Data Analysis of TAMA - University of Tokyot-munu.phys.s.u-tokyo.ac.jp/sympo2002/proc/kanda.pdf ·...
Transcript of Data Analysis of TAMA - University of Tokyot-munu.phys.s.u-tokyo.ac.jp/sympo2002/proc/kanda.pdf ·...
N.KandaMiyagi Univ. of Education
Data Analysis of TAMANobuyuki Kanda
the TAMA collaborationsMiyagi University of Education, National Astronomical Observatory, University of Tokyo, University ofElectro-Communications, High Energy Accelerator Research Organization, Osaka University, Kyoto
University, Max-Planck-Institut fur Quantenoptik, National Research Laboratory of Metrology, HirosakiUniversity, Kinki University,
Tokyo Denki University, Osaka City University, Tokai University,Tohoku University, Niigata University, Hiroshima University
TAMA symposium2/6/2002
Hongo, Tokyo
N.KandaMiyagi Univ. of Education
Data Taking History
Period actual data amount
Data Taking 1 (DT1) 8/6-7/1999 (one night) ~3 + ~7 hours continuous lock
DT2 first Phyics run 9/17-20/1999 31 hours
DT3 4/20 - 4/23/2000 13 hours
DT4 8/21 - 9/3/2000 167 hours
DT5 3/1 - 3/8/2001 111 hours
Test Run 1 6/4 - 6/6/2001Longest stretch of continuous lock is 24hours 50min.The interferometer can operate in daytime of weedkay.
DT6 8/1 - 9/20/2001 1038 hours with duty cycle 86%h ~ 5×10-21 [1/√Hz]
1999-12-2
5
Detector Schematics (DAQ)
AD
C(S
ON
Y te
ktro
nix
VX
4244
)
SCSI
2GB
2 GB * 4 Disks1st level spool disk
DLT tape system(5 tapes at once)
UNIX Workstation(SUN Enterprise 450, OS Solaris 2)
VXI crate
intranet
2GB 2GB 2GB
Main IF signal * 7chL- feedback,Calibration Referenceetc.
20 kHz trigger signal(generated from 10MHz GPS clock)
IRIG-B time coded signal
MIX BUS Interface
• Continuous Data Taking• IF Calibration Signal Record
On-line analysis (1)
— Data acquisition system —
11
HDAQ raw data
MDAQ raw data
EPICS raw data
Mirror of HDAQ
crhsun1
online analysis (MF, SNR)
online analysis (IF diag., mdaq)
signals
signals
fft
mdaq
crhsun2
signals
VME
DAQ front end
ADC-SCSI
VXI-MXI
DLTHardDisk
Acquisition ArchiveSpool & Transfer
cresun1
NFS server
NFS server
~daq/online_monitor/copyRecent
DAQ processes(tqStart)
DAQ processes
DAQ processes (tqEpics)
scp
EPICS print taskprint out
NFSNFS
NFS
/export/home*/R*F*.raw
/usr4/
~daq/hdaq_mirroring
/import/usr4/onlmon/spool/
/data/hdaq/
/data/mdaq/
/import/data/hdaq/
/import/data/hdaq//import/data/mdaq
~daq/src/tqEpics/copyEdata
~daq/online_monitor/startMon
~daq/epics_monitor/startEmon
Monitor: HdaqMon
Monitor: MdaqMon
Monitor: EpicsMon
display
Monitor: ArchiveMon
N.KandaMiyagi Univ. of Education
DAQ and pre-process
DAQ systemDAQCalibration
accuracy ∆h/h ~1%
Pre-ProcessCut-out observation data
remove unlock or tuning periods
h(f) : strain equivarent spectrum calculusv(t)-> v(f) T(f) -> h(f)
Analysis
N.KandaMiyagi Univ. of Education
Two issues of the data analysis
Detector evaluation
IF operation stability, gaussianity
Detection Feasibility
GW search
Stochastic
Binary Coalescence
Burst
Continuous
N.KandaMiyagi Univ. of Education
Sensitivity / Data amount,and possible issues
before DTSimulationMethod study
DT1 / DT2 : a few night runCalibrationOff-line data selectionStability, non-Gaussinanity checkMatched filter analysis
DT3 / DT4 : more than 100 hours, operation in nightVeto analysisIF diagnosisContinuous analysis
DT5 : preparation of 1000 hoursDT6 : 1000 hours
Fast IF diagnosisExpected SNR monitor
= Real time monitor, feedback to the detector operation
02.2.6
7
Calibration
1.0
1.2
1.4
1.6
1.8
2.0
Sep 1920:00
Sep 1922:00
Sep 2000:00
Sep 2002:00
Sep 2004:00
Sep 2006:00
Gai
n
Time
Drift of openloop TF at 625 Hz
40
30
20
10
0
yield
-4 -2 0 2 4
delta G between 16 frames/mean at time [%]
'measured'
'gauss fit (result: 1sigma
Gain drift for night ~ 30%
Calibration accuracy ∆h/h ~ 1% @625Hz
run12 (DAQ 2)
02.2.6
6
DAQ 2 & DAQ 3: Noise Instability
noise average around 1kHz
run12 (DAQ 2)
run34 (DAQ 3)
run35 (DAQ 3)
9 hours
1e-21
1e-20
1e-19
1e-18
08/01 08/02 08/03 08/04 08/05 08/06 08/07 08/08
SN
R
1e-21
1e-20
1e-19
1e-18
08/08 08/09 08/10 08/11 08/12 08/13 08/14 08/15
SN
R
1e-21
1e-20
1e-19
1e-18
08/15 08/16 08/17 08/18 08/19 08/20 08/21 08/22
SN
R
1e-21
1e-20
1e-19
1e-18
08/22 08/23 08/24 08/25 08/26 08/27 08/28 08/29
SN
R
1e-21
1e-20
1e-19
1e-18
08/29 08/30 08/31 09/01 09/02 09/03 09/04 09/05
SN
R
1e-21
1e-20
1e-19
1e-18
09/05 09/06 09/07 09/08 09/09 09/10 09/11 09/12
SN
R
1e-21
1e-20
1e-19
1e-18
09/12 09/13 09/14 09/15 09/16 09/17 09/18 09/19
SN
R
1e-21
1e-20
1e-19
1e-18
09/19 09/20 09/21 09/22 09/23 09/24 09/25 09/26
SN
R
tuning or unlock stateh at 1 kHz
h at 100 Hz
Sensitivity History of Data Taking 6
0 10 20 30 40 50 60 70 80 90 100N/O
Obs
HDAQ Data Analysis and Monitor (ver 2.10), Run : 107, File : 1200 Start : 52155 MJD, Sep 02, 2001, Sun, 09:35:21 JST, Current : 52155 MJD, Sep 02, 2001, Sun, 11:23:29 JST
Ob
serv
atio
n
S
tatu
s
Obs.
0 10 20 30 40 50 60 70 80 90 1000
0.10.20.30.40.5
Dar
k P
ort
P
ow
er 0.07192 [V]0.0117 [V
rms]
0 10 20 30 40 50 60 70 80 90 100
100
Op
en−L
oo
p
Gai
n
1.676432.3 [deg]
0 10 20 30 40 50 60 70 80 90 10010
−21
10−20
10−19
10−18
No
ise
Lev
el (
935H
z)
9.762e−21 [1/Hz1/2]
0 10 20 30 40 50 60 70 80 90 100
0
1
2
Exc
ess
(G
auss
ian
ity)
Time (min.)
0.2223
On-line analysis (2)
— On-line monitor —
• One of the monitor screens for HDAQ
· Obs. status, power, servo gain, averaged noise level, excess.
12
0 5 10 15 20100
102
104
Co
un
ts
Excess
c2=1(F.A. <10ppm)
DT6 total
Theoritical distribution(Simulated Gaussian noise)
0 1 2 3100
101
102C
ou
nts
Excess
c2=1(F.A. <10ppm)
DT6 total
Theoretical distribution(Simulated Gaussian noise)
Quiet period
(nornalized)
c2=0.19(F.A. 3%)
(10hours, Aug. 06 night )
Interferometer diagnosis (6)
— Gaussianity evaluation —
• Excess (Gaussianity)
— evaluated in every 1 min.
· Gaussianity parameter
c2 : 1 (F.A < 10ppm)
— 14.0%.
· Quiet hours
· Threshold F.A.: 3%
— 11.3%.
→ Slightly larger
Consistent distribution.
⇓· Investigation with
the other signals.
9
N.KandaMiyagi Univ. of Education
Detector evaluation: Veto Analysis
HDAQ signals (DT4)
Y`i\c4 L-��
Y`i\c6 ��]gRK w���
Y`i\c9 d%V%�¦��
Y`i\c: ¢¶�¨��
Y`i\c; L+��
Y`i\c= X%T^%[K �����
<���K��Q¬D��>
FFT
,G(f)
<h(f)300Hz>
¦�
300Hz
¦�
<h(f)700Hz>
700Hz
<h(f)1kHz>
¦�
1kHz
vetovetovetoveto
<h(f)300Hz>
¦�
300Hz
¦�
<h(f)700Hz>
700Hz
<h(f)1kHz>
¦�
1kHz
Pass
«~ $ y
��
(5.7#5.8��)
V(f)
V(t)V(t) V(t) V(t)
¹¡K�£(5.4��)&���B®p£F "?M'GDM·u(CK·uJ§EMZ*WQ��DM)"
���K��K©j(5.3��)
S1frame = Σ{V(t) } K¦�²¯ 2
V(t)
h(f) ´��J <h(f)300Hz>#<h(f)700Hz># <h(f)1kHz>Q��(5.5��)
10HzªqF³�DM
h(f) KUb]
log
log1/2Hz
Hz
300Hz 700Hz 1kHz
1]d*_µJS1frame = Σ{V(t) } K ��
1]d%_=3.2768s
V(t)
s
V(t)
s
|�±K2�¡K ¸QGM
}�
¦�
S1frame
��²¯J�A
¦�
S1frame
Nσ Nσ
¦�
S1frame
ln
¦�
S1frame
ln
����¤�M°@
xth
{�JL®¥{�JL®¥{�JL®¥{�JL®¥
¦�
S1frame
��²¯J�OI@
V(t)
2
図5.18:
veto解析の流れ
47
26x10-6
24
22
20
18
16
S1frame[V2]
500040003000200010000
N
図 5.5: ADCチャンネル 4·地面振動信号の振幅 2乗和 S1frame 縦軸は振幅 2 乗和 S1frame[V
2]、横軸は計算したフレーム数 (×3.2768 秒)を表している。
図 5.6: ADCチャンネル 4·地面振動信号の振幅 2乗和 S1frameの度数分布 縦軸は度数、横軸は振幅 2乗和 S1frame[V
2]を表している。赤実線はガウスでフィッティングしている。この分布は正規分布に従うと見なす。
37
80
60
40
20
0
[%]
5654525048464442
run
AND
ch2
ch3
ch4
ch6
図 5.26: 干渉計の状態を表す信号別に表した 1kHz付近の < h̃(f) >1kHz 分布の µh1k±5σh1kの範囲外での雑音除去率 縦軸は信号の取り除かれた割合、横軸は run番号である。ch2は実験フロアの音響、ch3はレーザー強度、ch4は地面振動、ch6はダークポートの干渉光強度信号、ANDは干渉計の状態を表す信号の条件すべてを合わせた場合を表す。取り除く前の < h̃(f) >1kHz
分布の µh1k±5σh1k 以外の度数を 100[%]とした。
図 5.27: veto前後での < h̃(f) >700Hz の度数分布の変化白抜きのグラフがデータを veto前、網掛けのグラフが veto後の度数分布。縦軸は logスケールで、赤実線はガウスにフィッテイングしてある。平均値より離れたデータほど雑音が取り除かれていることがわかる。
53
60
50
40
30
20
10
0
SN
R
10-12 3 4 5 6 7 8
1002 3 4 5 6 7 8
1012
Expected SNR to binary mergers at 10kpc
[Msolar]
2001/8/3DT6
2001/6/2
2000/9/4DT4
mass of a member star
05
10152025303540
08/01 08/02 08/03 08/04 08/05 08/06 08/07 08/08
SN
R
05
10152025303540
08/08 08/09 08/10 08/11 08/12 08/13 08/14 08/15
SN
R
05
10152025303540
08/15 08/16 08/17 08/18 08/19 08/20 08/21 08/22
SN
R
05
10152025303540
08/22 08/23 08/24 08/25 08/26 08/27 08/28 08/29
SN
R
05
10152025303540
08/29 08/30 08/31 09/01 09/02 09/03 09/04 09/05
SN
R
05
10152025303540
09/05 09/06 09/07 09/08 09/09 09/10 09/11 09/12
SN
R
05
10152025303540
09/12 09/13 09/14 09/15 09/16 09/17 09/18 09/19
SN
R
05
10152025303540
09/19 09/20 09/21 09/22 09/23 09/24 09/25 09/26
SN
R
tuning or unlock state10-10 Msolar
1.4-1.4 Msolar0.5-0.5 Msolar
Expected SNR history for typical GW event at 10kpc away
Histogram of Expected SNR for DT6
0.5Msolar-0.5Msolar
10Msolar-10Msolar
1.4Msolar-1.4Msolar
Total Timetime expected SNR > mean - 1s
expect. SNR expect. SNR
expect. SNR
entr
y by
bin
(cal
cula
ted
for
1 m
in.)
entr
y by
bin
(cal
cula
ted
for
1 m
in.)
80.51%82.37%83.32%
for 0.5-0.5Msolarfor 1.4-1.4Msolarfor 10-10Msolar
-> Efficent operation status is more than 80%
N.KandaMiyagi Univ. of Education
GW search: Binary inspiral
Matched filterOne step search
Hierarchical search
Wavelet
Resampling
講演者: 神田展行 (宮城教育大 ) 日時: 12/8/2001 シート12
連連連連星星星星のののの出出出出すすすす重重重重力力力力波波波波 (2)
-15
-10
-5
0
5
10
15時
空の
歪み
(相
当)
15s1050[sec]
20
10
0
-10
x10-1
8
18.0017.9817.9617.9417.92s
拡大
講演者: 神田展行 (宮城教育大 ) 日時: 12/8/2001 シート18
ママママッッッッチチチチドドドドフフフフィィィィルルルルタタタターーーー法法法法 (2)
20
10
0
-10
-20
x10-1
5
18.0017.9517.9017.85
20
10
0
-10
-20
x10-1
5
18.0017.9517.9017.85
20
10
0
-10
-20
x10-1
5
18.0017.9517.9017.85
20
10
0
-10
-20
x10-1
5
18.0017.9517.9017.85
信号
ベストマッチ (黒線が予想波形)
Matched filter• Detector outputs:
: known gravitational waveform (template): noise.
• Outputs of matched filter:
• noise spectrum density
• signal to noise ratio • Matched filtering is the process to find optimal
parameters which realize
s t Ah t n t( ) ( ) ( )= +h t( )n t( )
ρ ( , , , . . . )~ ( )
~( )
( )
*
m m ts f h f
S fd fc
n1 2 2= z
max ( , , ,...), , ,...m m t c
c
m m t1 2
1 2ρFH IK
SNR = /ρ 2
Post-Newtonian approximation
( )nS f
Matched filtering analysis
tRead data
FFT of dataApply transfer function
Conversion to stain equivalent data
overlaps
Evaluate noise spectrum near the data( )nS f
1 2( , , )ct m mρ
max ( , , , .. . ), , ,...m m t
cc
m m t1 2
1 2ρFH IK
Event list
2 2c ct t
c c ct t t∆ ∆− ≤ ≤ +∆ ; 25ct ms
52 sec
00.7.1
10
DAQ2: Preliminary Result of Binary SearchPreliminary Result of Binary Search
with SNRthreshold = 7.2 (which corresponds to 6.2 kpc for 1.4-1.4Msolar event, 2.9 kpc for 0.5-0.5Msolar event, optimal incident direction and polarization),
2 events survived / 2.5 expected background-> 0.59 events/hour in C.L.90%
1
10
100
Number of events/bin
80604020
SNR2
background expect NBG = 2.5
χ2 < 2.5 χ2 < 1.5 χ2 < 1.0 fitting to χ2 < 1.5
prelim
inary
2/ρ χ)
Log
10[N
umbe
r of e
vent
s]
12.7
Discussion•Coalescing compact binary search of TAMA300, 1000 hours data and LISM data, are progressing.
•We have not observed events, which significantly exceed the threshold in both TAMA and LISM’s independent analysis.
•Even in the case if there are no significant events, we can still obtain upper limit to the event rate in the data using e.g. Poisson statistics.
•With 1000 hours data, we will be able to set an upper limit ~0.004 events /hours
c.f.: Caltech 40m : 0.17/hours (90% C.L.)
TAMA DT2 : 0.59/hours
TAMA DT4 : 0.020/hours
01.12.10
20
Wavelet
testing code (discrete)
expected SNR evaluation
(H.Yakura, N.Kanda)
11/9/97 3
TAMA group / Department of Physics, Miyagi University of Education Nobuyuki Kanda
A principle of the Resampling method
for GW from binary starts
To recognize GW as a sinusoidal wave, distort time axis according to the GW frequency changing.
-> “Re-sample” ADC sampling data
dt = omega(t) / omega_cutoff
resampling interval : dt
omega(t) : frequency prediction of GW at t
(-> template of GW)
omega_cutoff : cutoff frequency of resampling
• must be chosen less than coalescence frequency
• in current study, we choose cutoff as500 Hz in Keplar Motion in this analysis ( near by 1000 Hz in GW )
"resampling" points
constant sampling interval
distort along the chirp freq.
11/9/97 13
TAMA group / Department of Physics, Miyagi University of Education Nobuyuki Kanda
Check3. How many templates we need?
6x10-21
5
4
3
2
1
exc
ess
heig
ht
in F
FT
-20x10-3
-10 0 10
template arrival time parameter [sec]
4
6
81
2
4
6
810
2
4
6
8100
2
S/N threshold
arrival time interval
Typical interval: ∆t0 ±5msec (or =itaration interval/5msec)
01.12.10
18
Continuous GW from1987A remnant
(K. Soida)
01.12.10
19
(K. Soida)
N.KandaMiyagi Univ. of Education
New issue : Coincidence
Coincidence between two or more detectors !Event candidates list comparison
coherence of ρρρρ(t) = (signal | GWtemplates)
constraint of arrival time, mass
assumption of waveform
correlation of full time series data h(t)
01.12.10
21
TAMA-LISM coincidence
Coincidence between LISM (20mIF in Kamioka mine) and TAMA300
Distance: 220 km (maximum delay for arrival time : 0.73msec)
LISM:
h ~ 8 × 10-20 [1/√Hz]
8/1 - 8/23/ 2001
9/3 - 9/17/2001 total 777 hours
TAMA LISM data reading data reading
Matched filter Matched filter
TAMA event list LISM event list
keep the events in the common lock parts
TAMA event list LISM event list
for common lock parts for common lock parts
coincident event search
TAMA-LISM Analysis Algorithm
2,, ,c lism lism lism lismlismt M η ρ χ2, , , ,η ρ χctama tama tama tama tamat M
• Data length analyzed~ 121 hours
)),,((max 21,2,1
…… ctmm
tmmc
ρ
2 2c ct t
c c ct t t∆ ∆− ≤ ≤ + 3.27ct s∆
)),,((max 21,2,1
…… ctmm
tmmc
ρ
2 2c ct t
c c ct t t∆ ∆− ≤ ≤ + 3.27ct s∆
Results of coincident event search
TAMA LISM 65672 events 56725 events
After -veto31 events
After -veto3 events
After -veto0 event
ct
, ,ct ηM
, , ,ct η ρM
Results of onestep search for common lock parts
N.KandaMiyagi Univ. of Education
LIGO-TAMA
International cooperation
How about combined performance ?stability
fake rate estimation
Where is promising search ?mass region
kind of sources
N.KandaMiyagi Univ. of Education
Summary
We prepared, and developed IFGW detector
analysis issues: IF evaluation from the view of event detector,
Event search (observational limit, experience of
real data)
We proceed to realistic event detection:Coincidence