Unbiased AllSky Search Michigan as of August 17, 2003 D' Chin, V' Dergachev, K' Riles

1
Unbiased All-Sky Search (Michigan)as of August
17, 2003 D. Chin, V. Dergachev, K. Riles

• Analysis Strategy (Quick review)
• Measure power in selected bins of averaged
  periodograms
• Bins defined by source parameters (f, RA, d)
• Estimate noise level statistics from
  neighboring bins
• Set raw upper limit on quasi-sinusoidal signal
  on top of empirically determined noise
• Scale upper limit by antenna pattern correction,
  Doppler modulation correction, orientation
  correction
• Refine corrected upper limits further with
  results from explicit signal simulation

2
Overview of Data Pipeline for Unbiased All-Sky CW
Search
Raw Data
Simulated Data
Create 30-min SFTs
Creation of Power Statistic
Loop over frequency and sky Determine search
range and control sample range Determine upper
limit on detected power Apply efficiency
corrections (Doppler, AM, orientation Determine
limit on h0 and store
Determine efficiency corrections
3
Creation of Power Statistic and finding upper
limits
Raw Data
Simulated Data
LHO LDAS
1-minute, 30-min calibrated SFTs (Xavier)
1800-second raw data
2048(?)-second calibrated SFTs (Greg)
(Vladimir)
1800-second calibrated SFTs
(cross check)
Medusa system
(cross check)
Power Statistic Creation (Vladimir) Simplest
Average calibrated powers bin-by-bin Allow
summation of raw and simulated SFTs Apply epoch
vetoes (high noise, bad calibration, artifacts)
Michigan computers
Upper Limits Finder (Dave et al)
4
Upper Limits Finder (schematic)

• Loop over values of f0 (SSB frame), RA, sin(d) in
  steps of
• ¼ (17 mHz, 0.5 mHz)
• Determine the freq bin(s) of search and the
  large control range (nearly neighboring)
• Compare total power in bin(s) and compare with
  histogram to find upper limit (2s) on detected
  power in h
• Apply efficiency corrections (Doppler
  modulation, antenna pattern, worst orientation)
  to find 95 C.L. upper limit on h flux at earth
• Store upper limit

fDetected
Measure
Counts
Upper limit
Power (h2)
5
SFT Generation
Problem Need 30-minute SFTs but calibration
drifts non-negligible at times. Want a method to
use 1-minute calibration a coefficients with
minimal new artifacts.
Present baseline calibration procedure
(stitched) Create 1-minute SFTs (high-passed
Tukey-windowed Apply 1-minute calibration info,
window again in Fourier domain Compute inverse
transforms, window again, and stitch to make
30-minute interval Compute SFT from 30-minute
interval Machinery is in place with flexible
control of parameters Tukey window ramp
intervals High-pass and low-pass
filtering Strong-line suppression (mean-padding
in Fourier domain) Trouble Periodic windowing
introduces 1/60 Hz residual comb Optimum
tradeoff not yet clear Xavier working on
alternative approaches (averaging
Dirichlet kernel) Coordinating closely Stay
tuned
6
Flow chart for generating 30-minute calibrated
SFTs using the stitching method (Vladimir)
7
Validation of SFTs
Verified code works correctly for constant,
unfiltered calibration (comparing stitched SFT to
1800-sec SFT) Varying filter and Tukey window
parameters in time and freq domain, including up
to extreme of Hann window. Trying to find
reasonable tradeoff between data retention and
small artifacts from 1/minute modulation
(sharp windowing leads to 1/60 Hz residual comb
structure) Hann window defines asymptote for
smooth window behavior, but large data loss Tukey
window with sharp ramp (? rectangular window)
defines asymptote for high data retention, but
1/60 Hz artifacts
8
Validation of SFTs
Established that Tukey window with sharp ramp
preserves calibration line magnitude and phase
well But bigger worry is leakage of large lines
into spectral troughs Can see appreciable
(1-10) variation in trough noise, even with
aggressive filtering of low-frequency,
high-frequency, line noise Ultimate figure of
merit How well we can reconstruct
hardware-injected pulsar signals (hardware
injection automatically accounts for calib
variation) Work underway Also coordinating
closely with Xavier on his Dirichlet-kernel and
averaging methods
9
Detailed running history of SFT and power
statistic development with validation tests can
be found on Vladimirs home page http//tenaya.ph
ysics.lsa.umich.edu/volodya/
10
Sample Power Statistic (derived from 701 SFTs)
L1 200-400 Hz 60 Hz harmonics violin modes
prominent, but other artifacts present too

• Expect exponential distribution for power in each
  SFT bin.
• Kolmogorov-Smirnov statistic using median to
  define reference curve
• (but expect 0.02 for 701 points!)

11
Status of programs Program Author Status SFT
maker Vladimir Complete Under validation
(SFTs generated for H1, H2, L1) Power
statistic maker Vladimir Baseline complete
evolving (Statistic generated
for L1) Antenna pattern code Dave Complete -
LALified Freq modulation code Keith Crude
baseline complete Search engine Keith In
progress