The OpenAirInterface Software Alliance is showcasing its latest developments with NVIDIA technology in the O-RAN virtual exhibition 2023. 

In this demo, the OAI team integrated the NVIDIA Aerial software development kit’s layer 1 inline hardware accelerator with the OpenAirInterface layers 2 and above to build an accelerated 5G virtual RAN. The kit provides a full L1 high-PHY implementation of 5G NR compatible and interoperable with the O-RAN 7.2x fronthaul split. This architecture allows the connection to a commercial Open Radio Unit (O-RU) and uses the FAPI interface that can talk to a third-party L2+ stack. The NVIDIA Aerial SDK runs on commercial-off-the-shelf (COTS) components such as general-purpose servers with NVIDIA GPUs and NICs. All of this runs in containers with complete access to the source code for algorithm optimization and development.

The demo setup is depicted in the figure below. The gNB (O-CU and O-DU) runs on a COTS Dell server with an NVIDIA A100 Tensor Core GPU and an NVIDIA ConnectX-6 Dx SmartNIC. The server is connected to a fronthaul network using an O-RAN 7.2x split, PTP grandmaster, and an O-RU, which further on sends and receives signals to/from a commercial 5G phone (a Quectel UE in this demo).

There are two containers running on the edge server: one for the NVIDIA Aerial layer 1 and one for OpenAirInterface Layer 2 and 3, more precisely the MAC scheduler, the RLC as well as PDCP and RRC forming a complete gNB. The core network runs on a separate server and also in a virtualized environment. All the different network functions such as AMF, SMF, and UPF run in different containers. There is also a container running an edge application, in our case just an iPerf server.

For the moment, the demo is supporting frequency range one, 30 kHz subcarrier spacing with a bandwidth of 100 MHz. TDD config is 2.5-millisecond periodicity, 3-millisecond downlink, and 1-millisecond uplink. We support 2 layers of downlink, 1 uplink, and 1 cell. In the future, we expect to increase these numbers to 4 layers of downlink, 2 layers of uplink, and up to 3 cells.

About the OpenAirInterface Software Alliance

The OpenAirInterface Software Alliance (OSA) is a nonprofit organization founded in 2014 by EURECOM, a research institute based in the South of France. The Alliance manages and promotes the OpenAirInterface (OAI) open-source software that offers 4G and 5G and Core Network stacks as well as orchestration and management and control software. OAI implements 3GPP and the O-RAN specifications.

The OAI software development is organized into three project groups: Radio Access Network (RAN), Core Network (CN), and MOSAIC5G (M5G). Another project called CI/CD allows OAI to control the quality of all software produced within the Alliance. Each project group is composed of an engineering team following and achieving the objectives defined in its roadmap. The OSA stands out thanks to its large international community of contributors and users. The OAI software is used by many organizations around the globe for research and testing purposes as well as for building blocks of systems for various 4G/5G use cases, a growing number of them industrial.

For end-to-end deployments and control, OAI enables 5G deployment including the 5G gNB, Core Network, and RAN intelligent controller (RIC) capability thanks to O-RAN specified E2 and RIC software, itself a part of the OAI stack.

NVIDIA Aerial 

NVIDIA recently joined OpenAirInterface Software Alliance’s strategic board. OAI is collaborating with NVIDIA to advance wireless research by leveraging inline GPU-accelerated 5G Virtual and Open RAN solutions with complete access to the source for L1, L2, and L3  and a platform that interoperates in O-RAN 7.2x configuration with commercial radios and handsets.    

Going from simulation to deployed networks in real-time, over-the-air systems typically requires the use of hardware description languages. Innovation cycles are often constrained by the time-consuming process of FPGA programming. OAI’s integration of OAI Layer 2+ with the NVIDIA Aerial PHY Layer (L1 high-PHY) signals the development of a completely software-defined, C/C++ programmable 5G base station, enabling rapid prototyping and improved feature velocity without the need to program FPGAs.   

Channel coding, including LDPC, channel estimation, channel equalization, and modulation are all programmed in C/C++ programming languages and can be modified and re-compiled quickly for experimentation, as well as for changing the real-time behavior of a deployed base station. There are no fixed-function hardware accelerators used in our software-defined platform.  

Traditionally, purpose-built custom hardware has been used for wireless communications research, algorithm development, and network deployment. The Aerial SDK offers complete inline PHY Layer acceleration leveraging NVIDIA GPUs and NVIDIA  SmartNICs as a convenient, cost-effective, and performant platform alternative. This simplifies the development and testing of new 5G technology and applications.