A Real Time Controller Based On a Pulse Stream Neural System

1 This paper describes a real-time signal processing system based on an Arti cial Neural Network chip, which has been developed for and tailored to non-linear control applications. The chip described, which uses Pulse Stream computation principles, has been designed and manufactured using a standard 1.5 m digital CMOS technology. The system can compute up to 140MCPS, with a Nyquist frequency of about 70kHz. The prototype of a controller board containing one such chip has been developed and fully tested. The board can be directly interfaced to a host computer and the whole system supports neural and cognitive training algorithms. The system is currently being used to assess the performance of these algorithms for realtime control in several applications.