Consider implementing delay computations

[kiranshila]

Hey! In piecing together the code for our telescope, we are going to need a set of routines to compute delays to perform radio interferometry. We have an excellent engine written by a colleague here, but were wondering if this is the kind of thing that should/could be in SuperNOVAS, considering SuperNOVAS already has all the frame and time transformation logic. Would this be something you'd be interested in? I think CALC11 is the other "standard" codebase used for such things.

[attipaci]

Hi Kiran!

I fully understand the need for interferometric delays. We have the same application at the SMA (for our 8 antennas). I've been slowly thinking about migrating some of the interferometric code into SuperNOVAS in a generic way. For example, the novas_xyz_to_uvw() / novas_geodetic_to_cartesian() functions are a step towards that goal (the w coordinate being the delay for an absolute or relative x,y,z station coordinates for a source observed at apparent HA/Dec).

Interferometric delays are a tricky thing though. The above novas_xyz_to_uvw() gives geometric delays for the stations. But then you need to add delays specific to the cables wiring the stations, the receivers within, for the correlator electronics, digital delays, and for atmospheric refraction (for the observing wavelength). These components can be defined in different ways, so I'm not completely sold on having a catch-all solution that works for all interferometers the same. We should think about what if any of those extras can be generalized for SuperNOVAS. I'm certainly not against that so long as it's generic, and as long as we can produce robust unit testing for all of it...

[attipaci]

I'll convert this into a discussion...

[kiranshila]

So the code I linked I believe is pretty generic - insofar that I don't think there's anything specific about wavelength / telescope setup / etc. So, here I'm talking strictly of geometric delay from baseline projection plus gravitational delay (monopole) from massive planets. The linked code also includes quadrupole terms, but (at least for us), I doubt they matter. The tricky thing I don't know immediately how to solve is for the rates. In the linked code, they use Jax autodiff to simply take the derivatives of the delay functions. While this is possible in C (with Enzyme, a project that I really love), I wouldn't want to add a rather complex external dep and require LLVM haha.

I might have a go at trying to just implement that _far_field_delay function in C, using your transformations.

[attipaci]

Hi Kiran,

Take a look at PR #289. It's a work in progress, but it provides better support of interferometric projections (including geometric delays) for both the C99 API and for the new C++ API. I still have to add unit tests for the new functions / methods, so it will be a few days at least before the PR is 'ready'.

The nice part of this is that the projections and delays don't just apply to the far field. You could use these function / methods to calculate projections and delays even for objects that are relatively close with respect to the interferometric baselines, where each station has a somewhat different line of sight.

Let me know if this helps, or if there is more you'd like to see...