InSAR phase of TanDEM-X interferogram is negative

I process a large amount of TanDEM-X CoSSC data ( >1000 image pairs) and for some image pairs, the phase of the interferogram is mirrored to the negative side, i.e., I can adjust it by multiplying the distribution with -1 in order to “fix it” and finally obtain positive heights. I thought it was due to which satellite that is master/slave in the bistatic setup, but lately I have noticed that this does not consistently solve this problem for all pairs. Hence I wonder if someone else noticed this, and moreover, could suggest a solution?