uav deployment strategy.zip_html5

<html xmlns="http://www.w3.org/1999/xhtml"> <head> <meta charset="utf-8"> <meta name="generator" content="pdf2htmlEX"> <meta http-equiv="X-UA-Compatible" content="IE=edge,chrome=1"> <link rel="stylesheet" href="https://static.pudn.com/base/css/base.min.css"> <link rel="stylesheet" href="https://static.pudn.com/base/css/fancy.min.css"> <link rel="stylesheet" href="https://static.pudn.com/prod/directory_preview_static/62527a6774bc5c0105d527af/raw.css"> <script src="https://static.pudn.com/base/js/compatibility.min.js"></script> <script src="https://static.pudn.com/base/js/pdf2htmlEX.min.js"></script> <script> try{ pdf2htmlEX.defaultViewer = new pdf2htmlEX.Viewer({}); }catch(e){} </script> <title></title> </head> <body> <div id="sidebar" style="display: none"> <div id="outline"> </div> </div> <div id="pf1" class="pf w0 h0" data-page-no="1"><div class="pc pc1 w0 h0"><img class="bi x0 y0 w1 h1" alt="" src="https://static.pudn.com/prod/directory_preview_static/62527a6774bc5c0105d527af/bg1.jpg"><div class="t m0 x1 h2 y1 ff1 fs0 fc0 sc0 ls0 ws0">arXiv:1809.04963v1 [cs.IT] 13 Sep 2018</div><div class="t m1 x2 h3 y2 ff2 fs1 fc1 sc0 ls0 ws0">1</div><div class="t m1 x3 h4 y3 ff2 fs2 fc1 sc0 ls0 ws0">Placement Optimization for UAV-Enabled Wireless</div><div class="t m1 x4 h4 y4 ff2 fs2 fc1 sc0 ls0 ws0">Networks with Multi-Hop Backhauls</div><div class="t m1 x5 h5 y5 ff2 fs3 fc1 sc0 ls0 ws0">Peiming Li and Jie Xu</div><div class="t m1 x6 h6 y6 ff2 fs4 fc1 sc0 ls0 ws0">(*)</div><div class="t m1 x7 h7 y7 ff3 fs5 fc1 sc0 ls0 ws0">Abstract—Unmanned aerial vehicles (UAVs) have emerged as</div><div class="t m1 x8 h8 y8 ff4 fs5 fc1 sc0 ls0 ws0">a promising solution to provide wireless data access for ground</div><div class="t m1 x8 h8 y9 ff4 fs5 fc1 sc0 ls0 ws0">users in various applications (e.g., in emergency situations).</div><div class="t m1 x8 h8 ya ff4 fs5 fc1 sc0 ls0 ws0">This paper considers a UAV-enabled wireless network, in which</div><div class="t m1 x8 h8 yb ff4 fs5 fc1 sc0 ls0 ws0">multiple UAVs are deployed as aerial base stations (BSs) to serve</div><div class="t m1 x8 h8 yc ff4 fs5 fc1 sc0 ls0 ws0">users distributed on the ground. Different from prior works that</div><div class="t m1 x8 h8 yd ff4 fs5 fc1 sc0 ls0 ws0">ignore UAVs’ backhaul connections, we practically consider that</div><div class="t m1 x8 h8 ye ff4 fs5 fc1 sc0 ls0 ws0">these UAVs are connected to the core n etwork through a ground</div><div class="t m1 x8 h8 yf ff4 fs5 fc1 sc0 ls0 ws0">gateway node via rate-limited mult i-hop wireless backhauls. We</div><div class="t m1 x8 h8 y10 ff4 fs5 fc1 sc0 ls0 ws0">also consider that the air-to-ground (A2G) access links from UAVs</div><div class="t m1 x8 h8 y11 ff4 fs5 fc1 sc0 ls0 ws0">to users and the air-to-air ( A 2A) backhaul links among UAVs are</div><div class="t m1 x8 h8 y12 ff4 fs5 fc1 sc0 ls0 ws0">operated over orthogonal frequency bands. Under this setup, we</div><div class="t m1 x8 h8 y13 ff4 fs5 fc1 sc0 ls0 ws0">aim to maximize the common (or minimum) throughput among</div><div class="t m1 x8 h8 y14 ff4 fs5 fc1 sc0 ls0 ws0">all the ground users in the downlink of this network subject to</div><div class="t m1 x8 h8 y15 ff4 fs5 fc1 sc0 ls0 ws0">the ﬂow conservation constrain t s at t he UAVs, by optimizing</div><div class="t m1 x8 h8 y16 ff4 fs5 fc1 sc0 ls0 ws0">the UAVs’ deployment locations, jointly with the bandwidth</div><div class="t m1 x8 h8 y17 ff4 fs5 fc1 sc0 ls0 ws0">and power allocation of both the access and backhaul links.</div><div class="t m1 x8 h8 y18 ff4 fs5 fc1 sc0 ls0 ws0">However, the common throughput maximization is a non-convex</div><div class="t m1 x8 h8 y19 ff4 fs5 fc1 sc0 ls0 ws0">optimization problem that is difﬁcult to be solved optimally. To</div><div class="t m1 x8 h8 y1a ff4 fs5 fc1 sc0 ls0 ws0">tackle this issue, we use the techniques of alternating optimization</div><div class="t m1 x8 h8 y1b ff4 fs5 fc1 sc0 ls0 ws0">and successive convex programming (SCP) to obtain a locally</div><div class="t m1 x8 h8 y1c ff4 fs5 fc1 sc0 ls0 ws0">optimal solution. Numerical results show that the proposed design</div><div class="t m1 x8 h8 y1d ff4 fs5 fc1 sc0 ls0 ws0">signiﬁcantly improves the common throughput among all ground</div><div class="t m1 x8 h8 y1e ff4 fs5 fc1 sc0 ls0 ws0">users as compared to other benchmark schemes.</div><div class="t m1 x7 h7 y1f ff3 fs5 fc1 sc0 ls0 ws0">Index Terms—Unmanned aerial vehicle (UAV), wireless net-</div><div class="t m1 x8 h8 y20 ff4 fs5 fc1 sc0 ls0 ws0">works, mult i-hop backhauls, deployment optimization, band-</div><div class="t m1 x8 h8 y21 ff4 fs5 fc1 sc0 ls0 ws0">width and power all ocation.</div><div class="t m1 x9 h9 y22 ff2 fs6 fc1 sc0 ls0 ws0">I. INTRODUCTION</div><div class="t m1 x8 ha y23 ff4 fs8 fc1 sc0 ls0 ws0">U</div><div class="t m1 xa h9 y24 ff2 fs6 fc1 sc0 ls0 ws0">NMANNED aerial vehicles (UAVs), also known as</div><div class="t m1 xa h9 y25 ff2 fs6 fc1 sc0 ls0 ws0">(a.k.a.) drones, have found a wide rang e of applications</div><div class="t m1 x8 h9 y26 ff2 fs6 fc1 sc0 ls0 ws0">in, e.g., cargo delivery, aerial inspe ction, precision agriculture,</div><div class="t m1 x8 h9 y27 ff2 fs6 fc1 sc0 ls0 ws0">and tra fﬁc monitoring. Among others, employing UAVs as</div><div class="t m1 x8 h9 y28 ff2 fs6 fc1 sc0 ls0 ws0">aerial communicatio n platforms to assist terrestrial wireless</div><div class="t m1 x8 h9 y29 ff2 fs6 fc1 sc0 ls0 ws0">communications has recently emerged as one of the key tech-</div><div class="t m1 x8 h9 y2a ff2 fs6 fc1 sc0 ls0 ws0">nologies for the ﬁfth-generation (5G) cellular networks, which</div><div class="t m1 x8 h9 y2b ff2 fs6 fc1 sc0 ls0 ws0">has attracted a lot of interests from both academia and industry</div><div class="t m1 x8 h9 y2c ff2 fs6 fc1 sc0 ls0 ws0">[1, 2]. For example, UAVs can be used as aerial base stations</div><div class="t m1 x8 h9 y2d ff2 fs6 fc1 sc0 ls0 ws0">(BSs) to provide basic w ireless data access for remote areas</div><div class="t m1 x8 h9 y2e ff2 fs6 fc1 sc0 ls0 ws0">and in emergency situations (e.g., after natural disasters), as</div><div class="t m1 x8 h9 y2f ff2 fs6 fc1 sc0 ls0 ws0">well as to enhance the network capacity in terrestrial hot spots</div><div class="t m1 x8 h9 y30 ff2 fs6 fc1 sc0 ls0 ws0">[3–7]. On the other hand, UAVs can be utilized as aerial relays</div><div class="t m1 x8 h9 y31 ff2 fs6 fc1 sc0 ls0 ws0">to help far-apart ground users exchange information [8, 9], and</div><div class="t m1 x8 h9 y32 ff2 fs6 fc1 sc0 ls0 ws0">as access points in the sky for information dissemination and</div><div class="t m1 x8 h9 y33 ff2 fs6 fc1 sc0 ls0 ws0">data collection with ground nodes (e.g., sensors and actuators</div><div class="t m1 xb hb y34 ff2 fs7 fc1 sc0 ls0 ws0">Part of this paper has been submitted to the IEEE International Conference</div><div class="t m1 x8 hb y35 ff2 fs7 fc1 sc0 ls0 ws0">on Communication Systems (ICCS), Chengdu, China, Dec. 19–21, 2018.</div><div class="t m1 xb hb y36 ff2 fs7 fc1 sc0 ls0 ws0">This work was supported in part by the National Natural Science Foundation</div><div class="t m1 x8 hb y37 ff2 fs7 fc1 sc0 ls0 ws0">of China (Project N o. 61871137).</div><div class="t m1 xb hb y38 ff2 fs7 fc1 sc0 ls0 ws0">The authors are with the School of Information Engineering, Guang-</div><div class="t m1 x8 hb y39 ff2 fs7 fc1 sc0 ls0 ws0">dong U niversity of Technology, Guangzhou, 510006, China (e-mail: peimin-</div><div class="t m1 x8 hb y3a ff2 fs7 fc1 sc0 ls0 ws0">glee@outlook.com, jiexu@gdut.edu.cn). J. Xu is the corresponding author.</div><div class="t m1 xc h9 y3b ff2 fs6 fc1 sc0 ls0 ws0">in Internet- of-things (IoT) networks) [10]. Besides w ireless</div><div class="t m1 xc h9 y3c ff2 fs6 fc1 sc0 ls0 ws0">communications, UAVs can also b e used as aerial platforms</div><div class="t m1 xc h9 y3d ff2 fs6 fc1 sc0 ls0 ws0">for wireless power transfer (WPT) [11, 12], wireless powered</div><div class="t m1 xc h9 y3e ff2 fs6 fc1 sc0 ls0 ws0">communication networks (WPCN) [13], and mobile edge com-</div><div class="t m1 xc h9 y3f ff2 fs6 fc1 sc0 ls0 ws0">puting (MEC) [14]. I n the industry, various companies have</div><div class="t m1 xc h9 y40 ff2 fs6 fc1 sc0 ls0 ws0">launched their UAV-assisted wireless communication projects,</div><div class="t m1 xc h9 y41 ff2 fs6 fc1 sc0 ls0 ws0">and some preliminary prototype s include Facebook’s Aquila</div><div class="t m1 xc h9 y42 ff2 fs6 fc1 sc0 ls0 ws0">[15] and Nokia’s ﬂying-cell (F-Cell) [16].</div><div class="t m1 xd h9 y43 ff2 fs6 fc1 sc0 ls0 ws0">As compared to conventional terrestrial wir e le ss communi-</div><div class="t m1 xc h9 y44 ff2 fs6 fc1 sc0 ls0 ws0">cations, UAV-assisted wireless communications have the fol-</div><div class="t m1 xc h9 y45 ff2 fs6 fc1 sc0 ls0 ws0">lowing advantages. First, UAV-enabled aerial comm unication</div><div class="t m1 xc h9 y46 ff2 fs6 fc1 sc0 ls0 ws0">platforms can be quickly deployed on demand, and thus are</div><div class="t m1 xc h9 y47 ff2 fs6 fc1 sc0 ls0 ws0">cost-effective and suitable for emergency scenarios, e.g., when</div><div class="t m1 xc h9 y48 ff2 fs6 fc1 sc0 ls0 ws0">the terrestrial wireless infrastructures are damaged due to</div><div class="t m1 xc h9 y49 ff2 fs6 fc1 sc0 ls0 ws0">natural disasters. Next, the air-to-ground (A2G) wireless chan-</div><div class="t m1 xc h9 y4a ff2 fs6 fc1 sc0 ls0 ws0">nels between UAVs and ground nodes normally have much</div><div class="t m1 xc h9 y4b ff2 fs6 fc1 sc0 ls0 ws0">stronger line-of-sight (LoS) links than conventional ground-</div><div class="t m1 xc h9 y4c ff2 fs6 fc1 sc0 ls0 ws0">to-gro und (G2G) wireless channels; as a result, th e aerial BSs</div><div class="t m1 xc h9 y4d ff2 fs6 fc1 sc0 ls0 ws0">are expected to provide better wireless coverage and higher</div><div class="t m1 xc h9 y4e ff2 fs6 fc1 sc0 ls0 ws0">communication throughpu t than ground BSs. Furthe rmore, due</div><div class="t m1 xc h9 y4f ff2 fs6 fc1 sc0 ls0 ws0">to the fully controllable mobility in three-dimensio nal (3D)</div><div class="t m1 xc h9 y50 ff2 fs6 fc1 sc0 ls0 ws0">airspace, UAVs can adaptively change their locations over time</div><div class="t m1 xc h9 y51 ff2 fs6 fc1 sc0 ls0 ws0">for redu cing the distances with intended ground users, so as</div><div class="t m1 xc h9 y52 ff2 fs6 fc1 sc0 ls0 ws0">to further improve the communication performance.</div><div class="t m1 xd h9 y53 ff2 fs6 fc1 sc0 ls0 ws0">In general, UAV-enabled aerial wireless communica tion</div><div class="t m1 xc h9 y54 ff2 fs6 fc1 sc0 ls0 ws0">platforms can be classiﬁed into two categories depending</div><div class="t m1 xc h9 y55 ff2 fs6 fc1 sc0 ls0 ws0">on whether UAVs are quasi-stationary or fully mobile. For</div><div class="t m1 xc h9 y56 ff2 fs6 fc1 sc0 ls0 ws0">quasi-stationary UAVs, prior works [ 4–7, 9, 17, 18] focused</div><div class="t m1 xc h9 y57 ff2 fs6 fc1 sc0 ls0 ws0">on the optimization of their deployment locatio ns for com-</div><div class="t m1 xc h9 y58 ff2 fs6 fc1 sc0 ls0 ws0">munication performance improvement. For instance, [6] and</div><div class="t m1 xc h9 y59 ff2 fs6 fc1 sc0 ls0 ws0">[17] op timized the two-dimensional (2D)/3D placemen t of</div><div class="t m1 xc h9 y5a ff2 fs6 fc1 sc0 ls0 ws0">UAVs to maximize the network revenue (i.e., the number of</div><div class="t m1 xc h9 y5b ff2 fs6 fc1 sc0 ls0 ws0">users served by UAVs under given quality-of-service (QoS)</div><div class="t m1 xc h9 y5c ff2 fs6 fc1 sc0 ls0 ws0">requirement at each user) and minimize the number o f UAVs</div><div class="t m1 xc h9 y5d ff2 fs6 fc1 sc0 ls0 ws0">for maintaining wireless coverage, respectively. [7] optimized</div><div class="t m1 xc h9 y5e ff2 fs6 fc1 sc0 ls0 ws0">the UAVs’ ﬂying altitude to maximize the communication</div><div class="t m1 xc h9 y5f ff2 fs6 fc1 sc0 ls0 ws0">throughput for multiuser unicasting and multicasting commu-</div><div class="t m1 xc h9 y60 ff2 fs6 fc1 sc0 ls0 ws0">nications b y considering directional antennas at UAVs with</div><div class="t m1 xc h9 y61 ff2 fs6 fc1 sc0 ls0 ws0">adjustable beamwidth. Moreover, [4] and [5] analyzed the</div><div class="t m1 xc h9 y62 ff2 fs6 fc1 sc0 ls0 ws0">average performance of quasi-stationary UAV-enabled wire less</div><div class="t m1 xc h9 y63 ff2 fs6 fc1 sc0 ls0 ws0">networks via the stochastic geometry theory. For fully mobile</div><div class="t m1 xc h9 y64 ff2 fs6 fc1 sc0 ls0 ws0">UAVs, existing works [3, 19, 22] proposed to dynamically</div><div class="t m1 xc h9 y65 ff2 fs6 fc1 sc0 ls0 ws0">control th eir locations over time, a.k.a. trajectories, to impr ove</div><div class="t m1 xc h9 y66 ff2 fs6 fc1 sc0 ls0 ws0">the communication rates with different users at d ifferent time.</div><div class="t m1 xc h9 y67 ff2 fs6 fc1 sc0 ls0 ws0">For instance, [19] and [22] studied the UAV-enabled mobile</div><div class="t m1 xc h9 y68 ff2 fs6 fc1 sc0 ls0 ws0">relaying ne tworks, where a UAV-mounted relay n ode can adap-</div><div class="t m1 xc h9 y69 ff2 fs6 fc1 sc0 ls0 ws0">tively control its trajectory jointly with the wireless resource</div><div class="t m1 xc h9 y6a ff2 fs6 fc1 sc0 ls0 ws0">allocation to maxim iz e the en d-to-end communication rate</div><a class="l" rel='nofollow' onclick='return false;'><div class="d m2"></div></a></div><div class="pi" data-data='{"ctm":[1.568627,0.000000,0.000000,1.568627,0.000000,0.000000]}'></div></div> </body> </html>

<div id="pf2" class="pf w0 h0" data-page-no="2"><div class="pc pc2 w0 h0"><img class="bi x0 y0 w1 h1" alt="" src="https://static.pudn.com/prod/directory_preview_static/62527a6774bc5c0105d527af/bg2.jpg"><div class="t m1 x2 h3 y2 ff2 fs1 fc1 sc0 ls0 ws0">2</div><div class="t m1 x8 h9 y6b ff2 fs6 fc1 sc0 ls0 ws0">from the sourc e node to the destination node. In [3], the</div><div class="t m1 x8 h9 y6c ff2 fs6 fc1 sc0 ls0 ws0">authors proposed a new cyclical multiple access scheme that</div><div class="t m1 x8 h9 y6d ff2 fs6 fc1 sc0 ls0 ws0">schedules the multiuser communication based on the UAV’s</div><div class="t m1 x8 h9 y6e ff2 fs6 fc1 sc0 ls0 ws0">trajectory, in which an interesting throughput-delay tradeoff</div><div class="t m1 x8 h9 y6f ff2 fs6 fc1 sc0 ls0 ws0">is revealed. The authors in [9] and [18] used the machine</div><div class="t m1 x8 h9 y70 ff2 fs6 fc1 sc0 ls0 ws0">learning techniques to construct radio maps for A2G wireless</div><div class="t m1 x8 h9 y71 ff2 fs6 fc1 sc0 ls0 ws0">channels, and accordingly optimized the UAVs’ trajectory for</div><div class="t m1 x8 h9 y72 ff2 fs6 fc1 sc0 ls0 ws0">communication rate maximization. Further more, [1 1] and [12]</div><div class="t m1 x8 h9 y73 ff2 fs6 fc1 sc0 ls0 ws0">optimized the UAV’s trajectory for maximizing the energy</div><div class="t m1 x8 h9 y74 ff2 fs6 fc1 sc0 ls0 ws0">transfer e fﬁciency towards multiple ground n odes in a UAV-</div><div class="t m1 x8 h9 y75 ff2 fs6 fc1 sc0 ls0 ws0">enabled WPT system. [13] jointly optimized the UAV’s tra-</div><div class="t m1 x8 h9 y76 ff2 fs6 fc1 sc0 ls0 ws0">jectory and th e transmission reso urce allocation to maximize</div><div class="t m1 x8 h9 y77 ff2 fs6 fc1 sc0 ls0 ws0">the communication rates of multiple ground nodes in a UAV-</div><div class="t m1 x8 h9 y78 ff2 fs6 fc1 sc0 ls0 ws0">enabled WPCN system, in which the energy consumption at</div><div class="t m1 x8 h9 y79 ff2 fs6 fc1 sc0 ls0 ws0">each node cannot exceed the wir eless energy harvested from</div><div class="t m1 x8 h9 y7a ff2 fs6 fc1 sc0 ls0 ws0">the UAV. It is worth noting that b oth quasi-stationary and</div><div class="t m1 x8 h9 y7b ff2 fs6 fc1 sc0 ls0 ws0">mobile UAVs have advantages and disadvantages. In particular,</div><div class="t m1 x8 h9 y7c ff2 fs6 fc1 sc0 ls0 ws0">mobile UAVs can exploit the UAVs’ fully-controllable m obil-</div><div class="t m1 x8 h9 y7d ff2 fs6 fc1 sc0 ls0 ws0">ity to achieve higher communication performance than quasi-</div><div class="t m1 x8 h9 y7e ff2 fs6 fc1 sc0 ls0 ws0">stationary UAVs, but they need more sophisticated trajectory</div><div class="t m1 x8 h9 y7f ff2 fs6 fc1 sc0 ls0 ws0">control over time w ith non-causal information (e.g. , channel</div><div class="t m1 x8 h9 y80 ff2 fs6 fc1 sc0 ls0 ws0">state information) required in general. By contrast, q uasi-</div><div class="t m1 x8 h9 y81 ff2 fs6 fc1 sc0 ls0 ws0">stationary UAVs only need to a priori determine the UAVs’ de-</div><div class="t m1 x8 h9 y82 ff2 fs6 fc1 sc0 ls0 ws0">ployment locations over a certain period of time, and thus may</div><div class="t m1 x8 h9 y83 ff2 fs6 fc1 sc0 ls0 ws0">have much lower complexity for practical implementation. In</div><div class="t m1 x8 h9 y84 ff2 fs6 fc1 sc0 ls0 ws0">this paper, w e focus our study on the deployment optimization</div><div class="t m1 x8 h9 y85 ff2 fs6 fc1 sc0 ls0 ws0">problem for quasi-stationary UAV-enabled wireless networks.</div><div class="t m1 x7 h9 y86 ff2 fs6 fc1 sc0 ls0 ws0">This paper particularly considers a scenario when multip le</div><div class="t m1 x8 h9 y87 ff2 fs6 fc1 sc0 ls0 ws0">UAVs are employed as quasi-stationary aerial BSs to serve</div><div class="t m1 x8 h9 y88 ff2 fs6 fc1 sc0 ls0 ws0">multiple users distributed on the groun d. In practice, this</div><div class="t m1 x8 h9 y89 ff2 fs6 fc1 sc0 ls0 ws0">network faces various technical challenges. For instance, how</div><div class="t m1 x8 h9 y8a ff2 fs6 fc1 sc0 ls0 ws0">to provide UAVs w ith reliable backhaul con nections to the core</div><div class="t m1 x8 h9 y8b ff2 fs6 fc1 sc0 ls0 ws0">network is a challenging task to be tackled, as UAVs in the</div><div class="t m1 x8 h9 y8c ff2 fs6 fc1 sc0 ls0 ws0">sky are difﬁcult to have wireline backhauls as conventional</div><div class="t m1 x8 h9 y8d ff2 fs6 fc1 sc0 ls0 ws0">terrestrial BSs. In the literature, although there are some</div><div class="t m1 x8 h9 y8e ff2 fs6 fc1 sc0 ls0 ws0">prior works studying the UAV deployment optimization in</div><div class="t m1 x8 h9 y8f ff2 fs6 fc1 sc0 ls0 ws0">UAV-e nabled wireless networks (e. g., [4–7, 9, 17, 1 8]), they</div><div class="t m1 x8 h9 y90 ff2 fs6 fc1 sc0 ls0 ws0">only focused on the A2G access link s between UAVs and</div><div class="t m1 x8 h9 y91 ff2 fs6 fc1 sc0 ls0 ws0">ground users by ignorin g backhau l connections. To ﬁll such a</div><div class="t m1 x8 h9 y92 ff2 fs6 fc1 sc0 ls0 ws0">research gap , we co nsider that different UAVs ar e connected</div><div class="t m1 x8 h9 y93 ff2 fs6 fc1 sc0 ls0 ws0">via wireless multi-hop backhauls with the core network. Under</div><div class="t m1 x8 h9 y94 ff2 fs6 fc1 sc0 ls0 ws0">this setup, we aim to optimize the multiple UAVs’ deployment</div><div class="t m1 x8 h9 y95 ff2 fs6 fc1 sc0 ls0 ws0">locations for network performance optimization, subject to</div><div class="t m1 x8 h9 y96 ff2 fs6 fc1 sc0 ls0 ws0">such rate-constraine d wireless backhauls. This problem, to our</div><div class="t m1 x8 h9 y97 ff2 fs6 fc1 sc0 ls0 ws0">best knowledge, has not been investigated yet. Notice that</div><div class="t m1 x8 h9 y98 ff2 fs6 fc1 sc0 ls0 ws0">there is one related work [23] that used UAVs to provide</div><div class="t m1 x8 h9 y99 ff2 fs6 fc1 sc0 ls0 ws0">multi-hop wireless back hauls for connecting small ground</div><div class="t m1 x8 h9 y9a ff2 fs6 fc1 sc0 ls0 ws0">BSs with the core n etwork, in which the formulation of a</div><div class="t m1 x8 h9 y9b ff2 fs6 fc1 sc0 ls0 ws0">group of UAVs is optimized f or maximizing the utility of</div><div class="t m1 x8 h9 y9c ff2 fs6 fc1 sc0 ls0 ws0">the backhaul network. Different from the p rior work [23],</div><div class="t m1 x8 h9 y9d ff2 fs6 fc1 sc0 ls0 ws0">this paper considers both A2G access links and a ir-to-air</div><div class="t m1 x8 h9 y9e ff2 fs6 fc1 sc0 ls0 ws0">(A2A) backhaul link s in the UAV-enabled wireless ne twork,</div><div class="t m1 x8 h9 y9f ff2 fs6 fc1 sc0 ls0 ws0">for which how to jointly optimize the UAVs’ deployment</div><div class="t m1 x8 h9 ya0 ff2 fs6 fc1 sc0 ls0 ws0">locations and wireless resource allocation (e.g., transmit power</div><div class="t m1 x8 h9 ya1 ff2 fs6 fc1 sc0 ls0 ws0">and bandwidth alloca tion) is still an open problem. This thus</div><div class="t m1 x8 h9 ya2 ff2 fs6 fc1 sc0 ls0 ws0">motivates our investigation in this work.</div><div class="t m1 x7 h9 ya3 ff2 fs6 fc1 sc0 ls0 ws0">For the purpose of exposition, in this pape r we consider</div><div class="t m1 x8 h9 ya4 ff2 fs6 fc1 sc0 ls0 ws0">that the UAVs are connected to the core network through a</div><div class="t m1 xc h9 ya5 ff2 fs6 fc1 sc0 ls0 ws0">ground gateway node via wireless multi-hop backhauls. We</div><div class="t m1 xc h9 ya6 ff2 fs6 fc1 sc0 ls0 ws0">suppose that the A 2G access links from UAVs to users and the</div><div class="t m1 xc h9 ya7 ff2 fs6 fc1 sc0 ls0 ws0">A2A backhaul links among UAVs are operated over orthogonal</div><div class="t m1 xc h9 ya8 ff2 fs6 fc1 sc0 ls0 ws0">frequency band s to avoid severe co-channel interference . Due</div><div class="t m1 xc h9 ya9 ff2 fs6 fc1 sc0 ls0 ws0">to the multi-hop connection, each UAV is subject to the so-</div><div class="t m1 xc h9 yaa ff2 fs6 fc1 sc0 ls0 ws0">called ﬂow conservation constraints, i.e., the total outgoing</div><div class="t m1 xc h9 yab ff2 fs6 fc1 sc0 ls0 ws0">ﬂow from each UAV (to ground users thr ough A2G link s</div><div class="t m1 xc h9 yac ff2 fs6 fc1 sc0 ls0 ws0">or to other UAVs throug h A2A links) should be no larger</div><div class="t m1 xc h9 yad ff2 fs6 fc1 sc0 ls0 ws0">than the tota l incoming ﬂow to that UAV (fr om the gateway</div><div class="t m1 xc h9 yae ff2 fs6 fc1 sc0 ls0 ws0">node or from other UAVs). In this case, the communication</div><div class="t m1 xc h9 yaf ff2 fs6 fc1 sc0 ls0 ws0">rates of those ground users are fundamen ta lly constrained</div><div class="t m1 xc h9 yb0 ff2 fs6 fc1 sc0 ls0 ws0">by the limited rates of A2A backhaul links. T herefore, it</div><div class="t m1 xc h9 yb1 ff2 fs6 fc1 sc0 ls0 ws0">is crucial to properly ba la nce the rate trade off between the</div><div class="t m1 xc h9 yb2 ff2 fs6 fc1 sc0 ls0 ws0">A2G access links versus the A2A backhaul links via efﬁcient</div><div class="t m1 xc h9 yb3 ff2 fs6 fc1 sc0 ls0 ws0">joint UAV deployment optimization and wireless resource</div><div class="t m1 xc h9 yb4 ff2 fs6 fc1 sc0 ls0 ws0">allocation. In particular, we aim to maximize the c ommon</div><div class="t m1 xc h9 yb5 ff2 fs6 fc1 sc0 ls0 ws0">(or minimum) throughput among all grou nd users in the</div><div class="t m1 xc h9 yb6 ff2 fs6 fc1 sc0 ls0 ws0">downlink of this network, by jointly optimizing the UAVs’</div><div class="t m1 xc h9 yb7 ff2 fs6 fc1 sc0 ls0 ws0">deployment locations a nd the wireless power and bandwid th</div><div class="t m1 xc h9 yb8 ff2 fs6 fc1 sc0 ls0 ws0">allocation in both A2A backhaul and A2G access links,</div><div class="t m1 xc h9 yb9 ff2 fs6 fc1 sc0 ls0 ws0">subject to the individual transmit power constraints and th e</div><div class="t m1 xc h9 yba ff2 fs6 fc1 sc0 ls0 ws0">ﬂow c onservation c onstraints at the UAVs, as well as the</div><div class="t m1 xc h9 ybb ff2 fs6 fc1 sc0 ls0 ws0">total bandwid th constraints. However, the common throughput</div><div class="t m1 xc h9 ybc ff2 fs6 fc1 sc0 ls0 ws0">maximization is a non-convex optimization problem that is</div><div class="t m1 xc h9 ybd ff2 fs6 fc1 sc0 ls0 ws0">generally difﬁcult to be solved optimally. To tackle this issue,</div><div class="t m1 xc h9 ybe ff2 fs6 fc1 sc0 ls0 ws0">we employ the techniques of a lternating optimiz ation and</div><div class="t m1 xc h9 ybf ff2 fs6 fc1 sc0 ls0 ws0">successive convex programming (SCP) to optimize the UAVs’</div><div class="t m1 xc h9 yc0 ff2 fs6 fc1 sc0 ls0 ws0">deployment locations and th e wireless resource allocation in</div><div class="t m1 xc h9 yc1 ff2 fs6 fc1 sc0 ls0 ws0">an alternating manner for obtaining a locally optimal solution.</div><div class="t m1 xc h9 yc2 ff2 fs6 fc1 sc0 ls0 ws0">Numerical results show that the proposed design signiﬁcantly</div><div class="t m1 xc h9 yc3 ff2 fs6 fc1 sc0 ls0 ws0">improves the common throughput among all g round users as</div><div class="t m1 xc h9 yc4 ff2 fs6 fc1 sc0 ls0 ws0">compare d to other benchmark schemes.</div><div class="t m1 xd h9 yc5 ff2 fs6 fc1 sc0 ls0 ws0">It is worth noting that with wireless backhauls, the stud-</div><div class="t m1 xc h9 yc6 ff2 fs6 fc1 sc0 ls0 ws0">ied UAV-enabled wireless networks are reminiscent of the</div><div class="t m1 xc h9 yc7 ff2 fs6 fc1 sc0 ls0 ws0">UAV-e nabled relaying networks [8, 9]. However, the prior</div><div class="t m1 xc h9 yc8 ff2 fs6 fc1 sc0 ls0 ws0">works [8, 9] only considered one UAV relay node with two</div><div class="t m1 xc h9 yc9 ff2 fs6 fc1 sc0 ls0 ws0">communication hops. By contrast, this paper considers more</div><div class="t m1 xc h9 yca ff2 fs6 fc1 sc0 ls0 ws0">general multi-hop co nnections for no t o nly wireless backhauls</div><div class="t m1 xc h9 ycb ff2 fs6 fc1 sc0 ls0 ws0">among multiple UAVs but also A2G ac cess links from UAVs to</div><div class="t m1 xc h9 ycc ff2 fs6 fc1 sc0 ls0 ws0">ground user s. Therefore, our proposed design is more suitable</div><div class="t m1 xc h9 ycd ff2 fs6 fc1 sc0 ls0 ws0">for UAV-enabled cellular networks that aim to provide wire le ss</div><div class="t m1 xc h9 yce ff2 fs6 fc1 sc0 ls0 ws0">data ac cess for a wide range of areas with distributed groun d</div><div class="t m1 xc h9 ycf ff2 fs6 fc1 sc0 ls0 ws0">users in e.g. emergency situations.</div><div class="t m1 xd h9 yd0 ff2 fs6 fc1 sc0 ls0 ws0">The r emainder of this paper is organized as follows. Section</div><div class="t m1 xc h9 yd1 ff2 fs6 fc1 sc0 ls0 ws0">II introduces the system model of the UAV-enabled wire-</div><div class="t m1 xc h9 yd2 ff2 fs6 fc1 sc0 ls0 ws0">less networks with multi-hop backhauls, and formulates the</div><div class="t m1 xc h9 yd3 ff2 fs6 fc1 sc0 ls0 ws0">common-throughput maximization problem of our interest.</div><div class="t m1 xc h9 yd4 ff2 fs6 fc1 sc0 ls0 ws0">Section III proposes an efﬁcient algorithm to obtain a locally</div><div class="t m1 xc h9 yd5 ff2 fs6 fc1 sc0 ls0 ws0">optimal solution by applying the techniques of alternating</div><div class="t m1 xc h9 yd6 ff2 fs6 fc1 sc0 ls0 ws0">optimization and SCP. Section IV presents numerical results</div><div class="t m1 xc h9 yd7 ff2 fs6 fc1 sc0 ls0 ws0">to validate the performance of our pr oposed design. Finally,</div><div class="t m1 xc h9 yd8 ff2 fs6 fc1 sc0 ls0 ws0">Section V concludes this paper.</div><div class="t m1 xe h9 yd9 ff2 fs6 fc1 sc0 ls0 ws0">II. SYSTEM MODEL AND PROBLEM FORMULATION</div><div class="t m1 xd h9 yda ff2 fs6 fc1 sc0 ls0 ws0">As shown in Fig. 1, we c onsider a UAV-enabled wireless</div><div class="t m1 xc hc ydb ff2 fs6 fc1 sc0 ls0 ws0">network, in which M UAVs are deployed as aerial BSs to</div><div class="t m1 xc hc ydc ff2 fs6 fc1 sc0 ls0 ws0">serve K ground users, and the UAVs are connec ted to the</div></div><div class="pi" data-data='{"ctm":[1.568627,0.000000,0.000000,1.568627,0.000000,0.000000]}'></div></div>