Random networks of symmetrically coupled, excitable elements can self-organize into coherently oscillating states if the networks contain loops (indeed loops are abundant in random networks) and if the initial conditions are sufficiently random. In the oscillating state, signals propagate in a single direction and one or a few network loops are selected as driving loops in which the excitation circulates periodically. We analyze the mechanism, describe the oscillating states, identify the pacemaker loops and explain key features of their distribution. This mechanism may play a role in epileptic seizures.PACS numbers: 89.75. Hc, 05.65.+b, 87.19.lj, 87.19.xm The coherent oscillation (CO) of a collection of units that are non-oscillatory on their own is relevant to biological and physical sciences: CO has been identified and analyzed in populations of excitable biological cells ( [7]. In contrast with well-studied synchronization phenomena of self-oscillating units [8], in these cases the ability to oscillate derives from the interactions of the elements. CO can occur also on complex networks if the nodes are excitable [9][10] or even monostable [9], provided that the network contains loops, and that the directional symmetry of couplings is somehow broken to allow a signal to propagate in a single direction around a loop [10].Networks of these types include some neural [11][12] and genetic regulatory networks.[9] Some studies of excitable networks have been inspired by target and spiral waves in continuous media. [13][7] [10] In complex networks, loops are both generic and abundant. [15][14] While short loops of length L ≪ N (where N is the network size) are rare in large random networks, ones with L ln N occur generically in numbers growing exponentially with N . Their number also grows roughly exponentially with L up to a maximum at a length L m ∼ N.[15] [14] In this letter we show that random excitable networks readily self-organize into a CO state following a transient phase during which one or a few loops are dynamically selected as driving (or pacemaker) loops. We describe the mechanism of signal propagation and gain an understanding of the resulting distributions of the oscillating states and associated driving loops.We consider networks of diffusively coupled, excitable elements with dynamics described by the Bär model [16] N is the number of nodes, u i and v i are dynamical variables, D is the coupling strength, and A ij is the adjacency matrix. a, b, and ε are parameters, for which we adopt the values a = 0.84, b = 0.07, and ε = 0.04. In the absence of coupling, the individual nodes display excitable dynamics, with a stable equilibrium at (u, v) = (0, 0) and an excitation threshold u th ≈ 0.1. The dynamical equations were integrated numerically using a fourth-order Runge-Kutta algorithm with time step ∆t = 0.1. For the topology, we chose the undirected random regular network of degree 3 (RRN3), where nodes are randomly and symmetrically connected with the constraint that all have the same degr...