Enabling Green Cellular Networks: A Review and Proposal Leveraging Software-Defined Networking, Network Function Virtualization, and Cloud-Radio Access Network

Abstract

The increasing demand for enhanced communication systems, driven by applications such as real-time video streaming, online gaming, critical operations, and Internet-of-Things (IoT) services, has necessitated the optimization of cellular networks to meet evolving requirements while addressing power consumption challenges. In this context, various initiatives undertaken by industry, academia, and researchers to reduce the power consumption of cellular network systems are comprehensively reviewed. Particular attention is given to emerging technologies, including Software-Defined Networking (SDN), Network Function Virtualization (NFV), and Cloud-Radio Access Network (C-RAN), which are identified as key enablers for reshaping cellular infrastructure. Their collective potential to enhance energy efficiency while addressing convergence challenges is analyzed, and solutions for sustainable network evolution are proposed. A conceptual architecture based on SDN, NFV, and C-RAN is presented as an illustrative example of integrating these technologies to achieve significant power savings. The proposed framework outlines an approach to developing energy-efficient cellular networks, capable of reducing power consumption by approximately 40 to 50% through the optimal placement of virtual network functions.

1. Introduction

The advancement of Information and Communication Technology (ICT) has revolutionized and improved communication, leading to an exponential increase in data traffic and data usage. According to Ericsson’s report, mobile data usage per month per device is projected to increase significantly across different regions [1]. Considering the Compound Annual Growth Rate (CAGR), the mobile data traffic per smartphone is expected to grow annually at an average rate of 17% in North America, 16% in Western Europe, and 15% in Central and Eastern Europe from 2022 to 2029 [1]. Figure 1 shows the mobile data traffic per smartphone in GB for the same period.

The growth in data consumption can be attributed to several factors. The increasing number of smart electronic devices such as smartphones, wearable electronics, and sensors significantly enhance data creation and use. The worldwide adoption of IoT devices, such as smart home appliances and industrial sensors, generates considerable data. Nowadays, cloud services are an integral part of crucial applications and provide benefits such as users being able to store data, run different applications, and access services in the cloud environment. These dependencies on cloud infrastructure enhance data consumption. The roll-out of the 5G cellular networks promises enhanced throughput and low latency that will support data-intensive activities like High-Definition (HD) video streaming, Virtual Reality (VR), and Augmented Reality (AR). The COVID-19 pandemic accelerated remote work and online learning, which boosted data consumption. In summary, the convergence of technological enhancements, the growth of smart devices, and the changing behavior of users enhance the exponential growth of data consumption [2,3,4].

This tremendous growth in data traffic and data usage has compelled the development of more robust and efficient network infrastructures to support the increasing demand for data consumption. However, these advancements in telecommunication networks have also led to a significant increase in energy consumption, Carbon Dioxide (CO₂) emission, and climate change, posing a serious environmental threat. Telecom service provider Ericsson is proactively addressing climate change, a pressing global issue, by leveraging technology to reduce carbon emissions across their service portfolio, including telecommunication network infrastructure [5]. The ICT industry significantly contributes to global CO₂ emissions, accounting for approximately 2% of total emissions [6]. This includes the energy consumption of data centers as well as the entire lifecycle of devices, networks, and infrastructure. The energy demand associated with ICT operations is substantial. Radio Access Networks (RANs), which form the backbone of cellular communication, play a crucial role and consume over 60% of the energy used by the entire ICT industry [6]. Figure 2 depicts the power consumption breakdown of various sections within a cellular network. The RAN of the cellular network consumes 73% of the total power and the rest of the components (i.e., core, data centers, etc.) consume 27% of the total power [7]. Optimizing RANs for efficiency is essential to reduce their environmental impact [6]. Therefore, the need for sustainable and green telecommunication networks is more pressing than ever.

Wireless communication is growing rapidly and is widely used to fulfill ongoing demands; however, most network deployment designs have not focused on being energy efficient to reduce CO₂ emissions. That is where green communication comes in. Green communication, in the context of cellular technology, focuses on establishing environmentally sustainable and energy-efficient practices within cellular networks [8]. It aims to reduce energy consumption by optimizing cellular network components such as base stations, antennas, and data centers. Techniques include low power modes during idle periods and enhancing hardware efficiency. Cellular networks are also adopting renewable energy sources such as solar and wind to power base stations. Along with these, attention is given to efficient resource allocation, load balancing, and intelligent network management based on traffic demand. In simple terms, the goal of green communication is to find a balance in delivering data while using the least amount of energy possible [9,10,11,12,13]. To mitigate these challenges, emerging paradigms, such as SDN, NFV, and C-RAN, offer evolutionary potential. These technologies enable dynamic resource management, hardware abstraction, and centralized processing, which are key enablers for energy-efficient networks.

This paper aims to explore the potential of SDN, NFV, and C-RAN in creating a sustainable and green cellular network. These technologies hold the promise of transforming the traditional telecom network infrastructure into a more flexible, scalable, and energy-efficient one, thereby paving the way toward a green network. In addition, this paper serves as a comprehensive review of existing approaches and initiatives aimed at achieving energy efficiency in cellular networks. Figure 3 shows the block diagram of cellular architecture based on SDN, NFV, and C-RAN. In cellular communication, NFV can be utilized in mobile core networks for deploying telecommunication functions as virtual functions [14]. C-RAN can be used in the Baseband Unit (BBU) pool and Remote Radio Unit (RRU) network to lower the total cost of ownership and improve network performance. SDN can enhance the data plane, control plane, and application plane for handling data traffic, controlling network topology and defining requirements, respectively [15].

SDN decouples the control plane from the data plane and provides a centralized control that enables efficient network management and reduces the network’s energy consumption [16]. On the other hand, NFV virtualizes network functions and runs them as software in data centers or clouds, leading to reduced network power consumption and deployment cost [17]. C-RAN, a cloud-based architecture for the radio access network, offers centralized processing, collaborative radio, and real-time cloud infrastructure, resulting in enhanced spectral efficiency and energy efficiency. Detailed information about SDN, NFV, and C-RAN is presented in Section 3, Section 4, and Section 5, respectively, in this paper.

In this paper, the functionality and complexity of these technologies are delved into, and how they can be leveraged to create energy-efficient communication networks is discussed. The following key points are targeted:

• A comprehensive review of existing initiatives and research efforts focused on reducing energy consumption in cellular networks.
• The collective contribution of SDN, NFV, and C-RAN to reducing energy consumption in cellular networks.
• Existing challenges and limitations in the convergence of these technologies for energy-efficient communication networks.
• A cellular architecture is proposed based on SDN, NFV, and C-RAN to make the cellular network power efficient.

This paper follows the following structure: Section 2 presents related work to make the cellular network energy efficient. Section 3 provides information about SDN and how it will help cellular networks. Section 4 elaborates on the NFV. Section 5 provides the information about C-RAN. Section 6 presents the SDN, NFV, and C-RAN convergence challenges and some possible solutions. Section 7 includes the proposed cellular architecture based on SDN, NFV, and C-RAN. Along with these, it provides information about supporting organizations. Section 8 concludes the paper with some future work.

2. Related Work and Motivation

This section provides an overview of prior research work related to energy-efficient cellular networks. Table 1 provides a comparative analysis of key research contributions in the field of energy-efficient cellular networks. It categorizes various studies based on their focus areas, including SDN, NFV, C-RAN, and other green communication strategies. This classification helps highlight existing gaps and supports the argument for integrating SDN, NFV, and C-RAN to enhance network sustainability.

Chia et al. [18] focused on making off-grid cellular base stations more sustainable by utilizing renewable energy sources, such as solar power, and energy storage. The size of solar panels, converters, and batteries was examined based on the macro Long-Term Evolution (LTE) base station’s daily energy consumption requirements. In this study, an optimal solar design, energy output, and cost were considered using the HOMER tool [19]. The results demonstrated that this approach guaranteed 100% energy independence and long-term stability for LTE base stations. Tahsin et al. [20] introduced an optimal energy-sharing framework as Multi-Objective Linear Programming (MOLP). This framework primarily targeted two goals: the energy collected at the base station and the load on the base station in future time slots. Deep Q-learning was used to predict the energy collected at the base station, and the results were simulated in MATLAB version R2017b. The obtained results showed that the considered multi-operator energy cooperation surpassed the current methods in terms of deployment and management cost, performance, and energy efficiency. Tahsin et al. [21] presented an overview of sustainable and energy-efficient cellular base stations. The cellular base station system models and the possibilities of using renewable energy solutions were investigated. Along with these, locations where renewable energy-based base stations could be deployed were proposed. A. Jahid et al. [22] explored a hybrid energy-sharing system for LTE renewable energy-powered base stations using physical power lines. In this architecture, each LTE base station was equipped with its own renewable energy source and storage, and access to electricity/energy was shared via power cables to reduce the load on traditional energy sources. The proposed architecture was simulated for the LTE cellular system, and the output showed a significant reduction in the average energy consumption. Similarly, A. Jahid et al. [23] explored the practicality of using solar power and wind turbines attached to an energy storage system in an LTE base station located in a remote area of Bangladesh. M. S. Hossain et al. [24] explored the combination of solar Photovoltaic (PV) and biomass resources to power off-grid LTE cellular base stations in Bangladesh. A hybrid energy-sharing system was used, and it was concluded that this approach was cost-effective and energy efficient.

J. An et al. [25] introduced an Ultra-Dense Heterogeneous Network (UDHN), and the considered architecture was simulated. A random access method was proposed to improve network efficiency, reduce signaling overhead, and enhance energy efficiency. The simulated results confirmed the effectiveness of the considered network architecture. The authors also highlighted the main challenges faced by the UDHN. Z. Hasan et al. [26] presented a quick overview of methods such as heterogeneous networks and cell-zooming to make cellular communication more energy efficient. Research challenges were discussed, and techniques for greener cellular communication were suggested. Malathy et al. [27] considered energy-efficient resource allocation, network planning, the use of renewable energy resources, and C-RAN to make 5G and beyond 5G cellular networks more energy efficient. Isfaq B. S. et al. [28] presented a comprehensive review of approaches that could be considered to make 5G communication networks more energy efficient. Techniques such as energy harvesting, resource allocation, massive Multiple Input Multiple Output (mMIMO), Device-to-Device (D2D) networks, spectrum sharing, and ultra-dense networks for 5G green cellular communication were explored. A multi-tier network architecture for 5G was also proposed to make it more energy efficient. Fatima Salahdine et al. [29] presented a detailed overview of saving the energy of base stations by using sleep modes. A mechanism was provided to put the base station into sleep mode for ultra-dense cellular networks and wake them up when needed. Challenges and solutions related to energy-efficient cellular networks were also discussed. S. Jamil et al. [30] provided a brief overview of green cellular communication, considering techniques like D2D communication, mMIMO systems, Heterogeneous Networks (HetNets), and Green-IoT (G-IoT). Challenges related to cost, bandwidth, and spectral efficiency were also addressed. Similarly, S. Buzzi et al. [31] presented a detailed overview of making cellular communication green. Strategies such as mMIMO, D2D communication, Millimeter Wave (mmWave), HetNet network deployment, and C-RAN were considered. Hardware solutions to develop an energy-efficient smart grid were also explored. B. Mao et al. [32] explored the importance of green communication and presented a research overview on Artificial Intelligence (AI)-based green cellular communication. The use of AI-based systems to manage the cellular network and enhance energy efficiency was emphasized. The authors also investigated the use of Machine Learning (ML), especially deep learning, to enhance 6G cellular technology.

Sofana Reka et al. [33] presented the deployment of a smart grid for 5G cellular systems to reduce power consumption. Similarly, F. O. Ehiagwinal et al. [34] explored techniques such as smart grids, HetNet network architecture, and sleep modes of base stations to develop energy-efficient cellular communication. Additionally, it was mentioned that environmental activists, regulators, and the government could encourage eco-friendly base stations in Nigeria. Q. Wu et al. [35] and M. Feng et al. [36] presented a survey on developing green cellular communication by utilizing the sleep mode technique for base stations. Energy harvesting was also considered to power up the base stations. Y. Alsaba et al. [37] provided information about ongoing research on beamforming for energy harvesting in cellular communication networks. The performance of the beamforming system in energy-harvesting cellular communication was also investigated. Alsharif et al. [38] simulated the base station sleep mode mechanism to enhance energy efficiency for 5G and LTE communication systems. The Particle Swarm Optimization (PSO) algorithm was utilized to maintain coverage using LTE base stations when 5G base stations were switched off. The outcome of this simulation work showed that 3.52 kW of energy per day could be conserved while maintaining high data rates when 5G base stations were switched off. Similarly, U. K. Dutta et al. [39] focused on implementing Self-Adaptive-Scheduling (SAS) algorithms to switch 5G base stations on and off to enhance energy efficiency and reduce CO₂ emissions.

Nicola Piovesan et al. [40] carried out a survey on energy-efficient 5G networks and emphasized the use of energy-harvesting hardware. S. Guo et al. [41] presented the challenges of minimizing power consumption in green energy-powered C-RAN. This challenge was represented by Mixed Integer Linear Programming (MILP), and a two-phase heuristic polynomial-time algorithm was introduced to solve the problem. L. M. P. Larsen et al. [42] carried out a simulation to investigate energy consumption in cellular crosshaul networks. Possible C-RAN configurations were explored. It was concluded that carrying out more data processing before transmission was more energy efficient while enhancing the network. The study’s findings emphasized the importance of choosing the right functional split, as the scenarios differed in energy consumption by a factor of 40. Additionally, the authors explored a study in [43] to select the right RAN architecture shared with other operators to enhance energy efficiency. In this study, it was investigated how these approaches could be applied in real networks, along with current trends in research.

K. N. R. S. V. Prasad et al. [44] and Miao Yao et al. [45] provided insight into developing energy-efficient mMIMO technology for cellular systems. Some limitations of mMIMO technology were also mentioned. Similarly, Shunging Zhang et al. [46] considered mMIMO and HetNet architecture, along with orthogonal frequency division multiplexing and non-orthogonal aggregation, to enhance energy efficiency in cellular communication systems. A. Bohli et al. [47] presented a detailed overview of green cellular communication, considering mMIMO, mmWave, and HetNet mechanisms. To address the power consumption challenge in ultra-dense Small Cell Networks (SCNs) for 5G wireless communication, M. M. Mowla et al. [48] suggested the use of mMIMO and Passive Optical Networks (PONs). S. R. Danve et al. [49] presented different types of base stations and explored various techniques to enhance the energy efficiency of cellular base stations. An algorithm to save power was also provided. A. Jahid et al. [50] performed a simulation and proposed an approach called Dynamic Point Selection Coordinated Multipoint (DPS CoMP) to balance cellular network loads for enhancing throughput and energy efficiency. The simulation results showed that the considered approach reduced on-grid power consumption by 22% and enhanced energy efficiency by 32%. Alimi et al. [51] presented a detailed review of potential techniques to improve energy efficiency in 5G cellular communication while reducing Operating Expenses (OpEx) and CO₂ emissions. The authors primarily focused on developing mMIMO and C-RAN mechanisms to reduce power consumption in cellular networks.

Bojkovic et al. [6] presented a 5G green communication network based on energy harvesting, HetNet architecture, and SDN. Dawadi et al. [52] presented the current ICT development in Nepal. Simultaneously, the evaluation of cellular networks to make them green was explored, utilizing SDN and Internet Protocol Version 6 (IPv6). Alsharif et al. [53] examined mMIMO, mmWave, HetNet, D2D, and SDN mechanisms to develop green cellular communication. Usama et al. [54] reviewed recent research to develop green cellular communication. SDN, NFV, sleep modes of base stations, and ML techniques were considered to reduce power consumption in cellular networks. Zhang et al. [55] provided an overview of recent research on the energy efficiency of 5G cellular communication. Techniques such as green energy harvesting, smart grids, mMIMO, D2D, C-RAN, and NFV were considered. Additionally, the latest developments in the standard energy efficiency of the 3rd Generation Partnership Project (3GPP) within the context of key 5G green technologies were explored. Dlamini T. et al. [56] explored the advantages of SDN and NFV in cellular networks. Different network architectures based on Mobile Edge Computing (MEC) were proposed, utilizing NFV to deploy network functions. D. A. Temesgene et al. [57] provided a detailed review of softwarization in densely deployed RANs. The applicability of ML in future cellular networks, along with SDN and NFV, was explored to curtail energy consumption in cellular network architecture. E. J. Kitindi et al. [58] provided an overview of Wireless Network Virtualization (WNV), SDN, and C-RAN technologies to develop future cellular networks. A general WVN cellular network architecture using SDN was proposed. Challenges and research issues related to WNV- and SDN-based cellular networks were also discussed. Similarly, Chih-Lin I. et al. [59] explored China Mobile’s vision and potential solutions for future cellular communication. Information about C-RAN, SDN, NFV, and ultra-dense networks was provided.

Motivation

The motivation for our research paper stems from the observation that much of the existing research in the field has predominantly focused on two key areas: the utilization of renewable energy sources to power base stations and the development of smart grids through hardware solutions. These approaches focus on changing the energy resources but do not significantly address curtailing energy consumption in cellular communication. Some authors have also explored technologies such as mMIMO, mmWave, HetNet, and dynamic base station management strategies based on traffic patterns. While these areas have seen significant attention, we noted that the potential of technologies like C-RAN, SDN, and NFV remains underexplored in the context of cellular network sustainability.

Although a few papers have introduced the fundamentals of C-RAN, SDN, and NFV, there is a notable gap in fully harnessing their capabilities within the broader cellular network architecture. This research gap serves as a source of inspiration for our work. Our objective is to make these networks more sustainable and energy efficient, leveraging the full potential of these technologies to address the evolving needs of the telecommunications industry. The following sections explore these technologies, beginning with the role of SDN in enabling programmable and energy-efficient networks.

Table 1. State-of-the-Art.

Literature (Ref.)	Hardware Solutions/ Smart Grid	Renewable Energy Source/ Energy Harvesting	mMIMO	mmWave	HetNet	Cell Zooming	Beamforming	D2D Communication	AI/ML	Sleep Modes Basestation	C-RAN	SDN	NFV
Bojkovic et al. [6]		X			X							X
Chia et al. [18]	X
Tahsin et al. [20]									X
Alsharif et al. [21]	X	X
A. Jahid et al. [22]	X	X
A. Jahid et al. [23]		X
M. S. Hossain et al. [24]		X
J. An et al. [25]					X			X			X
Z. Hasan et al. [26]		X			X	X
Malathy et al. [27]		X					X	X			X
Ishfaq Bashir Sofi et al. [28]		X	X		X		X	X
Fatima Salahdine et al. [29]			X		X		X			X
S. Jamil et al. [30]			X		X			X
S. Buzzi et al. [31]	X	X	X	X	X			X			X
B. Mao et al. [32]									X
Sofana Reka et al. [33]	X
F. O. Ehiagwinal et al. [34]	X				X					X
Q. Wu et al. [35]		X	X					X		X
M. Feng et al. [36]		X	X		X					X
Y. Alsaba et al. [37]		X					X
Alsharif et al. [38]										X
U. K. Dutta et al. [39]										X
Nicola Piovesan et al. [40]	X	X
S. Guo et al. [41]											X
L. M. P. Larsen et al. [42]											X
L. M. P. Larsen et al. [43]			X							X	X
K. N. R. S. V. Prasad et al. [44]			X	X	X
Miao Yao et al. [45]			X
Shunging Zhang et al. [46]			X		X
A. Bohli et al. [47]			X	X	X
M. M. Mowla et al. [48]	X		X
S. R. Danve et al. [49]	X		X			X
A. Jahid et al. [50]	X	X
Alimi et al. [51]			X								X
Dawadi et al. [52]												X
Alsharif et al. [53]			X	X	X			X				X
Usama M et al. [54]									X	X		X	X
Zhang et al. [55]	X	X	X					X			X		X
Dlamini T. et al. [56]		X							X				X
D. A. Temesgene et al. [57]		X									X	X	X
E. J. Kitindi et al. [58]											X	X	X
Chih-Lin I. et al. [59]							X				X	X	X
This Paper	X	X	X	X	X	X	X	X	X	X	X	X	X

Table 1. State-of-the-Art.

Literature (Ref.)	Hardware Solutions/ Smart Grid	Renewable Energy Source/ Energy Harvesting	mMIMO	mmWave	HetNet	Cell Zooming	Beamforming	D2D Communication	AI/ML	Sleep Modes Basestation	C-RAN	SDN	NFV
Bojkovic et al. [6]		X			X							X
Chia et al. [18]	X
Tahsin et al. [20]									X
Alsharif et al. [21]	X	X
A. Jahid et al. [22]	X	X
A. Jahid et al. [23]		X
M. S. Hossain et al. [24]		X
J. An et al. [25]					X			X			X
Z. Hasan et al. [26]		X			X	X
Malathy et al. [27]		X					X	X			X
Ishfaq Bashir Sofi et al. [28]		X	X		X		X	X
Fatima Salahdine et al. [29]			X		X		X			X
S. Jamil et al. [30]			X		X			X
S. Buzzi et al. [31]	X	X	X	X	X			X			X
B. Mao et al. [32]									X
Sofana Reka et al. [33]	X
F. O. Ehiagwinal et al. [34]	X				X					X
Q. Wu et al. [35]		X	X					X		X
M. Feng et al. [36]		X	X		X					X
Y. Alsaba et al. [37]		X					X
Alsharif et al. [38]										X
U. K. Dutta et al. [39]										X
Nicola Piovesan et al. [40]	X	X
S. Guo et al. [41]											X
L. M. P. Larsen et al. [42]											X
L. M. P. Larsen et al. [43]			X							X	X
K. N. R. S. V. Prasad et al. [44]			X	X	X
Miao Yao et al. [45]			X
Shunging Zhang et al. [46]			X		X
A. Bohli et al. [47]			X	X	X
M. M. Mowla et al. [48]	X		X
S. R. Danve et al. [49]	X		X			X
A. Jahid et al. [50]	X	X
Alimi et al. [51]			X								X
Dawadi et al. [52]												X
Alsharif et al. [53]			X	X	X			X				X
Usama M et al. [54]									X	X		X	X
Zhang et al. [55]	X	X	X					X			X		X
Dlamini T. et al. [56]		X							X				X
D. A. Temesgene et al. [57]		X									X	X	X
E. J. Kitindi et al. [58]											X	X	X
Chih-Lin I. et al. [59]							X				X	X	X
This Paper	X	X	X	X	X	X	X	X	X	X	X	X	X

3. What Is SDN?

Software-Defined Networking (SDN) is an architectural approach to computer networking that enhances flexibility, scalability, and programmability by decoupling the network’s control plane from the data plane, as illustrated in Figure 4b,c. In traditional networking, both the control and the the data planes are tightly integrated within networking devices such as switches and routers (Figure 4), limiting the flexibility and programmability of the network. SDN manages networks through software applications [60,61]. It employs software controllers, guided by Application Programming Interfaces (APIs), to interact with the hardware infrastructure and direct the flow of network traffic [62]. This approach offers a more flexible and efficient method for managing and controlling network operations.

OpenFlow-based transformations (Figure 4b) are executed by the control plane (e.g., ONOS or Ryu), which operates under the OpenFlow protocol. These transformations are supported by OpenFlow-compatible vendors such as HP and Cisco. Additionally, P4 networking devices (Figure 4c) enable programmability in the data plane while maintaining a similar control plane architecture.

An SDN network is distinguished by its agility, effectively combining proactive and reactive capabilities to swiftly adapt to changing network conditions. Centrally managed through one or more controllers, SDN provides a unified view of the network, simplifying management tasks. By enabling programmatic configuration, SDN allows administrators to define network behavior through software-defined policies, thereby enhancing flexibility and automation. SDN embraces open standards and vendor neutrality that fosters interoperability and innovation, allowing organizations to seamlessly integrate diverse network components. These characteristics collectively make SDN networks highly adaptable, efficient, and suitable for addressing evolving networking challenges.

3.1. Working Principle

Communication between the SDN controller and the data planes is facilitated by the OpenFlow protocol. This protocol is a standard method for the SDN controller to interact with networking devices, such as routers and switches, whether they are OpenFlow-based or use P4Runtime. Through the OpenFlow or P4Runtime protocols, the SDN controller instructs these network devices on how to handle incoming data packets, make forwarding decisions, and modify packet headers if necessary. OpenFlow, a crucial component of SDN, provides centralized control, enabling SDN controllers to manage network devices and enforce policies efficiently. Its programmability allows administrators to dynamically define forwarding rules, ensuring that the network can be customized and adapted to meet evolving requirements. By decoupling the control and data planes, OpenFlow enhances the network’s flexibility and scalability. Operating on a flow-based forwarding model, OpenFlow classifies traffic into flows and directs devices on how to process packets. This contributes to the agility, efficiency, and adaptability of SDN networks [63,64].

3.1.1. OpenFlow

In 2008, the OpenFlow protocol was created by researchers from several universities, including Stanford, MIT, and Princeton. The goal was to allow researchers to try out new network protocols on their campus networks. Later, SDN and OpenFlow gained prevalent academic and industry adoption. Several commercial network switch vendors now integrate the OpenFlow API into their products. Major tech players like Microsoft, Facebook, Verizon, Google, Yahoo, and Deutsche Telekom have come together in funding the Open Networking Foundation (ONF) to promote SDN through open standards development [65]. In 2012, Google started using SDN to connect its data centers worldwide because it offered significant flexibility for managing traffic between data centers.

Figure 5 shows the specification of OpenFlow; from 2009, many specifications belonged to OpenFlow. From 2009, it was the very first one to be v1.0 using a single flow table and IPv4. If we look forward, v1.1 has a group table, v1.2 is the IPv6, v1.3 is a meter table (apart from other things), v1.4 is the synchronized table, and, finally, v1.5 is the egress table [63].

3.1.2. P4 Language

P4 was first introduced in a research paper presented at the 2014 Special Interest Group on Data Communication Computer Communication Review (SIGCOMM CCR) conference [66]. Subsequently, the inaugural P4 workshop was conducted in June 2015 at Stanford University. An updated version of P4, known as $P{4}_{16}$, was introduced between 2016 and 2017, supplanting the previous specification, $P{4}_{14}$. Figure 6 shows the different released versions of the P4 language.

Figure 6. P4 from 2018 to 2023 [67].

Figure 6. P4 from 2018 to 2023 [67].

P4Runtime and OpenFlow are used for network programming, but they serve different purposes and operate at different levels of abstraction. On the one hand, OpenFlow is a communications protocol that gives access to the forwarding plane of a network switch or router (or firewall) over the network. It enables network controllers to determine the path of network packets across a network of switches. While OpenFlow is a protocol that enables SDN by giving administrators software-based access to the flow control tasks provided by switches and routers, P4 is a high-level language for programming protocol-independent packet processors. P4 is a field-specific programming language developed for managing data packet forwarding planes in network devices in pretty much anything that P4 can build [68]. Unlike regular languages like C or Python, P4 is specifically made for handling network data efficiently. It does not support any specific protocols. On the contrary, it authorizes users to define the required protocols in the program [68]. P4 enables the flexibility to reconfigure the data plane. It has been gaining increasing attention due to its alignment with the next-generation SDN standards, characterized by Open Interfaces and complete data plane programmability. As a result, P4 has witnessed widespread adoption across academia and industry in recent years [69].

3.2. Architecture of SDN

The architecture of SDN can be divided into three layers, as shown in Figure 7:

• Application Layer: This top layer contains all the applications that need to communicate with the network. These applications can include network topology builders, logging and monitoring tools, network Access Control Lists (ACLs), and other network services [64,70]. The applications are agnostic to the Southbound protocols, whether they use OpenFlow or P4Runtime.
• Control Layer: This central layer houses the SDN controller, which manages the flow of data traffic in the network devices based on the requirements of the application layer. The SDN controller uses Northbound APIs to communicate with the application layer and Southbound APIs to interact with the infrastructure layer [64,70].
• Infrastructure Layer: This bottom layer consists of networking devices such as switches and routers. It is responsible for forwarding packets based on the decisions made by the control layer [64,70].

The architecture of SDN consists of five key elements that collaborate to form a flexible, programmable, and efficient network. These elements are described below [61,64]:

• Applications and Services: These programs and systems use the network to communicate and interact with the SDN controller via the Northbound Interface to request network services.
• Northbound Interface: This communication link between the SDN controller and the applications allows applications to request network services and receive network information.
• Network Operating System (NOS): The software running on the SDN controller provides the necessary functionality for managing the network.
• SDN Controller: The heart of the SDN architecture, the SDN controller acts as the network’s “brain”, making packet routing decisions. It interacts with network devices through the Southbound Interface and communicates with applications via the Northbound Interface. Examples of controllers include ONOS, OpenDaylight, Floodlight, IRIS, POX, and Ryu [71].
• Southbound Interface: This communication link between the SDN controller and network devices transmits instructions from the controller to the devices and provides the controller with network state information.
• Network Devices: The physical infrastructure, including switches, routers, and other networking hardware, receives instructions from the SDN controller and forwards or drops data packets accordingly.

These elements collectively contribute to a flexible, programmable, and centrally managed network infrastructure. Although there are similarities between SDN and the Open Systems Interconnection (OSI) model, it is crucial to recognize that they represent different paradigms, and their mapping is not exact.

Figure 7. Architecture of SDN [72].

Figure 7. Architecture of SDN [72].

3.3. How SDN Will Help Cellular Network 5G and Beyond

SDN is anticipated to be a key player in the evolution and operation of 5G networks and beyond. SDN, in conjunction with 5G, will revolutionize network capabilities, offering significant opportunities for network operation, which allows for dynamic allocation of network resources, thereby improving network efficiency and flexibility [73]. Integrating SDN with 5G and beyond networks can enhance AI and automation capabilities, leading to more efficient network management and operation. Furthermore, SDN can reduce associated costs by minimizing the need for physical infrastructure and allowing for the more efficient use of network resources [74]. SDN can also enhance the delivery of network-based services, IT systems, and applications, improving user experience and enabling new services. Despite the limitations related to performance and scalability of a centralized controller approach in SDN for extensive 5G and beyond networks, advancements in SDN technology are addressing these challenges, making it a promising solution for managing the complex and large-scale networks of 5G and beyond [75]. SDN can also be integrated with other technologies like NFV to enhance the capabilities of cellular networks further. In essence, SDN is seen as an enabling technology that can help realize the huge promises of a 5G network by providing a flexible, programmable, and centrally managed network infrastructure [76].

Padros-Garzon et al. [77] presented an SDN network in which the OpenFlow data plane was controlled. Moreover, functions were executed on OF commodity switches within a logically centralized data center. Additionally, an SDN controller (SDNC) was employed. A virtualized Serving Gateway (vS-GW) was deployed on top of the Northbound SDN as an application. However, the Serving Gateway User Plane (S-GW UP) was carried by the Regional Router (RR). The UP referred to the data path, which was inherently managed by the data plane. It functioned as the network component responsible for anchoring user sessions, managing mobility, and providing access to external networks. The architecture was found to be similar to the one proposed by Ameigeiras et al. [78]. Furthermore, the approach closely resembled the work of Guerzoni et al. [79], in which an SDN control plane was used to manage OpenFlow devices, primarily in the core network rather than in the access network. Yao et al. [45] proposed the management of OpenFlow switches by the SDN control plane; however, their approach was implemented within the core network.

B. Allen et al. [80] proposed a method to connect trains and data centers using SDN and cellular technology. A general controller was placed on the train (local), while an external controller (data center) was suggested for additional functionality via REST API, as shown in Figure 8. Silva et al. [81] also investigated seamless handover in vehicular networks. Vehicular Ad Hoc Networks (VANETs) frequently experience interruptions, and SDN integration was expected to mitigate this issue. Two different SDN architectures were proposed, and it was observed that the one with lower complexity provided better performance in terms of delay reduction, packet loss mitigation, and minimized network overhead. Additionally, for P4 (Figure 4c), a local controller was suggested for handling reactive forwarding, while the remaining functionalities (such as telemetry) were managed at the data center.

3.4. Challenges and Solutions

SDN offers the advantage of easy network programmability and the ability to create dynamic traffic flow policies. However, this advantage also introduces potential security vulnerabilities. The overall security of SDN was analyzed by Kreutz et al. [82]. It was concluded that the centralized nature of the controller and the programmability of the network introduce new threats related to network security, necessitating novel countermeasures. In SDN architecture, the centralized view of the network has been identified as a potential security drawback [83]. For instance, a Denial-of-Service (DoS) attack on a centralized controller managing a large network of multiple devices would be more destructive than a targeted attack on an individual router. To mitigate these challenges, several techniques have been proposed, including replication, diversity, and secure components [84].

Ensuring high-quality services in advanced SDN-enabled cellular networks remains a challenge. SDN is utilized to create network slicing, enabling support for different applications. Automated Quality-of-Service (QoS) provisioning per application or service is required. Automating QoS is particularly crucial when multiple wired and wireless technologies share the same network slice [85]. A Software-Defined Low Latency (SDLL) framework based on SDN was proposed by S. Messaoudi et al. [86] to provision QoS and guarantee ultra-low latency in 5G and beyond networks.

In SDN-enabled cellular networks, effective monitoring of performance metrics for both physical and virtual networks is essential. To address this challenge, Network Tomography (NT) has been explored as a solution. NT estimates network performance by analyzing data from a selected subset of network elements rather than requiring measurements from every part of the network. This targeted approach provides a cost-effective and efficient method for monitoring network health and diagnosing issues [87].

4. What Is NFV?

NFV is a concept brought by network service providers. It is utilized in the core and C-RAN of the cellular network [14]. In 2012, a significant evolution occurred when the world’s seven leading telecom operators, including AT&T, Orange, Deutsche Telekom, and China Mobile, collaborated to release a white paper based on network function virtualization. The main objective of this document is to highlight the key advantages and associated challenges bound to NFV development and deployment. This white paper serves as the foundation of the NFV concept, underlining its potential to remodel the telecom industry by virtualizing network functions and abstracting them from dedicated hardware [17].

The European Telecommunications Standards Institute (ETSI) has taken responsibility for the standardization of NFV and established a group known as the ETSI Industry Specification Group (ISG) for NFV. Since 2012, ETSI has continued to enhance the development of NFV by releasing significant milestones known as “Releases”. These Releases are issued approximately every two years. The ETSI Releases are essential for guiding the adoption of NFV within the telecom industry, promoting interoperability and tackling the challenges and complexities of virtualization [88].

As we know, every mobile network operator has a significant amount of specialized equipment and hardware dedicated to performing different network functions. These functions have dedicated hardware to support their functionality. These specialized middle boxes are hard to manage, and it is not a scalable solution to expand the network. When new mobile technologies like 5G and beyond came along, it used to mean a significant amount of work for operators. They had to change out hardware and redesign their entire network. Along with these, this dedicated hardware consumes considerable power [17]. Utilizing the NFV, these network functionalities can be performed in software that is installed in Virtual Machines (VMs), as shown in Figure 9.

4.1. Working Principle of NFV

The NFV can be defined as a concept in the telecommunication field that aims to virtualize and abstract network functions, such as load balancing, firewalls, and routers, from dedicated hardware applications and run them as software on standardized hardware or in a virtual machine. By utilizing virtualization technology, network operators can deliver more cost-effective, flexible, and scalable network services. The working principle of NFV is based on techniques such as virtualization, abstraction, orchestration, and SDN integration.

NFV uses virtualization technologies, such as hypervisors and containerization, to develop virtual instances of network functions. NFV also uses an approach defined as an abstraction to separate or decouple the network functions from the underlying dedicated hardware. It enables network functions to be implemented as software applications and makes them more flexible and adaptable. To manage the different Virtual Network Functions (VNFs), NFV requires a Management and Orchestration (MANO) system to install, configure, and manage the VNFs. NFV mainly works integrated with SDN, where SDN provides the network control plane and allows the dynamic configuration and management of network resources, while NFV prioritizes the virtualization and deployment of network functions.

In summary, the working principle of NFV involves the virtualization of network functions, abstracting them from dedicated hardware and orchestrating their deployment on standardized hardware platforms.

Table 2. Differences between NFV and SDN [89].

Category	NFV	SDN
Concept	Abstracts network functions from conventional devices and encapsulates them as software.	Separates the forwarding plane from the control plane to enable automated and programmable network control.
Goal	Service providers propose replacing distributed network devices with consolidated ones.	To achieve network hardware devices’ programmability and centralized management and control.
Key Aspects	1. Established procedure. 2. Hardware-based forwarding functions are detached from dedicated hardware.	1. An open and programmable control plane 2. Hardware-based traffic forwarding, and decision-making within the control plane.
Conflict or Not		The fusion of NFV and SDN introduces a novel network model. NFV enables adaptable service orchestration, while SDN realizes unified management and configuration of network functions.

represents key differences between the NFV and SDN and

Table 3. Differences between NFV and proprietary network equipment/devices [89].

Category	NFV	Proprietary Network Equipment/Devices
Hardware Used	Generic x86-based servers, versatile storage devices, and adaptable switching equipment are utilized.	Dedicated devices are used.
Hardware-Software Separation	Software is separated from hardware and provided as module components.	Hardware and software are closely integrated, with software functions relying on dedicated hardware.
Receptiveness	Universal hardware foundation and standardized interfaces enable an open ecosystem through collaboration among multiple parties.	Relying on dedicated services results in a closed system, making it challenging to onboard third-party partners.
Network Resilience	General-purpose hardware and resource virtualization technologies enable dynamic adjustments of both software and hardware resources to meet specific service demands.	Dedicated devices do not align with virtualization technologies, hindering resource-sharing and flexible scaling capabilities.
Upgradation	Device upgrades occur swiftly, primarily involving software enhancements.	Deployment of network devices is time-consuming, necessitating both software and hardware provisioning.
Operation and Management	Virtualizes hardware resources and automates operations and management intelligently.	Upgrading and replacing devices is a complex process, as maintenance involves manual or semi-manual preparations and configurations through the CLI or web-based systems.
Service Organizations	NFV networks are deployed according to service requirements and can be dynamically orchestrated with flexibility.	Traditional networks operate with relative independence. Converting service requirements into network specifications is not swift, resulting in a sluggish network response.

shows the differences between NFV and traditional devices.

4.2. Architecture of NFV

The NFV architecture encompasses several essential components that are employed together to enable the virtualization and management of network functions. Figure 10 represents the NFV reference architecture with essential components that are shown below:

• Virtualized Network Functions (VNFs): VNFs are the virtual instances of network functions implemented as software-based functions that were traditionally implemented using dedicated hardware. Network functions such as routing, firewall, and network optimization can be implemented using software and deployed and managed on the data centers or in the cloud.
• NFV Infrastructure (NFVI): NFVI is responsible for providing the underlying physical and virtual infrastructure for hosting the VNFs. It includes the following elements:

  (a)

  Compute Resources: NFVI provides general-purpose servers or cloud-based compute instances where VNFs run.

  (b)

  Network and Storage Resources: NFVI uses virtualized network components like connectivity, switches, routers, VLANs, and storage (including cloud storage) to handle data and configurations. These resources form the backbone for managing information within NFVI.

  (c)

  Virtualization Layer: The virtualization layer is responsible for providing virtualization technologies, such as hypervisors and containerization, to run VNFs on available physical hardware.

• Element Management System (EMS): This manages the physical network equipment, including legacy hardware, which is used for the deployment of NFV. It ensures coordination between virtualized and traditional network elements.
• Operations Support System (OSS) and Business Support System (BSS): These systems provide end-to-end network service management, billing, and provisioning.
• MANO: Management and orchestration in network function virtualization, is divided into three key functional blocks:

  (a)

  NFV Orchestrator (NFVO): The NFVO is responsible for VNF lifecycle management, handling deployment and scaling coordination. It integrates with NFVI to provide the required resources and communicate with the VNF Manager (VNFM) for VNF-specific tasks.

  (b)

  VNF Manager (VNFM): VNFM manages the lifecycle of VNFs, which includes instantiation, configuration, and termination (on/off) of VNFs. VNF Manager communicates with NFVO and VNF itself to manage the functionality of VNFs.

  (c)

  Virtualized Infrastructure Manager (VIM): VIM manages the virtual resources required for VNFs. It identifies, allocates, and handles faults in physical and virtual resources.

These blocks work together to efficiently control and organize NFV deployments to ensure reliability and agility in network services.

4.3. How NFV Will Help Cellular Network

The virtualization technique provides the capability to move the network functions onto virtualized environments, leading to numerous benefits for power consumption reductions in cellular communication systems. Hardware abstraction helps the operator to choose more energy-efficient hardware to consolidate network functions into fewer and more powerful servers. By running multiple virtual network functions on a single physical server, NFV makes server consolidation easier. Comparatively, this consolidation improves the utilization of hardware resources and reduces the power consumption of the network [17]. Based on the workload, NFV dynamically allots resources such as memory and processing power. This mechanism reduced power consumption, especially during times of low network activity or changes in the network traffic. Along with these, NFV also enables the use of low-power processors and hardware optimized for specific network functions [91].

Figure 11 illustrates the power consumption in traditional network appliances used for deploying various network functions, as well as in a virtualized system. It has been shown that virtualization techniques can reduce power consumption by approximately 40 to 50%, depending on the dynamic placement of VNF chains [92]. An architecture for 5G networks that supports NFV was proposed by A. N. Al-Quzweeni et al., and it was found that virtualization can result in energy savings of up to 38% [93]. Additionally, R. Mijumbi et al. [94] utilized Bell Lab’s GWATT tool to quantify the energy savings achieved by implementing NFV technology in different parts of the cellular network. Their findings indicated that virtualizing the core and RAN can lead to reductions in power consumption by 22% and 17%, respectively. Furthermore, NFV plays a significant role in deploying edge computing resources closer to end-users at the network’s periphery. This approach reduces the need for long-distance data transmission, thereby decreasing the power consumption associated with data transport.

NFV can also be applied in RAN. With the introduction of C-RAN, computing resources are being deployed closer to user equipment, facilitating the execution of critical network functions and applications in a virtualized environment. Virtualized RAN leverages NFV principles by distributing virtualized baseband functions across multiple servers, thereby enhancing network efficiency and flexibility [95].

Figure 11. Power consumption in traditional and virtualized functions. Virtualized functions consume approximately 50% less power compared to traditional network functions [96].

Figure 11. Power consumption in traditional and virtualized functions. Virtualized functions consume approximately 50% less power compared to traditional network functions [96].

4.4. Challenges and Solutions

NFV has the potential to provide multiple advantages to the cellular communications system, but it also has some challenges related to security, performance degradation, interoperability issues, and vendor lock-in [97].

4.4.1. Core: Challenges and Solutions

Incorporating NFV into the core of cellular networks introduces new attack surfaces and security vulnerabilities, and these are because of the resource pool based on cloud computing and open network architecture. This can impact the functionality of NFVs [97]. These vulnerabilities can be handled by developing robust security measures such as encryption, authentication, access control, and Intrusion Detection Systems (IDSs) to safeguard NFV infrastructure and VNFs. Regular security audits and updates are also crucial to avoid possible risks [98].

The virtualization overhead and resource-sharing in NFV infrastructure may cause performance degradation, affecting QoS and Quality-of-Experience (QoE) [97]. By leveraging performance optimization techniques such as selecting the specialized hardware components to upgrade the performance of specific computational tasks, and developing network function offloading mechanisms, the virtualization overhead could be reduced. In addition, optimal performance for critical network functions could be maintained by fine-tuning resource allocation and scheduling algorithms [99].

NFV implies the integration of diverse hardware and software components from multiple vendors, leading to compatibility challenges and interoperability issues. Along with this, dependency on specific NFV hardware vendors or proprietary solutions may result in vendor lock-in, limiting flexibility and hampering innovation. The following measures could be taken: Set up standardized interfaces and protocols for deploying the NFV to enhance the interoperability between different hardware components from different vendors [97]. Vendor lock-in risks could be tackled by embracing open standards and interoperable solutions. Promote vendor neutrality and flexibility by collaborating with multiple vendors and leveraging open-source NFV initiatives such as OpenStack [100] and OPNFV [101]. Ensure NFV ecosystem compatibility by participating in industry forums and consortia.

Rapid network expansion or sudden traffic fluctuations may affect the scalability. The scalability could be managed by adapting resource allocation automatically based on workload demands and network conditions by implementing dynamic resource orchestration and scaling mechanisms [102]. Using cloud-native architectures and containerization technology, NFV deployments can be made more agile and scalable [103]. To ensure the successful deployment and operation of virtualized cellular networks, operators must address these challenges and implement appropriate solutions.

In an NFV network environment, it is crucial to have the same level of reliability and availability of network services as in a traditional network environment. Figure 12 shows the mechanism to support the reliability and availability of an NFV-based network. This is achieved through fault prevention, which includes strategies to avoid errors and operational failures; fault detection, which involves identifying and diagnosing failures; and resiliency, which ensures service recovery after a failure through redundancy, migration, and protection of VNFs [104].

4.4.2. RAN: Challenges and Solutions

In a RAN, some critical tasks require swift execution. These include Layer 1 processing tasks like bit interleaving, modulation, and encoding. Additionally, the scheduling function, which organizes the order of data packet transmissions and Hybrid Automatic Repeat Request (HARQ)-related baseband processing are also time-critical. There is a need to ensure that the total delay plus jitter of the fronthaul does not exceed 3 milliseconds for HARQ-baseband processing to avoid a loss in air interface throughput [105]. However, virtualizing functions like these could introduce delays, impacting the QoS and user experience [105]. Two acceleration methods, in-line and look-aside, address these challenges. In-line acceleration handles Layer 1 tasks before data reaches the main CPU, freeing it for higher-layer processing. It is efficient and reduces delays. Look-aside acceleration allows the CPU to handle Layer 1 processing while a separate card manages specific functions, providing flexibility to offload tasks [106]. Thus, the in-line accelerator will handle one process at a time, whereas the look-aside accelerator can handle multiple dataflows. On the other hand, the look-aside accelerator will also use more time for transmission between accelerator and CPU, whereas the the in-line option is more scalable due to the separation of accelerator and CPU [107]. Benefits and caveats of the two processing options are examined in [108].

Both in-line and look-aside accelerations have their advantages and disadvantages, and the decision depends on the RAN’s requirements and the characteristics of time-sensitive tasks. Some RAN vendors, including Nokia, believe in in-line acceleration, whereas other vendors, including Ericsson, believe in look-aside acceleration [109].

5. What Is C-RAN?

Cloud-RAN is a RAN architecture that centralizes and virtualizes the baseband processing. The RAN architecture widely adopted today is referred to as distributed RAN, where the Radio Unit (RU) is located in the antenna tower, handling radio functions such as digital-to-analog conversion and up-conversion; the remaining processing is handled in the BBU. The BBU is located in a shelter close to the antenna tower and handles both time-critical and non-time-critical baseband processing. RANs represent a huge amount of under-utilized resources since the users tend to move around between different areas during the day. The users will usually be at home in the morning, then they go to work and come back home again in the evening, leading to much un-utilized capacity, since the mobile network is critical infrastructure that cannot be switched off when not in use. Hence, other ways must be found to increase resource utilization and add savings to the energy bill.

A proposed solution is to move the baseband processing into a centralized data center, a BBU pool, where virtualized baseband processing can share the same hardware and assign additional capacity to areas where it is needed when it is needed. Hence, as described in

Table 4. Difference between Centralized, Virtualized, and Cloud-RAN.

Category	Centralized RAN	Virtualized RAN	Cloud RAN
Concept	Baseband processing is moved away from the cell sites to a central baseband pool.	Baseband processing is virtualized, running as software instances independently from the underlying hardware.	Virtualized baseband processing is centralized in a datacenter. Deployment options are agile.
Benefits	Reduced site footprint, improved cell cooperation, and shared cooling mechanisms.	Load balancing, agile service deployment, faster updates.	In addition to the benefits derived from centralized and virtualized RAN, Cloud-RAN also benefits from dynamic capacity assignment, improved scalability, and increased resource utilization.
Drawbacks	Large capacity and latency requirements for the transport network connecting radio functions to centralized baseband processing (fronthaul network).	Complexity of virtualized functions and challenges in running time-critical RAN functions on COTS hardware.	Cloud-RAN faces the same drawbacks as virtualized RAN, but its agile deployment options can reduce fronthaul complexity seen in centralized RAN.

, the C-RAN architecture incorporates both concepts of centralized and virtualized RAN. The concept was first proposed by IBM in 2010. Later, 3GPP evolved for more agile deployment opportunities [110]. Here, the BBU was divided into a Distributed Unit (DU), handling the more time-critical functions, and a Centralized Unit (CU), handling more relaxed inter-site communication. Figure 13 illustrates the evolution of Cloud-RAN technology, highlighting the transition from distributed RAN to centralized RAN, and this progression has led to Cloud-RAN, where baseband processing is centralized and virtualized. Figure 14 illustrates the future mobile network architecture, where radio processing is divided into three units: RU, DU, and CU [110].

5.1. Working Principle of C-RAN

Centralization is a concept that has been adopted in many sectors to achieve higher utilization of limited resources, healthcare, for example. But centralization always comes with a price; the transport. In the case of mobile networks, it is data that needs to be transported from the RU on the cell site to the DU in a centralized data center. The current distributed RAN installations use the Common Public Radio Interface (CPRI) protocol for transport between the RU and BBU to encapsulate the very raw signal sent from the RU. The signal transmitted in the CPRI protocol has only been through a few operations inside the RU, namely frequency down conversion, sampling, and analog-to-digital conversion [111]. Hence, due to the very raw nature of the signal transmitted between the RU and the DU, the load is very high and continuously occupies the transport medium, while the latency requirements are strict [111]. Thus, the distance between RU and DU is bound to the latency requirements as well as the availability of high-capacity infrastructure. Several solutions to overcome the huge transport capacity problem of C-RAN exist; for example, compression [112], high capacity infrastructure like fiber or mmWaves [113], or to add more functions to the RU [111]. The latter solution is referred to as functional splits, and multiple approaches exist. The low layer split, separating the RU and DU functions, has not yet been determined, and multiple approaches exist. On the other hand, the high layer functional split separating the DU and CU has already been settled, to include the Packet Data Convergence Protocol (PDCP) and Radio Resource Control (RRC/IP) in the CU [110].

When the data finally reach the data center, they will meet a virtualized format of the baseband processing. Baseband processing functions can be virtualized either as virtual machines or containers, both enabling multiple virtual software instances to share the same hardware. The hardware used can be Commercial-Off-The-Shelf (COTS) servers, either in a private enterprise data center or in a public cloud where capacity is rented on demand. In both cases, using COTS hardware for baseband processing will also make it possible for other applications to use the same hardware when the load in the mobile network is low.

5.2. Architecture of C-RAN

The C-RAN architecture divides the baseband processing into the DU and CU to meet more agile deployment opportunities. Hence, a number of functional placement scenarios exist, taking into account the time-criticalness of particularly the functions in the DU. The functional placement scenarios vary from having only the RU on the cell site to having all virtualized baseband processing on the cell site additionally for the benefit of extreme low-latency scenarios. The Next Generation Mobile Networks Alliance (NGMN) has defined six functional placement scenarios, where the baseband processing is handled on either the cell site, aggregation site, or distributed site [114]. These functional placement scenarios are illustrated in Figure 15.

Due to the various functional placement scenarios, deployment of C-RAN becomes very agile, and the specific deployment can adapt to solve a specific problem in a specific area. One thing that needs to be considered is the transport network in-between the fronthaul connecting the RU to the DU and the midhaul connecting the DU to the CU [111]. Hence, when deploying C-RAN, the use of the capacity-demanding fronthaul network must be determined by the availability or opportunity for deployment of a high-capacity link between the RU and the DU. Hence, if such a link is not existing or possible to deploy, the DU, and maybe also the CU, must be left on the cell site.

C-RAN is currently the architecture suggested in specifications by Open Radio Access Network (O-RAN) and ITU-T [115], IEEE’s Next Generation Fronthaul Interface [116] and Small Cell Forum [117]. Hence, it is a widely discussed architecture for future mobile network deployments.

5.3. How Can C-RAN Improve the Sustainability of Cellular Networks?

The initial concept of C-RAN separated the very power-consuming amplifier from the baseband functions to be installed in the antenna mast. This was done to minimize the loss provided by the cable connecting the RU to the baseband, greatly reducing the size of the amplifier and also to be able to benefit from air cooling in the RU. This was the first step towards centralized processing and led to a huge reduction in energy consumption.

C-RAN will have many smaller contributions to the energy consumption budget. These will be related to the centralization and virtualization of the network. Hence, C-RAN will benefit from both. Centralizing the baseband processing will enable improved cell cooperation, leading to improved spectral efficiency, and thus, there might be small energy savings to catch due to reduced re-transmissions and lowered interference. Moving all processing to one location will bypass the need for air conditioning locally on the cell sites, as it will only be necessary in the data centers. However, this is only for countries with very warm climates. Virtualizing the baseband processing will enable easier installation of new power-saving features as well as the opportunity to run network functions on any hardware, maximizing the utilization of hardware resources, and thus saving energy from potentially less hardware when the load is consolidated. In C-RAN, the baseband processing is centralized and virtualized, which will give the additional benefit of utilizing users’ movement patterns; hence, baseband processing from sites, which users move into at different times of the day, can share the processing hardware because they utilize the resources at different times of the day. Furthermore, the agile capacity assignment of C-RAN will make sure no capacity is wasted because it is assigned to the area where it is needed. However, it must be noted that, according to the study from NGMN [118], baseband processing is only responsible for 6% of the RAN energy consumption.

5.4. Challenges and Solutions

The huge drawback of C-RAN is the fronthaul network, connecting the RU to the DU. Originally, RUs used the CPRI protocol [119] for transport over the fronthaul network. However, the CPRI protocol induces a constant and very high load on the fronthaul network, since it is synchronous and scales by the number of antennas. Hence, the solution has been to add more functions to the RU, known as functional splitting. Various approaches to functional splits exist, proposed by enhanced Common Public Radio Interface (eCPRI) [119], small cell forum [117], O-RAN Alliance [120], and 3GPP [110]. The more functions added to the RU, the more complex the RU and less shared processing in the data center. Conversely, having fewer functions added to the RU results in higher requirements to the fronthaul network in terms of capacity and latency. This is illustrated in Figure 16. So far, the O-RAN alliance and small cell forum are the only organizations focusing on only one or two functional split options. Small cell forums propose a division of the physical layer and the MAC layer. Hence, all of the physical layer processing must be handled in the RU, whereas the Media Access Control (MAC) and the above Radio Link Control (RLC) are handled in the DU. For the sake of O-RAN, they propose a functional split 7-2x category A and B, where the difference is that the category B radio unit includes beamforming and is more suitable for high-traffic areas [120].

In the case of hardware for virtualized RAN functions, there is currently significant attention being paid to the CPUs handling physical layer processing [109]. This has already been described in the NFV Section 4.4.2 under Challenges and Solutions. This problem is also affected by the functional split since it is the physical layer functions that require heavy processing. Hence, using the PHY/MAC split proposed by the small cell forum will eliminate the discussion on in-line and look-aside acceleration.

6. Pitfalls and Potentials

The convergence of SDN, NFV, and C-RAN portrays a significant change in the designing, deploying, and management of telecommunication networks. This convergence promises to enhance the scalability, flexibility, and efficiency of the network operations. Along with these, it intends to decrease the Capital Expenditures (CapEx) and OpEx of the entire telecommunication infrastructure.

However, the convergence of these technologies brings some challenges that need to be considered for successful implementation. Following are some challenges and potential solutions:

• Interoperability and Standardization: Different vendors may implement the supporting hardware and software related to SDN, NFV, and C-RAN technologies using their proprietary protocols and interfaces, resulting in interoperability issues. To avoid the interoperability issue, there should be a strong collaboration between the different stakeholders. The telecom industry should promote organizations like ONF [64] and ETSI [121] that support the standardization of these tools and technologies [122].
• Scalability: As telecom networks expand in size to support the increasing demand for connectivity raises the complexity of the network and leads to scalability issues related to SDN, NFV, and C-RAN. Dynamic scaling mechanisms can be implemented to solve the scalability issues. Technologies like orchestration can also be employed to handle the scalability issue [123].
• Network Performance and Latency: Introducing the virtualization and centralization control of network functions may increase the latency and affect real-time applications, for instance, voice and video streaming. To mitigate latency issues, SDN- and NFV-based edge computing capabilities can be deployed near the end-users. The use of efficient routing algorithms and optimization of network architecture can also minimize latency [124].
• Management and Orchestration: It can be challenging to manage and orchestrate virtualized network functions across distributed environments. Utilizing comprehensive management and orchestration platforms can provide centralized control and automation capabilities to ease network operations. These platforms also support multi-vendor environments [125].
• Resource Utilization and Optimization: Amplifying resource utilization in virtualized environments, whether it is computing, storage, or networking, is crucial for achieving cost-effectiveness. Utilizing intelligent resource allocation algorithms and analytics-driven optimization techniques can help optimize resource utilization. Furthermore, network-slicing technologies can facilitate efficient resource allocation for specific applications [126].
• Regulatory and Compliance Issues: Ensuring compliance with regulatory requirements and standards during the implementation of SDN, NFV, and C-RAN solutions introduces some challenges due to the dynamic nature of virtualized networks. It is crucial to be updated with the new rules and regulations set by the regulatory bodies. It is also important to utilize features such as network segmentation and encryption to protect data and avoid compliance issues [127].

Addressing these challenges requires a concerted effort from industry stakeholders, including network operators, equipment vendors, standardization bodies, and regulatory authorities. By overcoming these hurdles, the convergence of SDN, NFV, and C-RAN has the potential to revolutionize telecommunications networks, enabling greater agility, efficiency, and innovation.

6.1. Reducing the Energy Consumption

SDN, NFV, and C-RAN can collectively reduce energy consumption and carbon footprint in the cellular communication network. Here are some features of these technologies that will enhance energy efficiency: SDN provides centralized control of network resources using a software-based controller. It can manage network resources based on real-time network traffic demands and optimize network routing, leading to a more efficient way of using network equipment and reducing the energy needed for data transfer. Using the SDN, energy-aware routing and scheduling algorithms can be formulated that can reduce the power consumption of network devices by prudently selecting the most energy-efficient paths to route the traffic. The SDN-based centralized controller also helps to power down unnecessary networks during periods of low demand, such as un-utilized base stations being switched off.

NFV virtualizes multiple network functions and deploys them as software instances on standard server hardware or the cloud, instead of dedicated physical machines. As NFV consolidates multiple network functions onto a shared hardware infrastructure, it reduces energy consumption and carbon footprint by reducing the number of physical devices required. The 5G network functions, such as Unified Data Management (UDM), Application Function (AF), Network Repository Function (NRF), and many others, can be virtualized to enhance the energy efficiency of the communication network. NFV also helps for the dynamic scaling of network functions to support varying demands by deploying or decommissioning virtual network functions based on the load on the network. This results in reductions in capital and operating expenditures, including energy consumption [95].

C-RAN centralizes the baseband processing functions of multiple RRUs in one data center/cloud/centralized location, and it connects to radio units via high-speed fiber optic cables. This mechanism allows for harmonized resource allocation and interference management, reduces the need for physical network infrastructure, and improves energy efficiency in data transmission and processing. For instance, an investigation [128] has shown that using C-RAN architecture leads to approximately 83% energy reduction in the BBUs.

Collectively, SDN, NFV, and C-RAN provide an efficient way to reduce energy consumption in telecommunication networks by offering centralized control, optimizing resource allocation, reducing the need for physical network infrastructure, and enhancing energy efficiency in data transmission and processing. Nevertheless, the real impact may be different depending on practical implementation and usage scenario.

6.2. Virtualization Feasibility of Network Functions

The feasibility of virtualizing network functions is a critical consideration in networking architectures. While some functions seamlessly transition to virtualized environments, others pose significant challenges due to their complexity, real-time requirements, or resource-intensive nature. Evaluating the virtualization feasibility involves assessing factors such as latency requirements, resource utilization, and the ability to maintain desired levels of QoS and QoE. By carefully examining these factors, network architects can determine the viability of virtualizing specific functions and develop strategies to address challenges encountered during the virtualization process. In this subsection, we tried to differentiate the network functions of Core and RAN based on the complexity of virtualizing them.

6.2.1. Core Functions Easy for Virtualization

The UDM manages user data and subscription information such as credentials, user identification, and access authorization. The UDM works with the Unified Data Repository (UDR), which is a centralized storage in a 5G network for storing user-related data, network information, and service-related parameters to provide access to various 5G network functions [129,130]. These network functions are basically deployed in a cloud-based infrastructure, and they could be virtualized [131]. The Authentication Server Function (AUSF) handles user authentication, security key management, and ensuring secure communication between the network and the user equipment [129,130]. This network function could be easily virtualized [132].

The AF supports traffic routing and interacts with the policy framework [129,130]. AF is primarily software-based, which makes it ideal for virtualization. The NRF is responsible for service discovery and maintaining network function profiles [129,130]. As it primarily deals with control plane information, it can be virtualized without impacting QoS and QoE [133]. Functions like the Network Slice Selection Function (NSSF) and Network Exposure Function (NEF), not directly engaged in data plane operations, are also suitable candidates for virtualization without compromising performance metrics [134].

6.2.2. Core Functions Complex for Virtualization

The User Plane Function (UPF) handles packet routing and forwarding, and it is crucial to ensuring uninterrupted operation [129,130]. Even though UPF can be virtualized, doing so may impact QoS and QoE, especially in scenarios with high data rates or strict latency requirements [134]. Therefore, careful consideration and proper resource allocation are necessary to avoid degradation of QoS and QoE [134]. Similarly, the Access and Mobility Management Function (AMF) is responsible for managing connections and mobility [129,130]. Like UPF, AMF is time-sensitive as it must maintain seamless connectivity for mobile users [129,130]. Although AMF can be virtualized, achieving low latency and high QoS/QoE in a virtualized environment could be challenging [135]. Additionally, the Session Management Function (SMF) handles session establishment and maintenance. While SMF can also be virtualized, maintaining high QoS/QoE presents challenges due to its need for rapid response to session changes.

6.2.3. Virtualization of RAN Functions

In a 5G RAN architecture, the BBU functionality is split into the CU and the DU [136]. The BBU can be virtualized using the OpenAir-Interface (OAI) software platform. However, virtualizing the BBU can be challenging due to the high processing and timing requirements on certain functions [137].

In a RAN, multiple functions can be virtualized [124]. The CU oversees multiple base stations and manages network resources. Virtualizing the CU is easier than the DU because it has lower processing requirements [138]. The control plane and user plane within the CU can both be virtualized using off-the-shelf hardware [139]. Conversely, the DU handles real-time Layer 1 and Layer 2 scheduling tasks [136]. Virtualizing the DU poses challenges due to strict latency requirements and the necessity for real-time operations. Nonetheless, technological advancements have facilitated the connection of the DU to the radio via eCPRI, which can indeed be virtualized [139].

As per our knowledge, while documenting this article we did not find a specific list of network functions that could not be virtualized using NFV. It is important to note that the decision to virtualize a network function depends on various factors, such as the nature of the function, performance requirements, and cost considerations. The feasibility of virtualizing a network function is a subject of ongoing research and development in the field of 5G and beyond cellular networks [140,141,142].

7. Proposed Architecture for Cellular System Based on SDN, NFV, and C-RAN

Building on the energy efficiency challenges identified in Section 3, Section 4 and Section 5 (SDN scalability, NFV performance overhead, C-RAN fronthaul limitations) and the convergence pitfalls mentioned in Section 6, we propose a conceptual integrated architecture, as shown in Figure 17.

A comprehensive review of energy-efficient cellular networks utilizing SDN, NFV, and C-RAN technologies is presented as the primary focus of this article. The proposed architecture shown in Figure 17 demonstrates how these enabling technologies could be integrated into future cellular networks. This schematic representation is intended to serve as a conceptual foundation rather than an implemented solution, as the current work is focused on theoretical analysis and a survey of existing approaches. It is recognized that quantitative validation through simulations or field testing would be required to fully assess the proposed architecture’s performance and energy efficiency benefits. These important practical evaluations have been planned as future work, with more details outlined in Section 8. Here is the functionality of each element of the proposed architecture presented in Figure 17:

• 5G Core Control Plane With Virtualized Core Functions: The core of the 5G and beyond cellular network can be fully virtualized since flexibility is more important for operators to support different use cases. Virtualization of the core will help to develop quicker solutions/applications, and the developed solutions can be deployed and tested faster. It could also pave the way for more innovative and flexible network services, as network operators would have more freedom to customize and optimize their networks based on a single, unified platform [97].
  From an energy-efficiency perspective, virtualizing the 5G core delivers three significant benefits. First, hardware consolidation replaces dedicated appliances with virtualized functions, reducing power consumption by up to 40% through improved resource utilization [93]. Second, dynamic scaling enables core functions like UDM and AUSF to automatically adjust resources based on demand, eliminating energy waste during low-traffic periods. Third, by concentrating virtual resources in data centers rather than in distributed hardware, operators achieve 35% savings in cooling costs through optimized thermal management [95].
• SDN Controller: The SDN controller oversees the entire network element and manages the resource allocation based on the demands. This is a central point that manages the data flow in the network through SDN principles, making the network programmable and more adaptable to varying traffic patterns and demands [97]. The SDN controller serves as the centralized intelligence for network-wide resource management, directly addressing the critical energy efficiency challenges identified in Section 6. By dynamically scaling resources to match real-time traffic patterns, it eliminates energy waste from over-provisioning, reducing idle power consumption compared to static architectures through optimal VNF placement [93]. The controller’s global network view enables intelligent traffic routing that minimizes transmission hops, cutting routing energy while maintaining QoS requirements.
• Backhaul and Fronthaul Link: Fiber optic cables and microwave links are two options for the backhaul connections that link the core network to the RAN architecture. Fiber optic cables are more energy efficient under heavy load conditions, while microwave links are better under low load conditions [48]. Fronthaul connects the BBUs in the C-RAN architecture to RRUs on cell towers, typically using high-bandwidth, low-latency connections such as fiber cables.
• Cloud vRAN: The elements of C-RAN could also be virtualized to curtail the use of energy. The C-RAN elements are given below:

  –

  BBU: Processes the baseband signal and is part of the C-RAN that can be centralized in a data center to serve multiple radio sites.

  –

  PHY (Physical Layer): The layer in the BBU that handles the physical connection to the network.

  –

  MAC/RLC: These layers manage multiple access protocols and data transfer reliability.

  –

  RRC/PDCP: These layers manage radio resources and the convergence of data from different sources.

• NFV in C-RAN: When NFV and C-RAN are combined, they offer more energy-efficient, flexible, and scalable cellular networks. In C-RAN, functions with stringent latency requirements, such as Digital Signal Processing (DSP), are deployed on the dedicated hardware and co-located with RRHs. On the other hand, BBU functionalities such as packet scheduling and user management could be virtualized since these functions are not as latency sensitive and could be decoupled from hardware and deployed as software instance [95].
• User Equipment and Applications: This represents the devices and applications that use the cellular network, such as smartphones, wearables, and vehicles.

This integrated architecture demonstrates a practical implementation of SDN, NFV, and C-RAN technologies for next-generation cellular networks. Using SDN programmable control, NFV flexible virtualization, and centralized processing for RAN using C-RAN, the proposed cellular architecture addresses the critical energy efficiency challenges identified throughout this article while maintaining the performance requirements of 5G and beyond.

Table 5. Architecture components and their solutions to key challenges.

Architectural Component	Solved Challenge	Technical Approach	Validation
Cellular Architecture with SDN Controller	Dynamic QoS Provisioning	OpenFlow/P4Runtime Programmability	Google’s SDN Deployment [65]
Virtualized Core Functions	Hardware Dependency	NFV MANO Lifecycle Management	38% Energy Savings in 5G Cores [93]
BBU Pool with PHY/MAC Split	Fronthaul Capacity vs. Latency Tradeoff	O-RAN/C-RAN	Mavenir’s Open vRAN [143]
NFV-Managed RRUs	RAN Virtualization Overhead	In-line Acceleration for L1 Processing	Samsung’s 22% DU Power Reduction [143]
SDN-NFVO Integration	Multi-Vendor Interoperability	Standardized ETSI Interfaces	OPNFV Testbed Results [101]
Hierarchical Resource Pool	Energy-Inefficient Resource Fragmentation	Dynamic VNF Scaling Plus C-RAN Load Balancing	83% BBU Energy Savings [128]

presents the key components of the proposed cellular architecture and their role in addressing some challenges in the development of a sustainable cellular network. The integration of centralized SDN control, the virtualized network functions of NFV, and the resource optimization of C-RAN plays a key role in improving network efficiency and reducing energy consumption. This holistic approach demonstrates how software-driven architectures can significantly improve scalability, flexibility, and sustainability in next-generation cellular networks.

7.1. Interaction Between SDN Controller, NFV MANO, and C-RAN

The SDN controller is responsible for effectively managing the network resources and data flow across the network infrastructure, while the network functions within the core control plane have their specific roles in managing different aspects of the network’s signaling and control logic, as shown in Figure 17. Here is how the SDN controller and network functions interact and manage the network functionality:

• C-RAN Management:

  –

  Centralization: The SDN controller can centralize the control of the C-RAN infrastructure, allowing for the pooling of BBUs that can be dynamically assigned to RRUs based on the current network load and demand.

  –

  Network Optimization: Through the SDN controller, the network can optimize the routing of traffic between the RRUs and BBUs. It can also manage the split of control and user plane functions to improve performance and efficiency.

  –

  Dynamic Configuration: The SDN controller enables the dynamic reconfiguration of network resources in response to changing traffic patterns. This will help C-RAN to allocate and reallocate radio resources in real-time.

• NFV Management:

  –

  Oversees the Underlying Network Infrastructure: In the proposed architecture, the SDN controller does not directly handle the core network functions. Instead, it oversees the underlying network infrastructure. This infrastructure allows various network functions such as signaling, session management, and authentication to communicate with each other and with the radio access network. The SDN controller ensures that the data plane where actual data flows aligns smoothly with the control plane where decisions are made, resulting in an efficient and agile network [104].

  –

  Interconnectivity: Figure 18 shows the layered architecture of the integration of SDN and NFV systems. It is similar to SDN architecture and consists of infrastructure, control, and application layers. It utilizes the principle of NFV to facilitate the implementation and management of network functions. The SDN controller orchestrates network resources to ensure proper communication among VNFs. Under the management of the VIM, the controller can modify network behavior as required, responding to network user requests [97]. The SDN controller and NFV MANO work together to improve network services. SDN enhances NFV by providing better traffic steering and service chaining. MANO is responsible for managing and orchestrating the virtual network resources and connections between VNFs for a complete network service. This requires the SDN controller and MANO to collaborate for efficient traffic routing [104].

  –

  Deployment and Orchestration: The SDN controller, in collaboration with NFV orchestration tools, can deploy and manage the lifecycle of VNFs, such as scaling out or in, based on the network’s requirements. As depicted in Figure 19, the network orchestration function is utilized to establish network service chaining policies. These policies are also shared with the SDN controller within the NFVI networking layer through the NFV MANO framework. This collaboration provides efficient traffic routing and enhances overall network services [104].

  –

  Policy Enforcement: The SDN controller can enforce network policies at a granular level, directing specific types of traffic to pass through certain VNFs for processing, such as firewalls and load balancers.

• Integration of SDN with C-RAN and VNF:

  –

  Flexibility and Scalability: By integrating SDN with C-RAN and VNFs, the network gains flexibility and scalability, allowing it to support a wide range of services and adapt to changes in traffic patterns or network conditions.

  –

  Resource Utilization: The SDN controller enhances resource utilization by matching computing and radio resources with network demands in real-time, which is critical for the efficiency of both C-RANs and VNFs.

In summary, NFV and SDN both help make networks more flexible and adaptable, but they do different jobs that work well together. NFV focuses on making network functions virtual, which means they are more flexible and efficient. SDN, on the other hand, provides the tools to manage and organize these virtual functions across the network. The SDN controller acts as the brain of the network. It makes decisions to make sure everything runs smoothly, including both the C-RAN and the VNFs. This teamwork between NFV and SDN is crucial for delivering fast, reliable, and adaptable service, especially with the demands of advanced cellular networks like 5G and beyond.

7.2. Supporting Organization

Several companies are leading the development and implementation of virtualized core and vRAN technologies for cellular networks. Samsung, Huawei, Nokia, LG, Ericsson, Qualcomm, ZTE Corporation, NEC Corporation, Verizon, Orange, AT&T, and Cisco Systems are all key players in the field of virtualized core technology [144]. Samsung, Ericsson, Dell, HPE, Intel, Red Hat, Wind River, Mavenir, and Deutsche Telekom are part of a collaborative ecosystem innovating in 5G vRAN [138]. Samsung is actively involved in innovations in Evolved Packet Core (EPC), 4G/5G common packet core, IP Multimedia Subsystem (IMS), Mission Critical Push to Talk (MCPTX), and Cloud Management Systems [143]. Huawei provides a range of solutions, including virtualized core technologies. Mavenir and Deutsche Telekom have deployed Open vRAN in Neubrandenburg, Germany. These companies are shaping the future of 5G networks [143,145].

8. Conclusions

In this paper, a comprehensive survey of energy-efficient cellular networks was presented, focusing on the transformative potential of SDN, NFV, and C-RAN to reduce power consumption and enable sustainable 5G and beyond networks. Through a systematic analysis of the existing literature, key trends, challenges, and solutions in green telecommunications were identified. A conceptual architecture that integrates SDN, NFV, and C-RAN was discussed, demonstrating the potential of these technologies to significantly improve energy efficiency. However, the primary contribution of this work is considered to be the thorough review of state-of-the-art approaches rather than the detailed implementation of a specific architecture.

In Table 1, prior works are categorized based on their contributions to hardware efficiency, renewable energy integration, and emerging paradigms. This structured classification provides valuable insights into existing research efforts, highlighting areas where further advancements are needed. The convergence of SDN, NFV, and C-RAN has been identified as a promising pathway to achieving energy savings of approximately 40 to 50% through dynamic resource allocation, hardware consolidation, and centralized processing. Nevertheless, several critical challenges remain, including issues related to interoperability in multi-vendor SDN/NFV environments and the absence of standardized frameworks to ensure scalability.

It is recommended that future research focuses on the following areas:

• Quantitative validation: Simulations or testbed implementations should be conducted to evaluate the real-world energy efficiency of the proposed architectures under diverse traffic conditions.
• AI-driven optimization: Machine-learning techniques should be employed to enhance SDN/NFV orchestration, enabling more intelligent resource management.
• Standardization efforts: Interoperability gaps, particularly in Open RAN and hybrid cloud-edge deployments, should be addressed to facilitate seamless integration.
• 6G integration: The convergence with emerging technologies, such as terahertz communication and quantum networking, should be explored to further advance sustainability goals.

By addressing these challenges, the transition from theoretical frameworks to practical implementations can be facilitated, ensuring that green networking principles become embedded in future cellular infrastructure. This review is intended to serve as a foundational reference for researchers and practitioners aiming to develop energy-efficient cellular networks, thereby promoting the adoption of sustainable practices within the telecom industry.