The gridding process involves convolution followed by re-sampling. The interpolation usually uses a convolution relationship to place (u, v) points on a regular grid having sides which are powers of two. Gridding is needed because an FFT requires power of two array sizes. The effect of gridding using a convolution function on the SNR is rather complicated (Crane and Napier, p. 160). Possibly useful gridding convolution functions include (1) pillbox (two-dimensional circular step), (2) truncated exponential, (3) truncated sinc, (4) exp*truncated sinc, and (5) truncated spheroidal function (Sramek and Schwab).
At the edge of the image, the SNR is reduced by aliasing noise into the image and dividing the image by the Fourier transform of the (u, v) plane convolving function.
The grid size is usually chosen such that the dirty beam is 3 to 4 pixels in diameter in order that the dirty beam be well-sampled but the number of CLEAN components kept to a minimum.
References
Crane, P. C. and Napier, P. J. "Sensitivity." Ch. 7 in Synthesis Imaging in Radio Astronomy: Third Summer School, 1988 (Ed. R. A. Perley, F. R. Schwab, and A. H. Bridle). San Francisco, CA: Astronomical Society of the Pacific, 1989.
Sramek, R. A. and Schwab, F. "Imaging." Ch. 6 in Synthesis Imaging in Radio Astronomy: Third Summer School, 1988 (Ed. R. A. Perley, F. R. Schwab, and A. H. Bridle). San Francisco, CA: Astronomical Society of the Pacific, 1989.
