As previously shown, the direct extension of the impulse invariance principle to Volterra kernels has to be modified in order to provide a condition for the exact modeling of mixed-signal chains. At first sight this would seem to seriously complicate the otherwise simple discrete-time realization of separable kernels (among which bilinear kernels are of particular importance). We show here, however, that this not the case. By defining a cascade operator, the structure of a generalized impulse invariance can be unveiled, leading to a realization without an inordinate increase in computational complexity.