Automatic diagnosis of mobile communication networks under imprecise parameters

In the last years, self-organization of cellular networks is becoming a crucial aspect of network management due to the increasing complexity of the networks. Automatic fault identification, i.e. diagnosis, is the most difficult task in self-healing. In this paper, a model based on discrete bayesian networks (BNs) is proposed for diagnosis of radio access networks of cellular systems. Normally, inaccuracies are unavoidable in the parameters of the model (interval limits for discretized symptoms and probabilities in the BN). In order to enhance the performance of BNs, a methodology to model the “continuity” in the human reasoning is presented, named smooth bayesian networks (SBNs). SBNs are intended to decrease the sensitivity of diagnosis accuracy to imprecision in the definition of the model parameters. An empirical research campaign has been carried out in a live GSM/GPRS network in order to assess the performance of the proposed techniques. Results have shown that SBNs outperform traditional BNs when there is inaccuracy in the model parameters.