(Our) pulsar search technique
G. Petit (BIPM), E. Davoust (OMP), T. Fayard (CNES/GRGS)
Workshop pulsars 3-4 Mai 2007
Résumé • Acquisition system • Search for MSPs • “Standard” (survey) mode search: processing technique and sensitivity • “Multi-scan” (cluster) mode search: processing technique and sensitivity • Prospects
Workshop pulsars 3-4 Mai 2007
Acquisition system • RT signal 1-bit sampled at (presently) 100 Mb/s
• ADLink board passes the data on a PC hard disk
•V2004: one 25-MHz frequency band, one polarization (50 Mbps), operating from October 2004 to April 2006. •V2006: one 50-MHz frequency band, one polarization (100 Mbps), operating since July 2006. •Next upgrade: one 50-MHz frequency band, two polarizations (2x100 Mbps) => Same processing requirements, improved sensitivity. Workshop pulsars 3-4 Mai 2007
Two strategies to search for MSPs •
“Standard” (survey) mode: a 540 square degree area in the galactic plane – Longitude 20-110°, based on DM criteria, Latitude +/- 3° – 160 s obs, scan at set declination, 50 MHz bandwidth – About 22000 points, 1200 RT hours – Detection sensitivity of order 1.5 mJy
•
“Multi-scan” (cluster) mode: repeated observations in selected clusters – preferably not visible from Arecibo, close, with high stellar density, with no pulsar detected so far, with high DM . – Typically 4 consecutive 160-s observations – Detection sensitivity of order 0.8-0.9 mJy – Also used to re-observe survey candidates
Workshop pulsars 3-4 Mai 2007
“Standard mode” search: processing technique (1) • 160 s => 2 Go/obs • 160 correlations sums (i.e. 160 312.5 kHz bands) computed for 5x106 32 µs intervals => 800Mo/obs passed to processing computer(s). • The processing realizes a coherent folding for (of order) 106 trial periods, 100 trial DM, 10 trial Pdot, 10 pulse phase => 1010 elements in the search space. • Computing needs: – Using multi-level folding => (of order) 1012 operations, 10 hours of standard CPU, for ms PSR search down to 0.65 ms. – 1 hour on RT ≡ 7 CPU x days. – 2-3 Go RAM per process would vastly decrease the disk access.
Workshop pulsars 3-4 Mai 2007
“Standard mode” search: processing technique (2) • 1010 elements in search space => 1010 results (spectral power) with observed statistical dispersion (standard deviation σ). • Results with power close to 7σ (and above!) identified for closer inspection: Graph in (P,DM) coordinates
Validation on known pulsar: detection of J1807-2459A
Workshop pulsars 3-4 Mai 2007
Validation of the sensitivity in “standard mode” • •
1937+21 in 60 s: average SNR = 33, median SNR = 29. Implies SNR=7 for a 1.5 mJy pulsar, in one 160 s observation
1937+21 SNR for 60 s * 50 MHz: Ave=33.6 60 50 SNR
40 30 20 10 0 2006.5
2006.7
2006.9 Year
Workshop pulsars 3-4 Mai 2007
2007.1
2007.3
“Multi-scan mode” search: processing technique (1) • Increase integration time to improve sensitivity. • One single long integration (as in standard mode) not optimal: – With constant period, processing needs increase about as T3/2. – Searching for binary pulsars (variable P) becomes impractical.
• Chosen procedure: – – – –
Several (≥ 4) consecutive 160-s observations Independent processing of each 160-s observation. Matching of candidates peaks between all observations. Basic matching is easy for constant period, not so easy for binary search => use the Hough transform (see Aulbert arXiv:astroph/0701097v1)
Workshop pulsars 3-4 Mai 2007
“Multi-scan mode” search: processing technique (2) The Hough transform (figure from Aulbert, 2007) • •
Search for (a,b) in yi = axi + b, 1≤ i ≤ n Transform the problem to finding intersection of n lines b = axi - yi • In our case: xi is time (sequence of observation) yi is observed Period a is dP/dt b is P0, period at reference time • •
the
Each candidate peak for each observation yields one segment in the (a,b) plane, with slope = the observation number and some SNR. A cumulated SNR is computed for all points in the (a,b) plane. Highest SNRs are identified. Workshop pulsars 3-4 Mai 2007
Validation of the technique in “multi-scan mode” PSR 1744-24A 63.5
•
•
63
62.5 62
61.5 61 0.6
0.7
0.8 0.9 Binary phase
1
1.1
PSR 1744-24A detection in "cluster mode" 39 37 SNR over 3*160 s
•
1744-24A observed in 18 160s observations: typical SNR = 20 on each observation. “Multi-scan mode” processing based on Hough transform is applied on 16 groups of 3 observations. Method successful despite large Pdot values (one of the largest among all ms PSRs). Detection SNR in “multi-scan mode” consistent with the expected addition of SNR, except when the variation of P is strongly not linear (when the binary phase is 1.0 or 0.5).
P-11500 / µs
•
35 33
Eclipse begins
31
P not linear
29 27 25 0.6
0.7
Workshop pulsars 3-4 Mai 2007
0.8 0.9 Binary phase
1
1.1
Validation of the sensitivity in “multi-scan mode” • •
Several known pulsars detected in “multi-scan mode” “Observed flux” in Table below assumes SNR=8 for 0.8-0.9 mJy SNR
“Observed flux” /mJy
Notes
2.8
9
0.9-1.0
Very variable
11.563
2.5
30-35
3-3.5
Outside eclipse
J1807-2459A NGC6544
3.059
1.1
8.4
0.8-0.9
B1821-24
NGC 6626
3.054
1.1
8
0.8-0.9
B1620-26
NGC 6121
11.076
1.6
19
1.9-2.1
B1820-30A
NGC 6624
5.440
0.7
7.5
0.8
J1701-3006A NGC 6266
5.242
0.4
ND
-
Pulsar
Cluster
J2051-0827
-
4.509
J1748-24A
Terzan 5
•
Period Expected /ms flux /mJy
Implies SNR ≥8 for a 0.8-0.9 mJy pulsar, in four 160 s observation
Workshop pulsars 3-4 Mai 2007
Prospects • Dedicated acquisition system and processing techniques validated • Two search programs under way: – wide survey in the galactic plane: 22000 points, 18 % completed • No outstanding candidate so far
– search in some 15 selected clusters • Two candidate signals, will be reobserved
• Detection sensitivity (about independent of period and DM) – wide survey: 1.5 mJy – cluster search: about 0.8 mJy
Workshop pulsars 3-4 Mai 2007