This paper presents a novel method for modeling the functionality of a mixed-signal system, and analyzing its signal integrity (SI) at a high-level of abstraction with SystemC-AMS. Our model includes on a unique platform a functional module and a non-functional module. The functional module represents the operative behavior of the system and the non-functional module, based on neural network techniques, displays the SI characteristics of the system. The proposed method is demonstrated by modeling field bus communication system with two nodes. We achieved an error of about 3% for the neural network based Time Data Flow (TDF) model with respect to a RLC Electrical Linear Networks (ELN) model.