step;
|
---|
373 |
|
---|
374 | y = a* klocont
|
---|
375 | + b* khicont
|
---|
376 | + (a*a*a-a)*higainklo
|
---|
377 | + (b*b*b-b)*higainkhi;
|
---|
378 |
|
---|
379 | if (y > abmax)
|
---|
380 | {
|
---|
381 | abmax = y;
|
---|
382 | time = x;
|
---|
383 | }
|
---|
384 | }
|
---|
385 |
|
---|
386 | #if 0
|
---|
387 | const Float_t pedes = ped.GetPedestal();
|
---|
388 | const Float_t halfmax = pedes + (abmax - pedes)/2.;
|
---|
389 |
|
---|
390 | //
|
---|
391 | // Now, loop from the maximum bin leftward down in order to find the position of the half maximum.
|
---|
392 | // First, find the right FADC slice:
|
---|
393 | //
|
---|
394 | klo = maxpos;
|
---|
395 | while (klo > maxpos-fStartBeforeMax)
|
---|
396 | {
|
---|
397 | if (*(first+klo) < (Byte_t)halfmax)
|
---|
398 | break;
|
---|
399 | klo--;
|
---|
400 | }
|
---|
401 |
|
---|
402 | //
|
---|
403 | // Loop from the beginning of the slice upwards to reach the halfmax:
|
---|
404 | // With means of bisection:
|
---|
405 | //
|
---|
406 | x = (Float_t)klo;
|
---|
407 | a = 1.;
|
---|
408 | b = 0.;
|
---|
409 | klocont = (Float_t)*(first+klo);
|
---|
410 | khicont = (Float_t)*(first+klo+1);
|
---|
411 |
|
---|
412 | step = 0.5;
|
---|
413 | Bool_t back = kFALSE;
|
---|
414 |
|
---|
415 | while (step > fResolution)
|
---|
416 | {
|
---|
417 |
|
---|
418 | if (back)
|
---|
419 | {
|
---|
420 | x -= step;
|
---|
421 | a += step;
|
---|
422 | b -= step;
|
---|
423 | }
|
---|
424 | else
|
---|
425 | {
|
---|
426 | x += step;
|
---|
427 | a -= step;
|
---|
428 | b += step;
|
---|
429 | }
|
---|
430 |
|
---|
431 | y = a*klocont
|
---|
432 | + b*khicont
|
---|
433 | + (a*a*a-a)*fHiGainSecondDeriv[klo]
|
---|
434 | + (b*b*b-b)*fHiGainSecondDeriv[khi];
|
---|
435 |
|
---|
436 | if (y >= halfmax)
|
---|
437 | back = kTRUE;
|
---|
438 | else
|
---|
439 | back = kFALSE;
|
---|
440 |
|
---|
441 | step /= 2.;
|
---|
442 |
|
---|
443 | }
|
---|
444 | time = (Float_t)fHiGainFirst + x;
|
---|
445 |
|
---|
446 | #endif
|
---|
447 | dtime = 0.035;
|
---|
448 |
|
---|
449 | }
|
---|
450 |
|
---|
451 |
|
---|
452 | // --------------------------------------------------------------------------
|
---|
453 | //
|
---|
454 | // Calculates the arrival time for each pixel
|
---|
455 | //
|
---|
456 | void MExtractTimeFastSpline::FindTimeLoGain(Byte_t *first, Float_t &time, Float_t &dtime,
|
---|
457 | Byte_t &sat, const MPedestalPix &ped) const
|
---|
458 | {
|
---|
459 |
|
---|
460 | const Int_t range = fLoGainLast - fLoGainFirst + 1;
|
---|
461 | const Byte_t *end = first + range;
|
---|
462 | Byte_t *p = first;
|
---|
463 | Byte_t max = 0;
|
---|
464 | Byte_t maxpos = 0;
|
---|
465 |
|
---|
466 | //
|
---|
467 | // Check for saturation in all other slices
|
---|
468 | //
|
---|
469 | while (p<end)
|
---|
470 | {
|
---|
471 | if (*p > max)
|
---|
472 | {
|
---|
473 | max = *p;
|
---|
474 | maxpos = p-first;
|
---|
475 | }
|
---|
476 |
|
---|
477 | if (*p++ >= fSaturationLimit)
|
---|
478 | {
|
---|
479 | sat++;
|
---|
480 | break;
|
---|
481 | }
|
---|
482 | }
|
---|
483 |
|
---|
484 | if (sat)
|
---|
485 | return;
|
---|
486 |
|
---|
487 | if (maxpos < 1)
|
---|
488 | return;
|
---|
489 |
|
---|
490 | Float_t pp;
|
---|
491 |
|
---|
492 | p = first;
|
---|
493 | fLoGainSecondDeriv[0] = 0.;
|
---|
494 | fLoGainFirstDeriv[0] = 0.;
|
---|
495 |
|
---|
496 | for (Int_t i=1;i<range-1;i++)
|
---|
497 | {
|
---|
498 | p++;
|
---|
499 | pp = fLoGainSecondDeriv[i-1] + 4.;
|
---|
500 | fLoGainSecondDeriv[i] = -1.0/pp;
|
---|
501 | fLoGainFirstDeriv [i] = *(p+1) - 2.* *(p) + *(p-1);
|
---|
502 | fLoGainFirstDeriv [i] = (6.0*fLoGainFirstDeriv[i]-fLoGainFirstDeriv[i-1])/pp;
|
---|
503 | }
|
---|
504 |
|
---|
505 | fLoGainSecondDeriv[range-1] = 0.;
|
---|
506 |
|
---|
507 | for (Int_t k=range-2;k>0;k--)
|
---|
508 | fLoGainSecondDeriv[k] = fLoGainSecondDeriv[k]*fLoGainSecondDeriv[k+1] + fLoGainFirstDeriv[k];
|
---|
509 | for (Int_t k=range-2;k>0;k--)
|
---|
510 | fLoGainSecondDeriv[k] /= 6.;
|
---|
511 |
|
---|
512 | //
|
---|
513 | // Now find the maximum
|
---|
514 | //
|
---|
515 | Float_t step = 0.2; // start with step size of 1ns and loop again with the smaller one
|
---|
516 | Float_t lower = (Float_t)maxpos-1.;
|
---|
517 | Float_t upper = (Float_t)maxpos;
|
---|
518 | Float_t x = lower;
|
---|
519 | Float_t y = 0.;
|
---|
520 | Float_t a = 1.;
|
---|
521 | Float_t b = 0.;
|
---|
522 | Int_t klo = maxpos-1;
|
---|
523 | Int_t khi = maxpos;
|
---|
524 | Float_t klocont = (Float_t)*(first+klo);
|
---|
525 | Float_t khicont = (Float_t)*(first+khi);
|
---|
526 | time = upper;
|
---|
527 | Float_t abmax = khicont;
|
---|
528 |
|
---|
529 | //
|
---|
530 | // Search for the maximum, starting in interval maxpos-1. If no maximum is found, go to
|
---|
531 | // interval maxpos+1.
|
---|
532 | //
|
---|
533 | while (x<upper-0.3)
|
---|
534 | {
|
---|
535 |
|
---|
536 | x += step;
|
---|
537 | a -= step;
|
---|
538 | b += step;
|
---|
539 |
|
---|
540 | y = a*klocont
|
---|
541 | + b*khicont
|
---|
542 | + (a*a*a-a)*fLoGainSecondDeriv[klo]
|
---|
543 | + (b*b*b-b)*fLoGainSecondDeriv[khi];
|
---|
544 |
|
---|
545 | if (y > abmax)
|
---|
546 | {
|
---|
547 | abmax = y;
|
---|
548 | time = x;
|
---|
549 | }
|
---|
550 |
|
---|
551 | }
|
---|
552 |
|
---|
553 | if (time > upper-0.1)
|
---|
554 | {
|
---|
555 |
|
---|
556 | upper = (Float_t)maxpos+1.;
|
---|
557 | lower = (Float_t)maxpos;
|
---|
558 | x = lower;
|
---|
559 | a = 1.;
|
---|
560 | b = 0.;
|
---|
561 | khi = maxpos+1;
|
---|
562 | klo = maxpos;
|
---|
563 | klocont = (Float_t)*(first+klo);
|
---|
564 | khicont = (Float_t)*(first+khi);
|
---|
565 |
|
---|
566 | while (x<upper-0.3)
|
---|
567 | {
|
---|
568 |
|
---|
569 | x += step;
|
---|
570 | a -= step;
|
---|
571 | b += step;
|
---|
572 |
|
---|
573 | y = a* klocont
|
---|
574 | + b* khicont
|
---|
575 | + (a*a*a-a)*fLoGainSecondDeriv[klo]
|
---|
576 | + (b*b*b-b)*fLoGainSecondDeriv[khi];
|
---|
577 |
|
---|
578 | if (y > abmax)
|
---|
579 | {
|
---|
580 | abmax = y;
|
---|
581 | time = x;
|
---|
582 | }
|
---|
583 | }
|
---|
584 | }
|
---|
585 |
|
---|
586 | const Float_t up = time+step-0.055;
|
---|
587 | const Float_t lo = time-step+0.055;
|
---|
588 | const Float_t maxpossave = time;
|
---|
589 |
|
---|
590 | x = time;
|
---|
591 | a = upper - x;
|
---|
592 | b = x - lower;
|
---|
593 |
|
---|
594 | step = 0.025; // step size of 165 ps
|
---|
595 |
|
---|
596 | while (x<up)
|
---|
597 | {
|
---|
598 |
|
---|
599 | x += step;
|
---|
600 | a -= step;
|
---|
601 | b += step;
|
---|
602 |
|
---|
603 | y = a* klocont
|
---|
604 | + b* khicont
|
---|
605 | + (a*a*a-a)*fLoGainSecondDeriv[klo]
|
---|
606 | + (b*b*b-b)*fLoGainSecondDeriv[khi];
|
---|
607 |
|
---|
608 | if (y > abmax)
|
---|
609 | {
|
---|
610 | abmax = y;
|
---|
611 | time = x;
|
---|
612 | }
|
---|
613 |
|
---|
614 | }
|
---|
615 |
|
---|
616 | if (time < klo + 0.01)
|
---|
617 | {
|
---|
618 | klo--;
|
---|
619 | khi--;
|
---|
620 | klocont = (Float_t)*(first+klo);
|
---|
621 | khicont = (Float_t)*(first+khi);
|
---|
622 | upper--;
|
---|
623 | lower--;
|
---|
624 | }
|
---|
625 |
|
---|
626 | x = maxpossave;
|
---|
627 | a = upper - x;
|
---|
628 | b = x - lower;
|
---|
629 |
|
---|
630 | while (x>lo)
|
---|
631 | {
|
---|
632 |
|
---|
633 | x -= step;
|
---|
634 | a += step;
|
---|
635 | b -= step;
|
---|
636 |
|
---|
637 | y = a* klocont
|
---|
638 | + b* khicont
|
---|
639 | + (a*a*a-a)*fLoGainSecondDeriv[klo]
|
---|
640 | + (b*b*b-b)*fLoGainSecondDeriv[khi];
|
---|
641 |
|
---|
642 | if (y > abmax)
|
---|
643 | {
|
---|
644 | abmax = y;
|
---|
645 | time = x;
|
---|
646 | }
|
---|
647 |
|
---|
648 | }
|
---|
649 |
|
---|
650 | const Float_t pedes = ped.GetPedestal();
|
---|
651 | const Float_t halfmax = pedes + (abmax - pedes)/2.;
|
---|
652 |
|
---|
653 | //
|
---|
654 | // Now, loop from the maximum bin leftward down in order to find the position of the half maximum.
|
---|
655 | // First, find the right FADC slice:
|
---|
656 | //
|
---|
657 | klo = maxpos;
|
---|
658 | while (klo > maxpos-4)
|
---|
659 | {
|
---|
660 | if (*(first+klo) < (Byte_t)halfmax)
|
---|
661 | break;
|
---|
662 | klo--;
|
---|
663 | }
|
---|
664 |
|
---|
665 | //
|
---|
666 | // Loop from the beginning of the slice upwards to reach the halfmax:
|
---|
667 | // With means of bisection:
|
---|
668 | //
|
---|
669 | x = (Float_t)klo;
|
---|
670 | a = 1.;
|
---|
671 | b = 0.;
|
---|
672 | klocont = (Float_t)*(first+klo);
|
---|
673 | khicont = (Float_t)*(first+klo+1);
|
---|
674 | time = x;
|
---|
675 |
|
---|
676 | step = 0.5;
|
---|
677 | Bool_t back = kFALSE;
|
---|
678 |
|
---|
679 | while (step > fResolution)
|
---|
680 | {
|
---|
681 |
|
---|
682 | if (back)
|
---|
683 | {
|
---|
684 | x -= step;
|
---|
685 | a += step;
|
---|
686 | b -= step;
|
---|
687 | }
|
---|
688 | else
|
---|
689 | {
|
---|
690 | x += step;
|
---|
691 | a -= step;
|
---|
692 | b += step;
|
---|
693 | }
|
---|
694 |
|
---|
695 | y = a*klocont
|
---|
696 | + b*khicont
|
---|
697 | + (a*a*a-a)*fLoGainSecondDeriv[klo]
|
---|
698 | + (b*b*b-b)*fLoGainSecondDeriv[khi];
|
---|
699 |
|
---|
700 | if (y >= halfmax)
|
---|
701 | back = kTRUE;
|
---|
702 | else
|
---|
703 | back = kFALSE;
|
---|
704 |
|
---|
705 | step /= 2.;
|
---|
706 |
|
---|
707 | }
|
---|
708 |
|
---|
709 | time = (Float_t)fLoGainFirst + x;
|
---|
710 | dtime = fResolution;
|
---|
711 | }
|
---|
712 |
|
---|
713 | // --------------------------------------------------------------------------
|
---|
714 | //
|
---|
715 | // In addition to the resources of the base-class MExtractor:
|
---|
716 | // MJPedestal.MExtractor.Resolution: 0.003
|
---|
717 | // MJPedestal.MExtractor.RiseTime: 1.5
|
---|
718 | //
|
---|
719 | Int_t MExtractTimeFastSpline::ReadEnv(const TEnv &env, TString prefix, Bool_t print)
|
---|
720 | {
|
---|
721 | Bool_t rc = kFALSE;
|
---|
722 |
|
---|
723 | if (IsEnvDefined(env, prefix, "HiGainWindowSize", print))
|
---|
724 | {
|
---|
725 | SetResolution(GetEnvValue(env, prefix, "Resolution", fResolution));
|
---|
726 | rc = kTRUE;
|
---|
727 | }
|
---|
728 | if (IsEnvDefined(env, prefix, "LoGainWindowSize", print))
|
---|
729 | {
|
---|
730 | SetRiseTime(GetEnvValue(env, prefix, "RiseTime", fRiseTime));
|
---|
731 | rc = kTRUE;
|
---|
732 | }
|
---|
733 |
|
---|
734 | return MExtractTime::ReadEnv(env, prefix, print) ? kTRUE : rc;
|
---|
735 | }
|
---|
736 |
|
---|
737 | void MExtractTimeFastSpline::Print(Option_t *o) const
|
---|
738 | {
|
---|
739 | *fLog << all;
|
---|
740 | *fLog << GetDescriptor() << ":" << endl;
|
---|
741 | *fLog << " Resolution: " << fResolution << endl;
|
---|
742 | *fLog << " RiseTime: " << fRiseTime << endl;
|
---|
743 | MExtractTime::Print(o);
|
---|
744 | }
|
---|