terrain flattening tries to correct the radiometric distortions caused by different local incidence angles.
That means, that originally areas facing the sensor are pictured brighter and the other ones darker. Terrain flattening then uses a DEM to re-calculate the actual radiometry based on the area illuminated per pixel.
If your DEM is bad, you could make it worse, theoretically - especially at ridges facing towards or away from the sensor. You can try the ALOS World DEM for a comparison. It also has 30 meters but was made from optical stereoscopy
I made a similar comparison and found the best results gained from Terrain Flattening on GRD or SLC data:
Another thing is that creating Sigma out of Gamma is not a good idea. At least to my knowledge, Gamma0 is the final stage that can be reached. Why going back to Sigma?