Exjobbspresentation: Generative AI for Anomaly Diagnosis in 5G NR Scheduling Logs

5G NR base stations produce detailed scheduling logs that record per-slot decisions, channel measurements, and HARQ feedback at sub-millisecond granularity. Diagnosing anomalies in these logs currently requires hours of manual expert inspection per file. This thesis presents a system that automates this process end-to-end. Raw binary traces are parsed into a relational database, where statistical and machine learning methods detect anomalies at both the individual-record and temporal levels. A correlation-based module differentiates root causes with similar signatures, a rule engine maps findings to verified diagnostic patterns, and a Large Language Model (LLM) agent synthesizes the evidence into causal explanations. A feedback mechanism allows engineers to confirm or correct diagnoses, building a case store for future reference. On 11 logs from a controlled lab environment, the system classified all anomaly patterns correctly, diagnosed two engineer-confirmed failures, and surfaced subtle events that conventional alarms missed. Diagnosis time dropped from roughly one hour to 30 seconds, allowing engineers to quickly validate results and focus on resolution rather than investigation.

Handledare: Johan Thunberg (EIT), Ali Moradian (Ericsson) 

Examinator: Michael Lentmaier (EIT)