X-Pipeline Review Meeting Monday 28 July 2008 09:00 Pacific / 12:00 Eastern
Agenda: Monday 28 July 2008 09:00 Pacific / 12:00 Eastern
Updates on reviewers' analyses of GRBs.
- Instructions and suggested GRB assignments: [ text ].
Code walk-through: m2html documentation for revision 2301 [ web ]
- We are currently at line 712 of xdetection.m.
Attendance: Patrick, Gareth, Xavi
- Starting at xdetection line 712: Right before data conditioning and ffts
- 723-730 Create cell arrays to contain results for different analysisTimes. Cell arrays are arrays that can hold miscellaneous data types, very flexible
- 739-936 loop over analysisTimes
- 755 If using real data: Use Linear Predictor Error Filter (LPEF) to condition data in xcondition, can train filter using analysis data (line 756) or can load pre-computed filter (line 765). Only difference between xcondition calls is final arg which determines LPEF coeffiecients. In production analysis we train LPEF using analysis data, i.e., line 756
- xdetection reviewed to line 777, moving to look at xcondition.
- Patrick: Commenting and doc needs improving (ticket created)
- Xavi: Worried that filter is not long enough
- Patrick: Take 256s of time domain data, FFT it, square it to obtain something like PSD, IFFT it and take first second of this autocorrelation function to train filter using inbuilt matlab functions
- Xavi: suggesting increasing whitening time upwards from 1s, he uses 4s in matched filter
- Patrick: Ideas for tests of this code (see end of minutes)
- Patrick: Also noticed some over-whitening (i.e., notching) of spectral lines, not worried about this since over-whitening is preferred to under-whitening. Xavi agrees
- Patrick: explains input args to xcondition.m training mode is set to 'search' for production runs
- xcondition lines 100-112: hard coded params
- Xavi: are these filters FIR?
- Patrick: No, filtering takes place in freq domain
- xcondition lines 114-235: check input args
- xcondition line 188-191: analysisTime is length of FFT which will be applied to data, check that whitening time is longer than this analysisTime so that on this scale the data appears white
- xcondition line 218: In production analysis we always set training mode to 'search' Other options used in coherent follow up searches where we want to train filters while avoiding a short stretch of data containing event candidate of interest
- xcondition lines 255-265: Resample data using standard matlab function resample.m
- Xavi: concerned that ignoring only data near NYQUIST frequency might not be sufficient to avoid effects of aliasing
- Finished review of xcondition at line 277
- Tests to be performed:
- i) Post some plots of conditioned data using different whitening times, compare Gaussianity.
- ii) Check that conditioned data is found to be white by independent method
- iii) Check that output statistics follow chi-squares
- iv) Plot autocorrelation function for different whitening times
- v) Investigate effects of aliasing at frequencies below NYQUIST (see comments ~line 255)