Especially if you want to compare images of different dates, multi-temporal speckle filtering is not a good idea. It wouldn’t be effective on three images either.
Coregistration only makes sense to me if you want to compare single pixels for their properties. For general scatter plots (e.g. incidence angle vs. Sigma0), this is not required.
If you want to compare backscattering coefficients (your first workflow), the normalization is no longer required. It is kind of a relic from earlier versions (see lveci’s answer here)
If you want to correct radiometric errors caused by topography, you should run Terrain Flattening (as suggested here), but its results strongly depends on the used DEM.
I would select the SRTM 1 Sec for the Terrain Correction because of its higher spatial resolution.