I.3. The Tianshan Radio Experiment for Neutrino Detection
Olivier Martineau-HuynhTREND workshop,
April 19, 2013
Why TREND?Why Ulastai?
TREND-21CMA site
Beijing
Urumqi
Ulastai
• Ulastai, Tianshan mountains, XinJiang autonomous province• Site of the 21CMA radio interferometer (Epoch of Reionization)
AM emitters : forbidden zone
Electronics noiseAntenna measurmentGalactic emissionmodel
Background electromagnetic level ≡
Galactic emission
21CMA • Interferometer of 10000
antennas • Built in XinJiang by Wu
XiangPing (NAOC) in 2007• Obj: study of the Epoch of
ReIonization (1st stars)
N
4 km
E
S
W
3 km
The 21CMA setup
• Signals from 127 antennas added after phasing
N
E
S
W
1 pod of 127 antennas
DAQ room
• Transfer to acquisition room through optical fiber (2 fibers/ pod)
• Sampling at 200MSamples/s & online FFT & correlation
Setup, infrastructure & support fully available for TREND!
TREND setup & DAQ
The TREND-50 setup• 50 antennas deployed in summer-automn 2010, total surface
~1.5km² (+ 3 scintillators).• Stable operation since March 2011.
TREND-50~1.5 km²
East arm: antennas 101-120West arm: antennas 121-138, 140South arm: antennas 148-158Scintillators: 139, 175(@pod S15), 176 (@podS16)
TREND acquisition chain
Filter 50-200MHz
LNA 24dB
40dBFilter 50-100MHz
Optical transmitter
buffer 200MB
disk
antenna
~100m coax cable
<4km fiber Optical recevier
ADC 8b 200MS/s
Electronic boardPod
DAQ room
TREND acquisition chain
fiber to the DAQ room
(<4km)
64dB ampli+ 50-100MHz
filter
Optical receiver
Antenna 101
U183300G disk
Triggered events Data file
u101
Time fileu101
scans ADC
8bits200
MS/s
Circular buffer
200MB
Circular buffer
200MBcomputeru101
CPU(trigger)
64dB ampli+ 50-100MHz
filter
Optical receiver
Antenna 158
Triggered events Data fileu158
Time fileu158
scans ADC
8bits200
MS/s
Circular buffer
200MB
Circular buffer
200MBcomputeru158
CPU(trigger)
50 parallel & identical chains
Optical fiber100m<d<4km
1 triggered event = 4 words in time file &
1024 samples in data file
DAQ room
computeru203
(ADCs init)
start signal
50 parallel & identical chains
Buffer structure
• 2 buffers on each machine u101 – u158.
• 1 buffer is 200MB (ie 228 bytes).
• ADC flow synchronous on all machines thanks to start
signal sent simultaneously to all ADCs by machine u001.
• ADC running at 200MSamples/s (5 ns per sample)
Buffer = 1.34 s of data
• If buffer full, ADC flow to second buffer.
• When buffer fully analysed (see next slide), buffer cleared.
• If buffer full before 2nd one cleared, both are dumped.
Buffer 200MB(268 435 456 samples)
1.34 s of data withADC @ 200MSamples/s
subbuffer n°1 (1024 samples)
subbuffer n°262 144
subbuffer n°262 143
subbuffer n° 2 (1024 samples)
Buffer structure
Trigger principle / Buffer analysisT0:On each machine uNANT (NANT in 101-158):
– For every new buffer, compute snoise over 1024 samples of 1st sub-buffer.
– If sample i with amplitude Ai > N x snoise , trigger of level T0 on this antenna. 6<N<10.
T1:If at least 4 antennas have T0 triggers within a causal time window (Dt<DL/c), then data is sent to disk on machine u183 for all antennas with T0s
– to data file R[NRUN]_A[NANT]_data.bin: 1024 samples centered on sample i
– to time file R[NRUN]_A[NANT]_time.bin : • UNIX second of sample i• Buffer index of sample i• Subbuffer index of sample i• Position of sample i in subbuffer.
5
snoise
N x snoise
Code in C with MPI for Master/Slaves dialogs. Master program in python. Securities to check computer & ADC card status, buffer status, machines dropping out of DAQ. Monitoring tools to check status.
TREND monitoring
Log files• Process log files to compute live DAQ time,
stalled DAQ & dumped buffers (BuildStat.m).
Power Spectrum Density• PSD |FFT|²
OK
Bad environment
No gain
Monitoring of the DAQ chain & environment…Also used for calibration.
Calibration 1: relative calibration
• Baseline calibration
– Hyp: senv >> selec
then sbline ~K senv => K a sbline at time t
𝐴𝐶𝑎𝑙=𝐴𝐴𝐷𝐶
𝜎𝑏𝑙𝑖𝑛𝑒
DA/A = 15%
Absolute calibration(under development)
Use load measurement• PSDload: power spectum
density with input = 75Ω load.
• PSDref with input = antenna right after load.
• PSDcurrent with input = antenna at time t.
GdB (t) = PSDload + PSDcurrent (t) - PSDref
Calib comparison
OK but sometimes really off & not as good as relative callibration on plane tracks (~20%).We need to improve our calibration if we want to study shower later profiles.
Absolute callibration with calibrator
A solution for TREND? (balloon or helicopter)
Pierre CHAUVEAU’s internship (?)
TREND RADIO ENVIRONNEMENT
Local sidereal time
Sign
al n
oise
leve
l
Galactic plane @ 408 MHzMajor radio source: thermal emission from the Galactic plane.
Visible in Ulastai sky between 15h & 23h LST.
TREND antennas clearly exhibit an increased noise level when the Galactic plane is in the sky
TREND offline data treatment
Radio background
TREND antenna
TREND-50 antennas radio array:2011-2012 data220 real days analysed
1.2 1010 triggers recorded1.4 109 coincidences ~100Hz background event rate over whole array (physical origin)
Expected EAS trigger rate:~2 events/day for E>1018 eV
Background rejection is a key issue for EAS radio-detection
(& major goal for TREND phase 1)
EAS signal Background
Ex: signal pattern:
Elm background point sources
• Distinct features compared to EAS radio signals.→ Localized background can be rejected through data processing.
Shower axis
Background discriminationEAS radio signals features:- Focused signal spot on ground (decreasing
when moving away from shower axis)- ~ Flat wavefront- Short (<500ns) pulses- Isolated pulses- Random in time- Random in direction- Come from the sky- Polarized ( shower direction & Bgeo)- Frequency signature (?)
Background radio signals features- If distant: ~constant amplitude
- If close: curved wavefront - Usualy long pulses- Usually repetition- Possible pattern for consecutive triggers (50Hz)- Point sources or trajectories- Mostly from ground- Non polarized- Frequency signature possible…
EAS and background have distinct features → data processing written to efficiently select candidates & reject background
DATA PRE PROCESSING : STAGE 1
• Informations on run set-up (antennas, positions, delays….)
• Rejection of « empty » events (bug DAQ, corrected 02/2012)
• Identification of coincidences between antennas
Time (s)
Coin
cide
nce
rate
(Hz)Example of run: 3577
- 49 antennas- Total duration: ~38h30- Number of triggers: 4.452.938- Number of coincs: 633.918
DATA PRE PROCESSING : STAGE 2• Main source of pollution: power line
• Typical signature: Δt between 2 events = n.10 ms
• On run 3577: 67.406 coincidences remaining (89% efficiency)
• Influence on acceptance? (mainly cross-point events, estimated dead-time ~10%)
10 ms
Time (s)
Coin
cide
nce
rate
(Hz)
Could be implemented as an on-site pre-treatment (C program)
reduce data volume & ease up offline treatment
DATA PRE PROCESSING : STAGE 2• Signal waveform analysis:
Create « boxes » around over-threshold parts of the signal
- Total ToT- Number of boxes- Boxes ToT- Pre-trigger ToT- Central box (at expected trigger time)- …
Informations available for signal rejection in stage 3
• Rejection of « bad » signals (if mult<4, coincidences are rejected)
• For run 3577, 22.244 remaining coincidences (66% efficiency)
Event reconstruction• Reconstruct wave associated with coincidences of 4+ antennas.• Plane (direction ,q j) & spherical (source x0, y0, z0) wave front
hypotheses.• Antenna trigger times corrected through signal inter-correlation
treatment.• “Delay plot” to check reconstruction quality.
• For all reconstructions, parameters saved into DST (Matlab file)
RADIO PERFORMANCES
• Plane track reconstruction : - 3037 events in 4 minutes- Θ > 60°- Max multiplicity: 40
Total angular resolution <1.5° on the track.(and improves with smaller zenithal angle)
mult ≥ 22 antennasσ = 0.7°
Reconstruction quality should be checked more systematically (in particular single antenna effect wrong delay correction)
DATA PRE PROCESSING : STAGE 3
• Track identification (Δθ/Δφ < 10°, Δt < 1m) and rejection• For run 3577, 432 remaining coincidences (98% efficiency)
Data pre-processing: stage 4
• For surviving events: – shower profile reconstruction– Data saved to DST (‘Candidates’ field)
• Final cuts (local treatment):– More selective cuts on signal shape– More selective cuts on neighbours– Cut on amplitude difference Amax/Amin>1.5– Cut on pattern (visual)
Final list of EAS candidates
Conclusion on data treatment
• Whole process consists in rejecting background.
• Trading between cut efficiency / candidate survival probability / data treatment power.
• This layout does not allow for efficient cut on pattern need for strict cuts on neighbours (very unefficient!)
EAS selection could certainly be optimized…
Effect on EAS radio signals to be studied with MC simulations (see
below)
TREND results & present activities
The TREND status
• 2009: 6 antennas protype– Test setup for principle validation.– 25 CR candidates detected.
TREND-15 setup (2010)• 15 log-periodic antennas + 3 particle detectors (same DAQ
algorithm).
400 m
800
m
Ardouin et al., Astropart. Phys 34, 2011 <arXiv:1007.4359>
First EAS identification with autonomous radio
arrayNants θradio θscints ϕradio
ϕscints
4 61±3 67±5 359±2 3±4
4 52±1 49±3 195±2 191±4
5 42±1 36±3 55±4 56±5
4 45±1 49±3 12±1 10±5
7 56±2 53±4 323±2 331±5
Some radio EAS candidates are coincident with scintillator coincidences + direction recons match!
Selection of radio EAS candidates with dedicated algorithm
Radio data(subset)
Reconstruction of 3-fold scintillator coincidences EAS
Scintillator data
• Independant trigger & analysis of scint data (EAS) & radio data (EAS radio candidates).
The TREND-50 setup• 50 antennas deployed in summer-automn 2010, total surface ~1.5km².
TREND-50~1.5 km²
TREND-15(2010)
TREND-6(2009)
All according to previously presented, except: Antennas 148, 150-152, 157: log periodic (21CMA type)Antennas 149, 153, 155: proto butterflyAntennas 148-155, 158: elec board @ pod
Stable operation between Jan 14, 2011 and Dec 6, 2012: 691 days8 data taking campaigns: 574 days (83%)DAQ running time: 320 days (56%) Data analyzed so far: 220 days (69%) [100% off-line treatment]Acceptance = 103 km².days 69 days with full array @100% efficiency (31%)
TREND-50 is a BAD & UNRELIABLE setup.Significant effort in the last 5 years on maintenance, with mitigated success…That is certainly the price to pay for a (nearly) cost-free detector!
TREND-50 EAS candidates search• 2011-2012 dataset:
– 140 EAS candidates found (2/ live day)
Valid candidate?
Statistical response
EAS candidate search• Observed distribution very different from background one.
All events EAS candidates
Deficit for j = 90 & 270°All events
West90°
Expected distribution
• Antennas sensitive to East-West polarization only (x axis):
EEAS . x skymap West90°
North0°
Experimental distribution
30°
60°
90°
30°
60°
90°
Looks promising, but requires better modeling: Full simulation of radio signal & system responseto produce expected EAS distribution.[on-going work]
South180°
East270°
A a EEAS .x a vBgeo .x
EAS detection simulation• Shower simulation (CONEX): FORTRAN
– All sky: 5°<q<85° (step:10°) & 0<j<360° (step=20°)– At least 2 energies: E=5 1017 & 1018eV.– At least 100 core positions per direction
2x8x18x100 = 28’800 shower simulations2 modes:– 1D pure longitudinal MC+ODE => fast (2-3 s/shower).– Full 3D MC => very slow @ ultra high energies (~2+ h / shower @ CCIN2P3 with intel compiler)…
Probably still room for tuning though).
• Radio signal simulation ~10 antenna signals per core position -> 288000 signal simulations
2 modes again:– 1D: EVA/FORTRAN or custom Python code. Fast : ~2-3 s/antenna– 3D: EVA/FORTRAN. Very slow (~3-4 h/antenna) BUT 3D is the only accurate method close to shower axis (~100 m) due to strong Tcherenkov effect.
• Antenna response (NEC2): FORTRAN+Python interface. Fast (2-3 s/antenna).
TREND simulation: toolsCode in FORTRAN/C++ + Python 2.6
1D option fast: ~10s / antenna 3D option very slow: ~3-4h/ antenna
CPU needs: • 28800 showers & 288000 antenna signals
Massive computing needs!
Go for distributed computing. France-Asia Virtual Organisation (IN2P3, IHEP, KEK,KISTI). DIRAC as middleware.
Major effort undergoing.
EAS candidate search
• Ultimate test for EAS detection: rotate antennas to N-S
Same background distribution (random polar) but EAS expected distribution radicaly different.
Antennas rotated December 2012.Background distribution unchanged.
EAS candidates search just began.
Validation will open the way for a detailed study of EAS radio characteristics. Valuable because first self-triggered sample ever!
Conclusion
• TREND-50 on the way to confirm EAS radio-detection with limited background contamination: 1st objective validation.
• Poor reliability, stability & callibration may be a major handicap for detailed & reliable study of EAS radio characteristics.