Neural self-organization for the packet scheduling in wireless networks

This paper aims at introducing possibilities of self-organization for packet-switched networks at the scheduling management level. By discussing already existing scheduling strategies, we identify several contrasting needs that should be jointly addressed. The employment of the same algorithm for the whole network leads to low performance. On the other hand, adjusting the management cell-by-cell is not feasible and requires coordination. Hence, we propose a general framework for the scheduler, which can be easily tuned by means of a neural network. In this way, cells are grouped and self-similarities are identified, so that the differentiation in the management is lighter and a significant performance improvement can be achieved. Moreover, a general tunability of the system is introduced, allowing to cut the right trade-off between system complexity and QoS in a simple way.