250 likes | 491 Views
The CLEAN algorithm in the RHESSI software. Sa”m Krucker. Works well for point sources, but also works ok for extended sources Relatively fast easy to use. CLEAN approximates image with point sources using the RHESSI Point Spread Function (PSF)
E N D
The CLEAN algorithm in the RHESSI software Sa”m Krucker • Works well for point sources, but also works ok for extended sources • Relatively fast • easy to use • CLEAN approximates image with point sources using • the RHESSI Point Spread Function (PSF) • 2) Convolves the point sources with Gaussian (CLEAN beam; • PSF without side lobes)
RHSI point spread function PSF grid 5 grid 6 grid 7 grid 8 grid 9 1024” 1024” PSF is slightly different for different locations in image psf = o getdata( class_name=‘hsi_psf’, xy_pixel=xy ) psf = his_annsec2xy(psf, o )
Summing individual PSF gives PSF of reconstructed image FWHM for grid 3-9 9.8” 15.5” 7.1” UNIFORM weighting more weight to finer grids FWHM smaller NATURAL weighting same weight for all grids FWHM larger for comparison: finest single grid i.e. using grid 3 throu 9 in BACKPROJECTION and CLEAN gives a net resolution larger than for grid 3 alone.
default is UNIFORM weighting Relative weights (finest grid equals 1): 1, 0.57, 0.33, 0.19, 0.11, 0.06, 0.04, 0.02, 0.01 Not using coarse grids does not change much. Including fine grid without clear signal makes image noisy.
Unresolved source looks similar to PSF grid 8 only, 2002 July 23, 00:38 UT, 20 s, 40-100 keV
….. depending on counting statistics 5000 counts 20 000 counts 100 000 counts
PSF is used to subtract side lobes = -M* observations - max(obs)* PSF = RESIDUALS
PSF is replaced by clean beam Clean beam is GAUSSIAN with same FWHM as the PSF hsi_clean_beam_polar.pro
CLEAN BEAM is used to approximate PSF = -M* observations - max(obs)* PSF = RESIDUALS CLEANED image = max(obs)* clean beam + RESIDUALS + =M*
Problem for more complex source structures:maximum in image likely contains side lobes of other sources. SOLUTION: Only a fraction (CLEAN_FRAC) of the maximum M is subtracted at a time. The same process is repeated over several iterations.
CLEAN stops: • No more flux to clean, i.e. absolute maximum in residual map is negative • Or if keyword CLEAN_NITER is set, after CLEAN_NITER iterations
CLEAN parameters • CLEAN_FRAC: fraction of maximum that is subtracted (GAIN) default is 0.1 (good for point sources) for complex/extended sources lower values can be used (takes longer) • CLEAN_NITER: maximal number of iterations default is 33 (good/fast for first run) if not all flux is cleaned, increase clean_niter also: more iteration needed if clean_frac small
run CLEAN with default values im = o getdata( image_alg=‘CLEAN’) CLEAN_NITER = 33 CLEAN_FRAC = 0.1
CLEAN after 1 iteration (40-80 keV, g3-6, 20 s) Backprojection Residuals component map selected max CLEANed map Residuals component map selected max
CLEAN after 7 iterations Backprojection Residuals component map selected max CLEANed map Residuals component map selected max
CLEAN after 33 iterations Maximum number of iterations reached. Backprojection Residuals component map selected max CLEANed map Residuals component map selected max
run CLEAN with more iterations im = o getdata( clean_niter=111 ) To restart CLEAN from iteration 34 im = o getdata( clean_niter=111, /clean_more )
CLEAN after 38 iterations Absolute maximum is negative Backprojection Residuals component map selected max CLEANed map Residuals component map selected max
Select clean boxes im = o getdata( /clean_mark_box ) A plotman window is open and clean boxes can be selected.
With clean boxes after 65 iterations Backprojection Residuals component map selected max CLEANed map Residuals component map selected max
More keywords: • CLEAN_MORE_ITER: to run more iteration after clean stopped. Clean_niter must be increased. • CLEAN_MARK_BOX: allows to select clean boxes. Point sources are only selected inside these boxes. • CLEAN_NO_CH2=0: compares observed modulations with modulation from convolved point source model. chi2 test calculated • To run clean without displays: o set,clean_show_maps=0 o set,clean_progress_bar=0
BAD keywords: • CLEAN_MEDIA_MODE: residuals are NOT added at the end • CLEAN_NEGATIVE_MAX_TEST: clean does not stop when there is a negative absolute maximum.
Possible mistakes • CLEAN_FRAC too large • Not all flux is cleaned increase number of iterations • CLEAN boxes: be careful! Selecting boxes without reel emission can introduce artificial sources. Check with other imaging algorithm. • It is important to select good imaging parameters for backprojection (pixel_size, image_dim, ….) • Pixel size too large compared to grids used • Grid/detector without clear signal is used and images are noisy eliminate this grid
Effect of the use of grids without signal grid 3-9 grid 1, 3-9 grid 1 alone Post flare loops, 4-9 keV To get backprojections of individual grids/dectectors: image_grid_i = o getdata( class_name=‘his_bproj’, this_det=i) image_grid_i = his_annsec2xy ( image_grid_i, o )
Conclusion • CLEAN approximates image with point sources using • the RHESSI Point Spread Function (PSF) • 2) Convolves the point sources with Gaussian (CLEAN beam; • PSF without side lobes) • Works well for point sources, but also for extended sources • Relatively fast; easy to use Important KEYWORDS: CLEAN_NITER CLEAN_FRAC CLEAN_MARK_BOX