1. Multi-Connectivity to Improve Quality of Service
The improvement of QoS is the first objective associated with the use of multi-connectivity in 5G and future networks.
1.1. Existing Solutions to Improve QoS
Various papers have already explored the use of MC to improve the QoS. More specifically, these papers have focused on four main topics: perceived throughput improvement 
, latency reduction 
, reliability 
, and availability improvement 
The improvement of the user’s perceived throughput is usually presented as one of the intrinsic benefits of multi-connectivity (bandwidth additions). Therefore, various papers have focused on this subject such as 
. In this work, the authors of 
introduced a new mechanism that aimed to maximize the overall system throughput for DL in 5G networks and beyond. To achieve this, they first formulated the optimization problem in a sum-of-ratios form. Then, to solve this non-deterministic polynomial hard (NP-hard) problem, they introduced an iterative algorithm based on the Karush–Kuhn–Tucker (KKT) conditions, which allowed them to define a sub-optimal solution. The evaluation carried out by the authors demonstrated the benefits of this solution in terms of throughput maximization. However, the scenario considered by the authors seems simplistic (single macro-cell BS, lack of mobility). Moreover, the inter-macro-cell BSs’ handover, which affects the performance, is not addressed in this proposal. Similarly, the authors of 
tried to improve the throughput perceived by users. The main contribution of this entry was the definition of an architecture enabling the use of millimeter waves (high-throughput) in a multi-connectivity scenario while guaranteeing a high level of connectivity. To achieve this, the authors focused on the definition of an MC framework based on the existing standard, which should allow a UE to connect simultaneously to a 4G BS and a 5G BS supporting millimeter waves. The authors also introduced an algorithm for the optimal selection of secondary BSs that aimed to maximize connectivity for UEs. An important limitation of this work, which focused on the definition of an architecture integrating millimeter waves, was that it only considered non-standalone 5G architecture (and not standalone architecture). Therefore, it is impossible to measure the potential long-term benefits of the proposed solution. Moreover, the proposed selection algorithm was simplistic and inflexible (definition of fixed thresholds). The authors of 
also looked at the use of millimeter waves in an MC scenario. In particular, to deal with the limitations of millimeter waves (range, frequent disconnections), they proposed a mechanism to solve the related link-scheduling problem. The proposed iterative algorithm was based on temporal decomposition. It aimed to maximize the throughput perceived by the users for each of these time windows while taking into account the power limits of the milliliter BS. The solution and experiments introduced in this entry were more comprehensive than those described in 
. However, the approach described here appeared to be costly both in terms of control (exchange between devices) and management (number of handovers). It could therefore be interesting to evaluate the overheads associated with this approach and try to minimize them.
Latency reduction is the second potential benefit of the use of MC in 5G and future networks 
. To achieve latency reduction, the authors of 
have introduced an MC architecture enabling different RATS to be managed simultaneously: 4G LTE, 5G NR, Wi-Fi and WiGig. The overall management of this architecture is based on the use of an SDN controller. This should allow the selection of the most suitable RAT according to the user’s needs and the performance guaranteed by each RAT such as latency, bandwidth, etc. The main interest of this entry is that it proposes a software architecture that could allow fine-grained QoS management for each user or even each application. However, the proposed solution is still in the draft stage. Therefore, it could be interesting to use this proposal as a basis for defining a powerful software solution for MC scenarios. The authors of 
sought to find a trade-off between latency reduction and resource utilization in a multi-connectivity scenario. To achieve that, they proposed to activate multi-connectivity only for users requiring it (critical applications). Thus, they defined a heuristic latency-aware MC configuration algorithm integrating a critical latency threshold enabling the determination of the packets for which the activation of multi-connectivity is relevant. This solution, limiting the use of multi-connectivity, naturally demonstrated significant gains in terms of the use of resources. However, this solution did not take into account the network load in the latency threshold calculation. Similarly, only one RAT was considered. Thus, it might be interesting to consider new parameters and a more complete architecture to improve the performance of the proposed solution. The authors of 
provided a comprehensive evaluation of the performance of a 4G/5G cellular architecture including MC services applied to latency reduction. To achieve this, they used the Measuring Mobile Broadband Networks in Europe (MONROE) testbed and implemented the LTE-SQ (basic MC management) algorithm with different configurations such as preferred RAT, number of simultaneous connections, etc. This entry focused on the performance evaluation of MC and described a framework and a set of measurements that could be used as a starting point for further work in this field. Thus, rather than limitations, this entry provided areas for improvement such as user preferences, network history, etc.
Reliability improvement also seems to be a potential benefit of multi-connectivity 
. Reliability is generally improved by duplicating the packets transmitted by/to the UEs 
. Thus, for critical applications, each packet is simultaneously transmitted to different BSs (primary/secondary) thanks to multi-connectivity. These different papers have the same objective, that is, to upgrade the standardized MC architecture to enable an efficient duplication of packets at the master base station level and, more specifically, at the PDCP (Packet Data Convergence Protocol) level. In addition, these papers, which focus on a theoretical definition of architecture, raise various questions related to the optimization of the RLC layer for PDCP duplication and the handover management in a duplication scenario. Going further, the authors of 
proposed a machine learning-based RAT selection that aimed to improve reliability in an MC scenario. In this solution, packet duplication was carried out by taking into account the performance of each of the RATs and the history of the users. Thus, packets were duplicated only when necessary (reliability of the different RATs at a given time for a given user) and by selecting the RATs that guaranteed the highest level of reliability. This work was the first contribution that aimed to improve the reliability of communications through MC. However, the authors considered in this entry that all users have the same QoS requirements (homogeneous traffic). Moreover, the overheads associated with this machine learning-based solution (computation, latency) have not been evaluated. Therefore, many challenges remain in this field.
Finally, improved availability is the last potential benefit of multi-connectivity. However, similar to reliability improvement, this topic has not been studied extensively so far and only two papers provide relevant solutions 
. The solution proposed in 
focused on the availability of services (slices) at the network edge. Specifically, the authors aimed to ensure that a given service (slice) needed by a given user was available at the primary and secondary BSs to which this user could connect. To achieve this, they proposed a handover algorithm that aimed to ensure the transfer of the user’s data and required services. This mechanism, intended to be integrated into the 3GPP reference architecture, is simple. However, this entry introduced an interesting idea, that is, the dynamic management of slices in a multi-connectivity scenario. To support all the applications, it would seem necessary to consider different improvements, in particular, the pre-deployment of services according to the mobility of the users and the management of the users’ requirements (latency, bandwidth). The authors of 
described availability as the ability of the network to meet the users’ needs in terms of latency, packet loss rate, and bandwidth. Therefore, they proposed a framework based on an exhaustive search method to maximize the communication link utilization and to select the most appropriate communication channel for each user/application. This more classical solution also offers interesting perspectives reliability improvement through MC. However, the scenarios considered for the evaluation seem to be unrealistic (architecture, mobility) and the framework defined in this entry is based on simple and inflexible mechanisms.
1.2. Discussion Regarding QoS
Different papers have already tried to improve QoS in 5G networks and beyond through MC (see Table 1
). As a result, the ideas of perceived throughput improvement 
, latency reduction 
, reliability 
, and availability improvement 
have also been considered. Therefore, these papers, considering the different approaches, different RATS (cellular/Wi-Fi), and different tools (machine learning, heuristic, etc.), seem to be an interesting starting point for further research in this area. However, several limitations can be identified:
Table 1. Comparison of existing solutions to improve QoS.
First, it might be interesting to develop solutions that would jointly address the different QoS parameters (throughput, latency, jitter reliability, availability). Indeed, these parameters are intrinsically linked to each other. For example, solutions developed to improve reliability (packet duplication) will have an impact on network latency and available throughput. Similarly, maximizing the use of network capacity could have an impact on other QoS parameters. Thus, the simultaneous consideration of different QoS parameters would enable the definition of more efficient and effectively deployable solutions.
A more specific but equally important point would be to develop more solutions to improve reliability and availability. Indeed, as mentioned in the previous section, work in these areas is still in its early stages. However, these parameters will be essential for future cellular applications and, in particular, URLLC applications. Moreover, the improvement of these parameters through MC opens the way to interesting solutions (optimal RAT selection, efficient packet duplication, mobility management, etc.).
Finally, the issues related to the optimal management of services at the edge of the network (edge computing, slicing) have so far only been addressed in the context of availability improvement. However, due to the back-haul limitations, this optimal positioning of services seems essential in terms of latency, throughput, and reliability. Consequently, the management of services (mobility, positioning) should also be studied for these other parameters.
2. Multi-Connectivity to Improve Energy Efficiency
The improvement of energy efficiency is the second objective associated with the use of multi-connectivity in 5G networks and beyond.
2.1. Existing Solutions to Improve Energy Efficiency
Different research projects have already combined the ideas of energy efficiency improvement and multi-connectivity 
First, some researchers, considering that multi-connectivity could have a negative impact on the energy consumption of cellular networks, have proposed solutions that aimed to improve the energy efficiency of this technology 
. Therefore, the authors of 
have proposed a scheme to optimize both the user-cell association and the energy consumption of the network in an MC scenario. In particular, the authors defined a two-time-scale Lyapunov optimization mechanism enabling the efficient activation and deactivation of small-cell base stations depending on the queue and channel-state information of the UEs. Therefore, this solution aims to optimize both cell activation and user association to reduce energy consumption. The evaluation carried out demonstrates the gains of the solution in terms of energy consumption as well as its viability (low calculation times). However, two major limitations could be identified: the absence of a large-scale evaluation for the computational delays induced by the solution and the fact that the mobility of UEs was not considered (both in the definition of the solution and in the evaluation). Similarly, the authors of 
focused on energy optimization in an MC scenario. However, the approach considered was completely different. Indeed, the proposed solution was based on three emerging technologies in the energy area: Energy Harvesting (EH), Energy Sharing (ES), and Wireless Power Transfer (WPT). The authors considered base stations powered by solar panels and introduced a Smart Energy Management (SEM) module that aimed to associate energy management and load balancing. Thus, they defined an optimization problem by taking into account the UE requirements and the amount of energy available for the small cell BS. This solution, combining two areas (network and energy), offered interesting perspectives. However, the solution defined was simple and could be optimized (delays, performance). Moreover, the architecture considered, powered mainly by solar panels, seems unrealistic and should have supported alternative options to ensure the proper functioning of the network.
On the contrary, other research studies have argued that multi-connectivity could improve energy efficiency in 5G and future networks 
. Therefore, the authors of 
sought to demonstrate how MC could improve the overall energy efficiency of 5G networks by decoupling UL and DL UE association. To do so, they first formulated the MC-based user–cell association optimization problem and proved that it was NP-hard. Then, they introduced an algorithm that aimed to solve the optimization problem in polynomial time by combining the LP Relaxation-Rounding (LPR-R) and Generalized Assignment Problem (GAP) heuristics. The evaluation provided by the authors confirmed the benefits of this solution in terms of performance (throughput) and energy consumption. However, this solution did not take into account the number of handovers between base stations, which is generally considered an important factor in energy consumption. Furthermore, beyond the split between UL and DL flows, it would have been interesting to look at the split between different applications (Network Slicing) by taking into account their respective requirements. The authors of 
also attempted to demonstrate that MC could improve the energy efficiency of a network. To do this, the authors first defined different scenarios, a first scenario where MC did not offer any gains in terms of bandwidth (lower bound), a second scenario where the gain was weak (low performance bound), and a third scenario where the gain was important (upper bound). Then, they introduced five common algorithms to ensure an efficient user–cell association, that is, the max bitrate, max signal-to-interference-plus-noise ratio, max bitrate-energy efficiency, max clustered bitrate, and analytic hierarchy process. The experiments carried out in this entry demonstrated that these different MC algorithms could guarantee reduced energy consumption compared to a simple connectivity scenario. However, as this entry focused on the energy consumption assessment, it would have been interesting to take into account solutions optimizing energy consumption in a simple connectivity case and also to measure the impact of multi-connectivity at the UE level.
Finally, some research projects have focused on the definition of an optimal multi-connectivity solution that solves the trade-off between performance and energy consumption 
. Therefore, the objective of the solution proposed by the authors of 
was twofold and focused on guaranteeing acceptable delays while minimizing energy consumption by considering two main assumptions: (1) delays are induced by back haul and (2) small-cell BSs consume less energy than macro-cell BSs. Thus, they formulated a multi-objective optimization problem enabling them to meet both objectives through an estimation of the capacity of the base stations (small-/macro-cell) and the requirements (latency/bandwidth) of the users. The authors then tried to demonstrate the benefits of their solution but the scenario considered was too simplistic. Indeed, only one macro-cell BS and a limited number of interference-free small-cell BSs were deployed. Thus, inter-macro cell mobility and a more extensive study of the solution are necessary. The objective of the authors of 
was very similar. Indeed, they also tried to optimize the global consumption of the network while taking into account the delays involved in inter-base station communications. However, the solution proposed here (
) was more advanced than the solution described in 
. Indeed, the authors took into account the mobility of users between small-cell BSs and proposed an accurate estimation of the BSs’ load level based on an efficient exchange of information between UEs and BSs. Thus, this solution could offer an effective trade-off between performance and energy efficiency. However, it could be interesting to assess the impact (latency, energy consumption) of the communications defined for the estimation of the base station load, both at the UE and base station level. The solution proposed by the authors of 
also focused on the definition of a trade-off between performance and energy efficiency. Similar to the authors of 
, the authors of 
considered the use of solar panels to power small-cell BSs. However, in this paper, an important element that was not addressed in 
was taken into account, that is, the possible performance degradation associated with the use of solar-panel-powered BS. Therefore, in this article, the user–cell association was not only dealing with energy consumption but also with the performance required by users. To enable this, a layered algorithm using the Karush–Kuhn–Tucker (KKT) conditions and providing a joint optimization of the traffic scheduling and power allocation was introduced. The proposed solution seems to offer interesting performances but could go further by considering a dynamic management of the energy produced by the solar panels (variation according to time and demand) and by taking into account the idea of energy sharing introduced in 
2.2. Discussion Regarding Energy Efficiency
Different papers have already addressed jointly the ideas of energy consumption and multi-connectivity. As a result, some papers have sought to reduce the impact of multi-connectivity 
, whereas others have demonstrated its benefits 
or attempted to trade off performance and energy efficiency 
. These papers, considering different approaches and different tools (cf. Table 2
), seem to be an interesting starting point for further research in this area. However, several limitations can be identified:
Table 2. Comparison of existing solutions to improve energy efficiency.
First, it might be very interesting to look at the use of multiple RATs in this context. Indeed, existing papers have focused on the use of cellular technologies, whereas other technologies (WiFi, Bluetooth, etc.) could potentially reduce the overall energy consumption of the network thanks to multi-connectivity 
. Performance evaluation and definition of new mechanisms for these heterogeneous architectures with the objective of energy consumption reduction would appear to be a relevant topic of study.
In the same way, it would be interesting to determine the optimal positioning of the master and secondary nodes for MC scenarios, in the case of both pure cellular and heterogeneous networks. Indeed, as noted by the authors of 
, an efficient architecture could lead to a significant reduction in network consumption and there are many possibilities for optimization in this area as multi-connectivity is an emerging concept.
Finally, in line with the proposition introduced in 
, it could be interesting to look at the use of software/centralized approaches in this context. Indeed, C-RAN architectures (II.B.3) represent the future of cellular network management and could, perhaps, offer better management of mobility and load balancing. Consequently, it might be interesting to apply this idea to energy management.
3. Multi-Connectivity to Improve Fairness
Fair use of the network is the third objective associated with the use of multi-connectivity in 5G and future networks.
3.1. Existing Solutions to Improve Fairness
Different research projects have combined the ideas of fairness improvement and multi-connectivity: 
For all these papers, the underlying idea is the same: improving fairness in user access while maximizing the total throughput utility. To achieve this, the proposed solutions are based on a widely used scheduling algorithm, the Proportional Fair (PF) algorithm, which aims to maximize the per-user throughput’s logarithmic sum (PF metric, 
). However, this scheduling algorithm was designed for simple associations between UEs and base stations and not for a multi-connectivity environment. Moreover, UE association in the MC context has been proven to be NP-hard 
. Therefore, each entry in the literature proposed an evolution of the PF algorithm that aimed to guarantee high performance and fairness in this context.
The solution introduced in 
integrated two main improvements compared to the standard PF algorithm. First, the authors considered that the PF metric was, in an MC case, equal to the average throughput received by the UE from all the network nodes it was simultaneously connected to. They also considered that a centralized controller calculates the average throughput of all users, whereas a distributed approach is generally used by the PF algorithm. Then they defined three heuristic algorithms that should enable the association of UEs in a multi-connectivity context. These algorithms that require controller intervention, aimed to offload traffic from macro-cell base stations to small-/femto-cell base stations and to define for each UE, thanks to a stable matching scheme, two associations with base stations offering a maximum Reference Signal Received Power (RSRP). This system presented two main limitations. First, it was based on a non-standardized architecture that could be complex to deploy. Second, the additional costs (latency) potentially associated with this architecture were not evaluated.
The authors of 
, focused on LTE-WLAN aggregation and developed a similar solution. The main interest of this paper was the fact that different deployment scenarios were considered (ideal and non-ideal back haul). This ensured a high level of performance in realistic scenarios. The defined algorithm was simple and reproduced a “Water-Filling” mechanism. The amount of data transmitted by the UE to the macro-cell base station was proportional to the UE peak capacity ratio for this macro-cell base station and the throughput of this UE for the small-cell base station/WiFi AP. The main limitation of this approach was that it was only reactive and therefore did not predict UE behavior, inducing latency. Moreover, the deployment of this solution within the reference architecture was not studied and this integration could be complex.
The solution proposed in 
, introduced a new constraint compared to the scenarios considered in 
The approach defined in 
was another solution proposed in the literature that aimed to improve fairness using multi-connectivity. To solve the non-convex cell-association problem, the authors, considering a distributed architecture, introduced a matching game algorithm with externalities, enabling it to reach a local optimum. Moreover, to mitigate interference and improve fairness among users, they also defined a joint cell-selection and power-control algorithm. This iterative algorithm deals with power control and cell selection separately and sequentially. This solution presented interesting performances. However, it was weakly positioned compared to existing works and the reference architecture. Therefore, it could be complex to deploy. In addition, the authors only considered multi-connectivity between 4G base stations, limiting the scope of application of this method.
3.2. Discussion Regarding Fairness
Research papers dealing with the use of multi-connectivity for improving fairness have introduced different implementable methods 
. These papers, considering different tools (game theory, water-filling, decomposition), as well as different control solutions (centralized, distributed), provide an interesting starting point for future research in this field (cf. Table 3
). However, several limitations can be identified:
Table 3. Comparison of existing solutions for improving fairness.
Even if different control solutions have already been considered, none of the existing works have proposed a comparison between a centralized and a distributed approach. This could help to determine an optimal solution in the MC context. Moreover, for centralized control, it could be interesting to consider the development of solutions based on SDN technology. Indeed, this standardized technology could be an efficient way to globally manage the network and share resources fairly among users.
The idea of fair resource allocation among users has not been considered yet for 5G sliced networks. However, this idea could be interesting and could lead to a more complex model and additional constraints related to the QoS requirements of each of the user slices. Thus, in this Network Slicing context, enriched proposals could be defined.
Although different tools have been considered so far, Artificial Intelligence techniques have not been used by any of the proposals. However, in this volatile multi-connectivity environment that involves the resolution of a complex optimization problem, the evaluation of the performance of such algorithms could be interesting, in particular for systems based on centralized control.
4. Multi-Connectivity to Improve Mobility Management
Efficient mobility management is a fourth objective associated with the use of multi-connectivity in 5G networks and Beyond.
4.1. Existing Solutions to Improve Mobility Management
Different research projects have combined the ideas of mobility management and multi-connectivity: 
One of the first publications on the subject 
, demonstrated the benefits of an MC approach compared to traditional handover solutions. To achieve this, the authors proposed a comparison of the performance of MC and traffic-steering approaches by considering the relevant parameters for mobility, that is, Radio Link Failure (RLF), outage, and throughput. They also studied the impact of the MC system itself, taking into account different instantiation delays for the secondary base station, as well as the presence of small-cell 4G base stations co-located with the 5G base station.
In addition, various papers such as 
focused on the use of the MC approach to improve mobility management. Among these different proposals, the authors of 
proposed the simplest system, which used a deep-learning algorithm called Long Short-Term Memory (LSTM), to determine the mobility patterns of each UE and thus anticipate future movement trends. This information was then used to determine which base station should be used to establish a secondary connection with the UE. This approach provided interesting performance in terms of the throughput and handover prediction accuracy. However, the additional costs (calculation, latency) associated with the application to each UE of a deep learning-based solution and the impact of these additional costs were not evaluated. In 
, a richer model was proposed and three main contributions were introduced. First of all, instead of predicting the mobility of the UEs, the authors suggested measuring the channel quality at different base stations for each UE. As a result, depending on the actual network state, optimal associations were identified. They also recommended the use of a local coordinator to manage traffic between the different cells. Based on DC architecture, they finally defined a simplified handover procedure. This system provided attractive performance but its evaluation was only possible by considering a simulated semi-statistical channel model. Moreover, the integration of such a solution in the standardized DC architecture should be studied. In the same direction, the authors of 
proposed a complete framework for MC primarily designed to limit the number of handovers. The first interesting idea introduced by these authors was to consider the different parameters to manage multi-connectivity, such as the received signal strength, UE velocity, 4G/5G channel states, 4G/5G signal path loss, and BS load. They also showed how their solution could be integrated into the reference architecture. The second important element in this entry was the definition of a handover management mechanism. This mechanism was based on a reward function derived from the parameters introduced above and a Markov decision process. This approach could be an effective way to manage mobility in 5G and future networks. Nevertheless, it might be interesting to compare this proposal with mechanisms based on intelligent approaches that could guarantee greater flexibility and reliability such as 
Finally, the authors of 
looked at the application of multi-connectivity for future sliced 5G networks. More specifically, they proposed an improvement of the 3GPP EN-DC standard based on the LTE-LWA protocol, enabling the coordination of both 4G/5G bases stations and WLAN APs within the radio access network. This solution, completed with an evolved 3GPP UE session establishment call flow, was then used by these authors to define a protocol guaranteeing a higher availability and more efficient mobility management of UE slices. This solution, which extended the use of multi-connectivity to new areas (Network Slicing), had significant value. However, the proposed solution was simple and reactive, as slices were only deployed on user demand. As a result, the performance of such a solution would be limited and unsuitable for low-latency and high-reliability applications.
4.2. Discussion Regarding Mobility Management
Research papers dealing with the use of multi-connectivity for mobility management have demonstrated the benefits of this approach 
and introduced implementable algorithms 
and frameworks 
. These papers, which consider different scenarios (RATs, architecture) as well as different tools (AI, Markov, etc.), provide an interesting starting point for future research in this field (cf. Table 4
). However, several limitations can be identified:
Table 4. Comparison of existing solutions for improving mobility management.
Given the available tools, no framework has proposed a modular solution based on a high-performance predictive approach. This might appear as an interesting idea. Moreover, the level of performance (throughput, delay, packet loss) of such proactive MC systems should be evaluated.
The number of RATs considered in mobility management has so far been limited. Indeed, apart from 
that aimed to integrate WLAN APs, the other papers only focused on cellular RATs. The integration of satellite and WLAN RATs (outside the Network Slicing scope) or even LPWA RATs would therefore seem relevant for confirming the relevance of the MC approach.
In the continuity of 
, it could also be interesting to propose other solutions using MC to manage UE slices. Indeed, this question is currently a key priority 
. To achieve this and more broadly to manage mobility, it might be useful to consider the SDN approach. The technology, which guarantees a high level of flexibility, could be applied to the reduction of delays induced by mobility. Indeed, these delays lead to a considerable decrease in the benefits associated with the multi-connectivity approach 
5. Multi-Connectivity to Improve Spectrum and Interference Management
Efficient spectrum and interference management is the fifth objective associated with the use of multi-connectivity in 5G networks and beyond.
5.1. Existing Solutions to Improve Spectrum and Interference Management
Different research projects have combined the ideas of spectrum and interference management and multi-connectivity: 
For many of these studies, the objective was to determine how interference could be minimized to enable effective multi-connectivity in 5G cellular networks and beyond 
. To achieve this, the authors of 
introduced a user-centric approach called back-haul state-based distributed transmission. The proposed heuristic scheme in a dual-connectivity scenario, enables the UE to autonomously manage power transmission using two main criteria, the current back-haul links’ loads and the communication channels’ quality indicators. As demonstrated by the authors, this energy-efficient solution significantly reduced co-channel interference. However, this approach did not embrace the idea of cooperation between UEs although this could make it more effective. In addition, communication between one UE and two base stations was the only case considered. In the same way, the authors of 
proposed an interesting nonlinear interference cancellation mechanism designed to enable multi-connectivity and higher throughput for IoT gateways. This solution was divided into two main phases. First, the harmonic interference and extra intermodulation components were represented using a Neural Network (NN) model. Then, this model was used by the receiver to extract these components and limit interference. An interesting aspect of this approach was that the authors proposed a hardware implementation demonstrating high performance. Nevertheless, the additional costs associated with the Neural Network (NN) model were not assessed. Moreover, as the authors focused on IoT gateways, the scope of application of such a solution is reduced. Finally, the authors of 
, involved in the European 5G-ALLSTAR project, dealt with cellular-satellite multi-connectivity. In this context, this entry presented three main contributions. First, frequency bands that could potentially be used to integrate satellite communications in 5G networks were identified. Then, a new channel modeling solution enabling the simulation of both satellite and cellular communications using the ray-tracing rendering technique was defined. Finally, three approaches were introduced for interference management using the application of a filtering technique at the transmitter side, an improved beam-forming technique, and a radio resource management algorithm. The approach proposed in these papers was interesting but the integration of such a solution in 5G networks will have to be studied. Moreover, it will also be necessary to ensure coexistence with other 5G RATs.
On the other hand, some studies focused on the potential benefits of multi-connectivity for spectrum and interference management 
. The authors of 
defined a new multi-connectivity architecture based on the selection of local anchors. This architecture was used to enable a soft and efficient handover. In this system, the selected anchors, which corresponded to femto-cell base stations, managed the surrounding cells called virtual cells and the UE–cell association. This multi-layered architecture improved UE mobility management and eliminate inter-cell interference. This user-centric approach showed interesting performance in terms of the EU spectrum efficiency and HOF rate. However, the additional costs associated with such a system, which involved a large number of exchanges between base stations, as well as the possibility of implementing such an architecture must be studied. The authors of 
also investigated the use of virtualization and multi-connectivity to limit interference. In fact, considering that the Network Slicing architecture guaranteed the isolation between the different slices and therefore prevented inter-slices interference, the authors defined a solution based on this architecture. Then, taking into account users’ QoS requirements, they proposed a one-to-many matching game used to solve the user-to-slice association problem. This solution was promising for intra-cell interference management as inter-slice interference was avoided but it showed some significant limitations in terms of spectral efficiency and inter-cell interference management.
5.2. Discussion Regarding Spectrum and Interference Management
Research papers have demonstrated how multi-connectivity operation could be improved thanks to efficient spectrum and interference management 
. In contrast, other papers have proved that multi-connectivity could be useful for spectrum and interference management 
. These papers, which considered different tools (AI, game theory) as well as different control approaches (UE, infrastructure) and different RATs (cellular, satellite), provide an interesting starting point for future research in this field (cf. Table 5
). However, several limitations can be identified:
Table 5. Comparison of existing solutions for improving spectrum and interference management.
Both infrastructure- and user-centric solutions have been considered to more efficiently manage spectrum and interference. However, hybrid approaches that combine EU-level and infrastructure-level decision making are increasingly being used nowadays, in particular, for inter-cell UL interference 
. Therefore, it might be relevant to look at these approaches in a multi-connectivity scenario.
As noticed by the authors of 
, Network Slicing architecture could be an efficient way to limit interference, as inter-slice isolation is ensured. However, in this paper, only intra-cell interference was studied. Therefore, it might be useful to examine the impact of Network Slicing on inter-cell interference management.
Wireless back hauling, combined with an efficient beam-forming technique, is a promising way to minimize interference and increase spectrum efficiency in ultra-dense 5G networks 
. Nevertheless, this idea has not been considered so far in a multi-connectivity scenario and interactions between base stations and multi-layered architecture could serve this idea.