wireless snsor network.zip_C++_wireless_源码/资料下载-pudn.com

<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/625058e674bc5c01055e33f8/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/625058e674bc5c01055e33f8/bg1.jpg"><div class="t m0 x1 h2 y1 ff1 fs0 fc0 sc0 ls0 ws0">IEEE INTERNET OF THINGS JOURNAL, VOL. 7, NO. 1, JANUARY 2020 505</div><div class="t m0 x2 h3 y2 ff1 fs1 fc0 sc0 ls2 ws0">Mobility Predictions for IoT Devices Using</div><div class="t m0 x3 h3 y3 ff1 fs1 fc0 sc0 ls3 ws0">Gated Recurrent Unit Network</div><div class="t m0 x4 h4 y4 ff1 fs2 fc0 sc0 ls4 ws0">Abebe Belay Adege , Hsin-Piao Lin, and Li-Chun Wang , Fellow, IEEE</div><div class="t m0 x5 h5 y5 ff3 fs3 fc0 sc0 ls6 ws0">Abstract—Wireless and mobile technologies, as well as their</div><div class="t m0 x1 h5 y6 ff4 fs3 fc0 sc0 ls6 ws0">users, are growing rapidly as the Internet of Things (IoT) prod-</div><div class="t m0 x1 h5 y7 ff4 fs3 fc0 sc0 ls2 ws0">ucts, such as sensor-network technologies, mobile devices, and</div><div class="t m0 x1 h5 y8 ff4 fs3 fc0 sc0 ls2 ws0">supporting applications, become widely dispersed. Owing to the</div><div class="t m0 x1 h5 y9 ff4 fs3 fc0 sc0 ls2 ws0">dynamic changes in the wireless networks and the exponential</div><div class="t m0 x1 h5 ya ff4 fs3 fc0 sc0 ls2 ws0">growth of the IoT products, which make it difﬁcult to locate</div><div class="t m0 x1 h5 yb ff4 fs3 fc0 sc0 ls2 ws0">large quantities of users and devices, providing accurate tracking</div><div class="t m0 x1 h5 yc ff4 fs3 fc0 sc0 ls2 ws0">and trajectory predictions in open and highly condensed wireless</div><div class="t m0 x1 h5 yd ff4 fs3 fc0 sc0 ls2 ws0">networks is extremely difﬁcult. An adaptive and scalable system</div><div class="t m0 x1 h5 ye ff4 fs3 fc0 sc0 ls2 ws0">is required to offer accurate location-based services (LBSs) for</div><div class="t m0 x1 h5 yf ff4 fs3 fc0 sc0 ls2 ws0">the success of IoT. To enhance the attainment of IoT, we propose</div><div class="t m0 x1 h5 y10 ff4 fs3 fc0 sc0 ls7 ws0">a hybrid of principal component analysis (PCA) and gated recur-</div><div class="t m0 x1 h5 y11 ff4 fs3 fc0 sc0 ls2 ws0">rent unit (GRU) algorithms for mobility predictions in a wireless</div><div class="t m0 x1 h5 y12 ff4 fs3 fc0 sc0 ls2 ws0">urban area. During the system development processes, we ﬁrst</div><div class="t m0 x1 h5 y13 ff4 fs3 fc0 sc0 ls7 ws0">collect an LTE signal from three unmanned aerial vehicle base</div><div class="t m0 x1 h5 y14 ff4 fs3 fc0 sc0 ls6 ws0">stations (UAV-BSs), the Wi-Fi signal strength from each reach-</div><div class="t m0 x1 h5 y15 ff4 fs3 fc0 sc0 ls2 ws0">able Wi-Fi access points (APs), and channel information from the</div><div class="t m0 x1 h5 y16 ff4 fs3 fc0 sc0 ls2 ws0">Wi-Fi signal media. We then apply PCA to reduce the number</div><div class="t m0 x1 h5 y17 ff4 fs3 fc0 sc0 ls2 ws0">of Wi-Fi features and to decrease signal noise. Next, we train the</div><div class="t m0 x1 h5 y18 ff4 fs3 fc0 sc0 ls2 ws0">GRU algorithm to develop models that can predict the mobility</div><div class="t m0 x1 h5 y19 ff4 fs3 fc0 sc0 ls2 ws0">of IoT device users. Finally, we evaluate the tracking and trajec-</div><div class="t m0 x1 h5 y1a ff4 fs3 fc0 sc0 ls2 ws0">tory models. To evaluate the proposed techniques, we compare</div><div class="t m0 x1 h5 y1b ff4 fs3 fc0 sc0 ls2 ws0">the common parameters of the GRU with those of other deep</div><div class="t m0 x1 h5 y1c ff4 fs3 fc0 sc0 ls2 ws0">learning types. The proposed technique provides plausible and</div><div class="t m0 x1 h5 y1d ff4 fs3 fc0 sc0 ls6 ws0">state-of-the-art results for mobility predictions of IoT devices in</div><div class="t m0 x1 h5 y1e ff4 fs3 fc0 sc0 ls2 ws0">a wireless environment.</div><div class="t m0 x5 h5 y1f ff3 fs3 fc0 sc0 ls2 ws0">Index Terms—Deep learning, gated recurrent units (GRUs),</div><div class="t m0 x1 h5 y20 ff4 fs3 fc0 sc0 ls2 ws0">mobility predictions, principal component analysis (PCA).</div><div class="t m0 x6 h6 y21 ff1 fs4 fc0 sc0 ls8 ws0">I. INTRODUCTION</div><div class="t m0 x1 h7 y22 ff4 fs6 fc0 sc0 ls2 ws0">I</div><div class="t m0 x7 h6 y23 ff1 fs4 fc0 sc0 lsa ws0">NTERNET OF THINGS (IoT) is the interconnection of</div><div class="t m0 x7 h6 y24 ff1 fs4 fc0 sc0 lsa ws0">intelligent objects for universal computing, and through the</div><div class="t m0 x1 h6 y25 ff1 fs4 fc0 sc0 lsb ws0">use of the Internet, it enables smart devices to provide access</div><div class="t m0 x1 h6 y26 ff1 fs4 fc0 sc0 lsa ws0">to all users [1]. IoT has opened up tremendous opportuni-</div><div class="t m0 x1 h6 y27 ff1 fs4 fc0 sc0 lsb ws0">ties for novel applications, improving our lives, and creating</div><div class="t m0 x1 h6 y28 ff1 fs4 fc0 sc0 lsb ws0">a wide variety of smart systems, such as smart buildings, smart</div><div class="t m0 x1 h6 y29 ff1 fs4 fc0 sc0 lsb ws0">homes, and smart factories [2], [3]. IoT allows all things to</div><div class="t m0 x8 h8 y2a ff1 fs5 fc0 sc0 lsc ws0">Manuscript received August 17, 2019; revised September 21, 2019;</div><div class="t m0 x1 h8 y2b ff1 fs5 fc0 sc0 ls2 ws0">accepted October 7, 2019. Date of publication October 17, 2019; date of</div><div class="t m0 x1 h8 y2c ff1 fs5 fc0 sc0 lsc ws0">current version January 10, 2020. This work was supported by the Ministry</div><div class="t m0 x1 h8 y2d ff1 fs5 fc0 sc0 ls2 ws0">of Science and Technology through Pervasive Artiﬁcial Intelligence Research</div><div class="t m0 x1 h8 y2e ff1 fs5 fc0 sc0 ls2 ws0">Labs, Taiwan, under Grant MOST 108-2221-E-027-020 and Grant MOST</div><div class="t m0 x1 h8 y2f ff1 fs5 fc0 sc0 lsd ws0">108-2634-F-009-006. (Corresponding author: Abebe Belay Adege.)</div><div class="t m0 x8 h8 y30 ff1 fs5 fc0 sc0 ls2 ws0">A. B. Adege is with the Department of Electrical Engineering and Computer</div><div class="t m0 x1 h8 y31 ff1 fs5 fc0 sc0 ls2 ws0">Science, National Taipei University of Technology, Taipei 10608, Taiwan</div><div class="t m0 x1 h8 y32 ff1 fs5 fc0 sc0 ls2 ws0">(e-mail: abbblybelay@gmail.com).</div><div class="t m0 x8 h8 y33 ff1 fs5 fc0 sc0 ls2 ws0">H.-P. Lin is with the Department of Electronic Engineering, National Taipei</div><div class="t m0 x1 h8 y34 ff1 fs5 fc0 sc0 ls2 ws0">University of Technology, Taipei 10608, Taiwan (e-mail: hplin@ntut.edu.tw).</div><div class="t m0 x8 h8 y35 ff1 fs5 fc0 sc0 ls2 ws0">L.-C. Wang is with the Department of Electrical and Computer</div><div class="t m0 x1 h8 y36 ff1 fs5 fc0 sc0 ls2 ws0">Engineering, National Chiao Tung University, Hsinchu 300-10, Taiwan</div><div class="t m0 x1 h8 y37 ff1 fs5 fc0 sc0 ls2 ws0">(e-mail: lichun@cc.nctu.edu.tw).</div><div class="t m0 x8 h8 y38 ff1 fs5 fc0 sc0 ls2 ws0">Digital Object Identiﬁer 10.1109/JIOT.2019.2948075</div><div class="t m0 x9 h6 y5 ff1 fs4 fc0 sc0 lsa ws0">exist through Internet connections. Because we are highly con-</div><div class="t m0 x9 h6 y39 ff1 fs4 fc0 sc0 lsa ws0">nected with the Internet and large numbers of IoT devices have</div><div class="t m0 x9 h6 y3a ff1 fs4 fc0 sc0 lsa ws0">been produced, human daily activities are highly dependent on</div><div class="t m0 x9 h6 y3b ff1 fs4 fc0 sc0 lsa ws0">IoT services. In the coming years, increasing numbers of cur-</div><div class="t m0 x9 h6 y3c ff1 fs4 fc0 sc0 lsa ws0">rently available IoT devices will become connected to each</div><div class="t m0 x9 h6 y3d ff1 fs4 fc0 sc0 lsa ws0">other, and the global economic impact is expected to increase</div><div class="t m0 x9 h6 y3e ff1 fs4 fc0 sc0 lsb ws0">by $2 trillion. Although IoT will lead to an easier life for the</div><div class="t m0 x9 h6 y3f ff1 fs4 fc0 sc0 lsb ws0">Internet community [4], the explosion of IoT devices is lead-</div><div class="t m0 x9 h6 y40 ff1 fs4 fc0 sc0 lsb ws0">ing to higher signal interference in wireless networks owing to</div><div class="t m0 x9 h6 y41 ff1 fs4 fc0 sc0 lsa ws0">the additional protocols, channels, and communication man-</div><div class="t m0 x9 h6 y42 ff1 fs4 fc0 sc0 ls2 ws0">agement requirements. In wireless networks, in parallel with</div><div class="t m0 x9 h6 y43 ff1 fs4 fc0 sc0 lsa ws0">the expansion of IoT, Wi-Fi and mobile data and trafﬁc have</div><div class="t m0 x9 h6 y44 ff1 fs4 fc0 sc0 lsa ws0">grown rapidly, causing unavoidable interference and unsuc-</div><div class="t m0 x9 h6 y45 ff1 fs4 fc0 sc0 lsa ws0">cessful implementations of the IoT-based applications, such as</div><div class="t m0 x9 h6 y46 ff1 fs4 fc0 sc0 lsa ws0">channel modeling and location-based services (LBS). Because</div><div class="t m0 x9 h6 y47 ff1 fs4 fc0 sc0 lsa ws0">LBS serves to locate IoT devices, apply resource management,</div><div class="t m0 x9 h6 y48 ff1 fs4 fc0 sc0 lsa ws0">and supply a higher quality of service (QoS), it is a key tech-</div><div class="t m0 x9 h6 y49 ff1 fs4 fc0 sc0 lsb ws0">nology for wireless IoT applications [1], [5]. Thus, adaptive,</div><div class="t m0 x9 h6 y4a ff1 fs4 fc0 sc0 lsa ws0">accurate, robust, and scalable technologies are vital for aiding</div><div class="t m0 x9 h6 y4b ff1 fs4 fc0 sc0 lsa ws0">an IoT world.</div><div class="t m0 xa h6 y4c ff1 fs4 fc0 sc0 lsb ws0">Satellite-based positioning, such as a global positioning</div><div class="t m0 x9 h6 y4d ff1 fs4 fc0 sc0 lsb ws0">system (GPS), global navigation satellite system (GLONASS),</div><div class="t m0 x9 h6 y4e ff1 fs4 fc0 sc0 lsa ws0">and Galileo, commonly offer localization services [6].</div><div class="t m0 x9 h6 y4f ff1 fs4 fc0 sc0 lsb ws0">Nevertheless, the satellite-based applications are inappropri-</div><div class="t m0 x9 h6 y50 ff1 fs4 fc0 sc0 lsa ws0">ate for urban and hierarchical geolocation applications owing</div><div class="t m0 x9 h6 y51 ff1 fs4 fc0 sc0 lsb ws0">to the line-of-sight (LoS) problems, the requirement of sig-</div><div class="t m0 x9 h6 y52 ff1 fs4 fc0 sc0 lsa ws0">niﬁcant power and additional devices, and their sensitivity to</div><div class="t m0 x9 h6 y53 ff1 fs4 fc0 sc0 lsa ws0">occlusions. GPS, for example, requires a signiﬁcant amount</div><div class="t m0 x9 h6 y54 ff1 fs4 fc0 sc0 lsa ws0">of power and needs LoS services, and it is too difﬁcult to</div><div class="t m0 x9 h6 y55 ff1 fs4 fc0 sc0 lsa ws0">develop a model that can offer accurate mobility predictions in</div><div class="t m0 x9 h6 y56 ff1 fs4 fc0 sc0 lse ws0">highly revealing and changing urban areas [7]. Localizing the</div><div class="t m0 x9 h6 y57 ff1 fs4 fc0 sc0 lsa ws0">moving objects in a wireless network incurs higher data ﬂuctu-</div><div class="t m0 x9 h6 y58 ff1 fs4 fc0 sc0 lsa ws0">ations, which causes different objects to be erroneously located</div><div class="t m0 x9 h6 y59 ff1 fs4 fc0 sc0 lsa ws0">or tracked because of the complex wireless areas owing to</div><div class="t m0 x9 h6 y5a ff1 fs4 fc0 sc0 lsa ws0">signiﬁcant interference, blocking, and shadowing effects.</div><div class="t m0 xa h6 y5b ff1 fs4 fc0 sc0 lsb ws0">Mobility prediction is a signiﬁcant technique for dif-</div><div class="t m0 x9 h6 y5c ff1 fs4 fc0 sc0 lsa ws0">ferent applications, such as locating, tracking, predicting</div><div class="t m0 x9 h6 y5d ff1 fs4 fc0 sc0 lsa ws0">future routes, or patterns of IoT devices in sequences of</div><div class="t m0 x9 h6 y5e ff1 fs4 fc0 sc0 lsb ws0">time slots, and location-based advertisements and resource</div><div class="t m0 x9 h6 y5f ff1 fs4 fc0 sc0 ls2 ws0">allocations [6], [8], [9].</div><div class="t m0 xa h6 y60 ff1 fs4 fc0 sc0 lsa ws0">Mobility prediction is implemented through location updat-</div><div class="t m0 x9 h6 y61 ff1 fs4 fc0 sc0 lsb ws0">ing and location prediction strategies, where the location</div><div class="t m0 x9 h6 y62 ff1 fs4 fc0 sc0 lsa ws0">updating strategy is used to predict the current locations based</div><div class="t m0 x9 h6 y63 ff1 fs4 fc0 sc0 lsb ws0">on the available records, which refers to tracking prediction.</div><div class="t m0 x9 h6 y64 ff1 fs4 fc0 sc0 lsb ws0">By contrast, a trajectory prediction is a dynamic strategy in</div><div class="t m0 x9 h6 y65 ff1 fs4 fc0 sc0 lsb ws0">which the operation estimations of future locations of IoT</div><div class="t m0 x9 h6 y66 ff1 fs4 fc0 sc0 lsa ws0">devices are based on user movements or patterns. This type</div><div class="t m0 xb h8 y67 ff1 fs5 fc0 sc0 ls3 ws0">2327-4662</div><div class="t m0 xc h8 y68 ff1 fs5 fc0 sc0 ls2 ws0">c</div><div class="t m0 xd h8 y67 ff5 fs5 fc0 sc0 ls2 ws0"> 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.</div><div class="t m0 xe h8 y69 ff1 fs5 fc0 sc0 ls2 ws0">See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.</div><div class="t m0 xf h9 y6a ff6 fs7 fc0 sc0 ls2 ws0">Authorized licensed use limited to: XIDIAN UNIVERSITY. Downloaded on June 07,2020 at 15:16:03 UTC from IEEE Xplore. Restrictions apply. </div><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></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/625058e674bc5c01055e33f8/bg2.jpg"><div class="t m0 x1 h2 y1 ff1 fs0 fc0 sc0 ls1 ws0">506 IEEE INTERNET OF THINGS JOURNAL, VOL. 7, NO. 1, JANUARY 2020</div><div class="t m0 x1 h6 y6b ff1 fs4 fc0 sc0 lsa ws0">of strategy is affected by the speed and system resources</div><div class="t m0 x1 h6 y6c ff1 fs4 fc0 sc0 lsb ws0">used in mobile networks. The trajectory prediction in mobile</div><div class="t m0 x1 h6 y6d ff1 fs4 fc0 sc0 lsa ws0">technology is used for efﬁcient base station deployment, risk</div><div class="t m0 x1 h6 y6e ff1 fs4 fc0 sc0 lsa ws0">management, controlling the consumption of resources, and</div><div class="t m0 x1 h6 y6f ff1 fs4 fc0 sc0 lsb ws0">improving the quality of the decisions.</div><div class="t m0 x5 h6 y70 ff1 fs4 fc0 sc0 lsa ws0">A mobility prediction has been highly emphasized for pub-</div><div class="t m0 x1 h6 y71 ff1 fs4 fc0 sc0 lsb ws0">lic service provisioning, urban planning, and mobile network</div><div class="t m0 x1 h6 y72 ff1 fs4 fc0 sc0 lsa ws0">resource management [10] as the streaming sources of mobile</div><div class="t m0 x1 h6 y73 ff1 fs4 fc0 sc0 lsa ws0">networks and IoT devices are exponentially enlarged. Thus,</div><div class="t m0 x1 h6 y74 ff1 fs4 fc0 sc0 lsa ws0">divergence applicability of IoT devices and excessive mobile</div><div class="t m0 x1 h6 y75 ff1 fs4 fc0 sc0 lsa ws0">data products have encouraged further research in this area.</div><div class="t m0 x1 h6 y76 ff1 fs4 fc0 sc0 lsb ws0">However, designing an accurate system for mobility prediction</div><div class="t m0 x1 h6 y77 ff1 fs4 fc0 sc0 lsa ws0">is extremely challenging owing to the dynamical changes</div><div class="t m0 x1 h6 y78 ff1 fs4 fc0 sc0 lsa ws0">of wireless networks, the heterogeneous nature of data, and</div><div class="t m0 x1 h6 y79 ff1 fs4 fc0 sc0 lsa ws0">changes in location over time [9]. The Wi-Fi-based RSS val-</div><div class="t m0 x1 h6 y7a ff1 fs4 fc0 sc0 lsa ws0">ues are commonly used in indoor positioning, although they</div><div class="t m0 x1 h6 y7b ff1 fs4 fc0 sc0 lsa ws0">can also be applied in urban and smart areas that have numer-</div><div class="t m0 x1 h6 y7c ff1 fs4 fc0 sc0 lsa ws0">ous IoT connections [11]. However, using Wi-Fi technology</div><div class="t m0 x1 h6 y7d ff1 fs4 fc0 sc0 lsa ws0">alone for outdoor positioning causes a nonrobust position-</div><div class="t m0 x1 h6 y7e ff1 fs4 fc0 sc0 lsb ws0">ing model owing to multipath fading, reﬂections, diffraction,</div><div class="t m0 x1 h6 y7f ff1 fs4 fc0 sc0 lsb ws0">and interference effects. A cellular network by contrast com-</div><div class="t m0 x1 h6 y80 ff1 fs4 fc0 sc0 lsa ws0">monly uses outdoor positioning, which is less accurate owing</div><div class="t m0 x1 h6 y81 ff1 fs4 fc0 sc0 lsa ws0">to higher signal ﬂuctuations. Heterogeneous networks that do</div><div class="t m0 x1 h6 y82 ff1 fs4 fc0 sc0 lsa ws0">not vanish in the presence of other networks (e.g., cellular</div><div class="t m0 x1 h6 y83 ff1 fs4 fc0 sc0 lsb ws0">networks, channel state information, and Wi-Fi networks) are</div><div class="t m0 x1 h6 y84 ff1 fs4 fc0 sc0 lsa ws0">useful to ﬁll in the breaks in other networks and can pro-</div><div class="t m0 x1 h6 y85 ff1 fs4 fc0 sc0 lsa ws0">vide a better LBS in areas where high signal damage occurs.</div><div class="t m0 x1 h6 y86 ff1 fs4 fc0 sc0 lsb ws0">In [10], the Wi-Fi, Bluetooth, 4G LTE, and magnetic signals</div><div class="t m0 x1 h6 y87 ff1 fs4 fc0 sc0 lsa ws0">were collected and used interactively to provide an accurate</div><div class="t m0 x1 h6 y88 ff1 fs4 fc0 sc0 lsa ws0">LBS using the simultaneous localization and mapping (SLAM)</div><div class="t m0 x1 h6 y89 ff1 fs4 fc0 sc0 lsb ws0">algorithm.</div><div class="t m0 x5 h6 y8a ff1 fs4 fc0 sc0 lsa ws0">Mobility prediction can be computed through the aid of</div><div class="t m0 x1 h6 y8b ff1 fs4 fc0 sc0 lse ws0">image and video data; however, this approach has an unpre-</div><div class="t m0 x1 h6 y8c ff1 fs4 fc0 sc0 lsa ws0">dictable effect when the camera, observer, or observed objects</div><div class="t m0 x1 h6 y8d ff1 fs4 fc0 sc0 lsa ws0">move or cross each other, and its accuracy easily defects from</div><div class="t m0 x1 h6 y8e ff1 fs4 fc0 sc0 lsb ws0">blurry light [4]. A visual-based trajectory requires too many</div><div class="t m0 x1 h6 y8f ff1 fs4 fc0 sc0 lsa ws0">resources and a greater light intensity dependence compared</div><div class="t m0 x1 h6 y90 ff1 fs4 fc0 sc0 lsb ws0">with the radio-frequency-based mobility predictions. The costs</div><div class="t m0 x1 h6 y91 ff1 fs4 fc0 sc0 lsa ws0">of these technologies are too expensive to apply everywhere</div><div class="t m0 x1 h6 y92 ff1 fs4 fc0 sc0 lsa ws0">for cheaper IoT services in wider areas. The technical difﬁcul-</div><div class="t m0 x1 h6 y93 ff1 fs4 fc0 sc0 lsb ws0">ties, lack of adaptability in wider communities, and operational</div><div class="t m0 x1 h6 y94 ff1 fs4 fc0 sc0 lsb ws0">complexity, along with higher costs, create a paradigm in</div><div class="t m0 x1 h6 y95 ff1 fs4 fc0 sc0 lsa ws0">IoT for the use of images or video for LBS in urban and</div><div class="t m0 x1 h6 y96 ff1 fs4 fc0 sc0 lsb ws0">smart city environments. Thus, radio-signal-based mobility</div><div class="t m0 x1 h6 y97 ff1 fs4 fc0 sc0 lsa ws0">prediction has become a common way to offer an LBS in</div><div class="t m0 x1 h6 y98 ff1 fs4 fc0 sc0 lsa ws0">a wireless environment because it can be applied in a wider</div><div class="t m0 x1 h6 y99 ff1 fs4 fc0 sc0 lsb ws0">range, and is cost-effective for the ﬁtting of IoT business</div><div class="t m0 x1 h6 y9a ff1 fs4 fc0 sc0 lsa ws0">models. A radio-signal-based LBS, for example the use of</div><div class="t m0 x1 h6 y9b ff1 fs4 fc0 sc0 lsa ws0">Wi-Fi and cellular networks, is independent of the brightness</div><div class="t m0 x1 h6 y9c ff1 fs4 fc0 sc0 lsb ws0">and requires less storage space. It is also easy to capture the</div><div class="t m0 x1 h6 y9d ff1 fs4 fc0 sc0 lsb ws0">required data using mobile phones, sensors, and RFID [5].</div><div class="t m0 x5 h6 y9e ff1 fs4 fc0 sc0 lse ws0">Different machine learning techniques have been proposed</div><div class="t m0 x1 h6 y9f ff1 fs4 fc0 sc0 lsb ws0">to offer accurate positioning services using a radio signal</div><div class="t m0 x1 h6 ya0 ff1 fs4 fc0 sc0 lsa ws0">data set, including the k-nearest neighbors (K-NNs) [12],</div><div class="t m0 x1 h6 ya1 ff1 fs4 fc0 sc0 lsa ws0">a hybrid of support vector machines (SVMs) and deep</div><div class="t m0 x1 h6 ya2 ff1 fs4 fc0 sc0 lsb ws0">neural networks (DNNs) [13], and the artiﬁcial neural</div><div class="t m0 x1 h6 ya3 ff1 fs4 fc0 sc0 lsb ws0">networks (ANNs) [14]. However, the shallow-learning algo-</div><div class="t m0 x1 h6 ya4 ff1 fs4 fc0 sc0 lsa ws0">rithms, such as K-NN, SVM, and n ANN, lack ﬂexibility,</div><div class="t m0 x9 h6 ya5 ff1 fs4 fc0 sc0 lsb ws0">adaptability, and learning capacity owing to the lower learn-</div><div class="t m0 x9 h6 ya6 ff1 fs4 fc0 sc0 lsa ws0">ing capacity and are unable to adapt to dynamic changes</div><div class="t m0 x9 h6 ya7 ff1 fs4 fc0 sc0 lsb ws0">in the radio signals [15]. Moreover, owing to the growth of</div><div class="t m0 x9 h6 ya8 ff1 fs4 fc0 sc0 lsa ws0">numerous sensors and the spreading of information sources in</div><div class="t m0 x9 h6 ya9 ff1 fs4 fc0 sc0 lsa ws0">IoT eras, applying traditional methods for tracking and trajec-</div><div class="t m0 x9 h6 yaa ff1 fs4 fc0 sc0 lsb ws0">tory prediction in the complex wireless networks is becoming</div><div class="t m0 x9 h6 yab ff1 fs4 fc0 sc0 lse ws0">a challenge.</div><div class="t m0 xa h6 yac ff1 fs4 fc0 sc0 lse ws0">Zhang et al. [16] applied various machine learning</div><div class="t m0 x9 h6 yad ff1 fs4 fc0 sc0 lse ws0">approaches, such as SVM, K-NN, and local linear emend-</div><div class="t m0 x9 h6 yae ff1 fs4 fc0 sc0 lsb ws0">ing (LLE), for comparison with the DNN algorithm in terms of</div><div class="t m0 x9 h6 yaf ff1 fs4 fc0 sc0 lsa ws0">the root mean-square-error (RMSE), and DNN-based localiza-</div><div class="t m0 x9 h6 yb0 ff1 fs4 fc0 sc0 lsa ws0">tion was shown to outperform other methods. Adege et al. [17]</div><div class="t m0 x9 h6 yb1 ff1 fs4 fc0 sc0 lsa ws0">compared DNN with SVM-Poly and K-NN for localiza-</div><div class="t m0 x9 h6 yb2 ff1 fs4 fc0 sc0 lsb ws0">tion in wireless environments, and the results indicate that</div><div class="t m0 x9 h6 yb3 ff1 fs4 fc0 sc0 lsa ws0">DNN outperforms other traditional approaches owing to its</div><div class="t m0 x9 h6 yb4 ff1 fs4 fc0 sc0 lsb ws0">higher learning capacity from noisy signal distributions [18].</div><div class="t m0 x9 h6 yb5 ff1 fs4 fc0 sc0 lsa ws0">According to [19], the traditional approaches, such as particle</div><div class="t m0 x9 h6 yb6 ff1 fs4 fc0 sc0 lsb ws0">ﬁlter, Kalman ﬁlter, and Gaussian process, are characterized by</div><div class="t m0 x9 h6 yb7 ff1 fs4 fc0 sc0 lsa ws0">poor learning to a long time sequence data set. Thus, they are</div><div class="t m0 x9 h6 yb8 ff1 fs4 fc0 sc0 lsb ws0">limited to provide accurate and scalable mobility predictions</div><div class="t m0 x9 h6 yb9 ff1 fs4 fc0 sc0 ls2 ws0">in wireless environments as data in wireless environments are</div><div class="t m0 x9 h6 yba ff1 fs4 fc0 sc0 lsa ws0">heterogeneous, changed continuously, and easily affected by</div><div class="t m0 x9 h6 ybb ff1 fs4 fc0 sc0 lsa ws0">environmental factors. Consequently, these approaches unable</div><div class="t m0 x9 h6 ybc ff1 fs4 fc0 sc0 lsa ws0">to adapt signal ﬂuctuations in complex environments. In gen-</div><div class="t m0 x9 h6 ybd ff1 fs4 fc0 sc0 lsb ws0">eral, despite shallow learning, deep learning has varieties of</div><div class="t m0 x9 h6 ybe ff1 fs4 fc0 sc0 lsa ws0">hidden layers (HLs) and a number of optimizers and activa-</div><div class="t m0 x9 h6 ybf ff1 fs4 fc0 sc0 lsa ws0">tion functions. These make the DNN have various options for</div><div class="t m0 x9 h6 yc0 ff1 fs4 fc0 sc0 lsa ws0">optimizing the system performances. In [18] and [20], DNN</div><div class="t m0 x9 h6 yc1 ff1 fs4 fc0 sc0 lsb ws0">is particularly appropriate when the data are heterogeneous,</div><div class="t m0 x9 h6 yc2 ff1 fs4 fc0 sc0 lsa ws0">have different formats, and exhibit complex and hierarchical</div><div class="t m0 x9 h6 yc3 ff1 fs4 fc0 sc0 lsb ws0">natures. DNN is also more preferable to offer more accu-</div><div class="t m0 x9 h6 yc4 ff1 fs4 fc0 sc0 lsb ws0">rate results whenever there will be longer time-series data</div><div class="t m0 x9 h6 yc5 ff1 fs4 fc0 sc0 lsb ws0">sets. Thus, DNN is more appropriate than shallow learn-</div><div class="t m0 x9 h6 yc6 ff1 fs4 fc0 sc0 lsb ws0">ing as mobile data are collected from various sources and</div><div class="t m0 x9 h6 yc7 ff1 fs4 fc0 sc0 lsa ws0">complex in nature. Moreover, shallow learning cannot handle</div><div class="t m0 x9 h6 yc8 ff1 fs4 fc0 sc0 lsa ws0">high-dimensional states of the mobile databases while DNN</div><div class="t m0 x9 h6 yc9 ff1 fs4 fc0 sc0 lsa ws0">can. Consequently, most researchers have recently focused on</div><div class="t m0 x9 h6 yca ff1 fs4 fc0 sc0 lsb ws0">DNN-based applications especially when the data sets are too</div><div class="t m0 x9 h6 ycb ff1 fs4 fc0 sc0 lsa ws0">complex and dynamic [14].</div><div class="t m0 xa h6 ycc ff1 fs4 fc0 sc0 lsb ws0">To address these issues, we propose a hybrid of a principal</div><div class="t m0 x9 h6 ycd ff1 fs4 fc0 sc0 lsa ws0">component analysis (PCA) and gated recurrent unit (GRU)-</div><div class="t m0 x9 h6 yce ff1 fs4 fc0 sc0 lsb ws0">based mobility prediction (tracking and trajectory prediction)</div><div class="t m0 x9 h6 ycf ff1 fs4 fc0 sc0 lsb ws0">systems using Wi-Fi and cellular data sets in an urban area.</div><div class="t m0 x9 h6 yd0 ff1 fs4 fc0 sc0 lsa ws0">The larger dimension data sets require higher computational</div><div class="t m0 x9 h6 yd1 ff1 fs4 fc0 sc0 lsa ws0">time, consume larger storage spaces, and cause over ﬁtting</div><div class="t m0 x9 h6 yd2 ff1 fs4 fc0 sc0 lsa ws0">problems due to noise data. PCA and LDA are the common</div><div class="t m0 x9 h6 yd3 ff1 fs4 fc0 sc0 lsa ws0">linear projection approaches for feature reduction to unsu-</div><div class="t m0 x9 h6 yd4 ff1 fs4 fc0 sc0 lsa ws0">pervised and supervised data types, respectively [21]. Since</div><div class="t m0 x9 h6 yd5 ff1 fs4 fc0 sc0 lsa ws0">our data do not have distinct classes, we use the PCA algo-</div><div class="t m0 x9 h6 yd6 ff1 fs4 fc0 sc0 lsb ws0">rithm to integrate with the GRU network as a hybrid approach</div><div class="t m0 x9 h6 yd7 ff1 fs4 fc0 sc0 lsa ws0">for improving system performances, such as accuracy and</div><div class="t m0 x9 h6 yd8 ff1 fs4 fc0 sc0 lsb ws0">performance rates.</div><div class="t m0 xa h6 yd9 ff1 fs4 fc0 sc0 lsb ws0">A GRU network can mitigate complex problems using</div><div class="t m0 x9 h6 yda ff1 fs4 fc0 sc0 lsa ws0">its update and reset gates. Its performance was checked</div><div class="t m0 x9 h6 ydb ff1 fs4 fc0 sc0 lsa ws0">for different applications, such as speech recognition [22]</div><div class="t m0 x9 h6 ydc ff1 fs4 fc0 sc0 lsb ws0">and activity recognition in wireless networks [23]. In each</div><div class="t m0 x9 h6 ydd ff1 fs4 fc0 sc0 lsb ws0">application, GRU can offer state-of-the-art results. However,</div><div class="t m0 x9 h6 yde ff1 fs4 fc0 sc0 lsa ws0">despite the tremendous numbers of wireless products and IoT</div><div class="t m0 xf h9 y6a ff6 fs7 fc0 sc0 ls2 ws0">Authorized licensed use limited to: XIDIAN UNIVERSITY. Downloaded on June 07,2020 at 15:16:03 UTC from IEEE Xplore. Restrictions apply. </div><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a></div><div class="pi" data-data='{"ctm":[1.568627,0.000000,0.000000,1.568627,0.000000,0.000000]}'></div></div>

<div id="pf3" class="pf w0 h0" data-page-no="3"><div class="pc pc3 w0 h0"><img class="bi x0 y0 w1 h1" alt="" src="https://static.pudn.com/prod/directory_preview_static/625058e674bc5c01055e33f8/bg3.jpg"><div class="t m0 x1 h2 y1 ff1 fs0 fc0 sc0 ls1 ws0">ADEGE et al.: MOBILITY PREDICTIONS FOR IoT DEVICES USING GRU NETWORK 507</div><div class="t m0 x1 h6 y6b ff1 fs4 fc0 sc0 lsb ws0">devices, a hybrid PCA-GRU algorithm for tracking and tra-</div><div class="t m0 x1 h6 y6c ff1 fs4 fc0 sc0 lsb ws0">jectory predictions using a hybrid of Wi-Fi signal from access</div><div class="t m0 x1 h6 y6d ff1 fs4 fc0 sc0 lsa ws0">points (APs), and cellular data from unmanned aerial vehicle</div><div class="t m0 x1 h6 y6e ff1 fs4 fc0 sc0 lsb ws0">base stations (UAV-BSs) technologies to provide scalable and</div><div class="t m0 x1 h6 y6f ff1 fs4 fc0 sc0 lsb ws0">optimum mobility prediction services were not investigated.</div><div class="t m0 x5 h6 y70 ff1 fs4 fc0 sc0 ls2 ws0">Kim et al. [24] showed that a fully connected DNN incurs</div><div class="t m0 x1 h6 y71 ff1 fs4 fc0 sc0 lsa ws0">vanishing and exploding problems because such types of</div><div class="t m0 x1 h6 y72 ff1 fs4 fc0 sc0 lsa ws0">networks in longer time-series data are easily deﬂected when-</div><div class="t m0 x1 h6 y73 ff1 fs4 fc0 sc0 lsa ws0">ever more than one local minima or local maxima occurs.</div><div class="t m0 x1 h6 y74 ff1 fs4 fc0 sc0 lsa ws0">The modeling of IoT device-users when in motion is overly</div><div class="t m0 x1 h6 y75 ff1 fs4 fc0 sc0 lse ws0">challenging because data and locations are changed through</div><div class="t m0 x1 h6 y76 ff1 fs4 fc0 sc0 lsa ws0">time more frequently. In [22] and [25], GRU has updated and</div><div class="t m0 x1 h6 y77 ff1 fs4 fc0 sc0 lsa ws0">reset gates, and as a result, it has a higher capacity to adapt</div><div class="t m0 x1 h6 y78 ff1 fs4 fc0 sc0 lsa ws0">to data ﬂuctuations and provide efﬁcient generalization. It is</div><div class="t m0 x1 h6 y79 ff1 fs4 fc0 sc0 lsa ws0">characterized by high learning capability, captures complex</div><div class="t m0 x1 h6 y7a ff1 fs4 fc0 sc0 lsa ws0">data, and easily adapts to dynamic and temporal data changes.</div><div class="t m0 x1 h6 y7b ff1 fs4 fc0 sc0 lsb ws0">Adapting nonlinear data with a high learning rate in a time</div><div class="t m0 x1 h6 y7c ff1 fs4 fc0 sc0 lsb ws0">series, signiﬁcantly large and complex data sets are also the</div><div class="t m0 x1 h6 y7d ff1 fs4 fc0 sc0 lsa ws0">fundamental properties of the GRU network. Our proposed</div><div class="t m0 x1 h6 y7e ff1 fs4 fc0 sc0 lsa ws0">technique handles multiple IoT devices or objects when offer-</div><div class="t m0 x1 h6 y7f ff1 fs4 fc0 sc0 lsb ws0">ing tracking and trajectory services in a wireless environment.</div><div class="t m0 x1 h6 y80 ff1 fs4 fc0 sc0 lsb ws0">We compared the GRU network with the long short-term</div><div class="t m0 x1 h6 y81 ff1 fs4 fc0 sc0 lsa ws0">memory (LSTM) and multilayer perceptron (MLP) algorithms</div><div class="t m0 x1 h6 y82 ff1 fs4 fc0 sc0 lsa ws0">to evaluate its performance. We choose MLP and LSTM to</div><div class="t m0 x1 h6 y83 ff1 fs4 fc0 sc0 lsb ws0">evaluate the GRU’s performances as these exist in similar type,</div><div class="t m0 x1 h6 y84 ff1 fs4 fc0 sc0 lsb ws0">which are supervised deep learning types. Moreover, these</div><div class="t m0 x1 h6 y85 ff1 fs4 fc0 sc0 lsa ws0">are the state-of-the-art technologies that differ in performance</div><div class="t m0 x1 h6 y86 ff1 fs4 fc0 sc0 lsb ws0">due to data natures and their internal structures. Additionally,</div><div class="t m0 x1 h6 y87 ff1 fs4 fc0 sc0 lsa ws0">MLP, LSTM, and GRU are the common recent algorithms</div><div class="t m0 x1 h6 y88 ff1 fs4 fc0 sc0 lsb ws0">for time-series data set evaluations, such as using network</div><div class="t m0 x1 h6 y89 ff1 fs4 fc0 sc0 lsb ws0">signals [15], [26]. Our proposed method makes the following</div><div class="t m0 x1 h6 y8a ff1 fs4 fc0 sc0 ls2 ws0">contributions.</div><div class="t m0 x5 h6 y8b ff1 fs4 fc0 sc0 lsb ws0">1) We integrate Wi-Fi signals, channel IDs, and cellular</div><div class="t m0 x10 h6 y8c ff1 fs4 fc0 sc0 lsb ws0">data to ﬁll in the gaps when there are signal alterations</div><div class="t m0 x10 h6 y8d ff1 fs4 fc0 sc0 lsa ws0">or ﬂuctuations.</div><div class="t m0 x5 h6 y8e ff1 fs4 fc0 sc0 lsa ws0">2) We use the deep learning algorithm to develop a scalable</div><div class="t m0 x10 h6 y8f ff1 fs4 fc0 sc0 lsb ws0">and accurate mobility prediction models for massive IoT</div><div class="t m0 x10 h6 y90 ff1 fs4 fc0 sc0 lsa ws0">devices in cost-effective techniques and technologies.</div><div class="t m0 x5 h6 y91 ff1 fs4 fc0 sc0 lsb ws0">3) We design a novel positioning model to provide state-</div><div class="t m0 x10 h6 y92 ff1 fs4 fc0 sc0 lsa ws0">of-the-art positioning performances of the IoT devices</div><div class="t m0 x10 h6 y93 ff1 fs4 fc0 sc0 lsa ws0">through integrations of various algorithms, such as</div><div class="t m0 x10 h6 y94 ff1 fs4 fc0 sc0 lsb ws0">feedforward neural network (FFNN), PCA, and GRU</div><div class="t m0 x10 h6 y95 ff1 fs4 fc0 sc0 lsb ws0">network.</div><div class="t m0 x5 h6 y96 ff1 fs4 fc0 sc0 lsa ws0">4) We adapt the deep learning algorithm for mobile big</div><div class="t m0 x10 h6 y97 ff1 fs4 fc0 sc0 lsa ws0">data (MBD) collected from a real environment, experi-</div><div class="t m0 x10 h6 y98 ff1 fs4 fc0 sc0 lsb ws0">mentally.</div><div class="t m0 x5 h6 y99 ff1 fs4 fc0 sc0 lsb ws0">The remainder of this article is organized as follows. In</div><div class="t m0 x1 h6 y9a ff1 fs4 fc0 sc0 lsb ws0">Section II, we discuss related works. Section III presents</div><div class="t m0 x1 h6 y9b ff1 fs4 fc0 sc0 lsa ws0">the details of the proposed technique. The experimen-</div><div class="t m0 x1 h6 y9c ff1 fs4 fc0 sc0 lsb ws0">tal setup of the data collection processes is discussed in</div><div class="t m0 x1 h6 y9d ff1 fs4 fc0 sc0 lsa ws0">Section IV. The experimental results and discussions are shown</div><div class="t m0 x1 h6 y9e ff1 fs4 fc0 sc0 lsb ws0">in Section V. Finally, some concluding remarks and possible</div><div class="t m0 x1 h6 y9f ff1 fs4 fc0 sc0 ls2 ws0">future works are given in Section VI.</div><div class="t m0 x11 h6 ydf ff1 fs4 fc0 sc0 ls8 ws0">II. R</div><div class="t m0 x12 h6 ye0 ff1 fs5 fc0 sc0 ls10 ws0">ELATED WORKS</div><div class="t m0 x5 h6 ye1 ff1 fs4 fc0 sc0 ls2 ws0">The mobility predictions in wireless networks are very chal-</div><div class="t m0 x1 h6 ye2 ff1 fs4 fc0 sc0 lsb ws0">lenging since the place and time are changed recklessly [27].</div><div class="t m0 x9 h6 ye3 ff1 fs4 fc0 sc0 lsa ws0">The speed and directions of the motion of the IoT device and</div><div class="t m0 x9 h6 ye4 ff1 fs4 fc0 sc0 lsb ws0">the dynamical oscillations of wireless network trafﬁc bring</div><div class="t m0 x9 h6 ye5 ff1 fs4 fc0 sc0 lsa ws0">on the challenge of designing a robust model. Deep-learning-</div><div class="t m0 x9 h6 ye6 ff1 fs4 fc0 sc0 lsb ws0">based positioning in wireless network used to minimize the</div><div class="t m0 x9 h6 ye7 ff1 fs4 fc0 sc0 lsb ws0">hardware complexity as it is used to shift hardware-based</div><div class="t m0 x9 h6 ye8 ff1 fs4 fc0 sc0 lsa ws0">to application-based positioning approach [18]. For example,</div><div class="t m0 x9 h6 ye9 ff1 fs4 fc0 sc0 lsb ws0">a tracking system for wireless network users is now commonly</div><div class="t m0 x9 h6 yea ff1 fs4 fc0 sc0 lsb ws0">using mobile-based applications using radio signals emitted by</div><div class="t m0 x9 h6 yeb ff1 fs4 fc0 sc0 lsa ws0">mobile devices, APs or BSs. The scheme in [7] used a hybrid</div><div class="t m0 x9 h6 yec ff1 fs4 fc0 sc0 lsa ws0">of Wi-Fi and GPS data collected from the sensor network</div><div class="t m0 x9 h6 yed ff1 fs4 fc0 sc0 lsa ws0">and applied a crowdsourcing technique for online localization.</div><div class="t m0 x9 h6 yee ff1 fs4 fc0 sc0 lsa ws0">Their system performance was relatively better in accuracy</div><div class="t m0 x9 h6 yef ff1 fs4 fc0 sc0 lsa ws0">and power consumption than using GPS only. The system</div><div class="t m0 x9 h6 yf0 ff1 fs4 fc0 sc0 lsa ws0">performance shows that 92% of the localization performance</div><div class="t m0 x9 h6 yf1 ff1 fs4 fc0 sc0 lsa ws0">covers an error of less than 15 m. However, they evaluated the</div><div class="t m0 x9 h6 yf2 ff1 fs4 fc0 sc0 lsa ws0">system performances at the steady states rather than moving</div><div class="t m0 x9 h6 yf3 ff1 fs4 fc0 sc0 lsa ws0">objects. Farazi and Behnke [28] applied the LSTM algorithm</div><div class="t m0 x9 h6 yf4 ff1 fs4 fc0 sc0 lsa ws0">for detection, tracking, and identiﬁcation of the visual robot.</div><div class="t m0 x9 h6 yf5 ff1 fs4 fc0 sc0 lsb ws0">They evaluated the proposed system through classiﬁcation by</div><div class="t m0 x9 h6 yf6 ff1 fs4 fc0 sc0 lsa ws0">taking the detected object as a system evaluation technique.</div><div class="t m0 x9 h6 yf7 ff1 fs4 fc0 sc0 lsa ws0">When the detected objects are increased from 3 to 10, their</div><div class="t m0 x9 h6 yf8 ff1 fs4 fc0 sc0 lsa ws0">proposed technique’s performances decline from 96.2% to</div><div class="t m0 x9 h6 yf9 ff1 fs4 fc0 sc0 lsa ws0">66.5%. Thus, the proposed technique has a scalable problem.</div><div class="t m0 x9 h6 yfa ff1 fs4 fc0 sc0 lsb ws0">It is also a vision-based approach, which is not cost effective,</div><div class="t m0 x9 h6 yfb ff1 fs4 fc0 sc0 lsa ws0">and it has NLoS problems.</div><div class="t m0 xa h6 yfc ff1 fs4 fc0 sc0 lsb ws0">In [29], the indoor-based tracking and trajectory prediction</div><div class="t m0 x9 h6 yfd ff1 fs4 fc0 sc0 lsb ws0">using CNN for feature extraction and LSTM for motion</div><div class="t m0 x9 h6 yfe ff1 fs4 fc0 sc0 lsa ws0">predictions was discussed. The authors used additional</div><div class="t m0 x9 h6 yff ff1 fs4 fc0 sc0 lsa ws0">devices, such as barometer and magnetometer, which increase</div><div class="t m0 x9 h6 y100 ff1 fs4 fc0 sc0 lsb ws0">expenses. The visual-based tracking is also highly suspected</div><div class="t m0 x9 h6 y101 ff1 fs4 fc0 sc0 lsa ws0">in the LoS problem, and it is computationally complex [4].</div><div class="t m0 x9 h6 y102 ff1 fs4 fc0 sc0 lsb ws0">Their system performance shows a precision of 93% to the</div><div class="t m0 x9 h6 y103 ff1 fs4 fc0 sc0 lsb ws0">localization error threshold of 20 m. The computational time</div><div class="t m0 x9 h6 y104 ff1 fs4 fc0 sc0 lsa ws0">of CNN-LSTM was 1.6 s, which is slower due to the longer</div><div class="t m0 x9 h6 y105 ff1 fs4 fc0 sc0 lsa ws0">processing of the CNN and LSTM algorithms. In [19], the tra-</div><div class="t m0 x9 h6 y106 ff1 fs4 fc0 sc0 lsa ws0">jectory of the relative motion of pedestrians has been computed</div><div class="t m0 x9 h6 y107 ff1 fs4 fc0 sc0 lsb ws0">using the LSTM network. The authors used 2-D laser data sets</div><div class="t m0 x9 h6 y108 ff1 fs4 fc0 sc0 lsa ws0">to evaluate their system performance in an indoor environment.</div><div class="t m0 x9 h6 y109 ff1 fs4 fc0 sc0 lsa ws0">Their proposed system performance varies through numbers</div><div class="t m0 x9 h6 y10a ff1 fs4 fc0 sc0 lsa ws0">of observed objects and periods of observations. The accu-</div><div class="t m0 x9 h6 y10b ff1 fs4 fc0 sc0 lsa ws0">racy ranged from 0.26 up to 2.46 m for 3.2–10.0-s periods of</div><div class="t m0 x9 h6 y10c ff1 fs4 fc0 sc0 lsb ws0">observation.</div><div class="t m0 xa h6 y10d ff1 fs4 fc0 sc0 lsb ws0">In [8], [12], and [14], the schemes proposed the weighted</div><div class="t m0 x9 h6 y10e ff1 fs4 fc0 sc0 lsb ws0">KNN (WKNN), the KNN, and ANN algorithms, respectively,</div><div class="t m0 x9 h6 y10f ff1 fs4 fc0 sc0 lsa ws0">for tracking services. However, these kinds of approaches are</div><div class="t m0 x9 h6 y110 ff1 fs4 fc0 sc0 lsa ws0">inadequate to offer accurate and scalable LBS in dynamic and</div><div class="t m0 x9 h6 y111 ff1 fs4 fc0 sc0 lsb ws0">ﬂexible wireless environments since they have low learning</div><div class="t m0 x9 h6 y112 ff1 fs4 fc0 sc0 lsb ws0">capacity compared with the state-of-the-art algorithms, such</div><div class="t m0 x9 h6 y113 ff1 fs4 fc0 sc0 lsa ws0">as LSTM and GRU [16], [17]. In [30], the LSTM algorithm</div><div class="t m0 x9 h6 y114 ff1 fs4 fc0 sc0 lsa ws0">was applied for mobility predictions through ﬁve sequences of</div><div class="t m0 x9 h6 y115 ff1 fs4 fc0 sc0 lsb ws0">inputs. The authors obtained the minimum errors of 11.5 m.</div><div class="t m0 x9 h6 y116 ff1 fs4 fc0 sc0 lsa ws0">The precision of their system performance was unable to offer</div><div class="t m0 x9 h6 y117 ff1 fs4 fc0 sc0 lsa ws0">more than 90% of classiﬁcation accuracy. In [1], the trajec-</div><div class="t m0 x9 h6 y118 ff1 fs4 fc0 sc0 lsa ws0">tory prediction for the IoT localization was enhanced, and the</div><div class="t m0 x9 h6 y119 ff1 fs4 fc0 sc0 lsa ws0">system performance has been changed from 1- to 4.6-m errors</div><div class="t m0 x9 h6 y11a ff1 fs4 fc0 sc0 lsa ws0">for cell radios of 1 and 5 m, respectively. The authors imple-</div><div class="t m0 x9 h6 y11b ff1 fs4 fc0 sc0 lsa ws0">mented the proposed technique in indoor environments using</div><div class="t m0 x9 h6 y11c ff1 fs4 fc0 sc0 lsb ws0">only nine APs.</div><div class="t m0 xf h9 y6a ff6 fs7 fc0 sc0 ls2 ws0">Authorized licensed use limited to: XIDIAN UNIVERSITY. Downloaded on June 07,2020 at 15:16:03 UTC from IEEE Xplore. Restrictions apply. </div><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m1"></div></a></div><div class="pi" data-data='{"ctm":[1.568627,0.000000,0.000000,1.568627,0.000000,0.000000]}'></div></div>