Open rlbyrne opened 5 years ago
Do you have the xy real and imaginary images?
I'm wondering if somehow the x vs y absolute (global) phase offset that is fit on the data to minimize V is not getting applied to the model visibilities, and might create this behavior?
Here's the real XY model image: and the imaginary:
For reference, here is the real XY dirty: and imaginary:
@miguelfmorales I don't think we should be applying calibration solutions to models.
The changes in PR #166 seem to resolve the imaging artifact in Stokes V. Here are the polarized models created from that branch: Stokes I: Stokes Q: Stokes U: Stokes V:
The model used in this test is completely unpolarized. Do we expect to be seeing this much model power in Q and U?
The brightest source in the images I posted is Fornax and is therefore modeled as an extended source. We were wondering if the power seen in Q and U is from the source being extended (could nearby component sidelobes add coherently to produce false power in Q and U?). I reran this simulation with just one point source with Fornax's position and flux. Here're the results: Stokes I: Stokes Q: Stokes U: Stokes V:
The colorbar has changed since the last simulation so it's hard to tell how much better Q and U are doing here. There's definitely a feature in linear pol at the source location. Would we expect the source to be completely reconstructed in Stokes I? Could we be getting intrinsic errors in the other Stokes modes just because of the finite size of the degridding kernel?
Looking at the ratios of the colorbars, I think the Q&U are much lower relative to I than they were for Fornax. I'm not sure whether the residual flux in Q&U are a concern or not.
The power at the source in Stokes Q is at ~1% of I. The hole in Stokes U is a fraction of a percent of I.
Hi all,
As I had the RTS simulator open I thought I'd see what we get using a shapelet model of Fornax - here I've simulated just Fornax A at about 185MHz for phase I config, output uvfits and then shoved them through WSClean (CLEANed but with no beam correction). Long story short it looks like we see essentially the same results, with Stokes Q at about 1%, and a negative hole of < 1% in U. I don't understand why we see leakage at all, given you get leakage from a mismatched beam model (to my understanding), and the beam model is internally generated in the RTS so it should all be consistent, but it's there.
[image: comparison_all_baselines.png]
On Thu, 25 Jul 2019 at 07:10, Ruby Byrne notifications@github.com wrote:
The power at the source in Stokes Q is at ~1% of I. The hole in Stokes U is a fraction of a percent of I.
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/EoRImaging/FHD/issues/164?email_source=notifications&email_token=AB4O3LDF5X3BQ2YIUOJX2B3QBDONNA5CNFSM4H7KCOF2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD2X3OEY#issuecomment-514832147, or mute the thread https://github.com/notifications/unsubscribe-auth/AB4O3LAC2OJV4PT6H7YWAHLQBDONNANCNFSM4H7KCOFQ .
That's very interesting @JLBLine! Can you try adding the image to the thread again? It didn't come through.
@rlbyrne Can you get the same fractional power numbers for your Fornax model?
Oh strange, ok I'll insert and attach it hopefully it comes through
[image: comparison_all_baselines.png]
On Thu, 25 Jul 2019 at 20:34, Bryna Hazelton notifications@github.com wrote:
@rlbyrne https://github.com/rlbyrne Can you get the same fractional power numbers for your Fornax model?
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/EoRImaging/FHD/issues/164?email_source=notifications&email_token=AB4O3LC57LRXLRK3POD7LP3QBGMUZA5CNFSM4H7KCOF2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD2ZKUAQ#issuecomment-515025410, or mute the thread https://github.com/notifications/unsubscribe-auth/AB4O3LDR3KPXJUVAPESALGDQBGMUZANCNFSM4H7KCOFQ .
No it still didn't. Did you do it on GitHub or through email?
Ah through email, I'll try github now...
Thanks! That's very interesting
Thank you, @JLBLine! This is super interesting. The fact that it agrees so well with our Fornax-as-a-point-source simulation suggests that this might be a "real" systematic (instead of just an FHD implementation bug). We're worried that our extended Fornax model simulation doesn't look the same (e.g. Fornax is positive in U). I'm running some tests to hunt this down.
In the meantime we merged PR #166 because it definitely resolved a couple bugs, but let's keep this issue open until we figure out why we're modeling Stokes I power in Q and U.
We're worried that our extended Fornax model simulation doesn't look the same (e.g. Fornax is positive in U).
Ignore this part. I was simulating the wrong model here (I was modeling Fornax with polarized components).
I tried to do two tests to figure out why both my and Jack's unpolarized Fornax simulation has power in Q and U.
First, I tried switching the X and Y beams in FHD. Here's the result: Stokes I: Stokes Q: Stokes U: Stokes V:
With the correct beams we were seeing positive power at the source location in Q and negative in U. Now we see the reverse: a hole in Q and positive power in U. I think this indicates that the power reconstructed in Q and U comes from asymmetries between the X and Y beams.
I also tried and failed to set the Y beam equal to X beam (my output images were very weird). If I get that test working it'll help confirm this.
Could you change the color stretch to +/- 2? The current Stokes Q and U images are very faint.
I replotted just the area around the source with colorbar limits of +/- 2. Here's what I got for the regular run: And the flipped beam run:
This is related to #40
Imaging errors occur in Stokes V (and possibly other polarizations) for models of Stokes I and polarized sources. These errors are not visible when all baselines are imaged but become clear when only short baselines are imaged.
Here is an example of the Stokes V model image created only with baselines shorter than 50 meters. The bright source is Fornax A, but the dipole structure seen here appears for other bright sources as well. All sources in this model are unpolarized.
The calibrated data does not show this Stokes V structure. Here is the dirty Stokes V image made with baselines shorter than 50 meters: