Neural network circuit implementation using operational amplifiers and digital potentiometers

Abstract

Implementations of Artificial Neural Networks (ANN) have been advancing for almost three decades and their importance has been marked by the different methods used in their construction, their applications, and comparisons in terms of speed, costs, and performance between implementations made by software and hardware. As analog implementations of ANN have been shown to have good levels of performance, high processing speed, low power consumption, small size, and low cost, they have played an important role in the development of new designs. This work presents a proposal to design a circuit implementation of an ANN by using Operational Amplifiers (Opamps) and digital potentiometers to create a network that can be trained by using an external training system. This, based on circuit analysis and training algorithm by the back propagation (BP) approach.

The proposed design will be simulated in the circuit simulator Proteus. The circuit is tested using the logical gates benchmark problem to verify its performance with the BP learning algorithm.

The results of this work demonstrate that it is possible to create a neural network using analogous components. Furthermore, it shows good performance when implementing the training algorithm using digital potentiometers. As future work is expected to improve the performance of training to create a controller based on neural networks and thus, perform the control of a dynamic system.