SLC interpolation method

[mattiacallegari]

I changed the interpolation method in the SLC coregistration method from BILINEAR to BISINC21 in all the SNAP graphs. BISINC21 is the suggested interpolation for SLC, but it uses more computational resources. This needs to be checked in terms of performance.

Reference SNAP documentation:
Generally the accurate 21-point truncated sinc interpolation is encouraged to use for complex image interpolation. This, of course, will take longer processing time than using other interpolation method. With other interpolation method, however, some oscillation effect could be introduced in the interpolated image and in the coherence image generated later on.

[jdries]

@mattiacallegari we indeed see a major performance impact after this change.
It could also be an option to turn this into a parameter? Then we can more easily compare it in production, and even allow advanced users to fall back to a cheaper version if desired.

[mattiacallegari]

Yes, the interpolation method could be specified as an input parameter. The available interpolation methods in the coregistration step (S-1 Back Geocoding) are:

• nearest neighbour
• bi-linear
• cubic convolution
• 5-points bi-sinc
• 11-points bi-sinc
• 21-points bi-sinc
• bi-cubic

In the future, I think the best would be to use an already coregistered dataset as input. Indeed, every burst could be coregistered with a (virtual) reference and create a new coregistered SLC dataset. Some work in this direction is currently developed here