LAUSR

Self-organizing networks (SONs) have been deployed typically with the SON functions (SFs) as rule-based controllers that evaluate metrics and decide actions based on a set of rules. Atop SFs, SON coordination ensures that the SFs do not conflict with one another during operations. However, as SONs evolve toward cognitive network management, with SFs as cognitive, learning-based agents, called cognitive functions (CFs), coordination gets extremely complicated. The coordinator must track the multiple dynamic agents each having non-deterministic behavior owing to the learning. Yet, if smart enough, the agents could also learn to minimize the conflicts or at least the related effects negating need for a coordinator. We present a distributed coordination approach called synchronized cooperative learning, where CFs learn to minimize their effects on one another. We apply the concept to a network with CFs for mobility robustness optimization and mobility load balancing, with results showing that using this approach, the functions learn to find a good compromise between maximizing individual metrics and minimizing cross effects on their peers.