Current-Mode Neural Networks For Solving Nonlinear Programming Problems

Abstract

In this paper, neural networks for online solution of linear and nonlinear programming problems are presented. Current mode circuits are used in the design of networks. Transimpedance techniques based current conveyors are used for implementing the neurons. Two types of neurons are used in these networks: one being an integrator and the other being used as constraint. Many methods are suggested in this paper to implement networks weights by using current mode circuits such as Operational Transconductance Amplifiers. Network parameters are explicitly computed based upon problem specifications, to cause the network to converge to an equilibrium that represents a solution. Simulation results using Electronic workbench EDA version S.Oa-1996 shows the optimization solutions are obtained almost in real time, acceptable if they are compared with theoretical values, and are stable.