One of the advantages of the MDWDF approach is that discrete numerical integration routines are arrived at by applying coordinate transformations and spectral mappings (or integration rules along the transformed coordinate directions) to the MD circuit form of the original model problem. Indeed, the principal result of Chapter 4, in §4.10 is that digital waveguide networks on regular grids can be constructed by essentially the same means.
On the other hand, the expanded signal flow graph for a digital waveguide network is a network of lumped 
-ports in its own right; in fact, the original formulation of the DWN (and TLM structures) is lumped. In the flow graph for a MDWD network, however, the port structure is lost. We showed in §4.9 how several DWNs, differing perhaps in grid density or the choice of coordinate could be joined through the use of passive interfaces. Here Fettweis' approach falters, because it is not clear how to generate an structure on an irregular grid from a MD representation (a similar problem, that of terminating a MDWD network in hexagonal coordinates (see §3.3.3) at a straight boundary has been discussed in great detail in [210], but in that case, it was necessary to resort to active elements even for a passive termination!)
A possible direction here might make use of boundary network modeling, as outlined immediately previously; i.e., treat an interface as a MD boundary network in its own right between two separate MDWD networks operating using different grid arrangements.