Hardware/Software Co-design Approach for an ADALINE Based Adaptive Control System Shouling He Department of Electrical and Computer Engineering, Penn State Erie, Erie, PA 16563 Email: sxh63@ psu.edu Xuping Xu Department of Electrical and Computer Engineering, Penn State Erie, Erie, PA 16563 Email: Xuping-Xu@ psu.edu Abstract—In this paper, we report some results on hard- ware and software co-design of an adaptive linear neuron (ADALINE) based control system. A discrete-time Pro- portional-Integral-Derivative (PID) controller is designed based on the mathematical model of the plant. The pa- rameters of the plant model are identified on-line by an ADALINE neural network. In order to efficiently and economically implement the designed control system, a Field Programmable Gate Array (FPGA) chip is em- ployed to process the measured data and generate control signals. Moreover, a microprocessor is exploited to per- form the core computation of the ADALINE algorithm. Throughout the paper, we design and test the control system for a permanent magnetic DC motor. Our ex- periment results demonstrate the effectiveness of the pro- posed approach. It is worth noting that the experimental bed in the present paper can also be used as a low-cost general prototype to satisfactorily test adaptive control systems, owing to the benefit of software and hardware co-design. Index Terms — Hardware, software, ADALINE, FPGA, adaptive control. I. INTRODUCTION PID controllers have been widely used over the past five decades due to their simplicity, robustness, effec- tiveness, and applicability for a broad class of systems. Despite the numerous control design approaches that have appeared in the literature, it is estimated that nowadays PID controllers are still employed in more than 95% of industrial processes [1]. Likewise, adap- tive tuning of the parameters of PID controllers in nonlinear and time-varying systems has been of great interest to control engineers. With the introduction of on-line identification of process models and momen- tary tuning of PID gains, particularly with some state- of-the-art parameter predictive approaches, the compu- tational burden involved in the tuning becomes very heavy. Since a low-cost microcontroller with limited resources may not satisfy the requirements of many demanding applications, expensive and multifunctional microprocessors need to be used. Consequently, the challenges of software management and energy con- sumption cannot be avoided in real-time control sys- tems. On the other hand, with the rapid development of microelectronics, hardware design devices such as field programmable logic arrays (FPGAs) and complex pro- grammable logic devices (CPLDs) provide possibilities of fast and easy digital implementation of controllers in embedded systems. The strength of programmable logic devices is their high speed and parallel hardware implementation structures. Particularly, due to their digital nature, they can be conveniently processed in various digital manners. Furthermore, by varying the hardware configuration during operation time without changing the number of gates and the power demand, controllers based on such hardware devices can adapt to different operating conditions. Such a feature is par- ticularly useful for tolerance control with redundant function setups [2]. In view of these, such hardware technologies open the door for faster and more flexible control system design. Albeit attractive, pure hardware implementation does not provide a complete solution to system design. Due to the lack of common mathematical function li- braries for those hardware devices, control and system parameter identification algorithm have to be imple- mented in time consuming and expensive processes. Moreover, some identification and control algorithms may be better implemented serially in software. In this regard, we note that successful design of a practical control system should be characterized by the follow- ing: a fully developed adaptable and robust algorithm, efficient hardware design combined with flexible soft- ware structure. In this paper, we introduce a system design approach based on the aforementioned algorithm and implemen- tation guidelines. First, a PID controller is designed in discrete-time domain using the mathematical model of the physical plant. The gains of the PID controller, which are functions of the uncertain and varying pa- rameters of the plant, are identified by an ADALINE JOURNAL OF COMPUTERS, VOL. 3, NO. 2, FEBRUARY 2008 29 © 2008 ACADEMY PUBLISHER