Introduction

The first step in SAR interferogram processing is often the resampling of one complex SAR image to map it onto a second image to within an accuracy of about a tenth of a resolution element.

Although the implementation may be different, resampling can be viewed as consisting of two steps:

1.   reconstruction of the continuous signal from its sampled version by convolution with an interpolation kernel i(x,y), and
2.   sampling of the reconstructed signal at the new sampling grid.

This scheme holds even in many cases, where the convolution (step 1) is not obvious. For example, nearest neighbor and Lagrange-type interpolation of equidistantly sampled data can be considered as a convolution with particular kernels. The choice of the interpolation kernel (especially its length) requires a trade-off between interpolation accuracy and computational efficiency. This brief communication shows that straightforward system theoretical considerations give objective criteria for choosing or designing interpolation kernels for interferometric processing.