E2 Service Models in O-RAN: A Specification-Grounded Tutorial and Implementation Analysis

Abstract

The E2 interface and its associated E2 Service Models (E2SMs) are the primary mechanisms through which the Near-Real-Time RAN Intelligent Controller (Near-RT RIC) observes and controls Radio Access Network (RAN) behavior in O-RAN. Despite their central role in enabling xApp-driven closed-loop optimization, the structure of the E2SM framework and its realization across open-source platforms remain poorly documented. This paper provides a consolidated, specification-grounded tutorial and assessment on the O-RAN E2SM ecosystem. We characterize the architectural template shared by all five standardized E2SMs (E2SM-KPM, E2SM-RC, E2SM-CCC, E2SM-LLC, and E2SM-NI) through direct analysis of O-RAN Working Group (WG)3 specification body text, establishing a verified presence matrix of supported service types and a precise account of the RC-specific Style<inline-formula> <tex-math notation="LaTeX">$\to $ </tex-math></inline-formula>Action ID<inline-formula> <tex-math notation="LaTeX">$\to $ </tex-math></inline-formula>RAN Parameter hierarchy unique to E2SM-RC. We present a source-code-derived implementation analysis of open-source RAN stacks (OCUDU and OpenAirInterface (OAI)) and Near-RT RIC platforms (FlexRIC, OSC RIC, and micro-ONOS RAN Intelligent Controller (<inline-formula> <tex-math notation="LaTeX">$\mu $ </tex-math></inline-formula> ONOS-RIC)), distinguishing capabilities that are declared in service model advertisements from those functionally executed against the RAN stack, and explaining the architectural reasons behind persistent gaps. The analysis covers 29 KPM metrics across Distributed Unit (DU) and Central Unit (CU) function blocks, the style-level control surface of E2SM-RC and E2SM-CCC, and the emerging implementations of other service models. By mapping what specifications define to what implementations deliver, this paper provides xApp developers, platform integrators, and researchers with a precise operational picture of where O-RAN intelligence is realizable today and what changes are needed to support AI-native closed-loop control at scale.