Enhancing RRM optimization using a priori knowledge for automated troubleshooting

The paper presents a methodology that combines statistical learning with constraint optimization by locally optimizing Radio Resource Management (RRM) or system parameters of poorly performing cells in an iterative manner. The statistical learning technique used is Logistic Regression (LR) which is applied on the data in the form of RRM-KPI (Key Performance Indicator) pairs. LR extracts closed form (functional) relations, known as the model, between KPIs and RRM parameters. This model is then processed by an optimization engine which proposes a new RRM parameter value. The RRM parameter value is reinserted in the network/simulator to generate corresponding KPI vector constituting generated RRM-KPI pair. First, only the a priori RRM-KPI pairs which are based upon the a priori model information are used for the model extraction. Then, as the optimization iterations progress, the generated pairs are given more importance in model extraction and the model is iteratively refined. The use of the a priori knowledge has the advantage of avoiding wrong initial models due to noisy data, allows much faster convergence and makes it more suitable for the off-line implementation. The proposed method is applied to troubleshoot an Inter-Cell Interference Coordination (ICIC) process in a LTE network which is based on soft-frequency reuse scheme.