Professional Documents
Culture Documents
a r t i c l e i n f o a b s t r a c t
Article history: In this paper, a statistical channel model is proposed based on the second moment of Received Signal
Received 12 July 2017 Strength Indicator (RSSI) in an outdoor communication channel. The medium under study is a grass field
Received in revised form where the RSSI data are collected in different distances and orientations using a set of in-house built
22 December 2017
transmitter-receiver sensors. The validity of the constructed sensors is confirmed since the first moment
Accepted 8 January 2018
Available online 31 January 2018
of RSSI data follows the well-known Friis model. The proposed model presents an additional relationship
between the variance of RSSI data and distance. To demonstrate the application of this statistical rela-
tionship, we have investigated the localization problem of a hidden node using extended Kalman filter
Keywords:
Localization (EKF). Compared to the conventional EKF in which the covariance matrix of measurement noise is fixed,
Channel modeling this matrix can be updated online using the proposed model. The experimental and simulation results of
Measurement noise two different scenarios, which are fixed hidden node and mobile hidden node, show that the proposed
Wireless sensor network model improves the accuracy of RSSI localization from 10 to 22 percent in different situations.
Extended Kalman filters © 2018 Elsevier B.V. All rights reserved.
1. Introduction and communication costs, and network and anchors density. Nev-
ertheless, it cannot be said what algorithm is the best because the
Localization of wireless sensors has received significant atten- localization algorithms are severely affected by the application sce-
tion for the past few decades and has been exploited in many nario and each one can be the best in a certain application scenario.
applications such as environmental monitoring [1], habitat mon- One main category of localization algorithms is called range-
itoring [2], mobile robot tracking [3], wireless video capsule based algorithms in which the WSN consists of some anchors,
endoscopy [4], and military applications [5]. It has always been which are wireless sensors with known coordinates, and a hidden
essential for the humankind to gather information from the envi- node, which its coordinate is going to be localized. Range-based
ronment in order to make a proper decision about what has algorithms generally include two phases which are ranging and
happened or is going to happen. However, data gathering in many location computation [9]. Time of Arrival (TOA), Time Difference of
places is a dangerous, difficult, or time consuming job, and wire- Arrival (TDOA), Angle of Arrival (AOA), and Received Signal Strength
less sensor networks (WSNs) are one of the best substitutions in Indicator (RSSI) are some of the measurement techniques that are
such conditions. Thus the wireless sensors of a WSN are sometimes used in the first phase [10]. However, RSSI is one of the cheapest
inaccessible or hidden from sight. measurement techniques since there is no need to add any extra
So far different localization algorithms have been proposed and devices [11]. In this method, which has been widely used, the RSSI
applied considering energy efficiency, hardware limitations, sim- signal is measured based on the received signal strength as follows
plicity, performance, accuracy and so on [6,7]. Ref. [8] presented [12]:
three basic evaluation metrics to compare the performance of local- Received Power in Watt
ization techniques which are localization accuracy, computation RSSI (dBm) = 10 log (1)
0.001 Watt
RSSI is based on the fact that the received signal strength at the
∗ Corresponding author.
receiver is affected by the attenuation of the medium which is in
E-mail addresses: am.ka1981@gmail.com (A. Karimi Alavijeh),
turn a function of distance and the medium characteristics [13].
ramezani@mci.sdu.dk (M.H. Ramezani), ali.karimi6586@gmail.com However, this method suffers greatly from noise. Channel modeling
(A. Karimi Alavijeh). is an approach to study the behavior of measurement noise. Having
https://doi.org/10.1016/j.sna.2018.01.015
0924-4247/© 2018 Elsevier B.V. All rights reserved.
284 A. Karimi Alavijeh et al. / Sensors and Actuators A 271 (2018) 283–289
Fig. 2. Different positions of transmitter around the receiver where the data were
gathered.
where PL (d) is the RSSI at the receiver, P̄L (d0 ) is the average RSSI
strength at the reference distance d0 , n is the path loss exponent, d
is the distance between transmitter and receiver, and is a logarith-
mic Gaussian noise that models fading. Having wide measurements
in different distances, the parameters P̄L (d0 ) and n can be obtained
by minimizing the mean square error (MSE) between the model
output and the empirical measurements [14]. The transmitter was
placed around the receiver at angles 0◦ , 90◦ , 180◦ , and 270◦ to deal
with the heterogeneity of the antenna, and 11 determined distances
10 m, 20 m, up to 110 m. The RSSI was then gathered each time for
100 s with sample time 0.1 s (1000 samples). Fig. 2 shows the dif-
ferent positions of transmitter around the receiver where the data
were gathered.
A curve with the form of (2) was then fitted on them, and the
parameters of the model were calculated as below:
xk = Axk−1 + B (13)
k−1
R = diag (a + b log(di )) , i = 1, 2, 3, . . ., q. (16) The enhancement of localization is evaluated by the Root Mean
Square (RMS) of the position errors using (17):
5. Experimental and simulation results N
1
RMS loc = m2act [k] − m2est [k] (17)
This section provides the evaluation of the proposed chan- N
k=1
nel model by defining two different scenarios; fixed hidden node
and mobile hidden node. In the first scenario, the hidden node is where N is the number of samples in each repeat of simulation
fixed and we have experimental data (since we already have the (N = 1000), and mact [k] and mest [k] are the actual and the estimated
measured data at 44 different positions around the hidden node). positions of the hidden node, respectively. In order to investigate
The second scenario evaluates the impact of using the proposed the impact of using the proposed model, all estimations will be
model on the improvement of localization when the hidden node performed by two algorithms; CCEKF and VCEKF. The covariance
is mobile. Since there are no measured data in this case, the eval- matrix of measurement noise is constant in the CCEKF algorithm
uation is carried out by using simulation in which the behavior of while the VCEKF algorithm uses (16) to update this matrix on-line.
the measured noise is generalized to the whole area under study. In
order to validate the simulation results of the second scenario, the 5.1. Scenario 1
first scenario is evaluated by simulated data as well as real data and
the results are compared. Achieving the similar results from exper- The hidden node is fixed at the origin and its position is esti-
iments and simulations in the first scenario confirms the validity of mated by both VCEKF and CCEKF with the same initiation:
the simulation. According to Fig. 2, a circular area of radius 110 m is
covered. In this area, the anchors are located at known positions but x̂[0] = [−1, −1]T
the hidden node could be everywhere (in mobile case). However, P̂[0] = diag([1, 1]) (18)
in our experiments, in the fixed hidden node scenario, the hidden
Q = diag([0.003, 0.003]).
node is located at the origin, so the distance between hidden node
and anchors could vary from 10 to 110 and In the mobile hidden R is calculated by using (16) in VCEKF, while its elements are
node scenario, the distance could vary from 0 to 220 m. Basically we constant in CCEKF but are calculated by using (16) at d = 55 m. The
have not fixed the distance between anchors and the hidden node. results of 1000 times localization for each set of anchors are shown
Therefore, it should be noted that the errors reported in Fig. 7 are in Fig. 7. As it can be seen, the localization error in VCEKF is less than
not corresponding to specific distances. CCEKF while both of them are decreasing by increasing the num-
There are real data for 44 different positions of the transmit- bers of anchors. The lower plot of Fig. 7 shows the improvement of
ter which can be used for localization in the first scenario. Since VCEKF vs. the number of anchors. Regarding this figure, the less the
localization is possible by having at least 3 distances (if all of the number of anchors is, the more improvement can be seen.
anchors are not along a line), we selected 1000 combinations of k In order to compare the results of localization by using the
anchors (k = 3, 4, . . ., 10) randomly for the both experimental and experimental and the simulated RSSI, the hidden node at the ori-
simulated data. All of the simulations in this paper were done in gin is localized again with the same positions for anchors but using
MATLAB environment. simulated data. The simulated RSSI is generated by (14) with an
288 A. Karimi Alavijeh et al. / Sensors and Actuators A 271 (2018) 283–289
Fig. 8. Localization of the fixed hidden node using simulated data. (a) The RMS of the Fig. 9. Localization of the mobile hidden node using simulated data. (a) The RMS of
localization error for the CCEKF and VCEKF algorithms vs. the number of anchors, the localization error for the CCEKF and VCEKF algorithms vs. the number of anchors,
(b) the percentage of the localization improvement. (b) the percentage of the localization improvement.
additive Gaussian white noise whose covariance is determined by Fig. 9 depicts the RMS error of CCEKF and VCEKF and the improve-
(16). Fig. 8 shows the results of localization by employing simu- ment percentage of localization of VCEKF compared with CCEKF
lated RSSI. As it can be seen, the trends of reduction in localization for different numbers of anchors (3–10). The results show that the
error and localization improvement are almost the same as the localization improvement is also true for moving target.
experimental results. Although, as expected the absolute value of
localization error is significantly less than the first one since (14) 6. Conclusion
is an approximate model. In reality, there are many sources for the
noise term in (2), such as orientation of the antennas, fading and In this paper, after gathering RSSI by employing wireless
reflection. Specially the orientation of antenna is really important sensors, two communication channel models based on the first
as it can be seen from the distribution of points in each vertical line moment and the second moment of RSSI data were extracted.
of Fig. 5. In order to generate noise in simulations, we have con- The first model was exploited to validate the correct operation
sidered an ideal case which is a Gaussian noise with zero mean and of the developed sensors according to the obtained parameters of
variance given by (4). This causes a perfect match between model the channel. The second model showed that the variance of RSSI
noise and estimated noise, hence we have a very small error in sim- grows up as the distance increases. By proposing a mathematical
ulations. However, regarding the relative error between VCEKF and relationship for the dependency of RSSI variance and distance, we
CCEKF, it can still show the improvement which fairly matches the investigated a localization problem using EKF to present an appli-
experimental results. cation for the proposed model. It is shown that by on-line updating
of the EKF covariance matrix the localization can be improved com-
5.2. Scenario 2 pared to the conventional EKF. The results of the simulations can
be summarized as follows. First, localization of a fixed hidden node
The second scenario has the same condition as the first scenario using VCEKF resulted in about 22 percent improvement in compar-
except that it considers a moving hidden node. The hidden node ison with CCEKF in the case of having only three anchors. Second,
moves on a circle with the radius of 15 m with constant angular despite the number of anchors, the localization error of VCEKF is
speed. The governing equations of the system are as (11) and the always less than CCEKF algorithm. Third, localization improvement
initial values are assumed to be as below: in VCEKF is more obvious for lower number of anchors. Fourth and
finally, the improvement of the localization is also valid for mobile
x̂[0] = [17, 0.05, 0.05, 1, 0.1, 0.1]T hidden nodes.
P̂[0] = diag 1, 1, 1, 1, 1, 1 (19)
References
Q = diag 0, 0, qxc , 0, 0, qyc
[1] H. Yang, Y. Qin, G. Feng, H. Ci, Online monitoring of geological CO2 storage and
where leakage based on wireless sensor networks, IEEE Sens. J. 13 (2) (2013)
556–562.
x = y = 4
[2] I. Stojmenovic, Handbook of Sensor Networks: Algorithms and Architectures,
1 (20) vol. 49, John Wiley & Sons, 2005.
qxc = qyc = × 3.5 × 10−3 [3] J. Zhang, S. Li, G. Lu, Q. Zhou, A new wireless sensor localization and pose
x 2 tracking system for an autonomous mobile robot, in: 2010 International
A. Karimi Alavijeh et al. / Sensors and Actuators A 271 (2018) 283–289 289
Conference on Mechatronics and Automation (ICMA), IEEE, 2010, pp. [25] G. Welch, B.D. Allen, A. Ilie, G. Bishop, Measurement sample time
1971–1975. optimization for human motion tracking/capture systems, Proceedings of
[4] E.S. Nadimi, V. Tarokh, Bayesian source localization in networks with Trends and Issues in Tracking for Virtual Environments, Workshop at the IEEE
heterogeneous transmission medium, Navigation 59 (3) (2012) 163–175. Virtual Reality 2007 Conference (2007).
[5] M.P. Ðurišić, Z. Tafa, G. Dimić, V. Milutinović, A survey of military applications
of wireless sensor networks, in: 2012 Mediterranean Conference on
Embedded Computing (MECO), IEEE, 2012, pp. 196–199. Biographies
[6] V. Garg, M. Jhamb, A review of wireless sensor network on localization
techniques, Int. J. Eng. Trends Technol. (IJETT) 4 (4) (2013).
[7] A. Mesmoudi, M. Feham, N. Labraoui, Wireless sensor networks localization
algorithms: a comprehensive survey, arXiv:1312.4082. Amir Karimi Alavijeh received the M.Sc. degree in Control
[8] J. Wang, R.K. Ghosh, S.K. Das, A survey on sensor localization, J. Control Theory Electrical engineering from the University of S. Beheshti,
Appl. 8 (1) (2010) 2–11. Tehran, Iran, in 2014. He was a design engineer in sev-
[9] J.-A. Jiang, X.-Y. Zheng, Y.-F. Chen, C.-H. Wang, P.-T. Chen, C.-L. Chuang, C.-P. eral companies where he designed power stations, and
Chen, A distributed RSS-based localization using a dynamic circle expanding worked on research based projects related to image pro-
mechanism, IEEE Sens. J. 13 (10) (2013) 3754–3766. cessing, energy verifying, power protection, and so on.
[10] B.-C. Seet, Q. Zhang, C.H. Foh, A.C. Fong, Hybrid RF mapping and Kalman He is currently working on developing a wireless sensor
filtered spring relaxation for sensor network localization, IEEE Sens. J. 12 (5) network and researching into localization using wireless
(2012) 1427–1435. sensor networks and OCR employing neural networks. He
[11] H. Karl, A. Willig, Protocols and Architectures for Wireless Sensor Networks, is now a university lecturer at IAU and PNU. His primary
John Wiley & Sons, 2007. research interests include control engineering, estimation
[12] V. Apte, Y.A. Powar, et al., Improving the accuracy of wireless LAN based techniques, robotics, and wireless communication sys-
location determination systems using Kalman filter and multiple observers, tems.
in: Wireless Communications and Networking Conference, 2006. WCNC 2006.
IEEE, vol. 1, IEEE, 2006, pp. 463–468. Mohammad Hossein Ramezani received the M.S. and
[13] T. Chuenurajit, D. Suroso, P. Cherntanomwong, Implementation of RSSI-based Ph.D. degrees both in Electrical engineering from Sharif
3D indoor localization using wireless sensor networks based on ZigBee University of Technology, Iran, in 2002 and 2008, respec-
standard, J. Inf. Sci. Technol. 3 (2) (2012). tively. In 2009, he joined the Faculty of Electrical
[14] A. Goldsmith, Wireless Communications, Cambridge University Press, 2005. Engineering, University of S. Beheshti, Iran as assistant
[15] V. Erceg, L.J. Greenstein, S.Y. Tjandra, S.R. Parkoff, A. Gupta, B. Kulic, A.A. Julius, professor. In the period of May 2014–April 2017 he has
R. Bianchi, An empirically based path loss model for wireless channels in been post doctorate fellow at the Faculty of Engineering,
suburban environments, IEEE J. Sel. Areas Commun. 17 (7) (1999) 1205–1211. University of Southern Denmark. Since March 2017, he
[16] H. Lee, S. Lee, Y. Kim, H. Chong, Grouping multi-duolateration localization has been assistant professor of Control Engineering at the
using partial space information for indoor wireless sensor networks, IEEE Mads Clausen Institute, Division of Mechatronics, Univer-
Trans. Consum. Electron. 55 (4) (2009). sity of Southern Denmark. His primary research interests
[17] X. Li, Y. Zhang, K. Xu, G. Fan, H. Wu, Research of localization and tracking are control engineering, mechatronics and signal process-
algorithms based on wireless sensor network, J. Inf. Comput. Sci. 8 (4) (2011) ing.
708–715.
[18] F. Gustafsson, F. Gunnarsson, N. Bergman, U. Forssell, J. Jansson, R. Karlsson,
P.-J. Nordlund, Particle filters for positioning, navigation, and tracking, IEEE Ali Karimi Alavijeh received the B.Sc. degree in Power
Trans. Signal Process. 50 (2) (2002) 425–437. Electrical Engineering from Azad University of Najafabad,
[19] C. Jie, RSSI-based indoor mobile localization in wireless sensor network, Int. J. Iran, in 2011. He is currently a DSP-based instrument
Digit. Content Technol. Appl. 5 (7) (2011) 408–416. designer. His research interests include aviation design
[20] Z. Xiong, F. Sottile, M.A. Spirito, R. Garello, Hybrid indoor positioning and sensor network enhancement (software and hard-
approaches based on WSN and RFID, in: 2011 4th IFIP International ware).
Conference on New Technologies, Mobility and Security (NTMS), IEEE, 2011,
pp. 1–5.
[21] G. Mao, Localization Algorithms and Strategies for Wireless Sensor Networks:
Monitoring and Surveillance Techniques for Target Tracking, IGI Global, 2009.
[22] P. Manual, XBee/xBee-PRO RF modules.
[23] S.S. Saab, Z.S. Nakad, A standalone RFID indoor positioning system using
passive tags, IEEE Trans. Ind. Electron. 58 (5) (2011) 1961–1970.
[24] J. Han, Q. Song, Y. He, Adaptive unscented Kalman filter and its applications in
nonlinear control, in: Kalman Filter Recent Advances and Applications,
InTech, 2009.