A QoE anomaly detection and diagnosis framework for cellular network operators

Abstract

Traditional anomaly detection and diagnosis framework of cellular network is on purpose to optimize KPIs (Key Performance Indicators). However, cellular network operators are now attaching great importance to the QoE (Quality of Experience) of OTT (Over the Top) services on their networks rather than current percent-based KPIs since KPI anomaly cannot represent QoE anomaly all the time. Currently, network operators cannot measure anomalous QoS (Quality of Service) metrics which have direct mapping relationships with QoE anomaly by Network-side instrumentation, let alone QoE anomaly. To address this limitation, this paper presents a QoE anomaly detection and diagnosis framework along with a case study to evaluate its feasibility. Our study, including QoE anomaly detection and cross-layer root cause analysis, are based on a month-long WeChat video call service dataset captured by our OTTCAP (Over the Top services capturing and analyzing Platform) under live DC-HSPA+ (Dual-Cell High Speed Packet Access Plus) network. Results of our work can be directly used by network operators to do QoE prediction and network optimization at Network-side.