AI-Assisted Trace Documentation Tool in Telecom

Sammanfattning

Within large-scale embedded telecom systems such as Ericsson's Radio Access Network (RAN), the absence of a structured mapping between trace events and source code locations forces engineers into manual, error-prone codebase navigation. This thesis investigates how AI-based coding assistants can be integrated into the development, documentation, and visualization of static inline trace wrappers for the 5G User Plane Control (UPC) component at Ericsson, and what impact this has on automation, developer effort, and documentation quality. A pipeline combining script-based wrapper generation with AI-assisted documentation using Amazon Q Developer was evaluated against Ericsson's existing development workflow. AI assistance reduced average developer time per file by approximately 71%, and trace interpretation time decreased considerably, demonstrating that the pipeline effectively reduces repetitive manual work. Nonetheless, AI-generated outputs required human validation to correct domain-specific and contextual inaccuracies, indicating that AI coding assistants complement rather than replace developer judgment in large-scale embedded systems.