Dynamic binary neural networks and storage of control signals for switching circuits

This paper studies basic dynamics and learning capability of the simple dynamic binary neural network. The network has the signum activation function and can exhibit various binary periodic orbits. In order to visualize the dynamics, we introduce the Gray-code-based return map. In order to store a desired binary periodic orbit, we present a simple learning algorithm based on the correlation learning. We then try to store a teacher signal corresponding to a typical control signal of a switching power converter. Performing numerical experiments, we have confirmed the storage of the teacher signal and its automatic stabilization.