IMPLEMENTATION OF A NEURAL-NETWORK CONTROLLER AND ESTIMATOR USING A DIGITAL SIGNAL-PROCESSING CHIP

In this paper, artificial neural networks with error backpropagation a re considered for the control of linear and nonlinear dynamics. A neur al network estimator and controller is constructed and trained off-lin e to learn the dynamic behavior and satisfactory control. The neural n etwork estimator generates estimates of the plant output and also prov ides the plant Jacobian for the neural network controller. For real-ti me control purpose, the neural network estimator and controller is imp lemented in assembly language using a Motorola DSP56001 digital signal processing chip. The time to convergence can be shortened by utilizin g the computational speed of the chip. The nonlinear activation functi on of the neural network is approximated and stored as a look-up table . Simulation results between the DSP and C language versions agree wel l without any noticeable degradations.