Buletinul Ştiinţific al Universităţii "Politehnica" din Timişoara Seria ELECTRONICĂ şi TELECOMUNICAŢII TRANSACTIONS on ELECTRONICS and COMMUNICATIONS Tom 49(63), Fascicola 1, 2004 Adaptive Interfaces Based on FPGA Implemented Artificial Neural Network Ştefan Oniga 1 , Virgil Tiponuţ, Atilla Buehman 1 , Daniel Mic 1 Abstract - The goal of this work is to build smart interfaces with learning and adaptive capability. The key element of the learning and adaptive behavior are artificial neural network (ANIN) blocks, implemented in FPGA using the System Generator tool for Simulink developed by Xilinx Inc. This tool allow the easy generation of hardware Description Language (I1DL) code from a system representation in Simulink. This VHDL design can then bc synthcsized for implementation in the Xilinx family of FPGA devices. The off-chip learning task is performed using Matlab and the ANN's wcights are transferred automatically from Matlab workspace to weights memory. Keywords: smart, neural network, adaptive, learning, FPGA, prosthetic I. INTRODUCTION The efforts made world wide by the large numbers of universities and research organizations that are involved in designing and building natural uscr interfaces it seems to be not enough because of the lack of adaptation and learning capabilitics. The use of neural networks to add learning and adaptive behavior to smart sensors is cssential and the FPGA implementation is an easy an attractivc way for hardware implementation. Among possible applications are intelligent computer peripherals enabling people with any kind of handicap to use computer and communicate, as any kind of industrial or domestic device with learning and adaptive capabilities. The goal of this work was to devclop hardware- software codesign platform enabling the fast development of smart interfaces with the addition of sensors, hardware modules that can bc easily connected and VHDL modules that can manage sensors, basic behaviours (ex: features extraction, pattern recognition, etc). Using this framework development of new smart devices needs only design and synlhesis of new VHDL drivers for the new sensors and new application-specific ANNs. This platform is based on low cost general purpose FPGA boards wilhout need for hardware design. This paper presents a new method for hardware implementation of artificial neural networks (ANN) in field programmable logic devices (FPGA) that can be uscd in smart sensors development. It also permits the development of the ANN's specific modules and libraries for System Generator tool. Main applicalions for such smart devices with embedded and hidden intelligence at user are in the prosthetic, automotive, "domotic" and automation field where the trend is to produce easy-to-use devices II. THE HARD WARE-SOFTWARE CODESIGN PLATFORM Smart devices must use multisensorial interfaces with natural, adaptive behavior and learning capability. The kcy for the adaptive and learning behavior are VHDL described neural networks. Any application of a new smart device should use these ANN modules to add adaptive and learning capability. The platform developed in order to provide a fast prototyping environment for adaptive interfaces is shown in Fig.I. and it was developed to facilitate the use of codesign teehniques. Other requiremcnts for the development platform are: • Lxchangeability of sensors, thanks to common interfaces for any class of VHDL drivers • Reusability of developed VHDL components • Reduced time to market The Aduc812 microcontroler is used to implement the Data Acquisition System and to adapt signal sensors to neural network input requirements. The reconfigurable device (XC2S50 Xilinx) is used to implement the neural networks and other logic blocks of the same application. The System Generator tool for Simulink developed by Xilinx Inc. allow the easy generation of hardware Description Language (HDL) code from a system representation in Simulink. This VHDL design can then be synthesized for implementation in the Xilinx family of FPGA devices. ' Nortli University of Baia Mare, Electrotechnical Department Dr. V. Babeş Slr., Nr. 62/A, 430083 Baia Mare. e-mail onigas@ubm.ro 2 Electronics and Telecommunications Faculty, Applied Electronic Department Bd. V. Pârvan Nr. 2,300223 Timişoara, e-mail, tiponuUoietc.utt.ro 236