A Simple Discrete-Time Recurrent Neural Network and its Application

This paper studies a simple recurrent-type neural network characterized by sparse binary connection parameters and signum activation function. Depending on the parameters, the network exhibits a variety of rotation-type periodic orbits of binary vectors. The network is well suited for precise analysis and FPGA based hardware implementation. Performing precise analysis in 8-dimensional cases, we have clarified the number and stability of the periodic orbits. Some of the orbits are applicable to control signal of octopod walking (spider) robots.