EasyLoc: RSS-Based Localization Made Easy

(1)

Procedia Computer Science 10 ( 2012 ) 1127 – 1133

The International Workshop on Cooperative Robots and Sensor Networks (RoboSense 2012)

EasyLoc: RSS-based Localization Made Easy

Maissa Ben Jamˆaa

a,b

_{, Anis Koubˆaa}

a,c

_{, Yasir Kayani}

a

a_{COINS Research Group, Al-Imam Mohamed bin Saud University (CCIS-IMAMU), Saudi Arabia.} b_{ReDCAD Research Unit, University of Sfax, Tunisia.}

c_{CISTER Research Unit, Polytechnic Institute of Porto (ISEP}_{/IPP), Portugal.}

Abstract

Localization based on Received Signal Strength (RSS) is a key method for locating objects in Wireless Sensor Networks (WSNs). However, current RSS-based methods are ineffective at both deployment and operation design levels since they (i.) usually require a labor-intensive pre-deployment profiling operations to map the RSS to either locations or distances and (ii.) rely on heavy processing operations. These two designs problems limit the possibility of implementing the localization technique on resources constrained sensor nodes and also restrict its scalability and practical use. In this paper, we tackle the challenge of devising a self-organizing and practical RSS-based localization technique that improves on previous approaches in terms of ease of deployment, ease of implementation while still providing a reasonable accuracy. To this end, we come up with a new solution, EasyLoc, a plug-and-play and distributed RSS-based localization method that requires zero pre-deployment configuration. The idea consists in exploiting the available distance information between anchors to derive an online and anchor-specific RSS-to-distance mapping. We show that, in addition to its simplicity, EasyLoc provides location errors of 90% less than 1m and an average error of 0.48m in small environment and 1.8m in large environment.

c

2012 Published by Elsevier Ltd.

Keywords:

Practical RSS-based localization, Zero pre-deployment conﬁguration, Low power wireless networks.

1. Introduction

Consider the problem of localizing a sensor node (referred as unknown node), assuming some anchor nodes with known locations are spread in the environment. This problem has been well-studied in the literature and several tech-niques have been devised. Among these techtech-niques, those based on Received Signal Strength (RSS) are considered as key methods in resource-constrained low-cost networks such as Wireless Sensor Networks (WSNs). In fact, RSS-based localization represents a non expensive solution since (i.) it merely relies on the built-in wireless transceivers and thus does not require any additional hardware as with other techniques (.e.g Time Diﬀerence of Arrival (TDOA)), and (ii.) it exhibits a low computational complexity as RSS can be directly read from the wireless transceiver (e.g. CC2420 transceiver of TelosB/MICAz motes).

Classical RSS-based localization techniques are roughly classiﬁed in two categories: (1) map-based and (2)

model-based. Map-based techniques consist in building a map of radio ﬁngerprints, each refers to a unique cell in the

deployment area. On the other side, model-based techniques aim at establishing a mathematical model capturing the

variation of the RSS as a function of the distance. For both mechanisms, it is central to pass through an oﬀ-line and

tedious environment proﬁling phase to collect empirical data to map the distance (or cells) to RSS. Such cumbersome Open access under CC BY-NC-ND license.

(2)

pre-deployment phase represents a major handicap constraining the wide adoption of RSS-based localization methods and its practical deployment. Further, the resulting static mapping is prone to error as it is not robust to the dynamics of the environment, thus leading to a loss of accuracy. Adapting the mapping to each change is highly cumbersome, in particular for map-based techniques, as it requires the iteration of the proﬁling operation.

2. Problem Statement

To cope with the tedious and cumbersome environment proﬁling phase, recent RSS-based localization studies, such as [1, 2, 3, 4, 5, 6, 7], have provided dynamic RSS to distance/location mapping solutions which rely on a runtime calibration process. Nonetheless the majority of these works, such in [1, 2, 4, 5], have considered high-power radios (WiFi, GSM), which have diﬀerent characteristics as compared to low-high-power radios in sensor networks [8]. Further, works in [1, 2, 3, 5, 6] have relied on centralized approaches as they require a global view on network measurements and also need intensive and heavy loaded processing operations of empirical data such as iterative Matrix manipulations (e.g: SVD, LMS) [2, 3, 5, 6] and Genetic Algorithms [5]. Centralized approaches are known to provide (near) optimal solutions but are not scalable. In addition, heavy loaded computations are complex and memory greedy which inhibit the possibility of their implementation in resources constrained sensor nodes. It turns out that although several works, such as [1, 2, 3, 4, 5, 6] have reduced the deployment phase complexity, they are still

diﬃcult to implement and too heavy to operate on sensor node with limited resources. In particular, authors in [7]

have considered practical issues in their design, but their approach was simply validated using MATLAB simulations

and oﬄine empirical data analysis where no implementation was reported. This raises questions about the ability of

this approach to be implemented.

In our work, we consider practical issues as a main concern in particular in terms of ease of deployment, imple-mentation and operation. We thus devise a solution with low complexity that we eﬀectively implement on TelosB motes. To meet this objective, we propose EasyLoc a lightweight, practical, scalable and distributed RSS-based local-ization method. Our problem can be formulated as follow: Given a set of existing anchor nodes with known location,

how to design a RSS based localization method for low power sensor networks that (1) is sufficiently accurate (2) can be easily deployed with no fingerprinting or pre-deployment profiling (3) adaptive to channel changes, (4) is fully distributed and does not require an intensive centralized computation (5) does not overwhelm the sensor node resources and (6) can be easily implemented on the stringent resources sensor nodes. We present our solution in the

next section. 3. EasyLoc

EasyLoc is a model-based plug-and-play localization method that has four main advantages: (i.) it establishes an anchor-speciﬁc RSS-to-distance mapping that helps to mitigate the eﬀect of hardware imperfections, (ii.) it is easy to deploy as it avoids any pre-deployment calibration phase and builds its mapping in runtime (iii.) it provides a dynamically updated RSS to distance mapping allowing for better localization accuracy and (iv.) takes advantage of simple existing algorithms.

As depicted in Figure 1, EasyLoc comprises two main phases: (i.) Online calibration phase and (ii.) Localization

phase.

3.1. Online calibration phase

The objective of the online calibration phase is to allow each anchor node to build its own RSS-to-distance map-ping by exploiting the knowledge of its distances to other neighbor anchors. This eliminates the need for building a global RSS to distance mapping through cumbersome environment proﬁling. Algorithm 1 presents the least-square RSS-to-distance mapping process running in each anchor node. Each anchor collects a raw of RSS measures via a number of probe messages (prbMsg) exchanged with its neighboring anchors (Step 1 in Figure 1(a)), and stores their averages in its neighbor table (neighTB) (Step 2 in Figure 1(a)). Then, it determines a linear mapping between the RSS and the logarithm of distance, using least-square linear regression (Step 3 in Figure 1(a)). The regression line representing the RSS-to-distance mapping deﬁnes the line that minimizes the sum of squared residuals of the linear

(3)

ID x y ARSS RSS RSS ase . . . . . . . . ID x y ARSS . . . . Log(d) Log(d) Log(d) RSS L (d) RSS libra tion Ph

Anchor nodes exchange probe messages

1

. . . .

Anchor nodes build their neighbor tables

2

3

Anchor nodes generate their model based _{on neighbor table stored data}

Log(d) Log(d)

Ca

l

(a) Online Calibration Phase

RSS RSS e Log(d) RSS Log(d) RSS RSS RSS Log(d) SS Log(d) RSS RSS RSS Log(d) RSS Log(d) RSS RSS RSS zat io n P ha se Log(d) Log(d)

Unknown node broadcasts location requests

1

Log(d) Log(d)

Anchor nodes map RSS and estimate the distance to unknown node

2

Log(d) Log(d)

Unknown node receives distance estimates and computes its location

3

Locali

z

(b) Localization Phase

Figure 1. EasyLoc Calibration and Localization Phases

regression model. It has to be noted that the linear regression model is justiﬁed by the fact that RSS and distance are related by the log-normal shadowing model expressed as:

RS S (d)[dB]= RS S (d0)− 10η log(d/d0) + Xσ[dB] (1)

The model is periodically updated each time a new couple (RSS,distance) feeds an anchor. This makes EasyLoc adaptive to environment changes as well as radio changes.

3.2. Localization phase

This phase aims at determining the absolute location of an unknown node using a trilateration technique, such as Min-Max [9] or Weighted Centroid (WC) [10]. Algorithm 2 represents the localization process running at the unknown node and Algorithm 3 describes the localization process running at each anchor node. Figure 1(b) illustrates the localization steps. First, The unknown node triggers the localization process by broadcasting a burst of w location requests (locReq) to neighbor anchors (Step 1 in Figure 1(b)). Anchor nodes, on their sides, measure the RSSs based on received locReq, average them (RSSAvg) and use their mapping models to determine the corresponding distance (Step 2 in Figure 1(b)). Then, the anchor node sends back to the unknown node a location response message containing the distance estimate, along with its location coordinates (Step 3 in Figure 1(b)). Subsequently, The unknown node waits at least for three distance estimates from anchors to be able to determine its location using trilateration. We point out that the unknown node must eliminate outliers (if any), which are distances greater than

a certain environment-speciﬁc threshold Distth, and rely only on relevant distances to estimate the location. This

helps to ﬁlter out wrong overestimated distances that might compromise the localization accuracy. The maximum distance threshold is environment-dependent and can be set to the maximum distance between any two nodes in the environment.

4. Experimental Results

4.1. Experiments Design

We have implemented EasyLoc on TelosB motes using TinyOS to evaluate its performance and to demonstrate its feasibility and eﬀectiveness through real experiments. Experimental data collection and data analysis of EasyLoc experiments were carried out using iLoc tool [11], a tool that we designed for automating localization experiments

(4)

Algorithm 1. Online Calibration Phase in an Anchor Node Input: Timer1 t1, probe burst size n, averaging window Aw Output: Anchor-speciﬁc RSS-to-distance linear mapping 1: thread probeTransmission (){

2: if (t1 seconds have elapsed) then 3: broadcast a burst of n prbMsg containing 4: the ID and the location coordinates X and Y 5: end if

6: }

7: thread probeReception (){ 8: if (a prbMsg is received) then

9: update neighbor table (neighTB) based on received data 10: (ID,X,Y, RSS[], AvgRSS, distance, Nb of received

prbMsg) 11: end if 12: }

13: thread mappingUpdate (){

14: if (Number of received prbMsg≥ a threshold (Aw)) then 15: Average RSS[] vector values, Update AvgRSS and 16: Generate mapping model parameters a and b using

re-gression

17: // model: RS S = a + b × log10(distance)

18: Clean RSS[] vector and Reset number of received prbMsg

19: end if 20: }

Algorithm 2. Unknown Node Localization Process

Input: Location request burst w, Inter-Packet Arrival time IPI Output: Location of the unknown node

1: send successively w location requests (locReq) with a period IPI // locReq message contains Unknown node ID and IPI information 2: wait()

3: if (at least three distance estimates≤ Distthare received) then

4: estimate location using trilateration (e.g. Min-Max). 5: end if

Algorithm 3. Anchor Node Localization Process Input: Timer1, Timer2 t2, RSS-to-distance mapping Output: Distance to the unknown node

1: if (a locReq is received) then 2: stop Timer1 and Timer2

3: set Timer2 period to t2 (t2= 2 ∗ IPI + ) // varies from an anchor to another 4: increment Nb of received locReq 5: end if

6: if (t2 seconds have elapsed without receiving a locReq) then 7: compute the average RSS (RSSAvg) of received locReq 8: determine the distance corresponding to RSSAvg

using RSS-to-distance mapping

9: send the measured distance to unknown node 10: stop Timer2 and relaunch Timer1 11: end if

Table 1. EasyLoc Algorithms

with WSNs. iLoc comprises two independent applications: (i.) Experiment Control, which is a Java application responsible for the experimental data collection and experiments conﬁguration, and (ii.) Data Analysis, which is a Matlab application that allows empirical data analysis to assess the statistical properties of RSS-based localization.

Our experiments were conducted in two indoor interference-free and static environments with diﬀerent extents: Small

environment of 2m2_{area size (1mx2m) and Large environment of 30m}2_{area size (4}_{.4mx6.8m), where a set of anchor}

nodes were deployed. The objective of considering two diﬀerent-size environments is to assess the impact of area size

on localization error. In each scenario, 12 unknown nodes locations were estimated using the Min-Max [9] algorithm (aka Bounding-Box algorithm). For each unknown node location, a number of locReq bursts (NbBurst) are transmitted to anchor nodes, with a burst size equals to w. Table 2 illustrates the experimental scenarios and settings.

Table 2. Experiment Settings Environment

Extent NbBurst w

Txp (dBm) NbA

Impact of Txp smalllarge 105 105 {0,-15,-25}{0,-15} 8 Impact of NbA

small 5 10

0 {4,6,8}

large 10 5 {4,8}

The performance of EasyLoc was evaluated by analyzing the (i.) distance error, which is the diﬀerence between the real and estimated distances between anchor nodes and the unknown node, (i.) location error, which is the dif-ference between the real and estimated location of the unknown node. Errors were assessed through their cumulative distribution functions (CDF) in addition to their means, Standard Deviations (StD) and 95% of Conﬁdence Interval (CI). The impacts of both transmission power (Txp) and number of anchors (NbA) on distance and location errors were investigated.

4.2. Results Analysis

The experimental study reveals several results that we summarize in the following observations.

Observation 1. Increasing the transmission power and the number of anchors reduces the location/distance errors.

It is clear from Figure 3 and Figure 4 that the location and distance estimation accuracy becomes worse for lower transmission powers, and for smaller numbers of anchors. In fact, it is interesting to observe that the CDF curves of location/distance errors grow more smoothly with low transmission powers (i.e. −15dBm and −25dBm) than with the maximum transmission power (0dBm). Further, it can be noted in Table 3 that the average location/distance errors are

(5)

minimized with 0dBm transmission power. This conﬁrms the feeling about the negative impact of low-power links on achieving good localization accuracy, which renders localization more challenging in such circumstances. However, it has to be noted that high transmission powers result in higher probability of outliers, which are the ratio of discarded distances due to exceeding the maximum threshold. The same observations also hold with respect to the impact of the number of nodes. The CDF curves in addition to average values of location/distance errors agree that increasing the number of anchors contributes to reducing localization error. In summary, the best accuracy occurs with the maximum transmission power 0dBm and maximum number of anchors (8 anchors). We refer to this conﬁguration as the best

case.

Observation 2. EasyLoc provides a good distance/location accuracy that is dependent on the environment size and comparable to those found with traditional ﬁngerprinting methods. It can be observed in Table 3 that the average

location error varies between 33cm and 71cm in the small 2m2_{environment, and between 1.25m and 2.9m in the large}

30m2_{environment, respectively. We also observe from Figure 3(d) and Figure 4(d) that, in best case, 90% of location}

errors are less than 1m in the small environment, and 90% of location errors are less than 3m in the large environment. This demonstrates that EasyLoc, in addition to its ease of deployment, achieves good localization accuracy propor-tional to the environment size and this accuracy is comparable to those found with RSS-based localization relying

on extensive proﬁling. For instance, in [12], the minimum location error reported was 90% less than 4m in a 100m2

environment.

Observation 3. the Min-Max trilateration technique reduces more eﬀectively the impact of distance errors on location errors with larger number of anchors regardless the transmission power. Indeed, we conclude from Table 3

location/distance errors decrease when the number of anchor nodes increases. The interesting point is that the location

errors for diﬀerent transmission powers are close to each other although corresponding distance errors are clearly more

diﬀerent. This infers that increasing the number of anchors will be able to compensate for the distance errors of low

transmission powers. This represents a positive issue for EasyLoc as it will allow for using low-power transmission without signiﬁcantly compromising location errors.

Observation 4. EasyLoc is adaptive to radio changes. It is known that RSSs signiﬁcantly vary over long time

spans [8]. It implies that static RSS-to-distance mapping will not be accurate over time. EasyLoc eﬀectively copes with this issue as it adapts to radio changes via periodic calibration refresh. Figure 2 shows how an anchor mapping curve evolves over time in response to radio changes for 5 diﬀerent locReq Bursts.

50 Small | Txp=-25dBm | NbA=8 -70 -60 -50 R SS locReq Burst Nb1 locReq Burst Nb2 locReq Burst Nb3 locReq Burst Nb4 locReq Burst Nb5 0 100 200 300 400 500 -100 -90 -80 R Distance (cm)

Figure 2. Mapping Curves Evolution for one Anchor

5. Conclusion

In this paper, we proposed EasyLoc, a novel RSS-based localization method that require zero pre-deployment proﬁling operations. The main idea was to exploit the knowledge of distance to neighbor anchors to derive an anchor-speciﬁc RSS to distance mapping. This demonstrates that RSS-based localization can be easily deployed, in contrast

(6)

1 Small | Tx power = 0 dBm 0.6 0.8 1 ba bil ity 0 0.5 1 1.5 2 2.5 0 0.2 0.4 Pr ob Nb Anchors=4 Nb Anchors=6 Nb Anchors=8 Distance Error (m)

(a) Impact of Nb of Anchors

1 Small | Nb Anchors = 8 0.6 0.8 1 ba bil ity 0 0.5 1 1.5 2 2.5 0 0.2 0.4 Pr ob tx power=0 dBm tx power=-15 dBm tx power=-25 dBm Distance Error (m) (b) Impact of Tx power 1 Small | Tx power = 0 dBm 0.6 0.8 1 ba bil ity 0 0.5 1 1.5 2 0 0.2 0.4 Pr ob Nb Anchors=4 Nb Anchors=6 Nb Anchors=8 Location Error (m) (c) Impact of Nb of Anchors 1 Small | Nb Anchors = 8 0.6 0.8 1 ba bil ity 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 Pr ob tx power=0 dBm tx power=-15 dBm tx power=-25 dBm Location Error (m) (d) Impact of Tx power Figure 3. Small environment: Localization accuracy

1 Large | Tx power=0 dBm 0.6 0.8 1 ba bil ity 0 2 4 6 8 0 0.2 0.4 Pr ob Nb Anchors=4 Nb Anchors=8 Distance Error (m)

(a) Impact of Nb of Anchors

1 Large | Nb Anchors=8 0.6 0.8 1 ba bil ity 0 2 4 6 8 0 0.2 0.4 Pr ob tx power=0 dBm tx power=-15 dBm Distance Error (m) (b) Impact of Tx power 1 Large | Tx power = 0 dBm 0.6 0.8 1 ba bil ity 0 2 4 6 8 0 0.2 0.4 Pr ob Nb Anchors=4 Nb Anchors=8 Location Error (m) (c) Impact of Nb of Anchors 1 Large | Nb Anchors=8 0.6 0.8 1 ba bil ity 0 1 2 3 4 0 0.2 0.4 Pr ob tx power=0 dBm tx power=-15 dBm Location Error (m) (d) Impact of Tx power Figure 4. Large Environment: Localization Accuracy

(7)

Table 3. Small and Large Environments Experiment Results Min Max Mean StD CI Outliers

% Small En vir onment: Estimated distances > 3m ar e ﬁlter ed Distance err ors NbA = 8 Txp=0dBm 0.0024 2.3686 0.452 0.4187 0.036 6.3063 Txp =-15dBm 0.0010 1.5797 0.637 0.3911 0.0363 1.1111 Txp=-25 dBm 0.0041 1.692 0.7534 0.383 0.0402 1.1331 Txp = 0dBm NbA=4 0.0264 2.3586 0.7262 0.4801 0.0781 34.6847 NbA=6 0.0042 2.3586 0.6761 0.5108 0.0585 21.8667 NbA=8 0.0024 2.3686 0.452 0.4187 0.036 6.3063 MinMax err ors NbA = 8 Txp=0dBm 0.02 0.8981 0.3378 0.2229 0.0522 6.3063 Txp =-15dBm 0.05 0.831 0.4443 0.2433 0.0616 1.1111 Txp =-25dBm 0.0472 0.8023 0.4148 0.2157 0.0598 1.1331 Txp = 0dBm NbA=4 0.0403 1.4235 0.7157 0.3259 0.0839 34.6847 NbA=6 0.0673 1.6046 0.6107 0.3193 0.0748 21.8667 NbA=8 0.02 0.8981 0.3378 0.2229 0.0522 6.3063

Min Max Mean StD CI Outliers % Lar ge En vir onment: Estimated distances > 7m ar e ﬁlter ed Distance err ors NbA = 8 Txp=0dBm 0.0020 6.3498 1.2341 1.2381 0.094 27.3719 Txp =-15dBm 0.0441 6.5942 2.3869 1.468 0.102 6.4706 Txp = 0dBm NbA=4 0.0052 6.5142 1.8644 1.4777 0.1689 38.75 NbA=8 0.0020 6.3498 1.2341 1.2381 0.094 27.3719 MinMax err ors NbA = 8 Txp=0dBm 0.0721 3.8446 1.2589 0.9491 0.1698 27.3719 Txp =-15dBm 0.1756 3.2047 1.8083 0.8185 0.1465 6.4706 Txp = 0dBm NbA=4 0.2415 6.6242 2.9147 1.717 0.3209 38.75 NbA=8 0.0721 3.8446 1.2589 0.9491 0.1698 27.3719

to tradition cumbersome methods. The eﬀectiveness of EasyLoc was experimentally demonstrated through

exten-sive measurements in two indoor environments. In the future, we will work towards consolidating the intelligence of the localization algorithm to ensure the correctness of the anchor-speciﬁc RSS-to-distance mapping and discard potentially faulty nodes. Second, we will assess the robustness of EasyLoc by considering large-scale deployments in diﬀerent building structure.

Acknowledgments

This work is funded by R-Track project under the grant 8-INF-2008 of the National Plan for Sciences and Tech-nology (NPST).

References References

[1] Y. Gwon, R. Jain, Error Characteristics and Calibration-Free Techniques for Wireless LAN-based Location Estimation, in: Proceedings of the second international workshop on Mobility management & wireless access protocols, MobiWac’04, ACM, New York, NY, USA, 2004, pp. 2–9.

[2] H. Lim, L. C. Kung, J. C. Hou, H. Luo, Zero-Conﬁguration, Robust Indoor Localization: Theory and Experimentation, in: Proceedings of the 25th IEEE INFOCOM conference, Barcelona, Spain, 2006.

[3] M. Dominguez-Duran, D. Claros, C. Urdiales, F. Coslado, Dynamic calibration and zero conﬁguration positioning system for WSN, in: The 14th IEEE MELECON conference, Ajaccio, 2008.

[4] A. Varshavsky, D. Pankratov, J. Krumm, E. Lara, Calibree: Calibration-Free Localization Using Relative Distance Estimations, in: Proceed-ings of the 6th International Conference on Pervasive Computing, Sydney, Australia, 2008.

[5] K. Chintalapudi, A. Padmanabha Iyer, V. N. Padmanabhan, Indoor Localization Without the Pain, in: Proceedings of the 16th ACM MobiCom conference, Chicago, Illinois, USA, 2010.

[6] J. Larranaga, L. Muguira, J.-M. Lopez-Garde, J.-I. Vazquez, An Environment Adaptive ZigBee-based Indoor Positioning Algorithm, in: Proceedings of the IPIN conference, Zurich, Switzerland, 2010.

[7] A. M. Bernardos, J. R. Casar, P. Tarro, Real Time Calibration for RSS Indoor Positioning Systems, East (September).

[8] N. Baccour, A. Koubaa, M. Zuniga, H. Youssef, C. Boano, M. Alves, Radio Link Quality Estimation in Wireless Sensor Networks: a Survey, ACM Trans. Sen. Netw. 8.

[9] K. Langendoen, N. Reijers, Distributed Localization in Wireless Sensor Networks: a Quantitative Comparison, Comput. Netw. 43. [10] J. Blumenthal, R. Grossmann, F. Golatowski, D. Timmermann, Weighted Centroid Localization in Zigbee-based Sensor Networks, in: IEEE

WISP, Alcala de Henares, 2007. [11] iLoc Overview (2012).

URLhttp://iloc.rtrackp.com

[12] G. Zanca, F. Zorzi, A. Zanella, M. Zorzi, Experimental Comparison of RSSI-based Localization Algorithms for Indoor Wireless Sensor Networks, in: Proceedings of REALWSN workshop, Scotland, 2008.