Finite wordlength effects analysis of a digital neural adaptive tracking controller for a class of nonlinear dynamical systems

In this paper, analysis of the finite wordlength effects for the implementation of a digital neural networks adaptive tracking controller and the update law of the weight matrices for a class of nonlinear system is proposed. Both the quantization and computational errors are considered. In the sense of the Lyapunov stability, weight matrices of the neural networks and the error between the identified states and real system states will converge during finite time interval. Moreover, a robust stability condition can be obtained in terms of the mantissa bit number. Based on this criterion, the existence of the controller can be guaranteed