CULo: Coordinate User Location System for Indoor Localisation

Teddy Mantoro1, Wendi Usino2, Andriansyah3

1Department of Computer Science, The Australian National University, ACT 0200, Australia 2Faculty of Information Technology, University of Budi Luhur, Jakarta, Indonesia

3Research and Information Technology Bureau, The Capital Market and Financial Institutions Supervisory Agency, Jakarta, Indonesia

This paper proposes CULo, an approach to estimate a coordinate user location in Location-Aware Computing Environments, by

measuring a user location based on proximate sensor data. Signal strength and signal quality of IEEE 802.11 (WiFi) are proposed to be used in collecting proximate sensor data to determine user location, after evaluating the proximate sensors such as Bluetooth, IrDa and WiFi. While service to the users continue in operation across changing circumstance and in a seamless manner, the users location are estimated and at the same time proximate sensors and precise sensors take role to localisation service delivery, directly, to the users location. Our previous work proposes the ηk-Nearest Neighbor algorithm to estimate symbolic user location and in this work, we propose the use of multivariate regression estimation in estimating a coordinate user location in indoor environment. The features for context aware location protocol, which has strong consideration to implement both approaches, in estimation user location, are also discussed.

Index Terms—User Location, Localisation, Ubiquitous Computing, Pervasive Computing.

I. INTRODUCTION

ocation-Aware Computing is a rapidly growing field in the area of Context-Aware Computing. User and equipment

location is the target of developing location-aware applications. Unfortunately a range of mobile devices (Laptop, PDA, Smart Phone) in the market are still lacking satisfactory location technology.

Location-Aware Computing which promises accuracy, economy and ease of deployment, are always seen to be under construction. Numerous location models have been proposed in different domains and can be categorized into two classes, i.e. symbolic or descriptive (hierarchical, topological) location such as a city or a room, and coordinate (Cartesian, metric or geometric) location such as GPS or active bat.

As user location is an important part in pervasive computing, symbolic location is preferred over coordinate location in the user’s daily activity. The use of coordinate location for human-serving can be converted into symbolic location, its more natural human location scale which, except in special cases, makes daily communication easier.

Precise sensors such as a pressure chair sensor, a magnetic phone sensor and a keyboard activity sensor are very straight forward in capturing precise and correct user location to be recorded at spatio-temporal sensor database, but Bluetooth, IrDA and IEEE 802.11 (WiFi) as proximate sensors are not.

Our environment was equipped with the sensors above, but when we evaluated three proximate sensors, we chose WiFi as the best sensor candidate for a user to transparently continue operating across changing circumstance in a seamless manner. Unfortunately WiFi’s signal strength and quality is found to fluctuate greatly in the estimation of user location even though there are no moving objects in the hot-spot area.

Location information does not contain user location only, it is a set of object data that needs user identity. The location

information can be recognized as an occurrence when it satisfies the standard format of the sensor data. The complete set of location sensor data contains user identity, location identity, time and state, and without user identity no sensor data will be recorded in the spatio-temporal sensor database.

User location information is very important in pervasive computing because it may change the paradigm: providing service directly to where the user is located, not to the server and then by the user accessing the service to the server such as POP3, IMAP or SMTP protocol on email service. It can change the way of thinking of delivering service, from delivering service without need to know user location to a new paradigm, delivery of service directly to user location based on user profile.

This paper presents various sensors for sensing user location in an Active Office followed by context-aware location protocol, location awareness in pervasive environment, proximate user location and multivariate regression for estimation of coordinate user location. Exploratory data analysis and the result of the multivariate regression are also presented.

II. VARIOUS SENSORS FOR SENSING USER LOCATION

As an implementation model of Intelligent Environment (IE) architecture, an Active Office was developed which consists of several normal rooms with minimal intrusive detectors and sensors, and minimal badging of people to make user interaction with the environment easier.

An Active Office is equipped with two types of sensors i.e. fixed/precise sensors and proximate sensors (Fig. 1). The magnetic door sensor, pressure chair sensor, magnetic phone sensor, keyboard and mouse activity sensor, and RFID are used as fixed sensors and The Bluetooth and WiFi are used as proximate sensors to detect current state/context and to understand/recognize several contexts such as context location,

L

ISAST Transactions on Computers and Software Engineering, No. 1, Vol. 2, 2008 Teddy Mantoro et al.: CULo: Coordinate User Location System for Indoor Localisation1

context activity and context actions for IE to provide service based on user situations.

FIG. 1 TYPE OF SENSORS AT THE ACTIVE

OFFICE

In identifying user identity and user location in IE, our approach is to use the wireless connections in devices that users normally carry for other purposes, for example, a smart phone or a phone PDA. We have not designed or used any other small sensor tag, such as the active badge/bat system [1,2,3] because we don’t want to add more items for the user to carry such as keys, glasses, mobile phone, etc. The smart phone or phone PDA is not only handy because it is small, but also because it could be used as user identity since it has a unique MAC address and is also a personal device. The location of these devices, and hence the person with them, is determined by a mixture of precise, proximate and predicted location sensors. The data from these sensors is turned into a predictor to precisely locate the device, and thus the person. Once a user is located such services can be delivered based on the current situation from a resources manager.

The more sensors in the small devices that the user carries, the more information the system gets to determine user location.

III. CONTEXT-AWARE LOCATION PROTOCOL

Each type of sensor has a different positioning technique to be recorded and each position has a different location property. The five protocol features proposed as con-text-aware location protocols are as follows:

A. Location Sensor Data

Raw location sensor data is any data item which is collected directly from the sensors that report user or equipment locations. Some sensors such as door sensor, chair pressure sensor, etc. can directly report precise location, in both coordinate location and symbolic location formats. But some sensors need to use complex calculation to estimate location, such as WiFi or Bluetooth, using signal strength and signal quality. After long calculation, sensor data only shows proximate location, which is not as accurate as the precise location.

B. Estimation Location

Estimation location uses proximate sensor data to convert to proximate location. In the case of the WiFi proximate sensor,

the WiFi signal strength and quality is used but is found to greatly fluctuate. It reaches to 33 dBm different in the same location in the range -99 to 0. A number of techniques can be used for this purpose, depending on what location information is needed. Multivariate Regression may be used for coordinate location, and an instance-based machine learning algorithm for symbolic location.

C. Standard Location Format

Location information can not just contain user location alone, it needs other data items to recognize as an occurrence in a database. Therefore, a complete set of location sensor data contains user identity, time and state. User identity or equipment identity is a key point to be recorded in user location database. The standard format of the instance of user location occurrence is below:

standard format = {user-id | equipment-id + location-id + time + state }

D. Fusion of Location and Presentation

The Context-Aware Computing environment by nature deals with large amounts of various sensor data. As time goes and pervasive applications grow, the amount of data that is collected, stored and processed will steadily grow as well. This is likely to make an application drop speed in responding to user queries. To over-come that problem a spatio-temporal data model is used to assist the mobile object query to gain access to the spatio-temporal database. The queries involve projection, selection, joining, and aggregation operation for the spatial and the temporal range of the mobile object.

E. Moving-Object Location (Mobility)

The pattern of user mobility or equipment change location developed from an object’s movement history [4]. The most probable location/position of an object location was predicted using an algorithm which analyzes the object location pattern in a spatio-temporal database combined with other semantic in-formation.

IV. LOCATION AWARENESS IN PERVASIVE COMPUTING

Basically in Location Awareness concept, it asks a simple question: “Where am I?” But in the Pervasive environment, that question reflects the user identity and the user location. Location Awareness brings two important concepts: Context Identity and Context Location. Both of the concepts have strong influences in the areas of Location Object, User Mobility and Localisation.

Location Object deals with finding user location in the precise, estimated or predicted location. User Mobility deals with the changes in a user’s location from one place to another, if there is no change of service then there is no user mobility [5]. Localisation deals with distributed data processing, i.e. local data when the user is on the move.

In IEs, the estimate location and predicted location are

ISAST Transactions on Computers and Software Engineering, No. 1, Vol. 2, 2008 Teddy Mantoro et al.: CULo: Coordinate User Location System for Indoor Localisation2

different concepts. Estimate location is the use of proximate sensors to estimate user location. In our environment, a machine learning algorithm based on Self Organizing Map [6] and also a Multivariate analysis approach to find the best match of user location are used in the Active Office. Predicted location is used when fixed sensors cannot sense user location in the Active Office. Predicted location is the use of probabilistic method to predict current user location based on patterns of the historical data of fixed sensors and proximate sensors data [7]. In our experiment, the routines are written in C++ and Python/PyGTK, and the spatio-temporal sensor database uses MySQL in a UNIX environment.

Context Location (loc) is a predicate relation between object and its location. In this case, the object requests its position or location from the IE by using the pull method in the environment.

loc(object, location) (1) The object itself can be a user, a chair, a phone, a door, a

room, a building, etc. For example: loc(u1,room1) loc(chair1, room1) loc(room1, bldg108) loc(bld10, ANU) In the case where the object is a user, when he moves from

one location to another, the predicate location could deduce user mobility.

mobilityi

locationobjectloc =∞

=)

1,(U (2)

for object as a person A user could get his location in the environment when he

brings his mobile device, such as a Mobile Phone or a PDA with either Bluetooth or WiFi enabled or both. The Bluetooth server as a master Bluetooth will scan the user’s Bluetooth enabled device as a client device. At the same time, when the user’s WiFi enabled device is in a certain area, the signal strength and signal quality will be measured to get the estimated location.

loc(u1,room1):-

btloc(room1,u1,p800) .or. wifiloc(PDA,u1,room1)

The ambiguity of possible reporting of multiple locations

for the user is resolved by using the latest user location. In the case where the user has two devices with two connectivity’s capability, using WiFi and Bluetooth for instance, the active office environment will check both devices, then use the latest user location and store it in the IE repository as the current location. The service status will capture directly from the

resources manager which accesses the IE repository and the user model. The IE repository and the user model are holding the information from every sensors/devices and the relationship between user identity and latest sensor sensors/devices data.

In the case of Bluetooth, the Bluetooth server scans the room until the client is found and then asks the resolve server to find the user identity (carry, see below). The Bluetooth server then sends the user identity, location and the MAC address of the Bluetooth mobile phone.

btloc(room1,u1,p800):-

btscan(room1,p800) .and. carry(p800,u1)

The environment tracks the mode of the mobile phone for

user activity and responds by checking status. btstat (room1,u1,p800, “silent”) :-

btloc (room1,u1,p800) .and. mode (p800, “silent”)

In the case of WiFi, the WiFi server is always in “on-

listening” mode. When a mobile WiFi client reports the signal strength and signal quality, it sends the estimation of user location, followed by resolution of the MAC address of the PDA which leads to the user’s identity.

wifiloc(PDA,u1,room):-

wifiscan(PDA, room1) carry(PDA,u1)

FIG. 2 Resolving: User Identification, Device

Identification and MAC Address Resolve-database contains 3 basic components for user-

identity purposes i.e.: user identification, device identification and MAC address of the device. Fig. 2 shows the possible resolution of user identity when a user carries a device. It is possible for a user to have m (many) devices and possible for a device to have n (many) MAC ad-dresses implying that it is possible for a user to have m x n (many) MAC addresses.

V. EVALUATION OF PROXIMATE USER LOCATION

In an Active Office, user location can be categorized as Precise Location, Proximate Location and Predicted Location. The Precise Location and Proximate Location are based on the

ISAST Transactions on Computers and Software Engineering, No. 1, Vol. 2, 2008 Teddy Mantoro et al.: CULo: Coordinate User Location System for Indoor Localisation3

sensor’s capability in covering an area and Predicted Location is based on the history data of both Precise Location and Proximate Location [8]. Fixed sensors such as a chair sensor, a phone sensor and a keyboard activity sensor are very straight forward for user location. The information which the sensors capture is precise and correct. For example, when a user sits on a chair with an embedded chair sensor in a certain room, a set of sensor data is available to record to the spatio-temporal database directly.

Proximate location, which is detected by Wireless PAN/LAN, has two functions, i.e. to access the network and to sense user location within the scale of a room or an office. Bluetooth and IrDA (Infrared), as a wireless personal area network, favours low cost and low power consumption, over a range and peak speed. On the other hand, WiFi as a wireless local area network, favours higher speed and greater range but has higher cost and power consumption. The range of Bluetooth to sense another Bluetooth in an indoor environment, such as an Active Office, is about 3 meters for class 2 and 25 meters for class 1. IrDA is not very useful in sensing a user’s location because when there are many users in the area that is covered by IrDA, only one user will be recognized. IrDA connections are limited to two devices with direct line of sight.

Bluetooth permits scanning between devices: when Bluetooth-capable devices come within range of one another, the location of one Bluetooth will be in the range of the other Bluetooth. Bluetooth-capable devices can be used as sensors or an access point to sense a user with Bluetooth capable devices. Our experiment unfortunately shows that Bluetooth signal strength is not useful enough to sense a user’s location. The Bluetooth access point can be used as a sensor for several rooms within the range without measuring any signal strength. For example, when a user is close to a certain access point, the user’s location will be proximately close to the access point and could represent user location from several rooms.

This situation makes WiFi the best candidate for estimating user location in an indoor environment such as home or office. WiFi’s signal strength and quality can be used in estimating user location. Unfortunately, in estimating user location, WiFi’s signal has significant variation in signal strength and signal quality even though there were no moving objects in the hot-spot area. Measurement of signals between Morning time and Afternoon time were found to be significantly different. It reached a 33 dBm difference in a range of -99 to 0 dBm of the signal strength. In our previous work, we introduced the ηk-Nearest Neighbor algorithm for use in instance-based learning methods of the machine learning algorithm to estimate symbolic user location. This approach used normalization training data and finding the maximum number of locations that appeared from the first nearest ten (k=10) of the ηk-Nearest Neighbor Algorithm. This approach was not based on triangulation, interpolation or mapping algorithm as in [9,10,11], using ‘trapping method’ to trap radio signals that lead to symbolic location. The result is quite promising, it

reached nearly 96% correctness in estimating symbolic user location [4,8].

For coordinate location, multivariate regression is proposed. The next section will discuss the preliminary analysis of multivariate regression for estimation of coordinate user location.

VI. MULTIVARIATE REGRESSION FOR ESTIMATION OF

COORDINATE USER LOCATION

A CULo is an approach to estimate a coordinate user location in Location-Aware Computing Environments, by measuring a user location based proximate sensor data. This approach uses multivariate regression for estimation of coordinate user location. The concept can be described as follows:

Let Y denote the n by p matrix of dependent or response variables, X denotes the n by (k+1) matrix of independent or predictor variables, B denotes the (k+1) by p matrix of unknown coefficients, and ε denotes the n by p matrix of errors; the multivariate regression could be represented by matrix notation as

Y = XB + ε (3)

The unknown coefficient matrix B could be estimated using least squares method by minimizing the trace, the sum of diagonal element, of the p by p matrix (Y - XB)'(Y-XB). The estimators of B are given by

B̂ = (X'X)-1X'Y (4)

Detailed discussion of multivariate regression can be seen in many multivariate statistics books, such as (Rencher 1998).

TABLE 1. EXAMPLE OF WIFI’S SIGNAL

STRENGTH AND QUALITY IN THE CORRIDOR Signal Strength Signal Quality

AP1 AP2 AP3 AP4 AP5 AP6 AP1 AP2 AP3 AP4 AP5 AP6 Corri-

dor

-49 -81 -91 -99 -99 -99 1.39 0.35 0.7 0 0 0 c1 -52 -82 -99 -99 -99 -99 1.39 0.35 0 0 0 0 c1

-57 -86 -99 -99 -99 -99 1.39 0.35 0 0 0 0 c1

-49 -91 -99 -99 -99 -99 1.39 0.35 0 0 0 0 c1 -57 -99 -99 -99 -86 -99 1.39 0 0 0 2.9 0 c1

-57 -99 -81 -99 -99 -99 1.39 0 0.7 0 0 0 c1

-43 -99 -99 -99 -99 -99 1.39 0 0 0 0 0 c2 -49 -99 -99 -99 -99 -99 1.39 0 0 0 0 0 c2

-44 -99 -85 -99 -99 -99 1.39 0 0.7 0 0 0 c2

-56 -90 -99 -99 -99 -99 1.39 0.35 0 0 0 0 c2 -42 -87 -99 -99 -99 -99 1.39 0.35 0 0 0 0 c2

-53 -99 -99 -99 -99 -99 1.39 0 0 0 0 0 c2

In the context of coordinate user location, the matrix X

consists of twelve variables, i.e. six signal strengths (SS1 to SS6) and six signal qualities (SQ1 to SQ6) that are results from 6 WiFi's access points, whereas the matrix Y represents the

ISAST Transactions on Computers and Software Engineering, No. 1, Vol. 2, 2008 Teddy Mantoro et al.: CULo: Coordinate User Location System for Indoor Localisation4

coordinate of location in Euclidian space and consists of two variables, i.e. abscissa and ordinate. Table 1 shows the example data that measured WiFi’s signal strength and quality in the square corridor in our building.

A. Exploratory Data Analysis

In this section, we examine the characteristics of signal strength and signal quality data. Please see the boxplot graphic (Fig. 3) for signal strength and signal quality. SS1, SS2, SS3 and SS5 (SS4 and SS6 are ignored due to homogenous constant data). See also the similar boxplot graphics for SS1, SS2 and SS3 that derive from boxplot graphics of signal strength (Fig. 4).

FIG. 3 BOXPLOT FOR WIFI’S SIGNAL STRENGTH

AND QUALITY

The graphics reveal that: 1. In terms of signal strength, only SS1, SS2, SS3 and (to a

small extent) SS5 are potentially important to be good predictors in estimating coordination location. For signal quality, SQ3 is the only variable that is important, the others can be ignored.

2. As expected, data variation is high in each location. The minimum five repeated measurements in each location show significant dispersion; some, in fact, demonstrate the existence of outliers. This in turn affects the quality of the variables as good predictors.

FIG. 4 BOXPLOT FOR SIGNAL STRENGTH 1, 2

AND 3

B. Results from Multivariate Regression

The corridor locations are transformed into abscissa (x) and ordinate (y) forming a rectangle. Corridor 1 is the starting point at (0,0), and each subsequent point in the corridor adds 2 points either to the x-axis or y-axis. The turning points are c9, c15 and c24. The numbers of points in each corridor are not the same: the first has 8 points, the second has 6 points, the third has 9 points and the fourth has 5 points (Fig. 5). This in turn means that the two additional points, c17-c21, must be transformed into (12,8), (12,7), (12,6), (12,5), and (12,4)

ISAST Transactions on Computers and Software Engineering, No. 1, Vol. 2, 2008 Teddy Mantoro et al.: CULo: Coordinate User Location System for Indoor Localisation5

respectively.

• Data with repetition. X = -15.92 - 0.16*SS1 + 0.15*SS2 - 0.19*SS3 +

2.17*SQ5 (R-square = 0.7742) Y = 32.91 - 0.15*SS1 + 0 7*SS2 + 0.21*SS3 -

6.39SQ2 - 2.21*SQ5 (R-square = 0.8295) The result indicates that SS1, SS2, SS3 and SSQ5 are consistently important to predict the coordinate, while SQ2 is only useful to predict the ordinate.

• Without repetition. X = - 0.25*SS1 - 0.20*SS3 (R-square = 0.8715) Y = 29.76 - 0.21*SS1 + 0.31*SS3 - 1.80*SQ5 (R-

square = 0.8940) In this case of without repetition, the result indicates that only SS1 and SS3 are consistently important to predict the coordinate, while SQ5 is only useful to predict the ordinate.

In general, both models, either with or without repetition, indicate that SS1 and SS3 are consistently important to predict the coordinate, while SQ5 is only useful to predict the ordinate. Repetition in this case is the selection of data in the same location based on the maximum signal strength and signal quality so that each location only has one representative data.

FIG. 5 The Multivariate Regression of WiFi’s Signal Strength and Quality of the Corridor

However, in terms of R-square (the coefficient of

determination) - the statistical measure of how well the estimated points (marked by the boxes in Fig. 5) approximates the actual points (marked by the diamonds in Fig. 5) where the maximum value of the R-square of 1.0 indicating the estimated points fit the actual points - the model without repetition is better than the model with repetition.

The coordinate user location using Multivariate Regression in our experiment gets 82.24% correctness in estimation of user location. These results imply that variation in data may be

reduced in order to improve the model performance by filtering the data chosen for the maximum signal strength and quality in a location. However, these models can only be applied in the corridor.

The result of the estimation of the symbolic user location (such as the use of ηk-Nearest Neighbour Algorithm, for instance k = 10) can not be compared with the result of the estimation of the coordinate user location (such as the use of the Multivariate Regression), because the symbolic user location is based on a name (such as room E213) and coordinate user location is based on abscissa (x) and ordinate (y). Moreover, different precision and scale are used in the measurements of the raw training data on both approaches.

C. Some Suggestions

• Use the combination of symbolic and coordinate user location estimation methods; the ηk-Nearest Neighbour Algorithm has given very good results in the estimation of symbolic user location. The results may possibly be enhanced by partitioning the areas covered by all the access points into grids. Each grid is then classified as a different symbolic location. For each grid, the coordinate user location is applied to predict the coordinate and has its own prediction function.

• Incorporate time and distance from each access point information into a regression model as additional predictor variables.

VII. SUMMARY

This paper describes location awareness and context-aware protocol in pervasive computing, through the various types of sensors with which our Active Office is equipped. User location information is an important aspect in location awareness, especially in an indoor environment, it can be categorised as Precise Location, Proximate Location and Predicted Location. The Precise Location and Proximate Location are based on the sensor’s capability in sensing an area and Predicted Location is based on the history data of both Precise Location and Proximate Location.

This paper also presents the evaluation of proximate sensors and proposes WiFi as the best sensor candidate for a user to transparently continue operation across changing circumstance in a seamless manner. Unfortunately WiFi’s signal strength and quality is found to fluctuate greatly in the estimation of user location even though there were no moving objects in the hot-spot area. Our previous work proposed the use of the ηk-Nearest Neighbor algorithm in estimation of symbolic user location and in this work we propose the use of multivariate regression in estimation of coordinate user location. Exploratory data analysis and the result of the multivariate regression for estimation of coordinate user location are also presented.

User location information, as a core of Location Awareness, is very important in pervasive computing because it can change the paradigm: providing service directly to where the

ISAST Transactions on Computers and Software Engineering, No. 1, Vol. 2, 2008 Teddy Mantoro et al.: CULo: Coordinate User Location System for Indoor Localisation6

user is located, not to the server and then by the user accessing the service to the server. It can change the way of thinking of delivering service, from deliver service without need to know user location to a new paradigm: deliver service directly to user location based on user profile.

REFERENCES

[1] Harter, A. and A. Hopper. A Distributed Location System for the Active Office. IEEE Network 8(1): pp. 62--70. 1994.

[2] Harter, A., A. Hopper, et al. The Anatomy of a Context-Aware Application. Wireless Networks 1: 1-16. 2001.

[3] Want, R., A. Hopper, et al. The Active Badge Location System. ACM Transactions on Information Systems Vol. 40 (No. 1): pp. 91-102. 1992.

[4] Mantoro T, W Usino. Improving the Accuracy of User Mobility Patterns for Intelligent Environments. The 3rd International Conference on Computer Science & Information Systems, ATINEV, Athens, Greece, July 23-24, 2007.

[5] Mantoro, T. and C. W. Johnson. User Mobility Model in an Active Office. LNCS 2875, European Symposium on Ambient Intelligence (EUSAI’03), Eindhoven, The Netherlands.

[6] Mantoro, T. Distributed Support for Intelligent Environments, PhD Dissertation, The Australian National University, Canberra, Australia, June 2005.

[7] Mantoro, T. and C. W. Johnson. Location History in a Low-cost Context Awareness Environment. Workshop on Wearable, Invisible, Context-Aware, Ambient, Pervasive and Ubiquitous Computing, ACSW 2003, Adelaide, Australia.

[8] Mantoro, T. and C. W. Johnson. Instance-Based Learning Methods for Estimation of Symbolic User Location in Pervasive Computing Environments. IEEE International Symposium on a World of Wireless, Mobile and Multimedia Networks (WoWMoM), Taormina, Italy. 2005.

[9] Bahl, P. and V. N. Padmanabhan. Radar: An in-Building RF-Based User Location and Tracking System. Proceedings of the IEEE Infocom 2000, Tel-Aviv, Israel.

[10] Hightower, J. and G. Borriello. Particle Filter for Location Estimation in Ubiquitous Computing: A Case Study. Ubiquitous Computing (UbiComp 2004), Nottingham, UK.

[11] Rencher, A. C. Multivariate Statistical Inference and Applications. New York, John Wiley Publisher. 1998.

[12] Schwaighofer, A., M. Grigoras, et al. GPPS: A Gaussian Process Positioning System for Cellular Networks. Neural Information Processing Systems, NIPS 2003, Vancouver and Whistler, British Columbia, Canada.

[13] Smalagic, A. and D. Kogan. Location Sensing and Privacy in A Context-Aware Computing Environment. IEEE Wireless Communications (October 2002).

ISAST Transactions on Computers and Software Engineering, No. 1, Vol. 2, 2008 Teddy Mantoro et al.: CULo: Coordinate User Location System for Indoor Localisation7