If the source is intrinsically phaseless (i.e. symmetric), then the measured phase should be 0 **where there is signal** (otherwise it should be random if the data reduction algorithm worked correctly). You can now estimate the S/N either by looking at channel to channel fluctuations in visamp or by looking at the phase fluctuations (channel by channel). If we assume that the source is phaseless then we should use the real part of the visibility, i.e. corr.visamp \cdot \cos(corr.visphi). If the phase varies just a little bit around 0, e.g. 20 degrees this is only a few percent effect, but the larger the phases fluctuation become the larger the fluctuations in the real part of the visibility will become.