Protoplanetary Disk Simulation - VLA-CASA5.4.0: Difference between revisions

From CASA Guides
Jump to navigationJump to search
← Older edit
Hmedlin (talk | contribs)
803 edits
Hmedlin (talk | contribs)
803 edits
 
(25 intermediate revisions by the same user not shown)
Line 1: Line 1:
== Overview ==
== Overview ==


The following tutorial shows how to adopt typical parameters for simulating Karl G. Jansky VLA data. We will use the same image as the ALMA tutorial [https://casaguides.nrao.edu/index.php/Protoplanetary_Disk_Simulation_(CASA_5.1) "Protoplanetary Disk Simulation"]. Follow this link to obtain the [https://casa.nrao.edu/Data/EVLA/simulation/ppdisk672_GHz_50pc.fits protoplanetary disk model image]. Model images are in units of Jy/pixel. Other simulation options, e.g., using component lists, or how to use the toolkit are explained in the [https://casaguides.nrao.edu/index.php/Simulating_Observations_in_CASA_5.1 Simulations in CASA] section of the CASAguides.  
The following tutorial shows how to adopt typical parameters for simulating Karl G. Jansky VLA data. We will use the same image as the [https://casaguides.nrao.edu/index.php/Protoplanetary_Disk_Simulation_(CASA_5.1) ALMA tutorial Protoplanetary Disk Simulation]. Follow this link to obtain the [https://casa.nrao.edu/Data/EVLA/simulation/ppdisk672_GHz_50pc.fits protoplanetary disk model image]. Model images are in units of Jy/pixel. Other simulation options, e.g., using component lists, or how to use the toolkit are explained in the [https://casaguides.nrao.edu/index.php/Simulating_Observations_in_CASA_5.1 Simulations in CASA] section of the CASAguides.  


Figure 1 shows the model that we will use for this simulation tutorial.  
Figure 1 shows the model that we will use for this simulation tutorial.  
Line 13: Line 13:
We will mostly use the {{simobserve}} and {{simanalyze}} tasks similar to the ALMA tutorials (we will follow the ALMA plotted image sequence). The ALMA model, however, has a specified frequency of 672GHz and we will adapt it to work for VLA frequencies.   
We will mostly use the {{simobserve}} and {{simanalyze}} tasks similar to the ALMA tutorials (we will follow the ALMA plotted image sequence). The ALMA model, however, has a specified frequency of 672GHz and we will adapt it to work for VLA frequencies.   


Note that {{simobserve}} has a few limitations, e.g., it cannot simulate different spectral windows (spw). If this is desired, each spw needs to be simulated separately, followed by a concatenation ({{concat}}) of all simulated MeasurementSets (MS). In addition, {{simobserve}} has no option to add pointing errors to the simulated data. All VLA configurations and the VLA receiver temperatures are, however, accessible in {{simobserve}}.  
Note that {{simobserve}} has a few limitations, e.g., it cannot simulate different spectral windows (spw). If this is desired, each spw needs to be simulated separately, followed by a concatenation ({{concat}}) of all simulated MeasurementSets (MS). In addition, {{simobserve}} has no option to add pointing errors to the simulated data. All VLA configurations and the VLA receiver temperatures are, however, accessible in {{simobserve}}.
 


== Q-band, 128MHz bandwidth, noiseless image, 1hr integration time, A-configuration, no deconvolution ==
== Q-band, 128MHz bandwidth, noiseless image, 1hr integration time, A-configuration, no deconvolution ==
Line 139: Line 138:
== Q-band, 128MHz bandwidth, noiseless image, 1hr integration time, A-configuration, cleaned ==
== Q-band, 128MHz bandwidth, noiseless image, 1hr integration time, A-configuration, cleaned ==


As a second step we will deconvolve this data set. This could be done inside the same project, but for a separation of our use cases, we will recreate the MS with {{simobserve}} and modify the {{simanalyze}} call:
As a second step we will deconvolve this dataset. This could be done inside the same project, but for a separation of our use cases, we will recreate the MS with {{simobserve}} and modify the {{simanalyze}} call.


<source lang="python">
<source lang="python">
Line 156: Line 155:
== Q-band, 128MHz bandwidth, 4mm pwv, 1hr integration time, A-configuration ==
== Q-band, 128MHz bandwidth, 4mm pwv, 1hr integration time, A-configuration ==


We now add environmental effects with an atmospheric precipitable water vapor of 4mm and a ground temperature for spillover of 270K. Such conditions are fairly good at the VLA site. We change the  ''user_pwr'' and ''t_ground'' parameters in {{simobserve}} accordingly. The task only adds thermal noise due to the sky brightness, but not phase noise due to pwv variations across the array. The ''sm.settrop'' tool can be used for the more sophisticated tropospheric models. A description is given in [https://casaguides.nrao.edu/index.php/Corrupting_Simulated_Data_(Simulator_Tool) Corrupting Simulated Data].
We now add environmental effects with an atmospheric precipitable water vapor (pwv) of 4mm and a ground temperature for spillover of 270K. Such conditions are fairly good at the VLA site. We change the  ''user_pwr'' and ''t_ground'' parameters in {{simobserve}} accordingly. The task only adds thermal noise due to the sky brightness, but not phase noise due to pwv variations across the array. The ''sm.settrop'' tool can be used for the more sophisticated tropospheric models. A description is given in [https://casaguides.nrao.edu/index.php/Corrupting_Simulated_Data_(Simulator_Tool) Corrupting Simulated Data].


<source lang="python">
<source lang="python">
Line 163: Line 162:
</source>
</source>


The {{simobserve}} output is identical to Figs. 2 and 3 since we do not change any parameters that influence the figure. Now let's analyze the MS:
The {{simobserve}} output is identical to Figures 2 and 3 since we did not change any parameters that influence the figure. Now let's analyse the MS.


<source lang="python">
<source lang="python">
Line 170: Line 169:
</source>
</source>


The results are displayed in Fig. 6. The noise is dominating the image no as expected.
The results are displayed in Figure 6. The noise is dominating the image as expected.


{|
{|
|[[Image:VLAsim-psim3-ana.png|600px|thumb|left|'''Fig. 6:''' Output of {{simanalyze}} after injecting noise.]]
|[[Image:VLAsim-psim3-ana.png|600px|thumb|left|'''Figure 6:''' Output of {{simanalyze}} after injecting noise.]]
|}
|}


Line 179: Line 178:
==== Comparison with the VLA Exposure Calculator ====
==== Comparison with the VLA Exposure Calculator ====


We can compare the simulation to the predicted VLA sensitivity via the [http://go.nrao.edu/ect VLA Exposure Calculator Tool (ECT)] ([http://go.nrao.edu/ect Description]). Using winter observing conditions (for good pwv), a medium elevation (cf. Fig. 3), 128MHz bandwidth at a frequency of 44GHz, the ECT predicts a beam size of 0.044" and which is compatible to our simulations (cf. Fig. 3 and 6). The ECT output is shown in Fig. 7.  
We can compare the simulation to the predicted VLA sensitivity via the VLA [http://go.nrao.edu/ect Exposure Calculator Tool (ECT)]. Using winter observing conditions (for good pwv), a medium elevation (see Figure 3), 128MHz bandwidth at a frequency of 44GHz, the ECT predicts a beam size of 0.044" and which is compatible to our simulations (see Figures 3 and 6). The ECT output is shown in Figure 7.  


{|
{|
|[[Image:VLAsim-etc.png|400px|thumb|left|'''Fig. 7:''' Exposure Calculation for 4mm pwr, 128MHz bandwidth]]
|[[Image:VLAsim-etc.png|400px|thumb|left|'''Figure 7:''' Exposure Calculation for 4mm pwr, 128MHz bandwidth]]
|}
|}


The rms provided in Fig. 6 is over the entire image and includes the source. It is therefore not representative to the thermal noise. To obtain a better noise figure, we now produce an image with the same parameters as {{simanalyze}}, but with a larger image size directly via {{tclean}}. Since the briggs parameter in {{simanalyze}} defaults to robust=0.5, we will use the same value here:
The rms provided in Figure 6 is over the entire image and includes the source. It is therefore not representative to the thermal noise. To obtain a better noise figure, we now produce an image with the same parameters as {{simanalyze}}, but with a larger image size directly via {{tclean}}. Since the briggs parameter in {{simanalyze}} defaults to robust=0.5, we will use the same value here.


<source lang="python">
<source lang="python">
Line 192: Line 191:
</source>
</source>


Opening the image in the {{viewer}}
Open the image in the {{viewer}}:


<source lang="python">
<source lang="python">
Line 199: Line 198:
</source>
</source>


and measuring the noise statistics away from the source (Fig. 8), gives a value of about 7e-5Jy/beam which is in very good agreement with the ECT predicted noise figure of 6.9e-5 Jy/beam.
Measuring the noise statistics away from the source (Figure 8) gives a value of about 7e-5Jy/beam which is in very good agreement with the ECT predicted noise figure of 6.9e-5 Jy/beam.


{|
{|
|[[Image:VLAsim-tcleanbig.png|600px|thumb|left|'''Fig. 8:''' Larger image of the simulation given in Fig. 6, measuring the rms noise in the purple rectangle regions away from the emission. A few low-level sidelobes are still present, however. ]]
|[[Image:VLAsim-tcleanbig.png|600px|thumb|left|'''Figure 8:''' Larger image of the simulation given in Figure 6, measuring the rms noise in the purple rectangle regions away from the emission. A few low-level sidelobes are still present, however. ]]
|}
|}


== Q-band, 8GHz bandwidth, 4mm pwv, 1hr integration time, A-configuration ==
== Q-band, 8GHz bandwidth, 4mm pwv, 1hr integration time, A-configuration ==


To compensate for noise, we will extend our bandwidth from 128MHz to 8GHz. The current implementation of {{simobserve}} has no channelization so the simulation will not have the better uv-coverage that is obtained in multi-frequency imaging. Consecutive simulations with different spectral windows, and combined MSs, however, can achieve that and we will show how to obtain a multi-channel MS below. In addition, for large channel widths, one needs to be careful not to introduce [https://science.nrao.edu/facilities/vla/docs/manuals/oss2016A/performance/fov/bw-smearing bandwidth smearing]. Our target is luckily near the phase center of the image and bandwidth smearing will be minimal.  
To compensate for noise, we will extend our bandwidth from 128MHz to 8GHz. The current implementation of {{simobserve}} has no channelization so the simulation will not have the better uv-coverage that is obtained in multi-frequency imaging. Consecutive simulations with different spw's, and combined MS's, however, can achieve that and we will show how to obtain a multi-channel MS below. In addition, for large channel widths, one needs to be careful not to introduce [https://science.nrao.edu/facilities/vla/docs/manuals/oss2016A/performance/fov/bw-smearing bandwidth smearing]. Our target is luckily near the phase center of the image and bandwidth smearing will be minimal.  


<source lang="python">
<source lang="python">
Line 214: Line 213:
</source>
</source>


The {{simobserve}} output again is identical to Fig. 3.  
The {{simobserve}} output again is identical to Figure 3.  


<source lang="python">
<source lang="python">
Line 221: Line 220:
</source>
</source>


The image (Fig. 9) is clearly better defined, as expected from a broader bandwidth. We also decreased the cleaning threshold to 0.05mJy and added a few more iterations to reach this threshold.  
The image (see Figure 9) is clearly better defined, as expected from a broader bandwidth. We also decreased the cleaning threshold to 0.05mJy and added a few more iterations to reach this threshold.  


{|
{|
|[[Image:VLAsim-psim4-ana.png|600px|thumb|left|'''Fig. 9:''' Imaging with 8GHz bandwidth.]]
|[[Image:VLAsim-psim4-ana.png|600px|thumb|left|'''Figure 9:''' Imaging with 8GHz bandwidth.]]
|}
|}


<!--  
<!--  
Line 232: Line 230:
==== Comparison with the VLA Exposure Calculator ====
==== Comparison with the VLA Exposure Calculator ====


We compare again the simulation to the predicted VLA sensitivity via the [http://go.nrao.edu/ect VLA Exposure Calculator Tool (ETC)] ([http://go.nrao.edu/ect Description]). Using the same conditions and parameters as above, except for 8GHz bandwidth, the ECT predicts a rms of 10.7<math>\mu Jy</math> (Fig. 10):  
We compare again the simulation to the predicted VLA sensitivity via the VLA [http://go.nrao.edu/ect Exposure Calculator Tool (ETC)]. Using the same conditions and parameters as above, except for 8GHz bandwidth, the ECT predicts a rms of 10.7<math>\mu Jy</math> (see Figure 10):  


{|
{|
|[[Image:VLAsim-etc8GHz.png|400px|thumb|left|'''Fig. 10''': Exposure time calculation for 4mm pwr, 8GHz bandwidth]]
|[[Image:VLAsim-etc8GHz.png|400px|thumb|left|'''Figure 10''': Exposure time calculation for 4mm pwr, 8GHz bandwidth]]
|}
|}


Note that the VLA exposure calculator switched to 3bit observing to accommodate the large bandwidth. The simulator, however, assumes 8bit samplers, which provide about 15% better sensitivity than VLA 3bit correlations (For a discussion, see the [https://science.nrao.edu/facilities/vla/docs/manuals/oss/performance/vla-samplers VLA Samplers] page).  
Note that the VLA ECT switched to the 3-bit samplers to accommodate the large bandwidth. The simulator, however, assumes 8-bit samplers, which provide about 15% better sensitivity than VLA 3-bit correlations (For a discussion, see the [https://science.nrao.edu/facilities/vla/docs/manuals/oss/performance/vla-samplers VLA Samplers] sections within the OSS).  


Similar to the case of 128MHz bandwidth, we produce a larger image to be able to measure the image rms in a signal-free region:
Similar to the case of 128MHz bandwidth, we produce a larger image to be able to measure the image rms in a signal-free region.


<source lang="python">
<source lang="python">
Line 247: Line 245:
</source>
</source>


opening the image in the {{viewer}}  
Open the image in the {{viewer}}  


<source lang="python">
<source lang="python">
Line 254: Line 252:
</source>
</source>


and measuring the noise statistics away from the source (Fig. 11), gives a value of about 1.3e-5Jy/beam which is in reasonable agreement with the exposure calculator prediction. The regions where we measured the noise still contains some residual sidelobe structure and deeper cleaning, especially with a clean mask would improve the noise figure further, bringing it even closer to the predicted value.  
Measuring the noise statistics away from the source (see Figure 11), gives a value of about 1.3e-5Jy/beam which is in reasonable agreement with the exposure calculator prediction. The regions where we measured the noise still contains some residual sidelobe structure and deeper cleaning, especially with a clean mask would improve the noise figure further, bringing it even closer to the predicted value.  


{|
{|
|[[Image:VLAsim-tcleanbig8GHz.png|600px|thumb|left|'''Fig. 11:''' Larger Image of the 8GHz simulation in Fig. 9, measuring the rms noise in the purple rectangle regions away from the emission.]]
|[[Image:VLAsim-tcleanbig8GHz.png|600px|thumb|left|'''Figure 11:''' Larger Image of the 8GHz simulation in Fig. 9, measuring the rms noise in the purple rectangle regions away from the emission.]]
|}
|}
-->
-->
Line 265: Line 263:
In reality, the 8GHz bandwidth is not confined to a single channel but distributed across many channels. This results in a spread out uv-coverage as it depends on the projected baselines expressed in number of wavelengths, so channels are slightly displaced from each other. Combining the channels in a multi-frequency synthesis (mfs), therefore does not only increase the sensitivity, but also the image fidelity through a better defined psf. Channelization and mfs imaging also reduces bandwidth smearing effects.  
In reality, the 8GHz bandwidth is not confined to a single channel but distributed across many channels. This results in a spread out uv-coverage as it depends on the projected baselines expressed in number of wavelengths, so channels are slightly displaced from each other. Combining the channels in a multi-frequency synthesis (mfs), therefore does not only increase the sensitivity, but also the image fidelity through a better defined psf. Channelization and mfs imaging also reduces bandwidth smearing effects.  


As mentioned above, {{simobserve}}, naturally, does not channelize the visibilities when the input image is not channelized.  
As mentioned above, {{simobserve}}, naturally does not channelize the visibilities when the input image is not channelized.  


In the following, we will channelize the MS and, to show the procedure, we will also apply a spectral index of <math>S\propto\nu^{-1.5}</math> at the same time.  
In the following, we will channelize the MS and, to show the procedure, we will also apply a spectral index of <math>S\propto\nu^{-1.5}</math> at the same time.  
Line 274: Line 272:
<source lang="python">
<source lang="python">
# In CASA
# In CASA
%cpaste
importfits(fitsimage="ppdisk672_GHz_50pc.fits", imagename="ppdisk_Q_50pc.im")
importfits(fitsimage="ppdisk672_GHz_50pc.fits", imagename="ppdisk_Q_50pc.im")
stats=imstat(imagename='ppdisk_Q_50pc.im')
stats=imstat(imagename='ppdisk_Q_50pc.im')
Line 283: Line 283:
     imhead(imagename="ppdisk_Q_50pc_spx_"+str(x)+".im", mode="put", hdkey="crval4", hdvalue=str(y)+"GHz")
     imhead(imagename="ppdisk_Q_50pc_spx_"+str(x)+".im", mode="put", hdkey="crval4", hdvalue=str(y)+"GHz")
     imhead(imagename="ppdisk_Q_50pc_spx_"+str(x)+".im", mode="put", hdkey="cdelt4", hdvalue="128MHz")
     imhead(imagename="ppdisk_Q_50pc_spx_"+str(x)+".im", mode="put", hdkey="cdelt4", hdvalue="128MHz")
comb=ia.imageconcat(outfile="ppdisk-combined_spx.im", infiles="ppdisk_Q_50pc_spx*.im",  axis=3, relax=True, tempclose=False, reorder=True, overwrite=True)
comb=ia.imageconcat(outfile="ppdisk-combined_spx.im", infiles="ppdisk_Q_50pc_spx*.im",  axis=3, relax=True, tempclose=False, reorder=True, overwrite=True)
comb.close()
comb.close()
--
</source>
</source>


Using this model cube in {{simobserve}} will produce the channelized visibilities:
Using this model cube in {{simobserve}} will produce the channelized visibilities.


<source lang="python">
<source lang="python">
Line 295: Line 296:
</source>
</source>


{{simobserve}} and {{simanalyze}} assumes that, if the model is a cube, the output will be a cube, too. In our case, however, we will produce a single multi-frequency synthesis continuum image so we have to use other, general CASA tasks for imaging and display.   
{{simobserve}} and {{simanalyze}} assumes that, if the model is a cube, the output will be a cube too. In our case, however, we will produce a single multi-frequency synthesis continuum image so we have to use other, general CASA tasks for imaging and display.   


Let's first have a look at the new uv-coverage. Each channel is displayed in a different color in Fig. 10:
Let's first have a look at the new uv-coverage. Each channel is displayed in a different color in Figure 10:


<source lang="python">
<source lang="python">
Line 305: Line 306:


{|
{|
|[[Image:VLAsim-uvcovermfs.png|400px|thumb|left|'''Fig. 10:''' uv-coverage of chanelized visibilities.]]
|[[Image:VLAsim-uvcovermfs.png|400px|thumb|left|'''Figure 10:''' uv-coverage of chanelized visibilities.]]
|}
|}


We will now create a continuum image with the mfs technique, and, to recover the spectral index, we will use the Multi-term Taylor expansion in frequency (mtmfs, see [https://casa.nrao.edu/casadocs/latest/synthesis-imaging/deconvolution-algorithms CASAdocs Deconvolution Algorithms]). We will use ''nterms=2'' to derive the spectral index. The spectral index map, however, is fairly sensitive to the scales in the image. A regular Hogbom clean based on point sources, typically creates maps with too steep spectral indices when applied to extended emission. To avoid this effect, we therefore use the multi-scale algorithm with a range of scales:  
We will now create a continuum image with the mfs technique, and, to recover the spectral index, we will use the Multi-term Taylor expansion in frequency (mtmfs, see [https://casa.nrao.edu/casadocs/latest/synthesis-imaging/deconvolution-algorithms CASAdocs Deconvolution Algorithms]). We will use ''nterms=2'' to derive the spectral index. The spectral index map, however, is fairly sensitive to the scales in the image. A regular Hogbom clean based on point sources, typically creates maps with too steep spectral indices when applied to extended emission. To avoid this effect, we therefore use the multi-scale algorithm with a range of scales.  


<source lang="python">
<source lang="python">
Line 314: Line 315:
tclean(vis='psimvla4mfs/psimvla4mfs.vla.a.noisy.ms', imagename='psimvla4mfs/8GHzmfs', imsize=[192, 192], cell='3.11e-3arcsec', specmode='mfs', gridder='standard', deconvolver='mtmfs', nterms=2, scales=[0,2,4,8,13,20,40],  niter=1000, threshold='2e-4Jy', weighting='briggs', robust=0.5)
tclean(vis='psimvla4mfs/psimvla4mfs.vla.a.noisy.ms', imagename='psimvla4mfs/8GHzmfs', imsize=[192, 192], cell='3.11e-3arcsec', specmode='mfs', gridder='standard', deconvolver='mtmfs', nterms=2, scales=[0,2,4,8,13,20,40],  niter=1000, threshold='2e-4Jy', weighting='briggs', robust=0.5)
</source>
</source>
As before, we are not attempting primary beam corrections as the image is very small compared to the primary beam (cf. Fig. 2).   
 
As before, we are not attempting primary beam corrections as the image is very small compared to the primary beam (see Figure 2).   


<!--
<!--


Let's also do a cube:
## This portion has been commented out the spectral index is very unreliable. In fact it is still very unreliable with tclean/mtmfs and depends a lot on how it is cleaned, in particular the multi scale parameters… so in the end Juergen decided not to compare the cube/specfit to mtmfs spectral indices. ##
 
Let's also make a cube.
<source lang="python">
<source lang="python">
# In CASA
# In CASA
Line 324: Line 328:
</source>
</source>


and we will use {{spxfit}} to calculate the spectral index from the cube for comparison. First, however, we need to smooth to a common beam across all frequencies, the parameter ''kernel="commonbeam"'' smoothes the entire cube to the lowest resolution in the cube (typically the channel with the lowest frequency):
We will use {{spxfit}} to calculate the spectral index from the cube for comparison. First, however, we need to smooth to a common beam across all frequencies, the parameter ''kernel="commonbeam"'' smooths the entire cube to the lowest resolution in the cube (typically the channel with the lowest frequency):


<source lang="python">
<source lang="python">
Line 336: Line 340:




'''Note, that for both, {{tclean}} with ''mtmfs'' and {{spxfit}} reliable results are only obtained for high signal-to-noise regions.'''
Note, that for both, {{tclean}} with ''mtmfs'' and {{spxfit}} reliable results are only obtained for high signal-to-noise regions.
 
## End of commented out portion ##


-->  
-->  


The images can be displayed with the viewer. Note that the image itself is called ''psimvla4mfs/8GHzmfs.image.tt0'' (Fig. 11) and the spectral index map ''psimvla4mfs/8GHzmfs.alpha'' (Fig. 12). In Fig. 12 we also show the contours of the image on the spectral index map, to check for the signal-to-noise at each position.  
The images can be displayed with the viewer using the contour button to load the images with the contour overlay. Figure 11 is '''8GHzmfs.image.tt0''' and Figure 12 is '''8GHzmfs.alpha''', both located under the directory ''psimvla4mfs''. In Figure 12, we show the contours of the image on the spectral index map to check for the signal-to-noise at each position.  


{|
{|
|[[Image:VLAsim-psim4mfstclean.png|400px|thumb|left|'''Fig. 11:''' The image of Fig. 9 but using multi-frequency-synthesis (mfs) imaging. ]]
|[[Image:VLAsim-psim4mfstclean.png|400px|thumb|left|'''Figure 11:''' Similar to the image of Figure 9, but using multi-frequency-synthesis (mfs) imaging. ]]
|[[Image:VLAsim-psim4mfstcleanalpha.png|400px|thumb|left|'''Fig. 12:''' The spectral index map with image contours overlaid.]]
|[[Image:VLAsim-psim4mfstcleanalpha.png|400px|thumb|left|'''Figure 12:''' The spectral index map with image contours overlaid.]]
|}
|}


As seen in Fig. 11, this method should produce a somewhat better fidelity than the non-mfs image shown in Fig. 9, although a direct comparison is difficult given the spectral index that we introduced. The spectral index map itself (Fig. 12), unfortunately, is dominated by noise, even at the brightest regions, partly due to the pwv that we assumed. This results in a relatively unreliable spectral index map overall, although the values scatter around the spectral index of -1.5 that we inserted earlier.
As seen in Figure 11, this method should produce a somewhat better fidelity than the non-mfs image shown in Figure 9, although a direct comparison is difficult given the spectral index that we introduced. The spectral index map itself (Figure 12), unfortunately, is dominated by noise, even at the brightest regions, partly due to the pwv that we assumed. This results in a relatively unreliable spectral index map overall, although the values scatter around the spectral index of -1.5 that we inserted earlier.


== Q-band, 8GHz bandwidth, 4mm pwv, 4hr integration time, A-configuration ==
== Q-band, 8GHz bandwidth, 4mm pwv, 4hr integration time, A-configuration ==


Let's improve the depth of the observations even further by going to 4h on-source integration, via '' totaltime='14400s' ''. Given that mfs imaging only marginally improved the image, for simplicity we will go back to using the original, unchannelized 8GHz bandwidth.   
Let's improve the depth of the observations even further by going to 4hr on-source integration, via '' totaltime='14400s' ''. Given that mfs imaging only marginally improved the image, for simplicity we will go back to using the original, unchannelized 8GHz bandwidth.   


<source lang="python">
<source lang="python">
Line 358: Line 364:
</source>
</source>


Fig. 13 shows a better uv-coverage (lower left image) and the 4h observation is also marked by a wider red bar in the top left image. This also improved the shape of the psf.  
Figure 13 shows a better uv-coverage (lower left image) and the 4hr observation is also marked by a wider red bar in the top left image. This also improved the shape of the psf.  


{|
{|
|[[Image:VLAsim-psim5-obs.png|600px|thumb|left|'''Fig. 13:''' Output of {{simobserve}} with 8GHz bandwidth of 4h of on-source observing time. ]]
|[[Image:VLAsim-psim5-obs.png|600px|thumb|left|'''Figure 13:''' Output of {{simobserve}} with 8GHz bandwidth of 4hr of on-source observing time. ]]
|}
|}


Line 369: Line 375:
</source>
</source>


Given the expected deeper image, we reduced the cleaning threshold to 0.02mJy, and the result is shown in Fig. 14.  The image now is an almost perfect representation of the true sky model convolved with the clean beam.  
Given the expected deeper image, we reduced the cleaning threshold to 0.02mJy, and the result is shown in Figure 14.  The image now is an almost perfect representation of the true sky model convolved with the clean beam.  


{|
{|
|[[Image:VLAsim-psim5-ana.png|600px|thumb|left|'''Fig. 14:''' Images after increasing the integration time to 4h.]]
|[[Image:VLAsim-psim5-ana.png|600px|thumb|left|'''Figure 14:''' Images after increasing the integration time to 4hr.]]
|}
|}


Line 378: Line 384:
==== Comparison with the VLA Exposure Calculator ====
==== Comparison with the VLA Exposure Calculator ====


We compare again the simulation to the predicted VLA sensitivity via the [http://go.nrao.edu/ect VLA Exposure Calculator Tool (ETC)] ([http://go.nrao.edu/ect Description]). Using 8GHz bandwidth 4h on-source, and a low elevation due to the longer observations, the ECT predicts a rms of ~7.1<math>\mu Jy</math> (Fig. 15):  
We compare again the simulation to the predicted VLA sensitivity via the VLA [http://go.nrao.edu/ect Exposure Calculator Tool (ETC)]. Using 8GHz bandwidth 4hr on-source, and a low elevation due to the longer observations, the ECT predicts a rms of ~7.1<math>\mu Jy</math> (see Figure 15):  


{|
{|
|[[Image:VLAsim-etc8GHz4h.png|400px|thumb|left|'''Fig. 15''': Exposure time calculation for 4mm pwr, 8GHz bandwidth]]
|[[Image:VLAsim-etc8GHz4h.png|400px|thumb|left|'''Figure 15''': Exposure time calculation for 4mm pwr, 8GHz bandwidth]]
|}
|}


Using the same settings but a medium elevation results in ~5.4 <math>\mu Jy</math>.  
Using the same settings but a medium elevation results in ~5.4 <math>\mu Jy</math>.  
Note that the VLA exposure calculator switched to 3bit observing to accommodate the large bandwidth. The simulator, however, assumes 8bit samplers, which provide about 15% better sensitivity than the 3bit samplers of the VLA (For a discussion, see the [https://science.nrao.edu/facilities/vla/docs/manuals/oss/performance/vla-samplers VLA Samplers] page).  
Note that the ECT switched to the 3-bit samplers to accommodate the large bandwidth. The simulator, however, assumes 8-bit samplers, which provide about 15% better sensitivity than the 3-bit samplers of the VLA. For a discussion, see the [https://science.nrao.edu/facilities/vla/docs/manuals/oss/performance/vla-samplers VLA Samplers] page.  


Similar to the case of 128MHz bandwidth 1h case, we produce a larger image to be able to measure the image rms in a signal-free region:
Similar to the case of 128MHz bandwidth 1hr case, we produce a larger image to be able to measure the image rms in a signal-free region.


<source lang="python">
<source lang="python">
Line 394: Line 400:
</source>
</source>


opening the image in the {{viewer}}  
Open the image in the {{viewer}}  


<source lang="python">
<source lang="python">
Line 401: Line 407:
</source>
</source>


and measuring the noise statistics away from the source (Fig. 16), gives a value of about 6mJy/beam which is slightly better but in very good agreement with the exposure calculator predictions between the two elevation settings.   
Measuring the noise statistics away from the source (Figure 16), gives a value of about 6mJy/beam which is slightly better but in very good agreement with the exposure calculator predictions between the two elevation settings.   
<!--
<!--
The regions where we measured the noise still contains some residual sidelobe structure and deeper cleaning, especially with a clean mask would improve the noise figure further, bringing it even closer to the predicted value.  
The regions where we measured the noise still contains some residual sidelobe structure and deeper cleaning, especially with a clean mask would improve the noise figure further, bringing it even closer to the predicted value.  
-->
-->
{|
{|
|[[Image:VLAsim-tcleanbig8GHz4h.png|600px|thumb|left|'''Fig. 16:''' Larger Image of the 8GHz simulation in Fig. 14, measuring the rms noise in the purple rectangle regions away from the emission.]]
|[[Image:VLAsim-tcleanbig8GHz4h.png|600px|thumb|left|'''Figure 16:''' Larger Image of the 8GHz simulation in Figure 14, measuring the rms noise in the purple rectangle regions away from the emission.]]
|}
|}


Line 413: Line 419:


<!--
<!--
===Q-band, 8GHz bandwidth, 18mm pwv, 1h integration time, A-configuration===
== Q-band, 8GHz bandwidth, 18mm pwv, 1hr integration time, A-configuration ==


We will now go back to 1h integration, but assume worse conditions of the VLA site with a precipitable water vapor of 18mm.  
We will now go back to 1hr integration, but assume worse conditions of the VLA site with a pwv of 18mm.  


<source lang="python">
<source lang="python">
Line 427: Line 433:
</source>
</source>


We see that the resulting image is almost identical to the pwv of 4mm (Fig. 7), with almost identical noise figures. The observations seem to be very little affected by the variations in the VLA site pwv. Note that the simulations do not include pointing errors due to stronger winds.  
We see that the resulting image is almost identical to the pwv of 4mm (Figure 7), with almost identical noise figures. The observations seem to be very little affected by the variations in the VLA site pwv. Note that the simulations do not include pointing errors due to stronger winds.  


{|
{|
|[[Image:VLAsim-psim6-ana.png|600px|thumb|left|Fig. 10: Assuming less than ideal weather conditions with pwv of 18mm.]]
|[[Image:VLAsim-psim6-ana.png|600px|thumb|left|Figure 10: Assuming less than ideal weather conditions with pwv of 18mm.]]
|}
|}


-->
-->


===Q-band, 8GHz bandwidth, 4mm pwv, 1h integration time, C-configuration===
== Q-band, 8GHz bandwidth, 4mm pwv, 1hr integration time, C-configuration ==


Our last Q-band simulation will use a precipitable water vapor of 4mm again, but we will now simulate the observations with the VLA C-configuration (specified with the ''antennalist'' parameter).  
Our last Q-band simulation will use a pwv of 4mm again, but we will now simulate the observations with the VLA C-configuration (specified with the ''antennalist'' parameter).  


<source lang="python">
<source lang="python">
Line 444: Line 450:
</source>
</source>


As expected, the extent of the uv-coverage decreases by a factor of about 10 and the synthesized beam size increases by the same factor, as seen in Fig. 17.  
As expected, the extent of the uv-coverage decreases by a factor of about 10 and the synthesized beam size increases by the same factor, as seen in Figure 17.  


{|
{|
|[[Image:VLAsim-psim6-obs.png|600px|thumb|left|'''Fig. 17:''' Output of {{simobserve}} using the C-configuration. ]]
|[[Image:VLAsim-psim6-obs.png|600px|thumb|left|'''Figure 17:''' Output of {{simobserve}} using the C-configuration. ]]
|}
|}


Line 455: Line 461:
</source>
</source>


Note that the output filenames will now change and reflect C-configuration in their names. The resulting images are displayed in Fig. 18, where the images now look like a point source with the degraded resolution. The surface brightness sensitivity of the observations, however, increases and we can clean to a lower threshold of 0.1mJy.  {{simobserve}} automatically detected that the beam is massively oversampled and that a larger image size will provide a better field of view.  
Note that the output file names will now change and reflect C-configuration in their names. The resulting images are displayed in Figure 18, where the images now look like a point source with the degraded resolution. The surface brightness sensitivity of the observations, however, increases and we can clean to a lower threshold of 0.1mJy.  {{simobserve}} automatically detected that the beam is massively oversampled and that a larger image size will provide a better field of view.  


{|
{|
|[[Image:VLAsim-psim6-ana.png|600px|thumb|left|'''Fig. 18:''' C-configuration results.]]
|[[Image:VLAsim-psim6-ana.png|600px|thumb|left|'''Figure 18:''' C-configuration results]]
|}
|}


The sharp edges in some of the images show the boundaries of the original model image which has non-zero for all pixels, the increased field of view was simply padded with masked values.
The sharp edges in some of the images show the boundaries of the original model image which has non-zero for all pixels. The increased field of view was simply padded with masked values.


To combine the data of two different array configurations, we refer to the [https://casaguides.nrao.edu/index.php/VLA_Data_Combination VLA Data Combination Guide].
To combine the data of two different array configurations, we refer to the [https://casaguides.nrao.edu/index.php/VLA_Data_Combination VLA Data Combination Guide].


== X-band, 4GHz bandwidth, 4mm pwv, 1h integration time, A-configuration ==
== X-band, 4GHz bandwidth, 4mm pwv, 1hr integration time, A-configuration ==


Finally, we change the frequency and perform an '''X-band (10GHz)''' A-configuration simulation. Since X-band has a maximum bandwidth of 4GHz, we will reduce it to this value. We will also lower the flux density of the model to 0.01mJy given that the brightness of the model will likely be lower, too.   
Finally, we change the frequency and perform an '''X-band (10GHz)''' A-configuration simulation. Since X-band has a maximum bandwidth of 4GHz, we will reduce it to this value. We will also lower the flux density of the model to 0.01mJy given that the brightness of the model will likely be lower, too.   
Line 474: Line 480:
</source>
</source>


When expressed in wavelengths, the uv-coverage and synthesized beam are somewhere in between that of Q-band A-configuration and C-configuration as expected (Fig. 19).  
When expressed in wavelengths, the uv-coverage and synthesized beam are somewhere in between that of Q-band A-configuration and C-configuration as expected (Figure 19).  


{|
{|
|[[Image:VLAsim-psim8-obs.png|600px|thumb|left|'''Fig. 19:''' Output of {{simobserve}} for X-band observations. ]]
|[[Image:VLAsim-psim8-obs.png|600px|thumb|left|'''Figure 19:''' Output of {{simobserve}} for X-band observations. ]]
|}
|}


Line 486: Line 492:




The larger primary beam of the X-band data is shown in Fig. 20, where the model image is barely visible in the center. The simulated images are shown in Fig. 21. The ring shape is still visible in the results although it is quite blurred. We note that the noise is considerably lower as expected from the lower band noise properties of the receiver and atmosphere. We adjusted the clean threshold accordingly. Again, {{simanalyze}} was overriding the image size as the psf would not have been adequately covered for the lower frequency.  
The larger primary beam of the X-band data is shown in Figure 20, where the model image is barely visible in the center. The simulated images are shown in Figure 21. The ring shape is still visible in the results although it is quite blurred. We note that the noise is considerably lower as expected from the lower band noise properties of the receiver and atmosphere. We adjusted the clean threshold accordingly. Again, {{simanalyze}} was overriding the image size as the psf would not have been adequately covered for the lower frequency.  


{|
{|
|[[Image:VLAsim-psim8-sky.png|500px|thumb|left|'''Fig. 20:''' X-band sky coverage]]
|[[Image:VLAsim-psim8-sky.png|500px|thumb|left|'''Figure 20:''' X-band sky coverage]]
|[[Image:VLAsim-psim8-ana.png|500px|thumb|left|'''Fig. 21:''' X-band simulation results.]]
|[[Image:VLAsim-psim8-ana.png|500px|thumb|left|'''Figure 21:''' X-band simulation results]]
|}
|}


Line 510: Line 516:
simobserve(project='psimvlatest',  skymodel='ppdisk-combined.im', inbright='3e-5Jy/pixel',  setpointings=True, integration='2s',  direction='J2000 18h00m00.031s -22d59m59.6s',  mapsize= '0.78arcsec', obsmode='int', antennalist='vla.a.cfg', hourangle='transit', totaltime='3600s',  thermalnoise='tsys-atm', user_pwv=4, t_ground=270.0, graphics='both', overwrite=True)
simobserve(project='psimvlatest',  skymodel='ppdisk-combined.im', inbright='3e-5Jy/pixel',  setpointings=True, integration='2s',  direction='J2000 18h00m00.031s -22d59m59.6s',  mapsize= '0.78arcsec', obsmode='int', antennalist='vla.a.cfg', hourangle='transit', totaltime='3600s',  thermalnoise='tsys-atm', user_pwv=4, t_ground=270.0, graphics='both', overwrite=True)
</source>
</source>


plotms(vis='psimvlatest/psimvlatest.vla.a.noisy.ms', xaxis='Uwave', yaxis='Vwave',coloraxis='channel')
plotms(vis='psimvlatest/psimvlatest.vla.a.noisy.ms', xaxis='Uwave', yaxis='Vwave',coloraxis='channel')

Latest revision as of 18:55, 19 March 2019

Overview

The following tutorial shows how to adopt typical parameters for simulating Karl G. Jansky VLA data. We will use the same image as the ALMA tutorial Protoplanetary Disk Simulation. Follow this link to obtain the protoplanetary disk model image. Model images are in units of Jy/pixel. Other simulation options, e.g., using component lists, or how to use the toolkit are explained in the Simulations in CASA section of the CASAguides.

Figure 1 shows the model that we will use for this simulation tutorial.

Figure 1: Model image of a protoplanetary disk in units of Jy/pixel that we use for this simulation guide.

The ALMA version of the tutorial describes CASA tools to derive the center of the image. We will use their results and specify direction='J2000 18h00m00.031s -22d59m59.6s' for all of our simulations. The image center can also be determined with the CASA viewer. Given that the VLA primary beams at the VLA frequencies are much larger than the image, the precise pointing direction center is less important.

We will mostly use the simobserve and simanalyze tasks similar to the ALMA tutorials (we will follow the ALMA plotted image sequence). The ALMA model, however, has a specified frequency of 672GHz and we will adapt it to work for VLA frequencies.

Note that simobserve has a few limitations, e.g., it cannot simulate different spectral windows (spw). If this is desired, each spw needs to be simulated separately, followed by a concatenation (concat) of all simulated MeasurementSets (MS). In addition, simobserve has no option to add pointing errors to the simulated data. All VLA configurations and the VLA receiver temperatures are, however, accessible in simobserve.

Q-band, 128MHz bandwidth, noiseless image, 1hr integration time, A-configuration, no deconvolution

Let's start with a simulation at 44GHz (Q-band), with a bandwidth of 128MHz, the largest possible bandwidth of a spectral window at the VLA. We will simulate observations with the VLA A-configuration as it provides the resolution that is needed for the disk to be well resolved. To start with, we do not add any noise to the data. 

# In CASA
simobserve(project='psimvla1', 
                    skymodel='ppdisk672_GHz_50pc.fits', 
                    inbright='3e-5Jy/pixel', 
                    incenter='44GHz', 
                    inwidth='128MHz' , 
                    setpointings=True, 
                    integration='2s',  
                    direction='J2000 18h00m00.031s -22d59m59.6s',  
                    mapsize= '0.78arcsec', 
                    obsmode='int', 
                    antennalist='vla.a.cfg', 
                    hourangle='transit', 
                    totaltime='3600s',  
                    thermalnoise='', 
                    graphics='both', 
                    overwrite=True)

project: The name of our project is psimvla1. All data will be stored in a directory that is created using the project name.

skymodel: The input model image in Jy/pixel units. We overwrite the fits header to assume that the model is valid for 44GHz with the incenter parameter and the bandwidth to 128MHz with inwidth. We also adjust the peak to a lower [math]\displaystyle{ 3\times10^{-5} }[/math]Jy/pixel value with the inbright parameter, as expected at the lower frequency.

setpointings: allows simobserve to derive the pointing positions by its own algorithm. Given that the primary beam at Q-band is about 1 arcminute (see the VLA Observational Status Summary (OSS)), and the size of the model is less than an arcsecond, a single pointing will be adequate.

integration: To avoid time smearing, we follow the guidance for data rates in the VLA OSS and assume 2s correlator integration time per visibility.

direction: the center of the map. For a single pointing this is equivalent to the pointing center.

obsmode: int is used for interferometric data such as VLA observations.

antennalist: the VLA configuration antenna position file. The files are available in CASA via 'vla.x.cfg' where 'x' is the name of the array configuration. Here 'vla.a.cfg' is the VLA A-configuration (the python command os.getenv("CASAPATH").split()[0]+"/data/alma/simmos/" shows the directory that contains all array configurations that are packaged in CASA).

hourangle: is used to simulate observations at a specific hour angle. We use 'transit' for culmination.

totaltime: This is the time on source.

thermalnoise: We leave this parameter empty for this noise-less simulation.

graphics: 'both' will show graphics on the screen and save them as png files in the project directory.

overwrite: True will overwrite previous results; be careful when running multiple setups as the files may have different names and only the files with the same names will be overwritten.


The output of the simulation is shown in Figures 2 and 3. The first image is the sky coverage which shows clearly that the primary beam exceeds the size of the model image by far. The other outputs are explained in the caption of Figure 3.

Figure 2: Sky Coverage
Figure 3: Output of simobserve. Upper left: Elevation vs Time of the modeled source (blue). The vertical green line marks the transit time and the red bar is the time span of the simulated observations. Upper right: The VLA antenna positions in units of meters to the central VLA reference point. Lower left: The uv-coverage of the simulations (note that they do not account for the bandwidth). Lower right: the psf of the observation with fitted Gaussian major and minor beam sizes.


The task simanalyze can now process the newly created MS and image the data: 

# In CASA
simanalyze(project='psimvla1', 
                    image=True, 
                    vis='psimvla1.vla.a.ms',  
                    imsize=[192, 192], 
                    interactive=False, 
                    niter=0, 
                    weighting='briggs',  
                    pbcor=False, 
                    stokes='I', 
                    analyze=True, 
                    showuv=False, 
                    showpsf=True, 
                    showmodel=True, 
                    showconvolved=True, 
                    showclean=True, 
                    showresidual=True, 
                    showdifference=True, 
                    showfidelity=True, 
                    graphics='both', 
                    overwrite=True)

project: The project (directory) name, same as in our call for simobserve.

image: True will image the visibilities with the following sub-parameters.

vis: The input MS inside the project directory. Note that simobserve's notation is to use the project name appended by the antenna configuration file name.

imsize: the number of pixels for the image dimensions. We use 192 for each axis. By default simobserve will use the pixel size of the model image as the cell size for each simulated image pixel (3.11 milli-arcseconds in our case).

interactive: If set to True, the cleaning will be interactive to allow the step by step setting of clean boxes and clean depths (see the VLA imaging topical guide). We will turn this feature off.

niter: number of clean iterations. We start with a dirty map and set niter=0.

weighting: (cf. Image weighting schemes). We will use briggs which defaults to a robust = 0.5.

pbcor: If set to True the image will be primary beam corrected. In our case of a very large primary beam (see Figure 1), primary beam correction will introduce very small corrections, so we turn it off.

stokes: the Stokes polarization to be calculated; we want Stokes I.

analyze: When set to True, simobserve will perform basic analysis of the generated images and create hard copies of the following displays.

showuv: We set this to False as we already show the uv-coverage in the simobserve output (see Figure 3).

showpsf: Although we already have a plot of that in Figure 3, we show it here again to have the comparison to the other plots.

showconvolved: creates a plot of the model convolved with the synthesized clean beam.

showclean: shows the deconvolved image (in our case we do not perform deconvolution, but show the plot to be consistent with subsequent runs).

showresidual: a plot of the residual image after deconvolution.

showdifference: displays the difference between output cleaned image and input model sky image convolved with output clean beam, i.e., it shows remaining clean artefacts.

showfidelity: a fidelity image, as defined by [math]\displaystyle{ \frac{I}{|I-T|} }[/math] where [math]\displaystyle{ I }[/math] is the observed image and [math]\displaystyle{ T }[/math] the sky model, see the simanalyze CASAdoc help. Note that the fidelity image is unit-less, but a bug in simobserve still shows it as Jy/beam.

The resulting plots of our first simulation are shown in Figure 4. Since this is a dirty image only (niter=0), the image and residual are the same dirty image. We also did not turn on any thermal noise and the image corruptions are all due to the incomplete uv-coverage of our simulation.

Figure 4: Simulated Images for noise-less Q-band observations. Upper left: the psf of the observations with the clean beam parameters. Upper center: the sky model in Jy/pixel. Upper right: The sky model convolved with the clean beam. Bottom left: the simulated image in Jy/beam. Bottom center: the residual image from deconvolution. Bottom right: the fidelity image (note that a bug in CASA shows two color wedges for this plot; the inner one is the correct one to use).

Q-band, 128MHz bandwidth, noiseless image, 1hr integration time, A-configuration, cleaned

As a second step we will deconvolve this dataset. This could be done inside the same project, but for a separation of our use cases, we will recreate the MS with simobserve and modify the simanalyze call. 

# In CASA
simobserve(project='psimvla2', skymodel='ppdisk672_GHz_50pc.fits', inbright='3e-5Jy/pixel', incenter='44GHz', inwidth='128MHz' , setpointings=True, integration='2s',  direction='J2000 18h00m00.031s -22d59m59.6s',  mapsize= '0.78arcsec', obsmode='int', antennalist='vla.a.cfg', hourangle='transit', totaltime='3600s',  thermalnoise='', graphics='both', overwrite=True)

simanalyze(project='psimvla2', image=True, vis='psimvla2.vla.a.ms',  imsize=[192, 192], interactive=False, niter=1000, threshold='1e-4Jy', weighting='briggs',  pbcor=False, stokes='I', analyze=True, showuv=False, showpsf=True, showmodel=True, showconvolved=True, showclean=True, showresidual=True, showdifference=True, showfidelity=True, graphics='both', overwrite=True)

The only difference to the earlier call of simobserve is our setting of niter=1000, threshold='1e-4Jy' which corresponds to a maximum number of 1000 clean iterations and a clean threshold of 0.1mJy. The threshold was chosen based on the previous image to be a few times above the rms of the image, and the niter is set large enough to reach the threshold. The residual image is now flatter as expected and the bulk of the sidelobes disappear.

Figure 5: Same as Figure 4, this time, however, with a cleaned image.

Q-band, 128MHz bandwidth, 4mm pwv, 1hr integration time, A-configuration

We now add environmental effects with an atmospheric precipitable water vapor (pwv) of 4mm and a ground temperature for spillover of 270K. Such conditions are fairly good at the VLA site. We change the user_pwr and t_ground parameters in simobserve accordingly. The task only adds thermal noise due to the sky brightness, but not phase noise due to pwv variations across the array. The sm.settrop tool can be used for the more sophisticated tropospheric models. A description is given in Corrupting Simulated Data. 

# In CASA
simobserve(project='psimvla3', skymodel='ppdisk672_GHz_50pc.fits', inbright='3e-5Jy/pixel', incenter='44GHz', inwidth='128MHz' , setpointings=True, integration='2s',  direction='J2000 18h00m00.031s -22d59m59.6s',  mapsize= '0.78arcsec', obsmode='int', antennalist='vla.a.cfg', hourangle='transit', totaltime='3600s',  thermalnoise='tsys-atm', user_pwv=4, t_ground=270.0, graphics='both', overwrite=True)

The simobserve output is identical to Figures 2 and 3 since we did not change any parameters that influence the figure. Now let's analyse the MS. 

# In CASA
simanalyze(project='psimvla3', image=True, vis='psimvla3.vla.a.noisy.ms', imsize=[192, 192], interactive=False, niter=1000, threshold='1e-4Jy', weighting='briggs',  pbcor=False, stokes='I', analyze=True, showuv=False, showpsf=True, showmodel=True, showconvolved=True, showclean=True, showresidual=True, showdifference=True, showfidelity=True, graphics='both', overwrite=True)

The results are displayed in Figure 6. The noise is dominating the image as expected.

Figure 6: Output of simanalyze after injecting noise.


Comparison with the VLA Exposure Calculator

We can compare the simulation to the predicted VLA sensitivity via the VLA Exposure Calculator Tool (ECT). Using winter observing conditions (for good pwv), a medium elevation (see Figure 3), 128MHz bandwidth at a frequency of 44GHz, the ECT predicts a beam size of 0.044" and which is compatible to our simulations (see Figures 3 and 6). The ECT output is shown in Figure 7.

Figure 7: Exposure Calculation for 4mm pwr, 128MHz bandwidth

The rms provided in Figure 6 is over the entire image and includes the source. It is therefore not representative to the thermal noise. To obtain a better noise figure, we now produce an image with the same parameters as simanalyze, but with a larger image size directly via tclean. Since the briggs parameter in simanalyze defaults to robust=0.5, we will use the same value here. 

# In CASA
tclean(vis='psimvla3/psimvla3.vla.a.noisy.ms', imagename='psim3-bigimage', imsize=[640, 640], cell='3.11e-3arcsec',  specmode='mfs',  gridder= 'standard', deconvolver='hogbom', weighting='briggs', robust=0.5, niter=1000, threshold='1e-4Jy')

Open the image in the viewer: 

# In CASA
viewer('psim3-bigimage.image')

Measuring the noise statistics away from the source (Figure 8) gives a value of about 7e-5Jy/beam which is in very good agreement with the ECT predicted noise figure of 6.9e-5 Jy/beam.

Figure 8: Larger image of the simulation given in Figure 6, measuring the rms noise in the purple rectangle regions away from the emission. A few low-level sidelobes are still present, however.

Q-band, 8GHz bandwidth, 4mm pwv, 1hr integration time, A-configuration

To compensate for noise, we will extend our bandwidth from 128MHz to 8GHz. The current implementation of simobserve has no channelization so the simulation will not have the better uv-coverage that is obtained in multi-frequency imaging. Consecutive simulations with different spw's, and combined MS's, however, can achieve that and we will show how to obtain a multi-channel MS below. In addition, for large channel widths, one needs to be careful not to introduce bandwidth smearing. Our target is luckily near the phase center of the image and bandwidth smearing will be minimal. 

# In CASA
simobserve(project='psimvla4',   skymodel='ppdisk672_GHz_50pc.fits', inbright='3e-5Jy/pixel', incenter='44GHz', inwidth='8GHz' , setpointings=True, integration='2s',  direction='J2000 18h00m00.031s -22d59m59.6s',  mapsize= '0.78arcsec', obsmode='int', antennalist='vla.a.cfg', hourangle='transit', totaltime='3600s',  thermalnoise='tsys-atm', user_pwv=4, t_ground=270.0, graphics='both', overwrite=True)

The simobserve output again is identical to Figure 3. 

# In CASA
simanalyze(project='psimvla4', image=True, vis='psimvla4.vla.a.noisy.ms',  imsize=[192, 192], interactive=False, niter=10000, threshold='5e-5Jy', weighting='briggs',  pbcor=False, stokes='I', analyze=True, showuv=False, showpsf=True, showmodel=True, showconvolved=True, showclean=True, showresidual=True, showdifference=True, showfidelity=True, graphics='both', overwrite=True)

The image (see Figure 9) is clearly better defined, as expected from a broader bandwidth. We also decreased the cleaning threshold to 0.05mJy and added a few more iterations to reach this threshold.

Figure 9: Imaging with 8GHz bandwidth.


Multi-Term-Multi-Frequency Synthesis Imaging

In reality, the 8GHz bandwidth is not confined to a single channel but distributed across many channels. This results in a spread out uv-coverage as it depends on the projected baselines expressed in number of wavelengths, so channels are slightly displaced from each other. Combining the channels in a multi-frequency synthesis (mfs), therefore does not only increase the sensitivity, but also the image fidelity through a better defined psf. Channelization and mfs imaging also reduces bandwidth smearing effects.

As mentioned above, simobserve, naturally does not channelize the visibilities when the input image is not channelized.

In the following, we will channelize the MS and, to show the procedure, we will also apply a spectral index of [math]\displaystyle{ S\propto\nu^{-1.5} }[/math] at the same time.

To start with, we need to produce a multi-channel model image. The following script converts our current model image from fits to the CASA image format with importfits. We then create 64 planes, apply the spectral index, and fix the headers accordingly to bring them into consecutive frequency order. Each channel is specified to a width of 128MHz (real VLA data would have typical channel widths of 2MHz). Finally, our individual planes are combined into a cube with the ia.imageconcat tool method. 

# In CASA
%cpaste

importfits(fitsimage="ppdisk672_GHz_50pc.fits", imagename="ppdisk_Q_50pc.im")
stats=imstat(imagename='ppdisk_Q_50pc.im')
maximum=stats['max'][0]
ratio=3e-5/maximum
for x in range (0, 64):
     y=40+x*0.128
     immath(imagename="ppdisk_Q_50pc.im", mode='evalexpr', expr="IM0*"+str(ratio)+"*(("+str(y)+"/40)^-1.5)", outfile="ppdisk_Q_50pc_spx_"+str(x)+".im")
     imhead(imagename="ppdisk_Q_50pc_spx_"+str(x)+".im", mode="put", hdkey="crval4", hdvalue=str(y)+"GHz")
     imhead(imagename="ppdisk_Q_50pc_spx_"+str(x)+".im", mode="put", hdkey="cdelt4", hdvalue="128MHz")
comb=ia.imageconcat(outfile="ppdisk-combined_spx.im", infiles="ppdisk_Q_50pc_spx*.im",  axis=3, relax=True, tempclose=False, reorder=True, overwrite=True)
comb.close()

--

Using this model cube in simobserve will produce the channelized visibilities. 

# In CASA
simobserve(project='psimvla4mfs',   skymodel='ppdisk-combined_spx.im',  setpointings=True, integration='2s',  direction='J2000 18h00m00.031s -22d59m59.6s',  mapsize= '0.78arcsec', obsmode='int', antennalist='vla.a.cfg', hourangle='transit', totaltime='3600s',  thermalnoise='tsys-atm', user_pwv=4, t_ground=270.0, graphics='none', overwrite=True)

simobserve and simanalyze assumes that, if the model is a cube, the output will be a cube too. In our case, however, we will produce a single multi-frequency synthesis continuum image so we have to use other, general CASA tasks for imaging and display.

Let's first have a look at the new uv-coverage. Each channel is displayed in a different color in Figure 10: 

# In CASA
plotms(vis='psimvla4mfs/psimvla4mfs.vla.a.noisy.ms', xaxis='Uwave', yaxis='Vwave',coloraxis='channel')
Figure 10: uv-coverage of chanelized visibilities.

We will now create a continuum image with the mfs technique, and, to recover the spectral index, we will use the Multi-term Taylor expansion in frequency (mtmfs, see CASAdocs Deconvolution Algorithms). We will use nterms=2 to derive the spectral index. The spectral index map, however, is fairly sensitive to the scales in the image. A regular Hogbom clean based on point sources, typically creates maps with too steep spectral indices when applied to extended emission. To avoid this effect, we therefore use the multi-scale algorithm with a range of scales. 

# In CASA
tclean(vis='psimvla4mfs/psimvla4mfs.vla.a.noisy.ms', imagename='psimvla4mfs/8GHzmfs', imsize=[192, 192], cell='3.11e-3arcsec', specmode='mfs', gridder='standard', deconvolver='mtmfs', nterms=2, scales=[0,2,4,8,13,20,40],  niter=1000, threshold='2e-4Jy', weighting='briggs', robust=0.5)

As before, we are not attempting primary beam corrections as the image is very small compared to the primary beam (see Figure 2).


The images can be displayed with the viewer using the contour button to load the images with the contour overlay. Figure 11 is 8GHzmfs.image.tt0 and Figure 12 is 8GHzmfs.alpha, both located under the directory psimvla4mfs. In Figure 12, we show the contours of the image on the spectral index map to check for the signal-to-noise at each position.

Figure 11: Similar to the image of Figure 9, but using multi-frequency-synthesis (mfs) imaging.
Figure 12: The spectral index map with image contours overlaid.

As seen in Figure 11, this method should produce a somewhat better fidelity than the non-mfs image shown in Figure 9, although a direct comparison is difficult given the spectral index that we introduced. The spectral index map itself (Figure 12), unfortunately, is dominated by noise, even at the brightest regions, partly due to the pwv that we assumed. This results in a relatively unreliable spectral index map overall, although the values scatter around the spectral index of -1.5 that we inserted earlier.

Q-band, 8GHz bandwidth, 4mm pwv, 4hr integration time, A-configuration

Let's improve the depth of the observations even further by going to 4hr on-source integration, via totaltime='14400s' . Given that mfs imaging only marginally improved the image, for simplicity we will go back to using the original, unchannelized 8GHz bandwidth. 

# In CASA
simobserve(project='psimvla5',   skymodel='ppdisk672_GHz_50pc.fits', inbright='3e-5Jy/pixel',  incenter='44GHz', inwidth='8GHz' , setpointings=True, integration='2s',  direction='J2000 18h00m00.031s -22d59m59.6s',  mapsize= '0.78arcsec', obsmode='int', antennalist='vla.a.cfg', hourangle='transit', totaltime='14400s',  thermalnoise='tsys-atm', user_pwv=4, t_ground=270.0, graphics='both', overwrite=True)

Figure 13 shows a better uv-coverage (lower left image) and the 4hr observation is also marked by a wider red bar in the top left image. This also improved the shape of the psf.

Figure 13: Output of simobserve with 8GHz bandwidth of 4hr of on-source observing time.
# In CASA
simanalyze(project='psimvla5', image=True, vis='psimvla5.vla.a.noisy.ms',  imsize=[192, 192], interactive=False, niter=1000, threshold='2e-5Jy', weighting='briggs',  pbcor=False, stokes='I', analyze=True, showuv=False, showpsf=True, showmodel=True, showconvolved=True, showclean=True, showresidual=True, showdifference=True, showfidelity=True, graphics='both', overwrite=True)

Given the expected deeper image, we reduced the cleaning threshold to 0.02mJy, and the result is shown in Figure 14. The image now is an almost perfect representation of the true sky model convolved with the clean beam.

Figure 14: Images after increasing the integration time to 4hr.


Comparison with the VLA Exposure Calculator

We compare again the simulation to the predicted VLA sensitivity via the VLA Exposure Calculator Tool (ETC). Using 8GHz bandwidth 4hr on-source, and a low elevation due to the longer observations, the ECT predicts a rms of ~7.1[math]\displaystyle{ \mu Jy }[/math] (see Figure 15):

Figure 15: Exposure time calculation for 4mm pwr, 8GHz bandwidth

Using the same settings but a medium elevation results in ~5.4 [math]\displaystyle{ \mu Jy }[/math]. Note that the ECT switched to the 3-bit samplers to accommodate the large bandwidth. The simulator, however, assumes 8-bit samplers, which provide about 15% better sensitivity than the 3-bit samplers of the VLA. For a discussion, see the VLA Samplers page.

Similar to the case of 128MHz bandwidth 1hr case, we produce a larger image to be able to measure the image rms in a signal-free region. 

# In CASA
tclean(vis='psimvla5/psimvla5.vla.a.noisy.ms', imagename='psim5-bigimage', imsize=[640, 640], cell='3.11e-3arcsec',  specmode='mfs',  gridder= 'standard', deconvolver='hogbom', weighting='briggs', robust=0.5, niter=10000, threshold='2e-5Jy')

Open the image in the viewer 

# In CASA
viewer('psim5-bigimage.image')

Measuring the noise statistics away from the source (Figure 16), gives a value of about 6mJy/beam which is slightly better but in very good agreement with the exposure calculator predictions between the two elevation settings.

Figure 16: Larger Image of the 8GHz simulation in Figure 14, measuring the rms noise in the purple rectangle regions away from the emission.



Q-band, 8GHz bandwidth, 4mm pwv, 1hr integration time, C-configuration

Our last Q-band simulation will use a pwv of 4mm again, but we will now simulate the observations with the VLA C-configuration (specified with the antennalist parameter). 

# In CASA
simobserve(project='psimvla6',   skymodel='ppdisk672_GHz_50pc.fits', inbright='3e-5Jy/pixel',  incenter='44GHz', inwidth='8GHz' , setpointings=True, integration='2s',  direction='J2000 18h00m00.031s -22d59m59.6s',  mapsize= '0.78arcsec', obsmode='int', antennalist='vla.c.cfg', hourangle='transit', totaltime='3600s',  thermalnoise='tsys-atm', user_pwv=4, t_ground=270.0, graphics='both', overwrite=True)

As expected, the extent of the uv-coverage decreases by a factor of about 10 and the synthesized beam size increases by the same factor, as seen in Figure 17.

Figure 17: Output of simobserve using the C-configuration.
# In CASA
simanalyze(project='psimvla6', image=True, vis='psimvla6.vla.c.noisy.ms',  imsize=[1728, 1728], interactive=False, niter=1000, threshold='1e-4Jy', weighting='briggs',  pbcor=False, stokes='I', analyze=True, showuv=False, showpsf=True, showmodel=True, showconvolved=True, showclean=True, showresidual=True, showdifference=True, showfidelity=True, graphics='both', overwrite=True)

Note that the output file names will now change and reflect C-configuration in their names. The resulting images are displayed in Figure 18, where the images now look like a point source with the degraded resolution. The surface brightness sensitivity of the observations, however, increases and we can clean to a lower threshold of 0.1mJy. simobserve automatically detected that the beam is massively oversampled and that a larger image size will provide a better field of view.

Figure 18: C-configuration results

The sharp edges in some of the images show the boundaries of the original model image which has non-zero for all pixels. The increased field of view was simply padded with masked values.

To combine the data of two different array configurations, we refer to the VLA Data Combination Guide.

X-band, 4GHz bandwidth, 4mm pwv, 1hr integration time, A-configuration

Finally, we change the frequency and perform an X-band (10GHz) A-configuration simulation. Since X-band has a maximum bandwidth of 4GHz, we will reduce it to this value. We will also lower the flux density of the model to 0.01mJy given that the brightness of the model will likely be lower, too. 

# In CASA
simobserve(project='psimvla7',   skymodel='ppdisk672_GHz_50pc.fits', inbright='1e-5Jy/pixel',  incenter='10GHz', inwidth='4GHz' , setpointings=True, integration='2s',  direction='J2000 18h00m00.031s -22d59m59.6s',  mapsize= '0.78arcsec', obsmode='int', antennalist='vla.a.cfg', hourangle='transit', totaltime='3600s',  thermalnoise='tsys-atm', user_pwv=4, t_ground=270.0, graphics='both', overwrite=True)

When expressed in wavelengths, the uv-coverage and synthesized beam are somewhere in between that of Q-band A-configuration and C-configuration as expected (Figure 19).

Figure 19: Output of simobserve for X-band observations.
# In CASA
simanalyze(project='psimvla7', image=True, vis='psimvla7.vla.a.noisy.ms',  imsize=[192, 192], interactive=False, niter=1000, threshold='3e-5Jy', weighting='briggs',  pbcor=False, stokes='I', analyze=True, showuv=False, showpsf=True, showmodel=True, showconvolved=True, showclean=True, showresidual=True, showdifference=True, showfidelity=True, graphics='both', overwrite=True)


The larger primary beam of the X-band data is shown in Figure 20, where the model image is barely visible in the center. The simulated images are shown in Figure 21. The ring shape is still visible in the results although it is quite blurred. We note that the noise is considerably lower as expected from the lower band noise properties of the receiver and atmosphere. We adjusted the clean threshold accordingly. Again, simanalyze was overriding the image size as the psf would not have been adequately covered for the lower frequency.

Figure 20: X-band sky coverage
Figure 21: X-band simulation results


Last checked on CASA Version 5.4.0.

Retrieved from "https://casaguides.nrao.edu/index.php?title=Protoplanetary_Disk_Simulation_-_VLA-CASA5.4.0&oldid=26888"