EoRImaging / FHD

Fast Holographic Deconvolution
BSD 2-Clause "Simplified" License
20 stars 10 forks source link

Discontinuities in frequency with diffuse HERA simulations #100

Open aelanman opened 6 years ago

aelanman commented 6 years ago

There are vertical discontinuities at (apparently) even spacing in waterfall plots from visibility simulations of diffuse emission, using using HERA beams and the GSM. For all of these nfreq_avg = 1024 and there are 1024 channels. Snapshots are 11 time samples and the total simulation covers 5 minutes.

So far I've determined:

The plots below are generated by taking 30min of simulated data, selecting the given baselines, averaging in time, taking the first derivative, normalizing each by its maximum, then shifting the curves up so they don't overlap.

bm2_14m 14m baselines with beam_model_version=2, with different orientations.

bm2_29m-ish Approx. 29.2m baselines with beam_model_version=2

bm2_ew-14m 14m east-west baselines with beam_model_version=2

bm2_waterfall Waterfall plot with beam_model_version=2

bm2_waterfall_uglier Bonus waterfall plot with extra jumpiness. beam_model_version=2

Figures below are all with beam_model_version=3. bm3_14m

bm3_29m-ish

bm3_ew-14m

bm3_waterfall

Each snapshot is only 11 time samples, and there are clear discontinuities between them.

I've also seen hints of this in the waterfall plots of EoR bubble simulations, which suggests to me that this might be related to how FHD handles healpix maps.

aelanman commented 6 years ago

An update --- These stripes do not appear when the instrument is set to MWA, so it seems to be caused at least in part by the beam model.

Another update --- These stripes don't appear when using EoR0_diffuse_model.sav instead of a full-sky GSM with a selection on it.

aelanman commented 6 years ago

I've run Adam Beardsley's code for plotting the high-resolution psf, and found that there is a sharp cutoff visible on the edges. This is especially obvious when I set a lower psf_resolution.

psf_resolution = 10: full_kernel_psfres-10

psf_resolution=100: full_kernel_psfres-100

Both of these are with beam_model_version=3 with HERA, on the beam_model_version_three branch. I haven't seen any sharp cutoffs with beam_model_version=2, so this might not be responsible for all of the discontinuities we're seeing.