Deep Learning-Based Multi-Fault Diagnosis for Self-Organizing Networks

Abstract

Having self-organizing ability is regarded as one of the vital features for modern wireless communication networks. Such self-organizing networks (SONs) thus draw significant attention in past years. As fault diagnosis is one of the essential functionalities for SONs, in this paper, we investigate the multi-fault and fault severity level diagnosis. Specifically, we propose deep learning-based approaches that can determine the faults and their corresponding levels by utilizing the network key performance indicators (KPIs). Furthermore, to enhance recall, we propose a loss function design that can effectively trade false alarm rate against recall. We conduct simulations adopting a practical setup to evaluate the performance. Results show that our proposed approaches can accurately diagnose multiple faults and determine their severity levels.