Distributed coupling complexity in a weakly coupled oscillatory network with associative properties

We present a novel architecture of an oscillatory neural network capable of performing pattern recognition tasks. Two established strategies for obtaining associative properties in oscillatory networks invoke either a physical, time constant or a global, dynamical all-to-all coupling. Our network distributes the complexity of the coupling between the spatial and the temporal domain. Instead of O(N²) physical connections or a global connection with O(N ²) frequency components, each of the N oscillators receives an individual coupling signal which is composed of N-1 frequency components. We demonstrate that such a network can be built with analog electronic oscillators and possesses reliable pattern recognition properties. Theoretical analysis shows that the scalability is in fact superior to the dynamic global coupling approach, while its physical complexity is greatly reduced compared to the individual time constant coupling.