D2D Multi-hop Multi-path Communications in B5G Networks: A Survey on Models, Techniques, and Applications

In 5G networks, device-to-device (D2D) communications have played an important role in enlarging the coverage, relaxing the workload of backhaul links of both macro base stations (MBSs) and small-cell base stations (SBSs), and serving the mobile users (MUs) local applications and services at high capacity. However, beyond 5G (B5G or 6G) networks will require disruptive solutions that can assist D2D communications to meet numerous advanced applications and services requested by dense MUs. One of the most e ﬃ cient solutions for D2D communications is multi-hop multi-path (MHMP). In this paper, we present a detailed survey of the so called D2D MHMP communications in terms of models, techniques, and applications for B5G networks. We discuss and propose the future research directions of D2D MHMP communications. All the models, techniques, and applications as well as future research directions of D2D MHMP communications provide the useful insights into B5G networks design and optimisation. of the Creative Commons Attribution license ( ), which permits unlimited use, distribution and reproduction in any medium so long as the original work is properly cited.


Introduction
In every generation of mobile networks, the carriers always face the problems of how to provide the proliferation of mobile users (MUs) with high quality of service (QoS) and high resource efficiency. The emerging 5G technologies have been mostly addressing the problems of high data rate, low latency, ubiquitous and massive connectivity, etc., but expensive deployment due to system architecture changes and upgrades. Therefore, the next generation of mobile networks, i.e., beyond 5G (B5G or 6G) networks will place a high priority on more intelligent solutions that can improve the QoS while do not change the system architecture. One of the most * Corresponding author. Email: vonguyenson@duytan.edu.vn promising solutions, which has been well studied in 5G networks and further developed in B5G networks is device-to-device (D2D) communications [1].
D2D communications enable direct transmission between two MUs over one hop in close proximity. This way not only efficiently mitigates the workload at both the macro base stations (MBSs) and the small-cell base stations (SBSs) but also improves system capacity to meet the high data rate of advanced applications and services. However, D2D communications over one hop cannot exploit the contents already cached at the other MUs around in dense B5G networks for multi-hop multi-path (MHMP) offloading, namely D2D MHMP communications [2,3].
In D2D MHMP communications, the local applications and services are fully taken to serve the MUs at higher quality of experience (QoE 1 ) rather than higher QoS traditionally. Thanks to shorter transmission per hop, the D2D MHMP communications reduce the transmission power of each transmitter of a hop. This in turn allows the transmitters to share their power consumption for longer lifetime as well as to mitigate the interference on the other receivers which re-use the same spectrum resource. Importantly, in many applications and services, e.g., emergency scenarios, the D2D MHMP communications expand the coverage of networks from the functional areas to nonfunctional ones, i.e., disaster areas, to serve as many MUs as possible. It is therefore crucial to further study on D2D MHMP communications to make them more powerful for allying to B5G networks.
In this paper, all the aspects of D2D MHMP communications from architectures, models, techniques, to applications are investigated insightfully. They are organised, analysed, and evaluated in a systematic and logical way. Together with the discussions on future research directions, we aim to provide both academic and industrial communities with a full understanding about D2D MHMP communications. This enables us to propose disruptive solutions for D2D MHMP communications to address many challenges raised by the dense MUs with high QoE demand in B5G networks.
The rest of this paper is organised as follows. In Section 2, we introduce an overview of D2D communications and techniques that can be utilised in D2D MHMP communications. Section 3 and Section 4 present the D2D MHMP communications architecture and related models, respectively. The detailed techniques of D2D MHMP communications are shown in Section 5. Section 6 is dedicated to presenting various useful applications of D2D MHMP communications. In Section 7, we discuss and propose some potential research directions of D2D MHMP 1 Unlike QoS, QoE [4] is more visual and sensitive for the MUs to evaluate the system performance. Some metrics of QoE are used including hit rate, latency, jitter, quality fluctuation, etc. communications in B5G networks. Finally, we conclude the paper in Section 8.

Overview of D2D Communications
In traditional mobile networks, all communications must go through the MBSs or the SBSs if better. This scheme is not quite reasonable for B5G in which dense MUs. The MUs often use high-speed data applications and services (e.g., social networks and e-learning, e-heath, video sharing, and games, etc.), and they may be located in close proximity. In this context, the MUs in short distances from each other can perform direct D2D communications which are expected to be much better than longer distances from MBSs or SBSs [1]. To name but a few, the benefits of D2D communications include: increasing the spectrum, energy, and coverage efficiencies [5,6]; high connectivity and low transmission time [7][8][9], and high QoS [10,11].
D2D communications can be classified based on the spectrum use manner as shown in Fig. 1. D2D communications occur in the licensed band of the cellular mobile networks (inband) or in the unlicensed band (outband). For inband communications, recent studies have focused on how to efficiently distribute the spectrum for D2D communications and cellular mobile networks (underlay inband mode). This research problem mainly focuses on minimizing the interference between D2D communications and cellular mobile communications [12][13][14][15][16]. Other solutions for addressing the interference, the authors suggest that a portion of the licensed band of cellular mobile networks can be devoted to D2D communications (overlay inband mode). In this mode, the spectrum allocation is very important so that the licensed band is pre-specified, but not wasted [17].
Meanwhile, the studies in [18][19][20][21] recommend the outband used for D2D communications rather than the inband. In outband communications, the coordination between two radio interfaces can be controlled by the MBSs (controlled outband D2D communications) or between D2D users themselves (autonomous outband D2D communications). It can be observed that D2D outband communications face the challenge of complicated coordination between the two bands because D2D communications usually occurs at the secondary radio interface (e.g., WiFi direct and Bluetooth). Therefore, in this paper, all the D2D MHMP communications are in inband applications and services.
In addition, D2D communications can be divided into three typical categories based on the way they communicate with each other including peer-topeer communications and cooperative communications. Although D2D communications bring many benefits to the network infrastructure and the network core, it is required a disruptive development to become more powerful, and thus can be applied to B5G networks. The promising solution developed for D2D communications is D2D MHMP communications is investigated in detail below.

D2D MHMP Communications Architecture
A typical architecture of D2D MHMP communications in B5G networks is illustrated in Fig. 2. It consists of a core network, MBS, SBSs, unmanned aerial vehicles (UAVs), vehicles, and MUs; hereafter, the vehicles and MUs are called end users (EUs). The B5G networks are able to provide advanced applications and services (ASSs) over multiple communication modes, e.g., from MBS, SBSs, and UAVs to EUs, vehicle-to-vehicle (V2V) communications, direct D2D communications, and D2D MHMP communications. In this architecture, the D2D MHMP communications are established based on the key techniques such as routing, relay selection (RS), caching, downlink spectrum resource sharing (DRS), and energy harvesting (EH). The routing and RS techniques play an important role in finding the best D2D MHMP communications from the source EUs (SUs) to the destination EUs (DUs). In addition, the caching, DRS, and EH techniques are applied to further utilising the storage, spectrum, and ambient energy resources to enhance the performance of D2D MHMP communications in B5G networks. In most of the D2D MHMP communications, inband is often used, and thus the interference may occur if using underlay mode.
It is obvious that the B5G networks can serve the dense EUs numerous advanced ASSs like video streaming, social networks, e-learning, e-health, smart transportation, etc. With D2D MHMP communications assisted, the B5G networks can be more efficiently applied to content delivery, public safety, environment and disaster management, IoT networks, and heterogeneous ad-hoc networks (HANETs). The main benefit of D2D MHMP communications is that it can expand the coverage of networks from functional areas to nonfunctional ones, e.g., in disaster and emergency scenarios [63].

D2D MHMP Communications Models
In this section, we introduce some models which are important and mostly used in D2D MHMP communications. Other ones such as wireless channel, transmission models, and energy consumption models 3 EAI Endorsed Transactions on Industrial Networks and Intelligent Systems 10 2020 -01 2021 | Volume 7 | Issue 25 | e3  [ [64][65][66][67][68][69] are assumed to be well-known and thus they are not considered. The general information of D2D MHMP communications models is shown in Table 1 and Fig. 3. In all systems, the models have to be built first, and then, the system performance can be analysed, optimized, and evaluated. The details of D2D MHMP communications models are discussed as follows.

Capacity Model
Capacity is the crucial metric used to evaluate the system performance of a proposed architecture, technique, or optimisation design. Most of the capacity models are derived originally from the Shannonlike capacity W log 2 (1 + γ m,n ) characterised by the signal-to-noise ratio (SNR) γ m,n = P m,n N 0 or signal-tointerference-plus-noise ratio (SINR) γ m,n = P m,n N 0 +I m,n [2, 3, 13, 29, 31, 33-40, 52-57, 62]. Here, W is the system bandwidth, P m,n and I m,n are the power of the incoming signal of interest and the power of the co-channel interference signal respectively at the receiving node of the n th D2D hop in the m th path, and N 0 is the power of additive white Gaussian noise. In addition, the capacity can be derived from the incoming data traffic at a relay node [30,32,[58][59][60]. In D2D MHMP communications, the capacity of the m th path, m = 1, 2, ..., M, consisting of N m D2D hops, is generally computed as below So, the best path is the one that provides the highest capacity given by It is noted that the above max − min capacity model is equivalent to the min − max inference model given in [13].

Coverage Model
Coverage is another important metric that has been mostly studied to evaluate the performance of D2D MHMP communications. The coverage of D2D MHMP communications is modeled based on the quality of the received signal, i.e., SNR or SINR, at the receiving node of a D2D hop. If the SNR (or SINR) is less than a given threshold (θ), it means that the transmitting node of the D2D hop cannot cover the receiving node [6, 22, 23, 25-28, 52, 54, 55, 62]. The coverage outage probability (COP) of the n th D2D hop in the m th path is computed relying on the SNR (or SINR) represented by γ m,n and the threshold θ, given by where P presents the probability. Next, we have the end-to-end (E2E) COP over H m D2D hops of the m th path, expressed as [24,53,56,61] From Eq. (4), we can select the best path with the lowest COP to guarantee the large coverage of the D2D MHMP communications. 4 EAI Endorsed Transactions on Industrial Networks and Intelligent Systems 10 2020 -01 2021 | Volume 7 | Issue 25 | e3

Delay Model
With D2D MHMP communications assisted, the system capacity and coverage can be improved, but it may cause longer delay due to MH communications and the process of MHMP communications establishment. In D2D MHMP communications, E2E delay is often formulated in many forms to be minimised or limited so as to meet the ultra-low delay of B5G networks. Particularly, in [45,49,57,59,61], the E2E delay is computed based on the time when a packet is transmitted at the SU and the time when it is successfully received at the DU. The E2E delay can be derived from the processing delay per hop, the average waiting delay per hop, and the number of hops [3,46]; the per-hop parameters of coding delay, propagation delay, and transmission delay [47]; or simply transmission delay [60,62]. In some delay models, the E2E delay further includes the MHMP control process [48,50,51,58].

Security Model
Unlike direct D2D communications, D2D MHMP communications are more vulnerable to the eavesdroppers because of more MH transmissions [41]. The secure models for D2D MHMP communications are thus seriously studied. Recently, physical layer security (PLS) has become a promising solution that can efficiently assist the cryptographic technologies done at higher layers of the protocol stack. In [43], an energy efficient secure model has been proposed for D2D MHMP communications. It is known in PLS that the secrecy capacity for the n th D2D hop in the m th path is given by where [x] + = max{x, 0} and C e m,n is the channel capacity of the eavesdropping hop associated with the n th D2D hop in the m th path.
From Eq. (5), by taking into account the secrecy capacity constraint C c , the secrecy outage probability (SOP) of the n th D2D hop in the m th path is given by Then, we have the E2E SOP over H m D2D hops of the m th path, expressed as So far, Eq. (7) allows us to select the best path with the lowest SOP to guarantee the security of the D2D MHMP communications.
Furthermore, in [42], the SOP of a D2D hop can be replaced with the trusted probability to compute the secrecy levels of D2D MHMP communications in a similar vein with the work in [43]. And it is interesting that the PLS combined with cooperative beamforming strategy can improve the secrecy performance of D2D MHMP communications [44].

D2D MHMP Communications Techniques
In this section, we investigate the key techniques that have been used in D2D MHMP communications. These techniques mostly focus on establishing the best D2D MHMP communications by using clustering or/and relay selection assisted routing (CRR) techniques and enhancing the resource efficiency by using DRS, caching, and EH techniques to gain higher system performance. The summary of D2D MHMP communications techniques is shown in Table 2 and Fig.  4. The details are presented as below. 5 EAI Endorsed Transactions on Industrial Networks and Intelligent Systems 10 2020 -01 2021 | Volume 7 | Issue 25 | e3 Quang-Nhat Tran et al.

Techniques for establishing D2D MHMP Communications
The clustering or/and relay selection assisted routing (CRR) techniques enable the EUs to access the contents from the SUs located far away over MHMP communications instead of within one hop over direct D2D communications. Depending on the models of D2D MHMP communications, different CRR techniques have been proposed in the literature to establish a D2D MHMP communications session. Under the assumption that the methods to create D2D MH communications, e.g., frequency reuse, frequency synchronization, crosstier interference mitigation, coded cooperative data exchange, energy efficient relay selection, D2D based clustering scheme, relay selection, bit error rate and fading analysis, etc., within one cell [22,23,25,28,52,65,[69][70][71][76][77][78][79][80] and cooperative D2D communications across multiple cells [64], can be utilised, the CRR techniques are detailed in the sequel.
In [9], the authors proposed a shortest path routing protocol in which all clustering, RS, and routing techniques are used for effective data dissemination, i.e., larger coverage and higher transmission efficiency. To do so, the target areas of dense EUs are first divided into different clusters for effective advertisement dissemination. Then, the routing path selection process is done to find the shortest path from the SU to the DU for effective data dissemination. During the routing path establishment process, the relay selection technique is deployed to find the set of successive forwarders who can relay the advertisement message successfully from the SU to the DU over the shortest path using geographical user information. In addition, the shortest path routing protocol can be simply done by selecting the set of relay nodes to form the path with minimum number of D2D hops [24], low symbol error rate and outage probability [29], max-min capacity [31], or with energy harvesting [36] and secrecy [42] added to enhance the system performance.
For the purpose of energy efficiency, the authors in [63,66] studied an energy-efficient clustering and routing framework. The framework starts with selecting the EUs that have higher remaining energy and computational power to act as the cluster heads (CHs). After that, the other EUs that can communicate with a particular CH is grouped into a cluster. Finally, together with optimal power allocated to the CHs for high energy efficiency, a path optimization problem is formulated and solved for finding the path with minimum end-to-end delay and longer network lifetime. Clustering, RS, and routing techniques have also been studied in [44,54,75]. However, the authors in these works further combined with the beamforming technique [44,75] and EH technique [54] at the relay nodes so as to improve the quality of received signal.

Techniques for Enhancing D2D MHMP Communications Performance
Downlink Spectrum Resource Sharing. As mentioned in Section 2, inband D2D communications have been often studied in the literature because the inband resource can be controlled and distributed easier than the outband resource can do. In inband, a D2D communication between the two EUs is established by reusing the spectrum resource shared by an arbitrary common cellular user (CU), i.e. underlay mode, or by managing the spectrum resource allocation between the D2D communication and the CU, i.e., overlay mode. The former always yields interference caused by the transmission of the D2D communication to the CU due to using the same spectrum resource, meanwhile the later does not [81,82].
In [68], the authors have used multiple modes to exploit both inband and outband for D2D MHMP communications. This way can obviously find the best spectrum resource sharing solution, but with very high complexity of radio interface control and management. Meanwhile, the authors in [69] have one utilised inband for D2D MHMP communications for the ease of radio interface control and management.
Despite the co-channel interference, the DRS in underlay mode has been drawn more studies in D2D MHMP communications thanks to its flexible design depending on only the interference limit of the CUs. In a simple sharing manner, a CU can share its spectrum resource with only one D2D hop [34] with less effect of co-channel interference. However, some works in [2,3,43,52,55,58] allow multiple D2D hops in different paths to reuse the same spectrum resource shared by a CU as long as the signal to interference-plus-noise ratio (SINR) at the CU is greater than or equal to a given threshold.
Caching. In D2D MHMP communications, the contents can be provided by caching technique at the MBS and the SBSs [34]. However, in dense EUs of B5G networks, many contents are already cached in the EUs. The cached contents in the EUs must be exploited to relax the workload of the MBS and the SBSs as well as improve the QoS of the system [2,3,30,46]. However, the caching techniques in [2,3,30,34,46] are all proactive caching in which the best SUs are assigned to serve the DU. Further reactive caching techniques that are more efficient are required to study in the context of D2D MHMP communications. It means that 6 EAI Endorsed Transactions on Industrial Networks and Intelligent Systems 10 2020 -01 2021 | Volume 7 | Issue 25 | e3  [41][42][43]53] the contents are placed in the best SUs in advance, with a certain caching density based on a prediction model of content popularity. And then, whenever a DU requests the content, D2D MHMP D2D communications are immediately established to serve the DU. Energy Harvesting. One of the most important problems of D2D MHMP communications is that many EUs agree to join the data transmission session from the SU to the DU. These EUs have to sacrifice part of the remaining energy for D2D MHMP communications. The promising technique to compensate the EUs for their energy consumption is EH. In particular, in [54,64], a relay EU acts as not only an information transfer (IT) node but also an energy harvesting node (energy harvested from the MBS in functional area) and an energy transfer node (to the next relay node in nonfunctional area). The process of EH and IT can be repeated over a path from the SU to the DU as studied in [36].

D2D MHMP Communications Applications and Services
As we can see from the earlier sections that D2D MHMP communications can efficiently support the MBSs and the SBSs to provide the EUs with many local advanced applications and services. In this section, typical D2D MHMP communications assisted applications and services, i.e., content delivery, public safety, disaster management, and IoT and HANET, as shown in Table  3 and Fig. 5, are presented in the sequel.

Content Delivery
In most studies, D2D MHMP communications are used for local content delivery applications and services (CASs) [2,3]. Conventionally, the MBSs or the SBSs are responsible for cooperatively serving the EUs the CASs. In addition, D2D communications are able to do so by direct transmissions over one hop. It is necessary that D2D MHMP communications are applied to enhance the hit rate of CASs, i.e., the success probability to hit a requested content, and thus provide the EUs with better chances to access the contents. To do so, D2D MHMP communications are equipped with proactive caching techniques to bring the contents closer to the EUs or/and passive caching techniques to exploit the contents already cached in the EUs for serving the others around. As a result, the EUs can feasibly access the contents from MBSs, SBSs, direct D2D communications, and D2D MHMP communications, at the highest QoE.

Public Safety
It is interesting that D2D MHMP communications have been widely applied to B5G networks for public safety purpose [28,38,52,67,69,76,84,85]. In this scenario, 5G networks may be damaged partially or fully, or the MBSs are overloaded due to the huge traffic requested from the EUs for emergency supports and communications [67]. Such public applications often get the emergency forces and local governments involved to ensure the citizens' safety. In order to meet the requirements of emergency communications like low latency, high data traffic exchange, connectivity maintenance, large coverage; routing algorithms [67], relay and clustering techniques [28,38,52,69,76], and even using UAVs [84,85,91]; have been further studied for D2D MHMP communications. As a result, B5G networks with D2D MHMP communications can 7 EAI Endorsed Transactions on Industrial Networks and Intelligent Systems 10 2020 -01 2021 | Volume 7 | Issue 25 | e3 assist the emergency forces and local governments in enhancing the performance of public safety activities.

Disaster Management
In emergency conditions like natural disasters, the numerous studies suggest that D2D MHMP is a key technology that helps to maintain the connection for rescue activities. The main reason is the mobile infrastructure is often destroyed fully or partially when a natural disaster strikes and D2D communications are proper solution for these cases [22,27,88]. The requirement to keep the communications ongoing is a critical matter attracting researches. Due to the damage of mobile networks during disasters, under limited energy and battery life of the mobile nodes, D2D MHMP communications are equipped with some special techniques, such as clustering, energy harvesting, cooperative beamforming [22,27,33,51,54,63,66,75,88]. Besides conventionally using the remaining mobile nodes of network infrastructure, UAVs are also deployed to establish D2D MHMP information exchanged for disaster management [53,85,89,91]. The UAVs can help to replace the role of the damaged MBSs in nonfunctional areas by connecting themselves to the MBSs in functional areas. The D2D MHMP communications are then connected to the UAVs to provide the EUs with useful information that may reduce human lost in natural disasters.

IoT and HANETs
With the proliferation of wireless devices, IoT and HANETs are mentioned frequently in many applications of the fourth industrial revolution. The advantage of D2D communications that allows the mobile nodes to directly communicate with each other plays an important role in IoT networks and HANETs. D2D communications are expanded to D2D MHMP communications for IoT networks by utilising routing algorithms [42,43,74]. The objective is to exploit the dense mobile nodes in IoT and HANETs to minimise the energy consumption. Especially, with the help of UAVs, D2D MHMP communications and IoT networks can be efficiently applied to natural disaster management [53]. However, the most important issue of D2D MHMP communications in IoT networks and HANETs is security. The authors in [41] have listed out all the threats related to data transmission over D2D MHMP communications in IoT networks and HANETs. Meanwhile, few of studies have proposed some methods for security in IoT [41,42].

Research Directions and Open Issues
D2D MHMP communications are considered as the promising solution to relax workload at backhaul link, enlarging the coverage and improving connectivity to deal with demand of users in both the number of connection and QoS. The above researches still have to be studied more to improve the QoE and the flexibility in advanced ASSs. Future research directions may focus on routing, energy efficiency, reactive caching, security, and even with UAVs assisted. In fact, there are some studies about UAVs that have been applied to D2D communications, but not many applied to D2D MHMP communications [53,89].
In routing, selecting paths in D2D MHMP communications in order to increase the reliability of the transmitted data and reduce the latency is an important problem that needs to be studied further. Besides, studying the proper number of hops in routing algorithms is also required to select the path so that latency and energy consumption can be minimized. Related to energy efficiency and reactive caching, only minimizing the energy consumption and deploying proactive caching are not enough. It is more efficient if energy harvesting techniques are added to prolong the lifetime of mobile devices, especially in emergency scenarios, and if reactive caching techniques applied to meet the dense EUs requesting extreme high data rate ASSs in B5G networks.
Importantly, in dense B5G networks, using D2D MHMP communications will still be an important point to be researched in terms of security. The reason is that the data transmitted over D2D MHMP communications is easier to be disclosed. Security for D2D MHMP communications is also a very serious issue. Finally, when D2D MHMP communications applied to emergency scenarios, public safety, transportation systems, and smart cities, the UAVs with optimal trajectories are deployed as intermediate SBSs to support the EUs in quick response without any interruptions.

Conclusion
We have witnessed a great achievement in cellular mobile communications that 5G networks have been launched and equipped with many emerging technologies. In B5G networks, besides studying new disruptive solutions that may change the system architecture and require a high cost, it is important to place a high priority on the ones that do not make the system any changes, but are able to ubiquitously serve the everhigh demand of non-stop proliferation of EUs. In this paper, we have made an insightful survey on D2D MHMP communications -a promising solution for B5G networks. The survey presents throughout not only the models, techniques, and applications but also the analysis of D2D MHMP communications to know how it works, what it benefits, and which open issues can be developed in future research directions. Our work 8 EAI Endorsed Transactions on Industrial Networks and Intelligent Systems 10 2020 -01 2021 | Volume 7 | Issue 25 | e3 D2D Multi-hop Multi-path Communications in B5G Networks: A Survey on Models, Techniques, and Applications can provide the useful insights into D2D MHMP communications design and optimisation to enhance the performance of B5G networks.