RAN Intelligent Controllers (RIC)

• RAN Intelligent Controllers (RIC) are software-defined, programmable control entities that enable closed-loop, data-driven management of disaggregated 5G radio access networks per O-RAN standards.
• They operate in a two-tier architecture with near-real-time and non-real-time controllers, leveraging standardized E2 and A1 interfaces for rapid configuration updates and resource allocation.
• RICs facilitate advanced SLA enforcement and optimization using real-time KPI monitoring and algorithmic techniques like linear programming and knapsack solutions, as validated on platforms like Colosseum.

A RAN Intelligent Controller (RIC) is a software-defined, programmable control entity introduced by the O-RAN Alliance to enable closed-loop, data-driven management of disaggregated 3GPP 5G (and beyond) Radio Access Networks. RICs orchestrate RAN behavior by ingesting rich Key Performance Measurements (KPMs), running advanced control and optimization logic, and issuing real-time configuration updates to the protocol stack. In O-RAN, RICs are the pivotal enabler of dynamic service-level adaptation, AI/ML-driven resource management, and vendor-agnostic network programmability, operating over open interfaces standardized as E2 (for low-latency control/telemetry) and A1 (for policy/model management).

1. Architectural Placement and Protocol Stack

RICs are realized in a two-tier architecture per O-RAN specifications:

• Near-Real-Time RIC (near-RT RIC): Implements control loops operating at 10 ms–1 s granularity, hosted either at the mobile edge or central location. Integrates with Distributed Units (DUs), Central Units (CUs), and the radio stack via the E2 interface. Hosts xApps (microservices) for fine-grained resource control—traffic steering, scheduling, handover, slicing enforcement, etc.
• Non-Real-Time RIC (non-RT RIC): Operates at >1 s timescales, often co-located with the Service Management and Orchestration (SMO) layer. Manages policy-based orchestration, long-term analytics, AI/ML model lifecycle, and policy dissemination through the A1 interface. Hosts rApps for strategic planning and offline learning.

The canonical protocol stack comprises:

• E2AP (E2 Application Protocol) above SCTP, managing set-up, subscription, reporting, and control transactions.
• E2 Service Models (E2SMs): Plug-in schemas for streaming telemetry (e.g., E2SM-KPM for KPI reports) and control (e.g., E2SM-RC for semi-persistent scheduling, E2SM-SLICE for slice control).
• E2SM payloads are generally encoded as protocol buffers within ASN.1-wrapped E2AP messages.

The gNB (OAI-based in (Moro et al., 2023)) is partitioned into CU-CP, CU-UP, and DU/MAC/PHY, running an E2 agent with split functionality between E2AP termination (e2sim) and E2SM-specific logic implemented as dedicated threads tightly integrated with the MAC scheduler.

2. Closed-Loop Control and Key Performance Metrics

The controller operates a closed feedback loop as follows:

• At periodic intervals (e.g., per TTI or every 10/100 ms), the E2SM-KPM agent on the gNB transmits per-UE reports detailing the Transport Block Size (TBS) and allocated Physical Resource Blocks (PRBs).
• The xApp within the near-RT RIC consumes these reports, computes spectral efficiency $\nu_u = \mathrm{TBS}_u/\mathrm{PRB}_u$, and applies an optimization to determine new PRB assignments or scheduling parameters.
• Control actions are dispatched as E2SM-RC messages via CONTROL_REQUEST primitives, specifying per-UE SPS patterns (periodicity, PRB count) or signals to disable SPS in favor of dynamic scheduling.
• The gNB applies configuration updates immediately, enabling rapid response to changes in user demand or SLA violations.

Key KPMs are:

• $\mathrm{TBS}_u$: per-TTI MAC-level transport block throughput for UE $u$.
• $\mathrm{PRB}_u$: PRBs allocated to $u$ in each scheduling interval.
• Derived $\nu_u$ for resource allocation efficiency.

3. Data-Driven SLA Enforcement and Optimization Formulations

RIC enables dynamic enforcement of Service Level Agreements (SLAs) via both soft and strict policies:

• Soft SLA Policy: Formulated as a constrained linear program that minimizes the aggregate SLA violation over all GBR UEs:

$\min_{\{P_u,v_u\}} \sum_{u\in U} v_u$

subject to

$$\sum_{u\in U} P_u \le C, \quad v_u \ge \mathrm{SLA}_u - P_u\nu_u, \quad P_u,v_u \ge 0$$

where $P_u$ is the allocated PRBs, $C$ is the total available PRBs, and $v_u$ quantifies per-UE violation.

• Strict SLA Policy: Cast as a 0-1 knapsack problem to maximize the weighted cardinality of UEs whose bitrates are fully satisfied:

$$\max_{\{x_u\}} \sum_{u\in U} W_u x_u, \quad x_u \in\{0,1\}$$

subject to

$$\sum_{u\in U}\left\lceil \tfrac{\mathrm{SLA}_u}{\nu_u}\right\rceil x_u \le C$$

$x_u$ is binary and $W_u$ encodes priority.

The xApp runs these optimizations every control interval (e.g., 100 ms), applying LP/knapsack solvers to select new PRB assignments, thereby balancing resource usage across GBR and elastic flows.

4. Integration into OAI 5G Stack and E2 Interface Extensions

Custom integrations to OpenAirInterface include:

• SPS Hooks: Added to DU/MAC for both DL and UL path, enabling per-UE forced PRB assignment over standard scheduling epochs.
• RC Agent: Implements a minimal protobuf-based E2SM-RC, exposing a programmable interface for per-UE PRB control.
• E2SM Agent/E2AP Termination: Runs as an ITTI thread, with direct access to MAC metrics for per-UE state, periodically emitting KPM reports to the e2sim process, which then serializes and dispatches E2SM messages to the RIC.

This architecture ensures rapid application of control decisions and direct mapping between protocol actions and on-air scheduling.

5. Experimental Evaluation: Colosseum Testbed and Performance Metrics

The proposed RIC/xApp platform was validated on the Colosseum emulator:

• Testbed: One OAI gNB (SA), single 5G Core, three UEs (OAI UE), one near-RT RIC.
• Scenario: Line-of-sight, static pathloss, n78, 40 MHz, 65 PRBs; full-buffer TCP traffic with target GBRs (15/10/5 Mbps) for the UEs.
• Control interval: 100 ms.

Evaluation metrics:

• Per-UE throughput, instantaneous and total SLA violation, and control loop latency.

Key outcomes:

• Soft policy achieves minimum total SLA violation (∼20% lower than strict, ∼60% lower than baseline PF scheduler).
• Strict policy maximizes the number of fully satisfied UEs at the expense of total SLA satisfaction.
• Baseline PF scheduler cannot meet multiple GBRs simultaneously, resulting in large aggregate violations.
• Control loop achieves convergence within a single control slot (100 ms), demonstrating suitability for real-time SLA enforcement.

6. Significance and Implications

The architectural and algorithmic principles demonstrated in this framework substantiate the claims central to O-RAN: that RIC-mediated, closed-loop control is feasible on standard-compliant 5G stacks at sub-second timescales, and allows robust, data-driven adaptation to heterogeneous service demands. The separation of protocol stack, service models (E2SMs), and xApp logic allows modular upgrades and easy integration of algorithmic advances.

This work further establishes realistic baselines for control latency and optimization problem sizes and illuminates the practical trade-offs between strict satisfaction and utility-maximization in multi-UE, multi-SLA scenarios. The explicit demonstration of a programmable E2 interface, fine-grained PRB control, and semi-persistent scheduling highlights the flexibility O-RAN brings for network operators intent on supporting differentiated service classes and emerging URLLC/eMBB/mMTC use cases (Moro et al., 2023).