svensl -
You can't define complex voltages such as V=a+j*b in VerilogA/MS.  There was some talk about adding complex support as an extension to better support RF, but this effort fizzled out.

zi_np's third argument, according to the LRM, is a vector of N pairs of real numbers, where the first number if the real part and the second the imaginary part.  If you put a non-zero imaginary part, then you have to put in the complex conjugate yourself (same real part, -imaginary part).  (One might argue that the simulator should add this in for you ... )

I've had some difficulties with simulators not really agreeing on what a "vector" is in some cases -- I was trying to use noise_table, and I wasn't able to use an array
 int noisevals[10];
I had to use a comma-separated list of numbers enclosed in braces {}, and the numbers had to be constants (fixed numbers, or I think I was able to use parameters).