Transcript Diapositiva 1 - Istituto Nazionale di Fisica Nucleare

S. Frasca
Baton Rouge, March 2007



Whole sky blind hierarchical search
(P.Astone, SF, C.Palomba - Roma1)
Targeted search (F.Antonucci, F. Ricci –
Roma1)
Binary source search (T.Bauer , J.v.d.Brand,
S.v.d.Putten – Amsterdam)

Our method is based on the use of Hough
maps, built starting from peak maps
obtained by the SFTs.
h-reconstructed data
Here is a rough
sketch of our
pipeline
Data quality
Data quality
SFDB
SFDB
Average spect
rum estimation
Average spect
rum estimation
peak map
peak map
hough transf.
hough transf.
candidates
coincidences
candidates
coherent
step
events
4

The software is described in the document
at http://grwavsf.roma1.infn.it/pss/docs/PSS_UG.pdf



Time-domain big event removal
Non-linear adaptive estimation of the power
spectrum (these estimated p.s. are saved together
with the SFTs and the peak maps.
Only relative maxima are taken (little less
sensitivity in the ideal case, much more robustness
in practice)

Periodogram of 222 (= 4194304 ) data of C7
Seconds in abscissa. Note on the
full piece the slow amplitude
variation and in the zoom the
perfect synchronization with the
deci-second.
1kHz band analysis: peak maps
• On the peak maps there is a further cleaning procedure consisting in
putting a threshold on the peaks frequency distribution
• This is needed in order to avoid a too much large number of
candidates which implies a reduction in sensitivity.
C7: peaks frequency distribution before and after cleaning
25


Now we are using the “standard” (not
“adaptive”) Hough transform
Here are the results
Parameter space
• observation time
C 6 : Tobs  13.87days C 7 : Tobs  3.37days
• frequency band
50Hz 1050Hz
• frequency resolution f  0.00095367 Hz
TFFT  1048.576 s
• number of FFTs
C 6 : N FFT  2286
• sky resolution
,   7.5 deg (@ 50Hz )  0.3 deg (@1050Hz)
 1
  
 f
• spin-down resolution
C 7 : N FFT  556
  100 yr (@50Hz)  2100 yr (@1050Hz)
fmax  1.58 108 Hz/s
C6 : N sd  40 f  4.06 10 10 Hz/s
C7 : N  10 f  1.76 10 9 Hz/s
sd
~1013 points in the parameter space are explored for each
data set
27
Candidates selection
• On each Hough map (corresponding to a given frequency
and spin-down) candidates are selected putting a
n   map
CR 
threshold on the CR

map
• The choice of the threshold is done according to the
maximum number of candidates we can manage in the next
steps of the analysis
• In this analysis we have used CRthr  3.8
• Number of candidates found:
C6: 922,999,536 candidates
C7: 319,201,742 candidates
28
1kHz band: candidates analysis
C6: frequency distribution of candidates (spin-down 0)
f [Hz]
29
C6: frequency distribution of candidates (spin-down 0)
f [Hz]
Sky distribution of candidates (~673.8Hz)
peaks frequency distribution
 [deg]
 [deg]
30
C6: frequency distribution of candidates (spin-down 0)
f [Hz]
Sky distribution of candidates (~980Hz)
peaks frequency distribution
 [deg]
 [deg]
31
C6: frequency distribution of candidates (spin-down 0)
f [Hz]
Sky distribution of candidates (881-889Hz)
peaks frequency distribution
 [deg]
 [deg]
32
C7: frequency distribution of candidates (spin-down 0)
Sky distribution of candidates (779.5Hz)
f [Hz]
peaks frequency distribution
 [deg]
 [deg]
33
red line: theoretical distribution
34
‘disturbed’
band
‘quiet’ band
Many candidates appear in
‘bumps’ (at high latitude), due
to the short observation
time, and ‘strips’ (at low
latitude), due to the
symmetry of the problem
35
Coincidences
• To reduce the false alarm probability; reduce also the
computational load of the coherent “follow-up”
• Done comparing the set of parameter values identifying
each candidate
• Coincidence windows: f  1, f  0,   2,   2
• Number of coincidences: 2,700,232
• False alarm probability: 2.2 107
band 1045-1050 Hz
36
‘Mixed data’ analysis
• Let us consider two set of ‘mixed’ data:
A6
B6
A7
C6
B7
C7
time
• Produce candidates for data set A=A6+A7
• Produce candidates for data set B=B6+B7
• Make coincidences between A and B
• Two main advantages:
• larger time interval -> less ‘bunches’ of candidates expected
• easier comparison procedure (same spin-down step for both sets)
40



In any log file there are mainly: comments,
parameters, “events” and statistics.
These are the log files of the SFDB
construction
There are information on big time events
and big frequency lines (as “events”)





























File D:\SF_DatAn\pss_datan\Reports\crea_sfdb_20060131_173851.log
started at Tue Jan 31 17:38:51 2006
even NEW: a new FFT has started
PAR1: Beginning time of the new FFT
PAR2: FFT number in the run
even EVT: time domain events
PAR1: Beginning time, in mjd
PAR2: Duration [s]
PAR3: Max amplitude*EINSTEIN
even EVF: frequency domain events, with high threshold
PAR1: Beginning frequency of EVF
PAR2: Duration [Hz]
PAR3: Ratio, in amplitude, max/average
PAR4: Power*EINSTEIN**2 or average*EINSTEIN (average if duration=0, when age>maxage)
stat TOT: total number of frequency domain events
par GEN: general parameters of the run
GEN_BEG is the beginning time (mjd)
GEN_NSAM the number of samples in 1/2 FFT
GEN_DELTANU the frequency resolution
GEN_FRINIT the beginning frequency of the FFT
EVT_CR is the threshold
EVT_TAU the memory time of the AR estimation
EVT_DEADT the dead time [s]
EVT_EDGE seconds purged around the event
EVF_THR is the threshold in amplitude
EVF_TAU the memory frequency of the AR estimation
EVF_MAXAGE [Hz] the max age of the process. If age>maxage the AR is re-evaluated
EVF_FAC is the factor for which the threshold is multiplied, to write less EVF in the log file
stop at Wed Feb 1 12:39:22 2006