1 | #!/usr/bin/python
|
---|
2 | #
|
---|
3 | # Werner Lustermann
|
---|
4 | # ETH Zurich
|
---|
5 | #
|
---|
6 | from ctypes import *
|
---|
7 | import numpy as np
|
---|
8 | from scipy import signal
|
---|
9 |
|
---|
10 | # get the ROOT stuff + my shared libs
|
---|
11 | from ROOT import gSystem
|
---|
12 | # fitslib.so is made from fits.h and is used to access the data
|
---|
13 | gSystem.Load('~/py/fitslib.so')
|
---|
14 | from ROOT import *
|
---|
15 |
|
---|
16 | class rawdata( object ):
|
---|
17 | """
|
---|
18 | raw data access and calibration
|
---|
19 | - open raw data file and drs calibration file
|
---|
20 | - performs amplitude calibration
|
---|
21 | - performs baseline substraction if wanted
|
---|
22 | - provides all data in an array:
|
---|
23 | row = number of pixel
|
---|
24 | col = length of region of interest
|
---|
25 | """
|
---|
26 | # constructor of the classe
|
---|
27 | def __init__( self, dfname, calfname, bslfname='' ):
|
---|
28 | """
|
---|
29 | open data file and calibration data file
|
---|
30 | get basic information about the data in dfname
|
---|
31 | allocate buffers for data access
|
---|
32 |
|
---|
33 | dfname : fits or fits.gz file containing the data including the path
|
---|
34 | calfname : fits or fits.gz file containing DRS calibration data
|
---|
35 | bslfname : npy file containing the baseline values
|
---|
36 | """
|
---|
37 | self.dfname = dfname
|
---|
38 | self.calfname = calfname
|
---|
39 | self.bslfname = bslfname
|
---|
40 |
|
---|
41 | # baseline correction: True / False
|
---|
42 | if len( bslfname ) == 0:
|
---|
43 | self.correct_baseline = False
|
---|
44 | else:
|
---|
45 | self.correct_baseline = True
|
---|
46 |
|
---|
47 | # access data file
|
---|
48 | try:
|
---|
49 | df = fits( self.dfname )
|
---|
50 | except IOError:
|
---|
51 | print 'problem accessing data file: ', dfname
|
---|
52 | raise # stop ! no data
|
---|
53 | self.df = df
|
---|
54 |
|
---|
55 | # get basic information about the data file
|
---|
56 | self.NROI = df.GetUInt( 'NROI' ) # region of interest (length of DRS pipeline read out)
|
---|
57 | self.NPIX = df.GetUInt( 'NPIX' ) # number of pixels (should be 1440)
|
---|
58 | self.NEvents = df.GetNumRows() # find number of events
|
---|
59 | # allocate the data memories
|
---|
60 | self.evNum = c_ulong()
|
---|
61 | self.trigType = c_ushort()
|
---|
62 | self.Data = np.zeros( self.NPIX * self.NROI, np.int16 )
|
---|
63 | self.startCells = np.zeros( self.NPIX, np.int16 )
|
---|
64 | # set the pointers to the data++
|
---|
65 | df.SetPtrAddress( 'EventNum', self.evNum )
|
---|
66 | df.SetPtrAddress( 'TriggerType', self.trigType )
|
---|
67 | df.SetPtrAddress( 'StartCellData', self.startCells ) # DRS readout start cell
|
---|
68 | df.SetPtrAddress( 'Data', self.Data ) # this is what you would expect
|
---|
69 | # df.GetNextRow() # access the first event
|
---|
70 |
|
---|
71 | # access calibration file
|
---|
72 | try:
|
---|
73 | calf = fits( self.calfname )
|
---|
74 | except IOError:
|
---|
75 | print 'problem accessing calibration file: ', calfname
|
---|
76 | raise
|
---|
77 | self.calf = calf
|
---|
78 | #
|
---|
79 | BaselineMean = calf.GetN('BaselineMean')
|
---|
80 | GainMean = calf.GetN('GainMean')
|
---|
81 | TriggerOffsetMean = calf.GetN('TriggerOffsetMean')
|
---|
82 |
|
---|
83 | self.blm = np.zeros( BaselineMean, np.float32 )
|
---|
84 | self.gm = np.zeros( GainMean, np.float32 )
|
---|
85 | self.tom = np.zeros( TriggerOffsetMean, np.float32 )
|
---|
86 |
|
---|
87 | self.Nblm = BaselineMean / self.NPIX
|
---|
88 | self.Ngm = GainMean / self.NPIX
|
---|
89 | self.Ntom = TriggerOffsetMean / self.NPIX
|
---|
90 |
|
---|
91 | calf.SetPtrAddress( 'BaselineMean', self.blm )
|
---|
92 | calf.SetPtrAddress( 'GainMean', self.gm )
|
---|
93 | calf.SetPtrAddress( 'TriggerOffsetMean', self.tom )
|
---|
94 | calf.GetRow(0)
|
---|
95 |
|
---|
96 | self.v_bsl = np.zeros( self.NPIX ) # array with baseline values (all ZERO)
|
---|
97 | self.smoothData = None
|
---|
98 | self.maxPos = None
|
---|
99 | self.maxAmp = None
|
---|
100 |
|
---|
101 |
|
---|
102 | def next( self ):
|
---|
103 | """
|
---|
104 | load the next event from disk and calibrate it
|
---|
105 | """
|
---|
106 | self.df.GetNextRow()
|
---|
107 | self.calibrate_drsAmplitude()
|
---|
108 |
|
---|
109 |
|
---|
110 | def calibrate_drsAmplitude( self ):
|
---|
111 | """
|
---|
112 | perform amplitude calibration for the event
|
---|
113 | """
|
---|
114 | tomV = 2000./4096.
|
---|
115 | acalData = self.Data * tomV # convert into mV
|
---|
116 |
|
---|
117 | # reshape arrays: row = pixel, col = drs_slice
|
---|
118 | acalData = np.reshape( acalData, (self.NPIX, self.NROI) )
|
---|
119 | blm = np.reshape( self.blm, (self.NPIX, 1024) )
|
---|
120 | tom = np.reshape( self.tom, (self.NPIX, 1024) )
|
---|
121 | gm = np.reshape( self.gm, (self.NPIX, 1024) )
|
---|
122 |
|
---|
123 | # print 'acal Data ', acalData.shape
|
---|
124 | # print 'blm shape ', blm.shape
|
---|
125 | # print 'gm shape ', gm.shape
|
---|
126 |
|
---|
127 | for pixel in range( self.NPIX ):
|
---|
128 | # rotate the pixel baseline mean to the Data startCell
|
---|
129 | blm_pixel = np.roll( blm[pixel,:], -self.startCells[pixel] )
|
---|
130 | acalData[pixel,:] -= blm_pixel[0:self.NROI]
|
---|
131 | acalData[pixel,:] -= tom[pixel, 0:self.NROI]
|
---|
132 | acalData[pixel,:] /= gm[pixel, 0:self.NROI]
|
---|
133 |
|
---|
134 | self.acalData = acalData * 1907.35
|
---|
135 |
|
---|
136 | # print 'acalData ', self.acalData[0:2,0:20]
|
---|
137 |
|
---|
138 | def filterSlidingAverage( self , windowSize = 4):
|
---|
139 | """ sliding average filter
|
---|
140 | using:
|
---|
141 | self.acalData
|
---|
142 | filling array:
|
---|
143 | self.smoothData
|
---|
144 | """
|
---|
145 | #scipy.signal.lfilter(b, a, x, axis=-1, zi=None)
|
---|
146 | smoothData = self.acalData.copy()
|
---|
147 | b = np.ones( windowSize )
|
---|
148 | a = np.zeros( windowSize )
|
---|
149 | a[0] = len(b)
|
---|
150 | smoothData[:,:] = signal.lfilter(b, a, smoothData[:,:])
|
---|
151 |
|
---|
152 | self.smoothData = smoothData
|
---|
153 |
|
---|
154 | def filterCFD( self, length=10, ratio=0.75):
|
---|
155 | """ constant fraction filter
|
---|
156 | using:
|
---|
157 | self.smoothData
|
---|
158 | filling array:
|
---|
159 | self.cfdData
|
---|
160 | """
|
---|
161 | if self.smoothData == None:
|
---|
162 | print 'error pyfact.filterCFD was called without prior call to filterSlidingAverage'
|
---|
163 | print ' variable self.smoothData is needed '
|
---|
164 | pass
|
---|
165 | cfdData = self.smoothData.copy()
|
---|
166 | b = np.zeros( length )
|
---|
167 | a = np.zeros( length )
|
---|
168 | b[0] = -1. * ratio
|
---|
169 | b[length-1] = 1.
|
---|
170 | a[0] = 1.
|
---|
171 | cfdData[:,:] = signal.lfilter(b, a, cfdData[:,:])
|
---|
172 |
|
---|
173 | self.cfdData = cfdData
|
---|
174 |
|
---|
175 | def findPeak (self, min=30, max=250):
|
---|
176 | """ find maximum in search window
|
---|
177 | using:
|
---|
178 | self.smoothData
|
---|
179 | filling arrays:
|
---|
180 | self.maxPos
|
---|
181 | self.maxAmp
|
---|
182 | """
|
---|
183 | if self.smoothData == None:
|
---|
184 | print 'error pyfact.findPeakMax was called without prior call to filterSlidingAverage'
|
---|
185 | print ' variable self.smoothData is needed '
|
---|
186 | pass
|
---|
187 | maxPos = np.argmax( self.smoothData[:,min:max] , 1)
|
---|
188 | maxAmp = np.max( self.smoothData[:,min:max] , 1)
|
---|
189 | self.maxPos = maxPos
|
---|
190 | self.maxAmp = maxAmp
|
---|
191 |
|
---|
192 | def sumAroundPeak (self, left=13, right=23):
|
---|
193 | """ integrate signal in gate around Peak
|
---|
194 | using:
|
---|
195 | self.maxPos
|
---|
196 | self.acalData
|
---|
197 | filling array:
|
---|
198 | self.integral
|
---|
199 | """
|
---|
200 | if self.maxPos == None:
|
---|
201 | print 'error pyfact.sumAroundPeak was called without prior call of findPeak'
|
---|
202 | print ' variable self.maxPos is needed'
|
---|
203 | pass
|
---|
204 |
|
---|
205 | sums = np.empty( self.NPIX )
|
---|
206 | for pixel in range( self.NPIX ):
|
---|
207 | min = self.maxPos[pixel]-left
|
---|
208 | max = self.maxPos[pixel]+right
|
---|
209 | sums[pixel] = self.acalData[pixel,min:max].sum()
|
---|
210 |
|
---|
211 | self.integral = sums
|
---|
212 |
|
---|
213 | def ReadBaseline( self, file, bsl_hist = 'bsl_sum/hplt_mean' ):
|
---|
214 | """
|
---|
215 | open ROOT file with baseline histogram and read baseline values
|
---|
216 | file name of the root file
|
---|
217 | bsl_hist path to the histogram containing the basline values
|
---|
218 | """
|
---|
219 | try:
|
---|
220 | f = TFile( file )
|
---|
221 | except:
|
---|
222 | print 'Baseline data file could not be read: ', file
|
---|
223 | return
|
---|
224 |
|
---|
225 | h = f.Get( bsl_hist )
|
---|
226 |
|
---|
227 | for i in range( self.NPIX ):
|
---|
228 | self.v_bsl[i] = h.GetBinContent( i+1 )
|
---|
229 |
|
---|
230 | f.Close()
|
---|
231 |
|
---|
232 |
|
---|
233 | def CorrectBaseline( self ):
|
---|
234 | """
|
---|
235 | apply baseline correction
|
---|
236 | """
|
---|
237 | for pixel in range( self.NPIX ):
|
---|
238 | self.acalData[pixel,:] -= self.v_bsl[pixel]
|
---|
239 |
|
---|
240 |
|
---|
241 | def info( self ):
|
---|
242 | """
|
---|
243 | print information
|
---|
244 | """
|
---|
245 | print 'data file: ', dfname
|
---|
246 | print 'calib file: ', calfname
|
---|
247 | print 'calibration file'
|
---|
248 | print 'N BaselineMean: ', self.Nblm
|
---|
249 | print 'N GainMean: ', self.Ngm
|
---|
250 | print 'N TriggeroffsetMean: ', self.Ntom
|
---|
251 |
|
---|
252 | # --------------------------------------------------------------------------------
|
---|
253 | class fnames( object ):
|
---|
254 | """ organize file names of a FACT data run
|
---|
255 |
|
---|
256 | """
|
---|
257 |
|
---|
258 | def __init__( self, specifier = ['012', '023', '2011', '11', '24'],
|
---|
259 | rpath = '/scratch_nfs/bsl/',
|
---|
260 | zipped = True):
|
---|
261 | """
|
---|
262 | specifier : list of strings defined as:
|
---|
263 | [ 'DRS calibration file', 'Data file', 'YYYY', 'MM', 'DD']
|
---|
264 |
|
---|
265 | rpath : directory path for the results; YYYYMMDD will be appended to rpath
|
---|
266 | zipped : use zipped (True) or unzipped (Data)
|
---|
267 | """
|
---|
268 | self.specifier = specifier
|
---|
269 | self.rpath = rpath
|
---|
270 | self.zipped = zipped
|
---|
271 |
|
---|
272 | self.make( self.specifier, self.rpath, self.zipped )
|
---|
273 | # end of def __init__
|
---|
274 |
|
---|
275 | def make( self, specifier, rpath, zipped ):
|
---|
276 | """ create (make) the filenames
|
---|
277 |
|
---|
278 | names : dictionary of filenames, tags { 'data', 'drscal', 'results' }
|
---|
279 | data : name of the data file
|
---|
280 | drscal : name of the drs calibration file
|
---|
281 | results : radikal of file name(s) for results (to be completed by suffixes)
|
---|
282 | """
|
---|
283 |
|
---|
284 | self.specifier = specifier
|
---|
285 |
|
---|
286 | if zipped:
|
---|
287 | dpath = '/data00/fact-construction/raw/'
|
---|
288 | ext = '.fits.gz'
|
---|
289 | else:
|
---|
290 | dpath = '/data03/fact-construction/raw/'
|
---|
291 | ext = '.fits'
|
---|
292 |
|
---|
293 | year = specifier[2]
|
---|
294 | month = specifier[3]
|
---|
295 | day = specifier[4]
|
---|
296 |
|
---|
297 | yyyymmdd = year + month + day
|
---|
298 | dfile = specifier[1]
|
---|
299 | cfile = specifier[0]
|
---|
300 |
|
---|
301 | rpath = rpath + yyyymmdd + '/'
|
---|
302 | self.rpath = rpath
|
---|
303 | self.names = {}
|
---|
304 |
|
---|
305 | tmp = dpath + year + '/' + month + '/' + day + '/' + yyyymmdd + '_'
|
---|
306 | self.names['data'] = tmp + dfile + ext
|
---|
307 | self.names['drscal'] = tmp + cfile + '.drs' + ext
|
---|
308 | self.names['results'] = rpath + yyyymmdd + '_' + dfile + '_' + cfile
|
---|
309 |
|
---|
310 | self.data = self.names['data']
|
---|
311 | self.drscal = self.names['drscal']
|
---|
312 | self.results = self.names['results']
|
---|
313 |
|
---|
314 | # end of make
|
---|
315 |
|
---|
316 | def info( self ):
|
---|
317 | """ print complete filenames
|
---|
318 |
|
---|
319 | """
|
---|
320 |
|
---|
321 | print 'file names:'
|
---|
322 | print 'data: ', self.names['data']
|
---|
323 | print 'drs-cal: ', self.names['drscal']
|
---|
324 | print 'results: ', self.names['results']
|
---|
325 | # end of def info
|
---|
326 |
|
---|
327 | # end of class definition: fnames( object )
|
---|
328 |
|
---|
329 |
|
---|
330 |
|
---|
331 | class histogramList( object ):
|
---|
332 |
|
---|
333 | def __init__( self, name ):
|
---|
334 | """ set the name and create empty lists """
|
---|
335 | self.name = name # name of the list
|
---|
336 | self.list = [] # list of the histograms
|
---|
337 | self.dict = {} # dictionary of histograms
|
---|
338 | self.hList = TObjArray() # list a la ROOT of the histograms
|
---|
339 |
|
---|
340 | def add( self, tag, h ):
|
---|
341 | self.list.append( h )
|
---|
342 | self.dict[tag] = h
|
---|
343 | self.hList.Add( h )
|
---|
344 |
|
---|
345 |
|
---|
346 | class pixelHisto1d ( object ):
|
---|
347 |
|
---|
348 | def __init__( self, name, title, Nbin, first, last, xtitle, ytitle, NPIX ):
|
---|
349 | """
|
---|
350 | book one dimensional histograms for each pixel
|
---|
351 | """
|
---|
352 | self.name = name
|
---|
353 |
|
---|
354 | self.list = [ x for x in range( NPIX ) ]
|
---|
355 | self.hList = TObjArray()
|
---|
356 |
|
---|
357 | for pixel in range( NPIX ):
|
---|
358 |
|
---|
359 | hname = name + ' ' + str( pixel )
|
---|
360 | htitle = title + ' ' + str( pixel )
|
---|
361 | self.list[pixel] = TH1F( hname, htitle, Nbin, first, last )
|
---|
362 |
|
---|
363 | self.list[pixel].GetXaxis().SetTitle( xtitle )
|
---|
364 | self.list[pixel].GetYaxis().SetTitle( ytitle )
|
---|
365 | self.hList.Add( self.list[pixel] )
|
---|
366 |
|
---|
367 | # simple test method
|
---|
368 | if __name__ == '__main__':
|
---|
369 | """
|
---|
370 | create an instance
|
---|
371 | """
|
---|
372 | dfname = '/data03/fact-construction/raw/2011/11/24/20111124_121.fits'
|
---|
373 | calfname = '/data03/fact-construction/raw/2011/11/24/20111124_111.drs.fits'
|
---|
374 | rd = rawdata( dfname, calfname )
|
---|
375 | rd.info()
|
---|
376 | rd.next()
|
---|
377 |
|
---|
378 | # for i in range(10):
|
---|
379 | # df.GetNextRow()
|
---|
380 |
|
---|
381 | # print 'evNum: ', evNum.value
|
---|
382 | # print 'startCells[0:9]: ', startCells[0:9]
|
---|
383 | # print 'evData[0:9]: ', evData[0:9]
|
---|