The paper you are referring to is addressing many subjects that are indeed covered by the new IPF version and part of the material is actually taken from the processor algorithm document. For example:
- II.A : denoising in range
- II.B and III.A - denoising in azimuth. The processor was having natively (i.e. since launch) the possibility to denoise in azimuth but it was an option not activated and the annotations wasn’t providing the azimuth noise vector. With the new IPF this is now the case. The description in the paper looks however correct
The noise power scaling which has been an issue for us (up to the new processor version) and the paper is actually showing well where we were having problems (III.B/C ). The paper is proposing a way to correct for the processor noise scaling issues based on the residual of the demonised signal. The approach can work OK but I believe it is limited to data over ocean. I am not quite sure how it will work over land scenes where we know that even away from noise floor the data is impacted by it .
The new IPF version is the result of a long activity aiming at providing more reliable denoising information everywhere (not only over ocean where it is relatively easy) that I can summarise in 3 steps:
NESZ estimation over the Doldrums, where we have analysed several thousands of bursts to find the few signal-free ones hence giving only noise. This allows to have the best NESZ estimation possible that we use to scale our range profiles. I expect that now the “swath balancing” issue is gone
Find a way to track the evolution of the noise power with the data-take to cope with the earth brightness temperature. In the past we were using only 2 values located at the start/end of the data take. The linear interpolation made between those two point can’t cope with the land/sea or sea-ice transition hapenning.
change the IPF to provide (relatively) easy to use denoising annotations
This is sumarised in http://sarcv.ceos.org/site_media/media/documents/04-esa-s1-wgcv-noise3.pdf.
In order to do that we needed to change the format. The new format introduces the azimuth vectors in the simplest way possible without completely breaking everything. Indeed the format changes are pretty minimum and doesn’t reflect the complexity behind especially for GRD.
The difficulty for GRD relies on the fact that the burst/swath are merged and effective denoising needs to know where the boundary are. Section III.D , overlook the fact that it was almost impossible with the previous annotation to identify burst/swath transition leading to artefacts like fast ripples at swath transitions as you can see them in fig 9. Those should be pretty reduced now.
Furthermore, the swath merging was not constrained and the merging point was constantly varying in azimuth to cope with the roll pointing changes. In the case of EW which has 5 swaths and are 1 min long it was leading to >70 blocks of variable size (some ridiculously small) in where to apply a proper azimuth/range denoising.
We reviewed all that to reduce the complexity on the user side, limiting the number of block to the extreme minimum. Like we did in for range we further add extra point in the azimuth vector at burst transition to avoid any interpolation error. The TN we made available explain how to retrieve them exactly.
The section III.E are also what we recommend to do. We limit ourselves to provide as accurate as possible vectors. Any local correction is let to the user for its own application.
In short, the approaches are complementary. The new products should ease and improve the results obtained in this paper.