1 |
|
// ------ PAMELA Digitizer ------ |
2 |
|
// |
3 |
|
// Date, release and how-to: see file Pamelagp2Digits.cxx |
4 |
|
// |
5 |
|
// NB: Check length physics packet [packet type (0x10 = physics data)] |
6 |
// |
// |
7 |
#include <sstream> |
#include <sstream> |
8 |
#include <fstream> |
#include <fstream> |
9 |
#include <stdlib.h> |
#include <stdlib.h> |
10 |
|
#include <stdio.h> |
11 |
#include <string.h> |
#include <string.h> |
12 |
#include <ctype.h> |
#include <ctype.h> |
13 |
|
#include <time.h> |
14 |
#include "Riostream.h" |
#include "Riostream.h" |
15 |
#include "TFile.h" |
#include "TFile.h" |
16 |
#include "TDirectory.h" |
#include "TDirectory.h" |
260 |
DigitizeAC(); |
DigitizeAC(); |
261 |
DigitizeCALO(); |
DigitizeCALO(); |
262 |
DigitizeTrack(); |
DigitizeTrack(); |
263 |
//DigitizeS4(); |
DigitizeS4(); |
264 |
DigitizeND(); |
DigitizeND(); |
265 |
// |
// |
|
// Create CPU header, we need packet type (0x10 = physics data) and packet length. |
|
|
// |
|
|
UInt_t length = (fCALOlength + fACbuffer + fTracklength)*2; |
|
|
DigitizePSCU(length,0x10); |
|
|
// |
|
266 |
// Add padding to 64 bits |
// Add padding to 64 bits |
267 |
// |
// |
268 |
AddPadding(); |
AddPadding(); |
269 |
|
// |
270 |
|
// Create CPU header, we need packet type (0x10 = physics data) and packet length. |
271 |
// |
// |
272 |
|
UInt_t length=2*(fCALOlength+fACbuffer+fTracklength+fNDbuffer+fS4buffer)+fPadding+fTOFbuffer+fTRIGGERbuffer; |
273 |
|
//UInt_t length=2*(fCALOlength+fACbuffer+fTracklength+fNDbuffer)+fPadding+fTOFbuffer+fTRIGGERbuffer; |
274 |
|
DigitizePSCU(length,0x10); |
275 |
if ( !i%100 ) std::cout << "writing event " << i << endl; |
if ( !i%100 ) std::cout << "writing event " << i << endl; |
276 |
WriteData(); |
WriteData(); |
277 |
}; |
}; |
1037 |
Int_t Digitizer::DigitizeTOF() { |
Int_t Digitizer::DigitizeTOF() { |
1038 |
//fDataTof: 12 x 23 bytes (=276 bytes) |
//fDataTof: 12 x 23 bytes (=276 bytes) |
1039 |
UChar_t *pTof=fDataTof; |
UChar_t *pTof=fDataTof; |
1040 |
|
Bool_t DEBUG=false; |
1041 |
|
|
1042 |
// --- activate branches: |
// --- activate branches: |
1043 |
fhBookTree->SetBranchStatus("Nthtof",1); |
fhBookTree->SetBranchStatus("Nthtof",1); |
1054 |
// ------ evaluate energy in each pmt: ------ |
// ------ evaluate energy in each pmt: ------ |
1055 |
// strip geometry (lenght/width) |
// strip geometry (lenght/width) |
1056 |
Float_t dimel[6] = {33.0, 40.8 ,18.0, 15.0, 15.0, 18.0}; |
Float_t dimel[6] = {33.0, 40.8 ,18.0, 15.0, 15.0, 18.0}; |
1057 |
Float_t dimes[6] = {5.1, 5.5, 7.5, 9.0, 6.0, 5.0}; |
//Float_t dimes[6] = {5.1, 5.5, 7.5, 9.0, 6.0, 5.0}; |
1058 |
|
|
1059 |
// S11 8 paddles 33.0 x 5.1 cm |
// S11 8 paddles 33.0 x 5.1 cm |
1060 |
// S12 6 paddles 40.8 x 5.5 cm |
// S12 6 paddles 40.8 x 5.5 cm |
1097 |
}; |
}; |
1098 |
Float_t atte1[48],atte2[48],lambda1[48],lambda2[48]; |
Float_t atte1[48],atte2[48],lambda1[48],lambda2[48]; |
1099 |
Int_t temp=0; |
Int_t temp=0; |
1100 |
|
// correct readout WM Oct '07 |
1101 |
for(Int_t i=0; i<48; i++){ |
for(Int_t i=0; i<48; i++){ |
1102 |
fileTriggerCalib >> temp; |
fileTriggerCalib >> temp; |
1103 |
fileTriggerCalib >> atte1[i]; |
fileTriggerCalib >> atte1[i]; |
|
fileTriggerCalib >> atte2[i]; |
|
1104 |
fileTriggerCalib >> lambda1[i]; |
fileTriggerCalib >> lambda1[i]; |
1105 |
|
fileTriggerCalib >> atte2[i]; |
1106 |
fileTriggerCalib >> lambda2[i]; |
fileTriggerCalib >> lambda2[i]; |
1107 |
fileTriggerCalib >> temp; |
fileTriggerCalib >> temp; |
1108 |
} |
} |
1115 |
// fine lettura dal file */ |
// fine lettura dal file */ |
1116 |
|
|
1117 |
//const Int_t nmax=??; = Nthtof |
//const Int_t nmax=??; = Nthtof |
1118 |
Int_t nh, ip, ipad, ipmt; |
Int_t ip, ipad; |
1119 |
|
//Int_t ipmt; |
1120 |
Int_t pmtleft=0, pmtright=0; |
Int_t pmtleft=0, pmtright=0; |
1121 |
Int_t *pl, *pr; |
Int_t *pl, *pr; |
1122 |
pl = &pmtleft; |
pl = &pmtleft; |
1123 |
pr = &pmtright; |
pr = &pmtright; |
1124 |
|
|
1125 |
/* ********************************** inizio loop sugli hit */ |
// TDC variables: |
1126 |
|
Int_t TDClast=4095; // no signal --> ADC ch=4095 |
1127 |
|
Int_t TDCint[48]; |
1128 |
|
Float_t tdc[48],tdc1[48],tdcpmt[48]; |
1129 |
|
for(Int_t i=0; i<48; i++) |
1130 |
|
tdcpmt[i] = 1000.; |
1131 |
|
Float_t thresh=10.; // to be defined better... (Wolfgang) |
1132 |
|
|
1133 |
|
// === TDC: simulate timing for each paddle |
1134 |
|
Float_t dt1 = 285.e-12 ; // single PMT resolution |
1135 |
|
Float_t tdcres[50],c1_S[50],c2_S[50],c3_S[50]; |
1136 |
|
for(Int_t j=0;j<48;j++) tdcres[j] = 50.E-12; // TDC resolution 50 picosec |
1137 |
|
for(Int_t j=0;j<48;j++) c1_S[j] = 500.; // cable length in channels |
1138 |
|
for(Int_t j=0;j<48;j++) c2_S[j] = 0.; |
1139 |
|
for(Int_t j=0;j<48;j++) c3_S[j] = 1000.; |
1140 |
|
for(Int_t j=0;j<48;j++) c1_S[j] = c1_S[j]*tdcres[j]; // cable length in sec |
1141 |
|
for(Int_t j=0;j<48;j++) c2_S[j] = c2_S[j]*tdcres[j]; |
1142 |
|
// ih = 0 + i1; // not used?? (Silvio) |
1143 |
|
|
1144 |
|
/* ********************************** start loop over hits */ |
1145 |
|
|
1146 |
for(Int_t nh=0; nh<Nthtof; nh++){ |
for(Int_t nh=0; nh<Nthtof; nh++){ |
1147 |
|
|
1151 |
FGeo[j]=0.; |
FGeo[j]=0.; |
1152 |
} |
} |
1153 |
|
|
1154 |
// ridefiniz. piano e pad per i vettori in C |
Float_t s_l_g[6] = {8.0, 8.0, 20.9, 22.0, 9.8, 8.3 }; // length of the lightguide |
1155 |
|
Float_t t1,t2,veff,veff1,veff0 ; |
1156 |
|
veff0 = 100.*1.0e8 ; // light velocity in the scintillator in m/sec |
1157 |
|
veff1 = 100.*1.5e8; // light velocity in the lightguide in m/sec |
1158 |
|
veff=veff0; // signal velocity in the paddle |
1159 |
|
|
1160 |
|
t1 = Timetof[nh] ; // Start |
1161 |
|
t2 = Timetof[nh] ; |
1162 |
|
|
1163 |
|
// Donatella: redefinition plane and pad for vectors in C |
1164 |
ip = Ipltof[nh]-1; |
ip = Ipltof[nh]-1; |
1165 |
ipad = Ipaddle[nh]-1; |
ipad = Ipaddle[nh]-1; |
1166 |
pmtleft=0; |
pmtleft=0; |
1167 |
pmtright=0; |
pmtright=0; |
1168 |
|
|
1169 |
//Paddle2Pmt((Int_t)ip, (Int_t) ipad, (Int_t*) &pmtleft, (Int_t*) &pmtright); |
if (ip<6) { |
1170 |
Paddle2Pmt(ip, ipad, &pmtleft, &pmtright); |
Paddle2Pmt(ip, ipad, &pmtleft, &pmtright); |
|
//Paddle2Pmt(ip, ipad, pl, pr); |
|
1171 |
|
|
1172 |
// per avere anche la corrispondenza pmt --> half board e canale |
// per avere anche la corrispondenza pmt --> half board e canale |
1173 |
// metodo GetPMTIndex(Int_t ipmt, Int_t &hb, Int_t &ch) // non lo usiamo x ora |
// metodo GetPMTIndex(Int_t ipmt, Int_t &hb, Int_t &ch) // non lo usiamo x ora |
1174 |
|
|
1175 |
/*calcola la pos media e il path all'interno della paddle */ |
// evaluates mean position and path inside the paddle |
1176 |
|
|
1177 |
Float_t tpos=0.; |
Float_t tpos=0.; |
1178 |
Float_t path[2] = {0., 0.}; |
Float_t path[2] = {0., 0.}; |
1179 |
//--- Strip in Y = S11,S22,S31 ------ |
//--- Strip in Y = S11,S22,S31 ------ |
1180 |
if(ip==0 || ip==3 || ip==4) |
if(ip==0 || ip==3 || ip==4) |
1181 |
tpos = (Yintof[nh]+Youttof[nh])/2.; |
tpos = (Yintof[nh]+Youttof[nh])/2.; |
1182 |
else |
else |
1183 |
if(ip==1 || ip==2 || ip==5) //--- Strip in X per S12,S21,S32 |
if(ip==1 || ip==2 || ip==5) //--- Strip in X per S12,S21,S32 |
1184 |
tpos = (Xintof[nh]+Xouttof[nh])/2.; |
tpos = (Xintof[nh]+Xouttof[nh])/2.; |
1185 |
else if (ip!=6) |
else //if (ip!=6) |
1186 |
printf("*** Warning: this option should never occur! (ip=%2i, nh=%2i)\n",ip,nh); |
printf("*** WARNING TOF: this option should never occur! (ip=%2i, nh=%2i)\n",ip,nh); |
1187 |
path[0]= tpos + dimel[ip]/2.; |
path[0]= tpos + dimel[ip]/2.; |
1188 |
path[1]= dimel[ip]/2.- tpos; |
path[1]= dimel[ip]/2.- tpos; |
1189 |
|
|
1190 |
// cout <<"Strip N. ="<< ipaddle <<" piano n.= "<< iplane <<" POSIZ = "<< tpos <<"\n"; |
// cout <<"Strip N. ="<< ipaddle <<" piano n.= "<< iplane <<" POSIZ = "<< tpos <<"\n"; |
1191 |
|
|
1192 |
/* per il momento metto un fattore geometrico costante*/ |
if (DEBUG) { |
1193 |
FGeo[0] =0.5; |
cout <<" plane "<<ip<<" strip # ="<< ipad <<" tpos "<< tpos <<"\n"; |
1194 |
FGeo[1] =0.5; |
cout <<"pmtleft, pmtright "<<pmtleft<<" "<<pmtright<<endl; |
1195 |
// FGeo[1] = atan(path[1]/dimes[ip])/6.28318; // frazione fotoni verso SX |
} |
1196 |
// FGeo[2] = atan(path[2]/dimes[ip])/6.28318; // e verso DX |
|
1197 |
|
// constant geometric factor, for the moment |
1198 |
|
FGeo[0] =0.5; |
1199 |
|
FGeo[1] =0.5; |
1200 |
|
// FGeo[1] = atan(path[1]/dimes[ip])/6.28318; // frazione fotoni verso SX |
1201 |
|
// FGeo[2] = atan(path[2]/dimes[ip])/6.28318; // e verso DX |
1202 |
|
|
1203 |
/* rimando la fluttuazione poissoniana sui fotoni prodotti |
/* rimando la fluttuazione poissoniana sui fotoni prodotti |
1204 |
sto studiando come funziona la funzione: |
sto studiando come funziona la funzione: |
1205 |
long int i = sto.Poisson(double x); */ |
long int i = sto.Poisson(double x); */ |
1206 |
// Npho = Poisson(ERELTOF[nh])*Pho_keV*1e6 Eloss in GeV ? |
// Npho = Poisson(ERELTOF[nh])*Pho_keV*1e6 Eloss in GeV ? |
1207 |
Npho = Ereltof[nh]*Pho_keV*10.0e6; // Eloss in GeV ? |
Npho = Ereltof[nh]*Pho_keV*1.0e6; // Eloss in GeV ? |
1208 |
|
|
1209 |
Float_t knorm[2]={0., 0.}; // Donatella |
Float_t knorm[2]={0., 0.}; // Donatella |
1210 |
Float_t Atten[2]={0., 0.}; // Donatella |
Float_t Atten[2]={0., 0.}; // Donatella |
1211 |
for(Int_t j=0; j<2; j++){ |
for(Int_t j=0; j<2; j++){ |
1212 |
QhitPad_pC[j]= Npho*FGeo[j]*effi*pmGain*echarge; |
QhitPad_pC[j]= Npho*FGeo[j]*effi*pmGain*echarge*1.E12; // corrected WM |
1213 |
knorm[j]=QhitPad_pC[j]/(atte1[pmtleft+j]*exp((dimel[ip]/2.*pow(-1,j+1))/lambda1[pmtleft+j]) + |
/* knorm[j]=QhitPad_pC[j]/(atte1[pmtleft+j]*exp((dimel[ip]/2.*pow(-1,j+1))/lambda1[pmtleft+j]) + |
1214 |
atte2[pmtleft+j]*exp((dimel[ip]/2.*pow(-1,j+1))/lambda2[pmtleft+j])); |
atte2[pmtleft+j]*exp((dimel[ip]/2.*pow(-1,j+1))/lambda2[pmtleft+j])); |
1215 |
|
Atten[j]=knorm[j]*(atte1[pmtleft+j]*exp(tpos/lambda1[pmtleft+j]) + |
1216 |
|
atte2[pmtleft+j]*exp(tpos/lambda2[pmtleft+j])); |
1217 |
|
QhitPmt_pC[j]= QhitPad_pC[j]*Atten[j]; |
1218 |
|
*/ |
1219 |
|
// WM |
1220 |
|
knorm[j]=atte1[pmtleft+j]*exp(lambda1[pmtleft+j]*dimel[ip]/2.*pow(-1,j+1)) + |
1221 |
|
atte2[pmtleft+j]*exp(lambda2[pmtleft+j]*dimel[ip]/2.*pow(-1,j+1)); |
1222 |
|
Atten[j]=atte1[pmtleft+j]*exp(tpos*lambda1[pmtleft+j]) + |
1223 |
|
atte2[pmtleft+j]*exp(tpos*lambda2[pmtleft+j]) ; |
1224 |
|
QhitPmt_pC[j]= QhitPad_pC[j]*Atten[j]/knorm[j]; |
1225 |
|
if (DEBUG) { |
1226 |
|
cout<<"pmtleft "<<pmtleft<<" j "<<j<<endl; |
1227 |
|
cout<<" atte1 "<<atte1[pmtleft+j]<<"lambda1 "<<lambda1[pmtleft+j]<<" atte2 "<<atte2[pmtleft+j]<<"lambda2 "<<lambda2[pmtleft+j] <<endl; |
1228 |
|
cout<<j<<" tpos "<<tpos<<" knorm "<<knorm[j]<<" "<<Atten[j]<<" "<<"QhitPmt_pC "<<QhitPmt_pC[j]<<endl; |
1229 |
|
} |
1230 |
|
} |
1231 |
|
|
1232 |
|
if (DEBUG) |
1233 |
|
cout<<"Npho "<<Npho<<" QhitPmt_pC "<<QhitPmt_pC[0]<<" "<<QhitPmt_pC[1]<<endl; |
1234 |
|
|
1235 |
Atten[j]=knorm[j]*(atte1[pmtleft+j]*exp(tpos/lambda1[pmtleft+j]) + |
QevePmt_pC[pmtleft] += QhitPmt_pC[0]; |
1236 |
atte2[pmtleft+j]*exp(tpos/lambda2[pmtleft+j])); |
QevePmt_pC[pmtright] += QhitPmt_pC[1]; |
1237 |
|
|
1238 |
QhitPmt_pC[j]= QhitPad_pC[j]*Atten[j]; |
// TDC |
1239 |
} |
t2 = t2 + fabs(path[0]/veff) + s_l_g[ip]/veff1 ; // Signal reaches PMT |
1240 |
|
t1 = t1 + fabs(path[1]/veff) + s_l_g[ip]/veff1; |
1241 |
QevePmt_pC[pmtleft] += QhitPmt_pC[0]; |
|
1242 |
QevePmt_pC[pmtright] += QhitPmt_pC[1]; |
TRandom r; |
1243 |
|
t1 = r.Gaus(t1,dt1); //apply gaussian error dt |
1244 |
} // **************************************** fine loop sugli hit |
t2 = r.Gaus(t2,dt1); //apply gaussian error dt |
1245 |
|
|
1246 |
|
t1 = t1 + c1_S[pmtleft] ; // Signal reaches Discriminator ,TDC starts to run |
1247 |
|
t2 = t2 + c1_S[pmtright] ; |
1248 |
|
|
1249 |
|
// check if signal is above threshold |
1250 |
|
// then check if tdcpmt is already filled by another hit... |
1251 |
|
// only re-fill if time is smaller |
1252 |
|
|
1253 |
|
if (QhitPmt_pC[0] > thresh) { |
1254 |
|
if (tdcpmt[pmtleft] == 1000.) { // fill for the first time |
1255 |
|
tdcpmt[pmtleft] = t1; |
1256 |
|
tdc[pmtleft] = t1 + c2_S[pmtleft] ; // Signal reaches Coincidence |
1257 |
|
} |
1258 |
|
if (tdcpmt[pmtleft] < 1000.) // is already filled! |
1259 |
|
if (t1 < tdcpmt[pmtleft]) { |
1260 |
|
tdcpmt[pmtleft] = t1; |
1261 |
|
t1 = t1 + c2_S[pmtleft] ; // Signal reaches Coincidence |
1262 |
|
tdc[pmtleft] = t1; |
1263 |
|
} |
1264 |
|
} |
1265 |
|
if (QhitPmt_pC[1] > thresh) { |
1266 |
|
if (tdcpmt[pmtright] == 1000.) { // fill for the first time |
1267 |
|
tdcpmt[pmtright] = t2; |
1268 |
|
tdc[pmtright] = t2 + c2_S[pmtright] ; // Signal reaches Coincidence |
1269 |
|
} |
1270 |
|
if (tdcpmt[pmtright] < 1000.) // is already filled! |
1271 |
|
if (t2 < tdcpmt[pmtright]) { |
1272 |
|
tdcpmt[pmtright] = t2; |
1273 |
|
t2 = t2 + c2_S[pmtright] ; |
1274 |
|
tdc[pmtright] = t2; |
1275 |
|
} |
1276 |
|
} |
1277 |
|
|
1278 |
|
if (DEBUG) |
1279 |
|
cout<<nh<<" "<<Timetof[nh]<<" "<<t1<<" "<<t2<<endl; |
1280 |
|
|
1281 |
|
} // ip < 6 |
1282 |
|
|
1283 |
|
}; // **************************************** end loop over hits |
1284 |
|
|
1285 |
|
// ====== ADC ====== |
1286 |
for(Int_t i=0; i<48; i++){ |
for(Int_t i=0; i<48; i++){ |
1287 |
if(QevePmt_pC[i] != 0.){ |
if(QevePmt_pC[i] != 0.){ |
1288 |
ADCtof[i]= (Int_t)(ADC_pC*QevePmt_pC[i] + ADCoffset); |
ADCtof[i]= (Int_t)(ADC_pC*QevePmt_pC[i] + ADCoffset); |
1289 |
if(ADCtof[i]> ADClast) ADCtof[i]=ADClast; |
if(ADCtof[i]> ADClast) ADCtof[i]=ADClast; |
1290 |
} else |
} else |
1291 |
ADCtof[i]= ADClast; |
ADCtof[i]= ADClast; |
1292 |
}; |
} |
1293 |
|
|
1294 |
|
|
1295 |
|
// ====== build TDC coincidence ====== |
1296 |
|
|
1297 |
|
Float_t t_coinc = 0; |
1298 |
|
Int_t ilast = 100; |
1299 |
|
for (Int_t ii=0; ii<48;ii++) |
1300 |
|
if (tdc[ii] > t_coinc) { |
1301 |
|
t_coinc = tdc[ii]; |
1302 |
|
ilast = ii; |
1303 |
|
} |
1304 |
|
|
1305 |
|
// cout<<ilast<<" "<<t_coinc<<endl; |
1306 |
|
// At t_coinc trigger condition is fulfilled |
1307 |
|
|
1308 |
|
for (Int_t ii=0; ii<48;ii++){ |
1309 |
|
// if (tdc[ii] != 0) tdc1[ii] = t_coinc - tdc[ii]; // test 1 |
1310 |
|
if (tdc[ii] != 0) tdc1[ii] = t_coinc - tdcpmt[ii]; // test 2 |
1311 |
|
tdc1[ii] = tdc1[ii]/tdcres[ii]; // divide by TDC resolution |
1312 |
|
if (tdc[ii] != 0) tdc1[ii] = tdc1[ii] + c3_S[ii]; // add cable length c3 |
1313 |
|
|
1314 |
|
} // missing parenthesis inserted! (Silvio) |
1315 |
|
|
1316 |
|
for(Int_t i=0; i<48; i++){ |
1317 |
|
if(tdc1[i] != 0.){ |
1318 |
|
TDCint[i]=(Int_t)tdc1[i]; |
1319 |
|
if (DEBUG) |
1320 |
|
cout<<i<<" "<<TDCint[i]<<endl; |
1321 |
|
//ADC[i]= ADC_pC * QevePmt_pC[i] + ADCoffset; |
1322 |
|
//if(ADC[i]> ADClast) ADC[i]=ADClast; |
1323 |
|
} else |
1324 |
|
TDCint[i]= TDClast; |
1325 |
|
} |
1326 |
|
|
1327 |
|
if (DEBUG) |
1328 |
|
cout<<"-----------"<<endl; |
1329 |
|
|
1330 |
|
// ====== write fDataTof ======= |
1331 |
UChar_t tofBin; |
UChar_t tofBin; |
|
// --- write fDataTof: |
|
1332 |
for (Int_t j=0; j < 12; j++){ |
for (Int_t j=0; j < 12; j++){ |
1333 |
Int_t j12=j*12; |
Int_t j12=j*12; |
1334 |
fDataTof[j12+0]=0x00; // TDC_ID |
fDataTof[j12+0]=0x00; // TDC_ID |
1341 |
fDataTof[jk12+4] = Bin2GrayTof(tofBin,fDataTof[jk12+4]); |
fDataTof[jk12+4] = Bin2GrayTof(tofBin,fDataTof[jk12+4]); |
1342 |
tofBin=(UChar_t)(ADCtof[k+4*j]%256); // ADC# (lsb) |
tofBin=(UChar_t)(ADCtof[k+4*j]%256); // ADC# (lsb) |
1343 |
fDataTof[jk12+5] = Bin2GrayTof(tofBin,fDataTof[jk12+5]); |
fDataTof[jk12+5] = Bin2GrayTof(tofBin,fDataTof[jk12+5]); |
1344 |
fDataTof[jk12+6]=0x00; // TDC# (msb) -- Wolfgang |
tofBin=(UChar_t)(TDCint[k+4*j]/256); // TDC# (msb) |
1345 |
fDataTof[jk12+7]=0x00; // TDC# (lsb) -- Wolfgang |
fDataTof[jk12+6]=Bin2GrayTof(tofBin,fDataTof[jk12+6]); |
1346 |
|
tofBin=(UChar_t)(TDCint[k+4*j]%256); // TDC# (lsb) |
1347 |
|
fDataTof[jk12+7]=Bin2GrayTof(tofBin,fDataTof[jk12+7]); |
1348 |
}; |
}; |
1349 |
fDataTof[j12+20]=0x00; // TEMP1 |
fDataTof[j12+20]=0x00; // TEMP1 |
1350 |
fDataTof[j12+21]=0x00; // TEMP2 |
fDataTof[j12+21]=0x00; // TEMP2 |
1410 |
somma+=pads[j]; |
somma+=pads[j]; |
1411 |
padid=paddle+somma; |
padid=paddle+somma; |
1412 |
*pl = padid*2; |
*pl = padid*2; |
1413 |
*pr = *pr + 1; |
// *pr = *pr + 1; |
1414 |
|
*pr = *pl + 1; // WM |
1415 |
}; |
}; |
1416 |
|
|
1417 |
void Digitizer::DigitizeAC() { |
void Digitizer::DigitizeAC() { |
1418 |
// created: J. Conrad, KTH |
// created: J. Conrad, KTH |
1419 |
// modified: S. Orsi, INFN Roma2 |
// modified: S. Orsi, INFN Roma2 |
1420 |
|
// fDataAC[0-63]: main AC board |
1421 |
|
// fDataAC[64-127]: extra AC board |
1422 |
|
|
1423 |
fDataAC[0] = 0xACAC; |
fDataAC[0] = 0xACAC; |
1424 |
fDataAC[64]= 0xACAC; |
fDataAC[64]= 0xACAC; |
1425 |
fDataAC[1] = 0xAC11; // main card |
fDataAC[1] = 0xAC11; |
1426 |
fDataAC[65] = 0xAC22; // extra card |
fDataAC[65] = 0xAC22; |
1427 |
|
|
1428 |
// the third word is a status word (dummy) |
// the third word is a status word (dummy: "no errors are present in the AC boards") |
1429 |
fDataAC[2] = 0xFFFF; //FFEF? |
fDataAC[2] = 0xFFFF; //FFEF? |
1430 |
fDataAC[66] = 0xFFFF; |
fDataAC[66] = 0xFFFF; |
1431 |
|
|
1432 |
const UInt_t nReg = 6; |
const UInt_t nReg = 6; |
1433 |
|
|
1434 |
// Registers (dummy) |
// FPGA Registers (dummy) |
1435 |
for (UInt_t i=0; i<=nReg; i++){ |
for (UInt_t i=0; i<=nReg; i++){ |
1436 |
fDataAC[i+4] = 0xFFFF; |
fDataAC[i+4] = 0xFFFF; |
1437 |
fDataAC[i+68] = 0xFFFF; |
fDataAC[i+68] = 0xFFFF; |
1442 |
fDataAC[63] = 0xABCD; |
fDataAC[63] = 0xABCD; |
1443 |
fDataAC[127] = 0xABCD; |
fDataAC[127] = 0xABCD; |
1444 |
|
|
1445 |
// shift registers, which one is with respect to PMT, where in |
// shift registers (moved to the end of the routine) |
|
// shift registers is a question of time relative trigger |
|
|
// In level2: hitmap, hitmap-status (synchronised with a trigger), |
|
|
// status |
|
|
|
|
|
for (UInt_t i=0; i<=15; i++){ |
|
|
fDataAC[i+11] = 0x0000; |
|
|
fDataAC[i+75] = 0x0000; |
|
|
} |
|
1446 |
|
|
1447 |
// singles counters are dummy |
Int_t evntLSB=Ievnt%65536; |
1448 |
|
Int_t evntMSB=(Int_t)(Ievnt/65536); |
1449 |
|
|
1450 |
for (UInt_t i=0; i<=16; i++){ |
// singles counters are dummy |
1451 |
fDataAC[i+26] = 0x0000; |
for (UInt_t i=0; i<=15; i++){ //SO Oct '07: // for (UInt_t i=0; i<=16; i++){ |
1452 |
fDataAC[i+90] = 0x0000; |
// fDataAC[i+26] = 0x0000; |
1453 |
} |
// fDataAC[i+90] = 0x0000; |
1454 |
|
fDataAC[i+26] = evntLSB; |
1455 |
|
fDataAC[i+90] = evntLSB; |
1456 |
|
}; |
1457 |
|
|
1458 |
// coincidences are dummy |
// coincidences are dummy (increment by 1 at each event) |
1459 |
|
// for (UInt_t i=0; i<=7; i++){ |
1460 |
|
// fDataAC[i+42] = 0x0000; |
1461 |
|
// fDataAC[i+106] = 0x0000; |
1462 |
|
// } |
1463 |
for (UInt_t i=0; i<=7; i++){ |
for (UInt_t i=0; i<=7; i++){ |
1464 |
fDataAC[i+42] = 0x0000; |
fDataAC[i+42] = evntLSB; |
1465 |
fDataAC[i+106] = 0x0000; |
fDataAC[i+106] = evntLSB; |
1466 |
} |
}; |
1467 |
|
|
1468 |
// increments for every trigger might be needed at some point. |
// increments for every trigger might be needed at some point. |
1469 |
// dummy for now |
// dummy for now |
1470 |
fDataAC[50] = 0x0000; |
fDataAC[50] = 0x0000; |
1471 |
fDataAC[114] = 0x0000; |
fDataAC[114] = 0x0000; |
1472 |
|
|
1473 |
// dummy FPGA clock |
// dummy FPGA clock (increment by 1 at each event) |
1474 |
|
/* |
1475 |
fDataAC[51] = 0x006C; |
fDataAC[51] = 0x006C; |
1476 |
fDataAC[52] = 0x6C6C; |
fDataAC[52] = 0x6C6C; |
1477 |
fDataAC[115] = 0x006C; |
fDataAC[115] = 0x006C; |
1478 |
fDataAC[116] = 0x6C6C; |
fDataAC[116] = 0x6C6C; |
1479 |
|
*/ |
1480 |
|
if (Ievnt<=0xFFFF) { |
1481 |
|
fDataAC[51] = 0x0000; |
1482 |
|
fDataAC[52] = Ievnt; |
1483 |
|
fDataAC[115] = 0x0000; |
1484 |
|
fDataAC[116] = Ievnt; |
1485 |
|
} else { |
1486 |
|
fDataAC[51] = evntMSB; |
1487 |
|
fDataAC[52] = evntLSB; |
1488 |
|
fDataAC[115] = fDataAC[51]; |
1489 |
|
fDataAC[116] = fDataAC[52]; |
1490 |
|
} |
1491 |
|
|
1492 |
// dummy temperatures |
// dummy temperatures |
1493 |
fDataAC[53] = 0x0000; |
fDataAC[53] = 0x0000; |
1502 |
fDataAC[i+119] = 0x1A13; |
fDataAC[i+119] = 0x1A13; |
1503 |
} |
} |
1504 |
|
|
1505 |
// We activate all branches. Once the digitization algorithm |
// We activate all branches. Once the digitization algorithm is determined |
1506 |
// is determined only the branches need to activated which involve needed |
// only the branches that involve needed information will be activated |
|
// information |
|
1507 |
|
|
1508 |
|
fhBookTree->SetBranchAddress("Ievnt",&Ievnt); |
1509 |
fhBookTree->SetBranchStatus("Nthcat",1); |
fhBookTree->SetBranchStatus("Nthcat",1); |
1510 |
fhBookTree->SetBranchStatus("Iparcat",1); |
fhBookTree->SetBranchStatus("Iparcat",1); |
1511 |
fhBookTree->SetBranchStatus("Icat",1); |
fhBookTree->SetBranchStatus("Icat",1); |
1551 |
// will fire. We will furthermore assume that both cards read out |
// will fire. We will furthermore assume that both cards read out |
1552 |
// identical data. |
// identical data. |
1553 |
|
|
1554 |
// If you develop you digitization algorithm, you should start by |
// If you develop your digitization algorithm, you should start by |
1555 |
// identifying the information present in level2 (post-darth-vader) |
// identifying the information present in level2 (post-darth-vader) |
1556 |
// data. |
// data. |
1557 |
|
|
1565 |
}; |
}; |
1566 |
|
|
1567 |
if (Nthcat>50 || Nthcas>50 || Nthcard>50) |
if (Nthcat>50 || Nthcas>50 || Nthcard>50) |
1568 |
printf("Error! NthAC out of range!\n\n"); |
printf("*** ERROR AC! NthAC out of range!\n\n"); |
1569 |
|
|
1570 |
|
// energy dependence on position (see file AcFitOutputDistancePmt.C by S.Orsi) |
1571 |
|
// based on J.Lundquist's calculations (PhD thesis, page 94) |
1572 |
|
// function: [0]+[1]*atan([2]/(x+1)), where the 3 parameters are: |
1573 |
|
// 8.25470e-01 +- 1.79489e-02 |
1574 |
|
// 6.41609e-01 +- 2.65846e-02 |
1575 |
|
// 9.81177e+00 +- 1.21284e+00 |
1576 |
|
// hp: 1 minimum ionising particle at 35cm from the PMT releases 1mip |
1577 |
|
// |
1578 |
|
// NB: the PMT positions are needed! |
1579 |
|
|
1580 |
// look in CAT |
// look in CAT |
1581 |
// for (UInt_t k= 0;k<50;k++){ |
// for (UInt_t k= 0;k<50;k++){ |
1639 |
|
|
1640 |
fDataAC[67] = fDataAC[3]; |
fDataAC[67] = fDataAC[3]; |
1641 |
|
|
1642 |
|
// shift registers |
1643 |
|
// the central bin is equal to the hitmap, all other bins in the shift register are 0 |
1644 |
|
for (UInt_t i=0; i<=15; i++){ |
1645 |
|
fDataAC[i+11] = 0x0000; |
1646 |
|
fDataAC[i+75] = 0x0000; |
1647 |
|
} |
1648 |
|
fDataAC[18] = fDataAC[3]; |
1649 |
|
fDataAC[82] = fDataAC[3]; |
1650 |
|
|
1651 |
// for (Int_t i=0; i<fACbuffer; i++){ |
// for (Int_t i=0; i<fACbuffer; i++){ |
1652 |
// printf("%0x ",fDataAC[i]); |
// printf("%0x ",fDataAC[i]); |
1653 |
// if ((i+1)%8 ==0) cout << endl; |
// if ((i+1)%8 ==0) cout << endl; |
1655 |
}; |
}; |
1656 |
|
|
1657 |
|
|
1658 |
void Digitizer::DigitizeND(){ |
void Digitizer::DigitizeS4(){ |
1659 |
// creato: S. Borisov, INFN Roma2 e MEPHI, Sept 2007 |
Int_t DEBUG=0; |
1660 |
// 4 bytes: 16bit header, 8bit trigPhysics, 16bit up&low background |
// creato: S. Borisov, INFN Roma2 e MEPHI, Sett 2007 |
1661 |
|
TString ciao,modo="ns"; |
1662 |
// ND header |
Int_t i,j,t,NdF,pmt,NdFT,S4,S4v=0,S4p=32; |
1663 |
fDataND[0] = 0x0000; |
Float_t E0,E1=1e-6,Ert,X,Y,Z,x,y,z,V[3],Xs[2],Ys[2],Zs[2],Yp[6],q,w,p=0.1,l,l0=500; |
1664 |
fDataND[1] = 0x000F; |
Xs[0]=-24.1; |
1665 |
|
Xs[1]=24.1; |
1666 |
|
Ys[0]=-24.1; |
1667 |
|
Ys[1]=24.1; |
1668 |
|
Zs[0]=-0.5; |
1669 |
|
Zs[1]=0.5; |
1670 |
|
Yp[0]=-20.; |
1671 |
|
Yp[2]=-1.; |
1672 |
|
Yp[4]=17.; |
1673 |
|
for(i=0;i<3;i++) |
1674 |
|
Yp[2*i+1]=Yp[2*i]+3; |
1675 |
|
srand(time(NULL)); |
1676 |
|
// --- activate branches: |
1677 |
|
fhBookTree->SetBranchStatus("Nthtof",1); |
1678 |
|
fhBookTree->SetBranchStatus("Ipltof",1); |
1679 |
|
fhBookTree->SetBranchStatus("Ipaddle",1); |
1680 |
|
|
1681 |
|
fhBookTree->SetBranchStatus("Xintof",1); |
1682 |
|
fhBookTree->SetBranchStatus("Yintof",1); |
1683 |
|
fhBookTree->SetBranchStatus("Xouttof",1); |
1684 |
|
fhBookTree->SetBranchStatus("Youttof",1); |
1685 |
|
|
1686 |
|
fhBookTree->SetBranchStatus("Ereltof",1); |
1687 |
|
fhBookTree->SetBranchStatus("Timetof",1); |
1688 |
|
NdFT=0; |
1689 |
|
Ert=0; |
1690 |
|
for(i=0;i<Nthtof;i++){ |
1691 |
|
if(Ipltof[i]!=6) continue; |
1692 |
|
Ert+=Ereltof[i]; |
1693 |
|
|
1694 |
|
if(modo=="ns") continue; |
1695 |
|
NdF=Int_t(Ereltof[i]/E1); |
1696 |
|
NdFT=0; |
1697 |
|
X=Xintof[i]; |
1698 |
|
Y=Yintof[i]; |
1699 |
|
Z=(Float_t)(random())/(Float_t)(0x7fffffff)-0.5; |
1700 |
|
//cout<<"XYZ "<<X<<" "<<Y<<" "<<Z<<endl; |
1701 |
|
for(j=0;j<NdF;j++){ |
1702 |
|
q=(Float_t)random()/(Float_t)0x7fffffff; |
1703 |
|
w=(Float_t)random()/(Float_t)0x7fffffff; |
1704 |
|
// cout<<"qw "<<q<<" "<<w<<endl; |
1705 |
|
V[0]=p*cos(6.28318*q); |
1706 |
|
V[1]=p*sin(6.28318*q); |
1707 |
|
V[2]=p*(2.*w-1.); |
1708 |
|
pmt=0; |
1709 |
|
x=X; |
1710 |
|
y=Y; |
1711 |
|
z=Z; |
1712 |
|
while(pmt==0 && (x>Xs[0] && x<Xs[1])&&(y>Ys[0] && y<Ys[1])&&(z>Zs[0] && z<Zs[1])){ |
1713 |
|
l=0; |
1714 |
|
while(pmt==0 && (x>Xs[0] && x<Xs[1])&&(y>Ys[0] && y<Ys[1])&&(z>Zs[0] && z<Zs[1])){ |
1715 |
|
x+=V[0]; |
1716 |
|
y+=V[1]; |
1717 |
|
z+=V[2]; |
1718 |
|
l+=p; |
1719 |
|
//cout<<x<<" "<<y<<" "<<z<<" "<<l<<endl; |
1720 |
|
//cin>>ciao; |
1721 |
|
} |
1722 |
|
if((x<Xs[0]+p || x>Xs[1]-p)&&(y>Ys[0]+p && y<Ys[1]-p)&&(z>Zs[0]+p && z<Zs[1]-p)){ |
1723 |
|
for(t=0;t<3;t++){ |
1724 |
|
if(y>=Yp[2*t] && y<Yp[2*t+1]){ |
1725 |
|
if(pmt==0)NdFT++; |
1726 |
|
pmt=1; |
1727 |
|
//cout<<NdFT<<endl; |
1728 |
|
break; |
1729 |
|
} |
1730 |
|
} |
1731 |
|
if(pmt==1)break; |
1732 |
|
V[0]=-V[0]; |
1733 |
|
} |
1734 |
|
q=(Float_t)random()/(Float_t)0x7fffffff; |
1735 |
|
w=1-exp(-l/l0); |
1736 |
|
if(q<w)break; |
1737 |
|
q=(Float_t)random()/(Float_t)0x7fffffff; |
1738 |
|
w=0.5; |
1739 |
|
if(q<w)break; |
1740 |
|
if((x>Xs[0]+p && x<Xs[1]-p)&&(y<Ys[0]+p || y>Ys[1]-p)&&(z>Zs[0]+p && z<Zs[1]-p))V[1]=-V[1]; |
1741 |
|
if((x>Xs[0]+p && x<Xs[1]-p)&&(y>Ys[0]+p && y<Ys[1]-p)&&(z<Zs[0]+p || z>Zs[1]-p))V[2]=-V[2]; |
1742 |
|
x+=V[0]; |
1743 |
|
y+=V[1]; |
1744 |
|
z+=V[2]; |
1745 |
|
l=0; |
1746 |
|
//cout<<x<<" "<<y<<" "<<z<<" "<<l<<endl; |
1747 |
|
//cin>>ciao; |
1748 |
|
} |
1749 |
|
} |
1750 |
|
} |
1751 |
|
Ert=Ert/0.002; |
1752 |
|
q=(Float_t)(random())/(Float_t)0x7fffffff; |
1753 |
|
w=0.7; |
1754 |
|
//E0=(Float_t)(4064./7.); |
1755 |
|
E0=4064./7.; |
1756 |
|
if(Ert<1) S4=0; |
1757 |
|
else S4=(Int_t)(4064.*(1.-exp(-(Ert-1.)/E0))); |
1758 |
|
i=S4/4; |
1759 |
|
if(S4%4==0) |
1760 |
|
S4v=S4+S4p; |
1761 |
|
else if(S4%4==1){ |
1762 |
|
if(q<w) S4v=S4-1+S4p; |
1763 |
|
else S4v=S4+1+S4p; |
1764 |
|
} else if(S4%4==2) S4v=S4+S4p; |
1765 |
|
else if(S4%4==3){ |
1766 |
|
if(q<w) S4v=S4+1+S4p; |
1767 |
|
else S4v=S4-1+S4p; |
1768 |
|
} |
1769 |
|
if (DEBUG) |
1770 |
|
cout<<"Ert_S4 = " << Ert << " --- S4v = " << S4v << endl; |
1771 |
|
fDataS4[0]=S4v;//0xf028; |
1772 |
|
fDataS4[1]=0xd800; |
1773 |
|
fDataS4[2]=0x0300; |
1774 |
|
//cout<<" PMT "<<NdFT<<" "<<NdF<<endl; |
1775 |
|
//cin>>ciao; |
1776 |
|
} |
1777 |
|
|
1778 |
|
|
1779 |
|
|
1780 |
|
void Digitizer::DigitizeND(){ |
1781 |
|
// creato: S. Borisov, INFN Roma2 e MEPHI, Sett 2007 |
1782 |
|
Int_t i=0; |
1783 |
|
UShort_t NdN=0; |
1784 |
fhBookTree->SetBranchStatus("Nthnd",1); |
fhBookTree->SetBranchStatus("Nthnd",1); |
1785 |
fhBookTree->SetBranchStatus("Itubend",1); |
fhBookTree->SetBranchStatus("Itubend",1); |
1786 |
fhBookTree->SetBranchStatus("Iparnd",1); |
fhBookTree->SetBranchStatus("Iparnd",1); |
1794 |
fhBookTree->SetBranchStatus("Timend",1); |
fhBookTree->SetBranchStatus("Timend",1); |
1795 |
fhBookTree->SetBranchStatus("Pathnd",1); |
fhBookTree->SetBranchStatus("Pathnd",1); |
1796 |
fhBookTree->SetBranchStatus("P0nd",1); |
fhBookTree->SetBranchStatus("P0nd",1); |
1797 |
|
//cout<<"n="<<Nthnd<<" "<<NdN<<"\n"; |
1798 |
|
for(i=0;i<Nthnd;i++){ |
1799 |
|
if(Iparnd[i]==13){ |
1800 |
|
NdN++; |
1801 |
|
} |
1802 |
|
} |
1803 |
|
//NdN=100; //only for debug |
1804 |
|
|
1805 |
UShort_t NdN=0; |
for(i=0;i<3;i++){ |
1806 |
for(Int_t i=0;i<Nthnd;i++) |
fDataND[2*i]=0x0000; |
1807 |
if(Iparnd[i]==13) |
fDataND[2*i+1]=0x010F; |
1808 |
NdN++; |
} |
1809 |
|
fDataND[0]=0xFF00 & (256*NdN); |
|
NdN=10; // test! |
|
|
fDataND[2]=0x0F00 & (NdN*256); |
|
|
//fDataND[2]=0xFFFF; //test |
|
|
fDataND[2]=0x0000; //background neutrons |
|
1810 |
} |
} |
1811 |
|
|
1812 |
|
|
1822 |
// printf("%0x ",fDataDummy[i]); |
// printf("%0x ",fDataDummy[i]); |
1823 |
//if ((i+1)%8 ==0) cout << endl; |
//if ((i+1)%8 ==0) cout << endl; |
1824 |
} |
} |
|
|
|
|
|
|
|
|
|
1825 |
}; |
}; |
1826 |
|
|
1827 |
|
|
1848 |
swab(fDataTrack,temp,sizeof(UShort_t)*fTracklength); // WE MUST SWAP THE BYTES!!! |
swab(fDataTrack,temp,sizeof(UShort_t)*fTracklength); // WE MUST SWAP THE BYTES!!! |
1849 |
fOutputfile.write(reinterpret_cast<char*>(temp),sizeof(UShort_t)*fTracklength); |
fOutputfile.write(reinterpret_cast<char*>(temp),sizeof(UShort_t)*fTracklength); |
1850 |
fTracklength=0; |
fTracklength=0; |
1851 |
// S4 |
// padding to 64 bytes |
|
// ...to be done... |
|
|
// ND |
|
|
memset(temp,0,sizeof(UShort_t)*1000000); |
|
|
swab(fDataND,temp,sizeof(UShort_t)*4); // WE MUST SWAP THE BYTES!!! |
|
|
fOutputfile.write(reinterpret_cast<char*>(temp),sizeof(UShort_t)*4); |
|
|
|
|
|
// |
|
|
// fOutputfile.write(reinterpret_cast<char*>(fDataDummy),sizeof(UShort_t)*fDummybuffer); |
|
|
// |
|
|
// padding to 64 bytes |
|
1852 |
// |
// |
1853 |
if ( fPadding ){ |
if ( fPadding ){ |
1854 |
fOutputfile.write(reinterpret_cast<char*>(fDataPadding),sizeof(UChar_t)*fPadding); |
fOutputfile.write(reinterpret_cast<char*>(fDataPadding),sizeof(UChar_t)*fPadding); |
1855 |
}; |
}; |
1856 |
// |
// S4 |
1857 |
|
memset(temp,0,sizeof(UShort_t)*1000000); |
1858 |
|
swab(fDataS4,temp,sizeof(UShort_t)*fS4buffer); // WE MUST SWAP THE BYTES!!! |
1859 |
|
fOutputfile.write(reinterpret_cast<char*>(temp),sizeof(UShort_t)*fS4buffer); |
1860 |
|
// ND |
1861 |
|
memset(temp,0,sizeof(UShort_t)*1000000); |
1862 |
|
swab(fDataND,temp,sizeof(UShort_t)*fNDbuffer); // WE MUST SWAP THE BYTES!!! |
1863 |
|
fOutputfile.write(reinterpret_cast<char*>(temp),sizeof(UShort_t)*fNDbuffer); |
1864 |
}; |
}; |
1865 |
|
|
1866 |
|
|
1953 |
|
|
1954 |
|
|
1955 |
|
|
|
|
|
1956 |
Float_t ADCfull; |
Float_t ADCfull; |
1957 |
|
Int_t iladd=0; |
1958 |
for (Int_t ix=0; ix<Nstrpx;ix++) { |
for (Int_t ix=0; ix<Nstrpx;ix++) { |
1959 |
Iview=Npstripx[ix]*2-1; |
Iview=Npstripx[ix]*2-1; |
1960 |
Nstrip=(Int_t)Istripx[ix]-1; |
Nstrip=(Int_t)Istripx[ix]-1; |
1961 |
ADCfull=AdcTrack[Iview][Nstrip] += Qstripx[ix]*fMipCor; |
if(Nstrip<fNstrips_ladder) iladd=0; |
1962 |
|
if((Nstrip>=fNstrips_ladder)&&(Nstrip<2*fNstrips_ladder)) iladd=1; |
1963 |
|
if((Nstrip>=2*fNstrips_ladder)&&(Nstrip<3*fNstrips_ladder)) iladd=2; |
1964 |
|
ADCfull=AdcTrack[Iview][Nstrip] += Qstripx[ix]*fMipCor[iladd][Iview]; |
1965 |
AdcTrack[Iview][Nstrip] *= SaturationTrack(ADCfull); |
AdcTrack[Iview][Nstrip] *= SaturationTrack(ADCfull); |
1966 |
|
|
1967 |
}; |
}; |
1970 |
for (Int_t iy=0; iy<Nstrpy;iy++) { |
for (Int_t iy=0; iy<Nstrpy;iy++) { |
1971 |
Iview=Npstripy[iy]*2-2; |
Iview=Npstripy[iy]*2-2; |
1972 |
Nstrip=(Int_t)Istripy[iy]-1; |
Nstrip=(Int_t)Istripy[iy]-1; |
1973 |
ADCfull=AdcTrack[Iview][Nstrip] -= Qstripy[iy]*fMipCor; |
if(Nstrip<fNstrips_ladder) iladd=0; |
1974 |
|
if((Nstrip>=fNstrips_ladder)&&(Nstrip<2*fNstrips_ladder)) iladd=1; |
1975 |
|
if((Nstrip>=2*fNstrips_ladder)&&(Nstrip<3*fNstrips_ladder)) iladd=2; |
1976 |
|
ADCfull=AdcTrack[Iview][Nstrip] -= Qstripy[iy]*fMipCor[iladd][Iview]; |
1977 |
AdcTrack[Iview][Nstrip] *= SaturationTrack(ADCfull); |
AdcTrack[Iview][Nstrip] *= SaturationTrack(ADCfull); |
1978 |
|
|
1979 |
}; |
}; |
2221 |
|
|
2222 |
void Digitizer::LoadMipCor() { |
void Digitizer::LoadMipCor() { |
2223 |
std:: cout << "Entering LoadMipCor" << endl; |
std:: cout << "Entering LoadMipCor" << endl; |
2224 |
|
Float_t xfactor=1./151.6*1.04; |
2225 |
|
Float_t yfactor=1./152.1; |
2226 |
|
|
2227 |
|
fMipCor[0][0]=140.02*yfactor; |
2228 |
|
fMipCor[0][1]=140.99*xfactor; |
2229 |
|
fMipCor[0][2]=134.48*yfactor; |
2230 |
|
fMipCor[0][3]=144.41*xfactor; |
2231 |
|
fMipCor[0][4]=140.74*yfactor; |
2232 |
|
fMipCor[0][5]=142.28*xfactor; |
2233 |
|
fMipCor[0][6]=134.53*yfactor; |
2234 |
|
fMipCor[0][7]=140.63*xfactor; |
2235 |
|
fMipCor[0][8]=135.55*yfactor; |
2236 |
|
fMipCor[0][9]=138.00*xfactor; |
2237 |
|
fMipCor[0][10]=154.95*yfactor; |
2238 |
|
fMipCor[0][11]=158.44*xfactor; |
2239 |
|
|
2240 |
|
|
2241 |
|
fMipCor[1][0]=136.07*yfactor; |
2242 |
|
fMipCor[1][1]=135.59*xfactor; |
2243 |
|
fMipCor[1][2]=142.69*yfactor; |
2244 |
|
fMipCor[1][3]=138.19*xfactor; |
2245 |
|
fMipCor[1][4]=137.35*yfactor; |
2246 |
|
fMipCor[1][5]=140.23*xfactor; |
2247 |
|
fMipCor[1][6]=153.15*yfactor; |
2248 |
|
fMipCor[1][7]=151.42*xfactor; |
2249 |
|
fMipCor[1][8]=129.76*yfactor; |
2250 |
|
fMipCor[1][9]=140.63*xfactor; |
2251 |
|
fMipCor[1][10]=157.87*yfactor; |
2252 |
|
fMipCor[1][11]=153.64*xfactor; |
2253 |
|
|
2254 |
|
fMipCor[2][0]=134.98*yfactor; |
2255 |
|
fMipCor[2][1]=143.95*xfactor; |
2256 |
|
fMipCor[2][2]=140.23*yfactor; |
2257 |
|
fMipCor[2][3]=138.88*xfactor; |
2258 |
|
fMipCor[2][4]=137.95*yfactor; |
2259 |
|
fMipCor[2][5]=134.87*xfactor; |
2260 |
|
fMipCor[2][6]=157.56*yfactor; |
2261 |
|
fMipCor[2][7]=157.31*xfactor; |
2262 |
|
fMipCor[2][8]=141.37*yfactor; |
2263 |
|
fMipCor[2][9]=143.39*xfactor; |
2264 |
|
fMipCor[2][10]=156.15*yfactor; |
2265 |
|
fMipCor[2][11]=158.79*xfactor; |
2266 |
|
|
2267 |
/* |
/* |
2268 |
for (Int_t j=0; j<fNviews;j++) { |
for (Int_t j=0; j<fNviews;j++) { |
2269 |
for (Int_t i=0; i<fNstrips_view;i++) { |
for (Int_t i=0; i<fNstrips_view;i++) { |
2556 |
|
|
2557 |
}; |
}; |
2558 |
|
|
|
Float_t Digitizer::SaturationTrack(Float_t ADC) { |
|
2559 |
|
|
2560 |
Float_t SatFact=1.; |
Float_t Digitizer::SaturationTrack(Float_t ADC) { |
2561 |
return SatFact; |
Float_t SatFact=1.; |
2562 |
|
if(ADC<70.) { SatFact=80./ADC; }; |
2563 |
|
if(ADC>3000.) { SatFact=3000./ADC; }; |
2564 |
|
return SatFact; |
2565 |
}; |
}; |
2566 |
|
|
2567 |
|
|