]>
Commit | Line | Data |
---|---|---|
4c039060 | 1 | /************************************************************************** |
2 | * Copyright(c) 1998-1999, ALICE Experiment at CERN, All rights reserved. * | |
3 | * * | |
4 | * Author: The ALICE Off-line Project. * | |
5 | * Contributors are mentioned in the code where appropriate. * | |
6 | * * | |
7 | * Permission to use, copy, modify and distribute this software and its * | |
8 | * documentation strictly for non-commercial purposes is hereby granted * | |
9 | * without fee, provided that the above copyright notice appears in all * | |
10 | * copies and that both the copyright notice and this permission notice * | |
11 | * appear in the supporting documentation. The authors make no claims * | |
12 | * about the suitability of this software for any purpose. It is * | |
13 | * provided "as is" without express or implied warranty. * | |
14 | **************************************************************************/ | |
15 | ||
16 | /* | |
17 | $Log$ | |
cc80f89e | 18 | Revision 1.4.8.2 2000/04/10 08:53:09 kowal2 |
19 | ||
20 | Updates by M. Ivanov | |
21 | ||
22 | ||
23 | Revision 1.4 1999/09/29 09:24:34 fca | |
24 | Introduction of the Copyright and cvs Log | |
25 | ||
4c039060 | 26 | */ |
27 | ||
8c555625 | 28 | //----------------------------------------------------------------------------- |
29 | // $Header$ | |
30 | // | |
31 | // | |
32 | // Origin: Marian Ivanov, Uni. of Bratislava, ivanov@fmph.uniba.sk | |
33 | // | |
34 | // Declaration of class AliTPCRF1D | |
35 | // | |
36 | //----------------------------------------------------------------------------- | |
cc80f89e | 37 | |
8c555625 | 38 | #include "TMath.h" |
39 | #include "AliTPCRF1D.h" | |
40 | #include "TF2.h" | |
41 | #include <iostream.h> | |
42 | #include "TCanvas.h" | |
43 | #include "TPad.h" | |
44 | #include "TStyle.h" | |
45 | #include "TH1.h" | |
46 | ||
47 | extern TStyle * gStyle; | |
48 | ||
49 | static Double_t funGauss(Double_t *x, Double_t * par) | |
50 | { | |
cc80f89e | 51 | //Gauss function -needde by the generic function object |
8c555625 | 52 | return TMath::Exp(-(x[0]*x[0])/(2*par[0]*par[0])); |
53 | } | |
54 | ||
55 | static Double_t funCosh(Double_t *x, Double_t * par) | |
56 | { | |
cc80f89e | 57 | //Cosh function -needde by the generic function object |
8c555625 | 58 | return 1/TMath::CosH(3.14159*x[0]/(2*par[0])); |
59 | } | |
60 | ||
61 | static Double_t funGati(Double_t *x, Double_t * par) | |
62 | { | |
cc80f89e | 63 | //Gati function -needde by the generic function object |
8c555625 | 64 | Float_t K3=par[1]; |
65 | Float_t K3R=TMath::Sqrt(K3); | |
66 | Float_t K2=(TMath::Pi()/2)*(1-K3R/2.); | |
67 | Float_t K1=K2*K3R/(4*TMath::ATan(K3R)); | |
68 | Float_t l=x[0]/par[0]; | |
69 | Float_t tan2=TMath::TanH(K2*l); | |
70 | tan2*=tan2; | |
71 | Float_t res = K1*(1-tan2)/(1+K3*tan2); | |
72 | return res; | |
73 | } | |
74 | ||
8c555625 | 75 | /////////////////////////////////////////////////////////////////////////// |
76 | /////////////////////////////////////////////////////////////////////////// | |
77 | ||
8c555625 | 78 | ClassImp(AliTPCRF1D) |
79 | ||
80 | ||
81 | AliTPCRF1D::AliTPCRF1D(Bool_t direct,Int_t np,Float_t step) | |
82 | { | |
cc80f89e | 83 | //default constructor for response function object |
8c555625 | 84 | fDirect=direct; |
85 | fNRF = np; | |
86 | fcharge = new Float_t[fNRF]; | |
87 | fDSTEPM1=1./step; | |
88 | fSigma = 0; | |
8c555625 | 89 | fGRF = 0; |
90 | fkNorm = 0.5; | |
91 | fpadWidth = 3.5; | |
92 | forigsigma=0.; | |
93 | fOffset = 0.; | |
94 | } | |
95 | ||
96 | ||
97 | AliTPCRF1D::~AliTPCRF1D() | |
98 | { | |
cc80f89e | 99 | if (fcharge!=0) delete [] fcharge; |
8c555625 | 100 | if (fGRF !=0 ) fGRF->Delete(); |
101 | } | |
102 | ||
103 | Float_t AliTPCRF1D::GetRF(Float_t xin) | |
104 | { | |
cc80f89e | 105 | //function which return response |
106 | //for the charge in distance xin | |
8c555625 | 107 | //return linear aproximation of RF |
108 | Float_t x = TMath::Abs((xin-fOffset)*fDSTEPM1)+fNRF/2; | |
109 | Int_t i1=Int_t(x); | |
110 | if (x<0) i1-=1; | |
111 | Float_t res=0; | |
112 | if (i1+1<fNRF) | |
113 | res = fcharge[i1]*(Float_t(i1+1)-x)+fcharge[i1+1]*(x-Float_t(i1)); | |
114 | return res; | |
115 | } | |
116 | ||
117 | Float_t AliTPCRF1D::GetGRF(Float_t xin) | |
118 | { | |
cc80f89e | 119 | //function which returnoriginal charge distribution |
120 | //this function is just normalised for fKnorm | |
8c555625 | 121 | if (fGRF != 0 ) |
122 | return fkNorm*fGRF->Eval(xin)/fInteg; | |
123 | else | |
124 | return 0.; | |
125 | } | |
126 | ||
127 | ||
128 | void AliTPCRF1D::SetParam( TF1 * GRF,Float_t padwidth, | |
129 | Float_t kNorm, Float_t sigma) | |
130 | { | |
cc80f89e | 131 | //adjust parameters of the original charge distribution |
132 | //and pad size parameters | |
8c555625 | 133 | fpadWidth = padwidth; |
134 | fGRF = GRF; | |
135 | fkNorm = kNorm; | |
136 | if (sigma==0) sigma= fpadWidth/TMath::Sqrt(12.); | |
137 | forigsigma=sigma; | |
138 | fDSTEPM1 = 10/TMath::Sqrt(sigma*sigma+fpadWidth*fpadWidth/12); | |
139 | sprintf(fType,"User"); | |
140 | // Update(); | |
141 | } | |
142 | ||
143 | ||
144 | void AliTPCRF1D::SetGauss(Float_t sigma, Float_t padWidth, | |
145 | Float_t kNorm) | |
146 | { | |
cc80f89e | 147 | // |
148 | // set parameters for Gauss generic charge distribution | |
149 | // | |
8c555625 | 150 | fpadWidth = padWidth; |
151 | fkNorm = kNorm; | |
152 | if (fGRF !=0 ) fGRF->Delete(); | |
153 | fGRF = new TF1("fun",funGauss,-5,5,2); | |
154 | funParam[0]=sigma; | |
155 | forigsigma=sigma; | |
156 | fGRF->SetParameters(funParam); | |
157 | fDSTEPM1 = 10./TMath::Sqrt(sigma*sigma+fpadWidth*fpadWidth/12); | |
cc80f89e | 158 | //by default I set the step as one tenth of sigma |
8c555625 | 159 | sprintf(fType,"Gauss"); |
160 | } | |
161 | ||
162 | void AliTPCRF1D::SetCosh(Float_t sigma, Float_t padWidth, | |
163 | Float_t kNorm) | |
164 | { | |
cc80f89e | 165 | // |
166 | // set parameters for Cosh generic charge distribution | |
167 | // | |
8c555625 | 168 | fpadWidth = padWidth; |
169 | fkNorm = kNorm; | |
170 | if (fGRF !=0 ) fGRF->Delete(); | |
171 | fGRF = new TF1("fun", funCosh, -5.,5.,2); | |
172 | funParam[0]=sigma; | |
173 | fGRF->SetParameters(funParam); | |
174 | forigsigma=sigma; | |
175 | fDSTEPM1 = 10./TMath::Sqrt(sigma*sigma+fpadWidth*fpadWidth/12); | |
176 | //by default I set the step as one tenth of sigma | |
8c555625 | 177 | sprintf(fType,"Cosh"); |
178 | } | |
179 | ||
180 | void AliTPCRF1D::SetGati(Float_t K3, Float_t padDistance, Float_t padWidth, | |
181 | Float_t kNorm) | |
182 | { | |
cc80f89e | 183 | // |
184 | // set parameters for Gati generic charge distribution | |
185 | // | |
8c555625 | 186 | fpadWidth = padWidth; |
187 | fkNorm = kNorm; | |
188 | if (fGRF !=0 ) fGRF->Delete(); | |
189 | fGRF = new TF1("fun", funGati, -5.,5.,2); | |
190 | funParam[0]=padDistance; | |
191 | funParam[1]=K3; | |
192 | fGRF->SetParameters(funParam); | |
193 | forigsigma=padDistance; | |
194 | fDSTEPM1 = 10./TMath::Sqrt(padDistance*padDistance+fpadWidth*fpadWidth/12); | |
195 | //by default I set the step as one tenth of sigma | |
8c555625 | 196 | sprintf(fType,"Gati"); |
197 | } | |
198 | ||
199 | void AliTPCRF1D::Draw(Float_t x1,Float_t x2,Int_t N) | |
200 | { | |
cc80f89e | 201 | // |
202 | //Draw prf in selected region <x1,x2> with nuber of diviision = n | |
203 | // | |
8c555625 | 204 | char s[100]; |
205 | TCanvas * c1 = new TCanvas("canRF","Pad response function",700,900); | |
206 | c1->cd(); | |
207 | TPad * pad1 = new TPad("pad1RF","",0.05,0.55,0.95,0.95,21); | |
208 | pad1->Draw(); | |
209 | TPad * pad2 = new TPad("pad2RF","",0.05,0.05,0.95,0.45,21); | |
210 | pad2->Draw(); | |
211 | ||
212 | sprintf(s,"RF response function for %1.2f cm pad width", | |
213 | fpadWidth); | |
214 | pad1->cd(); | |
215 | TH1F * hRFo = new TH1F("hRFo","Original charge distribution",N+1,x1,x2); | |
216 | pad2->cd(); | |
217 | gStyle->SetOptFit(1); | |
218 | gStyle->SetOptStat(0); | |
219 | TH1F * hRFc = new TH1F("hRFc",s,N+1,x1,x2); | |
220 | Float_t x=x1; | |
221 | Float_t y1; | |
222 | Float_t y2; | |
223 | ||
224 | for (Float_t i = 0;i<N+1;i++) | |
225 | { | |
226 | x+=(x2-x1)/Float_t(N); | |
227 | y1 = GetRF(x); | |
228 | hRFc->Fill(x,y1); | |
229 | y2 = GetGRF(x); | |
230 | hRFo->Fill(x,y2); | |
231 | }; | |
232 | pad1->cd(); | |
233 | hRFo->Fit("gaus"); | |
234 | pad2->cd(); | |
235 | hRFc->Fit("gaus"); | |
236 | } | |
237 | ||
238 | void AliTPCRF1D::Update() | |
239 | { | |
cc80f89e | 240 | // |
241 | //update fields with interpolated values for | |
242 | //PRF calculation | |
243 | ||
244 | //at the begining initialize to 0 | |
8c555625 | 245 | for (Int_t i =0; i<fNRF;i++) fcharge[i] = 0; |
246 | if ( fGRF == 0 ) return; | |
247 | fInteg = fGRF->Integral(-5*forigsigma,5*forigsigma,funParam,0.00001); | |
248 | if ( fInteg == 0 ) fInteg = 1; | |
249 | if (fDirect==kFALSE){ | |
250 | //integrate charge over pad for different distance of pad | |
251 | for (Int_t i =0; i<fNRF;i++) | |
252 | { //x in cm fpadWidth in cm | |
253 | Float_t x = (Float_t)(i-fNRF/2)/fDSTEPM1; | |
254 | Float_t x1=TMath::Max(x-fpadWidth/2,-5*forigsigma); | |
255 | Float_t x2=TMath::Min(x+fpadWidth/2,5*forigsigma); | |
256 | fcharge[i] = | |
257 | fkNorm*fGRF->Integral(x1,x2,funParam,0.0001)/fInteg; | |
258 | }; | |
259 | } | |
260 | else{ | |
261 | for (Int_t i =0; i<fNRF;i++) | |
262 | { //x in cm fpadWidth in cm | |
263 | Float_t x = (Float_t)(i-fNRF/2)/fDSTEPM1; | |
264 | fcharge[i] = fkNorm*fGRF->Eval(x); | |
265 | }; | |
266 | } | |
267 | fSigma = 0; | |
268 | Float_t sum =0; | |
269 | Float_t mean=0; | |
270 | for (Float_t x =-fNRF/fDSTEPM1; x<fNRF/fDSTEPM1;x+=1/fDSTEPM1) | |
271 | { //x in cm fpadWidth in cm | |
272 | Float_t weight = GetRF(x+fOffset); | |
273 | fSigma+=x*x*weight; | |
274 | mean+=x*weight; | |
275 | sum+=weight; | |
276 | }; | |
277 | if (sum>0){ | |
278 | mean/=sum; | |
279 | fSigma = TMath::Sqrt(fSigma/sum-mean*mean); | |
280 | } | |
281 | else fSigma=0; | |
282 | } | |
283 | ||
284 | void AliTPCRF1D::Streamer(TBuffer &R__b) | |
285 | { | |
cc80f89e | 286 | // Stream an object of class AliTPCRF1D. |
8c555625 | 287 | |
288 | if (R__b.IsReading()) { | |
289 | Version_t R__v = R__b.ReadVersion(); if (R__v) { } | |
290 | TObject::Streamer(R__b); | |
291 | //read pad parameters | |
292 | R__b >> fpadWidth; | |
293 | //read charge parameters | |
294 | R__b >> fType[0]; | |
295 | R__b >> fType[1]; | |
296 | R__b >> fType[2]; | |
297 | R__b >> fType[3]; | |
298 | R__b >> fType[4]; | |
299 | R__b >> forigsigma; | |
300 | R__b >> fkNorm; | |
301 | R__b >> fK3X; | |
302 | R__b >> fPadDistance; | |
303 | R__b >> fInteg; | |
304 | R__b >> fOffset; | |
305 | //read functions | |
306 | if (fGRF!=0) { | |
cc80f89e | 307 | delete [] fGRF; |
8c555625 | 308 | fGRF=0; |
309 | } | |
310 | if (strncmp(fType,"User",3)==0){ | |
311 | fGRF= new TF1; | |
312 | R__b>>fGRF; | |
313 | } | |
314 | ||
315 | if (strncmp(fType,"Gauss",3)==0) | |
316 | fGRF = new TF1("fun",funGauss,-5.,5.,4); | |
317 | if (strncmp(fType,"Cosh",3)==0) | |
318 | fGRF = new TF1("fun",funCosh,-5.,5.,4); | |
319 | if (strncmp(fType,"Gati",3)==0) | |
320 | fGRF = new TF1("fun",funGati,-5.,5.,4); | |
321 | R__b >>fDSTEPM1; | |
322 | R__b >>fNRF; | |
323 | R__b.ReadFastArray(fcharge,fNRF); | |
324 | R__b.ReadFastArray(funParam,5); | |
325 | if (fGRF!=0) fGRF->SetParameters(funParam); | |
326 | ||
327 | } else { | |
328 | R__b.WriteVersion(AliTPCRF1D::IsA()); | |
329 | TObject::Streamer(R__b); | |
330 | //write pad width | |
331 | R__b << fpadWidth; | |
332 | //write charge parameters | |
333 | R__b << fType[0]; | |
334 | R__b << fType[1]; | |
335 | R__b << fType[2]; | |
336 | R__b << fType[3]; | |
337 | R__b << fType[4]; | |
338 | R__b << forigsigma; | |
339 | R__b << fkNorm; | |
340 | R__b << fK3X; | |
341 | R__b << fPadDistance; | |
342 | R__b << fInteg; | |
343 | R__b << fOffset; | |
344 | //write interpolation parameters | |
345 | if (strncmp(fType,"User",3)==0) R__b <<fGRF; | |
346 | R__b <<fDSTEPM1; | |
347 | R__b <<fNRF; | |
348 | R__b.WriteFastArray(fcharge,fNRF); | |
cc80f89e | 349 | R__b.WriteFastArray(funParam,5); |
8c555625 | 350 | } |
351 | } | |
352 |