OpenAirInterface (OAI)

OAI delivers a complete, 3GPP-compliant implementation of both the Radio Access Network (RAN) and Core Network (CN). This includes the physical layer (PHY), MAC, RLC, PDCP, and RRC layers for the gNB/eNB, as well as the AMF, SMF, and UPF for the 5G Core. The stack runs entirely in software on commercial off-the-shelf (COTS) x86 servers, eliminating the need for proprietary hardware accelerators for research and development workflows.

• gNB and eNB: Full base station implementations for 5G NR and 4G LTE
• 5G Core Network: Modular microservice-based architecture with HTTP/2 signaling
• UE (User Equipment): A complete terminal-side stack for end-to-end testing

The OAI PHY layer performs real-time baseband processing on general-purpose CPUs, a significant engineering achievement. It implements critical signal processing chains including OFDM modulation/demodulation, turbo and LDPC channel coding, and MIMO detection. The soft-modem architecture leverages SIMD instruction sets (AVX2, AVX-512) and real-time Linux kernels with preemptive scheduling to meet the stringent sub-millisecond timing constraints of 4G and 5G waveforms.

• 5G NR numerologies: Support for flexible subcarrier spacing (15, 30, 60 kHz)
• Massive MIMO: Up to 64 antenna ports for beamforming research
• Real-time kernel: PREEMPT_RT Linux patches for deterministic scheduling

OAI is designed for hardware-in-the-loop (HIL) and system-level simulation workflows. It integrates directly with the ns-3 discrete-event network simulator, allowing researchers to replace simulated protocol layers with OAI's real software stack. This enables high-fidelity digital twin testing where real baseband processing interacts with simulated channel models, traffic generators, and mobility patterns. The combination of OAI and ns-3 is a cornerstone of reproducible wireless research.

• ns-3 LENA module: LTE/EPC Network Simulator co-developed with OAI
• Channel emulation: Feed IQ samples through simulated propagation models
• Scenario replay: Inject recorded RF data into the OAI receiver chain

OAI aligns with the O-RAN Alliance architecture, supporting the functional split between the Centralized Unit (CU), Distributed Unit (DU), and Radio Unit (RU). The OAI 5G gNB can be deployed in a disaggregated configuration using the F1 interface (between CU and DU) and the 7.2 split fronthaul interface toward an RU. This makes OAI a primary platform for developing and testing O-RAN Intelligent Controller (RIC) applications, including xApps and rApps for AI-driven RAN optimization.

• F1 functional split: 3GPP-compliant CU-DU separation
• E2 interface: Enables near-real-time RIC integration for control loops
• O-DU/O-CU testing: Reference implementation for O-RAN plugfest interoperability

OAI serves as a foundational testbed for AI-enhanced RAN research. Its open-source nature allows direct access to internal metrics—channel state information (CSI), MAC scheduler decisions, and buffer status reports—that are typically opaque in commercial equipment. Researchers use OAI to train and deploy deep reinforcement learning agents for resource allocation, neural network-based receivers, and predictive beamforming algorithms. The platform supports both offline training with collected datasets and online inference within the real-time processing pipeline.

• Dataset generation: Export labeled PHY and MAC data for supervised learning
• Custom scheduler API: Replace default algorithms with ML inference engines
• Federated learning: Distributed model training across multiple OAI instances

OAI uniquely supports over-the-air (OTA) transmission using standard software-defined radio (SDR) frontends such as the USRP X310/N310 and BladeRF. This enables full end-to-end network deployment where an OAI gNB communicates with commercial 5G handsets or the OAI UE soft-modem. Researchers can conduct real-world validation of algorithms, measure performance under live channel conditions, and demonstrate prototypes in field trials without any proprietary network equipment.

• SDR support: USRP B210, X310, N310; BladeRF 2.0 micro
• Commercial UE interoperability: Tested with COTS 5G phones and modules
• RF frontend abstraction: Modular driver architecture for new hardware integration