I think in the case of the simulation you are doing, you can reduce the filter’s cut-off frequency size to ” 0.3*bit rate ” (or even maybe a little less). In that case, the cut-off will be 15GHz and your channel spacing can even be 40GHz. You can use optical spectrum analyzers to observe how the channels are spaced and if that is enough or not. I hope I have made a clear point here.