You don't seem to be using the covariance of the value when you compute your random numbers, which is why things are messed up. There's nothing special about complex numbers; this is equivalent to trying to sample from a multivariable distribution.
If you know this is normal with a given covariance, you can use mvnrnd.
If it's something else, the best way to do it would be to sample real number first, then using the conditional PDF, sample the complex number afterwards. You need to use use the conditional PDF, though, not the unconditional PDF (which I think is what's going on here)
