Understanding software-defined-radios and -networks in 5G architectures

Article By : Kaue Morcelles and Brendon McHugh

Here's a look at the orchestration of 5G mobile service-based architectures implemented via open radio access network (O-RAN) technology.

The world of wireless communication is about to change, as 5G is finally reaching the end-consumer. One of the biggest 5G promises is massive device communication to power revolutionary IoT systems, such as autonomous vehicles, metaverse hardware, gaming virtual reality (VR), and smart factories. Some of the 5G technologies required for this revolution are Machine-to-Machine (M2M) communication, massive Machine Type Communication (mMTC), Ultra-Reliable Low-Latency Communications (URLLC), and enhanced Mobile Broadband (eMBB). In this context, the optimization of base stations is critical to provide low latency connections, optimal sharing of spectrum and processing resources, and dense small cell deployment.

Furthermore, 5G will provide convergent network communication across multi-technologies networks, and open communication systems to cooperate with satellites, cellular networks, clouds, data centers, and home gateways. Additionally, 5G systems will be autonomous and sufficiently able to adapt themselves depending on the required QoS to handle application-driven networks dynamically. In this context, we will discuss here the orchestration of 5G mobile service-based architectures (SBA) implemented via open radio access network (O-RAN) technology. This article also approaches the use of software-defined radios (SDR) and software-defined networks (SDN) in 5G, which enable network function virtualization (NFV), network slicing, cloud/edge computing, artificial intelligence (AI), and machine learning (ML).

5G Network Architecture

The first component of a 5G structure is the transport network, that connects the 5G RAN to the core network. It can be divided into three structures: the fronthaul, the midhaul, and the backhaul (see Figure 1). The distributed units (DU) are connected to the remote radio units (RRU) through the fronthaul network, where each DU can cover distances from a few kilometers to over 50 km, controlling multiple antennas. The midhaul performs an intermediary connection by linking the distributed units (DU) to the central unit (CU). Finally, the backhaul link connects the central units and remote/mobile systems to the core network. In addition to the transport network, the core 5G network contains several components for access and control. In an SBA architecture, the components are arranged in a set of interconnected Network Functions (NF), including the NF Repository Function (NRF), Network Slice Selection Function (NSSF), Policy Control Function (PCF), User Plane Function (UPF), Session Management Function (SMF), Access and Mobility Management Function (AMF), and Data Network (DN). At the user equipment (UE) side, access is controlled and performed through gNB nodes, that talk to AMF and UPF services via NG interfaces. The NG interface carries both user plane and control plane protocols: the user plane implements the PDU (Protocol Data Unit) session, and the control plane controls both the session and the connection to the network, including service request and transmission resources.

click for full size image

Figure 1: 5G network architecture consists of three structures. (Source: Per Vices)

To better understand the advantages of 5G, let us compare it with the giant 4G/LTE technology. Firstly, the core of 5G technology is fundamentally different, with the use of mmWave, massive MIMO connections, cloud-native software design, and high levels of system virtualization. Secondly, the 3GPP 5G is a service-based architecture, meaning that the system elements are defined as network functions (NF), that offer services to other NFs with authorized access. The service-based nature is more attractive than the 4G/LTE implementation, as it provides network slicing, function virtualization, cloud-based systems, and better compatibility with open-RAN technology. Moreover, the implementation of UPFs, to decouple gateway control and the user plane, and AMFs, to separate session management from connection and mobility management, are not found in 4G protocols. In 5G, the user and control planes are decoupled, as the UE traffic is 1000 times higher than in 4G. Finally, 5G systems allow the use of smaller and more specialized network cells, such as fempto cells and pico cells.

One of the most important aspects of 5G is the decoupling and virtualization of the RAN elements, which allows more intelligent, dynamic, and flexible networks different applications. At the front of the RAN development movement is the Open RAN (O-RAN) architecture. By opening the interface between RAN components, the O-RAN allows operators to combine different vendors in the same system, improving flexibility and giving the operator the freedom to work with the technology provider of choice. In the O-RAN, the base station is divided into two: the centralized unit (CU) and the distributed unit (DU) (Figure 2). The CUs are responsible for larger timescale functions, whereas the DUs operates in time-critical tasks. At the end of the chain, the remote radio units (RRU) manage all RF communication and components, such as modulation, coding, and interference avoidance. In terms of the protocol stack, the CU handles the higher layers, the DU manages lower layers, and the RRU deals with the physical layers. The open interfaces between CU and DU are called higher-layer splits (HLS), whereas the connections between DU and RRU are composed of lower-layer splits (LLS) interfaces. All O-RAN applications run on the RAN Intelligent Controller (RIC). The RIC platform provides abstraction over the RAN components, integrating optimization and automation algorithms.

click for full size image

Figure 2: Open RAN (O-RAN) architecture is shown. (Source: Per Vices)

Software-Defined Radio (SDR)

Software-defined radios, or SDRs, are radio systems composed of an analog radio front-end (RFE), an FPGA-based digital unit, and a mixed-signal interface, typically via ADCs and DACs. The RFE is responsible for receiving and transmitting the analog portion of the RF signal, which is discretized by the DAC/ADC interface. The RFE is an important part of the circuit, as it defines the signal range, number of channels, and bandwidth. The highest performance RFE in the market achieves 3 GHz of instantaneous bandwidth, using up to 16 independent channels. At the core of the SDR there is an FPGA configured with DSP capabilities: modulation/demodulation, up/down-converting, and data packetization. FPGAs are completely reconfigurable matrices of digital logic, so the same system can support several processing algorithms, state-of-the-art protocols, and even artificial intelligence without changing the hardware. SDRs provide low latency, flexibility, high interoperability (important for the 5G physical layer), and massive MIMO capabilities – useful for beamforming and spatial multiplexing. One commercial example is the Cyan SDR from Per Vices (Figure 3), which can be used as the core of 5G base stations and testbeds/emulators.

click for full size image

Figure 3: Per Vices Cyan can be used in 5G base stations. (Source: Per Vices)

In the 5G context, both RRUs and Baseband Units (BBUs) can contain one or more SDR units to perform radio-related functions, providing compatibility, interoperability, and flexibility. In gNodeB 5G BBUs, for instance, the connection to the RRUs is realized using eCPRI optic fibers. In these cases, the SDR must contain both eCPRI and Gigabit Ethernet (GBE) ports, as well as the ability to handle MIMO antennas. The RRU SDRs, on the other hand, are required to comply with the frequency range of the application, which can fall into either the FR1 or FR2 categories. The FR1 (Frequency Range 1) covers sub-6GHz frequencies (600 to 6000 MHz), whereas the FR2 (Frequency Range 2) covers frequency bands from 24.25 to 52.6 GHz. The FR2 band is applied in shorter range/higher bandwidth applications when compared to FR1. The RRU SDR must be selected and configured to work inside the desired spectrum. Small cells also profit from SDR implementations, as lightweight, low power, and compact complete RF solutions are readily available in the market.

The importance of SDR implementation arises from its role in O-RAN systems. Three of the most important O-RAN landmarks are disaggregation, virtualization, and softwarization, with the last being provided by SDRs. Softwarization is fundamental to achieve URLLC, eMBB, and mMTC functionalities. Moreover, SDR-based systems are flexible, upgradable, and interoperable, giving the operator control over the RAN without needing to constantly replace the hardware. SDRs can also comply with the instructions generated by the RICs, which is crucial for RAN optimization and automation.

Software-Defined Networking (SDN)

Software-defined networking (SDN) is the physical separation between the control plane functions and the forwarding function. A typical SDN architecture is divided into three parts: the application layer, the control layer – where the SDN controller operates – and the physical infrastructure. Layers communicate with each other through APIs (northbound APIs for the application-control communication and southbound for control infrastructure). SDN improves programmability and enables higher levels of network automation and optimization. It also provides cloud-like capability within the structure, allowing centralized computation and network control abstraction from the physical layer, data analytics algorithms, and system virtualization via virtual overlay networks. System virtualization enables one of the most important features in 5G: network slicing.

Network slicing refers to the division of the physical network into multiple virtual networks, that are unique and optimized for a particular service or application. Each virtual network, or slice, can be configured with only the specific resources required to perform a certain task, such as autonomous vehicles, IoT devices, and mobile services. The most obvious advantage of this technique is the optimization and tuning of the resource distribution to serve the needs of particular customers and market segments. Client-side services can be classified into eMBB, mMTC, and urLLC, with each category having its own throughput, bandwidth, latency, and robustness requirements (Figure 4). Network slicing is achieved using a combination of SDNs, SDRs, Network Functions Virtualization, data analytics, and automation. End-to-end automation, in particular, is fundamental to enable real-time network adaptability, design, and control over the massive number of slices within a single RAN.

click for full size image

Figure 4: This is an image of a 5G network slicing. (Source: Per Vices)

To design network slicing methods, network function virtualization (NFV) is crucial. This method enables the virtualization of the RAN and core network functions that were once performed by hardware, such as routing, scaling, security, and load balance. By realizing network functions in software, the operators can continuously update the network functions with state-of-the-art algorithms without needing to replace the hardware, saving time, reducing installation costs and customer disturbance. Moreover, NFV allows real-time re-purposing and redistribution of functions through the network slices, as well as inter-slice and intra-slice control of the RAN resources.

SDR and SDN/NFV for Optimizing Network Resources

The immense amount of data throughput necessary in 5G systems can easily overwhelm the most advanced LTE networks. For instance, a typical CPRI-based LTE fronthaul typically deals with channel bandwidths of around 10-20 MHz, which translates into around 10 Gbps in a 10 channels connection. 5G, on the other hand, deals with bandwidths in the range of 100 MHz to 500 MHz, and with massive MIMO scaling the fronthaul throughput can go up to the Tbps range. CPRI fibers are no longer enough, and optimized technologies, such as enhanced CPRI (eCPRI) are needed. In an eCPRI interface fronthaul, the physical layer functions are split between the RRU and the DU in an optimized proportion, thereby increasing the complexity of RUU while reducing the load on the fronthaul. The requirement for performance optimization is not restricted to the fronthaul, as both location, access, and management of resource instantiation depend greatly on the requirements of the service slice. In this context, SDR and SDN/NFV-based structures (Figure 5) can help.

There are several different types of orchestration and control for 5G optimization. The software-defined RAN (SD-RAN) community, for example, is developing O-RAN compatible open-source RIC controllers. The SD-RAN project focuses on the development of Near-Real-Time RICs (nRT-RIC) to optimize the dynamics and latency of network control, with the most prominent one being the open-source µONOS-RIC. Besides the open-source nature, the µONOS-RIC is compatible with AI/ML-based applications, that can be optimized for massive MIMO, Self-Organizing Networks (SONs), and smart Radio Resource Management (RRM). Another recently developed optimization technology is the Cross-Layer-Controller (CLC), which is applied in resource allocation and pairing between network slices, depending on real-time monitored RAN conditions. It acts via SDR and SDN systems to perform cross-layer management and adaptation from a set of target key performance indicators (KPIs).

click for full size image

Figure 5: SDN/NFV can be applied to 5G RANs to optimize performance. (Source: Per Vices)

In O-RAN-based architectures, the main objective of network optimization is to improve overall performance in various conditions, prevent network instability, and solve problems with minimal damage to the service. It does so by constantly measuring KPIs and crowdsourced information, and making decisions to control and adapt the cells accordingly. This prevents congestion, overloading, and interference, and reduces latency. In O-RAN, optimization is performed through nRT-RICs. External intelligence can run on top of nRT-RICs, making decisions based on AI/ML algorithms. AI/ML-driven nRT-RICs enable the use of advanced management algorithms, such as dynamic spectrum sharing (DSS) and NSSI resource allocation optimization.

In the O-RAN architecture, the Split Option 7-2x LLS complies with several optimization techniques, including beamforming optimization. Beamforming can be designed to increase both data throughput and the number of parallel connections, as well as improve the power efficiency and signal-to-noise ratio of the network, by focusing the RF beam to a specific location. Massive MIMO antennas play a major role in beamforming optimization. In these systems, the controller sets a global optimization objective, and each MIMO cell gives a partial contribution to the beam. SDR BBUs are fundamental for the dynamic and coherent coordination of MIMO antennas.

Current Research and 5G O-RAN Testbeds

O-RAN-oriented 5G architectures introduced several challenges in network design. Researchers are still trying to solve several technology bottlenecks, such as how to provide short overhead data access to AI agents, how to design robust data-driven control loops, and what are the exact roles and requirements of each RAN component. The SD-RAN community is one of the research teams trying to solve these issues. As mentioned before, SD-RAN developed an open-source nRT-RIC compatible with AI/ML applications, which provides the technology and abstraction necessary for data-driven control loops and intelligence allocation. The OpenRF Association, on the other hand, is aiming for the development of a highly interoperable 5G ecosystem, including both RF hardware and software, to reduce integration costs and time-to-market, while maintaining enough flexibility and customization. Both SD-RAN and OpenRF projects would not be feasible without the use of powerful SDRs and SDNs.

It is impossible to discuss 5G research without talking about the emulators, in particular the Colosseum testbed. The Colosseum is the largest network emulator testbed in the world, with 256 SDRs able to emulate up to 65536 RF channels (100 MHz). This massive system can work with GNU Radio, MATLAB, and most DSP technologies, and provides a great testing framework for AI/ML algorithms, MIMO systems, and O-RAN in general. The Colosseum can also simulate path loss, multi-path, and fading, providing RF conditions similar to the real-life environment. The Leonardo Bonati research team recently used the Colosseum to validate the feasibility of network control using deep reinforcement learning (DRL) agents running on top of nRT-RICs through xApps. The algorithm, which is O-RAN compatible, operates by selecting the best-suited scheduling policy for each RAN slice, considering both URLLC, MTC and eMBB. Compared to other approaches, the DRL system presented a 20% improvement in spectral efficiency and 37% reduction in buffer occupancy.


This article discusses the many aspects of 5G mobile SBA systems, including orchestration, implementation, management, and functionality, focusing on the Open-RAN architecture. The Open-RAN community is driving the development of novel 5G solutions through the use of open and disaggregated interface standards between RRUs and BBUs. In this context, SDRs and SDNs play a major role in the 5G revolution, by providing flexibility, interoperability, softwarization, and virtualization of the RNA – fundamental tools to enable unique 5G features, such as network slicing and DSS. SDRs are also highly present in the development of novel technologies and 5G research, being applied in softwarization, real-time monitoring and control, AI/ML applications, and large-scale RAN emulation.


This article was originally published on Embedded.


About the Authors

Kaue Morcelles is an electrical engineer, with emphasis on electronic design and instrumentation. Learning and writing about cutting-edge technologies is one of his passions

Brendon McHugh is a technical writer and Field Application Engineer. He possesses a degree in theoretical and mathematical physics from the University of Toronto.


Leave a comment