Interference Mitigation Strategies in Beyond 5G Wireless Systems: A Review

Abstract

Over the past few years, wireless communication has grown dramatically, and the consumer demand for wireless services has seen a significant jump. One of the main challenges for beyond fifth generation (B5G) networks is the increased capacity of the network. The continuously increasing number of network users and the limited radio spectrum in wireless technologies have led to severe congestion in communication channels. This issue leads to traffic congestion at base stations and introduces interference in the network, thereby degrading system capability and quality of service. Interference reduction has thus become a major design challenge in wireless communication systems. This review paper comprehensively explores interference management (IM) strategies in B5G networks. We critically analyze and summarize existing research on interference issues related to device-to-device communication, heterogeneous networks, inter-cell interference, and artificial intelligence (AI)-based frameworks. The paper reviews a wide range of methodologies, highlights the strengths and limitations of state-of-the-art approaches, and discusses standardized techniques such as power control, resource allocation, spectrum separation and mode selection, carrier aggregation, load balancing and cell range expansion, enhanced inter-cell interference coordination, coordinated scheduling and beamforming, coordinated multipoint, and AI-based interference prediction methods. A structured taxonomy and comparative summary are introduced to help categorize these techniques. Several related works based on their methodologies, shortcomings, and future directions have been critically reviewed. In addition, the paper identifies open research challenges and outlines key trends that are shaping future B5G IM systems. A comparative visualization is also provided to highlight dominant and underexplored optimization objectives across IM domains. This review serves as a valuable reference for researchers aiming to understand and evaluate current and emerging solutions for interference mitigation in B5G wireless systems.

1. Introduction

Mobile data traffic has increased several times in the last two decades due to the widespread use of data-hungry devices (smartphones, tablets, etc.) for content-rich multimedia applications, and the same trend is expected to continue [1,2,3]. In 2025, the number of mobile-connected devices is around $(1.5\times {10}^9)$, including a significant number of internet of things (IoT) devices [4,5]. Under such conditions, the beyond fifth generation (B5G) mobile system has gained significant interest in terms of providing improved connectivity and higher mobility for more diversified devices [6,7]. Figure 1 illustrates the global connections and the share of total cellular connections, drawn by the authors based on data from [8].

The following three broad views are highlighted and addressed by the B5G of cellular networks: (i) user-centric (by offering fast recovery from network failures), (ii) service provider-centric (providing mission-crucial monitoring service, sensors, roadside assistance service, and intelligent linked transport networks), and (iii) network-operator-centric (offering secure, uniformly-monitored, low-cost, scalable, and energy-efficient communication infrastructure) [9,10,11]. B5G networks are also perceived to understand the key features as follows:

• High-speed connectivity: Using high-speed data communication (gigabits per second for machines and users) can lead to achieving zero latency property [12]. Data speeds for third generation (3G), fourth generation (4G), fifth generation (5G), and B5G are shown in

  Table 1. Data speeds for 3G, 4G, and B5G.

  Network Type	Theoretical Download Speeds	Average Download Speeds	Peak Download Speeds
  B5G	50–100+ Gb/s	~1–5 Gb/s	10–20 Gb/s
  5G	10–50 Gb/s	130–240 Mb/s	599 Mb/s
  4G	300 Mb/s	32.5 Mb/s	90 Mb/s
  3G	42 Mb/s	8 Mb/s	~20 Mb/s

  [13,14].
• The 5G network represents the current mobile communication standard and serves as a key transitional phase toward B5G. It introduced substantial enhancements such as ultra-reliable low-latency communication (URLLC) [15], enhanced mobile broadband (eMBB) [16], and massive machine-type communication (mMTC) [17]. However, 5G still faces limitations in areas such as high-density user scenarios, coordination across heterogeneous networks (HetNets), and support for real-time (RT) AI-driven adaptation [18,19]. These challenges have driven the development of B5G architectures, which aim to address the shortcomings of 5G by integrating advanced features such as AI-based orchestration, support for non-terrestrial networks (NTNs), and greater scalability in ultra-dense (UD) environments [20,21].
• Ubiquitous connectivity: Smart living individuals need constant and ubiquitous mobile network connectivity to upload their activity data and IoT control commands to the network, creating a huge uplink (UL) in data flow reporting. B5G will provide robots with omnipresent networking options in the future [22].
• Crowded area services: By using multiple base stations (BSs) and the available backbone bandwidth (BW), the existing 4G networks offer crowded area services. Therefore, by implementing the small cell (SC) model and mobile edge computing, B5G can deliver these services reliably without any interruption [23].
• Remote area services: B5G technology aims to enhance services in remote areas. These services include video surveillance, smart city, e-health like telemedicine, and remote meters for collecting reading and billing purposes [24,25].
• Low power consumption: Improvements in B5G new radio have dramatically reduced energy consumption on the network side, particularly the sparser transmission of always-on signals [26,27].
• Intelligent handover: To support consumer mobility, the handover phase is a central element of cellular networks. Consequently, in the sense of cellular networks, handover management has always been a focal research point. With a minimum delay during the switching of the network, an intelligent handover is required [28].
• High capacity: The ability to work in both bands, i.e., lower bands (e.g., sub-6 GHz) and millimeter wave (mm-Wave) (e.g., 24 GHz and up), offers higher throughput and low latency capacity. B5G is expected to increase traffic capacity and network quality by 100× [29].

In the 35 GHz ultra-high frequency band, technology can transmit data at a speed greater than 1 Gbps, and the longest transmission distance can be around 2 km. Compared to 4G, B5G has thousands of times higher mobile data and 10 to 100 times the number of connected devices. Moreover, it has more battery power, and lower latency [30,31].

By employing emerging technologies such as SCs, mm-Wave, and massive multiple-input multiple-output (m-MIMO), it is expected that B5G cellular networks accommodate network load (NL) increment. Capacity gain can also be accomplished by the use of device-to-device (D2D) communication with spatial and frequency reuse (FR) [32,33,34].

There are many challenges in implementing B5G networks including security and privacy management, channel access control management, load balancing (LB), quality of service (QoS) management, handoff management, and interference management (IM) [35,36,37].

Among these challenges, IM is particularly critical due to the high user density, FR, and heterogeneous architecture of B5G networks. Interference degrades signal quality, reduces network capacity, and negatively impacts QoS. To maintain reliable and efficient communication, it is essential to understand the sources of interference and apply mitigation strategies tailored to different B5G components.

Although the B5G network architecture includes emerging components such as satellite communication and unmanned aerial vehicle (UAV) systems, this review primarily focuses on IM within traditional terrestrial B5G structures, including D2D, HetNets, inter-cell interference (ICI), and artificial intelligence (AI)-based frameworks. which are widely deployed and standardized by third-generation partnership project (3GPP).

Information theory seeks to describe the ability of any wireless network, i.e., the maximum possible collection of rates for effective communication [38]. The research community has turned its focus towards interference-limited networks, starting from the capability characterization of noise-limited networks. Interference is known to be the final contact barrier that has to be resolved [39]. Figure 2, shows a detailed view of possible sources of interference in a B5G cellular network, including satellites, aircraft, and ground-based elements that connect through different communication links [1].

In the B5G network, there are two kinds of interference; ICI that takes place between different cells, such as micro-cells and SCs. Here, mitigation of the interference requires using some techniques like coordinated multipoint (CoMP), power control (POC), and scheduling to ensure communications link performance. Another type is called intra-cell interference which takes place between devices within the same cell. Minimizing this type of interference is done using orthogonal resource allocation (RA) for data transmission [40,41,42].

In multi-tier and heterogeneous B5G cellular networks, the management of radio resources and interference would be a key research issue. In this context, conventional IM and radio resource strategies such as LB, POC, and channel allocation may not be successful and a new look at the problem of IM will be needed. Therefore, this paper presents a structured literature review on IM issues and mitigation strategies in B5G networks. The review is organized around three key terrestrial components D2D communication, HetNets, and ICI, along with a dedicated discussion of emerging AI-based frameworks approaches. These areas represent foundational challenges in B5G deployment and are closely aligned with 3GPP standardization efforts.

We conducted this paper as a literature review to provide a structured and critical analysis of existing research on IM in B5G networks. By synthesizing recent contributions and comparing established techniques, the review offers a comprehensive perspective that helps identify effective practices and open research questions, especially those relevant to real-world deployments and standardization. In addition, a comparative visualization is included to quantitatively illustrate how optimization objectives, such as SE, power consumption, and outage probability, are prioritized across different IM domains.

In this paper, we distinguish between IM and interference mitigation (IMI). IM refers to the broader set of techniques and strategies used to monitor, coordinate, and control interference in wireless networks, including detection, classification, and prevention. In contrast, IMI focuses specifically on techniques that reduce or suppress the impact of existing interference on communication quality, such as through POC, RA, beamforming (BF), and filtering. Thus, mitigation is a subset of the broader IM framework.

The organization of the paper is as follows: Section 2 highlights the research motivations and contributions of this paper. Section 3 presents a classification and in-depth analysis of IM strategies in B5G networks, covering key categories and the current research addressing each type. In Section 4, the proposed IMI schemes and techniques by the 3GPP are presented. Section 5 provides a critical analysis of the reviewed strategies and outlines open research challenges for future B5G networks. Finally, the conclusions are communicated in Section 6.

2. Motivations and Contributions

The increasing demand for wireless services and the limited availability of radio spectrum in B5G networks have led to severe congestion and interference, significantly degrading system performance and QoS. Among the many challenges facing B5G deployment, effective IM remains one of the most critical, particularly in the contexts of D2D communication, HetNets, ICI, and AI-based frameworks. This paper aims to provide a comprehensive and structured review of IM strategies that can enhance the performance, scalability, and reliability of B5G networks.

While 5G networks introduced key advancements such as URLLC, eMBB, and mMTC, they still face challenges in UD deployments, RT AI integration, and coordination across heterogeneous infrastructures. These limitations highlight the need for B5G systems that offer greater scalability, intelligence, and support for emerging NTN paradigms.

The main contributions of this review are summarized as follows:

• This paper presents a structured review of IM challenges in B5G networks, focusing on D2D communication, HetNets, ICI, and AI-based framework approaches, and introduces a unified taxonomy and comparative analysis of techniques.
• It critically evaluates existing and emerging IMI techniques by examining their methodologies, advantages, limitations, and applicability as reported in recent literature, supported by summary tables and visual illustrations.
• The paper examines 3GPP-standardized IM schemes by reviewing practical techniques proposed for interference control (ICO) in next-generation wireless systems.
• It identifies open research challenges and future directions, including spectrum allocation, POC, and LB, as well as federated learning (FL), cross-domain IM, and explainable AI, to support the development of more adaptive and intelligent IM strategies for B5G networks.
• A quantitative comparative analysis is introduced, highlighting the focus of existing IM literature on key optimization objectives (e.g., spectral efficiency (SE), power consumption) and revealing underexplored areas such as outage minimization and energy efficiency (EE), especially in AI-based frameworks.

3. Classification and Evolution of Interference Management Strategies in B5G Networks

This section presents a comprehensive overview of IM strategies in B5G wireless systems. It begins with a historical perspective on the evolution of IM techniques from earlier mobile generations, highlighting the transition from static and centralized methods used in 3G/4G to more adaptive, intelligent, and decentralized approaches in B5G. Following this background, the section provides a structured classification and detailed analysis of the key domains where interference significantly impacts system performance. These domains include D2D communication, HetNets, ICI, and AI-based frameworks. Each category is examined in terms of current research trends, mitigation strategies, and practical deployment challenges.

3.1. Background and Evolution of Interference Management Techniques

The evolution of IM strategies in cellular networks has been closely tied to the generational advancements from 3G to B5G. In early mobile systems such as 3G, IMI primarily relied on static methods, including fixed frequency planning, orthogonal channel assignment, and rudimentary POC [43,44]. These techniques were sufficient when user density was low and network topology relatively uniform.

With the introduction of 4G and long-term evolution systems, more dynamic IM methods emerged. These included fractional FR, inter-cell interference coordination (ICIC), and semi-static scheduling techniques [45,46,47], which SE and partially addressed the growing complexity of network traffic. However, such strategies still depend heavily on predefined rules and centralized control, making them less responsive to RT variations in user behavior and NL [48].

In the 5G era, and more critically in B5G networks, the limitations of these traditional approaches have become evident. The rise of ultra-dense networks (UDNs), mMTC, and HetNet architectures has introduced highly dynamic and interference-prone environments [49,50]. Traditional rule-based systems lack the agility and scalability to operate effectively under such conditions.

To overcome these challenges, AI-enabled and self-organizing IM strategies have gained traction in B5G research and standardization. Techniques such as reinforcement learning (RL), deep learning (DL), and FL allow for RT adaptation, proactive interference prediction (IPR), and decentralized decision-making [51,52,53]. These approaches enable the network to learn from historical data, dynamically allocate resources, and manage interference with greater precision across diverse deployment scenarios.

Figure 3 provides a high-level visual summary of this evolution, illustrating the shift from static, rule-based techniques in 4G to adaptive, AI-driven methods in B5G.

3.2. Classification and Analysis of Interference Management Strategies in B5G Networks

Interference has a significant impact on the availability and reliability of wireless communication systems. To effectively manage the complex interference environments in B5G networks, various strategies have evolved over successive mobile generations. This subsection builds on that overview by classifying and analyzing the core IMI strategies considered in this review.

The classification of IM issues into four main categories: D2D communication, HetNets, ICI, and AI-based frameworks is based on practical deployment challenges and current standardization efforts in B5G networks. These categories are selected because they represent the primary sources of interference affecting system performance and scalability, as recognized in ongoing 3GPP studies [54,55,56,57]. D2D communication introduces interference due to direct transmissions sharing licensed spectrum with cellular links [58,59]. HetNets, composed of multi-tier architectures (e.g., macro-cell, micro-cell, and SC), experience both co-tier and cross-tier interference due to dense and uncoordinated deployments [60,61]. ICI occurs in all densely packed cellular environments and disproportionately impacts cell-edge users under universal FR [62,63]. Finally, AI-based frameworks approaches are increasingly vital for enabling RT, adaptive IMI in complex and dynamic network conditions [64,65]. This classification reflects both industry priorities and recent research trends, enabling a structured and comprehensive evaluation of interference challenges and mitigation strategies in B5G systems.

3.2.1. Device to Device

In cellular networks, D2D communication is a new concept. Instead of traveling the signal through the core network, user equipment (UEs) is allowed to communicate using a direct link [66]. In a sparse area, D2D attempts to use the physical range of the transmitting devices to increase the signal for mobile devices. D2D communication can work together with cellular networks to complement each other. In terms of energy and spectrum, resource sharing between D2D and cellular communication is a crucial factor to be considered when designing D2D [67,68,69].

Content privacy and strong anonymity are the main advantages obtained from D2D. These are given because it is not the responsibility of the central storage to store the information exchanged. D2D communications are also capable of improving delay, throughput, fairness, and EE [70,71]. In addition, the performance of D2D is dramatically enhanced by improving the spectrum reuse (SR) and system throughput (ST) due to direct traffic routing [72,73]. Offloading the cellular traffic is achieved by switching the path from infrastructures to direct transmissions [74,75]. Figure 4 illustrates a typical D2D communication scenario within a dense SC network in B5G. It highlights the coexistence of D2D and cellular links, which introduces cross-tier interference challenges that must be addressed through efficient POC and RA schemes [76].

An essential issue that needs to be solved is the interference that underlies D2D contact with the cellular world. To deal with this issue, the mutual interference between D2D links and cellular links in HetNets of B5G technology needs to be managed in such a way that the UDNs must take place in the future network. UDN is a strategy in the B5G communication system. It consists of SCs which are capable of improving the coverage of the network and the capacity of the system [77]. In these SCs, the FR creates severe interference between SC links, macro cell links, and D2D links which is a critical topic when using the D2D and UDN [78].

While D2D communication offers substantial gains in SE and offloading, its real-world implementation faces challenges such as SR conflicts with licensed cellular bands and difficulty in managing interference in uncoordinated dense deployments. Moreover, POC in D2D is constrained by hardware limitations on UE power transmission (PT) and the need to comply with regional spectrum regulations.

An effective IMI system is required to enhance the network’s overall efficiency. In the past few years, researchers proposed various IM schemes that aim to minimize interference in D2D communication systems.

Alibraheemi et al. [75] proposed a Q-learning-based RA strategy for D2D communication networks, enabling efficient mode selection (MOS) under varying interference conditions. Their simulation results demonstrated improved SE and reduced interference levels, showcasing the potential of AI-driven resource adaptation in dense cellular environments. In [76], the authors analyzed D2D communication systems with energy harvesting (EH) over κ-μ shadowed fading conditions. Their work introduced a power distribution mechanism that enhances EE while controlling interference, particularly in environments with fluctuating signal conditions. Nugraha et al. [79] investigated the potency of two POC methods to minimize D2D interference. The first of them is focused on the level of fixed power or uses no POC to manage the transmitter’s power level. In the second method, adaptive POCs for two distinct systems are used. They concluded that both methods of POC led to network efficiency enhancement. In [80], the authors exploited the interference user level under multicasting circumstances. They stated that the proposed method achieves better performance of signal interference to noise ratio (SINR) and lower occupation of resources than conventional pairwise D2D transmission schemes. RA and IM problems of D2D in underlay communications are investigated in [81]. The study showed that by using the proposed method of determining adaptive sub-BW, the total number of cellular UE in outages has decreased. Xu et al. [82] presented a study on IM in heterogeneous cellular networks for D2D communications and concluded that the proposed IM solutions would effectively reduce interference among cellular and D2D connections. Liu et al. [83] proposed an efficient interference alignment (IA) algorithm for D2D communications based on channel state feedback concentration interference alignment (FCIA). The authors suggested that the FCIA would reduce the overhead of feedback and coordination of the transmitter. Ning et al. [84] designed a Stackelberg game-based ICO scheme with full FR in the context of D2D underlying mm-Wave SC network. They reported that the proposed system converges rapidly, holds SINR in a high range, and achieves outstanding performance throughput. Saha et al. [85] studied the issue of allocating resources for D2D pairs from cellular users in such a way that the overall interference is minimized according to a minimum target sum rate. They found that the proposed algorithm requires polynomial time when the interferences are uniform. D2D MOS criteria for each UE based on the maximum received signal strength (RSS) for controlling the D2D to cellular interference are presented in [86]. It was found that the MOS technique can minimize significant interference from the D2D transmitter to the cellular network. Yang et al. [87] considered vehicular communication nodes scenario in which the same resources of the spectrum are shared, causing interference with other network nodes. The results showed that the proposed scheme increases QoS satisfaction with high EE compared to conventional link selection and power adaptation IM schemes. In [88], the authors proposed a novel MOS algorithm for D2D in HetNet for B5G technology. It was concluded that when participating device UE are moving away from each other, the outage likelihood of D2D links is sharply increased.

As B5G networks expand into UD and multi-layered environments, traditional D2D IMI strategies are increasingly challenged by both terrestrial and non-terrestrial deployments. In future scenarios involving UAVs, high-altitude platforms, or satellite-assisted D2D links, interference patterns are affected by factors such as 3D spatial variability, Doppler shifts, and dynamic topologies [37,89]. Additionally, in UDNs, D2D transmissions face cross-layer interference from macro-cell users, neighboring D2D pairs, and SC backhaul links [90]. To manage this complexity, emerging research is turning to AI-driven frameworks, including RL and FL, which offer adaptive POC, MOS, and spectrum coordination based on RT and historical data [53,91,92]. However, challenges remain in ensuring low-latency inference, fairness across user types, and model robustness in distributed, heterogeneous environments [93,94]. Addressing these issues is crucial for realizing scalable, reliable, and interference-aware D2D communication in the evolving B5G ecosystem.

Table 2. Summary of related works of D2D in literature.

Issue	Methodologies	Advantages	Limitations/Future Works	Ref.
D2D MOS under dynamic interference.	Q-learning-based adaptive RA.	Improved SE and ICO in dense networks.	Limited scalability tested. Multi-agent extensions will be considered.	[75]
Power efficiency under harsh fading.	EH-aware RA over κ-μ shadowed fading channel.	Energy-efficient IMI with fading awareness.	Needs expansion to multi-user D2D systems with mobility.	[76]
Explore interference-related issues from D2D communications that impact the foundational cellular networks during downlink (DOL) and UL transmissions.	POC strategies such as fixed POC and adaptive POCs.	Enhancing the SINR and the performance of the proposed network.	The proposed POC strategies are capable of managing 100 pairs of D2D only to ensure a 0 dB level of SINR.	[79]
Investigate the user-level inter-channel interference in multicast schemes.	Groupwise scheme based on network coding.	Improving SINR performance and lower resource utilization compared to conventional pairwise D2D transmission methods.	Cell-level inter-channel interference was not considered.	[80]
To mitigate the dead-zone issue in HetNets.	UL/DOL decoupling user association (UA). UL fractional FR schematic. An innovative concatenated bipartite matching approach.	Minimizing the number of outage users.	The D2D-cell members are unaware of intra-cell and ICI in offline semi-distributed schemes.	[81]
To enhance the transmission density while ensuring the QoS for links in various communications.	Optimal deployment of SCs schemes.	Minimizing the effect of interference while satisfying the minimum QoS requirement. Maximizing the transmission capacity while satisfying the minimum QoS requirement.	The power allocation (PA) strategies for both macro and small BSs were not considered.	[82]
To minimize the overhead of the transmitter coordination by proposing an efficient IA.	Channel state feedback concentration.	Improving the time slot discovery by 6.2% compared with conventional communication. Increasing the sum rate by 50% compared with conventional communication.	The non-line-of-sight was not considered in this scenario.	[83]
Optimize the transmission power of D2D links, mitigate the interference generated by D2D communication to the mm-Wave SC, and fully utilize the BW of the mm-Wave spectrum.	IM strategy based on a Stackelberg game utilizing complete FR within the framework of D2D communication in a millimeter-wave SC network.	Maximizing the total throughput of the proposed system by keeping the SINR as high as possible.	The UL transmission was not considered in this study.	[84]
Minimize the effect of interference by allocating the resources to maintain the target data rate.	The two-phase combinatorial technique is proposed to determine a RA, taking into account the constraint of the total data rate.	Maximizing the total data rate of the proposed system.	The non-uniform interference will be considered. More efficient algorithms for identifying specific triples within decrementing structures will be provided.	[85]
To control the interference between D2D users and cellular users.	Maximum RSS-based MOS strategy.	Improving the area’s SE of the proposed scenario.	The DOL scenario was not considered.	[86]
Maximize the data rate transmission and minimize the PT to mitigate the effect of interference with other links while preserving the minimum QoS required.	A novel IM technique based on graph theory that concurrently addresses power adaptation, link selection, and MOS to maximize the number of communication links while minimizing power consumption.	Maximizing the EE and effectiveness of the proposed scheme.	The UL scenario was not taken into consideration.	[87]
To mitigate the effect of interference that affects the proposed scenario.	A novel MOS strategy for D2D-enabled HetNets.	Improving network reliability compared to traditional approaches.	The throughput and reliability of D2D link will be considered.	[88]

summarizes the related works of D2D discussed in this section.

3.2.2. Heterogeneous Networks

B5G and future wireless networks are expected to offer services with high quality, very low energy consumption, much-improved security, ultralow latency, ultrahigh data-rate, and massive connectivity [95,96]. Compared with the legacy systems, B5G and future communication systems will be not only more advanced but also more complex [60]. The integration of HetNets is considered one of the main solutions for achieving the objectives of B5G networking systems [97,98].

HetNet is a network densification technique to accommodate ever-increasing traffic in a limited geographical region where a large number of transmission nodes are installed [99]. The use of HetNets makes it possible to reuse the spectrum in space [100], which, on the other hand, results in higher interference scenarios [101]. As shown in Figure 5, the B5G heterogeneous cellular architecture involves overlapping layers of macro, micro, and SCs, creating complex interference patterns. Understanding this layout is critical for developing effective IMI techniques across tiers [1].

Minimizing the cell sizes in HetNet requires a wide variety of spectrum solutions and higher SE at each BS to satisfy future wireless network requirements for improved coverage, lower latency, and higher data rates [102]. In addition to the advanced modulation and coding [103], m-MIMO antenna techniques [104], and additional spectrum, raising the number of cells represents the most critical parameter to meet the increasing traffic demand [105]. Homogeneous cellular systems, however, face limitations when increasing the number of cells, such as increased BS cost and site acquisition problems in urban areas. HetNets can be seen as a more versatile and cost-effective approach to such constraints [106]. Furthermore, HetNets are considered an efficient approach for improving coverage and system efficiency for future cellular networks [28,107,108].

To achieve the quality of experience (QoE) and QoS, low cost, and high EE, the B5G HetNets must integrate multiple components including wireless fidelity and cellular radio access network (RAN) through the utilization of different radio access technologies over different carrier frequencies [109]. It is also attractive for operators to deploy HetNets, as it will allow them to provide expanded services and also offer new market opportunities. However, the densification of HetNets also faces the issue of usable BW scarcity, considering their potential advantages. To ensure the success of this essential technology, reliable and scalable solutions that allow operators to use the available spectrum are mandatory [90].

In the research and design of future cellular networks, the challenge of HetNet interference is considered one of the most critical topics. ICI analysis has always been a main research problem in homogeneous systems as well. Interference from neighboring cells in the conventional cellular grid has often caused performance degradation for users of the cell edge who have suffered from both slower reception of the signal and increased interference [110]. There are many reasons for the resurgence of HetNets interference. First, due to the number of cells in each tier does not allowing effective resource partitioning schemes, it is almost necessary to think about universal FR deployments. Furthermore, in the past, interference affected only the cell-edge users of the macro-cell, while the deployment of a HetNet and the number of BSs makes any user in the network suffer from or cause interference. Hence, in the HetNet case, the system-level implications of ICI are more relevant. Last but not least, the unplanned and even random nature of a HetNet deployment makes the problem of managing interference more challenging than the case of well-thought homogeneous deployments [111].

Owing to the uncoordinated nature of HetNet, IM is one of the most important challenges [112].

Despite the performance benefits of HetNets, their deployment is often limited by backhaul constraints, especially for SCs in rural or infrastructurally weak areas. Coordinating inter-tier interference is also hindered by the lack of centralized control and variable spectrum licensing conditions across regions. Hardware heterogeneity among macro cell and SC introduces further complexity in unified interference coordination.

In regions with limited infrastructure, such as rural or developing areas, HetNet deployment faces challenges in synchronization and coordination [113]. Power- and spectrum-efficient solutions that operate under limited backhaul conditions are vital for viable IM in these contexts [114].

Proper treatment of HetNet interference involves two main tasks: understanding the phenomenon of interference (i.e., modeling and study of interference) and techniques for minimizing it as effectively as possible. Several studies have been carried out in this field. Alzubaidi et al. [108] proposed a joint optimization technique combining 3D positioning of non-terrestrial base station (NTBS) and reconfigurable intelligent surface (RIS) reflection control in non-orthogonal multiple access (NOMA) HetNets. Their IMI approach significantly improved DOL coverage and SINR in multi-tier deployments. A deep RL approach for joint RA in 5G HetNets is investigated in [109]. Their method used a multi-agent learning framework to enhance EE while mitigating inter-tier interference, making it particularly suitable for UD B5G networks. Two IM schemes, blind IA and a hybrid topological IM orthogonal multiple access scheme are presented in [115]. It was found that both schemes reach at least twice the rate of time division multiple access. Imam and El-Mahdy [116] proposed a cross-tier UL interference synchronization algorithm for two-tier networks to reduce the interference signals of the macro-cell consumer at the receivers of femto-cells. The results showed that the proposed algorithm surpasses other IA algorithms as it greatly improves the potential of the signal to interference ratio (SIR) and the capacity of the system. A new algorithm that selects the most suitable IoT terminal pair and provides eNodeB with perfect signals to supply relay-assisted communication for IoT terminals is presented in [117]. It was found that the suggested algorithm increases the overall network throughput and the amount of simultaneously served information technologies. Abbas et al. [118] investigated UL coverage in the existence of ICI and jammer interference in multi-tier HetNets. The authors found that due to improved ICI and jammer interference reduction, the reverse frequency allocation (RFA) results in higher UL coverage compared to the no-RFA scenario. Fang et al. [119] presented a study on the security analysis of the suggested IM mechanism with beam forming, spectral reuse (SRE), and D2D communications for the DOL in a two-tier HetNet. The IM mechanism was found to enhance the user’s secrecy rate with the use of some SRE interference D2D communications interference, and beam forming technique. A technique of hybrid IM depending on dynamic enhanced inter-cell interference coordination (eICIC) and CoMP transition is proposed in [120]. It was found that the suggested scheme can attain the best SE both at macro-cell and hypothetical cell edges.

In [121], the authors proposed a novel joint BS UE organization and POC systems for HetNets. They found that the proposed system showed the advantage of innovative algorithms in both traffic off-loading and IM. Two energy-efficient power management and RA algorithms which are the EE fairness algorithm, and the global EE optimization algorithm are suggested by Meng and Liu [122]. The results showed that the algorithms are efficient in reducing interference but to a different degree. Shamaei et al. [123] developed a RA scheme depending on matching theory to minimize reciprocal interferences between D2D and cellular communications. They reported that the proposed scheme is capable of realizing attractive network achievement with much less cost and complexity. WU et al. [124] applied two universal filtered multi-carrier (UFMC)-based interference elimination schemes for the SC HetNet, where both the ICI and the intra-cell interference occur. The authors concluded that the suggested UFMC-based schemes can effectively reduce the interference and mitigate the impact of frequency offsets. An interference reduction system was investigated by Zhang et al. [125]. This study concluded that the suggested system attains a large enhancement in SE as compared with conventional systems.

The summary of previous studies of HetNets is presented in

Table 3. Summary of previous studies of HetNets in literature.

Issue	Methodologies	Advantages	Limitations/Future Works	Ref.
Interference in NOMA-based HetNets.	Joint optimization of NTBS position and RIS.	Improved DL SINR and extended coverage area.	Evaluation under user mobility (UM) and high-density traffic needed.	[108]
Joint RA and POC in dense HetNets.	Deep RL-based multi-agent framework.	Higher EE, interference-aware learning in dynamic settings.	High training cost and lack of cross-tier synchronization considered.	[109]
Reducing the impact of interference signals from macro-cell receivers on femto-cell users.	Cross-tier UL IA scheme. Zero-forcing (ZF) technique.	Maximizing the capacity of the system and SIR compared to other IA schemes.	The average selection IA applying ZF scheme at the femto-BSs will be applied. The cross-tier DOL IA scheme will be applied. The complexity of the proposed scheme will be considered.	[116]
To maximize the user data-rate and network resources.	Communication strategy based on relay selection scheme.	Increasing the network throughput and the number of concurrently served IoT devices by 44% and 20%, respectively.	An algorithm that uses a multi-hop relay will be developed to enhance connection availability and accommodate a greater number of IoT terminals concurrently.	[117]
Mitigate the ICI and jammer interference.	RFA scheme.	Improving SIR of the proposed scenario.	A sophisticated wideband jammer will be evaluated with variable PT to reduce the probability of coverage.	[118]
To improve the security rate of the cellular device in the presence of an espionage attack.	A m-MIMO technique is applied to BSs. BF technique.	Improving the data rate secrecy of the proposed system.	The UL scenario was not considered.	[119]
Controlling the interference in DOL HetNets.	Dynamic eICIC and CoMP-based hybrid IM technique.	Improving the SE and throughput of the proposed scenario.	A novel technique, like selecting coordinated BSs, will be considered. The UL scenario was not considered.	[120]
To maximize the network total rate while preserving standards for individual UEs or to enhance the minimum throughput of UEs.	Precise penalty techniques are integrated with iterative convex optimization.	Improving the traffic offloading and IM.	The UL transmission scenario was not considered.	[121]
To minimize the cross-tier interference of different BS cells to primary users and mitigate same-tier interference among BS cells	Fractional programming strategy and dual decomposition technique.	Maximizing the EE of the proposed scenario.	This work will be extended to unlimited cooperation among flying cells.	[122]
To minimize the effect of co-tier and cross-tier interference between D2D users and cellular users and optimize the performance of HetNet.	Matching theory-based distributed RA scheme.	Maximizing the SE of the proposed system by 93% compared to conventional schemes.	PA strategy will be considered.	[123]
To minimize the effect of inter-cluster and intra-cluster interference in SC HetNets.	Advanced waveform UFMC.	Mitigating the effect of frequency offsets. Minimizing the bit error rate (BER).	The DOL transmission scenario was not considered.	[124]
To mitigate the effect of co-tier and cross-tier interference in HetNet.	PA strategy. Joint co-tier and cross-tier IM technique.	Maximizing the SE compared to the traditional schemes.	The UL transmission scenario was not considered.	[125]

.

3.2.3. Inter-Cell Interference

ICI takes place at the level of the network system as a result of interaction between adjacent cells. Figure 6 demonstrates the ICI of cellular networks. This interaction is either because of operating on the same channel known as co-channel interference or because of the overlap between neighboring channels known as adjacent channel interference [126].

Moreover, the estimation can be used in the DOL precoding and UL data reception for the same cohesion interval. Pilot reuse causes serious interference issues which are mainly because of the high number of user terminals in the same interference area and the limited length of the coherence interval which results in the degradation in the sum rate of the system [127]. The aforementioned interference is considered a type of ICI in which users sharing the same pilot sequence are exposed to interference during the UL training session [128].

ICI minimizes ST and network capability and has a detrimental effect on the performance of cell-edge UE. Environmental attention is also a major issue, as carbon dioxide emissions continue to increase due to the dense spread of BSs [129]. Careful management of ICI is important to enhance B5G network performance. Over the past few years, various approaches have been used to mitigate its effects. Alam et al. [128] proposed a distance-based cell range extension technique combined with almost blank subframe (ABS) allocation. Their study showed this hybrid strategy successfully mitigated ICI and improved LB in 5G heterogeneous deployments. A DL model for narrowband physical random-access channel (NPRACH) detection in the presence of ICI is developed in [129]. Their method improved detection reliability in 5G UL scenarios, especially in dense environments with overlapping coverage areas. Gueguen et al. [130] proposed a novel inter-cellular media access control scheduling algorithm to promote the QoE in 4G and B5G wireless systems. The authors indicated that the proposed scheduler outperforms current solutions in different scenarios. In [131], the authors applied maximum rank planning (MRP) as a new inter-cell IM mechanism for the very dense uncoordinated spread of SCs targeted by B5G networks. They stated that the MRP outperforms FR planning when interference rejection combining receivers are used, realizing high achievement in terms of outage throughput gains in low and high-traffic load situations. Lei et al. [132] investigated a sparse code multiple access (SCMA) based on an UL ICI abolition mechanism to improve the achievement of the cell-edge users by decoding the preferred signal and the interference signal at the BSs. They indicated that the SCMA-based scheme is a promising ICI abolition technique for B5G wireless systems. A functional process of inter-cell rank coordination (RAC) given the dominant ratio of interference is introduced by Mahmood et al. [133]. The authors found that the suggested technique is computationally effective and needs minimal overhead control. Karimi et al. [134] proposed a distributed IM algorithm. The proposed ICI sub-space formation algorithm is less complex and has fewer issues coming from IA. They reported that the suggested method enhances the throughput in networks with large prevalent interference systems. Mehmood et al. [135] discussed a potential solution to reduce interference and maximize user throughput through a structured approach between the separate cells in the NOMA scope. The authors indicated that NOMA plays an important role, a crucial role in providing large ST and assuring fairness of the layout. In [136], the authors applied RFA to mitigate the impacts of ICI and intentional jammers-interference (IJs-I). They demonstrated that the proposed technique results in considerable improvement in UL coverage performance by reducing ICI and IJs-I. Zhang et al. [137] used wireless virtual embedding to virtualize wireless resources in high-dense networks and dynamically allocate resources to minimize ICI in the network. The authors found that the SINR of the total network is fewest and guarantees the quality of each user. Ayoob et al. [138] proposed a hybrid system for radio resource management (REM) to enhance the jitter and delay performance. It was found that a fully flexible time division duplex (TDD) is an applicable option for indoor SCs within the extent of the envisioned B5G technologies.

These studies demonstrate the potential of advanced ICI mitigation techniques, such as CoMP and coordinated BF, to improve cell-edge performance and throughput. However, the feasibility of implementing these solutions in real-world deployments is often limited by backhaul capacity, synchronization demands, and infrastructure availability especially in rural or legacy networks. Furthermore, in m-MIMO systems, challenges such as pilot contamination remain unresolved, particularly under dynamic UM scenarios.

In advanced B5G networks, ICI becomes more pronounced due to the coexistence of macro-cell and dense SC, as well as dynamic spectrum sharing and mobility patterns. The 3GPP has proposed several mechanisms for ICI mitigation, including eICIC and CoMP transmission and reception [139,140]. While eICIC primarily manages time-domain resource partitioning, CoMP focuses on spatial-domain coordination using shared channel state information (CSI) and joint processing (JP). These techniques have shown effectiveness in IM at cell edges, especially under FR-1 scenarios. In addition, the use of UAVs and the adoption of dynamic TDD introduce asymmetric UL/DOL interference that must be handled through adaptive scheduling and POC [141,142]. Emerging research also explores hybrid ICI mitigation schemes that combine BF, POC, and user clustering to reduce control overhead while maintaining SE [143,144].

The summary of previous studies of ICI is shown in

Table 4. Summary of previous studies of ICI in literature.

Issue	Methodologies	Advantages	Limitations/Future Works	Ref.
LB with ICI.	CRE with ABS.	Reduced ICI and improved load distribution across cells.	Integration with AI-based scheduling in HetNets will be considered.	[128]
Interference-robust RACH detection.	DL model for NPRACH in multi-cell UL.	Better access success under interference-heavy NB-IoT settings.	Needs validation for non-Gaussian noise and energy-constrained IoT deployments.	[129]
To mitigate the ICI and improve the satisfaction of users.	Mean cell packet delay outage ratio-based-resource allocation scheme.	Enhancing the QoE of the proposed system.	The optimum value of minimum BW will be considered. The optimal tradeoff regarding the BW allocation to the donor cell will be considered for enhancing the beneficiary cell’s assistance without compromising the donor cell’s QoE.	[130]
Mitigate the dominant inter-cell-interference.	Novel IM technique based on MRP.	Maximizing the ST.	The dynamic configuration techniques that maximize the rank used by each user in the network will be considered.	[131]
To enhance the performance of the edge user in UL HetNets.	UL ICI elimination scheme based on SCMA.	Minimizing the BER of the proposed scenario.	The DOL transmission was not considered.	[132]
To optimize the balance between enhancing spatial gain and augmenting interference robustness.	An inter-cell RAC technique.	Maximizing the ST by 65% compared with conventional schemes.	The UL transmission scenario was not considered.	[133]
To jointly project the transmitted signal over the desired area and align most of the interference onto predefined interference areas at the interfered receiver end.	ICI subspace coordination algorithm for MIMO communication.	Maximizing the ST by 28% compared with traditional algorithm.	An ultra-reliable, low-latency service scenario will be considered.	[134]
To minimize the effect of ICI and maximize end user throughput.	Adaptive fractional FR-based proportional fairness.	Maximizing the SINR and throughput performance of the end users while ensuring fairness preservation.	CoMP will be used for edge cell users. Error mitigation in NOMA successive interference cancelation will be used to minimize the intra-cluster interference.	[135]
To mitigate the effect of Intentional jammers and ICI in UL HetNets.	A strategy for mitigating proactive interference based on RFA.	Improving the QoS of the proposed scenario.	UA decoupling with RFA will be considered to improve the UL coverage. The effect of jamming on certain parts of the spectrum will be considered.	[136]
To mitigate the effect of ICI in UDNs.	A virtual embedding scheme.	Enhancing the QoE of the user’s network.	The PA strategy was not considered.	[137]
To maximize ST by eliminating the effect of ICI.	A hybrid ICIC scheme based on power and channel allocation.	Maximizing the throughput of the proposed system. Minimize the time delay and latency.	The RT deployment of the proposed scenario will be considered.	[138]

.

3.2.4. Artificial Intelligence-Based Frameworks

The increasing complexity and dynamic nature of B5G networks have introduced new challenges to conventional IM techniques. Traditional optimization-based approaches often struggle to adapt in RT to UD deployments, heterogeneous environments, and rapid UM. Recently, AI has emerged as a promising direction for developing adaptive, self-organizing IM strategies in B5G systems. AI methods, like machine learning (ML) [145], DL [146], and RL [147], can make smart choices by learning from past and current data, which helps better manage interference between different layers of the system.

Various AI-driven IM methods have been proposed in the literature. These approaches target RA, POC, UA, and BF in highly dense and heterogeneous environments. RL-based algorithms can dynamically adapt RA policies based on environmental feedback. DL models can recognize complicated interference patterns, and supervised learning algorithms have been used to improve link adaptation when the channel and interference conditions are known.

Hamden et al. [148] proposed a Q-learning-based RL technique to manage interference in UAV-assisted HetNets. The results demonstrated a significant improvement in SE and adaptability compared to traditional schemes. In [149], the authors introduced a multi-agent Q-learning framework for MOS and RA in D2D-enabled HetNets, which effectively mitigated interference in dynamic environments. A federated deep reinforcement learning (FeDRL)-D2D approach is developed in [150] to manage energy-efficient IMI in D2D-assisted sixth generation (6G) networks, demonstrating enhanced EE and reduced co-tier interference. Elsayed et al. [151] proposed a ML-based solution to address UA and inter-beam PA in mm-Wave B5G networks. The suggested model enhanced the total system sum rate and minimized interference. A hybrid ML algorithm is presented in [152] for joint link adaptation and POC in interference-limited networks, resulting in improved ST and interference resilience. In [153], the authors applied DL-enhanced virtual RA for vehicular B5G networks to support low-interference and QoS-guaranteed slicing. Their solution effectively maintained service isolation while minimizing resource collisions. Attar et al. [154] proposed a semi-distributed RL and matching game-based method for RA in D2D communications. The proposed scheme showed reduced interference and improved SE. In [155], the authors presented a deep RL algorithm for power allocation in dynamic HetNets, reducing outage rates and improving capacity. An AI-based data analytics is utilized in [156] for proactive IMI and adaptive reconfiguration in 6G networks. In [157], the authors suggested a joint Q-learning and network slicing framework for intelligent RA, achieving high SE while maintaining ICO across slices.

While AI-based approaches provide adaptive IMI, their practical deployment is hindered by high model complexity, lack of RT inference capabilities on edge devices, and the need for large-scale labeled datasets [158]. Privacy concerns and limited standardization further complicate their integration into existing RAN infrastructure [159].

In addition to these barriers, real-world deployment of AI-based IM frameworks faces several challenges. First, most supervised learning models depend on vast labeled datasets, which are difficult to collect in dynamic and privacy-constrained wireless environments [160]. Even when data are available, it may not reflect live non-stationary behavior across HetNet layers. Moreover, the deployment of DL models at the network edge is constrained by hardware limitations including memory, energy, and latency which hinders their RT performance in mobile or resource-limited nodes [161]. Another critical issue is the lack of explainability in AI decision-making. Most deep neural models operate as black boxes, limiting transparency and operator trust [162]. FL, while a promising privacy-preserving solution, introduces issues such as non-independent and identically distributed (non-IID) data distributions, model convergence problems, and increased communication overhead between devices and aggregation servers [163]. These limitations must be overcome to ensure that AI-based IM systems can move from simulation and lab settings to robust, scalable, and interpretable deployment in live B5G networks.

Table 5. Summary of related works of AI-based frameworks in literature.

Issue	Methodologies	Advantages	Limitations/Future Works	Ref.
To manage interference in UAV-assisted HetNets.	Q-learning-based RL algorithm.	Improving SE and adaptability under dynamic conditions.	Accurate channel estimation and stable convergence must be ensured.	[148]
To optimize RA in D2D-enabled HetNets.	Multi-agent Q-learning framework.	Enhancing ST and IMI in dynamic networks.	May require high training time for densely deployed environments.	[149]
To reduce interference while preserving data privacy.	FeDRL-D2D.	Increasing EE and maintaining privacy in D2D communication.	Model synchronization under heterogeneous data distributions needs refinement.	[150]
To minimize inter-beam interference in mm-Wave networks.	Supervised ML algorithm for UA and POC.	Enhancing the sum rate performance by 13–30% depending on NL.	Limited robustness in rapidly changing channel conditions.	[151]
To jointly optimize link adaptation and POC in B5G.	Hybrid ML algorithm.	Increasing throughput and spectrum utilization in interference-limited scenarios.	Interpretation and computational cost of the model remain a challenge.	[152]
To enable slicing and resource reuse in vehicular B5G networks.	DL-enhanced intelligent virtual RA.	Maintaining QoS with minimized interference across network slices.	Needs further evaluation in high-mobility vehicular environments.	[153]
To perform scalable and distributed D2D allocation.	RL with matching theory.	Reducing interference and signaling overhead while maintaining high QoS.	Scalability and responsiveness in UD conditions require more analysis.	[154]
To handle dynamic power allocation in 6G HetNets.	Deep RL-based framework.	Reducing outage rate and enhancing system capacity in RT environments.	Practical deployment constrained by latency and model update time.	[155]
To proactively control interference in future networks.	AI-driven network analytics and optimization.	Improving IPR and adaptive reconfiguration.	Dependent on data availability and accuracy.	[156]
To enhance spectrum and IM via slicing.	Joint Q-learning and network slicing framework.	Achieving high SE and isolation across users.	Large-scale testing for heterogeneous QoS demands is needed.	[157]

summarizes the key AI-based frameworks methods discussed in this section.

3.3. Taxonomy of IM Strategies in B5G Networks

To provide a holistic overview of the various IMI strategies discussed in the previous subsections, this subsection introduces a structured taxonomy. The classification is organized by key interference sources, namely D2D communication, HetNets, ICI, and AI-based frameworks.

Each category is mapped to representative mitigation strategies, categorized by their operational technique types (e.g., centralized, distributed, or learning-based) and the enabling technologies that support their implementation. This taxonomy serves as a visual and conceptual bridge between Section 3.1, Section 3.2, Section 3.3, Section 3.4 and the standardized IM schemes outlined in Section 4.

As shown in

Table 6. Taxonomy of IM strategies in B5G networks based on interference source, mitigation strategies, technique type, and enabling technologies.

Interference Source	Mitigation Strategies	Technique Types	Enabling Technologies
D2D	POC, MOS, RA.	Centralized, distributed.	Game theory, adaptive control, SINR tuning.
HetNets	Cell range expansion (CRE), LB, eICIC.	Coordinated Scheduling (COS), distributed.	Carrier aggregation (CA), ABSs, CoMP.
ICI	FR, CoMP.	Cluster-based, JP.	BF, RAC.
AI-Based frameworks	Predictive scheduling (PS), Dynamic RA, POC.	Supervised ML, RL, DL.	Radio access network intelligent controllers (RICs), FL, data analytics.

, IMI in B5G networks can be classified by interference source, mitigation strategies, technique type, and enabling technologies.

3.4. Comparative Summary of B5G Interference Mitigation Techniques

To complement the taxonomy presented in Section 3.5, this subsection provides a comparative summary of the IMI techniques considered in this study.

Table 7. Comparative summary of IMI techniques in B5G networks across key interference domains.

Interference Type	Common Mitigation Strategies	Strengths	Limitations	Example Techniques
D2D	POC, RA, MOS.	Low latency, SR.	Cross-tier interference, mode switching complexity.	Adaptive POC, MOS, D2D clustering.
HetNets	CRE, eICIC, LB.	Load offloading, improved coverage.	Coordination overhead (CO), control signaling.	ABS, CRE, CA.
ICI	CoMP, BF, COS.	Improves edge throughput, reduces overlap.	Requires tight BS coordination and backhaul.	Joint transmission CoMP, coordinated BF.
AI-Based Frameworks	Dynamic RA, PS, RL-based POC.	Adaptive, scalable, context-aware.	High complexity, data requirements, trust issues.	RL in self-organizing networks (SONs), RICs, FL.

highlights the key strategies, strengths, limitations, and representative techniques for each interference type addressed in Section 3.1, Section 3.2, Section 3.3 and Section 3.4.

3.5. Integration and Interplay Among Interference Management Strategies in B5G Networks

Modern B5G networks are expected to support UD connectivity, mMTC, and dynamic UM, all of which require the concurrent application of multiple IM strategies. While individual IM frameworks such as D2D communication, HetNets, and AI-driven control mechanisms have shown promising results, their joint integration offers significantly more robust and adaptive IMI.

For example, D2D communication within a HetNet architecture can benefit from RL-based MOS algorithms that dynamically determine whether a UE should operate in direct D2D mode or via a BS, depending on interference levels and link quality [148]. Similarly, AI-assisted RA can be leveraged in HetNets to coordinate inter-tier and co-tier interference by learning from historical data and adapting in RL [151].

However, hybrid IM approaches also introduce trade-offs. The complexity of coordination between centralized macro-cells, distributed SCs, and D2D nodes increases signaling overhead and demands accurate CSI [116]. Furthermore, combining multiple strategies often requires cross-layer optimization, where decisions made at the MAC or PHY layer must align with application-layer service requirements

Therefore, while standalone IM techniques provide modularity and ease of deployment, integrated frameworks enable more efficient SR, improved EE, and greater scalability at the cost of increased implementation complexity and coordination latency [150,156,157]. Future B5G networks may benefit from adaptive orchestration platforms that intelligently switch between standalone and hybrid IM modes based on traffic demands, mobility patterns, and environmental constraints.

3.6. Comparative Analysis of Optimization Objectives Across Interference Management Domains

To provide a quantitative perspective on the evolution and focus areas of IM strategies in B5G networks, Figure 7 presents a comparative analysis across four major IM categories: D2D communication, HetNets, ICI, and AI-Based frameworks. Each domain is evaluated based on six critical optimization objectives: maximizing data rate, maximizing EE, maximizing SE, maximizing SINR, minimizing power consumption, and minimizing outage probability.

The data underlying Figure 7 are extracted from the surveyed literature, for D2D [75,76,79,80,81,82,83,84,85,86,87,88], HetNets [108,109,115,116,117,118,119,120,121,122,123,124,125], ICI [128,129,130,131,132,133,134,135,136,137,138], and AI-based frameworks [148,149,150,151,152,153,154,155,156,157]. This analysis highlights clear trends: for instance, minimizing power consumption appears as a dominant objective across all domains, particularly in D2D and HetNet scenarios where energy-constrained UEs and SC deployments are common. Maximizing SE is another prevalent goal, particularly in AI-based frameworks where dynamic SR and intelligent scheduling are increasingly emphasized.

In contrast, objectives such as minimizing outage probability and maximizing SINR receive relatively less focus in AI-based studies, indicating an opportunity for future research to better integrate reliability-centric optimization in intelligent IM systems. Similarly, maximizing EE, while critical, is slightly less emphasized compared to maximizing data rate and SE, especially in ICI mitigation approaches where throughput often takes precedence.

This comparative analysis provides valuable insight into the current research emphasis across different IM strategies and highlights potential research gaps, such as the need for balanced attention to EE and reliability, particularly in AI-driven IM for B5G and beyond. It also partially addresses the lack of simulation by offering a systematic numerical survey to quantify research trends.

To further strengthen the technical analysis,

Table 8. Quantitative performance summary of IMI techniques.

IMI Technique	Performance Metric	Reported Improvement	Scenario/System Type	Ref.
Q-learning-based adaptive RA	SE	SE improvement under dynamic interference conditions	D2D	[75]
EH-aware RA over κ-μ shadowed fading	EE	Reduced power usage with stable link performance	D2D	[76]
Fixed and adaptive POC	SINR/network efficiency	Network efficiency enhancement	D2D	[79]
Groupwise network coding in multicast	SINR/resource utilization	Improved SINR; lower resource use	D2D	[80]
UL/DOL decoupled association and fractional FR	Outage users	Reduced number of outage UEs	D2D	[81]
Optimal SC deployment	QoS/capacity	Minimized interference; maximized capacity	D2D	[82]
Feedback-concentrated IA	Time-slot discovery/sum-rate	+6.2% slot discovery; +50% sum-rate	D2D	[83]
Stackelberg game FR in mm-Wave SCs	Convergence speed/SINR	Rapid convergence; maintained high SINR	D2D	[84]
Two-phase combinatorial RA algorithm	Sum rate/interference minimization	Maximized total data rate under uniform interference while satisfying target rate constraints	D2D	[85]
Maximum-RSS MOS	SE	Improved SE compared to baseline	D2D	[86]
Graph-theory joint PA and link selection	EE	Increased EE; reduced power use	D2D	[87]
MOS algorithm for HetNets	Outage probability	Outage likelihood variations observed	D2D	[88]
Joint NTBS and RIS optimization in NOMA HetNets	SINR/DOL coverage	Significant SINR gain; improved coverage radius	HetNets	[108]
Deep RL for joint RA and PA	EE	+22% EE gain; robust under dynamic load	HetNets	[109]
Cross-tier UL IA	SIR/capacity	SIR and capacity gains over IA schemes	HetNets	[116]
Relay-selection for IoT terminals	Throughput/concurrent users	+44% throughput; +20% users	HetNets	[117]
RFA	UL coverage	Higher UL coverage vs. no-RFA	HetNets	[118]
BF and SRE for secure D2D	Secrecy rate	Improved user secrecy rate	HetNets	[119]
Hybrid eICIC + CoMP	SE	Best SE at macro-cell and cell edges	HetNets	[120]
Joint BS-UE association and POC	Traffic offloading/IM	Enhanced offloading; reduced interference	HetNets	[121]
EE fairness and global EE optimization	EE	Significant EE gains	HetNets	[122]
Matching-theory RA	Network performance	Improved metrics with low complexity	HetNets	[123]
UFMC-based interference elimination	BER	BER reduction under frequency-offset	HetNets	[124]
Advanced interference reduction system	SE	Large SE enhancement	HetNets	[125]
CRE with ABS allocation	ICI/load distribution	Balanced cell load and reduced ICI	ICI	[128]
DL-based NPRACH detection under ICI	Access reliability/SINR	Enhanced RACH success rate under UL interference	ICI	[129]
Packet-delay outage-ratio scheduler	QoE	Enhanced QoE	ICI	[130]
MRP	Outage throughput	Throughput gains under load	ICI	[131]
SCMA-based UL ICI cancellation	BER	BER minimization at cell edge	ICI	[132]
Inter-cell RAC	Sum throughput	+65% ST vs. conventional	ICI	[133]
ICI sub-space coordination	Sum throughput	+28% ST vs. IA	ICI	[134]
Adaptive fractional FR for NOMA	SINR/throughput	SINR and throughput improvements	ICI	[135]
RFA against intentional jammers	UL coverage	Considerable UL coverage improvement	ICI	[136]
Wireless virtual embedding	SINR/QoE	Optimized SINR; guaranteed QoE	ICI	[137]
Hybrid REM for indoor SCs	Jitter/delay	Enhanced jitter and delay metrics	ICI	[138]
Q-learning in UAV-aided HetNets	SE	Significant SE improvement	AI	[148]
Multi-agent Q-learning for D2D	SE	+28–35% SE	AI	[149]
FeDRL-D2D federated DRL for EE	SE	+22% EE in D2D	AI	[150]
ML-based inter-beam power adaptation	Sum-rate	Throughput enhancement	AI	[151]
Hybrid ML for link adaptation and POC	Spectral throughput	Improved ST; robust to interference	AI	[152]
DL-enhanced virtual RA in B5G	Slice isolation/collision	Maintained QoS; reduced collisions	AI	[153]
RL + matching game for D2D RA	Interference/SE	Reduced interference; improved SE	AI	[154]
Deep RL for PA	Outage/capacity	Lower outage; higher capacity	AI	[155]
AI-driven data analytics for IM	IPR	Enhanced prediction accuracy	AI	[156]
Joint Q-learning and network slicing	SE/isolation	+20–25% SE with isolation	AI	[157]

provides a consolidated summary of reported performance gains from selected IMI techniques across recent studies. This summary highlights key performance metrics, quantifiable improvements, and their deployment scenarios in B5G networks.

4. Interference Mitigation Schemes and Techniques by 3GPP

As B5G networks evolve, IM remains a central challenge due to dense deployments, heterogeneous architectures, and increased SR. To address this, the 3GPP has developed a suite of standardized IMI schemes across multiple releases (from Rel-10 to Rel-18), targeting different components of the network. These include foundational techniques such as POC [164], COS and BF [164], and co-channel interference handling via eICIC and CoMP [165,166], as well as emerging AI-based frameworks for proactive IPR and REM [167,168].

As summarized in the taxonomy presented in Section 3.5, IMI in B5G spans a diverse set of strategies and enabling technologies, each with its own strengths and limitations. This provides a foundation for understanding how 3GPP has approached standardizing these solutions.

This section presents a synthesized overview of these techniques as categorized by key interference sources: D2D communication, HetNets, ICI, and AI-based frameworks. Although general D2D-related interference aspects are mentioned earlier in Section 3, this section introduces and focuses specifically on the corresponding 3GPP standardized mitigation mechanisms, ensuring clarity and continuity in the discussion. The following subsections detail how 3GPP approaches IM through both classical and intelligent strategies, offering a coherent and technically grounded analysis that encompasses core topics such as co-channel interference, POC, and antenna BF.

4.1. Interference Mitigation in D2D

3GPP introduced Proximity Services in release 12 [169], enabling D2D communication with IM mechanisms designed to coexist with traditional cellular users.

4.1.1. Resource Allocation

Resource sharing in D2D underlay communications poses a high risk of cross-tier interference. 3GPP proposes semi-static and dynamic scheduling approaches based on centralized or distributed architecture [170]. The BS may allocate orthogonal resources or allow reuse under SINR constraints to optimize SE while mitigating interference.

4.1.2. Power Control

To manage D2D PT and minimize its interference with cellular users, 3GPP specifies both open-loop and closed-loop POC methods [171]. Open-loop control adjusts power based on path loss, while closed-loop mechanisms allow finer adaptation using feedback from the network.

4.1.3. Spectrum Separation and Mode Selection

Splitting 3GPP permits both underlay and overlay D2D operation modes. While spectrum partitioning using TDM can mitigate interference [172], it may lead to underutilization of resources. Hence, hybrid resource partitioning with dynamic MOS based on interference thresholds is also encouraged.

4.1.4. Interference Mitigation for Vehicular-to-Everything and NTN-Enabled D2D

In addition to traditional D2D applications, 3GPP has expanded support for side-link (SL) communication in advanced scenarios such as vehicular-to-everything (V2X) and NTNs, starting from Release 14 and further enhanced in Release 16 and beyond [1,2,173,174]. For V2X, SL-based D2D is crucial for enabling low-latency and highly reliable communication between vehicles, particularly in high-mobility environments. However, these setups suffer from rapid link fluctuations, frequent topology changes, and cross-link interference, especially at road intersections or in multi-lane highway scenarios [175]. To address these challenges, 3GPP specifies advanced SL control resource configurations and dynamic resource reservation schemes that allow for proactive IMI [174,176].

In NTN environments such as UAV-assisted or satellite-based networks D2D communication faces unique impairments, including Doppler shifts, long propagation delays, and spatial beam overlap. These factors complicate POC and scheduling due to desynchronized link behavior across altitude levels [177]. 3GPP has initiated study items under Release 17 and 18 to explore Doppler compensation mechanisms, predictive beam steering, and adaptive link control to maintain SL reliability while minimizing interference in these dynamic topologies [178,179].

4.2. Interference Mitigation in HetNets

To support capacity scaling and improved coverage, HetNets composed of macro-, micro-, pico-, and femto-cells have become essential in B5G. However, their uncoordinated and multi-tier deployment introduces severe co-tier and cross-tier interference, which 3GPP addresses with the following strategies:

4.2.1. Carrier Aggregation

CA, introduced in 3GPP Rel-10 [180], allows UEs to utilize multiple component carriers across frequency bands. In HetNets, CA helps isolate macro-cell and SC transmissions by distributing traffic over primary and secondary component carriers, enhancing both SE and interference avoidance.

4.2.2. Load Balancing and Cell Range Expansion

3GPP supports load-aware handovers and offloading mechanisms via CRE and mobility biasing [113]. These strategies aim to balance user distribution across tiers while coordinated resource partitioning (e.g., ABS) ensures that expanded cells do not become interference limited.

4.2.3. AI-Driven Coordination in HetNets

The increasing complexity of HetNet deployments in B5G, especially involving dense and overlapping tiers of macro-cell, micro-cell, and SC, poses new coordination and interference challenges. To address these, 3GPP has introduced architectural support for AI-driven coordination through the integration of RICs, particularly within open RAN frameworks. Near-RT RICs are designed to operate at the control plane of the RAN and facilitate dynamic decision-making for RA, BF management, and IMI [181,182].

In the context of HetNets, near-RT RICs can support dynamic handover optimization by learning UM patterns and traffic distribution across cells. They can also perform cross-tier interference suppression, coordinating between macro-cell and SC by adjusting power levels, assigning beam directions, or reconfiguring scheduling patterns. Furthermore, traffic-aware radio RA is enabled by analyzing load conditions and user QoS requirements, ensuring balanced utilization of available spectrum across tiers [183]. These functions are realized through extensible applications (EAPP) deployed on RIC, which leverage RT telemetry, AI models, and standardized interfaces to implement feedback-driven optimization strategies [184].

The adoption of AI-enhanced RICs for HetNet coordination allows for more scalable and adaptive IM compared to traditional static or semi-static solutions. However, challenges remain regarding EAPP interoperability, AI model generalization across deployment scenarios, and maintaining low-latency responses under constrained edge computing environments [185].

4.3. Inter-Cell Interference Mitigation Techniques

ICI is especially prominent at the edges of cells in dense deployments. 3GPP has standardized several cooperative techniques to mitigate ICI.

4.3.1. Enhanced Inter-Cell Interference Coordination

eICIC [18,186] improves cell-edge performance by introducing ABS, allowing macro-cell and SC to operate on the same frequency with temporal separation. This is particularly effective in macro–pico configurations using CRE.

4.3.2. Coordinated Scheduling and Beamforming

COS enables neighboring BSs to exchange control signaling and schedule non-overlapping transmissions [187]. Sharing BF vectors between cells can also suppress interference from adjacent sectors by steering beams away from vulnerable users.

4.3.3. Coordinated Multipoint

Introduced in Rel-11 [188], CoMP allows JP or dynamic point selection (DPS) across a cluster of BSs. In UL CoMP, signals from a UE are jointly received by multiple cells, improving reception at the cost of increased backhaul complexity. TDD systems benefit further from channel reciprocity, reducing feedback overhead.

4.3.4. ICI in Dynamic Time Division Duplexing and Satellite-Enabled Deployments

The emergence of dynamic TDD and NTNs in B5G introduces new dimensions to ICI, particularly due to the increased variability and asymmetry of traffic. In dynamic TDD systems, the UL and DOL transmission patterns are no longer fixed, enabling flexible slot allocation based on RT traffic demand. However, this flexibility can cause severe asymmetric ICI, especially when neighboring cells independently schedule UL and DOL transmissions. Such mismatches lead to cross-link interference, where a cell’s DOL transmission interferes with another cell’s UL reception [189]

To mitigate this, 3GPP has proposed enhancements involving adaptive guard periods, inter-cell slot alignment, and UL/DOL configuration exchange mechanisms to coordinate dynamic TDD scheduling across BSs [190]. These mechanisms are particularly critical in dense deployments and in scenarios with high mobility, such as high-speed rail or UAV-supported cells.

In satellite-enabled deployments (e.g., low earth orbit constellations), ICI challenges stem from beam overlap, non-uniform footprint distribution, and link desynchronization between satellites and terrestrial elements. The long propagation delays and Doppler shifts further complicate the synchronization of time and frequency resources [191]. To address these, 3GPP Release 17 and ongoing Release 18 study items explore techniques such as beam nulling, adaptive inter-beam scheduling, and predictive Doppler compensation to reduce the impact of ICI in integrated satellite-terrestrial networks [179,192]

These advances reflect the need for coordinated and predictive IMI strategies that operate effectively under non-ideal, high-dynamic conditions, particularly where traditional synchronous RA is infeasible.

4.4. AI-Enabled Interference Mitigation

To address the increasing complexity and dynamic behavior of B5G networks, the 3GPP has initiated efforts to incorporate AI into the standardization of IMI techniques. AI techniques such as supervised learning, DL, and RL are gradually being considered for integration into 3GPP’s SONs and RICs frameworks.

4.4.1. AI-Enabled Radio Access Network Intelligent Controllers

As part of the open RAN alliance, which is closely aligned with 3GPP’s architecture, the integration of near-RT and non-RT RICs is proposed [193]. These controllers use AI/ML models to optimize interference-related functions such as beam management, resource block allocation, and LB [194]. AI agents in RICs continuously monitor interference levels and make autonomous decisions to mitigate it without manual reconfiguration, thus enabling intelligent and adaptive interference coordination [19].

4.4.2. AI-Based Interference Prediction and Traffic Steering

3GPP specifications (e.g., technical specifications 28.104 and 28.105) outline how AI-driven data analytics can be used to predict interference events before they occur [195]. This proactive approach allows networks to apply traffic steering, FR optimization, or antenna tuning to avoid potential interference zones. Such functionality is particularly effective in UDNs and supports enhanced QoE [196].

4.4.3. Network Slicing-Aware IM

To support diverse service types (e.g., eMBB, URLLC), 3GPP encourages AI-assisted network slicing [197]. AI models ensure interference-aware resource isolation between slices, reducing cross-interference. For example, AI can dynamically allocate spectrum to URLLC slices while ensuring latency and reliability targets are met, even under high-load conditions [198].

4.4.4. Reinforcement Learning in Self-Organizing Networks

In the 3GPP SON framework, AI-based RL agents are proposed to autonomously learn optimal ICO strategies [199]. These agents consider metrics such as SINR, throughput, and user density to adaptively manage power levels, adjust scheduling, or reconfigure cells. This aligns with the 3GPP goal of supporting zero-touch operation and automation in B5G networks [200].

As summarized in Figure 8, 3GPP’s standardized IMI strategies address the key challenges of B5G networks, including co-channel interference, dense and heterogeneous deployments, and dynamic traffic behavior. These solutions encompass both classical techniques such as POC [3], COS [42], and BF [201] and, AI-driven methods for RT adaptation and proactive IPR [202,203]. Together, they form a cohesive and adaptive framework that not only enhances SE and communication reliability but also aligns with the long-term vision of autonomous and intelligent RANs.

4.4.5. Standardization Challenges and Open Issues

While AI-based frameworks hold significant promise for dynamic and intelligent IM, their deployment in real-world B5G networks faces a number of standardization and technical challenges. One of the most prominent issues is the integration of FL into the RAN. Although FL offers privacy-preserving benefits by avoiding centralized data aggregation, it struggles with non-IID data across edge devices, which can hinder model convergence and stability [159]. Furthermore, frequent model updates in FL incur communication overhead and require secure aggregation protocols, which are still under investigation in 3GPP Release 18 study items [204].

Another major limitation lies in the lack of explainability and interpretability of AI models, particularly deep neural networks. These models often behave like black boxes, making it difficult for network operators to verify decisions related to IPR, RA, or POC. This undermines trust and raises regulatory compliance concerns in mission-critical applications [162].

Moreover, deploying AI at the network edge introduces latency and hardware limitations. Many IM decisions such as POC or dynamic scheduling require sub-millisecond response times, which can be difficult to achieve with existing AI inference frameworks on mobile devices or edge servers. Efforts to compress and optimize AI models, such as pruning and quantization, are being explored but are not yet standardized within 3GPP [158].

Lastly, cross-vendor interoperability remains an open issue. AI modules deployed across multi-vendor networks must be interoperable and compatible with diverse hardware and software implementations. 3GPP and the open RAN alliance are currently working toward standardized interfaces and common ML frameworks, but trust frameworks, model verification, and lifecycle management remain active areas of research and standardization [185].

Addressing these challenges is essential to ensure that AI-based IM techniques can transition from controlled simulations to scalable, trustworthy, and production-grade deployment in B5G and future 6G networks.

4.4.6. Standardization Mapping and Industry Maturity

Many of the IM techniques discussed in this review are addressed across 3GPP Releases 15 through 18. For instance, eICIC and CoMP, first introduced in earlier releases, have seen ongoing enhancements in Release 15 and beyond, including integration with dynamic scheduling and beam coordination. Dynamic TDD coordination has evolved in Release 16 with guard period adaptation and slot alignment for UL/DOL balancing [205].

The integration of AI and ML into RAN architectures is a significant aspect of Release 17 and 18. These releases introduce the concept of near-RT and non-RT RICs, aligning with the open RAN Alliance architecture, to enable AI-driven interference detection and mitigation [206]. FL for traffic steering and IPR is actively studied in Release 18 [207], although normative specifications are still under development.

AI-native IM techniques such as RL for BF or scheduling are being trialed in early deployments and open testbeds, but full industry adoption requires overcoming challenges related to model interpretability, cross-vendor interoperability, and trust management. These issues are reflected in joint efforts by 3GPP and open RAN to define key performance indicators, lifecycle hooks, and interfaces for AI explainability and orchestration [208].

A summary of key IM techniques mapped to their respective 3GPP releases and maturity levels is shown in

Table 9. Mapping of IM techniques to 3GPP releases and maturity levels.

IM Technique	3GPP Release	Standardized Under	Maturity Level
eICIC	Rel-10/Rel-15+	TS 36.300/38.300 series	Widely deployed
CoMP (JP, DPS)	Rel-11/Rel-15+	TS 36.819/TS 38.214	Deployed with limitations
Dynamic TDD coordination	Rel-16	TS 38.331	Field-tested in UDN trials
RIC-based AI (non-RT/near-RT)	Rel-17/Rel-18	TS 38.533/open RAN WG2/WG3	Pre-commercial/maturing
FL-based IM	Rel-18 (study)	TR 38.838/TS 28.104, 28.105	Research phase

.

Although 3GPP-standardized techniques such as CoMP and eICIC offer substantial theoretical gains, their effectiveness depends heavily on deployment conditions. For instance, CoMP requires robust and low latency backhaul connectivity and precise synchronization among BSs conditions not always available in rural or legacy systems. AI-driven RICs further depend on high computational capacity and continuous data collection, which may be difficult to sustain at the network edge.

4.5. Comparative Discussion and Trade-Offs

While a wide range of IMI techniques have been standardized by the 3GPP and proposed in the literature, their effectiveness, complexity, and suitability vary depending on deployment scenarios and infrastructure capabilities. For instance, eICIC, introduced in 3GPP Rel-10 [18,186], is well suited for macro–pico deployments and offers simple implementation using ABS. However, it may lead to spectral inefficiency in the macro cell during ABS periods. In contrast, CoMP, standardized in Rel-11 [188], can significantly enhance signal reception at the cell edge through joint transmission or DPS but requires high-capacity backhaul and precise synchronization, which limits its feasibility in less-connected or rural networks [209,210,211].

POC methods [164,171], including both open-loop and closed-loop approaches, are relatively simple and effective in IM in D2D and HetNet scenarios, especially where network infrastructure cannot support advanced coordination. However, these techniques often fall short in dynamic environments with rapidly varying interference patterns. Meanwhile, BF and COS [164,187] provide better directional IMI and temporal separation but require CSI sharing and synchronization, increasing signaling overhead and computational demands.

AI-based IM frameworks, such as RL, DL, and FL, have emerged as promising solutions capable of dynamically adapting to network conditions [147,150,155]. These methods outperform static rule-based approaches in dense and heterogeneous deployments by enabling RT IPR and proactive REM [156,167,196]. However, their practical deployment remains constrained by training complexity, inference latency, and the need for robust data collection mechanisms [212,213,214]. Additionally, challenges such as trustworthiness and explainability must be addressed before wide-scale adoption [215,216,217].

To visualize the internal dynamics of these advanced frameworks, Figure 9 presents a high-level pipeline illustrating how AI techniques integrate with network elements to perform adaptive IMI. The pipeline outlines how data are collected from distributed network elements (e.g., BSs and UEs), processed through centralized or edge-based AI modules, and used to train or update learning models within the non-RT RIC. The near-RT RIC then executes RT decisions such as PF adjustments, POC, or RA based on these models. This integration enables dynamic, context-aware IMI but also introduces challenges related to training complexity, inference latency, and standardization of interfaces between AI components and RAN nodes.

In real-world deployments, the effectiveness of IMI techniques can vary significantly depending on infrastructure capabilities, mobility patterns, and load conditions. As a result, there is growing interest in adaptive switching between IM strategies to ensure robust performance under varying scenarios. For instance, when high capacity backhaul and tight synchronization are available, CoMP offers substantial gains at the cell edge. However, in backhaul-limited or rural environments, fallback to eICIC with ABS or POC techniques may provide more feasible solutions [218,219]. Similarly, AI-based controllers, such as near-RT RICs, can monitor traffic dynamics and interference levels to orchestrate RT transitions between strategies (e.g., deactivating CoMP clusters during low-load periods or dynamically applying eICIC in response to mobility-induced ICI) [220].

Such context-aware orchestration of IM techniques represents a scalable approach for future 5G and 6G networks. It aligns with the goals of zero-touch automation and intelligent RAN slicing while addressing trade-offs between complexity, overhead, and SE [182]. Adaptive strategy switching ensures that IM solutions are not only optimized for performance but also for EE, hardware constraints, and operator policy.

Overall, trade-offs between implementation complexity, CO, adaptability, and SE must be considered when selecting an IMI technique. For example, eICIC is suitable for cost-sensitive SC deployments, CoMP is ideal for capacity-driven urban networks with strong infrastructure, and AI-based techniques offer long-term scalability and automation for UD or SONs. These comparative insights complement the analysis previously summarized in Table 7 (Section 3.4) and provide a contextual understanding of each technique’s strengths and limitations for B5G applications.

Figure 10 summarizes the trade-offs between major IMI techniques across key performance metrics. As illustrated, AI-based methods score high in scalability and SE but are challenged by complexity and data requirements [148,150,152,156,157,211,212,213,214]. At the same time, techniques like eICIC and POC offer simpler, infrastructure-friendly solutions at the cost of adaptability or precision [120,164,171,186].

To further clarify applicability,

Table 10. Suitability of various IMI techniques across different B5G deployment scenarios. Check marks (✓) indicate strong suitability, dashes (—) represent moderate feasibility with limitations, and crosses (✕) denote low practical viability.

Technique	Urban Dense	Rural	UAV/NTN	Infrastructure-Limited
CoMP	✓	✕	—	✕
eICIC	✓	—	✕	✓
POC	—	✓	—	✓
AI-Based	✓	—	✓	—

provides a deployment-based evaluation of IMI techniques across key scenarios, including urban dense, rural, infrastructure-limited, and aerial/NTN environments. The table highlights how techniques like CoMP thrive in well-connected urban settings [188,209,210,211] but face barriers in rural deployments, while POC and eICIC remain viable in cost-constrained or low-infrastructure contexts [120,164,171,186]. AI-based approaches offer versatile adaptability but are limited by high system demands [150,152,156,215,216,217,218,219,220].

4.6. Interference Management for NTN and 6G Emerging Architectures

As B5G networks continue to evolve, the integration of NTNs and the anticipated shift toward 6G AI-native architecture bring new challenges and opportunities for IM. 3GPP Releases 17 and 18 have initiated formal study items on new radio for satellite access and interworking between terrestrial and non-terrestrial components. In such deployments, interference arises from beam overlaps, Doppler shifts, and asynchronous transmissions between ground terminals and moving satellite platforms [179,221]. Effective IM solutions must consider PS, adaptive BF, and cross-layer coordination that accounts for orbital dynamics and propagation delays.

In parallel, AI-native networks envisioned in 6G will require fully automated IM mechanisms. These will rely on distributed intelligence embedded in the RAN and core to support RT IPR, mitigation, and resource orchestration [222]. Unlike conventional model-driven techniques, 6G IM frameworks are expected to use continual learning, self-evolving policies, and federated intelligence to adapt across highly dynamic, heterogeneous environments. These solutions will need to address multi-dimensional interference, including from aerial devices, IoT swarms, and multi-access edge nodes without centralized control.

Moreover, the convergence of communication and sensing, often referred to as integrated sensing and communication (ISAC), will add new complexity to interference modeling. In ISAC systems, signals used for data transmission also carry radar or positioning functions, which can interfere with coexisting transmissions unless tightly coordinated. Emerging 3GPP and international telecommunication union discussions on 6G recognize the need for spectrum-aware waveform design and dual-function resource scheduling to manage this new interference class [223,224]. Recent studies also explore AI-based IMI in ISAC systems, which enable intelligent coordination between sensing and communication tasks in dense, dynamic environments [225].

Together, these trends underscore the necessity for IM strategies that are not only adaptive and intelligent but also cross-domain, spanning terrestrial, aerial, and orbital layers. Advancing these solutions will be key to realizing the low-latency, high-reliability, and massive-connectivity goals of 6G.

5. Critical Analysis and Open Research Challenges

Although substantial progress has been made in IMI strategies for B5G networks [226,227,228], several gaps and limitations remain, especially in the context of large-scale deployments, RT responsiveness, and integration with emerging network paradigms. This section provides a critical assessment of the reviewed techniques and outlines open research challenges that need to be addressed for the practical and scalable implementation of IMI in future wireless systems.

5.1. Scalability and Real-World Applicability

Many IMI techniques particularly those involving game theory or mathematical optimization [229,230] perform well in controlled environments or simulations but may not scale efficiently in UD B5G deployments [231,232]. These methods often assume ideal conditions, such as perfect channel knowledge or static user distributions, which may not hold in dynamic real-world networks.

5.2. Trade-Off Between Performance and Computational Complexity

AI-based methods, including DL and RL [212,233], have shown great potential in achieving adaptive and intelligent ICO. However, these solutions frequently suffer from high computational overhead, slow convergence times, and the need for extensive training datasets. Their deployment on resource-constrained edge devices remains a challenge, especially in time-sensitive applications such as URLLC [213,214].

5.3. Coordination Overhead in Cooperative Techniques

Techniques like CoMP, eICIC, and joint scheduling [209,210,211] require significant coordination between BSs, which may lead to high signaling overhead and increased latency. In rural or infrastructure-limited regions, backhaul capacity and synchronization constraints can limit the feasibility of such solutions.

5.4. Interference Across Heterogeneous and Multi-Domain Networks

Most reviewed approaches focus on terrestrial IM. However, B5G networks are expected to incorporate NTNs, such as UAVs [37], high-altitude platforms [234], and low earth orbit satellites [235]. These additional layers introduce new interference dynamics (e.g., air-to-ground, satellite-to-terrestrial links) that are insufficiently addressed in current frameworks.

5.5. Lack of Standardized AI Integration

While the 3GPP has begun incorporating AI/ML techniques in SONs and RICs [236,237], there is still a lack of standardized, interoperable frameworks for AI-IMI. Challenges such as generalization across network topologies and hardware heterogeneity remain unsolved.

5.6. Security and Trustworthiness of AI Models

AI-based IMI solutions may be susceptible to adversarial attacks, model spoofing, or data poisoning [215,216,217]. These risks threaten the robustness and reliability of AI agents responsible for critical network decisions. Ensuring the security, transparency, and trust of AI models is essential for practical adoption.

5.7. Comparative Evaluation of IMI Techniques

While various IMI strategies have shown promising results in B5G networks, their deployment viability varies based on system constraints. A comprehensive evaluation of these techniques requires considering key trade-offs in terms of implementation complexity, scalability in dense network environments, and practical feasibility.

Table 11. Trade-off evaluation of IMI techniques in B5G networks.

Technique	Complexity	Scalability	Feasibility	Key Challenges
POC	Low	High	High	Static thresholds, limited adaptability
eICIC	Low–Medium	Medium	High	ABS inefficiency in non-macro–pico setups
CoMP	High	Low–Medium	Medium	High backhaul, CO
BF	Medium–High	Medium	Medium	CSI dependency, computation cost
AI-based (RL/FL)	High	High (with edge AI deployment)	Medium	Training time, inference latency, model trust

summarizes this comparison across several widely studied IMI techniques. For example, POC and eICIC are relatively simple and feasible for existing infrastructure but offer limited adaptability in UDNs. On the other hand, CoMP and BF deliver superior performance at the cost of high complexity and CSI dependency. AI-based methods like RL and FL show high adaptability and scalability but face deployment challenges related to training time, hardware acceleration, and explainability.

These trade-offs underscore the need for hybrid, context-aware IM frameworks capable of dynamically selecting or combining techniques based on RT network conditions.

This comparison demonstrates that no single IMI strategy is universally optimal. Future systems must dynamically balance these trade-offs, potentially through hybrid or AI-driven orchestration mechanisms adapted to deployment context.

5.8. Future Research Directions

To address these challenges, several promising research avenues are recommended:

5.8.1. Hybrid Approaches [238,239]

Combine rule-based and data-driven methods to balance efficiency and adaptability, particularly in scenarios with limited training data or RT constraints.

5.8.2. Federated and Decentralized Learning [240,241,242]

Implement federated RL or distributed ML frameworks that preserve data privacy while enabling scalable IMI without centralized coordination.

5.8.3. Cross-Domain IM [243,244]

Develop integrated frameworks that consider aerial, terrestrial, and satellite interference, enabling coordinated decisions across all layers of the B5G ecosystem.

5.8.4. Lightweight AI Models [245,246]

Design simplified yet effective AI models (e.g., tiny ML) tailored for deployment on resource-constrained devices such as edge nodes and IoT gateways.

5.8.5. Explainable AI [217,247]

Enhance the interpretability of AI decisions for ICO by developing transparent models that can be audited and trusted by network operators.

5.8.6. Standardized Benchmarks and Datasets [3,248]

Create open-source datasets, simulation platforms, and performance benchmarks to evaluate and compare IMI strategies under standardized B5G scenarios.

5.8.7. Optical Wireless Communication and Interference Considerations [249,250,251,252]

Optical wireless communication, including visible light communication, light fidelity (LiFi), and free-space optical communication, is gaining attention as a complementary technology in B5G and beyond due to its large unlicensed BW and low electromagnetic interference. In dense indoor or urban environments, LiFi offers interference-free operations compared to radio frequency (RF) systems, but it also introduces new interference types, such as optical beam misalignment, multi-user access interference, and noise from ambient light sources. To manage these, AI-based RA and dynamic beam steering methods are proposed, similar to RF-domain IMI. Recent research explores hybrid RF/optical systems where coordinated IM across domains enhance reliability and QoS. Future B5G networks may benefit from cross-domain interference awareness involving both RF and optical channels, especially in UD or latency-critical applications.

5.8.8. Real-World Deployment Considerations

Despite the significant theoretical advancements in IMI strategies, their practical deployment in real-world B5G environments faces numerous challenges. First, many advanced IM techniques, such as CoMP and DL-based scheduling, require precise synchronization and low-latency backhaul connectivity between BSs, which are often unavailable in rural or infrastructure-limited deployments [221]. Hardware heterogeneity, including variations in processing capabilities and memory at edge nodes, further complicates consistent IM implementation across network tiers [203]. In AI-based IMI frameworks, model complexity and inference latency hinder RT decision-making, especially when deployed on constrained edge devices such as IoT gateways and mobile UEs [203]. Additionally, cooperative techniques like joint transmission, centralized scheduling, and coordinated BF introduce significant signaling overhead, which can degrade performance under mobility and limited backhaul conditions [238]. Finally, achieving interoperability across multi-vendor RAN components and aligning with standardization efforts from 3GPP and the open RAN alliance remain essential for large-scale, reliable deployment [221]. Addressing these practical constraints is crucial to translating simulation-based performance gains into scalable and robust deployment outcomes.

5.8.9. Practical Challenges in AI-Based IM

Although AI-based IMI strategies such as DL, RL, and FL have demonstrated strong performance in dynamic B5G environments, their real-world adoption faces notable practical constraints. First, training high-performing AI models requires extensive and representative datasets, which are often unavailable or difficult to collect in mobile, privacy-constrained wireless networks [158,160]. Supervised and RL models further suffer from long training times and the need for retraining in response to changing interference conditions.

Second, deploying AI inference on edge devices presents significant limitations. Mobile UE, IoT nodes, and edge gateways typically have constrained compute, memory, and power budgets, which makes it difficult to run DL models with low latency [161]. This undermines the RT responsiveness required for functions like POC or dynamic RA in URLLC and mMTC scenarios.

Additionally, the inference cost is not only computational but also includes latency introduced by data movement, model complexity, and lack of hardware acceleration in edge platforms. Techniques like pruning, quantization, and tiny ML can help reduce model size, but they often trade off prediction accuracy or robustness under diverse interference conditions [241,242].

FL, while beneficial for preserving privacy and decentralizing learning, brings challenges of model divergence due to non-IID data, increased communication overhead for aggregation, and slow convergence [159,163]. These issues are amplified in dense HetNets with frequent topology changes.

Finally, the lack of explainability in AI decisions limits operator trust and compliance in mission-critical networks. Most DL models act as black boxes, offering little transparency into why certain IM decisions are made [162,253]. Regulatory and standardization gaps also persist regarding model lifecycle management, auditing, and cross-vendor interoperability.

Overcoming these challenges is essential to bridge the gap between theoretical AI potential and practical, scalable deployment in interference-critical B5G scenarios.

As highlighted in Figure 7, while maximizing SE and minimizing power consumption have been primary focuses across different IM domains, future research must also address the relatively underexplored objectives, such as outage probability minimization and EE optimization, especially in AI-based IMI frameworks.

Furthermore, while AI-based IMI techniques offer promising adaptability in dynamic B5G environments, their practical deployment introduces challenges that extend beyond technical feasibility. As discussed in Section 4.4.5, FL suffers from non-IID data across devices, leading to model divergence and slow convergence [254]. Additionally, performing AI inference at the edge is constrained by latency, compute limitations, and energy constraints, which hinder RT decision-making in interference-critical scenarios [158]. These challenges are compounded by a lack of standardization in AI interfaces, trust frameworks, and model lifecycle management, especially in multi-vendor RAN environments [255]. Overcoming these technical and standardization barriers is essential to transition AI-based IMI strategies from simulation testbeds to scalable, real-world B5G deployments.

In addition, the transition from traditional optimization to AI-native IM frameworks demands not only performance improvements but also explainable and trustworthy models. As discussed in Section 4.4.5, FL suffers from non-IID data across devices, leading to model divergence and slow convergence [254]. Performing inference at the edge is further constrained by latency, computing limitations, and energy constraints, hindering RT decision-making [158]. These issues are compounded by the lack of standardization in AI interfaces, trust frameworks, and lifecycle management, especially across multi-vendor RAN ecosystems [255]. Community-driven initiatives such as standardized key performance indicators for AI explainability in critical applications (e.g., URLLC and aerial/NTN), open testbeds, and joint development of interoperable EAPP for RICs can accelerate the path toward scalable, trustworthy B5G deployments.

Pursuing these directions will not only enhance IM in B5G networks but also contribute to the foundational technologies required for future 6G networks.

6. Conclusions

Interference during transmissions that utilize the same time-frequency wireless resources is one of the main issues in cellular networks. The interference issues have become more critical due to the continuous growth in traffic, density, and size of wireless networks, especially in B5G. Recently, new communication techniques have been developed to deal with interference, along with progress in the characterization of network limitations, which has improved our understanding and ability to enhance performance. This paper presents a comprehensive review of the IM issues in B5G cellular communications and highlights the significance of addressing interference in D2D communication, HetNets, ICI, and AI-based frameworks. Promising solutions have been extensively discussed, with a focus on the methodology, strengths and limitations, and open research challenges. Furthermore, the paper presents and analyzes the standardized IMI schemes proposed by the 3GPP. As a literature review, this study provides a comprehensive analysis of current research trends, approaches, and directions in B5G IM. The goal is to guide researchers by summarizing validated results and identifying areas that warrant further theoretical or simulation-based investigation. Key insights include the growing role of AI-based coordination, the shift toward distributed and predictive IM strategies, and the increasing complexity of managing interference across heterogeneous and NTNs. Challenges such as scalability, CO, and AI transparency remain open research problems. In addition, a comparative perspective on optimization priorities across IM domains reveals that while objectives like SE and power consumption have received significant attention, aspects such as outage minimization and EE remain underexplored. By addressing these gaps, future studies can advance toward building more adaptive, secure, and intelligent IMI frameworks. In conclusion, effective IMI, particularly through 3GPP-proposed techniques, represents a key step toward the successful deployment of future B5G wireless networks.