correlated noise

[Geoffrey Coram]

According to the Verilog-A LRM, one can do the following:

real n;
n = white_noise(pwr);
V(a,b) <+ c1*n;
V(c,d) <+ c2*n;

However, many simulators were (at least initially) unable to
handle this code. Thus, model developers have taken to
creating internal nodes:

electrical n;
I(n) <+ V(n); // 1-Ohm resistor
I(n) <+ white_noise(pwr);
V(a,b) <+ c1*V(n);
V(c,d) <+ c2*V(n);

All simulators I know of will support that formulation,
and it may be that these same simulators are able to
automatically convert the first formulation at least
into the second.

Obviously, when not doing a noise simulation, the
second formulation is wasteful; I believe there are
also simulators that can compute noise correlations
without introducing the extra node.

Should we recommend that model developers stop using the
(potentially) wasteful formulation?

-Geoffrey