Stochastic resonance in a system of two coupled threshold elements (neurons
) forming a small neural network is investigated numerically. Periodic sign
als at inputs of the elements are phase-shifted with respect to each other
up to a half of the period, but their frequencies and amplitudes are identi
cal. The signal-to-noise ratio at outputs of the elements has a maximum as
a function of the input noise intensity for any phase shift. For proper cou
pling, dependent on the phase shift, this ratio is enhanced over that of a
single uncoupled element. The enhancement is usually observed for positive
(excitory) coupling if the phase shift is less than one fourth of the perio
d, and for negative (inhibitory) coupling otherwise, but minor deviations f
rom these rules are possible for high periodic signal frequency. Adiabatic
theory of stochastic resonance in coupled threshold elements is also formul
ated which describes qualitatively the dependence of the signal-to-noise ra
tio on the coupling for various phase shifts.