Contents Introduction
Unmanned aerial vehicles (UAVs), or drones, can be controlled wirelessly or by flight software. In recent years, UAVs have become more capable, with increased flight range, payload handling, flight stability, and computational power. Depending on their construction, UAVs can have fixed or rotary wings, which are more commonly used commercially due to their lower cost, immediate deployment, and ability to access remote locations. UAVs are used in various fields, including the military [1], agriculture, geolocation, security, transportation, rescue, and communications [2].
UAVs find diverse applications in various fields. In the military sector, they are primarily utilized for surveillance, reconnaissance of unknown areas, espionage in enemy territories, information gathering, and material transport, including coordinated attacks [3], [4], [5]. Conversely, in civilian applications, UAVs play roles in geolocation, detection, and tracking of moving vehicles [6]. They are also instrumental in precision agriculture, performing tasks such as site-specific spraying, identification of water deficiency, and crop irrigation management [7].
Regarding the communication system for UAVs, it can be deployed either in isolation or forming networks. These networks are called FANETs and are part of the new generation of communication technologies, such as 5G and beyond [8]. They can function alongside existing infrastructures, such as mobile networks deploying micro or nano cells to improve communication in areas with poor coverage or shadow zones [9].
The UAV-enabled communications networks face several challenges, including path planning, routing, and placement. Path planning for UAVs involves finding the optimal and shortest route between the source and destination while avoiding obstacles, considering factors like wind speed and direction [10], [11]. Meanwhile, UAV routing protocols aim to ensure packet transmission between UAV nodes despite factors like mobility, changes in network topology, out-of-coverage range, and frequent link disconnection [12]. However, optimizing UAV placement poses a critical challenge in FANETs, as it enhances resource utilization, prolongs network lifetime, and optimizes diverse parameters of the UAV system.
A mathematical model is employed to abstract and represent system features, accommodating varying levels of reality. Different applications necessitate distinct models tailored to their specific parameters. However, introducing additional parameters may lead to NP-hardness, escalating the problem's complexity. Earlier research has offered insights into UAV placement and trajectory optimization in 2D contexts [13], [14]. However, adopting 3D models is prevalent for accurate representation in scenarios closely mirroring network topology, encompassing applications such as UAV-mounted base stations, UAV-IoT setups, UAV flying backhaul, and UAV access point communications [13].
The quantity of primary studies addressing 3D optimal UAV placement in communication systems is rapidly expanding, though their distribution remains scattered. Nonetheless, attempts have been made to systematize this knowledge, e.g., previous research has surveyed UAV placement, concentrating on mobile communication networks [15], [14]. Additionally, specific works have delved into alternative solution strategies, such as artificial intelligence [16]. In contrast, our systematic mapping study (SMS) takes a broader view of wireless applications and adopts a conventional optimization strategy.
To the best of our Knowledge, a comprehensive overview of 3D optimal UAV placement within UAV-enabled communication, encompassing scenarios like base stations, IoT environments, and access points in standalone or FANETs networks, is absent in the current state-of-the-art (SotA). Thus, we conducted an SMS to systematically organize the expanding body of knowledge and enhance the understanding of solutions employed to optimize the 3D placement of UAVs in various network scenarios. Subsequently, a comprehensive examination of potential internal, external, and constructive threats was undertaken to ensure the reliability of the results. Additionally, our investigation seeks to pinpoint research endeavors and areas lacking exploration while consolidating insights into the modeling process, optimization variables, features, and solution approaches harnessed within optimization theory to tackle this problem.
The rest of the paper is organized as follows. Section II provides information on a UAV-enabled communications system, the features used in the formulation problem, and related concepts considered in this study. Section III summarizes the related work and highlights the contributions of our study. Section IV describes the methodology, encompassing the underlying process flow and tasks of the SMS, including the research questions and classification scheme. Section V analyzes the possible risks to the integrity and accuracy of our study. Section VI presents the SMS results and responses to the research questions formulated. Section VII provides a detailed analysis of the trends identified, discusses the findings, and highlights any observed gaps for further investigation. Finally, conclusions are outlined in Section VIII.
Background
This section provides a concise overview of concepts for grasping this study. These encompass calculating optimal UAV placement, the variables involved in the mathematical models governing UAV placement, and other pertinent concepts.
A. UAV-Enabled Communications Systems
A UAV-enabled communications system is a network infrastructure that utilizes UAVs or drones to facilitate wireless communication among different devices or users and transmit data to a conventional network. This system leverages UAVs' mobility and aerial capabilities to extend the reach and coverage of communication networks as depicted in Fig. 1. This communications system serves several essential purposes, including providing services in remote or disaster-stricken areas, supporting emergency response operations, improving coverage in crowded events, and bridging gaps in existing networks [13]. The communication services operating on UAVs or FANETs enable voice, data, and video transmission, ensuring device, sensor, and user connectivity across various applications such as disaster management, public safety, agriculture, infrastructure monitoring, and search and rescue operations.
A UAV-enabled communication system comprises several vital components that enable wireless connectivity. These components may vary depending on the system design and purpose but typically include the user, wireless, and ground segments [14].
The user segment encompasses all wireless devices or users that connect to the UAV device. These users include cell phones, sensor nodes, computers, tablets, and smart devices. The users communicate with the UAV by transmitting data through wireless connections.
The wireless segment primarily consists of UAVs, which can be categorized based on their specific operations and purposes. These categories include base stations, access points, relays, and IoT collector functions. For example, a UAV equipped with the base station function provides cellular connectivity to users, enabling them to access cellular services. Similarly, if the UAV implements the IEEE 802.11 standard, it can offer Wi-Fi services to user devices. Additionally, UAVs can collect data from sensor nodes when using IoT technology. Furthermore, the UAV network can establish a backhaul connection with the traditional network through UAV-relay, ensuring continuous and uninterrupted communication links [17].
The ground segment comprises two main components: i) the Ground Base Station (GBS) and ii) the Ground Control Station (GCS). The GBS encompasses antennas, receivers, and transmitters facilitating information exchange between UAVs and the conventional network. It serves as the interface between the UAV gateway and the ground, enabling data transmission. On the other hand, the GCS acts as the central control center for managing and monitoring UAV operations. It provides operators with a user interface to control UAV flights, allowing them to adjust flight parameters, monitor UAV status, and maintain situational awareness. The GCS ensures efficient and smooth control over the UAVs' operations [14].
UAV networks or FANETs employ various communication topologies, including mesh, cluster, tree, and star. In a mesh topology, UAVs establish direct links with nearby counterparts. The cluster topology forms groups of UAVs with a designated leader. The tree topology features a central UAV as the root, connecting other UAVs in a hierarchical structure. The star topology relies on a central hub UAV for communication among all UAVs. Additionally, FANETs can operate with one or multiple UAVs, either connected to existing infrastructure or as a standalone network. In a standalone FANET, UAVs communicate directly, forming a self-organizing network without external infrastructure such as cellular towers or Wi-Fi access points. Fig. 1 illustrates a UAV-enabled communications system with two scenarios. The first scenario depicts a FANET consisting of multiple UAVs connected to a GBS through a backhaul link. It shows the uplink and downlink connections between the users and UAVs and potential threats like interference. The second scenario pictures a single drone network, which can operate independently or with a link to a GBS. These scenarios represent typical models that involve the modeling of the optimal 3D placement.
B. 2D UAV Communications
Communications based on UAVs have been modeled in 2D, particularly when determining the path planning from the origin to the service area. This approach involves analyzing the coverage area as a plane, allowing obstacles to be avoided. For instance, Wu et al. [18] models the trajectory in 2D for communication networks, while Mardani et al. [19] focuses on maximizing communication quality and ensuring uninterrupted video transmission by employing a 2D model. The study utilizes a gridded graph to model the flight area and subdivides cellular coverage into sections through rectangular tessellation. Similarly Kumar et al. [20] and Cakir et al. [21] present solutions where path planning is modeled in 2D, addressing issues related to quality of service. However, these solutions often overlook the consideration of channel models.
In real-world environments, UAVs operate in 3D, enabling them to maneuver vertically and navigate obstacles, including structures of varying heights, more effectively. This capability allows for a more realistic representation of the UAV communication model, and the signal propagation becomes relevant because of factors such as multipath effects and signal reflections. Furthermore, 3D modeling improves the accuracy and efficiency of communication systems.
C. UAV Channel Models
An accurate channel characterization plays a pivotal role in the performance placement optimization of efficient UAV-enabled communications. Nevertheless, a plethora of challenges are open in UAV channel modeling. For instance, complete channel models that include both large-scale and small-scale effects have yet to be thoroughly explored in the state-of-the-art. Hence, the main features of the two types of aforementioned fading channels for UAV communications are described below.
1) Large Scale Fading
Here, empirically-based propagation channel models emerge from numerous experimental measurements that have been performed to figure out relationships between propagation channel parameters and practical UAV setups [22]. In particular, we highlight two large-scale fading models: the air–to–ground (A2G) and the air–to–air (A2A) propagation models, which are introduced as follows.
a) A2G channel model: The A2G propagation channel focuses on large–scale statistics such as path loss and shadowing. Such a model characterizes the connection between the transmitter UAV and the ground users (please see Fig. 1). The A2G propagation channel is characterized by assuming the probabilistic LoS and NLoS links. Commonly, the path losses, denoted by 
\begin{equation*}
L=\left\lbrace \begin{matrix} d^{\alpha },& \text{LoS}-\text{Link} \\
\eta d^{\alpha },& \text{NLoS}-\text{Link}, \end{matrix}\right. \tag{1}
\end{equation*}
\begin{align*}
P_{\text{LoS}}=\frac{1}{1+A \exp (-B(\theta -C))}, \tag{2}
\end{align*}
\begin{align*}
L_{A2G}=P_{\text{LoS}}d^{\alpha }+\eta (1-P_{\text{LoS}})d^\alpha. \tag{3}
\end{align*}
Finally, the received signal-to-noise-ratio (SNR) for the A2G propagation model can be expressed as [24]
\begin{align*}
\text{SNR}^{\text{Rx}}=\frac{P_{\text{UAV-X}}}{L_{\text{A2G}}N_{\text{A2G}}}, \tag{4}
\end{align*}
b) A2A channel model: The A2A is a propagation model for inter-UAV communications and wireless backhaul links. Unlike A2G links, the A2A model include better and simpler channel features. The main characteristic of the A2A relies on the LoS conditions between UAVs that are required to communicate with each other. Usually, the free space propagation model is applied to model the path loss between two UAVs. In particular, the communication A2A involves connections between UAVs, where a particular UAV serves as a node relay (R), as depicted in Fig. 1. In the A2A model, the received SNR at the UAV can be formulated as [24]
\begin{equation*}
\text{SNR}_{\text{UAV-X/R}}^{\text{Rx}}=\frac{P_{\text{UAV-X/R}}}{L_{\text{A2A}}N_{\text{A2A}}}, \tag{5}
\end{equation*}
In the A2A and A2G models, the operation frequency is an essential concern because propagation features can vary greatly with the frequency band. The majority of commercial UAV communications deployments have been carried out with two typical bands, i.e., 2.4 GHz and 5.8 GHz bands. In these frequencies, tropospheric attenuations are negligible. Conversely, at higher frequency bands, i.e., millimeter-wave (mmWave), the signals suffer path loss and tropospheric attenuations [22]. In this context, a novel three-dimensional (3D) A2G propagation model for mmWave links between the UAV and the ground user was proposed in [25]. Herein, the new model, in addition to considering the typical channel parameters (e.g., distance, propagation angles, path delays), also assumes specific features of mmWave propagation at 28 GHz using ray tracing theory.
2) Small Scale Fading
The amplitude attenuation for LoS and NLoS, as well as the multipath contribution for the NLoS component, are essential for analyzing the small–scale propagation conditions. In this sense, appropriate channel models are necessary for the UAV placement optimization to be more realistic. Hence, next, we discuss several models commonly used for UAV communication in small-scale fading LoS and NLoS links.
a) NLoS fading channel. Rayleigh: is well-known as the baseline channel model of most theoretical research on wireless communications for characterizing only diffuse scattering environments without the existence of a direct specular component, as shown in Fig. 2. In the context of UAV-based communications, this channel model was analytically tested in a pioneering work [26] for multi-carrier relay-based UAV networks. In [27], theoretical studies indicated that multiple-access A2G communications channels follow the Rayleigh model for the UAV heading system. Other related works in [28], [29], [30] found that the amplitude of the signal in the A2G fading experienced Rayleigh on field measurements of channels in UAV systems for rich scattering environments.
Nakagami-
b) LoS fading channel. Rician: is the small-scale model where the random fluctuations of a radio signal transmitted over a wireless channel can be built at the receiver side as a superposition of one dominant specular component plus a group of multipath waves associated with diffuse scattering, as illustrated in Fig. 2. In the context of UAV communications, Rician is appropriate for higher UAV altitude platforms [34] and for the scattered multipath environments in low altitudes with LoS conditions [34]. In particular, in the Rician distribution, the Rician
Rician Shadowed: is a more general fading channel model that, in addition to the Rician channel characteristics, considers that the amplitude of the LoS dominant component experiences a fluctuation due to natural situations in different UAV scenarios [38]. Specifically, the term shadowed arises as a consequence of the human body shadowing due to user movement disturbing the specular LoS component, as exemplified in Fig. 2. Specifically, the term shadowed arises as a consequence of the human body shadowing due to user movement disturbing the specular LoS component, as exemplified in Fig. 2. The work in [39] provided measurement-based characterization of the human body's impact on ultra-low UAV-A2G channels. Here, it was found that in this scenario, the small fading channel includes the average signal strength, shadowing, and Rician
It is worth mentioning that all the works discussed above for both large and small fading have been carried out under the assumption of operating frequencies below 6 GHz. This fact has a lot to do with the small-scale channels considered (e.g., Rayleigh, Rice, Nakagami-
Some efforts have been oriented to formulate more accurate UAV system models by assuming higher operating frequencies, i.e., mmWave band. For instance, Zhu et al. [45] optimize multi-UAV placement to maximize the sum subject to a minimum rate constraint for each user within mmWave bands and assuming ideal beam patterns. The optimization involves fixed UAV altitudes and employs analog and digital beamforming variables. The study focuses solely on LoS paths or air-to-ground links, omitting consideration of NLoS components. The optimization techniques utilized encompass successive convex optimization and combinatorial optimization. Also, Zhu et al. [46] optimize the rate in a full-duplex communication setup, acting as a relay between two ground nodes as a backhaul link in a millimeter wave. The optimization involves UAV placement, power control, and analog beamforming, assuming LoS communication. The authors introduce the alternating interference suppression algorithm to design beamforming and power vector variables. Likewise, Zhu et al. [47] analyze 3D beamforming using a uniform planar array for mmWave in UAV communications, employing a phased uniform planar array to achieve flexible coverage. The authors divide the uniform planar array into sub-arrays, steering them towards adjacent sub-areas to obtain a wide beam, effectively covering the entire rectangular area. Notably, the analysis exclusively considers LoS path loss for mmWave bands.
D. UAV Placement-Based Use Cases
In the SotA, usefulness of 3D UAV placement extends to various domains, including IoT, base stations, relays, and Wi-Fi networks. In the realm of IoT, UAVs can serve as mobile data collectors. UAVs enhance communication networks as base stations by establishing temporary or remote connectivity. As relays, UAVs extend the signal range and connectivity between ground nodes, aiding in remote or challenging environments. Furthermore, UAVs equipped with Wi-Fi capabilities can create flying access points.
E. UAV 3D Optimal Placement Process
In a UAV-enabled communications system, the optimal placement of UAVs is essential for maximizing system performance. The placement determines strategic positions that optimize coverage, capacity, connectivity, and efficiency [13]. While some researchers focus on 2D calculations, disregarding the UAV's height, 3D analysis is not trivial. It incorporates additional features like channel models, interferences, and specific technologies not accounted for in the 2D analysis. Fig. 3 illustrates the overall process of calculating the optimal placement of UAVs. It consists of five stages, further elaborated in the following subsections.
1) Definition of the Objective to Be Optimized
Achieving optimal 3D placement for UAVs requires an optimized objective. It could involve maximizing coverage, minimizing energy consumption, and optimizing power transmission. Clarity in defining the optimization objective is essential for guiding the system formulation problem in the UAV placement realm. Table 1 shows the main objective to be optimized and reported in the state-of-the-art.
2) System Formulation
This stage involves formulating a UAV-enabled communications system consisting of technical features and factors to the system, i.e., the number of UAVs. These features form the basis for formulating the objective function. Different contributions formulate systems using a subset of such features and factors depending on the network topology and the problem being addressed. Table 2 illustrates some state-of-the-art features such as channel model, access technique, and interference technique.
3) Definition of the Optimization Problem
Every feature of the formulated system is mathematically represented by defining variables and constraints. Translating system features into mathematical terms is crucial for shaping the optimization problem. The optimization problem can be categorized based on the linearity of the objective function, the type of variables, and the constraints. In Table 3, we classify optimization problems into various categories and provide details regarding the number of constraints in each category.
4) Solution Strategy Definition
In this stage, various solutions can be employed to solve the optimization problem. It can include both heuristic and exact methods, depending on the complexity of the problem. Table 4 provides an overview of solution strategies and the computational complexity of the algorithms.
5) Simulation Settings
This stage involves configuring and setting the conditions necessary to assess the performance of the chosen solution strategy on the optimization problem. These settings encompass various elements, including the number of simulated UAVs, the operating frequency or technology employed (e.g., LTE, Wi-Fi), and different scenarios that impact the computational time and complexity of the running algorithm.
6) An Example of the State-of-The-Art
Taking Alzenad et al. [49] as a representative example from the literature, it illustrates the sequential stages in addressing the 3D UAV problem, considering its application as a UAV cellular base station. In the initial stage, the objective to be optimized revolves around the number of covered users while minimizing power transmission. The second stage involves system formulation, where characteristics encompass an Air-to-Ground channel model considering LoS and NLoS scenarios, focusing on a single UAV without backhaul links and not specifying access or interference techniques. The third stage categorizes the optimization problem as a Mixed-Integer Non-Linear Programming (MINLP), utilizing altitude and horizontal location as optimization variables, subject to two constraints. Despite its non-polynomial complexity, the exhaustive search algorithm is adopted for the solution strategy. Lastly, the simulation settings entail a UAV simulation at a 2 GHz frequency.
A generic mathematical model to minimize the transmission power begins by determining the achievable data rate
\begin{equation*}
R= W_{i}\cdot log_{2}\left(1+\frac{P_{UAV-X}\cdot G}{L_{A2G}\cdot N_{A2G}} \right), \tag{6}
\end{equation*}
\begin{equation*}
P_{\text{UAV-X},\min} = \left(2^{\frac{\beta }{W_{i}}}-1 \right) \frac{L_{A2G}\cdot N_{A2G}}{G}. \tag{7}
\end{equation*}
Given the initial locations of UAVs, the average total transmit power of the network is the sum of the transmit power of each UAV in (7). Here, the aim is to minimize the total transmit power of UAVs by finding the optimal locations of the UAVs associated with the corresponding altitudes while the data requirement for all the users is maintained. This optimization problem is challenging, so advanced algorithms (e.g., heuristic approaches) are needed, as illustrated in the rest of the paper.
Related Work
Existing surveys in the literature that concentrate on civilian applications. For instance, Shakhatreh et al. [2] primarily addresses mobile communications for 5G, outlining its future research directions. Moreover, it expands the scope by detailing agricultural precision, construction, and infrastructure inspection applications. In contrast, Ghamari et al. [50] extensively defines civilian UAV communications applications, emphasizing routing protocols, 5G networks, FANET networks, channel models, security, and spectrum management. The surveyed applications span a wide range, including uses in natural disasters, IoT, traffic monitoring, and surveillance.
Meanwhile, Zhang et al. [51] survey the advancements in integrating 5G millimeter-wave communications into UAV-assisted networks. This integration facilitates data transmission improvements of around 10 Gbits/s. The survey covers various aspects, including antenna techniques, propagation channel characteristics, access mechanism techniques, configurations, service management, and security. Additionally, the authors delineate challenges such as mmWave beam alignment and switching, dynamic resource allocation, and virtualization. Besides, Zhenyu et al. [52] provide a survey on mmWave beamforming for UAV communications, presenting an overview of relevant mmWave antenna structures. They delve into beamforming techniques for achieving 3D coverage connecting with cellular networks. The survey extends to channel modeling, encompassing the description of propagation characteristics in LoS and NLoS connections within mmWave links. The channel model adopted by the authors employs a combination of deterministic and heuristic approaches. Similary, Zhenyu et al. [53] comprehensively analyze mmWave communication scenarios, covering antenna structure, beam tracking, and channel models. The primary scenarios considered include access point, communication terminal, and backbone link. The analysis is remarkably detailed for relay communication in full duplex, multiple access scenarios, and challenges inherent to communication, such as UAV jitter and the fast movement of UAVs.
To the best of our knowledge, there is a lack of comprehensive literature studies explicitly addressing the optimal 3D placement of UAVs. However, secondary studies touch upon our area of interest as part of the broader trends and challenges in UAV communications. For example, Shahzadi et al. [54] discuss UAV-assisted wireless networks in the context of 5G and beyond, where UAVs are used as base stations; it also states system characteristics, use cases, and communication challenges. This study is not focused on UAV placement but also screened some related topics. Specifically, the authors categorize the UAV problem based on the number of UAVs involved and their specific functions, such as serving as base stations or relays.
Mishra et al. [14] present a survey on cellular-connected UAVs and their applications, covering communication requirements, innovations in 5G network architectures, and briefly mentioning 3D UAV placement. In contrast, our SMS focuses on analyzing UAV placement in the context of mobile communications, particularly on system characteristics, objective functions, and solution strategies associated with UAV placement.
Parvaresh et al. [16] is one of the states of art that resembles our SMS, focusing on deploying drones as 3D base stations applying artificial intelligence. The study describes the challenges of communications and legacy methodologies to compute 3D placement, focusing on AI/MI models. We analyze the features of the communications system that influence the objective function and solution algorithms based on traditional mathematical optimization and heuristic techniques.
The survey by Elnabty et al. [15] is closest to our SMS, focusing on the UAV placement in 3D for 5G networks. It describes the objectives to be maximized and the features involved in communications. Its discussion concerns 5G communications aspects such as artificial intelligence, energy consumption, and the UAV-aided vehicular network. On the contrary, our SMS analyzes the placement of UAVs in 3D for different technologies, additional parameters that form the objective function, the complexity of the problem, and the proposed solution algorithms are analyzed. Due to this, the field of analysis is extended not only to 5G cellular networks but to other emerging technologies, and as time continues, new research appears.
In summary, our SMS stands apart from existing surveys because it focuses on resolving the 3D UAV placement problem, assessing the objective to be optimized, objective functions for different technologies, optimization models used, application scenarios, and describing the solution strategy adopted, notably that our SMS benefits from being more up-to-date and encompassing a larger pool of contributions. We consider that the surveys above and our SMS are complementary and improve the field of view on the aspects that influence calculating the optimal 3D placement of UAVs.
Methodology
The research adheres to the established guidelines for conducting an SMS, which follows a systematic approach to provide an overview of the research area [55], which has also been used in similar studies [56]. The tasks involved in this research were organized into three main stages: planning, conducting, and reporting. Fig. 4 illustrates the implemented process flow for the SMS.
The planning stage started by formulating the scope of the study, including its main goal and research questions (RQs). Subsequently, we developed the SMS protocol based on the RQs, which encompassed the selection of a search strategy and search strings, establishment of exclusion and inclusion criteria for candidate papers, development of a classification scheme and extraction process, determination of visualization methods for the results, and identification of potential threats to validity. The protocol ensured the study's quality by incorporating peer-review validation and adhering to established standards, well-known vocabularies, and taxonomies relevant to the research field.
The conducting stage involved implementing the SMS protocol outlined during the planning stage. The initial tasks, namely paper selection and application of inclusion/exclusion criteria, were performed iteratively to fine-tune the search string and inclusion/exclusion criteria. Once these were validated through testing and peer review, a pool of candidate papers meeting the inclusion/exclusion criteria was obtained. Subsequently, the classification scheme was applied to categorize the subset of resulting papers.
The reporting stage involved analyzing the SMS results, addressing the RQs, identifying gaps and trends, and suggesting directions for future research. Section IV-A provides an overview of the study scope, while Section IV-B presents the paper selection strategy. The inclusion/exclusion procedure is described in detail in Section IV-C. Section IV-D and Section IV-E outline the classification scheme and procedure, respectively.
A. Scope of the Study
This study focuses on the intersection of UAV networks, 3D analysis, and optimal placement. This SMS aims to analyze the current state-of-the-art 3D UAV optimal placement in UAV communications and identify trends and challenges in the field. The study seeks to answer the following research questions:
RQ1. Which objective is to be optimized, and what system features are considered in modeling a 3D UAV placement problem?
Knowing the optimization objectives provides a specific goal and guides decision-making during the optimization process. Understanding the features included in the system models provides insights into how closely they resemble real-world scenarios and identifies potential factors that can be considered to enhance the accuracy and realism of the models.
RQ2. What is the structure of the reported mathematical system models?
Analyzing the structure of the objective function pursued by the models enables us to examine the communication parameters utilized and pertinent aspects of the system, including constraints on the objective function.
RQ3.What are the mathematical strategies used for solving the reported system models?
Investigating the mathematical strategies employed to solve the system models helps researchers understand whether the strategies converge to a single optimal solution or if suboptimal algorithms (heuristic solutions) are utilized. This knowledge sheds light on solution strategies and their computational complexity.
B. Paper Selection Strategy
We employed a database-driven strategy using the Scopus database to retrieve high-quality, peer-reviewed research literature, including journals, conference papers, and letters. The Scopus database is known for its extensive coverage of reputable digital libraries in the field of UAV research, such as IEEE Xplore, Springer Link, and ScienceDirect [57].
The research string was constructed based on the scope of the SMS and utilized well-established standard vocabularies and taxonomies in the research domain. It aimed to retrieve all relevant papers by incorporating the following strategies:
Inclusion of top-level terms like “aerial vehicle”, “air vehicle”, “drone”, “UAV”, and “remotely guided vehicles” to encompass a wide range of related literature.
Utilization of synonyms obtained from taxonomies and vocabularies, including the 2022 IEEE Thesaurus Version 1.02 [58], to ensure comprehensive coverage.
Incorporation of terms related to “3D” and “placement” to capture UAVs' dimension and positioning aspects in the network.
Our search string was the conjunction (AND) of the UAV networks, 3D analysis, and optimal placement. Each was, in turn, represented as a disjunction (OR) of all domain-related terms. Table 5 presents the final search string with the domains and related words. After a two-iteration process, we coin this search string and validate the results of each iteration with a test set of 9 relevant papers provided by an experienced UAV network researcher. After retrieving all related papers, the initial search string was refined by expanding it to include additional terms such as “3D,” “3-D,” and “3 D.” Compound terms were also used, such as “air vehicle,” to avoid retrieving irrelevant papers with isolated words. Once the search string was fully refined, a query was made on July 20, 2022, to search within the title, abstract, and keywords of papers, resulting in 1,997 papers.
C. Inclusion and Exclusion Procedure
As depicted in Fig. 5 filtered candidate papers using an inclusion and exclusion procedure. Some papers retrieved from the search query were outside the scope of our study, so we used both automated and manual processing to address this. The inclusion criteria involved automatic filtering based on Scopus and CADIMA tools, which considered English language papers, duplicated papers removed, and specific document types, such as conference papers, journals, and letters.
After applying the automatic process and obtaining 1898 candidate papers, we manually screened titles, abstracts, and full texts that consisted of the manual process. We aim to report primary contributions at the intersection presented in Section A, supported by the CADIMA tool.
Using CADIMA options, we labeled each paper with one of three values: “Include” for papers falling within the scope of our study, “Exclude” for those not meeting inclusion criteria or having exclusion criteria applied, and “Unclear” for cases where the screener had doubts. Before the primary screening, two researchers conducted a three-iteration pilot with 20 papers to standardize their criteria. After each iteration, they discussed and agreed upon any divergences in the labeled papers. They proceeded to the primary screening stage once they achieved a 95% success rate and Krippendorff's alpha inter-coder reliability coefficient of 0.9, quantifying the level of agreement among coders statistically.
The primary screening consisted of two stages based on titles-abstracts, and full texts. CADIMA provided different sets of papers for each step, and the screeners worked individually. However, in the first stages (title and abstract screening), at least two screeners reviewed 20% of the papers to ensure inter-coder reliability. In contrast, two screeners examined all the papers in the full-text screening. Any divergences in labeled papers were discussed and agreed upon by the team. After screening 1,898 papers in the title-based screening, a pool of 187 candidate papers was marked as included or unclear and passed to the next stage. In the full-text screening, the screeners read deeply, including the title, abstract, introduction, and conclusions. If necessary, they also read sections as formulation problems and results. This process resulted in the selection of 130 papers for the study.
D. Classification Scheme
The classification scheme used in this study was developed by refining existing recognized classifications through an iterative process. An initial scheme based on existing taxonomies in the research community to ensure comparability. From Elnabty et al. [15], we utilized the objectives to be optimized and the system features. Subsequently, the scheme underwent iterative refinement through the incorporation of additional categories. For example, we expanded the options within the interference technique and optimization variable sections to encompass a broader range of possibilities. The full-text screening involved a three-iteration pilot to classify a random sample of 21 papers. Each paper was coded by assigning one or more parameters to each of the five stages, as depicted in Fig. 6. The two researchers independently coded each paper and subsequently discussed any divergences. As needed, the classification scheme was refined during these discussions. Fig. 6 displays the final scheme used to consistently classify the 130 papers, allowing us to address RQ1, RQ2, and RQ3.
Parameter classification scheme for the five stages of the 3D UAV placement problem.
E. Classification Procedure
Utilizing the classification scheme from Subection IV-D, two researchers manually coded the dimensions for each of the 130 papers using CADIMA as a support tool. The coders followed a mandatory and adaptive depth reading approach, covering the title, abstract, introduction, and conclusions before coding. They reviewed the formulation problem and results section to ensure comprehensive coding if needed. The codification was performed in subsets of ten papers, with divergences discussed and agreed upon before proceeding to the following subset. Further details of the classification procedure are provided in the replication package (http://dx.doi.org/10.17632/wfbrvxm7v6.1).
Threats to Validity
In this section, we address potential threats to the validity of this study and describe the measures we have taken to mitigate or minimize them. While we diligently followed the SMS process to enhance the validity of the results and conclusions presented in this paper, we encountered specific challenges at various stages that warrant further discussion.
A. Construct Validity
When devising the paper selection strategy, we encountered two threats to the completeness of the study. Firstly, we addressed whether the database search strategy would retrieve all relevant papers. To mitigate this, we opted for the Scopus database, renowned for its comprehensive coverage of high-quality refereed research literature. Scopus indexes peer-reviewed papers from major digital libraries like IEEE Xplore, Springer Link, Science Direct, and ACM.
Secondly, we tackled the validity threats related to the search string, which could potentially exclude relevant papers if specific keywords were missing. To address this, we took two measures. Firstly, we utilized well-established standards, vocabularies, and taxonomies in the research field. Secondly, we conducted a thorough three-iteration validation process using 20 papers from a senior UAV optimal placement researcher to refine and enhance the search string. The final search string comprised the conjunction of the three keywords (UAV, 3D, and placement).
Though there is room for further improvements, such as incorporating different thesauri and alternative search strategies like snowballing, the substantial number of candidate papers initially identified (1997) provides valuable results and findings, offering a comprehensive overview of the state-of-the-art in 3D UAV optimal placement.
B. Internal Validity
The study faces inherent internal threats in the form of individual researcher bias, which could influence decisions on paper inclusion, classification, and analysis. To mitigate this, we implemented two key actions. Firstly, we standardized the criteria across the research team to ensure a consistent understanding. Secondly, we employed a cross-checking process where at least two researchers reviewed each paper.
To ensure the inclusion and exclusion criteria were applied during the screening procedure, we conducted an iterative pilot involving the research team to validate and establish a shared understanding of the criteria. The primary screening commenced after achieving a 95% success rate and a high Krippendorff alpha coefficient of 0.9.
Additionally, during the initial two stages of the primary screening (title and abstract screening), two researchers reviewed 30% of the papers, and the same approach was applied in the full-text screening to ensure inter-coder reliability. The full-text screening utilized a mandatory and adaptive approach, where the title, abstract, introduction, and conclusions were compulsory reading. These measures were taken to minimize the impact of individual researcher bias and ensure reliable and consistent evaluation throughout the study.
Regarding the classification procedure, we invested significant efforts in constructing a comprehensive and consistent classification scheme for the 130 papers. It involved utilizing existing recognized classifications and iteratively refining them through a three-iteration pilot. Furthermore, a mandatory and adaptive depth reading approach, like full-text screening, was employed during the classification process to ensure sufficient comprehension before classifying a paper, minimizing the risk of misclassification. By adopting these measures, we aimed to establish a robust and reliable classification scheme for the research papers.
When analyzing the results and drawing conclusions, we emphasize the collective findings of the entire research team rather than individual interpretations. The graphs were generated based on the team's classification results, and the observations and trends were derived explicitly from these results, ensuring a comprehensive and collaborative analysis.
C. External Validity
The results and conclusions of our study are specifically valid for UAV 3D optimal placement, focusing on UAV communications models, parameters used in communication, and optimization techniques.
To ensure the validity of our general conclusions despite the lack of consensus, we have made substantial efforts to establish a systematic SMS protocol. This protocol included detailed definitions of research questions, inclusion and exclusion criteria, and a classification scheme to enhance our general conclusions' validity.
Results
This section presents the SMS analysis results to answer our research questions. These findings are based on collective observations and trends identified by the research team, ensuring a comprehensive perspective rather than individual interpretations.
The following subsections address each of our research questions. Section VI-A discusses the optimization objective and the system features to model a 3D UAV optimal placement problem (RQ1). Section VI-B further presents the objective function in terms of concrete variables and constraints to model it mathematically, forming the features mentioned above and the involved complexity of the problem (RQ2). Lastly, Section VI-C outlines the mathematical approaches employed to solve the objective function and the parameters used for simulating the proposed model (RQ3).
A. RQ1. Which Objective is to be Optimized, and What System Features are Considered in Modeling a 3D UAV Placement Problem?
This subsection further details the reported objectives to be optimized and the system features considered in modeling the UAV 3D optimal placement. Many contributions in our study have focused on optimizing multiple objectives.
1) Objective to Be Optimized
The 3D UAV placement problem is generally a multi-objective optimization problem. Thus, several objectives have been targeted in the state-of-the-art depending on the specific application and requirements. As depicted in Fig. 7, we grouped the most common objective functions found in the literature into five major categories, i.e., network performance, service coverage, user experience, system scalability, and resource allocation. This grouping can help better understand the high-level goals pursued when optimizing the 3D UAV placement.
Although we analyzed each category in isolation, these are interdependent, i.e., the objective functions in one category can affect the performance of the objective functions in other categories. For example, optimizing for maximum coverage may require more UAVs, increasing energy consumption and cost, thus affecting the system's scalability. Likewise, optimizing resource efficiency may require compromising coverage, affecting the overall user experience.
Next, each category and objective function addressing the UAV placement problem reported in the stated-of-the-art are further detailed.
a) Network performance: The network performance category encompasses all the objectives to be optimized that pursue increasing the efficiency and effectiveness of the communication network established among the UAVs and other devices. Specifically, data throughput, task completion time, network lifetime, latency, deployment cost, and outage probability are some objectives to be optimized that we found in this study, as depicted in Table 6.
Data throughput
TABLE 6 Objective to Be Optimized for the Network Performance Category and the System Features Considered in the Mathematical Model (SCF= Small-Scale Fading, LSF= Large-Scale Fading, S= Single, M= Multiple)
Data throughput is the transmitted data over a communication channel per unit of time. A vast majority (53.3%) of analyzed works (i.e., [59], [60], [61], [63], [65], [67], [68], [69], [70], [71], [72], [75], [76], [78], [79], [80], [81], [82], [83], [84], [86], [87], [88], [89], [90], [91], [92], [93], [95], [96], [97], [98], [99], [101], [102], [103], [104], [105], [106], [107], [108]) focuses on data throughput. All the studies above aim to maximize data throughput by optimizing the placement of UAVs. On the other hand, some of them also pursue to ensure different secondary objectives. For example, Zhang et al. [96] pursue minimizing interference and maximizing signal strength while maximizing data throughput. Other works instead maximize solely data throughput. For example, Nishant et al. [60] provide on-demand coverage to the ground users to maximize the sum rate.
Maximizing data throughput is particularly relevant in real-time data transfer scenarios such as video streaming, remote monitoring, and teleoperation. Furthermore, it contributes to increasing network capacity, i.e., the number of users or devices the network can support simultaneously without sacrificing performance.
The task completion time
Task completion time is the duration to finish a task in a UAV-based network. By optimizing UAV placement, we can reduce task completion time and enhance overall system efficiency. A 2.3% of contributions (i.e., [62], [64], [73]) focus on optimizing the task completion time. For example, Sujunjie et al. [64] minimize the time required for the UAVs to complete the offloaded task by optimizing flying height and horizontal placement. Minimizing task completion time is particularly relevant in safety-critical and real-time scenarios such as rescue and disaster relief, remote monitoring, and teleoperation.
The network lifetime
The network lifetime is when a network can function efficiently without requiring maintenance or a component replacement. The ultimate goal is to have a UAV-based network operating continuously for an extended period. Only two contributions (i.e., [74], [77]) focus on optimizing the network lifetime. For example, Jiaxun et al. [74] and Malkawi et al. [77] aims to maximize the network lifetime in a scenario where the UAVs or their batteries' lifetimes are limited; such potential impairment could degrade network performance over time. Maximizing the network lifetime is particularly relevant in environmental monitoring and surveillance scenarios.
Latency
The latency or delay refers to the responsiveness and timeliness of communication between UAVs and other devices or systems. In the literature, only two contributions (i.e., [85], [109]) are focused on latency. The first has the objective function of delay among terminal devices, while the second deals with the average latency ratio. Zhang et al. [109] calculated the ratio between the average latency experienced in the network and a predefined latency threshold or target. A ratio of less than 1 indicates that the average latency in the network is below the target, implying good latency performance. Conversely, a ratio greater than 1 indicates potential latency issues. Optimizing UAV placement and network design minimizes delay among terminal devices by strategically locating UAVs, optimizing communication infrastructure, and reducing overall network delay [109]. For this case, the cost function involves minimizing distance, using efficient routing protocols, and deploying additional infrastructure. This ensures efficient and reliable communication, enabling prompt data transmission, supporting real-time applications, and enhancing UAV network user experience and system performance. On the other hand, Huang et al. [85] analyze the delay among terminal devices, which refers to the time it takes for data packets to travel between the source and destination devices in the network.
Outage probability
The outage probability pertains to the likelihood of a communication link or connection encountering an outage or failure, typically measured when the Shannon channel capacity drops below a specified threshold rate. In the literature, only two works (i.e., [66], [94]) delve into the outage probability. For example, the outage probability is employed as an objective function in the study by et al. [66]. In contrast, in the work of Alnagar et al. [94], the focus is on the asymptotic outage probability. The asymptotic outage probability refers to the outage probability in the asymptotic limit as the number of communication links or the duration of communication approaches infinity. Nevertheless, both studies leverage this objective to optimize relay placement in urban scenarios.
Minimizing the outage probability is paramount in scenarios entailing real-time data transfer, mission-critical communications, or safety-critical operations. This minimization provides valuable insights into system reliability, particularly within contexts marked by a high density of UAVs or prolonged operational duration.
Deployment cost
The deployment cost encompasses the expenses associated with setting up and maintaining network components, which include UAVs and their related infrastructure, like ground stations and communication links. It aims to guarantee the network's efficiency, effectiveness, and long-term sustainability by ensuring it is financially feasible. Solely Caillouet et al. [100] research efforts to optimize the deployment cost, ensuring that the number of UAVs in the air at low altitudes close to the base station and their moves are minimal.
It is especially significant in applications where the network is expected to operate for an extended period or needs to be deployed on a large scale. In many scenarios, the deployment cost is a critical factor that must be considered, particularly in resource-constrained environments like developing countries or disaster-stricken regions.
b) Service coverage: The service coverage category encompasses a range of objectives to optimize and expand the coverage area between UAVs and user devices on the ground. In this study, we have identified several specific objectives for optimization, including ground coverage, Power transmission, beam width angle of UAV, and coverage probability, as shown in Table 7. These objectives are further described as follows.
Coverage
Ground coverage refers to the extent or area of the ground that can effectively receive reliable communication signals from UAVs. Ground coverage directly impacts the UAV communication services' quality and effectiveness to ground users, often involving adjusting the placement and configuration of UAVs to maximize the area they can serve while minimizing potential coverage gaps or dead zones. A 14.61% of the contributions (i.e. [100], [110], [111], [112], [115], [121], [122], [123], [125], [126], [128], [129], [131], [132], [133], [134], [135], [140], [142]) focus on optimizing ground coverage or coverage probability, which to maximize the likelihood of successful communication. For example, Pankaj et al. [112] concentrate exclusively on optimizing the coverage probability in mobile UAV networks, considering both vertical and spatial directions. Similarly, Nithin et al. [125] focus solely on optimizing coverage area for aerial access points, determining optimal placement in both vertical and horizontal dimensions. In contrast, the following studies incorporate additional optimization objectives. Yang et al. [115] focus on enhancing cellular network coverage in areas with high population concentrations during specific events, such as concerts, to alleviate traffic congestion. Additionally, Chun-Hung et al. [122] optimize multi-cell coverage probability while considering spectral efficiency in the downlink network.
Power Transmission
Power transmission is the signal strength UAVs use to transmit data to ground users to minimize energy consumption. A 13.84% of the contributions (i.e. [49], [113], [114], [116], [117], [118], [119], [120], [124], [127], [130], [132], [136], [137], [138], [139], [141], [143]) analyzing the power transmission. For example, Alzenad et al. [49] focus on minimizing transmit power to achieve energy efficiency while maximizing the coverage for a specified number of users. They analyze horizontal and vertical optimizations separately. On the other hand, Wu et al. [114] concentrate on maximizing the number of UAVs while simultaneously minimizing transmission power, particularly in scenarios with many UAVs. Additionally, Pan et al. [116] aims to maximize the network's utility function by balancing the number of associated users and the transmission power of UAVs. In contrast, Hazim et al. [113] solely minimize total transmit power to provide wireless coverage for indoor users in a high-rise building.
Beamwidth Angle of the UAV
The beamwidth angle of a UAV pertains to the angular width of its directional antenna beam pattern used for transmission. Only Hashir et al. [137] focus on optimizing the beamwidth angle as an objective. It's worth noting that this work also optimizes the transmission power simultaneously. Less commonly considered aspects are the beamwidth angle to maximize coverage by reducing signal degradation caused by misaligned antennas.
c) System scalabality: The system scalability category encompasses a diverse set of optimization objectives to design a network capable of accommodating a growing user base, efficiently managing resources, optimizing user association strategies, and adapting to evolving demands. Within this SMS, we have identified specific objectives, including maximizing the number of covered ground users, the number of UAVs, user association, and the number of covered aerial users, as depicted in Table 8. These objectives are further detailed below.
Number of Covered Users
The number of covered users refers to ground users receiving communications services from UAVs. The number of covered users is significant in various scenarios, such as during emergency responses to natural disasters. Deploying UAVs in such situations enables emergency communication and surveillance, where the number of covered users directly impacts the number of individuals who can receive vital information, request assistance, and coordinate rescue operations. Additionally, in public events like concerts, festivals, or sports matches, UAVs can be utilized for crowd management, security, and real-time video streaming.
A 20% of the contributions (i.e. [49], [87], [116], [144], [145], [146], [147], [148], [149], [150], [151], [154], [156], [157], [159], [160], [162], [166], [167], [170], [171], [172], [173], [174], [176], [178]) focusing on optimizing the number of covered user. For example, Hayajneh et al. [146] presented a model featuring a single UAV functioning as a base station. They aimed to maximize the number of ground users while ensuring maximum Quality of Service (QoS), considering various frequency bands. On the other hand, Xu et al. [147] focused on maximizing the number of ground users. They took into account available UAVs, the number of frequency bands, and the distribution of ground users. In a different approach, Lai et al. [148] aimed to maximize the number of covered users while adhering to the constraint of providing the minimum required data rates per user. In contrast, Bor-Yaliniz et al. [149] optimized to maximize the network's revenue, a value directly linked to the number of users covered by the UAV.
Number of UAVs
The number of UAVs denotes the quantity of drones deployed within a communication network. Incorporating the number of UAVs into the objective function facilitates efficient resource allocation and cost-effectiveness, especially in resource-constrained environments. Optimizing the quantity of UAVs ensures that the network can operate optimally, avoiding unnecessary redundancy or resource overutilization. This approach enhances the utilization of available resources and promotes a more financially viable network deployment. In this SMS, 12.3% of the contributions (i.e., [75], [92], [100], [114], [117], [152], [153], [155], [158], [161], [163], [165], [168], [169], [175], [177]) focus on optimizing the number of UAVs within the network. For example, Sawalmeh et al. [117] address multiple objectives, aiming to maximize wireless coverage while minimizing the number of UAVs and their power transmission. Similarly, Qin et al. [152] strive to minimize the number of UAVs within coverage constraints to reduce costs. Additionally, Zhang et al. [75] minimize the number of UAVs and maximize coverage rates.
Association
Association refers to assigning UAVs to specific ground users within the network. Notably, Ye et al. [65] are the sole contributors to optimizing UAV association, considering time allocation to maximize throughput. Through the optimization of the association strategy within the objective function, we can achieve efficient UAV allocation to individual users or devices. This allocation can be based on proximity, signal strength, or QoS requirements. This strategic optimization contributes to the effective functioning of the network, enhancing its overall performance.
Number of covered Aerial Users
The number of covered aerial users refers to the number of UAVs providing communication services to other UAVs, assisting ground users. Aerial users can significantly impact the system's scalability. By incorporating the number of covered aerial users in the objective function, it becomes possible to account for the need to accommodate and serve these additional entities within the UAV network. Cherif et al. [164] exclusively focus on the connectivity problem of aerial users when they are solely served by aerial base stations, aiming to maximize the number of covered aerial users while constraining the spectrum-sharing policy.
d) Resource allocation: The resource allocations category involves the distribution of available resources within UAV networks. Within this category, we identify a range of optimization objectives, including minimizing energy consumption, optimizing time allocation, managing transmission time effectively, and maximizing the utilization of the available spectrum.These objectives play a crucial role in guaranteeing UAV communication networks' efficient operation and performance, as illustrated in Table 9 and expounded upon below.
Energy Consumption
Energy consumption refers to the power UAVs and associated equipment used to provide communication services during operation. This category accounts for 5.38% contributions of the SMS contributions (i.e. [66], [179], [180], [182], [183], [184], [185]). For example, You et al. [179] investigate the energy efficiency of UAVs functioning as base stations and strive to minimize the overall energy consumption. They employ tilted antennas for targeted service coverage in specific regions to achieve this. Similarly, Nouri et al. [180] maximize energy efficiency by minimizing the necessary number of UAVs. In contrast, Chen et al. [66] exclusively optimize energy consumption for data collection, complemented by additional objectives like maximizing capacity and minimizing outage probability. Minimizing energy consumption within the objective function becomes critical, given that UAVs typically rely on finite onboard battery reserves. This optimization goal is paramount for extending flight durations, prolonging operational periods, and reducing the necessity for frequent battery replacements or recharging. Such considerations hold particular importance in scenarios where UAVs must operate continuously for extended durations, as seen in missions related to surveillance, monitoring, or search and rescue operations.
Time allocation
Time allocation refers to assigning specific time slots for UAV communications between UAVs and ground users within the network. Time allocation is critical to optimizing the network's performance, ensuring that resources are used efficiently. Effective time allocation strategies can enhance communication quality, reduce latency, and improve the overall efficiency of UAV network operations. In this SMS, only two studies (i.e., [65], [137]) focus on optimizing time allocation. Both of these studies have multiple objectives within their optimization efforts. For example, Muhammad Hasher et al. [137] optimize the transmission time in the uplink communication between ground users and UAVs and the transmission power. On the other hand, Ye et al. [65] optimize the time allocation for both the downlink and uplink communication portions.
Maximize the Spectrum
Maximizing the spectrum involves optimizing the utilization of available frequency bands within the UAV network. This optimization objective focuses on efficiently allocating and managing the radio spectrum across the network's UAVs. Efficient spectrum management is critical for enhancing network capacity and overall performance while meeting wireless communication demands in various scenarios. Solely Sun et al. [181] concentrate on maximizing the spectrum efficiency while optimizing the 3D placement and user association. Their analysis involves a single UAV with communication links to macro base stations, aiming to use the available frequency bands in the UAV network effectively.
e) User experience: The user experience category encompasses objectives aimed at enhancing the quality of service, reliability, and overall user satisfaction when utilizing UAV communication services. Within this SMS, we have identified specific objectives related to user satisfaction and quality of experience, shown in Table 10, which are further elaborated below.
User satisfaction
User satisfaction aims to enhance users' overall contentment and positive experience using UAV communication networks. Only Yu et al. [186] propose an air-ground collaborative deployment strategy to optimize user satisfaction. Three ground users with different bandwidth requirements achieve this objective, and two UAVs with different bandwidths provide coverage. The authors get the users to associate with the UAVs, providing the bandwidth for each user. The objective function helps design UAV networks that provide efficient and reliable communication and ensure user satisfaction and a positive perception of the services offered.
Quality of experience
Quality of Experience (QoE) encompasses the overall user satisfaction and perception when using a UAV network. The primary objective is to prioritize parameters that enhance users' experiences, including minimizing latency, maximizing throughput, and ensuring reliable connectivity. This optimization approach aims to create a more satisfying and efficient user experience within UAV networks. Only one study by Bera et al. [187] focuses on optimizing QoE. This research delves into information delivery within cache-enabled UAVs, specifically analyzing the delay and latency in delivering information through UAV caches by minimizing the delay rate of information delivery.
2) System Features
This section introduces the system features to model a 3D UAV placement problem. These encompass the channel model, access, and interference techniques. Furthermore, we present the findings, analyzing transversally the type of network (i.e., standalone or multi-UAV network) and the presence and absence of a backhaul system when modeling a 3D UAV placement problem.
a) Channel Model: Findings show that two-channel models have been considered to model a 3D UAV placement problem: i) air-to-ground (A2G) for UAV-to-user communication and ii) air-to-air (A2A) for UAV-to-UAV communication. Standalone networks typically use A2G, while A2A suits networks with a backhaul system. For each channel model, a particular type of fading (i.e., large-scale or small-scale) and the presence or absence of line-of-sight (i.e., LoS or NLoS) have been identified. We further detail the reported channel models in the following subsections.
Air to Ground Channel Model (A2G)
Fig. 8 depicts the distribution of analyzed papers that model an A2G channel model. It considers the chosen fading approach, the presence or absence of LoS conditions, and the type of network under investigation.
Distribution of papers modeling an Air-to-Ground (A2G) channel model, categorized by fading approach, LoS or NLoS conditions, and network type.
A vast majority of research efforts (86.9%) focus solely on large-scale fading (first row in Fig. 8). In contrast, around 12.3% of the research delves into large and small-scale fading. However, it's worth noting that only one paper doesn't consider a specific channel model scenario. Conveniently, most large-scale fading studies (67.6%) specifically address LoS and NLoS communications challenges. Additionally, 19.2% of the papers incorporate a Free Space Loss model within the LoS communications context. Considering both LoS and NLoS allows to broaden the range of situations where optimal placement can be determined, including dense urban, urban, and rural areas.
As per LoS communication, a key feature is the assumption or assurance of an unobstructed line of sight between UAVs and users. To ensure that, several analyzed papers, including [62], [63], [88], [164], [179], use antennas with reconfigurable beamforming. Other papers, such as [77], [88], [160], [165], employ mmWave technology, known for its highly directive beam that facilitates clear communication between transmitters and receivers.
The Air-to-Ground channel model is used for NLoS communication and incorporates a sigmoid function based on environmental constants. These constants depend on the type of scenario (i.e., suburban, urban, and high-rise urban) and the angle -aka theta - formed by the Euclidean distance between the UAV and the ground users. Some papers, like [67], [93], [148], express NLoS in terms of theta. Other papers, including [65], [115], [119], [149], [163], [178], [180] describe NLoS in terms of the height (h) and coverage radius (r) in a 2D projection between the UAV and ground-based users. Furthermore, specific scenarios like suburban, urban, dense urban, and high-rise urban environments have determined the optimal theta values for maximizing coverage in [49], which subsequently serves as a reference in other studies like [147] and [144].
Specific channel models have been employed in various communication scenarios to cater to unique requirements. For example, as discussed in [145], the WINNER II channel model analyses the interference scenarios between UAVs and communication towers. Conversely, in [117], an outdoor-to-indoor channel model is employed, allowing UAVs to provide services to users within building environments. The standard log-normal shadowing model is used for LoS and NLoS in [68], [116] to mmWave, while the 3GPP TR 36.828 model is used for cellular networks in [117]. Finally, some studies have adopted channel models with interactive elements, like in [102], where an AtG model with a flexible Nakagami-m distribution is employed.
Air to Air Channel Model
Fig. 9 shows the distribution of analyzed papers modeling an Air-to-Air (A2A) channel model, besides the fading approach. LoS communication and UAV type network, this channel model minimally considers a backhaul link (i.e., a link to transmit user information to a traditional network), thus approximating a more realistic network.
Distribution of papers modeling an Air-to-Air (A2A) channel model, categorized by fading approach, LoS or NLoS conditions, and backhaul link.
As per the fading approach, most contributions (79.23%) consider neither small or large-scale fading to find the optimal placement of UAVs. For example, Adam and Zhang et al. [75], [154] optimized multiple UAVs as 5G base stations using geometric disks and artificial bee colony to position UAVs. Using K-means clustering, Nouri et al. [180] optimized UAV numbers for IoT devices.
In about 19.23% of the papers, the A2A channel model is incorporated along with a backhaul link. This link can exist between a UAV and a ground station [80] or between multiple UAVs acting as relays [78]. In such cases, determining the optimal UAV placement isn't solely influenced by ground users but also by the positions of other UAVs or the ground station. Conversely, only two contributions consider the A2A channel model without a backhaul link when determining the optimal UAV placement. For example, Kang et al. [81] employ the FSP model, and Cherif et al. [164] identify the optimal placement of UAV base stations for covering UAV users, also using the FSP model.
Around 16.15% of the papers have focused on large-scale fading analysis. These studies commonly employ the FSP model, assuming LoS communication without obstacles. Examples of such works include [59], [66], [76], [79], [80], [81], [90], [105], [106], [107], [109], [129], [141], [157], [164], [169], [171]. In contrast, other research, like [104], [120], [130], [139], delves into both LoS and NLoS analyses, considering scenarios where the propagation path is partially obstructed. Additionally, in [72], the channel model estimates parameters based on radio signal strength measurements. Furthermore, [162], [170] utilize established models from cellular networks, such as 3GPP or its variants like 3GPP TR 36.942 in [108].
Only one contribution by Alnagar et al. [94] conducted a small-scale fading analysis using a Rician model for LoS communication. It's worth noting that no works specifically focus on the Rayleigh channel model in this context because researchers were primarily interested in UAV placement with a clear line of sight. Rayleigh fading models are typically used for NLoS communications where there is no direct line of sight between UAVs and ground base stations.
b) Access and interference technique: Regarding access techniques, we found that 31.53% of the analyzed papers have modeled various techniques, including FDMA, OFDMA, TDMA, and NOMA. Specifically, 12.3% of the papers have focused on FDMA (i.e., ([60], [69], [74], [85], [89], [104], [109], [117], [131], [132], [146], [162], [170], [179], [186], [187]) as depicted in Fig. 10. For example, studies such as You et al. [179] and Hayanjneh et al. [146] employed FDMA, where UAVs divide the available bandwidth into sub-bands, with each frequency sub-band transmitting data independently to a ground user. The articles utilizing this technique consider multiple UAVs in the model to mitigate interference among ground users.
Distributions of papers on access versus interference technique and the number of UAVs used in the model.
At 10% of the papers (i.e., [61], [65], [67], [73], [76], [77], [93], [95], [106], [119], [129], [165], [182]) have employed the OFDMA technique. For Example, articles like Ye et al. [67] and Xu et al. [73] focus on OFDMA to serve many ground users. Most papers use multiple UAVs to model when using this technique, as it helps mitigate interference and improves the UAV network capacity.
Only five articles examined the TDMA access technique (i.e., [64], [75], [84], [137], [140]). This technique involves analyzing different time slots and scheduling transmission time to users. For example, Sun et al. [64] employed TDMA to minimize task completion time. In contrast, Valiulahi et al. [84] assumed that UAVs needed to serve their scheduled users during the transmission time following deployment, determining the number of users each UAV could serve.
NOMA technique was modeled in five articles (i.e., [63], [88], [139], [156], [180]), where it was applied to maximize sum rate, energy efficiency and fairness while working within spectrum limitations. For example, Zhu et al. [63] employed MIMO and NOMA to maximize UAV data rates and throughput in downlink transmission. In this case, they used successive interference cancellation as an interference technique for decoding ground users.
As per the interference technique, few research efforts have been led to this realm. Only 7.69% of the papers incorporated an interference analysis in their models (i.e., [61], [80], [81], [86], [95], [142], [153], [164], [169], [180]). For Example, Rahimi et al. [153] and Kang et al. [81] used the frequency reuse method to prevent interference in multi-UAV networks. Furthermore, Namvar et al. [86] used the Nash Bargaining Solution technique to alleviate interference for drones in small cells. Huang et al. citeID27 utilized cognitive communication as an interference method.
A limited number of studies, like [61], [95], [180], have examined access and interference techniques. Among these, two studies focus on applications involving multiple UAVs.
B. RQ2. What is the Structure of the Reported Mathematical System Models?
This subsection reveals the outcomes related to the structure of the objective function, which is defined by the type of problem and the optimization variables, as pursued in the 3D UAV optimal placement problem reported in the SotA. Additionally, the subsection presents our findings regarding the complexity of the problem, and many contributions in our study have focused on using multiple optimization variables.
1) Type of Problem
Objective functions in research can be categorized according to the type of problem. The first category is linear programming (LP). Only Qin et al. [152] employ LP in this study. They use the optimal angle between radios and the altitude of the UAV, resulting in a linear programming objective due to the use of linear variables in their function and constraints. When both binary and other linear variables are present in the function and constraints, the problem can be formulated as mixed-integer linear programming (MILP). At 7.69% of the contributions use this category (i.e., [105], [109], [130], [131], [141], [143], [146], [148], [150], [153]). For instance, Lai et al. [148] ensured LP by employing a predefined altitude for the objective function. To address the variability and provide UAV coverage for individual ground users, they transformed the problem into MILP. Table 11 presents studies with objective functions categorized as LP and NLP, outlining the optimization variables, solution strategies employed, and simulation parameters.
When the objective function or constraints exhibit nonlinearity, they fall into the category of Nonlinear Programming (NLP), which accounts for 34.61% of contributions (e.g., [61], [62], [69], [76], [77], [117], [118]). For instance, Huang et al. [113] and Zhou et al. [118] have nonlinear functions and constraints, primarily because of the nonlinearity present in the channel model. Similarly, Sawalmeh et al. [117] feature linear objective functions but nonlinear power constraints that depend on path loss. Table 12 categorizes the objective functions as NLP, detailing their corresponding solution strategies and simulation settings.
In cases where both objective functions or constraints are nonlinear and binary variables are present, these problems are classified as Mixed Integer Nonlinear Programming (MINLP), encompassing the majority of papers at 56.15% (e.g., [49], [64], [65], [110], [111], [115], [116], [144], [145], [147], [149], [179], [180]). Many of these studies use binary variables for user and device associations with UAVs or to ensure node coverage, introducing nonlinearity primarily through the channel model. The Table 13 displays the optimization variables, solutions, and simulation parameters for objective functions categorized under MINLP.
Fig. 11 illustrates that optimization variables used in different studies are categorized by the type of problem present in the objective function. Most of these studies fall into the MINLP category, with altitude and horizontal optimization variables being the prevalent combination. Notably, 69.23% of the papers that model multi-UAV scenarios are classified under the MINLP category. This complexity arises due to the need to find optimal solutions for multiple UAVs, contributing to the intricacy of the problem.
2) Problem Optimization
Once the system features are defined, the objective function to be max-minimized should be formulated. This formulation entails defining one or multiple variables that will be optimized to achieve the objective of RQ1, which varies depending on the scenario. Also, analyzing the complexity of the objective function and classifying the type of problem play a critical role in finding a mathematical solution for solving this type of problem.
Fig. 11 depicts the optimization variables defined in the analyzed papers. As shown, altitude is the predominant variable (59.23%) used in objective functions. Among these, 33.84% (e.g., [72], [110], [120], [144], [147], [174]) exclusively focus on this optimization variable, while the rest combine multiple optimization variables. Altitude is primarily employed when objectives revolve around ground coverage, data rate, throughput, and energy consumption. Its advantage lies in its capacity to control power transmission, thereby influencing these objective functions. For example, Omri et al. [110] optimize UAV altitude to maximize coverage and meet QoS requirements, while Esrafilian et al. [72] maximize throughput by optimizing altitude for optimal relay placement.
Conversely, bandwidth serves as the dominant variable (6.9%) within objective functions, as evidenced by studies like [73], [93], [105], [107], [108], [130], [131], [146], [184]. For example, Hayajneh et al. [146] optimize the allocation of available bandwidth to UAVs, aiming to maximize the number of served users. Similarly, Shakhatreh et al. [131] focus on maximizing wireless coverage to ground users by optimizing the allocation of bandwidth resources.
Typically, maximizing the objective function in research involves optimizing a single variable. However, approximately 35.84% of the papers delved into utilizing multiple variables, particularly regarding UAV placement. In such cases, the horizontal location is the second most optimized variable (21.53%, e.g., [49], [60], [64], [69], [112], [180]). It's important to note that this variable often operates in conjunction with the altitude variable and is seldom considered independently. For example, in the study of Guo et al. [69], they aim to maximize the sum rate, breaking down the problem into horizontal and altitude optimization problems. Similarly, Babu et al. [125] focus on optimizing the energy efficiency for the multiple aerial access points. However, Chen et al. [66] optimize solely the horizontal variable to maximize the outage probability when the UAV operates as a relay.
C. RQ3. What are the Mathematical Strategies Used for Solving the Reported System Models?
In this subsection, we describe the approach to solving the mathematical objective function and the setting parameters used in the simulation for optimizing UAV placement. Several researchers analyzed more than one solution strategy.
1) Solution Strategy
Analyzed papers have used exact and heuristic algorithms to solve the UAV 3D optimal placement problem. As depicted in Fig. 12, most researchers use heuristic algorithms to achieve polynomial time in problem resolution.
Approximately 23% of papers have employed exact algorithms, primarily when the objective functions are linear, as seen in cases like [100], [141] and [143]. In some contributions, exact algorithms have also been applied to nonlinear objective functions, such as NLP and MINLP. In these cases, mathematical development techniques or solvers have been used. For example, Sivalingam et al. [160] used a standard CVX solver to address a MINLP problem, and Moon et al. [166] employed a Mosek solver. Additionally, Huang [61] utilized the semi-definite relaxation technique to optimize UAV horizontal placement and transmission power. Interestingly, around 3% of the contributions have used exact and heuristic algorithms, e.g., [101], [103], [109], [141]. For instance, Zhang et al. [109] optimized bandwidth allocation and user association using the Karush-Kuhn-Tucker conditions and employed heuristic algorithms to find the optimal 3D placement.
Heuristic algorithms are the predominant choice in most papers (80.76%). These algorithms are primarily used to obtain approximate results while minimizing time complexity. Particle Swarm Optimization (PSO) or its variants are favored by 14.6% of researchers due to their capacity to reduce complexity, solve problems in polynomial time, and provide minimal errors. Notable articles like [113], [120], [121], [123], [146], [155], [158] leverage this technique. For example, Chen et al. [167] utilize Multi-PSO, an optimized PSO algorithm. Ozdag et al. [173] employ Optimal Fitness-value Search Approaches at Continuous-data range (OFSAC-PSO), built on PSO principles.
Another significant group of studies (5.38%), including [73], [80], [83], [84], [156] employ successive convex optimization, which solves the problem by breaking it down into a sequence of simpler problems, sometimes even incorporating non-convex constraints. Also, some studies like [60], [66], [89], [115], [116], [149] use Bisections research or their variant Golden section search or line search in polynomial time to resolve the MINLP. Additionally, certain studies, such as [60], [66], [89], [115], [116], [149], utilize Bisection research or their variants, like the Golden Section Search or Line Search. These techniques resolve MINLP problems efficiently within polynomial time.
Fig. 12 reveals that most researchers used heuristic algorithms achieve polynomial time in problem resolution. For example, paper [156] compares two algorithms and demonstrates which one operates in polynomial time. It's worth noting that approximately 17.69% of the contributions do not report the complexity of the algorithms they employ.
2) Simulation Settings
After formulating the objective function and selecting the solution strategy, multiple simulation settings, like frequency, wireless technology, and the number of simulated UAVS, need to be laid down.
Fig. 13 shows that 50% of the papers selected 2 GHz for their simulations, often related to cellular networks as a use case. The selection of this frequency range also impacts the choice of channel models, which are typically designed for this frequency. For instance, Xu et al. [147] employed a carrier frequency of 1950 MHz to simulate another cellular network band. Some studies related to 5G, such as Mohammed et al. [151] and Shen et al. [105], operated within the 3 to 3.5 GHz range. Another 5.34% (i.e., [66], [79], [110], [115], [149], [175], [178]) focused on simulations at 2.5GHz, associated with LTE bands. For instance, Chen et al. [66] combined it with 3GPP wireless technology.
On the other hand, some papers like Zu et al. [73] and Yi et al. [80] used 5 GHz for relaying purposes, while Mayor et al. [168] employed the 2.4 GHz frequency, typical of Wi-Fi wireless technologies. Additionally, certain papers ventured into the mm-wave frequency range, spanning from 30 GHz to 60 GHz. Notable examples within this range include [77], [88], [89], [122], [160], [165].
As per wireless technology, some researchers typically perform simulations and report results without specifying a particular wireless technology (83.84%). Instead, the focus is primarily on carrier frequency. Other researchers identify concrete wireless technology, such as cellular technologies like LTE and 3GPP, as evidenced in papers [161], [172], [174]. Wi-Fi at 2.4 GHZ is also employed in some papers such as [124], [125], [131], [158], [168]. We also found that Free Space Optics technology is used [103].
Analyzed papers have also considered the number of UAVs as part of simulation settings. Determining the number of users and UAVs to be simulated in the network is crucial to assessing an algorithm's applicability in various scenarios. This information provides insights into the scalability and processing time required to find a solution. Among the analyzed articles, 54.61% primarily focus on simulating a network with a single UAV providing user service. An additional 30% of the papers involve simulations featuring 2 to 10 UAVs giving service interestingly, while only five articles consider simulations incorporating more than 30 UAVs, e.g., [114], [155], [161], [163], [185].
The articles generally do not specify the simulation tools used for the objective functions. However, the most commonly used is Matlab software, which is utilized in papers [70], [89], [94], [109], [117], [134], [141], [146], [155], [167], [172], [173]. Additionally, solutions are implemented using Java and Python programming languages in [100].
Discussion
The analysis of optimal 3D placement for UAVs presents a promising avenue for research. We observed that the earliest study on this topic was published in 2016. There has been a growing interest, particularly in the last four years, during which 85% of the publications have centered around this subject, as depicted in the Fig. 14.
The growing attention to this problem is attributed to technological advances in UAV capabilities, efficient resource optimization, and the rising demand for improved connectivity. This research approach involves five key stages: defining the optimization objective, considering system features, formulating the optimization problem, developing solution strategies, and setting up the simulation parameters as depicted in Fig. 15.
UAV networks serve various use cases, and current research trends focus on optimizing their placement within cellular networks, as depicted in Fig. 16. These trends are predominantly motivated by the growing need for improved connectivity and their alignment with the evolution of cellular networks, particularly in the context of emerging standards like 5G. However, the increasing research focus on optimizing UAV placement in IoT contexts is notable. It can be attributed to the proliferation of connected devices and the agile deployment of UAVs in remote and hard-to-reach areas. These factors enable rapid network expansion, addressing the scalability and coverage demands of IoT applications.
A. Currents Trends
Current research trends in UAV network optimization prioritize objectives related to network performance, coverage, and system scalability, often aligned with the demands of emerging technologies. These objectives typically involve enhancing data transmission rates, accommodating more users, and extending network coverage. They are analyzed separately. Nonetheless, limited research explores multifaceted objectives, such as maximizing the number of users and coverage while enhancing transmission rates.
Research focuses predominantly on channel models for communication between UAVs and users, emphasizing large-scale fading models applicable to LoS and NLoS communications. Additionally, the system's characteristics include an analysis of access techniques, which, in turn, find indirect application in interference-free model environments, notably exemplified by the FDMA technique.
The nonlinearity due to the A2G channel model contributes to problem complexity and further escalates the complexity when the researchers analyze the small-scale fading analysis, which influences the optimal placement and the optimization objective, especially in the quality of service. Furthermore, ensuring that a UAV serves a ground user necessitates the inclusion of binary variables in the model. Researchers' prevalent approach to addressing such challenges is the utilization of heuristic algorithms that operate in polynomial time and often employ a strategy of breaking the objective function into more minor optimization problems that are solved iteratively, which offers a path to achieving solutions within polynomial time and reducing simulation duration instead of just solving the main problem.
The operating frequency is the primary input parameter in simulation settings, aligning with each UAV's use case (Fig. 16). Most simulations use frequencies around 2 GHz related to cellular networks, employing A2G channel models proposed by [48]. When higher frequencies like mmWave are simulated, the researchers use the FSL model assuming LoS communication. However, more complex stochastic models, like the Saleh-Valenzuela (S-V) channel models [188] or those propped in [189], are often employed to account for factors such as small-scale fading, large-scale fading, and channel characteristics that need more attention in UAV mmWave communications.
Finally, 3D UAV optimal placement holds significant potential to enhance UAV network performance. However, real-world optimization of these networks remains challenging, as it requires considering criteria not covered in existing research. Several challenges have been identified during this study, and they demand increased focus in future research to fully harness the benefits of 3D optimal placement in UAV networks.
B. Open Issues and Research Challenges
This section explores outstanding issues and challenges that promote new research in this field.
Further research is needed to analyze the quality of service (QoS) and deployment cost in UAV networks, especially in post-disaster scenarios: Currently, there is a scarcity of studies focused on these optimization objectives. The oversight of QoS considerations in the network poses the risk of suboptimal performance and an unsatisfactory user experience, particularly in critical real-time applications like video surveillance and medical emergencies. Additionally, there is often a neglect of exhaustive analysis regarding the implementation costs of the network, particularly evident in large-scale deployments. The omission of network costs extends beyond infrastructure and energy analysis, neglecting replacement and maintenance challenges. Continuing research in these areas is essential for ensuring the sustainability of UAV networks over time, particularly in the context of real-time applications where optimal QoS and cost-effective deployments are imperative.
The analysis of outage probability and spectrum optimization objectives still requires further research to complement existing cellular networks by providing additional capabilities to hotspot areas: Few studies have concentrated on these crucial optimization objectives. Neglecting the outage probability undermines the network's reliability, introducing instability in communications. Additionally, overlooking spectrum optimization results in a diminished transmission capacity. Hence, continuing research in these areas decreases the likelihood of data loss during transmission and enhances the efficient utilization of the essential spectrum in UAV networks characterized by multiple devices and diverse applications.
Integrating UAV communication systems in IoT networks facilitates infrastructure monitoring and data collection, e.g., in linear topologies: While a substantial portion of research focuses on addressing the 3D location problem within cellular-based networks, there is a notable gap regarding IoT models. Overlooking UAV communication models in the context of IoT neglects research opportunities, particularly in data acquisition for areas lacking coverage, especially in the context of massive sensors for environmental monitoring. As the growth trajectory of IoT networks accelerates, propelled by the increasing mobility of UAVs, there is a pressing need to extend research efforts into UAV communications for IoT networks. This integration promises to address data acquisition challenges, such as optimizing energy use in communications since getting closer to the data source reduces the need for devices to transmit signals over long distances.
Access and interference techniques are necessary in UAV communications models for 5G systems and beyond, in addition to the drone's massive deployment in FANETS networks: Regrettably, few researchers have simultaneously investigated both access and interference techniques within their UAV communication models. The oversight of these parameters can result in a network's inability to enhance its capacity, leading to unreliable communications affected by interference from UAVs and heterogeneous networks. Consequently, it is crucial to integrate these techniques to enhance connection quality, diminish the likelihood of data loss, and improve spectral efficiency.
Small-scale fading analysis to find the optimal 3D UAV location is needed for future research on dense urban environments, particularly on mmWave communications: Most channel models analyzed refer to large-scale fading using AtG path loss. However, small-scale analyses of LoS and NLoS communications are scarce, not considering impacts on signal quality and the design in the location of each UAV within the network in real scenarios. Thus, it is crucial for upcoming research to integrate small-scale fading analysis for both A2G and A2A communication, utilizing models such as generalized fading. This approach enhances the accuracy of placement calculations and provides a comprehensive understanding of signal quality for ground users.
Expanding 3D analysis in UAVs and ground user placement holds great potential for enhancing cellular networks in cities situated on hills or mountains for future research: Articles typically assume ground users to be on a horizontal plane for the UAV placement calculations. However, real-world scenarios often feature non-planar configurations, such as IoT sensors deployed on uneven terrain. Neglecting this consideration can result in inaccurate outcomes in practical environments. Consequently, future research should incorporate 3D analysis for UAVs and ground users, acknowledging the impact of distance on mathematical modeling.
Optimal placement for heterogeneous UAV communications requires analysis for further research, particularly in smart cities: Research addresses problems specific to certain technologies, such as cellular networks. However, failing to consider this diversity results in suboptimal resource utilization, given the array of devices employing different technologies. This oversight becomes particularly pronounced in variable and dynamic scenarios. Consequently, future research endeavors should delve into UAV communications encompassing various technology stacks, emphasizing heterogeneous communications.
Further research is necessary to advance the simulation of massive UAVs, especially for urban planning and mapping applications: While existing models propose solutions for UAV placement, simulations often involve only a limited number of UAVs. This limitation hinders the comprehensive assessment of system scalability, resource optimization, and network performance. Consequently, there is a need for additional research to tackle scenarios involving a high number of UAVs. This research should encompass the analysis of inherent challenges such as collision avoidance, interference management, and others, along with a thorough evaluation of network performance.
Enhancing the level of abstraction in UAV communication systems is crucial for their practical application in both civil and military environments, particularly in achieving mission customization: Many papers discussing system maturity exhibit diverse maturity levels and often assume specific characteristics that may not align with realistic models. Neglecting system parameters hinders the reflection of real-world conditions and compromises the accurate assessment of performance, signal quality, network latency, and overall reliability. Consequently, future work in this domain should strive for a level of abstraction that ensures the effectiveness of communication system design and enables informed decision-making during the implementation phase.
Conclusion
In this SMS, a comprehensive review was conducted on 130 papers from the Scopus database, out of the 1997 publications, to address the objectives outlined in the study and respond to the three research questions laid out during this investigation. The outcomes of this SMS encompass identifying the elements utilized for modeling UAV communications, classifying optimization objectives, and assessing the methods or algorithms employed to determine optimal placement.
This study serves as a valuable resource for the research community by offering a comprehensive overview of previous research endeavors, highlighting prevalent trends, and pinpointing critical knowledge gaps about the optimal placement of UAVs in communication systems. These considerations encompass optimization objectives, transmission features, algorithmic analyses, and use case scenarios. Unlike existing literature primarily focusing on 2D settings, our research deliberately delves into 3D path planning solutions, offering insight into recent developments.
Our findings underscore a strong focus on optimizing data rate and throughput. Still, they also unveil significant gaps in addressing crucial objectives like deployment costs, quality of experience, and spectrum optimization. Furthermore, our research underscores the importance of delving into system formulation aspects, including small-scale fading, access techniques, and interference management. Also, our study indicates that heuristic algorithms are the prevalent solution strategy, with limited research on simulation settings for large-scale UAV deployments.
Our future work aims to delve deeper into the three-dimensional analysis of UAVs. This will encompass considerations of quality of service, in-depth exploration of small-scale channel models, and adaptations to three-dimensional scenarios involving ground users. By embarking on this research direction, we anticipate contributing to the ongoing advancement of UAV-enabled communication networks.