I did not mean that you should replicate your DLL with discrete blocks, I meant that the problem you have can be replicated with PLECS native blocks. To me that indicates that the current methods are not very appropriate to do what you need.
In any case the integrator is quite tricky, probably any other model would give less issues (e.g. low-pass-filter). When the analysis provides perturbations in multiples of Fs/2, you will be sampling DC value, this is sent to the open loop integrator and the system will not have a Steady-state (which makes the AC sweep and impulse response to fail).