A neural network approach to distributed adaptive routing of LEO intersatellite link traffic

A neural network (NN) approach to distributed traffic adaptive routing in intersatellite link (ISL) trunk networks of emerging low Earth orbit (LEO) satellite systems is proposed. This approach is basically motivated by the idea that due to the high complexity of the problem it may show a superior performance compared to deterministic algorithm approaches. A good performance is shown for an implementation using classical multilayer perceptrons (MLP) with backpropagation learning (BPA) in a distributed fashion-one MLP being placed in each satellite. A deterministic adaptive routing scheme is used as reference in the learning phase of the MLP. With the learnt “knowledge”, the distributed MLPs can perform a source routing in such a manner that the global traffic flows are efficiently shaped to reduce worst case link loads and to increase average link utilization, at the expense of an only slight increase of average transfer delay