Drifting subpulse patterns in pulsar signals are frequently interpreted in terms of a model in which a rotating ring of sparks on the polar cap gives rise to emission from regions of the magnetsophere connected to the sparks by dipolar magnetic field lines. The spacing of drift-bands in time depends on the circulation rate of the polar cap pattern, but to first order the longitudinal phase dependence of the subpulse modulation should obey a frequency-independent relation determined by the geometrical configuration in a similar manner to the polarization position angle. We present here observations at 272-1380 MHz of PSR B0320+39 and PSR B0809+74, both of which show nearly linear drift in two longitude regions, separated by a region of reduced modulation and accompanied by a large step in the phase of the subpulse pattern. We show that the observation of Bartel et al. (1981) that the subpulse spacing for PSR B0809+74 was 1.8 times greater at 102.5 MHz than at 1720 MHz is most likely an artifact of the phase step, which is only present at high frequencies. The phase steps can be understood as a consequence of observing overlapping offset images of the polar cap spark pattern. We also detected more complicated, frequency-dependent behaviour that would require that the images do not simply differ by rotation about their centers. Detailed modelling of non-axisymmetric refraction or distorted magnetic fields is suggested as a means of pursuing an explanation for this phenomenon.