On the FACET facility at SLAC, the E203 collaboration uses Smith-Purcell radiation to reconstruct longitudinal beam profiles non-destructively. In order to increase the detected signal, the grating needs to be placed as close as possible from the beam, but without interfering with it. We report on a comparison of several reconstructions from the same dataset by using different values of the beam-grating separation and we give an estimate of the error in the profile determination due to position uncertainty. We also present how using a new motor system and a dedicated wire scanner can increase the accuracy of the beam-grating separation measurement.