MEC and SDN Enabling Technologies, Design Challenges, and Future Directions of Tactile Internet and Immersive Communications

Abstract

Tactile Internet (TI) is an innovative paradigm for emerging generations of communication systems that support ultra-low latency and highly robust transmission of haptics, actuation, and immersive communication in real time. It is considered a critical facilitator for remote surgery, industrial automation, and extended reality (XR). Originally intended as a flagship application for the fifth-generation (5G) networks, their strict constraints, especially the one-millisecond end-to-end latency, ultra-high reliability, and seamless adaptation, present formidable challenges. These challenges are the bottleneck for evolution to sixth-generation (6G) networks; thus, new architects and technologies are urgently required. This survey systematically discusses the most important underlying technologies for TI and immersive communications. It especially highlights using software-defined networking (SDN) and edge intelligence (EI) as enabling technologies. SDN improves the programmability, adaptability, and dynamic control of network infrastructures. In contrast, EI exploits intelligence-based artificial intelligence (AI)-driven decision-making at the network edge for latency optimization, resource usage, and service offering. Moreover, this work describes other enabling technologies, including network function virtualization (NFV), digital twin, quantum computing, and blockchain. Furthermore, the work investigates the recent achievements and studies in which SDN and EI are combined in TI and presents their effect on latency reduction, optimum network utilization, and service stability. A comparison of several State-of-the-Art methods is performed to determine present limitations and gaps. Finally, the work provides open research problems and future trends, focusing on the importance of intelligent, autonomous, and scalable network topologies for defining the paradigm of TI and immersive communication systems.

1. Introduction

The Internet has evolved from an information-sharing medium into a dynamic platform supporting many varied types of applications, from social networking and web browsing to high-definition video streaming and cloud computing. While the conventional Internet is mostly focused on data transmission and multimedia communication, the next generation breakthrough of technology is embodied in the Tactile Internet (TI), which extends significantly beyond transmitting audio, video, and data to enable real-time control, actuation, and touch-based interaction. TI is referred to as one of the most transformative technologies in the fifth-generation (5G) and beyond era, which has the potential to revolutionize industries such as healthcare, education, industrial automation, and entertainment [1].

The TI can be characterized as a network that enables real-time communication of haptic feedback, actuation commands, and control signals with end-to-end latencies in the order of one millisecond. Unlike the conventional Internet, which is designed mainly for information exchange, TI is designed based on interaction, facilitating human–machine collaboration through ultra-low-latency networks that can effectively bridge the digital and physical realms. The vision of TI goes beyond communication; it aims to create a global platform where physical skills and expertise can be transmitted remotely, enabling a new dimension of socio-technical interaction. Examples include remote robotic surgery, real-time cooperative vehicular control, immersive extended reality (XR) environments, and intelligent manufacturing systems [1,2].

While the conventional Internet is optimized for content delivery, such as streaming video or browsing data-hungry sites, it was not created to accommodate tight latency, reliability, or real-time control requirements. TI varies at its fundamental level in its purposes, structure, and applications.

Table 1. Comparison between the TI and the conventional Internet [1,2,3,4].

Aspect	Conventional Internet	Tactile Internet (TI)
Primary function	Information and multimedia delivery	Real-time control, haptic feedback, and skill transmission
Latency tolerance	Tens to hundreds of milliseconds are acceptable	End-to-end latency of ~1 ms required
Reliability	Best-effort service; packet loss compensated by buffering	Ultra-reliability (>99.999%) is essential
Human role	Primarily, information consumption and communication	Direct interaction in cyber-physical loops (human-in-the-loop)
Applications	Web browsing, file transfer, video/audio streaming	Remote surgery, autonomous driving, immersive XR, Industry 4.0
Architectural approach	Centralized/cloud-based, content-centric	Distributed/edge-based, control-centric

provides a comparison between the main aspects of the conventional Internet and the TI. The new properties of TI enable applications that are not feasible with the traditional Internet. Remote surgery, for example, enables a specialist to operate on a remote patient miles away with near-instantaneous haptic feedback. Autonomous transport enabled by TI provides cooperative maneuvers between automobiles with ultra-low latency networking to prevent collisions. Similarly, in gaming and XR, TI enables realistic haptic feedback, increasing the quality of user experience above audiovisual immersion [3]. By connecting human beings, machines, and processes together in real-time interactive systems, TI has the potential to transform society, bridge geographical boundaries, and make possible a new type of interaction beyond the reach of the conventional Internet.

The TI is a primitive change in communication paradigms and is the fourth generation of the conventional Internet. It goes beyond pure data transmission by supporting real-time, ultra-reliable, and low-latency transmission of haptic rendering, actuation, and interaction. In contrast to conventional networks, which are designed primarily for voice, video, and data applications, TI is interfacing the human-to-machine (H2M) paradigm and enables the user to interact with distant environments in real time [1] physically. This ability opens several potential uses, including remote surgery, industrial automation, unmanned systems, and immersive enhanced reality that appears to be a foundational element of future sixth-generation (6G) networks [2].

TI has been identified as a key use case for 5G by the Third Generation Partnership Project (3GPP) and the International Telecommunication Union (ITU) for IMT-2020 [3]. 5G networks can provide an end-to-end latency of ten ms, which is not enough for low-latency applications (e.g., augmented reality (AR) and virtual reality (VR), which require latency below five ms). With the evolution of 5G networks, reducing latency to one ms is critical for realizing high-time immersive interaction. This shift is a stepping stone to 6G, adding to the mix intelligent, distributed computing, and high-fidelity connectivity to unleash the promise of TI. The main steps to realizing TI and immersive communications are an organized sequence from optimizing 5G networks to 6G configurations, as shown in Figure 1. Also, Figure 2 provides the roadmap toward TI standardization [4,5].

Figure 1 presents the step-by-step technological development beginning with the optimization of 5G radio access networks as the foundation for ultra-reliable and low-latency communications. This is succeeded by core network optimization, where flexible architecture and programmable interfaces ensure high scalability and availability. The next phase emphasizes the application of edge intelligence (EI), where computing power is moved closer to the end-users, thereby reducing latency and allowing real-time analysis. These collectively enable the TI, which is the principal outcome of integrating the capabilities of the advanced radio, core, and edge towards ultra-low latency and high reliability for haptic and control applications. The path ultimately culminates with designing 6G architectures, where all these enablers come together as a holistic architecture for enabling future-generation services such as holographic communication, digital twins, and mass-scale autonomous systems.

While current networks support human-to-human (H2H) communication (e.g., voice and video calls) and machine-to-machine (M2M) communication (e.g., Internet of Things (IoT)-driven automation), TI introduces a new paradigm, i.e., H2M communication [6]. This new interaction model enables the user not just to see and hear the remote environment but also to touch and manipulate it in real time with high accuracy. Beyond enabling haptic interactions, TI is a key driver of immersive communications, where multi-sensory experiences such as touch, movement, and spatial awareness are seamlessly integrated with digital environments. Immersive communications will characterize the 6G era, augmenting virtual, augmented, and mixed reality (VR/AR/MR) environments, metaverse worlds, and telerobotics [7,8]. Such applications demand not only ultra-low latency and high reliability but also intelligent network adaptiveness and dynamic resource distribution to keep the user interfaces uninterruptible. Realizing these objectives requires State-of-the-Art networking technology and AI-based efficiencies. Also, novel technologies should be introduced, including software-defined networking (SDN)and EI.

Table 2. Key requirements of TI and immersive communications [7,8,9].

Feature	Tactile Internet	Immersive Communications
Latency	<1 ms	1–5 ms
Reliability	>99.999%	>99.9%
Bandwidth	Moderate	Very high (up to several Gbps)
Jitter tolerance	Very low	Low
Synchronization required	Yes	Yes
Interactivity	Real-time feedback	Real-time sensory engagement

summarizes the key requirements of TI and immersive communications.

Despite its vast potential, TI and immersive communications face several challenges that hinder their widespread deployment and scalability. The key research challenges include the following [7,8,9].

a.

Ultra-low latency and reliability: Delivering low sub-millisecond latency and ultra-reliability (99.99999%) is a major challenge.

b.

Network adaptability and flexibility: Conventional network designs are not able to evolve to accommodate dynamic resource allocation and real-time decision-making.

c.

Computational offloading and intelligence at the edge: Bringing cloud computing closer to the output device adds a latency problem. Hence, the trend is toward EI to provide real-time AI-based processing.

d.

Resource-intensive network orchestration: A scalable, adaptive network architecture requires seamless integration of SDN, MEC, network function virtualization (NFV), and AI-enabled optimizations.

e.

Security and trust: Introducing haptic communications and remote actuation systems generates new security risks and will need blockchain-assisted and AI-robust security infrastructures.

To this end, this work presents an exhaustive survey of the main enabling technologies for TI and immersive communication, with a particular emphasis on SDN and EI. The contributions of this work include the following.

• An in-depth analysis of the evolution of TI and immersive communications, highlighting their significance in 6G networks.
• A comprehensive overview of the difficulties of obtaining ultra-low latency, high reliability, and dynamic responsiveness, especially for real-time haptic and immersive applications.
• Reviewing the enabling technologies, specifically MEC, NFV, SDN, and AI-based optimization, but with particular emphasis on SDN and EI as the essential enablers.
• A comparison of leading research contributions, which compares and evaluates the present State-of-the-Art, its limitations, and open problems.
• An overview of future research opportunities, focusing on intelligent, scalable, and secure network constructs as guiding principles for developing TI and immersive communications.

Methodology

This survey follows a systematic and well-structured method of counting, selecting, assessing, and analyzing the current State-of-the-Art technologies and current research associated with the TI and immersive communications, emphasizing SDN and EI. The methodology is organized into the following significant steps.

(1)

Survey scope definition

The scope of this review is defined to include primary and advanced issues concerning:

• TI, including haptic, actuation, and sensory data in real-time communication.
• 6G immersive communications, including technologies that facilitate XR, VR/AR, and remote interaction.
• Design challenges of TI and 6G immersive communications, including latency, reliability, and flexibility.
• Enabling technologies, especially MEC, SDN, EI, NFV.

The survey incorporates both theoretical and applied research from academia and industry to bridge the gap between technological feasibility and deployment readiness.

(2)

Literature selection strategy

The digital libraries, databases, and global standardization reciprocities employed are IEEE Xplore. ACM Digital Library, SpringerLink, ScienceDirect, Scopus, Google Scholar, ITU Reciprocity, 3GGP Reciprocity, and ETSI Reciprocity. These databases were searched for the following terms: “Tactile Internet”, “SDN in TI”, “Edge Intelligence”, “Immersive communication”, “6G architecture”, “Challenges of TI”, “low-latency networks”, “Network slicing for TI”, “NFV and TI”, “AI-driven networks”, “haptic networks”. The inclusion and exclusion criteria of the references searched are as follows.

• Peer-reviewed articles, conference papers, whitepapers, and technical reports from 2015 to 2025 were included.
• Works focusing on latency reduction, network programmability, reliability enhancement, and system scalability in the TI context were included.
• Survey and review articles on TI, MEC, SDN, EI, and immersive communication were included.
• Articles lacking technical depth or experimental validation were excluded.
• Studies not directly addressing TI or immersive communication use cases were excluded.

(3)

Comparative evaluation

The survey develops a benchmarking framework involving enabling technologies, target use cases, KPI, and evaluation methods to compare and evaluate State-of-the-Art works. This comparison highlights the role of each technology, e.g., SDN and EI, in meeting TI’s stringent quality of service (QoS)/quality of experience (QoE) demands, integration levels of these technologies in emerging 6G infrastructures, performance improvements, and bottlenecks.

(4)

Open challenges and future trends

Based on the literature, this survey outlines future research directions, focusing on federated learning for distributed AI in TI environments, integration of digital twins for predictive control, 6G-native architectures with built-in SDN-EI capabilities, and autonomous and self-healing network topologies.

To increase the readability of the survey,

Table 3. Key notation.

Notation	Description	Notation	Description
3GPP	Third Generation Partnership Project	4G	Fourth generation
5G	Fifth generation	6G	Sixth generation
ITU	International Telecommunication Union	uRLLC	Ultra-reliable and low-latency communications
H2H	Human-to-human	IoT	Internet of Things
AR	Augmented reality	QoS	Quality of service
M2M	Machine-to-machine	VR	Virtual reality
EI	Edge intelligence	MR	Mixed reality
KPI	Key performance indicator	RAN	Radio Access Network
MTP	Motion-to-photon	SDN	Software-defined networking
QoE	Quality of experience	TSN	Time-sensitive networking
NFV	Network function virtualization	UDP	User Datagram Protocol
AI	Artificial intelligence	MEC	Mobile edge computing
DetNet	Deterministic networking	E2E	End-to-end
TCP	Transmission Control Protocol	eMBB	Enhanced Mobile Broadband
Gbps	Gigabits-per-second	MIMO	Multi-input–multi-output
DoF	Degree of freedom	NC	Network coding
RLNC	Random linear network coding	DCT	Discrete cosine transform
HetNet	Heterogeneous networks	LTE-A	Long-term evolution-advanced
CAPEX	Capital expenditure	UE	User equipment
WT	Wavelet transform	IP	Internet Protocol
OPEX	Operational expenditure	ML	Machine learning
GNNs	Graph neural networks	ISG	Industry Specification Group
OFDM	Orthogonal Frequency–Division Multiplexing	ETSI	European Telecommunications Standards Institute
XR	Extended reality	RL	Reinforcement learning
MRL	Meta reinforcement learning	CRL	Continuous reinforcement learning
FPGA	Field-programmable gate array	GPU	Graphics processing unit
CPU	Central processing unit	CDNs	Content delivery networks
GA	Genetic algorithm	RAT	Radio access technology
ESF	Edge sample forecast	MLP	Multi-layer perceptron
DTEC	Digital twin-empowered edge computing	ZSM	Zero-touch network and service management
IIoT	Industrial Internet of Things	QoE	Quality of experience
SSLAs	Security service-level agreements	NOMA	Non-orthogonal multiple access
V2X	Vehicle-to-everything	B5G	Beyond 5G
TIoT	TI of Things	mMTC	Massive machine-type communications
EPC	Evolved packet core network	APIs	Application programming interfaces
PoC	Proof-of-concept	VNF	Virtual network function
LSTM	Long short-term memory	ECI-TeleCaring	Edge computational intelligence-aided haptic-based Tele-caring system
ZTNs	Zero-touch networks	SFC	Service function chaining
OVS	Open vSwitch	RTT	Round-trip time
PoA	Proof-of-authority	RIS	Reconfigurable intelligent surfaces
DAGs	Directed acyclic graphs

introduces the abbreviations considered in this work.

2. Design Challenges of the Tactile Internet and 6G Immersive Communications

The evolution of the TI and its 6G immersive communications represents a fundamental paradigm change in communication networks to deliver real-time control, sensory interaction, and immersive experience over ultra-reliable and ultra-responsive networks. However, their realization and design are faced with enormous challenges that stem from the stringent requirements of these applications, including sub-millisecond latency, extremely high reliability, advanced security, and seamless integration with emerging technologies like augmented and virtual reality, digital twins, and intelligent automation.

Developing and designing systems that can support the TI and 6G-enabled immersive communications is challenging [7]. This is because there is a need to meet a highly stringent set of performance requirements that includes ultra-low latency (≤1 ms round-trip delay), ultra-high reliability (≥99.99999%), deterministic and synchronized response, massive scalability, high spectral and energy efficiency, and strong security guarantees, in dynamic and resource-constrained environments [9]. Figure 3 presents the potential layering structure of the TI systems with the enabling technologies and main requirements for each layer.

At the heart of all these issues lies the necessity for an entire architectural transformation [6]. TI and immersive systems are not merely enhancements of existing broadband networks; they are revolutionary shifts that involve two-way haptic feedback, human-in-the-loop control, and real-time remote interaction, and hence need new architectural and protocol design. One of the biggest technological challenges is achieving an end-to-end latency of less than one millisecond, much lower than is currently possible through the initial 5G deployments. The latency target addresses the timespan from encoding a signal via transmission to processing, actuation, and delivery of feedback. Practically, this requires innovative protocol stacks, edge computing offloading via MEC, and a rapid routing architecture with fewer intermediate hops [7].

Closely related to latency is the issue of reliability, which in the TI context must reach ultra-high levels to ensure that haptic feedback and control signals are transmitted error-free. Applications such as remote surgery or autonomous transport need to observe a packet error rate below 10⁻⁷. This is sufficient reliability, such that the failure of a single packet will not result in catastrophic failure, especially in haptic feedback systems, where losing data provides an incorrect or delayed physical reaction. Current transmission protocols, such as TCP (with retransmission delay) or UDP (without reliability guarantees), are inherently insufficient. For this purpose, new transport layer mechanisms such as deterministic networking (DetNet) and time-sensitive networking (TSN) are being explored to meet these needs [10].

Security is yet another imperative issue that becomes increasingly complex in heterogeneous, distributed, and latency-constrained environments. Unlike traditional Internet applications, the TI requires end-to-end secure communication in real time over decentralized and heterogeneous networks (HetNet). While conventional security models such as IPSec offer strong protection for standard applications, they fall short when utilized in haptic or immersive applications. For instance, cryptographic algorithms incur latency overheads unacceptable for sub-millisecond environments. AR, VR, and teleoperation applications require lightweight security architectures with no processing and communication latency [7]. Furthermore, the heterogeneity of devices, ranging from low-power haptic actuators to edge nodes involving AI, calls for a multi-layered, light-weight security architecture that can be dynamically tuned based on interaction context, particularly in immersive scenarios, e.g., XR. Therefore, new encryption algorithms, zero-trust architectures, and paradigms of secure computing that are sensitive to latency must be examined.

The concept of perceptual transparency adds another level of richness. In the applications of the TI, user experience quality cannot be defined as the raw latencies or reliability measures but as the smoothness with which the system imitates natural-world interaction [7]. That requires ultra-high-precision coordination between master and slave devices to allow the operator to perceive a complete and uninterrupted loop of interaction. Any deviation, delay, or jitter breaks the illusion of direct manipulation and undermines the immersive experience. Unproblematic system integration is critical to the success of TI deployments. The IoT platforms, legacy networks, cloud-native services, and mobile broadband networks need to be harmonized to meet the unique requirements of TI. More importantly, though, these systems must accommodate the current applications as well as pending applications based on real-time haptic communications and sensor fusion of the senses. This involves far-reaching design complexity from the aspects of backward compatibility, standardization, and architectural consistency.

Haptic device engineering is still another bottleneck. These devices, enabling remote touch and manipulation feedback, are characterized by their degrees of freedom (DoF). To serve the requirements of immersive communications, future haptic devices will need to provide more advanced interactions, improved accuracy, and quicker response times. This requirement stimulates the development of actuator and sensor technology as well as the ability of such machinery to negotiate through constrained and time-varying networks [11].

Regarding data encoding and decoding, applications of TI are quite unlike audio-visual communication. Haptic communication is inherently two-way and consists of several simultaneous data streams, like force, velocity, position, and torque. It creates a much larger quantity of data and needs highly energy-efficient coding techniques. Network coding (NC), especially random linear network coding (RLNC), is a promising candidate [12]. Unlike block-based systems, RLNC uses a sliding window approach and supports in-network recoding, which enhances flexibility and robustness. Further, lossy real-time compression algorithms based on discrete cosine transform (DCT) or wavelet transform (WT) are being considered to keep payload sizes minimal with perceptual quality ensured.

TI routing is another particular challenge. TI needs constant and determinable delay budgets compared to traditional IP networks, where there is some tolerance for latency variation. The packets should be transmitted with durations not exceeding 33 µs, which is a requirement much tighter relative to LTE or 5G OFDM symbol duration. This calls for developing new routing schemes capable of responding ultra-fast, minimizing hop counts to a bare minimum, and utilizing edge resources efficiently. Placing MEC nodes near user equipment (UE) significantly reduces the communication path and latency. However, orchestration and mobility management among the nodes are open research problems.

Table 4. Main challenges of TI and immersive communications.

Challenge	Description	Existing Limitation	Required Advancement
End-to-end latency	1 ms or less total round-trip delay	Current 5G latency > 5 ms	MEC, SDN, intelligent scheduling
Ultra-reliability	Packet error rate < 10−7	UDP/TCP is inadequate	New protocols, DetNet, TSN
Security	Multi-layered, latency-aware, context-sensitive security frameworks	IPSec is too heavy for real-time haptics	Lightweight cryptography, AI-driven threat detection
Perceptual transparency	Seamless and realistic interaction for the user	High jitter, poor synchronization	Real-time feedback loops, precision synchronization
Haptic device design	Increased DoF, better feedback resolution	Limited precision, low responsiveness	Advanced sensors, AI-based tactile rendering
Coding and decoding	Efficient compression and error-resilient transmission of bidirectional haptic signals	High payload, retransmission delays	RLNC, real-time compression (DCT/WT)
Routing and mobility	Ultra-fast, adaptive routing paths optimized for latency and reliability	IP-based routing is too slow	Edge-aware routing, dynamic topology adaptation
System integration	Interoperability with 5G, IoT, and cloud-native infrastructure	Fragmented standards, legacy systems	Unified architecture, AI-enabled orchestration

summarizes the specifications of the previously introduced challenges of developing TI and 6G immersive communications.

3. Mobile Edge Computing for Tactile Internet and Immersive Communications

Mobile edge computing (MEC) is a transformative communication technology that has been integrated into the radio access network (RAN) of next-generation cellular networks to enhance computational efficiency, reduce latency, and improve network performance. Through the decentralization of cloud computing infrastructure, MEC shifts the location of computational resources, storage, and processing capabilities from the far end of the network (toward the cloud data center) to the network’s edge, where the end-user resides [13]. In the standard traditional cloud computing architectures, data is sent over multiple network hops from end devices to some remote cloud servers, resulting in higher latency, higher bandwidth usage, and potential overload of the backbone network infrastructure. MEC mitigates these shortcomings by offloading computational tasks and network intelligence to the edge nodes (e.g., base stations, access points, and local data centers) to ensure that subscribers can access remote services within one or two communication hops at most. That close relationship with end-users lowers the data transmission latency and improves the real-time response capability of the applications [14].

MEC is an essential part of 5G and upcoming 6G networks and critically underlies the enabling of ultra-reliable low-latency communication (uRLLC), massive machine-type communication (mMTC), and enhanced mobile broadband (eMBB) services. By integrating MEC with AI and machine-learning (ML) capabilities, networks can dynamically optimize resource allocation, predict traffic patterns, and enhance overall network efficiency [15]. MEC was first introduced by the European Telecommunications Standards Institute (ETSI), which announced an Industry Specification Group (ISG) dedicated to designing and developing MEC technology for cellular networks in December 2014. In 2017, ETSI extended the concept of edge computing from being committed to cellular networks only to cover other access networks besides mobile, such as Wi-Fi [16]. This changes the MEC abbreviation from mobile edge computing to multi-access edge computing.

MEC is a paradigm shift from a centralized scheme, i.e., remote centralized cloud, to a distributed scheme. Introducing MEC technology to the edge of the access networks achieves various benefits, including the following [14,15,16,17].

• Reducing end-to-end latency of the requested services: MEC dramatically decreases the end-to-end latency needed for real-time applications such as autonomous driving, AR, VR, and robotic control over distance. This is reasonable since the communication distance is reduced.
• Reducing network congestion: This is achieved because MEC reduces the amount of traffic that passes through the core network. Part of the traffic is handled by MEC servers at the edge of the RAN, without going through the core network.
• Achieving higher bandwidth efficiency: By processing data at the edge, MEC minimizes unnecessary data transmission to centralized cloud servers, reducing backhaul congestion and improving network scalability.
• Localized data processing: Sensitive data can be processed more locally at its source, improving privacy, security, and regulatory compliance.
• Increasing the overall network flexibility: MEC servers can be deployed in many locations, and computing tasks can be handled locally without going through the core network.
• Increasing overall system reliability and availability: Reducing the communication distance and the number of involved nodes in the communication process increases overall system reliability.
• Seamless service continuity: MEC allows real-time hand-off control and context-aware services and therefore provides an uninterrupted user experience on an application level, such as competent healthcare, industrial automation, and intelligent transportation systems.
• Energy efficiency: Offloading task computation to the local MEC servers decreases the energy consumption of user devices, which is translated into a longer-lasting battery for mobile devices and a higher efficiency of IoT devices.

This comes at the expense of the overall system cost, which includes both capital expenditure (CAPEX) and operational expenditure (OPEX). MEC is a highly revolutionary technology that enables next-generation networks by extending computation and intelligence to its end users. With the previously introduced benefits of the MEC paradigm, it represents one of the main key-enabling technologies of TI, 6G immersive communications, and other uRLLC applications. MEC is a milestone enabler of the TI and immersive communications that cuts a long part of the road to the sub-millisecond challenge.

Challenges with MEC-Based RAN for Tactile Internet and Immersive Communications

Integrating MEC into RAN empowers the future of TI and immersive communications; however, it also establishes several important research issues. MEC is designed to decrease latency, maximize bandwidth utilization, and improve real-time data processing; however, there are many challenges with deploying MEC for TI and immersive communications. This subsection investigates the critical issues of the MEC-based RAN for TI and immersive communications, including the following.

• Optimal MEC server placement and location selection.
• Edge server architecture and resource allocation.
• Integration of EI for adaptive computing.
• Seamless integration of MEC with RAN components.
• Interoperability between MEC-enabled RAN and core networks.

A.

Optimal MEC server placement and location selection

MEC works by locating computing resources closer to the end user and thus reducing end-to-end latency. However, the optimum positioning of MEC servers in the RAN infrastructure remains challenging due to the following [18,19].

●

Heterogeneous network structures (e.g., urban, rural, and industrial settings).

●

Dynamic user mobility patterns.

●

Traffic load variations across different locations.

The two key research questions are where MEC servers should be deployed to achieve minimum latency and maximize computational efficiency, and how MEC placement strategies can adapt to changing network conditions and user density variations. AI-based dynamic MEC placement algorithms to optimize real-time resource allocation and hierarchical MEC deployment to balance ultra-low latency and cost-effectiveness can be potential solutions.

Table 5. Specifications of existing MEC placement strategies [18,19,20].

MEC Placement Strategy	Advantages	Challenges
Base station-level MEC deployment	Reduces latency significantly by placing servers at cell towers.	Increased CAPEX/OPEX due to the large number of MEC nodes required.
Centralized MEC clusters	Cost-effective and allows shared resource pooling.	Higher latency due to distance from end-users.
Hybrid MEC placement	Balances cost and latency by placing MEC nodes at selected base stations.	Complexity in managing resource allocation dynamically.
AI-driven adaptive MEC placement	Uses AI to adjust MEC placement based on real-time traffic analysis dynamically.	High computational overhead requires continuous learning and adaptation.

provides the existing MEC placement strategies, as well as the benefits and challenges associated with each strategy.

B.

Edge server architecture and resource allocation

The architecture of MEC servers should be optimized to handle heterogeneous workloads efficiently. TI and immersive communications applications (e.g., telemanipulation, haptic interaction, and ultra-low-latency video streaming) require high real-time processing. The key considerations for edge server design include the following [6,14].

(1)

Computational capacity (How to allocate different resources (CPU/GPU/FPGA) efficiently for diverse workloads?).

(2)

Storage and caching mechanisms (How to reduce data retrieval delays while maintaining data consistency across multiple MEC nodes?).

(3)

Energy efficiency (How to minimize power consumption in MEC-enabled networks?).

Virtualization and AI technologies can provide potential solutions to address these issues while designing and developing MEC servers. Virtualizing edge server architectures (using NFV and SDN) to optimize flexibility and deploying AI-based workload scheduling algorithms to improve computational efficiency dynamically are promising solutions for the evolution of MEC platforms [21].

Table 6. Specifications of existing MEC platforms [15,17].

Edge Server Type	Advantages	Challenges
General-purpose MEC servers	Supports diverse workloads and scalable architecture.	Increased power consumption and resource contention.
Specialized AI-driven MEC servers	Optimized for ML/AI-based decision-making.	Higher infrastructure costs and complexity.
Hybrid edge–cloud architecture	Balances local and cloud processing for optimized resource use.	Requires intelligent workload partitioning and scheduling.

provides the benefits and challenges of the existing common types of MEC platforms.

C.

Seamless integration of MEC with RAN components

The seamless integration of MEC within the RAN infrastructure requires efficient resource allocation and traffic offloading, seamless handover mechanisms for mobile users, efficient migration schemes, and coordination between MEC servers and RAN controllers.

Table 7. Different challenges with the MEC integration with RAN.

Integration Challenge	Impact on Performance	Potential Solutions
Dynamic user mobility	Increases handover frequency and service disruption.	AI-driven mobility prediction and seamless MEC handover mechanisms.
Network slicing complexities	Allocating dedicated slices for MEC workloads is complex.	SDN-based dynamic network slicing for flexible resource allocation.
Latency constraints	MEC integration must not introduce additional processing delays.	Lightweight virtualized MEC deployment with optimized latency-sensitive routing.

provides different challenges with the MEC integration into RAN. It also provides the impact of each challenge on network performance and the potential solutions for each challenge [22].

D.

Interoperability between MEC-enabled RAN and core networks

MEC-based RAN must efficiently interconnect with the 5G/6G core network to maintain seamless traffic flow and service continuity. However, core network integration presents challenges, including synchronization delays between MEC and centralized cloud servers, security vulnerabilities in edge-to-core data transmission, and load balancing between MEC and centralized computing resources.

Table 8. Challenges with the MEC integration of MEC/RAN with the core network.

Core Network Challenge	Impact	Potential Solution
Security and privacy	Increased risk of cyberattacks on MEC nodes.	Blockchain-based decentralized authentication mechanisms.
Data consistency and synchronization	Data loss and inconsistency across MEC and cloud servers.	AI-driven predictive caching and synchronization techniques.
Interoperability between multi-vendor MEC solutions	Vendor-specific architectures create fragmentation.	Open-source MEC frameworks for standardization (e.g., ETSI MEC).

summarizes the challenges of integrating MEC platforms with the core network [23].

4. Edge Intelligence for Tactile Internet and Immersive Communications

TI requires an end-to-end delay of a millisecond, including the time of receiving the feedback signal. Deploying MEC at the edge of RAN reduces the communication latency; however, it cannot achieve the required one-millisecond challenge. This is mainly due to light limitations. Signals travel with a maximum speed of light, and for a propagation latency of one millisecond, signals can travel only 300 km, neglecting any other delays. Considering communication on both sides, signals can travel only 150 km away for feedback signals. Considering other delays, the distance is reduced to 15 km with the deployment of MEC technology [24]. This distance should be extended for TI and immersive communication applications to cover communication between large cities and different countries. EI provides a solution for such a challenge. The introduction of EI can achieve the required sub-millisecond latency by building a model for the remote environment that can predict the feedback signal and mimic the response of the remote side. This model is referred to as model mediation. Using AI algorithms, it can predict feedback signals and reduce latency by avoiding the time required for waiting for feedback signals. The model updates itself periodically based on the received real feedback signal.

EI stands for the integration of AI and edge computing, e.g., MEC, by enabling real-time data processing, intelligent decision-making, and further autonomy within 6G networks. In contrast to conventional cloud-based configurations, EI minimizes both latency and privacy loss while also maximizing bandwidth by performing processing at the generation point [25]. EI can be introduced to TI and immersive communication networks to provide dynamic resource allocation, predictive traffic management, and automated service orchestration [7]. EI can be used to assist 6G immersive communications by reducing end-to-end latency, enabling real-time rendering, and supporting data personalization. Processing data locally minimizes the delay between user actions and system responses, which is crucial for a seamless experience. Also, edge nodes can handle the rendering of complex AR/VR scenes, reducing the load on user devices. Furthermore, AI algorithms at the edge can adapt content based on user preferences and behavior in real time.

Conventional deep learning-based models are highly computationally expensive and, therefore, are not efficient for edge deployment. Using lightweight AI models, federated learning, and edge–cloud cooperation, EI can overcome technical issues of use cases of these applications by providing new ways of connectivity and interaction. Lightweight neural networks, e.g., MobileNet, SqueezeNet, and TinyML, have been proposed for efficient execution on edge devices with constrained resources [26]. Furthermore, implementing EI for MEC-enabled RAN introduces challenges related to computational overhead for AI model inference at the edge, real-time data processing constraints, and scalability of edge AI models in large-scale networks [27].

Table 9. Challenges with the EI deployment for TI and immersive communication networks.

Challenge	Technical Considerations
Resource constraints	Limited processing power at edge nodes requires model optimization techniques such as pruning and quantization.
Security and privacy	Differential privacy and homomorphic encryption mitigate risks associated with decentralized AI.
Scalability	Dynamic resource allocation and SDN enable efficient scalability.
Energy efficiency	AI model compression and energy-aware scheduling optimize power consumption.
Heterogeneous hardware	Edge devices vary in computational capabilities, requiring adaptive AI models.
Data synchronization	Real-time updates must be synchronized across distributed edge nodes to ensure consistency.

provides a detailed discussion on the challenges associated with EI deployment for TI and immersive communications.

Table 10. Key components of EI frameworks [28,29,30].

Component	Description
Edge nodes	Devices or micro-data centers at the network periphery that execute AI models locally.
AI/ML model	A hybrid AI framework in 6G integrates reinforcement learning, deep learning, and graph neural networks (GNNs) for optimal network control and decision-making.
Federated learning	A decentralized AI training approach that enables model learning without centralized data aggregation.
Edge orchestration	Software that manages the deployment, scaling, and migration of edge services.
Digital twins	Virtual replicas of network entities are used for predictive analysis and real-time optimization.
5G/6G infrastructure	High-speed, low-latency connectivity supporting real-time edge AI processing.
IoT and sensor networks	Data sources that provide real-time inputs for AI-driven decision-making.
Neuromorphic computing	Brain-inspired AI chips are designed for energy-efficient edge inference.
AI acceleration hardware	Specialized processors like TPUs and GPUs for optimized AI computations.

provides the standard key components of the EI platforms. Edge orchestration is a main part of the EI platform, which involves managing the deployment, scaling, and migration of edge services. It consists of several key components, including service placement, load balancer, and fault tolerance modules. Service placement is used to decide where to deploy services based on user demand and resource availability. However, the load balancer module distributes workloads across edge nodes to prevent bottlenecks. Furthermore, the fault tolerance module is used to ensure continuous operation through redundancy and self-healing mechanisms [28].

Although EI emphasizes local processing, the need for cooperative approaches with the cloud cannot be avoided for some situations, such as when global context is needed, or extremely computationally demanding operations need to be performed. This collaboration can be achieved through offloading, federated learning, and edge caching. Offloading schemes are used for cloud offloading of computationally demanding tasks, while edge offloading of latency-sensitive tasks. Federated learning is a distributed ML paradigm in which edge devices jointly train a global model without the exchange of raw model data. Edge caching is used for embedding frequently accessed data at the edge to decrease latency and bandwidth consumption.

EI can be deployed for dynamic optimization of network parameters to achieve the required QoS and QoE for immersive communications and TI applications. This includes deploying reinforcement learning to estimate and predict network traffic and perform load balancing dynamically. Also, graph neural networks can be used to improve routing efficiency for complex network topologies, and multi-agent AI can be deployed to assist distributed decision-making for enhanced resiliency in TI and 6G immersive communication networks [28]. This can assist in achieving the required key performance indicators (KPI) required for TI and 6G immersive communication [31].

Table 11. EI solutions to meet the required KPIs of TI and 6G immersive communication.

KPI	Challenge	Solution	Discussion
Latency efficiency improvement	Achieving sub-millisecond latency for haptic feedback.	Edge caching	Pre-loading data and models at the edge to reduce processing time.
		Lightweight AI models	Using compact neural networks that can run efficiently on edge devices.
		Network slicing	Allocating dedicated network resources for TI applications.
Reliability and fault tolerance	Ensuring continuous operation in dynamic environments.	Redundant edge nodes	Deploying multiple edge nodes to provide failover capabilities.
		Self-healing algorithms	AI algorithms that can detect and recover from failures autonomously.
Security and privacy	Protecting sensitive data from unauthorized access.	Edge-based encryption	Encrypting data at the edge before transmission.
		Haptic codecs	Haptic codecs are designed to encode and decode procedures for transmitting and reconstructing tactile and kinesthetic feedback over communication networks.
		Federated learning	Training AI models without sharing raw data.

provides the EI solutions for different challenges associated with achieving the required KPI for TI and 6G immersive communications. Furthermore,

Table 12. EI solutions for critical challenges with TI and 6G immersive communication.

Challenge	Challenge Justification	Solution	Discussion
Bandwidth and computational requirements	AR/VR applications require high bandwidth and computational resources.	Edge rendering	Offloading rendering tasks to edge servers reduces the computational burden on user devices.
		Content delivery networks (CDNs)	Distributing content across multiple edge nodes to reduce latency and bandwidth usage.
User mobility	Ensuring consistent performance as users move across different network areas.	Edge handover	Seamless transfer of sessions between edge nodes as users move.
		Predictive caching	Anticipating user movement and pre-loading content at the nearest edge node.
Energy efficiency	Prolonging battery life for AR/VR devices.	Energy-efficient AI models	Using lightweight models that consume less power.
		Edge offloading	Reducing the computational load on user devices by offloading tasks to edge servers.

provides EI solutions for heterogeneous challenges with the development of TI and 6G immersive communications.

5. Related Studies of MEC/EI-Based Tactile Internet System

This section introduces the existing works for developing TI and associated services using an MEC-based network structure. These studies provide MEC-based RAN for ultra-low latency 5G applications and TI [26,27,28,29,30,31,32,33,34,35,36]. A. Ateya et al. [32] developed a novel multilevel MEC structure for TI and uRLL services. Unlike other literature, the authors provided two levels of edge servers at the RAN, besides a core network cloud. Each cellular base station is connected to either a mini-cloud edge server or a micro-cloud edge server. Micro-cloud servers have limited computing capabilities and should be deployed for cells with expected fair-to-medium workloads. Cellular cells with massive devices and traffic are expected to deploy mini-cloud edge servers with higher computing capabilities than micro-cloud edge servers. The system deploys a hierarchy scheme, such that each group of micro-cloud edge servers is connected via a high-speed fiber connection. Their system was simulated in terms of latency and achieved higher efficiency. The introduction of mini-cloud edge servers achieved many benefits in terms of latency and task blocking. This comes at the cost, both OPEX and CAPEX, of the system.

D.A. Meshram et al. [33] developed a telesurgery system based on 5G and MEC technology. The main contribution of this work is the design and implementation of robotic telesurgery with haptic feedback. The article describes robot-assisted telesurgery operations and the importance of such systems. The developed telesurgery system employs MEC, SDN, and NFV over 4G and 5G networks. The article does not specify a location for MEC servers. The experimental setup of the developed telesurgery system deployed MEC servers at multiple geographical locations for 4G networks and at multiple radio access technology (RAT) cell aggregations, which act as an edge of the core 5G network. The evaluation process of the system considered real-time medical video transmission over the proposed TI structure, which represents the novelty of this work compared to previous studies. The system was simulated for performance evaluations, and the QoS parameters were measured. Results indicated that the proposed structure achieved a 62% reduction in latency and a 108% increase in throughput compared to 4G systems.

A. Braeken and M. Liyanage [34] provided a framework for remote monitoring of patients over a 5G network. The system deploys the MEC paradigm as mini-cloud servers connected to each cellular base station. The system deploys IoT for patient monitoring and integrates IoT devices with the MEC layer. This IoT-MEC integration over a 5G system achieves bandwidth efficiency, latency, and system availability. The authors mainly considered the authentication process for such MEC-IoT integration. A key agreement between both MEC servers and the IoT network was developed for the remote monitoring system. The authentication scheme was built using a public key-based mechanism.

R. Gupta et al. [35] gave insights into telesurgery operations for healthcare 4.0 over TI. The first part of the work provided a detailed review of the evolution of tele-health care systems until 2020, as well as a list of available market robots used for telesurgery operations and their evolutions. The second part of the work considered the communication network for telesurgery operations. The work mentioned two main communication systems for telesurgery operations: 5G-enabled TI systems and traditional networks. The work compares telesurgery operations over traditional networks and the TI system. Since TI and other uRLL 5G services require an end-to-end latency of 1ms or less and a reliability of 99.999%, they provide an efficient communication network for telesurgery operations. The work provided architecture for 5G-enabled TI systems for telesurgery operations that achieve higher reliability and latency efficiency than other traditional networks. The authors considered MEC and SDN as the main parts of the developed TI structure.

M. Shao et al. [36] considered finding optimum locations for MEC servers. The investigated MEC paradigm deploys distributed MEC servers between base stations, i.e., the edge of RAN, and end devices. The authors considered the dynamic resource requirements of base stations and defined MEC server planning as a joint server placement and resource allocation optimization problem. They assumed that end device computing resources are collected by base stations that send resource requests to edge servers. The authors formulate the planning problem of MEC servers via an uncertain programming formulation. In their formulation, the placement of MEC servers and the allocation of MEC server resources are jointly determined. The server placement problem is an NP-hard problem; the authors introduced a learning-based framework for such a joint optimization problem. The proposed learning-based framework integrates the genetic algorithm (GA), stochastic simulation, and a neural network. The neural network links the GA and the stochastic simulation. The stochastic simulation is used for resource allocations; however, the GA is introduced to determine the server’s location. The proposed joint optimization problem for MEC planning was simulated for two main networks: a small-scale network and the real network of Shanghai Telecom. Simulation results of the developed joint optimization algorithm indicated the optimum MEC servers’ location when considering latency and base station demands.

M. Maier et al. [37] developed an infrastructure of small-cell networks based on FiWi and MEC for enhancing the HetNet of long-term evolution-advanced (LTE-A) and future 5G uRLLC, mainly TI. The work mainly considered enhancing LTE-A HetNets using FiWi side by side with edge intelligence, i.e., AI-based MEC servers. The authors developed an ML algorithm at the edge server level to achieve a seamless and immersive TI system. The developed AI algorithm is a multi-layer perceptron (MLP) artificial neural network. To compensate for the latency of the haptic feedback signal, authors have introduced an edge sample forecast (ESF) module based on the developed MLP framework. The developed ESF employed multiple-sample-ahead-of-time forecasting, which keeps the response time of a human operator small. The ANN is trained using a dataset of 6-DoF teleoperation traces that contains 59,710 force feedback samples, while the waiting deadline was set to 1 ms. For performance evaluation, another dataset of 1000 samples is used, and the results indicate that the developed MLP achieved a percentage improvement of 27.8. Upon the evaluation process, the authors found that decreasing the packet rate of haptic data decreases the average end-to-end latency.

Caolu Xu et al. [38] addressed the key challenges of wireless multi-user interactive VR of the metaverse, including ultra-low latency, high bandwidth, and computation-intensive tasks. The authors employed a cooperative edge-device computing paradigm to reduce motion-to-photon (MTP) latency constraints. Their method models serial-parallel task processing upon a foreground–background separation mechanism where rendering tasks are scheduled and prioritized. For optimizing sensor data age and mobile power consumption, the work jointly optimizes rendering and MEC resource allocation under constraints of MTP. A security-reinforced learning algorithm, AQM-CUP, has been introduced for real-time managing task queues with discarding out-of-date tiles and adjusting decisions in response to changing conditions. Experimentation verification proves improved immersion experience and reduced user power consumption, showing that AQM-CUP far exceeds conventional schemes in convergence time and performance.

Maria Crespo-Aguado et al. [39] designed a hyper-distributed IoT–edge–cloud platform for real-time digital twin applications in industrial and logistics environments, as a living lab for future 6G use cases. The platform has been constructed with private 5G connectivity combined with AI/ML-based orchestration to spread computing resources along the IoT in the edge, cloud continuum. It supports high-throughput, low-latency services and can host various digital twin applications, e.g., immersive remote driving. Performance trials confirmed near-theoretical data rates (up to 552 Mb/s downlink and 87.3 Mb/s uplink in the n78 band). In contrast, QoE trials confirmed the ability of the system to sustain immersive and interactive user experiences.

Jiadong Yu et al. [40] developed a dynamic resource management scheme for metaverse use cases of MEC-based VR content streaming through a continuous reinforcement learning (CRL) paradigm. Unlike conventional offline RL practices that ignore desynchronization between the virtual and the physical worlds, the proposed CRL-based approach adapts dynamically to a real-time shifting environment. A digital twin-empowered edge computing (DTEC) system was developed to maximize QoE through attention-based resolution adaptation and effective allocation of computing and bandwidth resources. Three CRL variants were introduced and compared, i.e., CDDPG, PER-CDDPG, and FPER-CDDPG. Among them, FPER-CDDPG exhibited the best performance in reducing latency, improving QoE, and ensuring scalability for a growing number of users.

Rodrigo Asensio-Garriga et al. [41] developed a secure and efficient resource management framework for vehicle-to-everything (V2X) communications in beyond 5G (B5G) networks according to the zero-touch network and service management (ZSM) paradigm. Considering the vulnerability of MEC infrastructure to DDoS attacks, especially in dynamic multi-domain V2X environments, the authors proposed an end-to-end slicing approach that integrates autonomous security measures in terms of security service-level agreements (SSLAs). Their solution spans the entire 5G architecture, vehicle devices, RAN, MEC, and core network, and dynamically deploys coordinated countermeasures upon identifying DDoS attacks. Experimental results showed that the framework can effectively detect and counter attacks with low latency, meeting the reliability needs of mission-critical V2X services. This work adds to the ETSI ZSM proof-of-concept for secure slice assurance.

Yeabsira A. Ashengo et al. [42] addressed the issue of task offloading in uRLLC-based 5G networks, wherein the UE lacks sufficient computational resources, and the MEC servers are resource-starved because of excessive usage. To strike a balance between task allocation, they proposed an asynchronous meta reinforcement learning (MRL)-based offloading framework that allocates tasks based on their urgency and inter-dependency. The bottom-up asynchronous MRL model facilitates rapid adaptation in heterogeneous and homogeneous settings. Simulation results demonstrated a 60% improvement over baseline methods in training latency, power efficiency, and resource utilization, reflecting its applicability to dynamic uRLL scenarios.

Xuguang Zhang et al. [43] proposed a structured solution with scalable haptic coding, deep learning-based imperceptible QoE awareness based on emotional immersion metrics, and an on-demand cooperative network resource scheduling strategy. The architecture optimizes resource utilization and maintains high user experience rates in heterogeneous IoT environments. The simulation results also confirmed the devised method for maintaining haptic communication in sparse network environments. Linlin Zhao et al. [44] investigated the statistical upper bounds of E2E latency of teleoperation systems by integrating grant-free transmission with K-repetition and MEC. The authors proposed a service process considering transmission success probabilities, pilot collisions, and computation delays. They derived latency violation probabilities using martingale theory and presented an optimal latency budget allocation strategy between the uplink and computation phases. The simulation results affirmed the correctness and applicability of the proposed latency-bound estimation approach.

The TI of Things (TIoT) is redefining real-time applications with haptic interaction and immersive experience in sectors like healthcare and manufacturing, where ultra-reliable low-latency communication is essential. Omar Alnajar et al. [45] proposed an end-to-end network slicing architecture encompassing dynamic resource scaling, intent-aware allocation, and proactive bandwidth optimization to meet these requirements. Focusing on performance limitations and data privacy in centralized deep learning, the authors explored federated learning using light models. The framework integrated blockchain, SDN, NFV, and MEC technologies to enhance federated learning.

Vaibhav Fanibhare et al. [46] introduced non-orthogonal multiple access (NOMA) as a key driver for the TI that needs ultra-low latency and real-time haptic communication. The authors proposed a downlink power-domain SISO-NOMA system with multiple users and a base station and derived analytical expressions for SINR, sum rate, and fair power allocation. Performance analyses under two-user and three-user case scenarios proved that SISO-NOMA significantly outperforms the conventional orthogonal multiple access schemes based on bit error rate, latency, and sum rate. The authors also investigated the outage probabilities under several fixed and fair power allocation methods. They compared the performance with a 4×4 MIMO-NOMA system based on zero-forcing beamforming to prove the potential of NOMA for dynamic spectrum usage and user fairness in TI environments.

Table 13. Comparison between recent MEC-based TI systems and their specifications.

Ref.	Tactile Application	Tactile Data	Cellular Network	MEC Location	KPI	Evaluation Process
[33]	Telesurgery	Real-time medical video	4G	No specific location	Latency Throughput	Simulation
[34]	Remote healthcare–Monitoring	_	5G	No specific location	Security	Simulation
[35]	Telesurgery and Remote Healthcare	_	5G	No specific location	_	No evaluation
[36]	Telesurgery	_	5G	No specific location	Latency	Simulation
[32]	No specific app.	_	5G	Multilevel structure	Latency	Simulation
[37]	No specific app.	Force feedback samples from 6-DoF teleoperation	5G	Distributed among ONU-BSs/MPPs	Latency	Simulation
[42]	No specific app.	Video data	5G	No specific location	Latency Throughput	Simulation

provides a comparison of the previously introduced works and the specification of each system.

6. Software-Defined Networking for Tactile Internet and Immersive Communications

SDN is a communication paradigm aimed at solving the next-generation cellular core networks’ evolving needs. It marks the beginning of a new era in telecommunication, or software-based networking, where traditional hardware-based solutions are revamped based on software-controlled and programmable technologies. In conventional networks, control logic is explicitly embedded into separate hardware chunks, i.e., routers and switches. On the other hand, SDN offers a key paradigm shift in terms of separating the control plane from the data plane [47]. The separation of architectures is a central innovation that reduces network management, enhances flexibility, scalability, and agility. The forwarding plane, or data plane, performs the actual forwarding and transmitting of data packets based on routing decisions. These decisions are no longer made locally by each device but are managed by the autonomous control plane. The control plane manages global network intelligence and traffic management decisions such as routing, load balancing, and policy enforcement [48].

Forwarding devices in the data plane, such as switches and routers, are simplified and rely on programmable control schemes provided by centralized SDN controllers. The controllers run on shared software platforms and dynamically manage network behavior, enabling rapid response to traffic patterns, policy changes, or network failures [47]. The data plane is connected to the control plane by open standard interface protocols, specifically by OpenFlow and ForCES [48]. Through these protocols, controllers can dynamically modify and alter the actions of the forwarding devices in real time. Furthermore, SDN exposes network functionality to applications using application programming interfaces (APIs), such that network operators can program, manage, and optimize the network [49]. Programmability paves the way for smart automation, policy-based management, and interoperability with advanced technologies such as NFV and edge computing [50].

Based on the control scheme, SDN networks can be categorized into two main classes: SDN with a single centralized SDN controller and SDN with multiple controllers. SDN with a single centralized controller is an SDN network that deploys only a single controller at the core network, and this centralized controller handles all traffic load. SDN controller can handle a certain amount of traffic at a time; however, after a threshold of traffic, the performance of the control scheme decreases [47]. This is mainly a function of controller type; however, all available SDN controllers have an upper limit of network traffic that can be handled. Thus, SDN networks with a single centralized controller can be used for small-scale networks, where the maximum expected traffic does not exceed the upper limit of traffic that can be handled at a time by the SDN controller. This puts constraints on network scalability. To improve the performance of a single centralized SDN controller, a part of the existing literature uses clustering schemes to logically divide the controller into multiple controllers. This achieves the performance of the single-controller SDN networks; however, it remains unsuitable for large-scale networks [51].

The second class of SDN networks is the SDN networks with a multiple controller scheme. These networks deploy multiple physical controllers to improve network performance at high traffic loads. This increases the overall network scalability, while it introduces other challenges. All these benefits provided by the multiple controllers make this scheme efficient for large-scale networks and networks that support dense deployment. However, this increases the overall network cost, besides the design challenges introduced due to multiple SDN controller deployments. The network cost includes both CAPEX and OPEX, while the main design challenge with the multiple controller scheme is the problem of controller placement. Controller placement is the term defined for SDN networks with multiple controllers, which refers to the locations of SDN controllers, allocations of OpenFlow switches with SDN controllers, and the required number of SDN controllers [51,52].

6.1. SDN as an Enabling Technology of the Tactile Internet and Immersive Communications

Next-generation communication paradigms, such as TI and immersive communications, require ultra-low latency, high reliability, and custom network architectures. Such applications require real-time machine–to–machine and human–to–machine interactions, and thus, traditional rigid and static network architectures fall short. SDN has been a key enabler of these requirements due to architectural flexibility, central control, and dynamic programmability. SDN represents an efficient paradigm for the core network of TI and immersive communications systems, due to its architectural flexibility, availability, central control, reliability, and dynamic programmability improvement over traditional core networks, e.g., evolved packet core network (EPC). Recent studies illustrate that SDN with multiple controllers should be deployed for the core network of TI, mainly to achieve the required level of network flexibility and to reduce the round-trip latency to one millisecond [53]. SDN addresses many of the limitations in traditional networks through the separation of control and data planes, enabling centralized control and programmable behavior. SDN offers the following benefits for TI and immersive communications [7,49,53].

• Real-time traffic engineering to meet latency constraints.
• Centralized monitoring and analytics for dynamic QoS adjustments.
• Programmable interfaces to integrate edge intelligence and tactile feedback loops.
• Network slicing for application-specific virtual networks.
• Prioritizing immersive data flows.
• Adapting to network conditions in real time.
• Enabling media-aware routing and dynamic path computation.
• Facilitating MEC integration to reduce latency.

Table 14. Mapping TI components to SDN capabilities [53].

TI Component	Functionality	SDN Support
Tactile devices	Sensing/Actuation	Dynamic flow rules and local routing
Gateway	Edge Aggregation	Latency-aware routing
Network controller	Orchestration	Global view, flow optimization
Tactile server	Processing and Feedback Generation	Load balancing, resource allocation

maps the key components of the TI to different features of the SDN. Furthermore, AI and ML models can be integrated into the SDN controller to enhance decision-making. The models can predict traffic patterns, detect anomalies, and dynamically reconfigure slices. This results in proactive scaling, predictive maintenance, and QoS optimization. In proactive scaling, the controller can anticipate traffic bursts and scale resources. AI/ML approaches can be used with the SDN to identify potential failures and reroute slices in advance using telemetry data [54]. Furthermore, such algorithms can adjust slice parameters to meet real-time dynamic QoS requirements.

6.2. SDN-Driven Network Slicing

Network slicing is a fundamental capability of next-generation telecommunication networks, such as 5G and 6G, through which many virtual networks can be constructed over one shared physical infrastructure. A virtual network, or “slice,” is designed to serve the distinct requirements of an individual service, user, or application. SDN is pivotal in realizing this vision by dynamically orchestrating and managing network resources in slices [55]. Legacy networks are monolithic, with a single one-size-fits-all model that is not effective at addressing the varying requirements of various applications. For instance, autonomous vehicles require ultra-low latency and high reliability, whereas video streaming requires high bandwidth. Network slicing addresses this by partitioning the physical network into logical slices, each of which is optimized for a particular use case [56].

One of the primary features of SDN is network slicing, wherein multiple virtual networks exist on a shared physical infrastructure. Each of them may be customized for different performance requirements, e.g., a slice with ultra-low latency for tactile applications and a wideband slice for immersive media.

Table 15. Network slice characteristics for TI and immersive communications.

Slice Type	Latency	Bandwidth	Priority Level	Applications
Tactile Slice	<1 ms	Moderate	Highest	Remote Surgery, Haptics
Immersive Slice	<10 ms	Very High	High	VR Streaming, 360° Video
Background Slice	>20 ms	Low	Low	Non-critical IoT Data

provides the characteristics of the network slices that can be deployed for TI and immersive communications. SDN introduces programmability and centralized control necessary to properly manage these slices [57]. The SDN controller is an intelligent center brain that hides the physical infrastructure and enables operators to define, deploy, and control slices based on SLAs [58].

SDN can achieve the following benefits for TI and immersive communications network slicing [55,56,57,58].

• The SDN controller dynamically allocates bandwidth, CPU, and storage based on real-time demand and SLAs.
• SDN can allow fine-grained policy control per slice, such as prioritization of emergency services.
• SDN controllers can route traffic efficiently across slices using real-time analytics and optimization algorithms.
• SDN can ensure that each slice operates independently, reducing interference and potential security breaches.

Figure 4 presents the structure of SDN-enabled network slicing that can be deployed for TI. An SDN-enabled slicing architecture typically consists of four main layers: the infrastructure layer, the virtualization layer, the control layer, and the service layer. The infrastructure layer comprises physical network resources such as switches, routers, and links. The virtualization layer abstracts the physical resources and partitions them into virtual components. SDN controller oversees the entire network, orchestrating the slices using programmable logic. The service layer interfaces with applications and defines the slice requirements (e.g., latency, bandwidth). Despite its advantages, several challenges must be addressed to efficiently integrate SDN and network slicing for TI and immersive communication applications. This includes the following [57,59]:

• Managing many concurrent slices requires high-performance controllers and efficient orchestration.
• Each slice must be isolated to prevent data leaks and unauthorized access.
• Diverse vendors and architectures call for unified standards for slice creation, management, and teardown.
• The centralized nature of SDN can introduce control latency that may impact time-sensitive slices.

6.3. Challenges with SDN-Based Core Network for Tactile Internet and Immersive Communications

The integration of SDN with the core network to support the TI and immersive communications services is predicted to transform next-generation networks. The services involve ultra-low latency, high reliability, and real-time interactive qualities, which are very difficult to design and operate. While SDN provides dynamic programmability, central control, and abstracted management of network resources, its application in the TI poses several research challenges.

A.

Controller placement problem

One of the most basic problems in SDN architecture for the TI is the controller placement problem (CPP). The responsiveness of SDN networks heavily relies on the physical placement of controllers [52]. In TI applications, where end-to-end latency must be below one millisecond, placement of controllers becomes even more critical.

Table 16. Impact of CPP on performance metrics of TI and immersive communications [51,52].

Metric	Centralized Controller	Distributed Controller	Hierarchical Controller
Latency	High	Low	Moderate
Reliability	Low	High	High
Complexity	Low	High	Moderate

summarizes the impact of CPP on performance metrics of TI and immersive communications networks. The major features of the CPP include the following [51,52].

• Incorrect placement can lead to latency in control message propagation.
• Demands a conflict between minimizing delay and maximizing fault tolerance.
• Placement should account for hierarchical layouts as the network increases to avoid bottlenecks.

Another main issue is the optimum number of SDN controllers. Too few controllers lead to overload and high latency, while too many lead to heightened synchronization overhead and cost. The optimum number of SDN controllers should be selected to achieve load balancing, failure resilience, and synchronization overhead.

B.

Integration between SDN and MEC-based RAN

MEC plays a key role in meeting the latency and computing requirements; however, integrating MEC-based RANs with SDN poses several challenges, including interoperability, coordination mechanisms, and orchestration. Since different vendors use different standards for MEC and SDN, integrating both technologies is challenging [59]. This includes defining the interfaces and standard APIs for SDN-MEC communication. Furthermore, such integration requires real-time cooperation between edge nodes and centralized SDN controllers and seamless orchestration of edge and core resources for optimal service delivery. Figure 5 provides a layered architecture of SDN-EI integration.

C.

Cost of SDN network (CAPEX and OPEX)

While SDN promises reduced OPEX due to centralized control and automation, the CAPEX for deploying SDN infrastructure, especially in legacy environments, remains high. Cost optimization strategies, such as virtualization to reduce hardware dependencies, incremental deployment in hybrid environments, and adoption of open-source SDN controllers and switches, can be used to reduce the cost of SDN networks [51].

D.

Traffic management to achieve the required QoE

Ensuring the optimum QoE for immersive and tactile applications requires advanced traffic management algorithms beyond the traditional QoS approach. Real-time haptic communications require guaranteed bandwidth, low jitter, and extremely low latency. Furthermore, SDN networks should be able to handle traffic management issues, including dynamic flow control to adapt to changing network conditions at short notice, separating services like video, voice, and haptic feedback, and preventing packet loss under highly dynamic topologies [60].

7. Related Studies of SDN-Based Tactile Internet System

This section introduces the existing works that consider developing the TI, or any associated service, using an SDN-based network structure. Many existing studies provide an SDN-based core network for immersive communications, ultra-low latency 5G, and TI. V. Fanibhare et al. [61] developed a novel framework for the TI based on a multilevel structure of edge computing units and SDN technology. The work mainly considered the traffic flow within the suggested TI framework. SDN network has been used to control the flow over the TI, reducing the round-trip time. The proposed framework can be viewed as two main networks: RAN with MEC and the core network with SDN. The RAN was designed using a multilevel MEC structure with micro-cloud and mini-cloud layers. However, the core network deployed a single centralized SDN controller to manage and control the network traffic. The system is simulated for performance evaluation over a modified CloudSim platform, i.e., iFogSim. The considered evaluation metrics were throughput and latency, and the results indicated that the system reduces the end-to-end latency by 20%.

A. Ateya et al. [62] developed another SDN-based framework for the TI system, which deployed both SDN and MEC technologies for the core network and the RAN, respectively. The considered SDN network deploys a single centralized SDN controller with the OpenFlow interface. All control messages between the SDN controller and OpenFlow switches have been introduced. The system has been evaluated over a modified version of the CloudSim platform, i.e., CloudSimSDN, for three considered simulation scenarios, with different topologies for each scenario. The system achieved over 80% reduction in the round-trip latency over traditional EPC.

J.A. Cabrera et al. [63] investigated the challenges associated with the TI system, including the dynamic change of network topology, the dense deployment of heterogeneous devices, and the end-to-end latency requirements. The work considered network softwarization as a key technology that can overcome such changes and realize such systems. Authors introduced an SDN/NFV paradigm to enable the TI and facilitate the deployment of new services. Moreover, a novel scheme for turning network coding into a software-based service, in a way that increases the throughput of the network and reduces the end-to-end latency. Another main objective of the work was to develop a testbed, i.e., 5G Lab Germany holistic testbed [64], to facilitate the proof-of-concept (PoC) of the TI system. The developed testbed was used to evaluate the performance of the developed TI system experimentally.

P.V. Mekikis et al. [65] investigated using NFV/SDN as a key solution to overcome the TI’s latency requirements. The work considered implementing a novel NFV experimental scheme for achieving ultra-low latency communications. The work mainly considered deploying NFV and SDN for industrial IoT (IIoT) as an example of ultra-low latency applications. The developed NFV/SDN was evaluated for industrial TI applications, and the results achieved a sub-millisecond end-to-end latency, which validates the TI from the latency perspective.

Polachan, K. et al. [66] developed a robotic telesurgery system based on SDN and MEC technologies. The work considered telesurgery a main TI application and introduced a novel framework to meet the announced requirements. Moreover, the developed architecture considers the haptic feedback. The developed SDN deploys a single centralized SDN controller, designed to generate a dedicated link once a patient is detected on the remote side. The multi-modal multicast service gateway confirmed the dedicated link to facilitate the surgery. The system was evaluated for QoS parameters, jitter, throughput, and delay. Results showed that the developed system achieved a 108% improvement in throughput, a 92% reduction in jitter, and a 62% reduction in delay of the communicated data compared with a developed telesurgery system running over a 4G network.

Z. Xiang et al. [67] developed a chain-based low-latency virtual network function (VNF) implementation (CALVIN) and a low-latency management framework for improving the latency efficiency of virtual machines that enable the TI system. The developed framework was implemented over an SDN network. The developed CAVIN groups the virtualized network functions into three groups: elementary, basic, and advanced functions. These functions are implemented in heterogeneous places, since the first two groups are processed in the kernel and the last group is processed in the user space. The system was evaluated over a developed NFV testbed built using traditional computing units, i.e., computers. Results indicated that the developed CALVAN reduces the round-trip latency to an average of 0.32 ms for packets with a size of 1400 bytes.

Muhammad Z. Islam et al. [68] developed an edge computational intelligence-aided haptic-based Tele-caring system (ECI-TeleCaring) to address the disadvantages of traditional remote elderly care, such as high latency and poor service assurance. Based on SDN and MEC, the system supports real-time, location-based activity forecasting using dual long short-term memory (LSTM) models deployed at the edge. This architecture significantly improved QoS and QoE by achieving communication delays of under 2.5 ms for most of the data packets. The system achieved transparency–stability balance in elderly care and offers an operational testbed for real-time observation, delay control in the network, and information exchange between human operators and telerobots.

Mahfuzulhoq Chowdhury et al. [69] proposed a new end-to-end network management and intelligent resource-slicing architecture for zero-touch networks (ZTNs) for 6G as well as non-6G service function chaining (SFC) applications. Unlike existing literature, this work incorporates user intent, time, and cost preferences into application execution over ZTNs using MEC, NFV, and SDN. The solution dynamically assigns virtual and physical resources to optimize task assignment, resulting in significant gains such as a 25.27% reduction in task implementation delay, a 6.15% improvement in energy efficiency, and an 11.52% improvement in the cost-effectiveness of the service compared to baseline solutions.

Hikmat Adhami et al. [70] investigated using SDN and MEC in LTE-Advanced (LTE-A) networks. The study addressed the limitations of current LTE-A latency, which exceeds the one-ms threshold critical for haptic communication. The authors simulated various network scenarios and measured performance against key indicators such as end-to-end delay, jitter, and throughput. The combined SDN-MEC scenario achieved dramatic improvements, reducing haptic delay to 2 ms, jitter to 0.01 ms, and throughput to 1000 packets per second, thereby demonstrating the potential of SDN and MEC convergence for TI applications. Vaibhav Fanibhare et al. [55] developed a network slicing framework, TINetS3, based on SDN and Open vSwitch (OVS) for dynamically allocating and managing slices with different traffic and application loads. The architecture includes topology, service, and emergency slicing customized algorithms to enable efficient use of physical infrastructure. Performance was evaluated with performance metrics, including throughput and round-trip time (RTT), indicating the framework’s ability to meet stringent TI requirements and outperform existing methods.

Table 17. Summary of the recently developed TI systems.

Ref.	Key Enabling Technology			Discussion of Key Features of the Existing Systems
	MEC	SDN	NFV
[33]	√	√	√	The location of MEC servers should be specific. The system has been evaluated compared to 4G systems; however, TI is a main use case of 5G systems, and it requires higher specifications than those of 5G systems, thus, it should be compared with 5G systems. The work considers real-time medical video transmission over the TI structure, which is novel.
[34]	√	✕	✕	The system considers the security issues for MEC-based IoT-5G systems; however, other parameters, e.g., latency, reliability, and availability, are out of scope. The work does not provide an end-to-end structure, which makes the core network blind and is a main part of the evaluation.
[36]	√	✕	✕	The MEC structure system does not consider the expected massive loads, since simulation results consider three MEC servers for distributed 30 base stations, which may be impractical in dense deployment scenarios of 5G systems. Some assumptions are impractical for real implementation.
[32]	√	✕	✕	The system provides a structure for the RAN only; however, the core network is out of scope. The considered metric is only latency; other KPIs are not included.
[37]	√	✕	✕	The location of edge intelligence servers is distributed among ONU-BSs/MPPs, which remains unspecified. The work mainly considers the link between optical and wireless networks, while other parameters and network layers are out of scope.
[42]	√	✕	√	The round-trip latency is large for uRLLC. MANO has not been considered in the evaluation process.
[62]	√	√	✕	The work represents a simple framework for the TI system; however, the end-to-end structure is unclear. The role of the 5G network is not clear in the developed framework. The system did not consider other parameters such as availability and reliability. The evaluation process was very simple and included no haptic data. The performance of the SDN has not been evaluated. Deploying a single centralized SDN controller is improper for large-scale networks; however, the work has not considered this scenario. Only a single simulation scenario with limited OpenFlow switches was considered.
[61]	√	√	✕	The work did not consider the network scale’s effect on the SDN network’s performance. The developed system was not modeled, and did not consider other KPI.
[63]	✕	√	√	The work considered a central aspect of the TI system, i.e., network coding; however, other main aspects, e.g., data offloading and traffic management, are not considered. The testbed is one of the first testbeds developed for experimental measures of the TI systems; however, it is simple and lacks a lot of hardware. Moreover, it was built based on a single SDN controller only. The work did not consider the performance of the SDN network.
[65]	✕	√	√	The work was built over a single centralized SDN controller with limited performance for large-scale networks. The work considered data in the experimental evaluation was simple and did not contain tactile data.
[66]	✕	√	√	The work did not consider the performance of the SDN network with the network scale and traffic. The evaluation process is simple, and the process setup is not clear. The system did not consider other parameters such as availability and reliability.
[69]	✕	√	√	The work mainly considers SDN for performing and improving the network slicing operation; however, other network services have not been considered. The work did not consider the performance of the SDN network. The system did not consider other parameters such as availability and reliability.
[55]	✕	√	√	The work did not discuss the network structure, the role, or the impact of the SDN. The only considered metric is the round-trip time; however, other KPIs have not been considered. The evaluation process is carried out using devices with limited resources.

compares the previously introduced works that consider developing TI systems, and the specifications of each system are introduced.

8. Integrating SDN, NFV, and MEC/EI for TI

Integrating SDN, NFV, and MEC/EI for TI and immersive communications is necessary. Each of these technologies has its advantages, and when orchestrated together harmoniously, they form a collective infrastructure that can deal with the ferocious demands of the Tactile Internet. Underpinning the integration is the fact that the TI is fundamentally a service-oriented model. The network not only needs to transmit data, but also dynamically change based on context, user location, and application requirements in order to return tactile feedback, haptic information, and control commands in a reliable manner within milliseconds/microseconds. Figure 6 provides a potential structure of MEC/EI, SDN, and NFV integration framework for TI and immersive communications applications. MEC/EI assists by bringing computation, intelligence, and decision-making closer to end users, thereby minimizing the physical and network distances haptic information has to travel. However, the effectiveness of MEC/EI depends on a programmable and tunable network infrastructure, where SDN and NFV enter as complementary technologies.

SDN provides centralized control and programmability to manage dynamic routing of tactile flows. Traffic behavior in applications of the TI can suddenly change depending on user activity or the mobility of devices. The SDN controller has a global understanding of network topology, available bandwidth, and congestion, allowing it to proactively divert flows or provide tactile priority packets over regular traffic. For example, in an application of teleoperation with haptic feedback, when the controller of a robotic arm switches to another network cell, an SDN controller may reprogram paths dynamically to ensure low latency and packet loss avoidance. Moreover, SDN’s high-granularity traffic engineering capability can enable the creation of virtualized slices of networks, each of which is optimized for specific tactile applications, with the trade-off requirements of latency, reliability, and throughput being balanced.

On top of SDN, NFV introduces service agility and flexibility with the virtualization of network functions essential to facilitate the TI. Traditional hardware appliances, such as firewalls, intrusion detection systems, or traffic optimizers, introduce unacceptable processing latency for haptic applications. NFV permits such functions to be achieved as VNFs either at the edge or distributed across the core network, with on-demand scaling or migration capabilities. For instance, a predictive caching or a content delivery VNF placed alongside a remote surgery suite can ensure that tactile and visual cues are synchronized and delivered with deterministic latency. Similarly, security VNFs can be selectively instantiated in the network slice dedicated to tactile traffic, enforcing hard policies without affecting other services. NFV also enables SFC, where haptic traffic is dynamically forwarded through an optimized chain of VNFs tailored to its specific needs, such as latency-aware compression, encryption, or haptic signal optimization.

The integration of SDN, NFV, and MEC/EI functions optimally when orchestrated together by an intelligent orchestration framework. Orchestration in this context refers to not just allocating resources but also adjusting the deployment of VNFs and routing policies in real time depending on network conditions as well as app needs. AI and machine learning-based algorithms embedded within MEC/EI nodes can predict traffic bursts, anticipate mobility patterns, and allocate resources ahead of time. For example, predictive models can preplace VNFs along likely user routes or optimize SDN-based routing to minimize end-to-end jitter. This AI-based orchestration guarantees tactile interactions have unbroken continuity even in the presence of dynamic networks or mass multi-user environments.

Some real-world use cases exemplify the imperative of such a cohesive approach. In remote surgery, the TI needs a microsecond-scale response time for haptic feedback to be accompanied by high-definition video streaming. MEC/EI locally processes haptic inputs, minimizing the signal travel distance. SDN ensures that these flows are optimized and prioritized throughout the network, and NFV provides low-latency compression and security capabilities to be placed near the edge. Similarly, for collaborative immersive robotics, there are various robots in a manufacturing facility that must talk in real time while being controlled remotely. SDN provides network programmability to maintain synchronized control channels, NFV gives virtualized control and monitoring with scalable capability, and MEC/EI ensures local intelligence to offset variability in the environment and haptic feedback processing. Without a united framework like this, the unyielding latency and reliability requirements of these applications would be unfeasible.

Resilience and redundancy are the other major features of the integration. The TI would not tolerate single points of failure. SDN facilitates fast failover capabilities, diverting tactile traffic instantaneously in case of a node or link failure. NFV complements it by allowing VNFs to be replicated or migrated on different MEC sites for continuity of service. MEC/EI further complements resilience with localized decision-making that can still function even during a temporary loss of connectivity with the central controller. Together, the multi-layered architecture ensures that tactile communications are not disrupted, deterministic, and invariant under varying network conditions.

Security and privacy are other areas where integration must happen. Tactile applications, particularly in healthcare or industrial automation, handle sensitive data and mission-critical operations. NFV facilitates the placement of security capabilities close to the user, e.g., encryption, intrusion detection, and anomaly mitigation, without incurring excessive latency. Control of SDN in the center ensures that these security rules are applied consistently across the network, while MEC/EI can provide AI-based anomaly detection, which learns local behavioral patterns. In combination, they not only ensure data integrity but also ensure security features do not interfere with the high-performance demands of tactile flows.

Network slicing tailored to tactile applications is also facilitated through the convergence of SDN, NFV, and MEC/EI. Each slice can uniquely combine VNFs, SDN routing policies, and edge intelligence features. For example, an industrial automation slice can support deterministic latency and reliability as a topmost priority, with predictive analytics VNFs at the edge predicting the behavior of machines. An augmented reality slice can have high-performance throughput for visual content with rendering capabilities placed inside MEC servers while still benefiting from SDN for optimal routing. Inherent adaptability of this approach allows multiple tactile applications to achieve support from shared physical infrastructure by service providers without affecting performance or end-user experience. Summing up,

Table 18. Potential roles of MEC/EI, SDN, and NFV in the integrated framework for TI.

Aspect	MEC	SDN	NFV
Core function	Brings computation/storage close to end-users	Decouples control and data planes, enabling programmability	Virtualizes network functions on commodity hardware
Primary benefit	Low latency and localized processing	Centralized control, dynamic routing, flexibility	Service agility, cost reduction, scalability
Deployment model	Edge servers at base stations or access nodes	Centralized controller with programmable devices	VNFs deployed on NFVI with MANO orchestration
Use cases	AR/VR, TI, IoT analytics	Network slicing, traffic engineering, and mobility management	Firewalls, IDS, load balancers, and content caching
Integration	Requires dynamic routing and orchestration	Directs traffic to edge resources	Enables lightweight network services at the edge

provides the main aspects of MEC/EI, SDN, and NFV technologies in the integrated framework for TI applications.

From the research perspective, there are several challenges that still hinder the establishment of completely integrated SDN-NFV-MEC/EI architectures for the TI. First, coordination overhead between the SDN controller, NFV orchestration, and EI mechanisms can cause delays unless properly optimized. Hierarchical or distributed SDN control frameworks and lightweight container-based VNFs are being considered to mitigate these overheads. Second, predictive orchestration requires high-quality AI models able to forecast traffic and haptic behavior in real time, required for intensive training and continuous adaptation. Third, QoS guaranteeing on an end-to-end level across heterogeneous networks like wireless and optical segments remains challenging, particularly during high mobility or network congestion. Finally, standardization and interoperability across multi-vendor environments are required for large-scale deployment, as haptic applications often traverse public and private networks with mixed technologies. Despite these obstacles, experimental and field deployments have demonstrated the feasibility of combined SDN, NFV, and MEC/EI frameworks for touch applications. Testbeds that implement distant robotic surgery, VR collaboration, and haptic gaming have achieved sub-millisecond latency by integrating edge computing, programmable networking, and virtualized functions.

9. Future Directions for MEC, EI, and SDN in TI and Immersive 6G Communications

Realizing TI and 6G immersive communications requires deploying many novel technologies to achieve the previously mentioned requirements. This section provides open research questions and potential future directions on MEC, EI, and SDN to assist TI.

9.1. Future Directions for MEC/EI

Integrating AI, quantum computing, and distributed computing models at the edge paves the way for a new generation of network intelligence and responsiveness. Figure 7 presents a summary of the future directions and open research in applying MEC and EI for TI and immersive communications.

The research directions in EI for TI and immersive 6G communications include the following:

a)

Neuromorphic computing

Neuromorphic chips mimic the brain’s neuronal organization, enabling real-time, parallel processing with extremely low power consumption. Neuromorphic chips are ideal for edge devices that perform quick inference without going to the cloud. For example, neuromorphic processors, e.g., Intel’s Loihi, can spot anomalies in industrial IoT or allow tactile haptic feedback in remote surgeries with millisecond latency [71,72].

b)

AI-driven edge orchestration

With the increasing complexity of edge networks, dynamic resource allocation and service deployment are needed. AI-based orchestration leverages real-time intelligence to maximize workloads, predict congestion, and react to network dynamics [73]. This is critical in use cases like vehicular edge networks, where delay-sensitive applications, e.g., cooperative driving, always need optimization.

c)

Quantum EI

The union of quantum computing and AI, quantum EI, promises exponential decision-making and learning rate boosts. Quantum-enhanced optimization algorithms and variational quantum circuits could accelerate pattern recognition and predictive modeling at the edge [74]. This is especially valuable in highly dynamic environments such as multi-agent robotic systems or swarms of drones.

d)

EI-native applications

EI-native design favors the placement of applications tightly coupled with local data processing. Instead of taking cloud-native applications and deploying them at the edge, application developers are building tools and systems that leverage local context, proximity computation, and distributed processing [75]. Examples include real-time object detection in AR glasses or emotion-detecting avatars in the metaverse.

e)

EI blockchain

EI blockchain brings an aspect of trust and security to edge devices with tamper-proof records and decentralized identity management. Lightweight consensus schemes such as proof-of-authority (PoA) and directed acyclic graphs (DAGs) allow for feasibility on resource-limited machines. Applications range from firmware update security in IoT devices to decentralized federated learning among edge nodes [76].

f)

Self-learning AI agents

Unlike static models, self-learning agents continuously refine themselves based on real-time feedback and are therefore optimally used for non-stationary data distribution-based applications [77]. Such agents can enhance user experience in immersive systems by learning user behavior, environmental states, or network states, thereby providing more natural haptic feedback or real-time content adaptation.

g)

Federated EI

Parallel model training across various edge nodes distributed to offer privacy-preserving real-time analytics [78].

h)

Energy-efficient AI

EI environments tend to be power-constrained. Improved lightweight deep-learning models, e.g., MobileNet, TinyML, and neuromorphic architecture, help reduce computational load and extend battery life [26]. These models are essential in wearables, sensor networks, and remote healthcare monitoring systems that must run for extended periods.

i)

Integration with 6G technologies

Perhaps the most revolutionary innovation is the combination of edge intelligence with 6G enablers like terahertz communication, reconfigurable intelligent surfaces (RIS), and quantum secure links. This enables holographic communication and tactile interaction to be sent with sub-millisecond latencies and ultra-reliable data transfer, which actualizes the vision of the TI.

9.2. Future Directions for SDN

SDN provides the programmability, flexibility, and abstraction necessary for controlling dynamic and complex network slices suitable for various TI applications. Emerging directions in this part include the self-organization of slices, blockchain-enabled slice integrity, and network management through a digital twin. Figure 8 presents the potential integration of such technologies to realize TI and immersive communications applications. Furthermore,

Table 19. Key challenges and potential solutions in SDN-MEC-based TI systems.

Challenge	Description	Potential Solution
Controller delay	Latency from centralized decision-making	Distributed and hierarchical SDN controllers
Scalability	Management complexity with dense IoT nodes	Federated control planes and AI-based load balancing
Security threats	Attack vectors in programmable networks	Blockchain, quantum cryptography, and threat detection AI
Edge orchestration	Inefficient MEC resource coordination	Unified SDN-MEC orchestration and open APIs

provides the key challenges and potential solutions in SDN-MEC-based TI systems. Coordinating SDN, TI, and 6G immersive communications requires integrated optimization across different communications layers. This includes the following directions.

• Adaptive routing schemes based on ML, failure prediction approaches, and self-driving traffic engineering. This includes developing AI-based traffic prediction schemes to accurately forecast service requirements.
• Developing programmable APIs for traffic steering. This includes proposing SDN-based interfaces to dynamically reroute traffic based on policy, latency, and device proximity.
• Introducing the quantum networking paradigm to extremely secure dense haptic interactions with entanglement-based networking connections for ultra-low-latency applications.
• Investigating cross-layer optimization for enriched QoE. This includes developing QoE-aware scheduling methods based on user input and immersive conditions.

10. Conclusions

TI and immersive communications systems are changing the future landscape of next-generation networks by posing unprecedented latency, reliability, and responsiveness demands. This survey has provided a holistic overview of the evolution of TI, its basic technologies, and its key enabling role in establishing fundamental use cases like remote surgery, industrial automation, and XR. With the communication ecosystem marching towards 6G, the need to address the demanding specifications of TI grows more imperative and daunting. This work highlighted the revolutionary capability of SDN and EI as the foundational enablers of TI. SDN increases flexibility and agility in the network, whereas EI provides context-driven decision-making at the edge to avoid latency and optimize resource utilization. This survey unveiled existing gaps, such as scalability, end-to-end orchestration, and insufficient intelligence at the network edge, through an extensive comparative analysis of State-of-the-Art solutions. The work also highlighted open research directions, such as the need for intelligent, autonomous, and secure network topologies with self-adaptation in highly dynamic networks. The integration of SDN, EI, and next-generation technologies provides a promising path towards making the vision of TI a reality. Future research must focus on creating interoperable, scalable, and AI-based infrastructures that not only facilitate real-time haptics and immersive interaction but also anticipate and adjust to the evolving needs of 6G and beyond.