Realization of desired super-stable spike-trains based on digital spiking neurons

This paper presents simple methods to realize desired super-stable periodic spike-trains. The spike-trains are generated by a digital spiking neuron consisting of two shift registers connected by a wiring. A demand for the spike-trains is described by a cost function of the wiring. In order to optimize the cost function, we present a simple evolutionary algorithm. The algorithm uses individuals corresponding to the wiring patterns and the number of individuals can vary flexibly. As an example, we use a cost function that evaluates the autocorrelation of the spike-trains. In order to super-stabilize the spike-trains, we present a deterministic wiring method. The super-stabilized spike-trains are robust (all the initial points fall instantaneously to it) and are applicable to an encoder in communication systems. Performing several numerical experiments, effects of the proposed methods are investigated.