Transcript Get PPT file

OVSA Expansion
Pipeline Imaging
Log-spiral array: “uv” distribution
Left: sampling of
spatial Fourier plane
Right: inner region of
sampled spatial Fourier plane
The imaging problem
• EOVSA will be imaging over probably the widest range of
spatial scales of any radio telescope: e.g., at 18 GHz,
largest scale is 2000”, nominal resolution 3” requires < 1”
cells, range is 2000. By contrast, in a typical VLA
configuration the range is 20-40; Nobeyama
Radioheliograph was about 200.
• I like to think about the problem in terms of the number of
pieces of information we measure versus the number of
resolution elements in each image: the more information per
resolution element, the better the images will be.
• At 18 GHz, there are of order 105 resolution elements if the
resolution is 3”, while a 13-element array has 78 visibilities
in a snapshot: factor of 1000 short.
Sensitivity
TTA/(N(t))
TA is 10000 K for 2 m dishes
N = 13,  = 0.12 MHz, t = 0.02 s
T  20 K
EOVSA configuration
Beam
5 GHz
Imaging models: have 3
frequency-dependent components:
1. Coronal
emission
derived from an
EIT image (2.6”
pixels)
2. Chromospheric disk
derived from
Zirin et al.
3. Optically
thick gyroresonance
sources derived
from an MDI
image.
Actual VLA Radio Images
VLA radio images on 1999 April 11. The 5 GHz image is a mosaic
of 23 different fields and took all day to make (resolutions 40”/12”)
Imaging deconvolution
• The raw product of an EOVSA observation is a set of
“visibilities”: complex numbers representing the spatial Fourier
transform of the sky brightness distribution on each baseline
between pairs of antennas in the array.
• These spatial Fourier terms can be combined to create an
image (the “dirty map”) that represents the sky brightness
convolved with the point source response of the telescope
(the “dirty beam”). Deconvolution is the process of removing
the effects of that convolution.
• Two most commonly used methods in radio astronomy:
• “CLEAN”: go through the image sequentially removing
fractions of the dirty beam at the locations of peaks in the
dirty image. Represent map as collection of point sources.
Works best for compact sources.
• Maximum entropy: determine a model that, when
convolved with the dirty beam and differenced with the dirty
map, maximizes some measure of entropy (smoothness).
Works best for large smooth sources.
Maximum entropy does a poor job on bright compact features
(left): better technique is to clean bright emission, subtract it
from data, run MEM on residual, then restore bright emission
(right).
EOVSA real-time pipeline
• Snapshot u,v coverage can image compact flares but not
solar disk
• Full 8-hour synthesis can produce a reasonable map of the
full disk (at lower frequencies, probably not at higher)
• General idea is to “maintain” full-disk images at each of
some set of frequencies, look for differences in real-time
data from these default images
• Procedure for default synthesis images is to first run
CLEAN to remove bright compact features, then MEM to
get smoother structure
• To make a full synthesis, need to handle solar rotation: this
needs a new routine (not currently possible in any package)
• For weak flares you would “subtract” preflare model from
visibilities before mapping flare.
• MIRIAD (mostly FORTRAN) is the current deconvolution
package in use, need to exploit multi-processor architecture
EOVSA imaging simulation
EOVSA simulations: residuals
EOVSA analysis …
• “Default” images serve as our normal real-time image
sequence – make them on some suitable time cadence,
completely automated, generate movies, etc
• Assume analysis code is in IDL, calibrated image
sequences at multiple frequencies are the default data
• Daily summary “light curves” as dynamic spectra
• Users can request image sequences for periods and
frequency range of interest
• Flux versus frequency, selected regions, polarization, etc1
• …..
1(Most
non-black-belt users will not do much else, eg, very
few users do anything quantitative with EIT)
Full FASR Log-spiral array configuration
Log-spiral array: dirty beam
Imaging procedures
• Mostly work with 5 GHz model (nominal 4” resolution) using
2048x2048 image with 1.2” cells
4096x4096 dirty beam
Model converted from brightness temperature to Janskies for
deconvolution steps, then back to K for analysis
Threshold for fidelity measure: 3000 K
• For CLEAN, subtract model disk from visibilities first, then
restore at end (can also “fit” disk)
• For MEM, use model disk as default image
• “Weighting” of the visibilities (we regrid the data onto a regular
grid in order to exploit FFTs: relative weights ascribed to the
gridded data can be based on local density, etc) plays an
important role in point source response because of the wide
range of spatial scales sampled.
• Simulations here in MIRIAD package (CARMA/ATNF/SMA),
earlier in AIPS package (NRAO).
Spatial resolution versus frequency
• Scattering in the solar atmosphere limits the achievable
resolution:
Scattering angle ~ 2
3 arcmin at 500 MHz, 15 arcsec at 1.5 GHz, 2 arcsec at 5
GHz
• At low frequencies (below 2 GHz) arcsecond resolution
is not possible. This means that we do not need to
design for high resolution at low frequencies.
• This also puts an upper limit on the number of antennas
we need: N antennas give N(N-1)/2 pieces of
information. Ideally we have one piece of information for
each resolution element. E.g., at 1 GHz resolution ~ 25”,
6400 resolution elements fill disk, 113 antennas give this
many baselines.
• At high frequencies finances will prevent us from
adequate sampling: 1” beam => 4. 106 res el => 2800
antennas
Antenna size
• Need to be able to image the full disk of the Sun from
100 MHz to 20 GHz: range of 200 in frequency
• A single antenna of diameter D meters has a field of
view
FWHM = 1030 arcmin / D (m) / freq (GHz)
•
•
•
•
at frequency f GHz. The Sun is 30 arcminutes across.
At 20 GHz need D < 1.7 m
At 2 GHz need D < 17 m
No single dish can handle the wide frequency range: a 2
m dish is not sensitive enough at lower frequencies.
This drives us to using three separate arrays to cover
the frequency range.
Frequency dependence of solar radio flux and of disk brightness
temperature
Solar TB diluted by primary beam size
Imaging measures
• The conventional measure used in imaging simulations is
“image fidelity”: (image – model)/model (many people use
the inverse; here small numbers mean good imaging
fidelity).
• We implement this by looking at:
the distribution of
(image – model)/model
for all pixels exceeding a threshold in the model
• We calculate the mean of the absolute value over all ratios,
the median of the absolute value, and the standard
deviation of the differences.
• Eventually we will calculate these measures for different
spatial scales within the image.
Coronal magnetography
• Need to cover 2-20 GHz.
• Assume that we have 4 × 500 MHz IFs, each 2
polarizations, covering 2 GHz at a time: need 9 settings
to cover frequency range.
• 5 second time resolution would give 0.5 seconds per
setting
• 32 MHz spectral resolution is about 10 G magnetic field
resolution: roughly 500 channels from 2-20 GHz.
• Need images of high quality to be able to use spatiallyresolved spectroscopy.
Synchrotron emission from flares
• Trigger flare mode when flux on compact spatial scales
exceeds specified level
• Time resolution of prime importance: bright emission
confined to small spatial region.
• If we have 4 × 500 MHz IFs, each 2 polarizations,
covering 2 GHz at a time, then 0.05 second integrations
at each setting gives 0.5 second resolution.
Solar disk radio brightness temperatures
Disk brightness temperatures:
600000 K at 0.3 GHz - optically thick corona
150000 K at 1 GHz
- optically thin corona
20000 K at 3.75 GHz - upper chromosphere + thin corona
13000 K at 9.4 GHz
- upper chromosphere
10000 K at 17 GHz
- chromosphere
7000 K at 86 GHz
- chromosphere
Time sharing and time resolution
• Assuming a correlator with finite instantaneous bandwidth,
need different observing modes to handle different science
goals.
• This forces trade-offs between spectral coverage and time
resolution
Coronal
magnetography
High spectral resolution from 220 GHz, low time resolution, high
image quality, continous
Synchrotron in flares Low spectral resolution, high time
(> 3 GHz)
resolution (1 second), only during
flares
Low frequency radio High time and spectral resolution
bursts (< 1 GHz)
all the time
Low frequency radio bursts
• To maintain high temporal resolution continuously, need 2 ×
500 MHz IFs, each 2 polarizations, constantly dedicated to
100-1100 MHz.
• 1 MHz spectral resolution would be nice, 4 MHz would get
most of the science.
• Will probably only have sensitivity to see coherent bursts at
this resolution.
FASR Instrument (Antennas)
Three arrays, 3 km baselines for HFA,
6 km baselines for LFA and LPDA
Array
Designation
Number of
Antennas
Frequency
Range
Antenna
Size
HFA
~100
2-24 GHz
2m
~60
0.2-3 GHz
6m
20-300 MHz
Logdipole
High Frequency Array
LFA
Low Frequency Array
LPDA
Log-Periodic Dipole Array
~40