Recognition of oscillatory signals using a neural network oscillator

The neural oscillator, composed of two mutually inhibiting types of neuron with adaptive property in one neuron, is fed by periodic inputs of varying amplitudes, frequencies and phases. Using a performance measure defined in the frequency domain, we show that the neural oscillator is able to accurately recognize the spatiotemporal content of the oscillatory input without any information loss. The simulation results demonstrate that such a neural oscillator may exhibit marked changes in its spatiotemporal pattern in response to changes in the oscillatory input. Under a strong entrainment condition, the network can differentiate small changes in the frequency of in-phase inputs by displaying profound changes in both the waveforms and relative phases of the neuronal activities. Similarly, changes in the phase relationship of the oscillatory inputs are manifested as significant changes in the amplitude of neuronal activities. The results suggest a plausible mechanism for the classification of oscillatory signals in biological neurons without the need for any quantitative measurements, and the feasibility of such a simple neural network architecture as a building block for oscillatory information processing