91
ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC
An Integrated RSS Approach for an Improved Location Estimation in Wireless Sensor Networks
A. I. Alhasanat1, Atef Abu Salim 1, B. S. Sharif2, C. C. Tsimenidis3 and J. A. Neasham3
1,2Department of Electrical and Computer Engineering, School of Electrical and Electronic Engineering,
1Nizwa University, Nizwa, Oman
2Khalifa University, UAE
3University of Newcastle Upon Tyne, Newcastle Upon Tyne, UK
Abstract: Received Signal Strength (RSS) based localization represents a simple and cost effective approach in Wireless sensor networks. However, the accuracy of sensor locations is significantly deteriorated as a result the uncertainties of pair-wise RSS measurements. This paper discusses the use of the transmitter-receiver Correlation, or Link Quality Indicator (LQI), along with RSS measurements in order to increase the reliability of the measured RSS and then obtain better location estimation. The proposed technique can be efficiently adopted by sensor nodes in a completely distributed manner. In addition, this research demonstrates the performance of the RSS measurements in a typical indoor environment, including obstructions, reflectors and people in motion. The validity of the proposed RSS localization model is verified for some location estimation methods, e.g. centroid, Ecolocation and Least Square Estimator (LSE) along with the corresponding Cramer´-Rau Bound (CRB). The results of this paper showed better, and CRB comparable, localization performance by using the proposed RSS enhancement approach with these algorithms.
Keywords: RSS, LQI, CRB
I.INTRODUCTION
RF ranges for localization in wireless sensor networks (WSN) are conventionally measured using methods such as Received Signal Strength (RSS), Time-of-Arrival (TOA), Angle-of-Arrival (AOA) and other hybrid algorithms [1], [2]. RSS measures the power of incoming signals and then, based on known transmitting power, the path loss can be computed to indicate the transmitter receiver separation distance.
With RSS-based localization techniques, the location accuracy deteriorates due to the fact that the RSS value suffers severely from the problem of channel multi-path fading and shadowing [3]. Another problem which occurs with the RSS-based technique is that the transmit power may not be accurately known as it is a function of the sensor battery voltage [5]. Despite this, RSS provides the cheapest and simplest technology since no additional hardware is needed, as is the case with the TOA and AOA methods [6]. Moreover, the RSS of RF signals is available in almost all existing wireless systems, even with small, low-power platforms such as WSNs [7].
92
ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC
It is the author’s belief that improvements can be made to RSS-based localization accuracy, either by refining RSS data or using a robust localization algorithm, without incurring any additional burden on sensor networks, which is obviously a practically desirable solution. Hence, this study focuses on the RSS scheme as the basic ranging technology for localization
in WSNs. The contributions of this paper is summarized as follows:
1) An RSS thresholding method based on the correlation between the transmitter and receiver node is proposed and implemented through practical experiments.
2) The proposed method, without any additional network overhead, increases the reliability of RSS measurements and, therefore, reports a significant improvement in estimated sensor locations.
3) The suggested approach is implemented by each sensor nodes individually and then can be adopted in a complete distributed manner.
The remainder of this paper is organized as follows. Next section introduced the problem statement.
Section III introduced the RSS improvement approach suggested in this paper. The results, including the RSS model validity and it is impacts on localization methods, are discussed in Section II and V, respectively. Finally, the conclusion and future work are drawn in Section VI.
II. PROBLEM STATEMENT
Signal noise and attenuation, as well as interference and impediments from other channels, represent a severe challenge in wireless radio communication. In the literature, channel propagation effects are categorized into small- and large-scale [3]. With small-scale propagation, signal variations may occur irrespective of the distance, even if the distance is very short. For example, if multiple signals with different amplitudes and phases arrive at the receiver, these signals will be added constructively and destructively as a function of frequency, causing what is often referred to as a frequency-selective fading problem [4]. The effect of this phenomenon significantly depends on the channel bandwidth. For example, with narrowband channels, multipath fading can be diminished by using a spread-spectrum method or time-averaging techniques [8].On the other hand, large-scale (sometimes called medium-scale [9]) propagation affects the variations of the Received Signal Strength (RSS) (strength and power are used inter-changeably in this paper) with respect to the transmitter-receiver separation distance.
These variations have been characterized as being due to path loss and shadowing [3].
93
ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC
With path loss, the radiated power is attenuated along the distance between the transmitter and receiver.
The shadowing is attributed to obstructions between the transmitter and receiver, which attenuate the signal power through absorption, reflection, scattering and diffraction. In general, channel models can be further classified into Line-of-Sight (LOS) and Non-Line-Of Sight (NLOS) channels.
A. Experiment Procedure
As shown in Figure 1, the measurement system consisted of a total of 25 Pixie nodes and one PICDEM-Z device. The Pixie devices were placed in a uniform-random strategy on the top of the cubicle walls (1.30m from the ground). Since the PICDEM Zigbee is equipped with an O/P connector, it was used as a network coordinator in order to collect the RSS values measured by Pixie nodes and, then, to send them to the PC for system analysis, channel parameter estimation and algorithm development. All nodes were in range of each other.
Using a Time Division Multiple Access (TDMA) scheme, each Pixie node transmitted 16-times, with each time corresponding to each channel consecutively. Meanwhile, the other Pixie nodes read the RSS value of the received packet for each channel and computed their average. This stage continued until all measured RSS values were transmitted to the coordinator. In the second stage, the PICDEM-Z node collected the measured RSS from all Pixies, combined them into a 25x25 RSS matrix and sent these data to the PC. This procedure was repeated every five minutes, and thus, in total, 100 RSS matrices were obtained and stored in the PC, which could then be read by the Matlab software.
Since RSS measurements error is not ergodic due to shadowing and NLOS channels [18], the above experiment procedure was conducted twice. The first time was during the evening and weekend, to ensure that the communication channels were mostly static. The other time, however, was during a busy week- day, where these channels were more exposed to variability as a result of people movements and interference from other sources.
III. RSS ENHANCEMENT APPROACH
Generally, errors in RSS data can be classified as time-varying or static. It has been shown by Patwari in [18] that the time-varying error, which is attributed to the additive noise and interference, can be reduced by averaging multiple RSS measurements over time. On the other hand, static errors are typically caused by arrangements of physical object (e.g. walls, partitions, etc). As this error is environment-dependent, for a particular environment it will be mostly constant over time. Since the environment structure is unpredictable, no general technique can be adopted to reduce such errors [8].
Shadowing effects, as explained in Section II, dominate the high variance in RSS values [4]. Hence, in an attempt made by Want in [23], to improve the overall RSS-based range measurements, a threshold RSS value was specified, below which the RSS value is unreliable and may not be considered
94
ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC
15 Pixie Node
20
Coordinator Node
21
19
12
15 16 17
18
22
(m)
9 14
13
12 11
y−coordina te
7 8 9 10
6
23
24
6 5 4 3
3
1 25 2
0
0 1 2 3 4 5 6 7
x−coordinate (m)
Fig. 1.Pixie nodes replacement
for distance estimation. In this approach, RSS measurements were improved as a result of ignoring the RSS values of high ambiguities; however, using this approach may limit sensor connectivity, in particular with large scale WSNs. For example, perhaps the transmitter is located at a far distance from the receiver;
this causes a low, not necessarily unreliable, RSS value.
In order to improve the RSS measurements, this paper suggests a new approach which is consist of a sequence of three phases: correlation thresholding, averaging and symmetrical RSS values. In the following subsequent sections, we discuss these phases and show how they were used to improve the RSS measurements resulting from this experiment.
95
ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC
A. CORR Threshold Technique
When the RSS and the CORR data are combined, they can probably indicate whether a low signal strength (low RSS) resulted from long transmitter-receiver separation distance or from high shadowing. If it is due to high shadowing, then the received packets could be rejected. In other words, higher CORR indicates higher Signal-To-Noise Ratio (SNR) and then more reliable RSS measurements.
According to this proposition, the reliability of the overall RSS measurements in the two experiments is slightly improved, as shown in Figure 2. In this figure, the measured RSS variance σ2 is plotted as a function of Corrth (from 80 to 110). It should be noted that the RSS variance of the day experiment is slightly decreased at high values of Corrth. However, the night measurement almost demonstrated a constant σ2 with respect to the Corrth. This result implies that using Corrth could be more effective in low SNR environments. Hence, further investigations in this direction could be useful.
Moreover, both experiments show considerable decline in the σ2 when the Corrth is larger than 100. As a consequence, better RSS values can be achieved if the Corrth is kept as high
Fig. 2. The measured RSS variance σ2 as a function of Corrth (from 80 to 110) for both day and night experiments.
96
ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC
TABLE I
1
4
Day
Experiment
RESULTING FROM THE NIGHT AND DAY EXPERIMENTS.
1 3
Night Experiment
)
2σ (
v
varianc
e 1
2 Experiment Time Method σ2 n
Original RSS 9.8 2
RSS
1 1
Night Correlated RSS 9.6 2
Measure d
Averaged RSS 8.4 2
1
0 Symmetrical
RSS 8.2 2
9 Original RSS 12.4 2
Day Correlated RSS 11.8 2
8
Averaged RSS 9.8 2 80 85 90
9
5 100 105 CORR Threshold
value (Corrth) Symmetrical
RSS 8.3 2
as possible, whilst maintaining the connectivity between all sensor nodes. In the experiment procedure, for instance, when Corrth >104, some pairwise RSS measurements disappeared. To avoid such a scenario, the Corrth was set to 104. After the measured RSS data satisfying the constraint CORR > corrth were considered, the resulting data yield σ2 = 2, σ2 = 9.6 for the night RSS data, and σ2 = 2, σ2 = 11.8 for the day RSS measurements.
97
ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC
B. Averaging RSS Measurements
In these experiments, sixteen RSS values, corresponding to sixteen channels at each reception, were averaged. After that, 100 different RSS matrices collected during the night experiment were averaged to produce a single RSS matrix of 25x25. The same was applied to the day experiment. The result showed that the day and night experiments fit the linear model with approximately the same channel parameters, i.e σ2 is 8.4 and n is 2. As a result of applying this method on the RSS measurements resulting from the correlation thresholding method, a decrease in the RSS variance is obtained, and then the data is used to fit the linear model at σ2 = 2 and σ2 = 8.3 for the night experiment, and n = 2 and σ2 = 9.8 for the day experiment.
0
− 50
−1 00
100 101
(a)
[dBm]
0
RSS
, −5 0
Strength
−1 00
100 101
Signal
(c) 0
Received −5
0
−1 00
100 101
(e) 0
−5 0
−1 00
100 101
98
ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC
(g)
Transmitter−Receiver Distance (d), [metres]
Measured RSS value
Linear Fit model
0
− 50
−1 00
100 101 (b)
[dBm]
0
RSS,
−5 0
Strength
−1 00
100 101
Signal
(d) 0
Recei ved
−5 0
−1 00
100 101
(f)
0
−5 0
−1 00
100 101
(h)
Transmitter−Receiver Distance (d), [metres]
99
ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC
C. Symmetrical RSS Measurements
Ideally, the RSS data represents a symmetrical matrix, i.e. pi;j = pj;i. In practice, however, due to sensor hardware variabilities as well as environment and channel factors, the measured RSS matrix p~ is not necessarily symmetric. Further reduction can be obtained if we consider a symmetric RSS matrix by computing the average of the two RSS values of the two nodes that share the same link. The result showed that the night and day experiments fit the linear model with approximately the same channel parameters, i.e σ2 is 8.3 and n is 2.
Table I summarizes the estimated channel parameters, corresponding to each method explained above, during the night and day experiments. In addition, Figure 3 shows the linear
Fig. 3. Linear fit model of the original measured RSS (a) night and (b) day, Correlated RSS of (c) night and (d) day, averaged RSS of (e) night, (f) day, and symmetrical RSS of (g) night, (h) day, with Corrth = 100.
fit model of RSS measurements as a function of inter-sensor distance. Plots in the left column of this figure represent the results for the night experiment as (a) original RSS, (c) correlated RSS with Corrth = 100, (e) averaged RSS and (g) symmetrical RSS. The column on the right show these results for the day experiment as (b) original RSS, (d) correlated RSS with Corrth = 100, (f) averaged RSS and (h) symmetrical
IV. CHANNEL MODEL VALIDITY
As mentioned in the beginning of this paper, one of the objectives of this experiment is to verify the RSS model and, therefore, determine whether it can be applied for location estimation purposes. To do so, using the Two-Sample T hypothesis test implemented by the SPSS software, we compared RSS residuals of data set generated by the theoretical model to the RSS measurements obtained in the day and night experiments.
For the original and refined RSS measurements of the night experiment, the conducted tests yielded p- values of 0.89 and 0.68, respectively. As a result, the null hypothesis at a significance level of 0.05 was verified in the two cases. However, in the day experiment, the results of the tests showed that the null hypothesis was rejected for each of the original and modified RSS measurements at the same significance level. This result indicates that the RSS measurements in the day experiment were not properly characterized by the theoretical RSS model. One possible explanation of this result is that in the day experiment the environment dynamics were not attributed by the simple shadowing factor. The experiment results presented in Section V, nevertheless, show Pixie locations of good accuracy.
100
ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC
V.IMPACT ON LOCALIZATION METHODS
We turn now to see the influence of using the RSS modification approach discussed in this paper on the performance of four algorithms: LSE [20], Ecolocation [24], Centroid [25] and Proximity [26]. The average location estimation errors of all algorithms with the original and modified RSS measurements are shown in Table II. In order to simplify this comparison, the improvement ratio of location estimation achieved by using the modified RSS measurements over the original ones is computed. Note that results shown in this table represent the location estimation errors due the four algorithms and for eight reference nodes selected randomly. The maximum possible accuracy is bounded by the CRB [22]. The CRB is also reported for each case considered in this experiment. Since the CRB uses the actual sensor location, computing lower bound does not necessary need a Monte Carlo simulation. Instead, the CRB will be computed only once based on the measured channel parameters listed in Table 2.
It can be clearly observed from this table that for all algorithms, except Centroid and Proximity in the night experiment, the refined RSS measurements considerably benefited the localization process of these algorithms. Ecolocation algorithm demonstrated the highest improvement ratio when the RSS improvement approach was applied. For example, in the night experiment, this method produced about 29% reduction in its location errors from 3.56 to 2.54 m. Similarly, in the day experiment, LSE reduced this error by 24% from 1.7 to 1.3 m. It is worth mentioning that LSE provides the best accurate position estimation method which is almost coincided with the
TABLE II
COMPARISON OF THE ORIGINAL/MODIFIED AND DAY/NIGHT RSS MEASUREMENTS IN TERMS OF AVERAGE LOCATION ERRORS FOR CENTROID (CEN), PROXIMITY (PRX), ECOLOCATION (ECO), AND LSE ALGORITHMS, ALONG WITH THE CORRESPONDING CRB.
CEN PRX ECO LSE CRB
Night Experiment
Original RSS 1.43 2.67 3.56 1.19 1.13 Modified RSS 1.43 2.67 2.54 1.10 1.00 Improvement ratio 0% 0% 29% 9% 13%
Day Experiment
Original RSS 1.72 2.64 2.98 1.7 1.27 Modified RSS 1.64 2.48 2.68 1.3 1.13 Improvement ratio 5% 7% 11% 24% 12%
corresponding CRB. This is due to the fact that in Gaussian error distribution, LSE behaves as efficiently as the lower bound at high Signal-To-Noise (SNR) ratio [27]. Moreover, the location error due to CRB was reduced by 13% and 11% for the night and day experiments, respectively.
101
ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC
On the other hand, since Centroid and Proximity are range free algorithms and not RSS-based distance estimation de-pendant, they may not take advantage of the modification performed on the RSS measurements. As a result, there was no improvement in the estimated locations when the RSS enhancement approach was used, despite the fact that in the day experiment the location error was inconsiderably reduced by 5% and 7% for Centroid and Proximity, respectively.
VI. CONCLUSION AND FUTURE WORK
In summary, this paper has discussed the RSS-based localization model and described how this model can be used to estimate distances between sensor nodes, from both the theoretical and practical perspectives.
The well-known statistical RSS model used in the literature was explained. In addition, the practical experiment which had been conducted in order to examine the usability of the log-normal RSS model adapted in this paper was introduced.
Although it was difficult to accurately model the RSS behavior since environment characteristics are unpredictable, a sequence of RSS refinement techniques was applicable to the measured RSS data resulting from the practical experiment. This, as a consequence, has shown a slight reduction in the RSS measurements variance for the night measurement, whereas a considerable reduction was achieved on the RSS measurements resulting from the day experiment. The influence of these results on the performance of centroid, Proximity, Ecolocoation and LSE methods were studied. A considerable localization accuracy was obtained with the RSS enhancement approach that is approximately matched with the best performance given by the Cramer´-Rau Lower Bound (CRLB).
REFERENCES
[1] K. Le, “On angle-of-arrival and time-of-arrival statistics of geometric scattering channels,” IEEE Trans. Veh. Technol., vol. 58, pp. 4257-4264, 2009.
[2] Y. Xie, Y. Wang, B. Wu, X. Yang, P. Zhu, and X. You, “Localization by hybrid TOA, AOA and DSF estimation in NLOS environments,” in the 72nd IEEE Vehicular Technology Conference, 2010.
[3] A. Goldsmith, Wireless Communications, Cambridge University Press, 2005.
[4] T. Rappaport, Wireless Communications: Principles and Practice, 2nd ed., Prentice Hall PTR, 2002.
[5] Raghavendra C.S., Sivalingam Krishna, and Znati Taieb, Wireless Sensor Networks, Springer,2004.
102
ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC
[6] X. Ji and H. Zha, “Sensor positioning in wireless ad-hoc sensor networks using multidimensional scaling,” in IEEE INFOCOM Proceedings, 2004, pp. 2652-2661.
[7] X. Li, “Collaborative localization with received-signal strength in wireless sensor networks,” IEEE Trans. Veh. Technol., vol. 56, pp. 3807-3817, 2007.
[8] N. Patwari, “Location Estimation in Sensor Network,” PhD Thesis, Department of Electrical Engineering, University of Michigan, 2005.
[9] N. Patwari, Y. Wang, “Relative Location in wireless networks,” in the IEEE Vehicular Technology Conference VTC, vol. 2, pp. 1149-1153, May 2001.
[10] H. Hashemi, “Indoor radio propagation channel,” Proceedings of the IEEE,Vol. 81 , Issue 7, pp. 943- 968, 1993.
[11] V. Trees, Detection, Estimation and Modulation Theory, New York: Wiley,1968.
[12] G. Stuber, Principles of Mobile Communication, Springer, New York, 2002.
[13] “PIC microcontroller with 2.4GHz IEEE 802.15.4 transceiver and Zig-Bee stack,” FlexiPanel Ltd, 2008.
[14] “2.4 GHz IEEE 802.15.4 / ZigBee-ready RF Transceiver,” Texas Instru-ments, Chipcon, 2006.
[15] Model 3115 Double-Ridged Waveguide Horn Manual, ETS LINDGREN,2002, Number: SWRS041B,URL:
www.ets- lindgren.com/manuals/3115.pdf.
[16] IEEE std. 802.15.4: Wireless Medium Access Control (MAC) and Physical Layer (PHY) specifications for Low Rate Wireless Personal Area Networks (LR-WPANs), IEEE, 2003, URL:http://standards.ieee.org/getieee802/download/802.15.4-2003.pdf
[17] K. Srinivasan and P. Levis,“RSSI is Under Appreciated,”the Proceedings of the Third Workshop on Embedded Networked Sensors (EmNets),2006.
[18] N. Patwari, A. O. Hero , M. Perkins, N. S. Correal, and R. J. O’Dea, “Relative location estimation in wireless sensor networks,” IEEE Trans. Signal Process., vol. 51, pp. 2137-2148, 2003.
[19] Yongguang2002 C. Yongguang and H. Kobayashi,“Signal Strength Based Indoor Geolocation,”
IEEE International Conference on Commu-nications ICC, 2002, pp. ”436-439”
[20] K. Yedavalli and B. Krishnamachari, “Sequence-based localization in wireless sensor networks,”
IEEE Transactions on Mobile Computing, vol. 7, pp. 81-94, 2008.
103
ISSN: 2005-4238 IJAST Copyright ⓒ 2020 SERSC
[21] Ladha, C. and Sharif, B. and Tsimenidis, C., “Mitigating propaga-tion errors for indoor positioning in wireless sensor networks,” in the IEEE International Conference on Mobile Adhoc and Sensor Systems (MASS2007), 2007, pp. 1-6.
[22] A. J. Weiss, “On the Accuracy of a Cellular Location System Based on RSS Measurements,” IEEE Trans. Veh. Technol., vol. 52, pp. 1508-1518, 2003.
[23] R. Want and A. Hopper and V. Falco and J. Gibbons, “ Active badge location system,” ACM Trans.
Inf. Syst., Vol. 10, 1992, pp. 91-102.
[24] K. Yedavalli, B. Krishnamachari, S. Ravulat, and B. Srinivasan, “Ecolo-cation: A sequence based technique for RF localization in wireless sensor networks,” in the 4th International Symposium on Information Processing in Sensor Networks, 2005, pp. 285-292.
[25] J. Sukhyun Yun, W. Chung and E. Kim, “Centroid Localization Method in Wireless Sensor Networks using TSK Fuzzy Modeling,” in the 8th International Symbosuim on Advanced Intelligent System ISIS, 2007.
[26] N. Bulusu, J. Heidemann, and D. Estrin, “GPS-less low-cost outdoor localization for very small devices,” IEEE Pers. Commun., vol. 7, pp. 28-34, 2000.
[27] Van Trees, Fundamentals of Statistical Signal Processing : Estimation Theory, PTR Prentice-Hall, 1993.