Spectroscopic Table of Contents/Roadmap

[iancze]

On the road to implementing full-featured modeling for ALMA spectral line observations, there are several things to consider. This issue collects the relevant sub-issues, since some of the proposed solutions might need to address multiple issues.

Issues

Doppler setting + barycentric corrections #32

• Can we do all of the pre-processing ahead of the training loop with mstransform or other band-limited interpolation schemes?
• Does this have any disastrous affect on the noise properties?
• If yes, how do we properly model Doppler set data?

Window function and spectral covariance #18

• Modeling the covariance across data channels in the likelihood function
• Modeling multiple execution blocks with differing spectral resolutions

"Loose" (ungridded) visibilities #17

• Necessary if we're going to need to interpolate to TOPO frame for each integration to solve Doppler setting Doppler setting #32. The gridding destroys time information, though we could split the gridding up into multiple integration blocks
• Important for self-calibration as well

Gaussian process for spectral covariance inference

• Rather than modeling the covariance directly, just infer it