turbulence model for separated flow in diffuser
 

Hi
I am simulating 3D compressible flow in a diffuser (Mach number =4) ...The CCM+user manual says that SA model is not appropriate for separated flows. I have tried SST K-omega and realizable k-epsilon (RKE) models as well. SST K-omega does not start the diffuser...however, RKE does work. But shows huge separated regions at the exit. SA and RKE show similar separation patterns and pressure distibutions near the diffuser contracted regions. the main difference in the 2 models is at the exit.
I am unable to decide which of these models, SA or RKE, is more reliable in my study. A couple of other softwares, like NUMECA, give good results with SA for separated flows as well. Can the same be applicable with CCM+ also?
Please advice.