Module convolve_uv 

FFT-based UV-plane beam convolution.

This is a port of racs_tools.convolve_uv.convolve and racs_tools.gaussft.gaussft. The “robust” mode is implemented: the FT of the convolving Gaussian is computed analytically at each UV point (no kernel image needed), which handles NaNs gracefully.

The convolution is generic over the floating-point element type FftFloat (f32 or f64): the transforms run in the same precision as the input image, so f32 data (the common radio-astronomy case) is transformed in f32 — roughly half the memory traffic and compute of an f64 transform — while genuine f64 data is honoured exactly. Plans and scratch buffers are reused across calls via FftPlans, which matters when convolving every channel of a cube at the same image size.

Structs

ConvolutionResult

FftPlans

Cached FFT plans for a fixed (nrows, ncols) image size.

Enums

ConvolveError

Traits

FftFloat

Floating-point element type a convolution can run in: f32 or f64.

Functions

convolve_uv

Convolve image from old_beam to new_beam in the UV plane.

convolve_uv_with_plans

Like convolve_uv, but reuses pre-built FftPlans instead of planning per call. plans must have been built for image’s dimensions.

fftfreq

numpy-compatible fftfreq(n, d).

gaussft

Compute the UV-plane filter that deconvolves old_beam and re-convolves with new_beam. Direct port of racs_tools.gaussft.gaussft.