AI-Driven Mobility Management in 5G and 6G Wireless Networks: A Survey

1. Introduction

Mobility management is a critical aspect of modern cellular networks, particularly in 5G and the upcoming 6G networks, as users are increasingly mobile and demand seamless service across diverse environments. In 5G, mobility management enables continuous device connectivity as they move between cells or network regions. It ensures that user equipment (UE) maintains optimal service quality, regardless of their location. Techniques such as HO management, user equipment tracking, and seamless service continuity are key components of mobility management. These systems leverage advanced algorithms to predict user movement patterns and optimize network resources, enabling uninterrupted communication even at high speeds, such as on trains [1,2], and during transitions between network areas. In 6G, mobility management will be further enhanced by AI and machine learning (ML) to offer even more dynamic, context-aware mobility solutions [3].

Load balancing is a vital technique for optimizing the performance and resource utilization of cellular networks, while load balancing is not explicitly addressed as a central theme, its implicit presence within the broader context of resource management and mobility management is undeniable [4]. Efficient handover (HO) management inherently relies on effective load balancing across base stations (BSs) and network resources to prevent congestion and ensure quality of service (QoS). In 5G, load balancing ensures that network traffic is evenly distributed across base stations, cells, or network slices to prevent congestion and enhance overall system performance [5].

Mobility management and load balancing are closely related in cellular networks, particularly in 5G and future 6G systems, as both aim to ensure optimal performance, reliability, and user experience in highly dynamic environments. While they focus on different aspects of network optimization, they often collaborate to address challenges in network traffic, resource allocation, and user mobility [6].

The relationship between mobility management and load balancing becomes evident when considering how user movement impacts network load and resource utilization [7,8]. As users move through different cells, the load on a specific cell or base station can increase or decrease depending on their location and the volume of data they generate. Mobility management systems, therefore, need to work in tandem with load-balancing algorithms to ensure that when a user moves into a new area or cell, the new cell is not overloaded and resources are appropriately allocated to maintain quality of service. Similarly, if a base station is nearing capacity, mobility management can trigger an HO to a less congested cell, thus supporting load balancing by redistributing users across the network.

Several key challenges complicate mobility management in 5G and future 6G networks. The deployment of mm-wave technology, while offering high bandwidth, suffers from high path loss and susceptibility to blockage, leading to frequent and potentially disruptive HO events [9]. Network densification, achieved by deploying a large number of small cells (e.g., picocells and femtocells), increases HO frequency, potentially leading to higher latency and HO failures [10]. Moreover, the increasing number of connected devices and the prevalence of high-mobility scenarios (e.g., high-speed trains) demand efficient and adaptive HO mechanisms. The stringent QoS requirements of 5G and the even more demanding requirements anticipated for 6G only intensify these challenges.

Traditional HO techniques often rely on signal strength measurements to trigger the HO process. However, these methods may not be suitable for the dynamic and complex environments of 5G and 6G networks. To address these limitations, various advanced HO techniques have been proposed, often incorporating ML algorithms to enhance decision-making [11,12,13] and advance deep learning (DL) [14,15,16,17]. These ML and DL algorithms can predict user movement patterns and network conditions, enabling more proactive HOs that take into account not only the user’s location but also current network load. By leveraging this information, the network can balance the load more efficiently, maintaining optimal connections while minimizing congestion and keeping overall network performance high.

This paper presents a comprehensive survey of mobility load management in 5G networks, exploring existing work, its limitations, and recent advancements in mobility management techniques. Additionally, we discuss several approaches that incorporate ML algorithms to enhance decision-making processes and adapt to dynamic network conditions. This article aims to provide a holistic understanding of mobility load management in 5G networks, offering valuable insights for researchers and practitioners seeking more efficient, adaptive network solutions. The main contributions of this article are as follows:

• Summarizing Related Surveys on Mobility Load Management in 5G Networks: We provide a detailed overview of existing surveys and studies on mobility load management, consolidating key insights and identifying gaps in the current literature.
• Comprehensive Discussion on Mobility Load Management: We delve into the principles, frameworks, mechanisms, and strategies of mobility load management in 5G networks, emphasizing its role in optimizing resource utilization and ensuring seamless connectivity during user mobility.
• Exploration of AI/ML Optimization Techniques in Mobility Management: We discuss various AI and ML techniques that have been proposed to enhance mobility management, including predictive analytics, reinforcement learning, and DL, highlighting their applications in improving HO decisions and load balancing.
• Identification of Key Challenges and Future Directions: We highlight the major challenges in implementing mobility management, such as signaling overhead, latency trade-offs, and complexity in ultra-dense networks. Additionally, we outline future research directions, including the development of hybrid metrics, energy-efficient load balancing, and advanced AI-driven optimization frameworks.

The paper is organized into several sections to provide a comprehensive and structured exploration of mobility load management in 5G networks. Section 2 summarizes and compares existing surveys on mobility load management, identifies gaps in the literature, and contextualizes this work within the broader research landscape. Section 3 delves into the principles and mechanisms of mobility load management, discussing its role in optimizing resource utilization and ensuring seamless connectivity during user mobility. Section 4 explores various frameworks that enable proactive mobility management in dynamic, heterogeneous network environments. Section 5 discusses several impactful innovations in mobility load management. Section 6 highlighted the key challenges and future directions in implementing mobility management. Finally, Section 7 concludes the paper by summarizing the key findings and reiterating the significance of AI-driven techniques and technologies in advancing mobility load management for 5G networks and beyond. Figure 1 presents the paper’s structural organization and conceptual framework.

2. Related Work

This section provides a comprehensive review of existing surveys and research efforts focused on mobility load management across various domains and ML techniques. Numerous surveys have explored mobility management in 5G networks, highlighting the importance of load balancing, HO optimization, and resource allocation [18,19,20,21,22,23,24]. These surveys have primarily focused on traditional approaches, such as signal-strength-based HO mechanisms, but often lack in-depth discussion of integrating AI/ML for dynamic decision-making. Other related works have begun to address this gap by incorporating ML techniques, such as reinforcement learning and deep neural networks, to enhance mobility management in ultra-dense and heterogeneous networks [11,12,25,26,27,28,29]. However, challenges such as scalability, real-time adaptability, and energy efficiency remain underexplored. This review contextualizes the current state of research, identifies limitations in existing approaches, and sets the stage for discussing advanced AI/ML-driven solutions in subsequent sections. Table 1 summarizes the related studies in mobility management.

In [9], the authors presented an overview of mobility management in 5G networks, focusing on key factors such as Dual Connectivity (DC), Carrier Aggregation (CA), the rapid growth of mobile connections, network diversity, connected drones, and ultra-dense networks. It also addressed challenges like inefficient optimization processes and the complexity of optimization operations. The study explored innovative HO techniques in 5G networks, highlighted major challenges in UE mobility, and suggested future research directions in mobility management.

In [11], the study introduced the concept of load balancing, outlining its purpose, functionality, and evaluation criteria, and presenting a basic operational model. It offered a comprehensive literature review of techniques for addressing load-balancing challenges in HetNets, including KPIs and the relationship between coverage and mobility robustness optimization. The paper also explores machine learning-driven load-balancing solutions, discusses current implementation challenges, and examines future operational considerations. The study would benefit from a more detailed analysis of the trade-offs involved in implementing machine learning-driven solutions, such as computational overhead, energy consumption, and latency.

In [12], the authors conducted a comprehensive analysis of ML algorithms applied to mobility robustness optimization (MRO) functions. The study meticulously outlined deployment scenarios, methodologies, evaluation criteria, HO control parameters (HCPs), key performance indicators (KPIs), simulation tools, and achieved outcomes. This effort provides a valuable reference for researchers in selecting appropriate optimization algorithms based on their performance results. Furthermore, the authors categorized common ML techniques used to optimize HCPs in MRO functions and introduced high-mobility-aware network topologies to improve system performance.

The study in [18] presented a detailed analysis of mobility management in 5G heterogeneous networks (HetNets), covering critical aspects such as radio resource control, UE initial access and registration, paging procedures, connected mode mobility, and beam management. The authors identify key challenges in 5G mobility management and propose potential solutions to improve system performance. Their findings are intended to inform researchers in developing more efficient mobility management strategies for 5G networks.

The authors in [19] presented a comprehensive analysis of HO optimization challenges and solutions in beyond 5G (B5G) networks. The study began with an overview of HO in legacy systems and transitions to the specific challenges expected in B5G, proposing future research directions. It also reviewed recent techniques and technological solutions for HO optimization management, emphasizing strategies to improve network performance in B5G environments. While the paper provides a thorough overview of HO challenges and solutions, it lacks critical depth in evaluating the feasibility and scalability of the proposed strategies. The paper would benefit from a more detailed discussion of the limitations of current approaches and a comparative analysis of alternative methods. Furthermore, the study could have explored the impact of emerging technologies, such as AI/ML, on HO optimization in B5G networks.

The authors in [25] addressed the challenges of HO management in 6G networks, focusing on issues arising from diverse devices, high mobility, and complex network conditions such as Ultra Dense Networks (UDNs) and mmWave communications. The study traces the evolution of HO from 4G to 6G, explains MRO, and critiques the limitations of traditional MRO approaches in dynamic environments. By integrating AI with MRO techniques, the authors propose a self-optimizing solution to enhance HO decision-making, thereby improving the overall performance of 6G networks.

The work in [30] provided a comprehensive review of Multi-Access Edge Computing (MEC) HO strategies, detailing the MEC reference architecture and various HO algorithms and techniques. The study discussed the challenges and gaps in current MEC HO processes and offered guidance for future research in this area. Additionally, the authors examined the ETSI-MEC reference architecture, focusing on HO techniques that enhance mobility and state relocation. The paper identifies key challenges and research gaps in MEC HO processes, providing valuable insights for future advancements. Similarly, a study in [31] provided a comprehensive overview of HO decision making in MEC systems, with particular emphasis on the challenges and solutions associated with the integration of communication, computation, and mobility management.

In [26], the authors explored intelligent HO optimization strategies using ML models to improve mobility and home office management in 5G networks. It provided an overview of current cellular network states, discusses various ML methods for HO optimization, and examines data availability for evaluations. The paper also identified key challenges and future research directions in this field. The paper could have delved deeper into the limitations of current approaches, such as the lack of standardized datasets, the challenges of real-time implementation, or the trade-offs between model complexity and performance. A more detailed discussion of open research questions, such as the integration of ML with emerging technologies like 6G or edge computing, would have further enriched the paper.

The work in [20] provided a comprehensive review of recent HO decision algorithms in HetNets. It categorizes these algorithms by decision technique and summarizes their input parameters, methodologies, and performance evaluations. The study highlighted technical challenges and opportunities related to HOs in HetNets and dense cellular networks, drawing key findings from recent research. The significance of this survey lies in its thorough overview of HO decision algorithms, which can serve as a foundation for developing more advanced HO optimization approaches. Additionally, while the study highlights technical challenges, it does not propose concrete solutions or innovative directions to address these challenges.

This study [27] provided an overview of 5G mobility management in ultra-dense small-cell networks, with a particular focus on HO management using reinforcement learning (RL) algorithms to improve HO performance. The paper also discussed future challenges and research directions for 5G ultra-dense small cell networks. The focus on RL is timely, given its growing relevance in dynamic and complex network environments. However, the paper lacks a critical evaluation of the limitations and practical challenges of applying RL to HO management.

The survey paper in [21] provided a comprehensive examination of HO and mobility management issues within 5G HetNets. It reviewed the existing literature, investigated key performance indicators (KPIs), and evaluated solutions for HO and mobility-related challenges, while considering applicable standards. The paper also assessed the performance of current models in addressing these issues, accounting for factors such as energy efficiency, reliability, latency, and scalability. Furthermore, it identified significant challenges in existing research models, provided detailed evaluations of their solutions, and offered recommendations for future research. While it identified challenges and provided recommendations for future research, the discussion remains somewhat generic and lacks innovative or forward-looking insights. For example, the paper does not explore the potential of emerging technologies such as AI/ML, edge computing, or 6G networks to address these challenges.

In [22], the paper examined HO management and associated challenges in 5G New Radio (NR) networks, emphasizing advancements from 4G to 5G. It identified key challenges such as QoS, Quality of Experience (QoE), throughput, delay, traffic load, authentication process attacks, and resource allocation. The study highlighted research gaps, particularly the lack of new advancements and the separation of security and HO issues in previous works. It aimed to analyze technical issues, provide an overview of proposed methods and recent advances, and outline future research directions. The focus on key challenges such as QoS, QoE, throughput, and security is timely and relevant, given the increasing complexity of 5G networks. However, the paper could have offered a more critical evaluation of the proposed methods and recent advances. For instance, it does not adequately discuss the limitations and trade-offs of existing solutions, including scalability, computational complexity, and adaptability to ultra-dense or dynamic network environments.

In [23], the paper offered a detailed review of HO management in future mobile Ultra-Dense HetNets, with a specific focus on dual connectivity in 5G technology. It explored various mobility and HO management techniques designed to ensure seamless user connectivity during mobility. Additionally, the study analyzed key challenges, opportunities, and potential solutions to address mobility issues in future mobile networks. The discussion of techniques for seamless connectivity is insightful and relevant to the evolving demands of future networks. However, the paper could have offered a more critical evaluation of the reviewed techniques.

In [24], the study introduced a HO scheme tailored for 5G vehicular networks, combining Media-Independent Handover (MIH) and Fast Proxy Mobile IPv6 (FPMIPv6) standards to support both predictive and reactive HO scenarios. Vehicles monitor their velocity and available networks to prepare for HO. Predictive HO is triggered when user satisfaction drops below a predefined threshold, while reactive HO is activated upon losing connectivity with the serving network, allowing the vehicle to connect to the highest-ranked available network. The implementation leverages an enhanced version of the FPMIPv6 protocol, and the methodology is evaluated through simulations within a 5G vehicular network architecture. This study presented a well-designed HO scheme for 5G vehicular networks that effectively integrates MIH and FPMIPv6 standards to address both predictive and reactive HO scenarios. However, it could have discussed potential integration challenges with emerging technologies, such as edge computing or AI/ML, to further enhance HO management in 5G vehicular networks.

In [28], the survey explored the current state of cellular networks, with a particular focus on mobility and HO management in 5G. It introduces a novel taxonomy for ML applications in HO management, categorizing data sources into visual data and network data. This taxonomy provided a structured and comprehensive overview of how ML is applied to HO management. Additionally, the survey discussed key challenges and outlined future research directions in this field.

The paper [29] examined various studies on HO management for connected drones in mobile communication networks. It offered a targeted review of drone networks, mobility management, and related research. The analysis suggested that implementing intelligent HO schemes leveraging ML/DL and automated robust processes can significantly mitigate HO challenges compared to conventional methods. However, many existing studies lack real-world validation, relying heavily on simulations that may not fully capture the complexities of dynamic drone environments.

3. Background

3.1. Handover Optimization Functions

The main objective of a self-optimization network is to maintain high system performance and execution quality while minimizing the need for manual administrator intervention. This approach not only reduces the workload for mobile network operators but also simplifies network maintenance and management. Self-optimization is essentially a process that dynamically adjusts network parameters based on measurements and the performance of UE and eNodeB (eNB).

During the operational phase, various functions are implemented to optimize network parameters, each designed to achieve specific objectives. Interestingly, some of these functions may adjust the same parameters but for different purposes. Figure 2 illustrates the self-optimization network functions and their relationships with adjustable parameters in Radio Resource Management (RRM). For instance, MRO and Load Balancing Optimization (LBO) both optimize HCPs, but they do so to meet distinct optimization goals.

3.2. Mobility Robustness Optimization

MRO focuses on identifying and resolving HO issues, such as connection failures, unnecessary HOs (UHOs), and HO ping pong (HOPP). These issues can lead to HO failures (HOF) and radio link failures (RLF), particularly during intra-LTE or inter-RAT mobility. To mitigate these problems, MRO adjusts HCPs based on the specific HO challenges encountered.

The proper configuration of HCPs is critical for ensuring smooth UE mobility. Incorrectly set HCPs can result in high rates of HOPP and HOF, leading to inefficient use of network resources and service disruptions. For example, if HCP values are set too high, HOs may be delayed or triggered incorrectly, which is common in high-mobility scenarios. While this reduces HOF rates, it increases the likelihood of HOPP. On the other hand, setting HCP values too low can lead to premature or incorrect HOs, which is common in low-mobility situations. This reduces HOPP but increases HOF rates.

To address these challenges, an adaptive technique is required to dynamically adjust HCPs based on the UE’s mobility status. In this context, the HOF is classified into the following types [33]: Too Early HO, Too Late HO, and Ping-Pong HO, as shown in Figure 3. Each type represents a specific failure scenario that must be carefully managed to optimize network performance and ensure seamless mobility.

3.3. Load Balancing Optimization

The primary goal of load balancing optimization (LBO) within self-optimization functions is to address network congestion by distributing traffic more evenly across cells. This is achieved by adjusting HCPs to facilitate HO actions. An adaptive LBO mechanism can significantly enhance system capacity compared to a non-optimized cell [34].

Additionally, optimizing load balancing reduces the need for manual network management, minimizing human intervention and preventing cell congestion. LBO is supported by three optional subfunctions, controlled by the operation and maintenance (O&M) system: load reporting, adaptive HO, and load-balancing actions based on HOs [35]. Depending on the operator’s strategy, one or more of these subfunctions can be implemented.

Several load-balancing techniques commonly used in networking can be adapted for HO management in 5G and 6G networks:

• Static Load Balancing: This method distributes network traffic based on pre-configured settings, considering factors like base station capacity and geographic location. While simple to implement, it lacks flexibility in dynamic network conditions. In environments with fluctuating traffic, static load balancing may lead to inefficiencies.
• Dynamic Load Balancing: This technique continuously monitors network load and adjusts the distribution of UE accordingly. While more responsive to changing conditions, it requires complex algorithms and can introduce additional overhead. Given the highly dynamic nature of 5G and 6G networks, especially in high-mobility scenarios, ML can play a key role in optimizing dynamic load balancing.
• Adaptive Load Balancing: Combining elements of both static and dynamic approaches, adaptive load balancing follows predefined rules while adjusting to real-time network conditions. This hybrid method offers a balance between simplicity and adaptability, making it a promising solution for HO management in 5G and 6G networks.

Each of these techniques has its strengths and weaknesses, and their suitability depends on the specific requirements and conditions of the network. For 5G and 6G networks, which demand high reliability, low latency, and efficient resource utilization, dynamic, adaptive load-balancing methods are likely to be more effective, especially when enhanced with advanced technologies such as ML.

3.4. Handover Control Parameters and Events

3.4.1. Handover control parameters

In cellular networks such as 5G and 6G, Time To Trigger (TTT) and HO Margin (HOM) are key parameters that play a crucial role in managing HOs, ensuring smooth user mobility while maintaining network efficiency [13,36]. TTT determines how long a mobile device must experience a weak signal before an HO is initiated, preventing unnecessary switches by waiting for a sustained drop in signal quality rather than reacting to temporary fluctuations. A shorter TTT means quicker HOs, while a longer TTT allows more time to see if the signal improves, reducing the risk of "ping-pong" HOs where the device rapidly switches between cells. HOM, on the other hand, sets the minimum difference in signal strength required between the current cell and the target cell for an HO to occur. This ensures that the device switches only to a cell with significantly better connectivity, avoiding premature or suboptimal HOs. Together, TTT and HOM work hand in hand to optimize HO decisions, especially in challenging environments such as high-speed travel or densely populated urban areas, where signal conditions can change rapidly. They also help with load balancing by steering traffic away from congested cells and toward less busy ones, improving overall network performance. In essence, TTT and HOM are essential for maintaining seamless connectivity, minimizing unnecessary HOs, and ensuring efficient resource management in modern cellular networks.

These two parameters control when an HO is triggered and which cell the UE is handed over to, thereby contributing to the network’s overall mobility management strategy.

• TTT ensures that an HO is only triggered after a sustained deterioration in signal quality, which helps to avoid unnecessary HOs.
• HOM, on the other hand, ensures that the HO only occurs when the target cell has a significantly better signal than the current cell, preventing premature or suboptimal HOs.

The relationship between TTT and HOM can be critical in high-speed or dense environments. For example, when a user is moving quickly (as in a vehicle), the signal might fluctuate rapidly, and TTT helps ensure that HOs are not triggered too early due to temporary dips in signal quality. At the same time, HOM ensures that the HO is to a cell that provides better connectivity, ensuring smooth service. In urban environments, where cells overlap and load balancing is important, both TTT and HOM work together to minimize the risk of HOs to cells that are too congested or weak.

3.4.2. Handover Control Events

The LTE mobility framework defines a set of measurement events, labeled A1–A6 and B1–B2, that enable the network to make informed HO decisions based on serving and neighbor cell quality. Events A1–A5 constitute the core intra-RAT mobility triggers that manage measurement activation, reporting, and HO execution within LTE, while Event A6 refines intra-RAT transitions by comparing neighbor and serving performance with tighter offsets for dense deployments. In contrast, Events B1 and B2 extend the mobility mechanism across different radio access technologies, allowing the system to initiate inter-RAT HOs either proactively (B1) or under degraded serving conditions (B2). Together, these events form a structured control framework that ensures seamless mobility, optimized resource utilization, and reliable service continuity in heterogeneous cellular environments. Figure 4 shows categories of mobility control events.

• Core Intra-RAT Mobility Events Various event types (A1-A5) for triggering measurement reports were considered in accordance with 3GPP 36.331 [37]. Figure 5 illustrates possible conditions triggering each of the events A1-A5. Typically, UE conducts radio measurements through two distinct phases: the idle state and the connected state. In the idle state, these measurements serve to support cell selection and re-selection. In connected state mode, the measurement uses HO and redirection mobility scenarios and makes an HOD, which A1 does to A5, depending on the predefined threshold. These events are essential in maintaining optimal connectivity and ensuring a smooth HO among serving gNBs of the 5G system as mobile users move within the network:
  Case 1: When a serving cell becomes better than threshold (A1) and a serving cell’s quality deteriorates below threshold (A2).
  In A1, the network continuously monitors the signal strength and quality of both the serving and target cells. When the serving cell’s signal quality exceeds the threshold, it indicates that the current connection is stable and satisfactory, and no HO is required. However, A2 indicates that the current connection is degrading, possibly due to increased interference or a greater distance from the cell tower. As a result, the network begins evaluating target cells to determine whether a better option exists for maintaining connection quality.
  Case 2: When a target cell becomes better than a serving cell (A3), and a target cell’s quality becomes better than a threshold (A4).
  A3 indicates that a target cell can provide a stronger and more reliable signal to the mobile device. The network then initiates preparations for a potential HO to ensure a seamless transition without compromising connectivity. However, A4 occurs when the mobile device moves further away from the serving cell or experiences increased interference. The network intensifies its assessment of target cells to identify the best alternative for maintaining a stable connection.
  Case 3: Serving becomes worse than threshold 1, and a target becomes better than threshold 2 (A5).
  Event A5 is a combination of Event A2 and Event A4. Event A5 provides an HO triggering mechanism based on the measurement report. It can be used to trigger a time-critical HO when a serving cell becomes weak and it is necessary to change towards the target cell, even if that change does not satisfy the criteria for an event A3 HO. It is important to note that the specific implementation and usage of these HO events can vary depending on the network operator, deployment scenario, and network configuration. The A1 to A5 HO events provide flexibility and adaptability to optimize HODs based on various network conditions and requirements in 5G networks. Table 2 summarizes the parameters and their ranges used in different events [38].
• Advanced Mobility Refinement Events
  Event A6 (Intra-frequency/Inter-frequency neighbour becomes better than serving with an offset) is an intra-RAT mobility reporting event used when a neighbour LTE cell becomes better than the serving cell by a configured offset. It enables fine-grained HO refinement and optimization without degrading the serving cell. It is mainly applied in dense deployments (macro–pico/small cell) to improve HO positioning and reduce ping-pong effects.
• Inter-RAT Mobility Events
  Event B1 (Inter-RAT threshold-based reporting: Neighbour RAT meets threshold) is an inter-RAT reporting event triggered when a neighbour cell belonging to another RAT (e.g., NR, WCDMA, GSM) exceeds a defined threshold. It does not require the serving cell to degrade and enables early offloading or proactive transitions between RATs for capacity, performance, or energy efficiency.
  Event B2 (Inter-RAT reporting: Serving becomes worse and neighbour becomes better) is an inter-RAT dual-condition event in which the serving cell must fall below a threshold while the target RAT cell exceeds its threshold. It is more conservative than B1 and is typically used to ensure robust fallback or HO transitions under poor serving conditions (e.g., coverage edge or degradation scenarios).

3.5. HO Decision Approaches

HO decision approaches play a crucial role in ensuring seamless connectivity in wireless communication networks, particularly in mobile and heterogeneous environments. These approaches determine the optimal time and target network for an HO based on various parameters such as signal strength, speed, interference, energy efficiency, and network policies. Traditional methods, such as RSS-based and speed-based algorithms, offer simplicity but may suffer from frequent HOs or poor decision-making in dynamic conditions. Advanced techniques, including learning-based, fuzzy-logic-based, and multiple-criteria-based approaches, enhance decision accuracy by considering multiple factors and adapting to network conditions. Context-aware and policy-based algorithms further refine HO decisions by incorporating user preferences and predefined rules. The choice of algorithm depends on the network environment, mobility patterns, and QoS requirements, with a trade-off between complexity and efficiency. Table 3 presents several HO decision approaches with their advantages and challenges.

3.6. HO Performance Metrics

In standardized 5G/6G mobility management, performance metrics are formally captured through Key Performance Indicators (KPIs) defined by 3GPP (TS 36.x, 37.x, 38.x series) and extended in O-RAN Alliance specifications for intelligent and open RAN deployments. These metrics enable the evaluation of HO performance in terms of reliability, latency, continuity, and stability.

3.6.1. HO Interruption Time (HIT)

Defined by 3GPP as the time during which the UE is unable to exchange user plane packets with any cell during HO:

$$HIT=t_{resume}-t_{suspend}$$

(1)

The $t_{suspend}$ is the time when user-plane data is suspended due to HO execution (e.g., path switch), whereas $t_{resume}$ is the time when user-plane data transmission resumes on the target cell. 3GPP requirement for URLLC aims for $HIT\le 0.5ms$, when feasible with multi-connectivity.

3.6.2. HOF Rate (HFR)

As defined in 3GPP failure reporting (RLF + HO failure events):

$$HOFRate={\displaystyle \frac{N_{RLF}+N_{HO\_fail}}{N_{HO\_attempt}}}\times 100\%$$

(2)

where $N_{RLF}$ represents the number of RLF events while, and $N_{HO,fail}$ captures the number of HOs that fail due to insufficient radio resources, signaling errors, or execution mismatches. $N_{HO\_attempt}$ denotes the total number of HO attempts initiated either by the network or by the UE [37,38,39].

3.6.3. Radio Link Failure (RLF) Rate

RLF events occur when link quality remains below configured thresholds beyond TTT:

$$RLFRate={\displaystyle \frac{N_{RLF}}{N_{UE}}}\times 100\%$$

(3)

RLFs are quantified through $N_{RLF}$, which counts the events triggered when the downlink link quality remains below critical thresholds for longer than the configured TTTr. RLF is a critical reliability metric tied with MRO.

3.6.4. HO Ping-Pong (HOPP)

3GPP defines ping-pong as successive HO events between cells within a time window:

$$HOPPRate={\displaystyle \frac{N_{PP}}{N_{HO\_success}}}\times 100\%$$

(4)

HO stability is commonly assessed through ping-pong measurements, where $N_{PP}$ denotes the number of rapid back-and-forth HOs occurring within a stability window and $N_{HO,success}$ represents the subset of attempts that result in successful completion. O-RAN acknowledges HOPP rate as a mobility instability metric for xApps/rApps.

3.6.5. Service Continuity

3GPP defines service continuity (SC) as the ability to maintain session-level connectivity without drops:

$$SC=1-{\displaystyle \frac{T_{service\_pause}}{T_{session}}}$$

(5)

SC is characterized by the session duration $T_{session}$ and the cumulative interruption period $T_{service\_pause}$ caused by HO execution. URLLC & VoNR particularly require $SC\approx 1$.

3.6.6. Throughput Degradation During HO

Throughput degradation is defined as:

$$\mathrm{\Delta }R=R_{pre}-R_{post}$$

(6)

Throughput-related metrics rely on the number of successfully delivered payload bits B within a measurement interval T, while $R_{pre}$ and $R_{post}$ refer to the throughput measured immediately before and after HO execution, respectively. This KPI is relevant for eMBB and ISR mobility, where buffering behavior impacts QoE.

3.6.7. Latency

Latency is decomposed into user-plane (UP) and control-plane (CP) components:

$$L=L_{UP}+L_{CP}$$

(7)

Latency is evaluated using the transmission and reception timestamps $T_{tx}$ and $T_{rx}$. Low CP latency is a Rel-17 objective for CHO and AI-assisted prediction.

3.6.8. Packet Loss Rate (PLR)

$$PLR={\displaystyle \frac{N_{lost}}{N_{sent}}}\times 100\%$$

(8)

where $N_{sent}$ and $N_{lost}$ represent the number of transmitted and lost packets during the same interval. Impacts real-time traffic and VoNR QoE.

4. Proactive Framework for Mobility Management

This section explores various frameworks that enable proactive mobility management in dynamic, heterogeneous network environments. By leveraging predictive analytics, ML, and real-time data processing, these frameworks aim to anticipate user movement, optimize HO decisions, and allocate resources efficiently before connectivity issues arise. Table 4 summarizes several mobility management frameworks.

In [40], the authors proposed a proactive HO mechanism for 6G networks to address the limitations of conditional HO (CHO) in balancing HOF and ping-pong effects, as shown in Figure 6. The approach uses AI/ML-based time-series forecasting to predict optimal HO timing and target cells, thereby enhancing mobility robustness and user-perceived throughput. Then it compares UE-side and network-side models for measurement predictions, demonstrating that the UE-side model offers significant improvements. This proactive HO mechanism represents a paradigm shift in mobility management, moving from reactive to predictive strategies.

In [41], the authors presented the Advanced Mobility Management and Utilization Framework (A-MMUF) as shown in Figure 7, which transforms mobility management from a reactive to a proactive approach. A-MMUF leverages Mobility Prediction Models (MPMs) to forecast user mobility attributes and traffic patterns, enabling improved HOs, reduced signaling overhead, and proactive automation for enhanced network performance. The effectiveness of A-MMUF depends on the accuracy of MPMs, which in turn is influenced by the choice of ML techniques and the quality of the training data. Through case studies on proactive HO, mobility load balancing, and energy savings, the results demonstrate significant performance gains and highlight the agility-accuracy tradeoff for optimizing practical deployment.

The authors in [42] proposed a proactive framework that integrates predictive analytics with dynamic routing to improve resource utilization and overall network performance. The system adopts a two-tier design that combines Speed-Optimized Long Short-Term Memory (SP-LSTM) networks for forecasting with Reinforcement Learning (RL) for adaptive routing as shown in Figure 8. The SP-LSTM component predicts potential congestion events, allowing the network to take preventive measures, while the RL module refines routing decisions based on these forecasts to sustain optimal performance. This continuously learning and adaptive design aligns well with the emerging requirements of 6G networks, including ultra-low latency, high reliability, and efficient management of heterogeneous network environments.

Another scheme, the Transmit Power Tuning-based HO Success Rate Improvement Scheme (TORIS), used a data-driven solution to reduce inter-frequency HOFs [43]. As shown in Figure 9, the TORIS integrates an AI-based prediction model that achieves high accuracy using a novel feature set informed by domain knowledge and enhanced with advanced data augmentation methods, such as Chow–Liu Bayesian Networks and Generative Adversarial Networks. These techniques effectively address class imbalance by targeting borderline samples, thereby improving model robustness. Performance comparisons with state-of-the-art AI models demonstrate that TORIS significantly enhances HO success rates.

In [44], AI-based beam-level and cell-level mobility management techniques for high-speed railway (HSR) communications are investigated. A compressed spatial multi-beam measurement scheme, developed using compressive sensing, is proposed to enhance spatial–temporal beam prediction accuracy without increasing measurement overhead. In addition, an AI-based proactive HO mechanism is introduced to predict HO events in advance, thereby reducing radio link failure rates in HSR scenarios.

The study further evaluates two deployment strategies for AI-enabled cell-level HO management, as shown in Figure 10. In the first strategy, the AI prediction model runs on the UE, and the resulting predictions are transmitted to the serving Base Station (BS) to support HO decisions. In the second strategy, the AI model is deployed on the network side, enabling temporal prediction and decision-making-based on historical measurement reports. The primary distinction between these approaches lies in the location of model deployment, while the inputs, outputs, and model architectures remain identical.

Recent work has also explored AI-driven mobility enhancement within O-RAN-based vehicular communication systems as shown in Figure 11. An intelligent O-RAN framework is proposed that uses an ML model to predict how long a vehicle will remain within the communication range of another vehicle, enabling proactive HO decisions. The study evaluates the performance of Gaussian Naive Bayes (GNB), K-Nearest Neighbors (KNN), and Neural Networks (NN) models based on their training and test accuracies [45]. These findings highlight the growing role of predictive intelligence in enabling more reliable and efficient mobility management for next-generation vehicular networks. Compared with traditional signal-based HO triggers, the predictive approach in O-RAN-based demonstrates superior adaptability to rapidly changing vehicular speeds and channel dynamics.

Building on other AI-driven mobility enhancement schemes proposed for vehicular and high-mobility environments, a Smart HO Strategy (SHS) as shown in Figure 12 is introduced to autonomously fine-tune HCPs, including HM and TTT, by evaluating real-time channel conditions using SINR [46]. This approach targets the HO challenges inherent to 5G mmWave wireless channels, which are highly susceptible to interference and rapid fluctuations. The objective is to ensure seamless mobility as UE transitions between BS in a dual-connectivity multi-radio network. The proposed algorithm is evaluated using a 5G mmWave statistical channel model that captures dynamic channel behaviour, including fading and Doppler effects.

Other recent studies have increasingly focused on proactive and prediction-driven mobility management for next-generation vehicular and mobile networks. A virtual-cell–based mobility management framework is presented, in which real-world vehicle mobility data are used to train a trajectory prediction model based on the LSTM-DR architecture, which integrates LSTM networks with the Dead Reckoning (DR) method. This framework operates within a centralized SDN controller and forms virtual cells via an active gNB selection algorithm, supported by a signaling procedure that minimizes overhead [47]. Complementing this direction, another study examines the impact of key HCPs including radio link failure, HOPP, HOF, and HO delay on 5G heterogeneous networks and proposes a Proactive Decision-Making (PDM) approach to improve cell-selection accuracy under diverse mobility conditions [48]. Further advancements include the AEPHORA framework, which leverages AI/ML-driven vehicular mobility prediction to jointly optimize proactive HO and resource allocation decisions, aiming to reduce system transmission power while meeting stringent QoS requirements for delay and reliability in dense V2X environments [49]. Additionally, a dual-connectivity mobility management approach is introduced for real-time service users, incorporating predictive resource reservation to enhance throughput and fairness, alongside a proactive HO mechanism that reduces delay and HO frequency. Computational complexity is also reduced by differentiating cell-center and cell-edge users through adaptive base-station connectivity [50].

Table 4. Summary of Proactive Mobility Management Frameworks

Ref.	Framework / Scheme	Key Techniques & Objectives	Key Contributions / Findings
[40]	Proactive HO Mechanism for 6G	AI/ML-based time-series forecasting. Predict optimal HO timing and target cells; reduce HOF & ping-pong.	UE-side prediction model significantly improves mobility robustness and throughput; shifts from reactive to predictive HO.
[41]	A-MMUF	Mobility Prediction Models (MPMs), ML-based prediction. Enable proactive HO, reduce signaling.	Demonstrates improved HO, load balancing, and energy savings; highlights agility–accuracy tradeoff.
[42]	Predictive Analytics + Dynamic Routing	SP-LSTM for forecasting, Reinforcement Learning for adaptive routing. Improve performance.	Predicts congestion events and refines routing proactively; aligns with 6G needs.
[43]	TORIS	AI-based prediction model, Chow–Liu Bayesian Networks, GAN-based augmentation.	Novel feature-engineering improves prediction accuracy; significantly enhances HO success rate.
[44]	AI-Based Mobility for HSR	Compressive sensing, AI-based HO prediction. Reduce RLF in high-speed railway scenarios.	Two AI model deployment strategies (UE-side vs network-side); improves spatial temporal beam prediction.
[44]	AI-Driven O-RAN for Vehicular	GNB, KNN, NN-based prediction. Predict vehicle communication time; enable proactive HO.	ML-based prediction outperforms traditional signal-based HO triggers; adapts to vehicular mobility.
[46]	SHS (Smart HO Strategy)	SINR-based evaluation, fine-tuning HM & TTT. Address HO challenges in mmWave mobility.	Achieves seamless mobility in dual connectivity; captures fading & Doppler effects.
[47]	Virtual-Cell via LSTM-DR	LSTM + Dead Reckoning, SDN controller. Predict trajectory and reduce signaling.	Forms virtual cells using gNB selection; lowers overhead in vehicular mobility.
[48]	PDM	Prediction-driven HO decision. Improve cell-selection accuracy under varying mobility.	Considers key HCP impacts (RLF, ping-pong, HO delay) to improve mobility KPIs.
[49]	AEPHORA Framework	AI/ML-driven vehicular mobility prediction. Joint proactive HO and resource optimization.	Reduces power consumption while meeting delay and reliability QoS in dense V2X.
[50]	Dual-Connectivity Predictive MM	Predictive resource reservation, adaptive connectivity. Enhance throughput and fairness.	Differentiates cell-center vs. edge users; reduces HO frequency and computational complexity.

Table 4. Summary of Proactive Mobility Management Frameworks

Ref.	Framework / Scheme	Key Techniques & Objectives	Key Contributions / Findings
[40]	Proactive HO Mechanism for 6G	AI/ML-based time-series forecasting. Predict optimal HO timing and target cells; reduce HOF & ping-pong.	UE-side prediction model significantly improves mobility robustness and throughput; shifts from reactive to predictive HO.
[41]	A-MMUF	Mobility Prediction Models (MPMs), ML-based prediction. Enable proactive HO, reduce signaling.	Demonstrates improved HO, load balancing, and energy savings; highlights agility–accuracy tradeoff.
[42]	Predictive Analytics + Dynamic Routing	SP-LSTM for forecasting, Reinforcement Learning for adaptive routing. Improve performance.	Predicts congestion events and refines routing proactively; aligns with 6G needs.
[43]	TORIS	AI-based prediction model, Chow–Liu Bayesian Networks, GAN-based augmentation.	Novel feature-engineering improves prediction accuracy; significantly enhances HO success rate.
[44]	AI-Based Mobility for HSR	Compressive sensing, AI-based HO prediction. Reduce RLF in high-speed railway scenarios.	Two AI model deployment strategies (UE-side vs network-side); improves spatial temporal beam prediction.
[44]	AI-Driven O-RAN for Vehicular	GNB, KNN, NN-based prediction. Predict vehicle communication time; enable proactive HO.	ML-based prediction outperforms traditional signal-based HO triggers; adapts to vehicular mobility.
[46]	SHS (Smart HO Strategy)	SINR-based evaluation, fine-tuning HM & TTT. Address HO challenges in mmWave mobility.	Achieves seamless mobility in dual connectivity; captures fading & Doppler effects.
[47]	Virtual-Cell via LSTM-DR	LSTM + Dead Reckoning, SDN controller. Predict trajectory and reduce signaling.	Forms virtual cells using gNB selection; lowers overhead in vehicular mobility.
[48]	PDM	Prediction-driven HO decision. Improve cell-selection accuracy under varying mobility.	Considers key HCP impacts (RLF, ping-pong, HO delay) to improve mobility KPIs.
[49]	AEPHORA Framework	AI/ML-driven vehicular mobility prediction. Joint proactive HO and resource optimization.	Reduces power consumption while meeting delay and reliability QoS in dense V2X.
[50]	Dual-Connectivity Predictive MM	Predictive resource reservation, adaptive connectivity. Enhance throughput and fairness.	Differentiates cell-center vs. edge users; reduces HO frequency and computational complexity.

5. Impactful Innovations in Mobility Load Management

5.1. ML/DL Handover Management

ML/DL has become central to next-generation HO optimization, enabling predictive, adaptive, and context-aware mobility management across complex terrestrial, aerial, and integrated networks. Recent studies demonstrate that DRL can jointly manage HOs of both terrestrial users and UAV relays through coordinated learning, where a combination of DDPG-based mobility control and DNN-based channel-quality prediction significantly increases system capacity while reducing unnecessary HOs in 6G NTN-UAV hybrid networks [51]. Other work proposes proactive ML-based HO mechanisms for UAV-IoT systems, where Multi-Agent Deep Q-Networks (MADQN) anticipate UAV dropouts by accounting for energy, computation load, and environmental dynamics, reducing failure-related interruptions by nearly 45% and improving task continuity during mobility events [52]. Similarly, connectivity-aware DRL frameworks integrate 3D path planning with predictive RSRP-based HOs, enabling UAVs to make proactive HO decisions aligned with future trajectories, thereby substantially reducing outage probability and HO frequency in cellular-connected UAV networks [53]. In heterogeneous terrestrial–satellite systems, DRL with D3QN-REM-DER architectures has been shown to optimize HO selection under rapidly shifting satellite footprints, improving throughput, reducing delay, and decreasing unnecessary HOs in multi-connectivity scenarios [54]. In dense terrestrial deployments, ML-based HO type prediction and adaptive tuning of HCPs using DQN and supervised learning achieves over 94% prediction accuracy while reducing ping-pong and early/late HO events, improving stability in ultra-dense 5G networks [55].

Complementary to these, ML-based self-optimization methods such as regression-tree-driven HO parameter tuning (e.g., ML-SOHOT) achieve up to 96% improvement in HO performance metrics across diverse mobility patterns [56]. Similarly, the study in [57] employed ML within 5G mobile networks, using a Logistic Regression model to predict HO events and thereby reduce unnecessary HOs. Finally, in O-RAN environments, hierarchical federated learning (HFL) mechanisms explicitly account for UE HOs during distributed training, enabling mobility-aware model updates that reduce training delays and resource consumption while maintaining model convergence under frequent HO conditions [58], and multi-agent DRL approaches such as QMIX jointly optimize UAV trajectories, interference mitigation, and HO frequency, reducing redundant HOs by over 90% while sustaining low latency and high QoS for both aerial and terrestrial users [59].

5.2. Make HOs Proactive and Faster

Although numerous Self-Organizing Network (SON) mobility-robustness features have been introduced in LTE radio networks, HO failures continue to occur in certain environments. These failures frequently contribute to elevated call-drop rates and large numbers of RRC re-establishments. Such issues typically arise when radio conditions fluctuate rapidly during the HO preparation phase or immediately after the HO command is issued. Traditional SON algorithms often lack the responsiveness needed to cope with these highly dynamic and unpredictable radio environments.

A major improvement can be achieved by enabling the UE to continuously monitor radio-link quality and autonomously select the optimal target cell from a predefined list of candidate neighbors. This capability, standardized in 3GPP Release 16 for both NG-RAN (5G SA NR) and E-UTRAN (LTE), is known as CHO. Figure 13 shows the CHO process between severing and candidate cells.

CHO allows the gNB to pre-configure one or more potential target cells with “execute-when…” conditions. Once the UE detects that the preconfigured trigger is satisfied, it executes the HO immediately minimizing signaling delay and significantly reducing HO failures. CHO was introduced in 3GPP Rel-16 and later enhanced for dual-connectivity and PSCell scenarios in TS. 37.340 Rel-19 [60], TS. 38.300 [35] and TS. 38.401 [61].

In CHO operation, the UE receives and stores a prepared RRCReconfiguration message from a candidate target cell instead of executing it immediately as in a conventional HO. This stored command includes one or more conditional triggers derived from radio-link measurements typically RSRP and RSRQ of serving and neighboring cells. The UE then continuously monitors these measurements and autonomously executes the stored HO command once the defined condition(s) are met. By eliminating the need for the UE to send a measurement report and wait for a network response, CHO reduces exposure to signaling failures in rapidly varying channel conditions.

CHO can incorporate multiple conditional triggers to support composite decision criteria, such as simultaneous signal-strength and signal-quality thresholds. By delegating part of the mobility-control logic to the UE, CHO enhances robustness, reduces HO failures, and improves latency performance across both LTE (E-UTRAN) and 5G (NG-RAN) systems.

In [62], the authors proposed an Advanced CHO scheme based on Epsilon-Greedy Q-learning, enabling the UE to dynamically optimize HO parameters using current network conditions and past outcomes. The approach, evaluated under various mobility and signal scenarios, demonstrated significant improvements in HO decision quality. Building on the need to model and quantify such improvements, the study in [63] introduced a mathematical framework that relates the user blocking probability to HO management parameters via Markov models and stochastic geometry, revealing trade-offs between reducing blocking probability and mitigating RLFs.

To address temporal efficiency in HO processes, [64] proposed an in-time CHO mechanism that leverages historical mobility data to predict user dwell time at the serving base station. Using a Multivariate Multi-output Single-step Prediction (MMSP) model with multi-task learning, in-time CHO minimizes unnecessary resource reservations while ensuring timely HO execution. Complementing these optimization efforts, [65] analyzed CHO performance using a Markov model that incorporates factors such as HO offsets, user velocity, channel fading, and dynamic obstacles, quantifying their collective impact on HO latency, packet loss, and failure probability.

5.3. Spread Load with Multi-Connectivity Instead of Hard Moves

In next-generation radio access networks, the concept of multi-connectivity (MC) and, in particular, the Multi-Radio Dual Connectivity (MR-DC) architecture plays a key role in enhancing HO robustness and enabling more efficient load management. According to 3GPP TS 37.340 [60], MR-DC allows a UE to be simultaneously configured with a Master Cell Group (MCG) on a Master Node and a Secondary Cell Group (SCG) on a Secondary Node, which may belong to different RATs (e.g., E-UTRA + NR) and may be connected over non-ideal backhaul as seen in Figure 14. By maintaining two parallel radio links, the UE can perform data duplication, split-bearer operation, or selective traffic steering, significantly reducing interruption time during mobility events.

Recent research consistently demonstrates that MC is a key enabler for robust mobility management in next-generation networks. MC enables UE to maintain simultaneous links with multiple base stations, enabling fuzzy-logic-based multi-criteria HOs for UAVs that jointly consider RSRP trends and cell load to reduce failures and stabilize throughput [66]. Other studies show that MC combined with dual-connectivity architectures (e.g., NR-DC) can significantly enhance reliability and latency performance in high-mobility scenarios by leveraging packet-level redundancy, packet duplication at the PDCP layer, and intelligent path selection based on signal strength and mobility context [67]. Additional work highlights that MC is particularly effective in ultra-dense and mmWave deployments, where link intermittency is severe, enabling smooth transitions, interference-resilient connectivity, and reduced signaling cost compared to single-connectivity HOs [46]. Complementing these findings, hierarchical DRL-based MC frameworks and MC-assisted resource management solutions further show measurable reductions in HO rate, outage probability, and service interruptions by dynamically optimizing active-set selection and traffic distribution across multiple RATs and frequency layers [68,69].

In emerging 6G scenarios, MC is expected to become even more critical due to the coexistence of diverse RATs, including NR, Wi-Fi, NTN/LEO satellite systems, and future sub-THz links. Multi-connectivity facilitates seamless multi-RAT aggregation, provides redundancy for high-frequency links that are vulnerable to blockage, and supports flexible mobility architectures such as conditional HOs, make-before-break transitions, and traffic splitting through the 5G/6G Core. Consequently, MC serves not only to enhance mobility robustness but also as a strategic mechanism for dynamic load balancing, energy-efficient routing, and predictive mobility control across next-generation networks.

5.4. 5G-NTN Integrated Mobility Management

The integration of terrestrial networks (TN) with Non-Terrestrial Networks (NTN) creates a dual-mobility environment where both users and low-earth orbit (LEO) satellites move rapidly, leading to frequent HOs and fluctuating coverage. Traditional HO mechanisms struggle with these dynamics, generating excessive signaling load and instability. Recent solutions introduce signaling-load-aware CHO and predictive HO preparation, which reduce peak signaling events and improve continuity across rapidly changing satellite footprints [70]. Complementary simulation and testbed work further demonstrates that predictive mobility models implemented within O-RAN RIC architectures enhance decision timing and robustness in integrated TN/NTN systems [71].

AI/ML-based prediction frameworks are increasingly important in NTN mobility. Hybrid model-aided learning approaches that fuse LEO satellite geometry with reinforcement learning (e.g., A2C) improve HO stability for high-mobility platforms such as aircraft, outperforming traditional methods by reducing unnecessary HOs and adapting to rapidly changing topological conditions [72]. Multi-factor adaptive HO strategies that incorporate elevation angle, remaining service time, beam behavior, and load conditions further enhance mobility robustness, providing better performance than RSS-only decision criteria, particularly for LEO satellite Internet systems with dynamic beams and narrow visibility windows [73].

Service-specific mobility behavior also influences NTN HO performance. mMTC services show resilience to high HO frequency, while eMBB traffic experiences significant degradation as elevation angles change and HO intervals shorten, underscoring the need for mobility-aware constellation and beam design [74]. To support research and optimization, the LEON simulator offers dynamic end-to-end HO modeling with standardized 5G/6G protocols across large-scale LEO constellations [75]. Broader analyses of TN-NTN coexistence highlight that seamless 6G mobility requires predictive intelligence, efficient beam management, and reduced UE measurement overhead to handle rapidly shifting satellite footprints effectively [76]. Mobility management in NTNs is fundamentally more complex than in terrestrial systems due to fast-moving LEO satellites, dynamic beams, and rapidly changing coverage footprints. Table 5 summarizes the key NTN mobility management aspects, associated challenges, and commonly adopted solutions that enable reliable HO performance in 5G-NTN environments.

5.5. 5G-WIFI Integrated Mobility Management

The convergence of 5G and Wi-Fi has introduced new opportunities for seamless mobility across heterogeneous access networks, especially in indoor and enterprise environments where both technologies coexist. As highlighted in Ericsson’s technical analysis, Wi-Fi remains dominant for indoor high-capacity best-effort traffic, while 5G NR provides predictable reliability, mobility support, and QoS enforcement, making coordinated HO essential for maintaining QoE when users transition between radio domains [77]. The increasing overlap between 5G NR and Wi-Fi 6/6E deployment scenarios creates strong motivation for intelligent mobility management that steers users between RATs based on signal quality, network load, and application requirements. Advanced reliability mechanisms, such as short block codes and concatenated coding techniques, have been shown to improve mobility robustness in 5G NR-U/Wi-Fi coexistence by reducing retransmission overhead and stabilizing link quality during transitions [78].

Recent research demonstrates the limitations of reactive, signal-threshold based vertical HOs between 5G and Wi-Fi. These methods often lead to throughput degradation when devices remain anchored to weak Wi-Fi due to default selection policies, or when sudden channel variations cause HO failures. To address this, AI-assisted and predictive HO solutions have emerged. For example, Yang et al. propose a cooperative MEC-assisted HO framework using deep reinforcement learning and QUIC-based connection migration, enabling seamless switching between 5G and Wi-Fi without TCP session interruption and achieving up to 96% of the optimal throughput across variable environments [79]. Similarly, the Predictive CHO framework integrates LSTM-based signal-quality forecasting into CHO logic, enabling proactive, RAT-aware mobility control in multi-RAT networks that include both 5G and Wi-Fi, thereby significantly reducing ping-pong events and failures under dynamic interference conditions [80]. Complementary testbed work integrating 5G NR, Wi-Fi, and LiFi confirms that reliable heterogeneous mobility relies heavily on core-assisted intelligence, RAT-aware selection, and tight control-plane coordination—principles expected to extend into 6G multi-connectivity systems [81].

End-to-end multi-RAT integration also plays a crucial role in enabling seamless 5G/Wi-Fi mobility. MultiNet6G demonstrates a unified 5G core integrating NR and multiple Wi-Fi systems through ATSSS and N3IWF, enabling soft vertical HOs with microsecond-level timing precision, illustrating the potential of a “network-of-networks” architecture for industrial mobility scenarios [82]. Complementary studies confirm that successful inter-technology mobility requires accurate radio telemetry during decision-making and robust coordination of 3GPP and non-3GPP interfaces to avoid service interruption during vertical HO operations [83].

Recent research highlights that static or threshold-based HO mechanisms are insufficient for heterogeneous 5G and Wi-Fi environments. Dynamic, intelligent HO strategies that leverage user mobility patterns, predicted signal evolution, and QoS requirements consistently outperform legacy methods. The newly uploaded study on heterogeneous 5G networks reinforces this by showing that static HO parameters fail to reflect real-time fluctuations across different cell types: macro cells, small cells, and Wi-Fi hotspots, resulting in dropped sessions, packet loss and latency. Their dynamic HO optimization, based on real-time analytics, mobility profiling, and multi-metric evaluation, achieves substantial improvements in success rate, latency, and QoE, underscoring the need for adaptive, context-aware vertical HO mechanisms in mixed 5G/Wi-Fi environments [84].

6. Challenges and Future Directions

AI has become a cornerstone in optimizing mobility management across various domains. Despite its potential, numerous challenges persist, and significant advancements are required to fully unlock AI’s capabilities in these areas. Below is an analysis of the challenges and future directions for AI-driven mobility management. By addressing these challenges and focusing on innovative solutions, AI can revolutionize mobility management, providing seamless, efficient, and reliable connectivity across diverse networks and applications.

6.1. Challenges

6.1.1. Privacy and Security Risks

AI-based mobility management systems handle vast amounts of sensitive data, making them vulnerable to privacy breaches and cybersecurity threats [85,86]. Emerging wireless systems face challenges such as cyber-attacks, including spoofing, jamming, and denial-of-service, can exploit vulnerabilities in HO authentication processes, leading to service disruptions, unauthorized access, or compromised data integrity [87,88,89]. Weak encryption during HOs or inconsistent security protocols across heterogeneous networks further increase the likelihood of breaches. Additionally, adversarial manipulation of AI algorithms used in HO decision-making can lead to incorrect authentication, posing a risk to seamless mobility. Mitigating these threats requires robust encryption methods, secure and lightweight authentication protocols [90,91], AI-driven anomaly detection, and standardized security frameworks to ensure secure and efficient HO authentication in diverse mobility environments. Also, blockchain technology enhances security and privacy by offering decentralized authentication, tamper-proof data exchange, and seamless interaction across heterogeneous networks [92].

6.1.2. Energy and Resource Constraints

significantly impact mobility and HO management across UAVs, ITS, NTNs, and cellular networks, as these systems require real-time decision-making while operating with limited power, computational capacity, and bandwidth [93,94]. UAVs face reduced flight times due to the energy demands of frequent HOs, while ITS must optimize vehicle-to-RSU interactions to conserve bandwidth and power. NTNs face constraints due to limited satellite power and processing capabilities, making efficient HO protocols critical. Similarly, cellular networks, especially in dense 5G environments, face increased energy demands for managing low-latency, high-frequency HOs. Solutions such as lightweight authentication protocols [95,96], edge computing for offloading processes, AI-driven predictive models, and resource-adaptive protocols are essential to balance performance with energy efficiency, ensuring seamless connectivity across these systems.

6.1.3. Data Accessibility and Standardization

ML implementations for mobility and HO optimization depend critically on the availability of datasets that are both sufficient and high-quality. Sufficiency means having a dataset large enough for effective ML training, while quality ensures the data is clean, free from missing entries, duplicates, or noise, to enable accurate learning [28]. However, obtaining user mobility history data is often challenging due to stringent data protection regulations, leading to a scarcity of real-world datasets [97]. To compensate, synthetic data generated through network simulations is commonly used, but these datasets may not accurately reflect real-world scenarios. Moreover, the lack of uniformity across datasets means that data generated on one platform often cannot be used on others. This emphasizes the need for standardized, high-quality datasets that serve as benchmarks to assess the accuracy and reliability of ML models for mobility predictions and HO optimization, ensuring their validity and cross-platform applicability.

6.1.4. Deployment Models

AI/ML-based mobility management faces the challenge of choosing between centralized and distributed deployment models. In centralized systems, dec-ision-making occurs at a central point, enabling efficient resource management and global optimization but often leading to higher latency, single points of failure, and scalability issues. Conversely, distributed deployment distributes processing across edge devices or local nodes, reducing latency and improving resilience but introducing complexities in coordination, data consistency, and resource allocation. Distributed ML approaches also preserve user privacy and lower UE energy consumption [98]. However, challenges such as coordinating decentralized learning and transmitting locally trained models across imperfect channels require further research to optimize decentralized ML implementations for mobility management and HO optimization [99]. Balancing these approaches requires hybrid solutions that leverage the strengths of both, such as edge computing for low-latency tasks and centralized servers for global insights, ensuring efficient and scalable mobility management.

6.1.5. Latency and Processing Delays

Latency and processing delays are significant challenges in mobility management, especially for real-time applications like HOs and autonomous decision-making [100,101]. Systems such as NTNs, ITS, UAVs, and 5G networks require ultra-low latency to ensure seamless connectivity and maintain service quality [102,103,104]. However, factors such as physical signal distances, high mobility, and complex network demands exacerbate delays, thereby impacting performance and reliability. Solutions such as edge computing, predictive AI models, decentralized architectures, ultra-low latency protocols, and optimized resource allocation are being explored to minimize delays and enable real-time responses [105,106]. These advancements are critical to achieving efficient and reliable mobility management in next-generation networks.

6.2. Future Directions

6.2.1. Data-Driven Mobility Management

Data-driven mobility management represents a transformative future direction, leveraging the power of big data and AI/ML to optimize network performance, enhance user experiences, and support seamless mobility in complex environments [107]. By collecting and analyzing vast amounts of real-time data from diverse sources such as user devices, network infrastructure, and environmental sensors, data-driven approaches enable predictive decision-making for HOs, routing, and resource allocation. This not only improves network efficiency but also reduces latency, minimizes energy consumption, and ensures scalability in highly dynamic networks.

The integration of advanced analytics, such as DL and reinforcement learning, allows mobility management systems to adapt to evolving user behaviors and network conditions dynamically [108].

Furthermore, data-driven models can facilitate personalized mobility solutions, optimize traffic flow in ITS, enhance satellite resource allocation in NTNs, and improve autonomous UAV operations. However, this approach requires addressing challenges such as data privacy, secure data sharing, and ensuring interoperability across heterogeneous networks. Future advancements in federated learning, edge computing, and blockchain technology hold promise for overcoming these hurdles, making data-driven mobility management a cornerstone for next-generation connectivity.

6.2.2. Integration with Emerging Technologies

Integrating emerging technologies such as blockchain and digital twins can significantly enhance AI-based mobility and resource management systems by addressing challenges such as real-time decision-making, data security, and interoperability [109]. The 5G/6G network infrastructure provides ultra-low latency, high bandwidth, and dynamic resource allocation through network slicing, enabling seamless HOs and optimized connectivity for diverse applications.

With the rapid advancement of AI, sensing and communication networks have increasingly incorporated AI technologies, becoming a key component of next-generation mobile communication systems. AI-driven integration of sensing and communication in 6G enables a paradigm shift in mobility management, offering unprecedented adaptability and efficiency [110]. Additionally, AI can leverage data from integrated sensing to identify obstacles, interference, or signal degradation, enabling the network to adapt resource allocation and adjust communication parameters.

By combining AI with 6G’s sensing-communication integration, HO management can move beyond traditional reactive methods to proactive, context-aware strategies. This not only improves user experience and network reliability but also enhances energy efficiency and reduces signaling overhead, making it a cornerstone of next-generation mobility management.

6.2.3. Digital Twins

Digital Twin is an emerging technology surrounded by many promises and potential to reshape the future of industries and society [111]. A digital twin creates a virtual replica of a physical system, such as a network, vehicle, or infrastructure, enabling real-time monitoring, predictive analytics, and decision-making. In mobility management, digital twins can revolutionize mobility optimization, resource allocation, and traffic flow management across diverse environments like UAVs, ITS, NTNs, and cellular networks [112,113,114].

With digital twins, mobility management systems can predict network behavior and preemptively optimize decisions. For instance, by modeling user mobility patterns, network conditions, and traffic loads, digital twins can identify potential HO bottlenecks and suggest proactive solutions to avoid disruptions [115]. Additionally, a digital twin (DT) optimizes performance and reliability in network resource management by simulating resource allocation strategies and analyzing their impact in real time [116]. This allows networks to dynamically adapt to changing conditions, efficiently balance loads, and ensure reliable service delivery even in complex and high-mobility environments.

However, challenges such as high computational demands, data synchronization, and standardization must be addressed to fully leverage digital twins in mobility management. Future advancements in edge computing, AI, and high-speed networks like 5G/6G will play a critical role in overcoming these limitations, making digital twins an essential tool for achieving efficient, adaptive, and resilient mobility management in next-generation networks.

7. Conclusions

This survey presented a comprehensive study of mobility load management in 5G networks, emphasizing its critical role in ensuring seamless connectivity, optimal resource utilization, and robust HO performance in increasingly dynamic and heterogeneous environments. The interplay between mobility management and load balancing has been illustrated by how user mobility directly influences spatial traffic distribution, interference conditions, and QoS provisioning. Through an extensive review of the literature, several limitations of traditional signal-driven HO strategies have been highlighted, particularly in massive small-cell deployments and high-mobility scenarios, where rapid topology changes and fluctuating traffic loads introduce significant performance challenges. Additionally, several mobility management frameworks have been discussed, highlighting various strategies that can improve seamless communication. Furthermore, innovations and emerging technologies, along with ML/DL-based multi-connectivity solutions, enhance mobility management efficiency. Multi-connectivity, particularly when 5G is combined with technologies and AI-driven frameworks such as O-RAN, NTN, UAV, LEO satellites, and Wi-Fi, has shown substantial improvements in HO accuracy, failure reduction, load distribution, and overall QoE. Despite these advances, several challenges remain, including signaling overhead, energy efficiency, scalability, and the need for reliable real-time data to support model training and inference.

Looking forward, future mobility load management solutions for 6G will increasingly rely on hybrid metric design, proactive and learning-enabled control, tighter integration across terrestrial, non-terrestrial, and Wi-Fi systems, and advanced multi-agent intelligence capable of coordinating mobility across diverse RATs and network layers. By consolidating existing research, identifying unresolved problems, and outlining promising directions, this survey aims to support the development of resilient, intelligent, and scalable mobility management frameworks that can meet the performance demands of next-generation wireless networks. Additionally, integrating emerging technologies such as blockchain and digital twins offers significant potential to enhance AI-based mobility and resource management systems by improving real-time decision-making, strengthening data security, and enabling seamless interoperability across heterogeneous network components.