Transient analysis of the two bit neural A/D converter

The authors show the evolution in time (time trajectories) towards the equilibrium points of a neural network on the example of a two bit A/D `neural´ converter. On the basis of these trajectories it is possible to: (1) examine the dependence of the quantity and location of the equilibrium points of the network on the shape of the transfer function of the cells, (2) determine the stability properties of a given equilibrium point, (3) estimate its domain of attraction, and (4) given the initial conditions, estimate corresponding trajectory bounds. The authors present a landscape of the two-dimensional Lyapunov energy function E given by D.W. Tank and J.J. Hopfield (1986) and adapted for the authors´ example. They also show the dependence of the trajectories on the initial conditions in the network, the shape of the transfer function of the cell and its delay