Inter national J our nal of Electrical and Computer Engineering (IJECE) V ol. 7, No. 3, June 2017, pp. 1255 1261 ISSN: 2088-8708 1255       I ns t it u t e  o f  A d v a nce d  Eng ine e r i ng  a nd  S cie nce   w     w     w       i                       l       c       m     Localization of Distrib uted W ir eless Sensor Netw orks using T w o Sage SDP Optimization Reza Shahbazian 1 and Sey ed Ali ghorashi 2 1,2 Cogniti v e T elecommunication Research Group, Department of Electrical Engineering, Shahid Beheshti Uni v ersity G. C., T ehran, Iran. 2 CyberSpace Research Institute, Shahid Beheshti Uni v ersity G. C., T ehran, Iran. Article Inf o Article history: Recei v ed Jan 2, 2017 Re vised May 9, 2017 Accepted May 22, 2017 K eyw ord: W ireless Sensor Netw ork Localization Optimization Semi-Definite Programming (SDP) V irtual Anchor ABSTRA CT In man y applic ations of wireless sensor netw ork (WSN), the location of sensors is a neces- sity to e v aluate the sensed dat a and it is not ener gy and cost ef ficient to equip all sensors with global positioning systems. In WSN localization, some sensors (called anchors) ar e a w are of their location. Then, the distance measurements between sensors and anchors are used to localize the whole netw ork. WSN localization is a non-con v e x optim ization, ho we v er , relaxation techniques such as semi-definite programming (SDP) are used to relax the opti- mization. T o solv e this problem, all constraints should be considered simultaneously and the solution comple xity order is O n 2 where n is the number of sensors. The comple xity of SDP pre v ents solving lar ge size problems. Therefore, it is necessary to reduce the problem size in lar ge and distrib uted WSNs. In this paper , we propose a tw o stage optimization to re- duce the solution time, while pro vide better accurac y compared with original SDP method. W e first select some sensors that ha v e the maximum connection with anchors and perform the localization. Then, we select some of these sensors as virtual anchors. By adding the virtual anchors, we decrease the number of constraints. W e propose an al gorithm to select virtual anchors so that the total solution comple xity and time decrease considerably , while impro ving the localization accurac y . Copyright c 2017 Institute of Advanced Engineering and Science . All rights r eserved. Corresponding A uthor: S. A. Ghorshi Department of T elecommunications, F aculty of Electrical Engineering, Shahid Beheshti Uni v ersity T ehran, 1983963113, IRAN +9829904135 a ghorashi@sb u.ac.ir 1. INTR ODUCTION W ireless sensor netw orks (WSN) pro vide f ast, quite cheap and reliable solutions to a lar ge number of in- dustrial, commercial and military applications, ranging from surv eillance [1] and tracking to disaster management, robotics and other tasks [2, 3, 4] . Kno wing the correct positions of sensor nodes is essential to man y applications in ne xt-generation sensor netw orks. Location a w areness refers to de vices that can passi v ely or acti v ely determine their location. Location a w areness without the acti v e participation of the de vice is kno wn as non-cooperati v e localization. Location a w areness enables ne w applications for ubiquitous computing systems and mobile phones. WSNs present no v el trade-of fs in system design; on one hand, the lo w cost of the sensors f acilitates massi v e scale and highly parallel computation. On the other hand, each sensor is lik ely to ha v e limited po wer , limited reliability , and only local communication with a modest number of neighbors. These application conte xts and potential massi v e scale mak e it unrealistic to rely on careful placement or uniform arrangement of sensors. Using globally accessible beacons or to equip sensors with GPS is not feasible considering cost or ener gy constraints. Therefore, ranging-base localization techniques are introduced for WSNs [5]. The goal of localization is to determine the ph ysical coordinates of a group of sensor nodes. These coordi nates can be global, meaning that the y are aligned with some e xternally meaningful system lik e GPS. The y can also be local, meaning that the sensor nodes only ha v e the y related position in J ournal Homepage: http://iaesjournal.com/online/inde x.php/IJECE       I ns t it u t e  o f  A d v a nce d  Eng ine e r i ng  a nd  S cie nce   w     w     w       i                       l       c       m     DOI:  10.11591/ijece.v7i3.pp1255-1261 Evaluation Warning : The document was created with Spire.PDF for Python.
1256 ISSN: 2088-8708 constellation. In localization problem using neighboring dist ance measurements, it is assumed that the accurate positions of some nodes are kno wn, called anchors. Using some partial pair -wise distance measurements and the localization algorithm, it is possible to locate all sensor nodes in the netw ork. The dis tance measurement between nodes could be performed in dif ferent methods. By kno wing the transmitted po wer , using the recei v ed signal strength (RSS), and considering the ef fect of path loss, shado wi ng, and other losses the distance between tw o sensors could be found [6]. Both time of arri v al (T O A) and time dif ference of arri v al (TDO A) could be used to estimate the distances between sensors. Ho we v er , unsynchronized sensors with multi-path ef fect, highly de grades the accurac y of this es timation [7]. The methods such as angle of arri v al (A O A) or direction of arri v al (DO A) are not applicable in pair -wise distance measurement and may be used for tar get localization in a direct formulation [7]. In the past, localization problems were solv ed algebraically and computed by least squares solution to h yper - bolic equations called multi-lateration. No w adays, the optimum solution for sensor netw ork localization is pro vided using optimization. One of the first proposed con v e x optimization models for wireless sensor netw ork localization is introduced [8] in which the problem is relax ed to a second order cone programming (SOCP) problem. Ho we v er , this method needs a lar ge number of anchors on the area boundary to ha v e acceptable performance. In 2004, Bisw as [9] proposed a semi-definite programming (SDP) relaxation method for WSN localization. The research sho ws that the problem of finding uni qu e solution to a noiseless non-linear system describing the common point of intersection of h yper spheres in real Euclidean v ector space, can be e xpressed as a semi-definite program via distance geometry . This method outperforms the SOCP [8] in terms of accurac y and a v erage estimation error . Bisw as also de v eloped a solution [10] to deal with noisy corrupted data, and SDP relaxation used to transform the problem into a con v e x optimization scenario. This method [10] is based on maximum lik elihood (ML) estimation. Ho we v er , ML based methods are usu- ally v ery time consuming in comparison with other estimation methods. T o reduce the solution time, a method called smaller SDP (SSDP) [11] w as proposed which further relax es the original SDP . In this method, a single semi-definite matrix cone is relax ed into a set of small-size semi-definite matrix cones. The non-con v e xity of the problem co v ered by introducing a con v e x objecti v e function [12]. This method is only ef fecti v e for noise-free measurement cases. The research on impro ving the performance of SDP is still an open issue. In recent de v elopments, researchers impro v e the performance of WSNs localization in terms of rob ustness and accurac y especially in non-line of sight (NLOS) and harsh en vironments [13]. Ho we v er , in all SDP solutions, when the size of SDP problem increases, the dimension of matrix cone increases, simultaneously and the amount of unkno wn v ariables increases, non-linearly . It is kno wn that the arithmetic operation comple xity of the SDP is at least O n 2 to obtain an approximate solution. This comple xity pre v ents solving lar ge size problems. Therefore, it w ould be v ery beneficial to reduce the SDP problem size. On the other hand, the impact of noise on distance measurement and est imation error is important and this ef fect v aries in v ersely with problem size. The moti v ation of this w ork is to reduce the required time to solv e the optimization problem, while increasing the localization accurac y . In this paper , we propose a tw o stage localization algorithm based on SDP for WSNs that is applicable in an y m odification of the original SDP approach. Simulation results demonstrate that the proposed method is v ery ef fecti v e and significantly decreases the solution time and impro v es the a v erage position error . The remainder of the paper is or g anized as follo ws. In section 2, the system model is described and the problem is formulated. Section 3 proposes a sol ution for the problem. In section 4, simulation re sults are presented and section 5 concludes the paper . 2. RESEARCH MODEL W e consider a WSN with m anchors (kno wn positions) and n sensors (unkno wn positions) in a tw o dimen- sional (2D) en vironment. The e xtension of this localization problem to hi gh e r dimensions is straightforw ard. Some notations used in this paper are as follo ws. I , e and 0 denote the identity matrix, the v ector of all ones and the v ector of all zeros, respecti v ely . The 2-norm of a v ector x is denoted by k x k . A positi v e semi-definite matrix X is represented by X 0 . The position of anchor nodes are presented by v ector V a = f a 1 ; a 2 ; :::; a m g and the Euclidean distance between x j and x i is denoted as d ij and between a k and x j is denoted by d j k as follo ws: d ij = k x i x j k and d j k = k x j x k k (1) 3. PR OPOSED METHOD The optimization problem can be e xpressed as follo ws: IJECE V ol. 7, No. 3, June 2017: 1255 1261 Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE ISSN: 2088-8708 1257 Find X 2 R 2 n S.t. Y ii 2 Y ij + Y j j = d ij 2 ; 8 ( i; j ) 2 N s Y j j 2 X T j a k + a k 2 = d j k 2 ; 8 ( j ; k ) 2 N a N s = f ( i; j ) j x i x j < r g ; N a = f ( j ; k ) j x j x k < r g Y = X T X (2) where r is the communication range, and X = [ x 1 ; : : : ; x n ] , d ij and d j k are the noisy range measurements. The problem in (2) is non-con v e x and may be relax ed using SDP method as presented in (3). Y X T X ! Z 0; Z = I 2 X T X Y (3) The idea behind the proposed method is simple. The nearest sensors to the anchors with most connections to the anchors ha v e the best measurements. W e first localize some of sensors wit h best measurements, and use these sensors with kno wn location as virtual anchors. In the second stage, the real and virtual anchors are both used to localize the rest of sensors in the netw ork. The proposed method is e xplained in T able 1. T able 1. Proposed Algorithm Step Number Operation Initialization Anchors with kno wn and sensors with unkno wn location 1 Calculate all the sensor -sensor and sensor -anchor distance using noisy measurements d ij , d j k 2 Select the sensors that are in communication range of at least tw o anchors 3 Estimate the location of selected sensors 4 Choose the sensors with most distance to e xisting anchors a s virtual anchor 5 Choose the sensors that are not selected in 2 6 Calculate all the measurements d ij and d j k with ne w virtual anchors 7 Estimate the location of remained sensors 4. RESUL T AND AN AL YSIS Computer simulations are used to e v aluate the perform ance of the proposed algorithm. W e consider a netw ork with 7 anchors and 100 sensors deplo yed randomly in a normalized area. Simulation is performed using CVX toolbox in MA TLAB softw are [14]. The a v erage estimation error is calculated as follo ws: A v erage Error = 1 n : n X j =1 k x j a j k (4) W e perform the simulation for dif ferent communication ranges. W e assume that the distance measurements are corrupted by noise denoted by noise f actor as calculated using (5) and (6). W e compare the results with original SDP used to solv e the WSN localization [13]. Simulation parameters are summarized in table 2. d ij = d ij : (1 + r andn (1) noise f actor ) (5) d j k = d j k : (1 + r andn (1) noise f actor ) (6) The simulation en vironment, the location of anchors, the e xact and estimated locations of sensor s using original SDP and proposed tw o-stage SDP are illustrated in figure 1 and figure 2, respecti v ely . In figure 1 and figure 2, the communication range and noise f actor are set to 0 : 3 and 0 : 15 , respecti v ely . The a v erage e x ecution time to solv e the optimization problem, highly depends to hardw are configuration. Ho we v er , the relati v e and normalized e x ecution time could be used as a criterion to compare the computational comple xity of the proposed algorithm. In this paper , we ha v e a v eraged the total e x ecution time o v er 50 netw orks with 1000 realizations. The proposed tw o stage SDP Localization of Distrib uted WSNs using T wo Sa g e SDP Optimization (Shahbazian) Evaluation Warning : The document was created with Spire.PDF for Python.
1258 ISSN: 2088-8708 T able 2. Simulation parameters used for localization V ariable v alue Simulation Area Normalized 1 1 Number of sensors 100 Number of Anchors 7 Communication range 0 : 3 0 : 35 0 : 4 0 : 45 Noise f actor 0 : 05 0 : 1 0 : 15 0 : 2 0 : 25 outperforms the original SDP [10, 13], by 12 % impro v ement in e x ecution time. One the other hand, the a v erage localization error sho ws 30 % impro v ement on a v erage. -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 Estimated Location of sensros Exact Location of sensos Anchors Figure 1. Simulation area, the e xact and estimated location of sensors using original one-step localization In figure 3, the ef fect of noise f actor in e x ecution time is e v aluated. This simulation demonstrates that the increase in noise f actor , highlights the dif ference between original SDP and the proposed algorithm in terms of e x ecution time. In all noise f actors, the solution time of the proposed algorithm is much less than the original SDP . The ef fect of communication range on a v erage localization error is presented in figure 4. As this simulation sho ws, increasing the communication range, decreases the localization error . The proposed tw o stage algorithm has better localization error in comparison with original SDP in all communication ranges. Figure 5, illustrates the ef fect of dif ferent noise f actors on a v erage localization error . This figure demonstrates that in all noise figures and corrupted measurements, our proposed tw o stage localization has reduced the a v erage loc alization error in compared with one stage original SDP localization [10, 13]. 5. CONCLUSION In this paper we studied the localization problem in wireless sensor netw orks. The localization could be interpreted as a non-con v e x optimization problem. Semi-definite programming ha v e widel y been used to relax and solv e the optimization problem. Ho we v er , the computational com p l e xity increases non-linearly with increasing the number of sensors. W e proposed a tw o stage SDP solution to sol v e the localization problem. W e first, performed the localization with some selected sensors and used some of localized sensors as virtual anchors. Simulation results confirm that the proposed algorithm impro v es the localization accurac y 30% on a v erage, while decreases the a v erage e x ecution time by 12%. IJECE V ol. 7, No. 3, June 2017: 1255 1261 Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE ISSN: 2088-8708 1259 -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 Estimated Location of Sensors Anchors Exact Location of Sensors Figure 2. Simulation area, the e xact and estimated location of sensors using proposed tw o-step localization 0.05 0.1 0.15 0.2 0.25 Noise Factor (Normalized)  0 0.5 1 Total Time (Normalized) Proposed Two Step SDP One Step SDP [10], [13] Figure 3. The ef fect of noise f actor on the e x ecution time of localization algorithms Localization of Distrib uted WSNs using T wo Sa g e SDP Optimization (Shahbazian) Evaluation Warning : The document was created with Spire.PDF for Python.
1260 ISSN: 2088-8708 0.3 0.32 0.34 0.36 0.38 0.4 0.42 0.44 0.46 Radio and Measurement Range (Normalized) 0.03 0.04 0.05 0.06 0.07 0.08 0.09 Average Error (Normalized) Figure 4. The ef fect of radio range on a v erage localization error 0.05 0.1 0.15 0.2 0.25 Noise Factor (Normalized)  0.08 0.1 0.12 0.14 0.16 0.18 0.2 Average Error (Normalized) Proposed Two Step SDP One Step SDP [10], [13] Figure 5. The ef fect of noise f actor on a v erage localization error IJECE V ol. 7, No. 3, June 2017: 1255 1261 Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE ISSN: 2088-8708 1261 REFERENCES [1] Cheng B, Hanck e GP . A service-oriented architecture for wireless video sensor netw orks: Opportunities and challenges. Industrial Electronics Society , IECON 2015-41st Annual Conference of the IEEE . 2015; 1: 2667- 2672. [2] Iyeng ar SS , Brooks RR. Distrib uted sensor netw orks: sensor netw orking and applications. Chapman and Hall/CRC . 2012. [3] Hajihoseini A, Ghorashi SA. Distrib uted T ar get Localization in W ireless Sensor Netw orks using Dif fusion Adap- tation. Indonesian Journal of Electrical Engineering and Computer Science . 2016; 3(3): 512-518. [4] Saf aie A, Shahbazian R, Ghorashi SA. Cooperati v e Impro v ed T ar get Localization in Harsh En vironments using Direction of Arri v al. Indonesian Journal of Electrical Engineering and Computer Science . 2016; 3(2): 420-427. [5] Janapati R, Balasw amy C, Soundararajan K. Localization of Cooperati v e WSN using Dist rib uted PSO with Opti- mum References. International Journal of Electrical and Computer Engineering (IJECE) . 2016; 6(6): 3094-3102. [6] V aghefi RM, Gholami MR, Buehrer RM, Strom EG. Cooperati v e recei v ed signal st rength-based sensor localiza- tion with unkno wn transmit po wers. IEEE T ransactions on Signal Processing . 2013; 61(6): 1389-1403. [7] F ardad M, Ghorashi SA, Shahbazian R. A no v el maximum lik elihood based estimator for bearing-only tar get localization. 22nd Iranian Conference on Electrical Engineering (ICEE), IEEE . 2014; 1: 1522-1527. [8] Doherty L, El Ghaoui L. Con v e x position estimation in wireless sensor netw orks. INFOCOM 2001. T wenti eth Annual Joint Conference of the IEEE Computer and Communications Societies . 2001; 3: 1655-1663. [9] Bisw as P , Y e Y . Semidefinite programming for ad hoc wireless sensor netw ork localization. Proceedings of the 3rd international symposium on Information processing in sensor netw orks. A CM . 2004; 1: 46-54. [10] Bisw as P , Lian TC, W ang TC, Y e Y . Semidefinite programming based algorithms for sensor netw ork localization. A CM T ransactions on Sensor Netw orks (T OSN) . 2006; 2(2): 188-220. [11] W ang Z, Zheng S, Bo yd S, Y e Y . Further relaxations of the SDP approach to sensor netw ork localization. T ech. Rep. . 2006. [12] Fidan B, Kiraz F . On con v e x i fication of range measurement based sensor and source localization problems. Ad Hoc Netw orks . 2014; 20(1): 113-118. [13] Ghari PM, Shahbazian R, Ghorashi SA. W ireless sensor netw ork localization in harsh en vironment s using SDP relaxation. IEEE Communications Letters , 2016; 20(1): 137-140. [14] Grant M, Bo yd S, Y e Y . Matlab softw are for disciplined con v e x programming. 2010; V ersion 1.21. BIOGRAPHIES OF A UTHORS Reza Shahbazian recei v ed the B.Sc. de gree in Electrical Engineering from the Iran Uni v ersity of Science and T echnology (IUST), T ehran, Iran, in 2008 and the M.Sc. de gree (with honors) in T elecommunicati ons Engineering from the IUST , in 2011. In past, he has been a member of the W ireless Communications Labora tory in the School of Electrical Engineering at the IUST . He is currently a member of Cogniti v e Radio Laboratory at the Shahid Beheshti Uni v ersity (SB U), T ehran, Iran, to w ork under supervision of Dr . Ghorashi. He is af filiated with IEEE as student member . In journal of supercomputing, sensor re vie w , wireless personal communications, and other scientific publications, he has serv ed as in vited re vie wer . Further info on his homepage: http://f aculties.sb u.ac.ir/shahbazian Sey ed Ali Ghorashi recei v ed his B.Sc. and M.Sc. de grees in Electrical Eng. from the Uni v ersity of T ehran, Iran, in 1992 and 1995, respecti v ely . Then, he joined SAN A Pro Inc., where he w ork ed on modelling and simulation of OFDM based wireless LAN systems and interference cancellation methods in WCDMA systems. Since 2000, he w ork ed as a research assoc iate at Kings Colle ge London on capacity enhancement methods in multi-layer W -CDMA systems sponsored by Mobile VCE. In 2003 He recei v ed his PhD at Kings Colle ge and since then he w ork ed at Kings Colle ge as a research fello w . In 2006 he joined Samsung Electronics (UK) Ltd as a s enior researcher and no w he serv es as an associate professor at Department of T elecommunications, F aculty of Electrical Engineering, Shahid Beheshti Uni v ersity at T ehran, Iran, w orking on wireless communications. Localization of Distrib uted WSNs using T wo Sa g e SDP Optimization (Shahbazian) Evaluation Warning : The document was created with Spire.PDF for Python.