International Journal of Software Engineering and Its Applications

Vol.8, No.1 (2014), pp.235-246
http://dx.doi.org/10.14257/ijseia.2014.8.1.21

Indoor Wireless Localization Using Kalman Filtering in
Fingerprinting-based Location Estimation System
Geon-Yeong Park, Min-Ho Jeon and Chang-Heon Oh
School of Electrical, Electronics, and Communication Engineering
Korea University of Technology and Education
gypark@koreatech.ac.kr, w1004me@koreatech.ac.kr and choh@koreatech.ac.kr
Abstract
As “smart” devices such as smart phone and smart TVs become widely distributed, various
studies on location-based services have been conducted. Such location-based services are
useless, however, unless the user’s location is known. A number of researchers have
examined methods to trace and determine indoor locations for indoor location-based
services. In particular, WALN has been examined in various studies because of its advantage
to use a frequency band available without advanced settings. This study suggests a new
indoor tracing method to reduce time delays upon location fingerprinting for point data
collection, which is a disadvantage of the existing Kalman filtering algorithm and
fingerprinting type location tracing algorithm. This study also compares its performance with
that of existing methods based on the collected data. As a result of the experiment, the fast
collection algorithm is presented as a solution to the problems of existing methods. It is
proven that the fast collection algorithm presented in this study is applicable to a location
tracing system in an actual environment.
Keywords: fingerprinting, location estimation, WLAN, fast collection, NLOS

1. Introduction
In recognition of the usefulness of Location-based Service (LSB), large-sized companies,
plants, and colleges are demanding services for its utilization [1]. Since Global Positioning
System (GPS) itself cannot provide sufficient data to determine the user’s location when he is
in a building, indoor location tracking to determine the user’s location in a building,
particularly when utilizing wireless local area network (WLAN), has been actively studied
[2]. This is because WLAN is being used in every place from college campuses to airports,
hotels, large-sized companies, and even private homes. This study introduces a WLAN-based
indoor location estimation system. This determines the user’s location in reference to received
signal strength indication (RSSI) collected from access points (AP) [3-7]. Fingerprinting is
the most common method used to determine the user’s location because its method for
location estimation is simple compared to angle of arrival (AOA), time difference of arrival
(TDAO), and time of arrival (TOA). Fingerprinting is used to link location-related
characteristics such as the RSSI received from a number of APs to the location rather than
relying on the accurate estimation of the angle or distance and to infer the location of such
characteristics. In this case, the wireless network interface card (NIC) and existing WLAN
infrastructure could be readily reused with no need for special hardware at a mobile station
(MS).
Fingerprinting consists of the offline step and real-time step. The offline step collects the
data used in determining location by using the fingerprinting method. In this step, the location

ISSN: 1738-9984 IJSEIA
Copyright ⓒ 2014 SERSC

On the other hand. As to RSSI characteristics presented by Kamol Kaemarungsi. this problem needs to be solved first of all. and dispersion. Hence. the collection of location fingerprint data would be quite efficient. and algorithm suggested in this study. The information collected in the real time step and the route of the user’s movement are used to update location fingerprinting data in consideration of the surroundings.International Journal of Software Engineering and Its Applications Vol. Based on the points above. distribution of received signal strength. it will take more time depending on the size of the building. a location fingerprint database needs to be established in consideration of the route of the user ’s movement and number of surrounding APs to form the fingerprinting type location estimation system. Multipath fading changes the received signals around the medium values at a certain location. standard deviation of the RSS. Chapter 5 presents the conclusion. The Hamming signal distance between the and for each in the database is computed. Chapter 3 explains the fast collection algorithm in the real-time step. The real time step of fingerprinting estimates the location based on the location information of the database collected in the offline step. Hence. One of the disadvantages of the offline step in fingerprinting is that it takes long to collect data and the values change over time. 236 Copyright ⓒ 2014 SERSC . the time dependency of received signal strength. received signals are presented and they are influenced by the combined effects of large-scale fading and small-scale fading [9]. a location fingerprint consists of vectors of various APs’ RSSI values at a certain location of ( ). Location fingerprint data collected in a static status may be different at each time of measurement due to the dense multiple channels of indoor radio waves and radio effects such as reflection. A mobile node (MN) acquires location information . This st udy suggests the fast collection (FC) algorithm to reduce the time of collecting location fingerprints in the offline step in the indoor location estimation system and to complement location fingerprints that might be changed over time. it involves less cost for the additional system compared to other indoor location estimation systems. When it comes to the RSSI characteristics that he studied. there must be a system to address this problem. the interference from multiple APs. special attention must be paid to their effect to the user’s body.8. effect of user’s orientation. Chapter 2 points out the problems of the existing methods and explains the fast collection algorithm in the offline step to solve such problems. In this study. stationary condition of the RSS. No. the effect of the user’s orientation. experiment environment. Setting up the database involves the mapping of each location point and location fingerprint data (RSSI). Since the collected data in the offline step and period for the establishment of the fingerprinting offline step may be changed over time. Algorithm of Fast Collection for Offline Step Data The fingerprinting-based indoor location estimation system estimates indoor location in reference to the RSSI of WLAN. Kamol Kaemarungsi has studied RSSI characteristics such as the effect of a user’s body. diffraction. this study suggests a method to collect offline data effectively in a fingerprinting-based indoor location estimation system. When a location fingerprint database must be established in the offline step based on the points above. Chapter 4 evaluates the RSSI collection and transfers software for the location fingerprint database.1 (2014) fingerprint or radio map database is established [8]. 2. and the independence of multiple RSSs. and so forth [10]. The data extracted in real time to solve the problem above is used as the location fingerprint data. As such.

which forms in Table 2 of the movement coordinates. there is no need for APcentric distribution.1 (2014) When the information of a moving object is collected and the database is established after all possible routes of the user’s movement are calculated. the time of establishing the location fingerprint database will be shortened. using the location information collected every hour will further shorten the offline step. (1) When the value of calculated in Expression 1 is . Afterwards. Figure 1. As shown in this figure. with the errors in the real-time step reduced. No. with a number for . To reduce the difference in the fingerprint method. however. For such location vector division. This theory will be proven in the following section before the fast collection algorithm is presented. k. Location vector division provides vector R.International Journal of Software Engineering and Its Applications Vol.8. A Comparison of AP RSSI Depending on the Location Figure 1 shows the RSSI values of APs collected during the movement. the object moves at similar rates in the same place. In addition. are significantly different. as to collected initially in Expression 1 and Vector R of collected thereafter. all the variables of Vector R included in are subtracted by vector R of to calculate the absolute value. The RSSI values of APs collected in a static status. thus. Copyright ⓒ 2014 SERSC 237 . and the measured RSSI values are in a similar scale. a new Vector is generated as many as . it is important to collect as many vector Rs as possible.

This indicates that RSSI values may be changed depending on the direction even right under AP. which indicates that when . Two interpretations are possible: first. Once the direction is designated. Table 1.International Journal of Software Engineering and Its Applications Vol. the fast collection algorithm suggested in this study calculates the vector in the way of location vector division.2 -37 -37. the received signals may go through interference.6 -49 -49.. .1 (2014) which is the value of Vector .4 -49. and determines its location in reference to Hamming distance in the section of . when . reference point (1m) (2m) (3m) (4m) (5m) (6m) (7m) (8m) (9m) (10m) AP 1 -47 -47. the value added to the variable of Vector – k – is low.2 -41 AP1 and AP3 are of almost a straight line.8 -48. the use of AP-centric distribution increases the value of . Location Vector Division Data of Vector (location) (0m) .4 -36. The direction is designated to a certain value.8 -50. is added to the generated Table 1 shows the data after the location vector division of the Table 2 in application of Expressions 1 and 2.8.4 -48.2 -47 -45. AP may exist in the range of .4 -40. No. In the second case.2 -50. To solve this problem.4 -42. and then the value of is generated through location vector division.8 -38. In this case.2 -48. and this value gets bigger as AP goes out of the center. As for the AP-centric distribution of the fast collection algorithm.6 -43. Hence. AP-centric distribution collects the data from four different directions with AP as the center as shown in Figure 3.8 -50. the difference in location vector division is complemented by the data above when . and the added variable increases accordingly. the value of AP centric distribution is inserted between and .6 -35.4 -47.8 -44. 238 Copyright ⓒ 2014 SERSC . when the AP from which the value of RSSI is collected is far. Table 2 shows the RSSI data right under AP. The added variable is .6 -49.6 -39. inserts it.6 -51 AP2 -45 -44 -43 -42 -41 -40 -39 -38 -37 -36 -35 - - (2) section in Section AP3 -33 -33.8 -34.2 -41 AP4 -53 -51.

the way to complement it in the real-time step also will be discussed.8. the location fingerprint. which is used in location estimation in the real-time step. Algorithm of Fast Collection for Real-time Step Data As stated earlier. (3) Copyright ⓒ 2014 SERSC 239 . No. Hence. RSSI Value of the Movement Route Measured Right Under AP Direction Front Right Back RSSI -35 -29 -41 -23 Figure 3. the environmental change over time also needs to be considered.1 (2014) Figure 2.International Journal of Software Engineering and Its Applications Vol. Location Vector Division Data of - Section Table 2. A Comparison of the RSSI of each AP on Li Depending on the Measuring Methods 3. To complement the disadvantage of data change over time. is revised by using Expression 3.

The computer application is developed by using C# in application of a native Wi-Fi API provided by Microsoft Developers Network (MSDN). No. and is the variable of Vector R to which will be compared. Otherwise. the information collected earlier remains as it is. One outstanding feature is the beacon message must be transmitted after the program is initialized when AP information is collected and then recollected. RSSI Information of an AP 4. The distribution traces form a lineal curve over time. Since information on APs may be different depending on the types of NIC. Performance Analysis 4. The RSSI value for a certain AP among those that were collected for 24 hours at one place may change from a minimum of -29dBm to a maximum of -33dBm in an adjacent area. WLAN-based indoor location estimation methods cannot be realized unless RSSI from APs is read.8. MAC address.International Journal of Software Engineering and Its Applications Vol. it is converged to one variable at one time zone as in Figure 1.1. link quality. information on SSID. this research has developed an RSSI collection and transfer that can be used to measure RSSI. the difference of the fingerprinting type estimation system can be reduced. and beacon period at a certain place is collected and transmitted to the server. The lineal distribution may be described with the example in Figure 4. Hence. and a number of studies have been conducted on this area. RSSI. it is vital to collect various types of NIC and information among common APs so that it can be applied to various devices. Development of the RSSI Collection and Transfer Software WLAN-based indoor location estimation methods are economical compared to other indoor location estimation methods. When the changing data collected in the offline step is adjusted according to the timing.1 (2014) in Expression 1 is the variable of Vector R. however. When Expression 1 is substituted for the changing amount. Although the change is irregular from 09:00 to 08:00 on the next day. Figure 4. As shown in Figure 5. It may change at every second irregularly. 240 Copyright ⓒ 2014 SERSC . frequency. the variable of the collected Vector R changes in a lineal form from -31dbm (09:00) to -33dBm (10:00) when it is observed each hour.

No. when signals from an adjacent corridor enter a room in the building. Hence. transmit. Based on the measuring method suggested by Kamol Copyright ⓒ 2014 SERSC 241 .International Journal of Software Engineering and Its Applications Vol.1 (2014) Figure 5. On each floor. and the data collected in the offline step include about 38-50 AP signals at one place. the RSSI signals received become very weak. there are 16 to 28 personal APs. This is because common APs are arranged at the same intervals in a row as shown in Figure 6.8. RSSI Reader 4. It is a four-story building in which the 1st to 3rd floors have 18 common APs for each and the 4th floor has 15. Figure 6. it takes about 10 seconds per step to collect.2. The 4th Floor of the 1st Engineering Building with the Locations of Published APs When all of the AP information is collected to establish a location fingerprint database. there are duplicated vector data even if only three data sets are used. and s ave AP information. information on APs with the highest strength of RSSI and common APs installed at the Korea University of Technology and Education is used. thus. Measurement Setup A laptop computer equipped with a WLAN card of Ralink technology as well as the RSSI collection and transfer software developed to collect offline data and estimate location was used to collect the RSSI information of AP within the 1st engineering building in the Korea University of Technology and Education. Further.

distance calculation.International Journal of Software Engineering and Its Applications Vol. 4. the user’s location information is received and the actual location is compared with the estimated location of the estimation program in application of the fast collection algorithm. As the range of measurement becomes larger. the signals received by the WLAN card are limited from -93dBm to 0dBm.8.4. 242 Copyright ⓒ 2014 SERSC . Figure 7. For accurate verification. To evaluate the fast collection algorithm suggested in this study. after the table to predict the route of movement in each direction and the fingerprint location database in the offline are prepared. The experimental result indicates that the suggested fingerprinting method involves increase in data collecting time at consistent rates because of continuous movement while existing fingerprinting methods involve an increase at inconsistent rates due to such variables as collected data. the information is collected at intervals of 1m while in the case of the suggested fingerprinting method the information is collected at intervals of 10 seconds. the measuring time increases accordingly.3. the difference increases up to 10 times. the time of collecting data in the existing fingerprinting type offline step and the time of the algorithm suggested in this study are compared. No. the object moves 1m per second or 10m per 10 seconds. Experimental Results Figure 8 shows the time of collecting AP information when four APs are selected at one place and the fast collection algorithm of fingerprinting is applied in comparison with existing fingerprinting methods. Routes within the Building 4. In the case of existing fingerprinting methods. etc.1 (2014) Kaemarungsi. As the time of measuring increases. transmission initialization. Experimental Methods As shown in Figure 7.

1 (2014) Figure 8. the range of errors is relatively similar. No. however. Since AP-centric distribution is not considered. In comparison with Figure 9. Figure 9. A Comparison of the Data Collecting Time in the Offline Step Figure 9 shows the range of signal errors at the moment of collecting RSSI from a moving object in application of the fingerprinting method without AP-centric distribution as a fast collection algorithm and an existing fingerprinting method. the range of signal errors of the fast collection algorithm in consideration of mobility is low.International Journal of Software Engineering and Its Applications Vol. RSSI Errors when the AP-centric Distribution is Applied Copyright ⓒ 2014 SERSC 243 .8. S ince the existing fingerprinting method does not consider variables of mobility. the fingerprinting method with the AP-centric distribution applied involves a narrower range of errors than the existing fingerprinting method. Figure 10 shows the range of signal errors at the moment of collecting the RSSI of a moving object in application of fingerprinting with the AP-centric distribution applied and an existing fingerprinting method.

and the data of indoor environments at various angles were collected and comparatively analyzed to verify the excellence of the fast collection algorithm. Figure 11. No. It turns out that the performance of the suggested method is better than that of existing fingerprinting methods in terms of tracing a moving object. RSSI Errors when the AP-centric Distribution is Applied Figure 11 shows the range of errors in comparison of or RSSI at the measuring location and Vector R in the location fingerprint database in application of the suggested fingerprinting method. The experimental result indicated that while the existing indoor location estimation systems in application of the fingerprinting 244 Copyright ⓒ 2014 SERSC . and then the results are illustrated.1 (2014) Figure 10. The Difference from the Actual Distance 5.International Journal of Software Engineering and Its Applications Vol. Conclusion To highlight the necessity of the fast collection algorithm suggested in this study. the values of Vector R collected by means of an existing fingerprinting method are compared to determine the range of errors.8. the problems of existing fingerprinting methods were addressed. To verify the excellence of the suggested algorithm.

R. Krishnan and A. Youssef. Electronics and Communication Engineering at the Korea University of Technology and Education. 10. Deaheon. (2012). 91. Miami. J. Einahrawy. pp. His research interests are in the areas of wireless communications. pp.-H. Hopper. 269. vol. Kaemarungsi and P. [4] C.S. [7] C. Korea.International Journal of Software Engineering and Its Applications Vol. International Journal of ACM Transactions on information systems. Virrantaus. 349. This was because the performance of the existing fingerprinting method in collecting data and tracing location was inferior or the experiment was conducted in a fixed condition. no. Kim. [5] J. vol. 35. Yim. Krishnakumar. [2] R. (2012). USA. Markkula. [8] M. References K. The study should be developed further in application of this algorithm and in reference to valid data regarding mobility at various locations. vol. [1] Authors Geon-Yeong Park. 7. (2001) December 3-6. [10] K. [6] H. A. “Bayesian indoor positioning system”. Oh. “Properties of Indoor Received Signal Strength for WLAN Location Fingerprinting”. In addition. W. wireless sensor network and real-time operating system. vol. Proceedings of the Second International Conference on Web Information System Engineering. USA. Korea. LA. Journal of information and communication convergence engineering. V. (1992). they did not prove to be useful for tracing a moving object. Wann and M. Want. A. Madigan. Currently. degree in Electrical. Agrawala and A. Terziyan. 10. Boston. U. Krishnamurthy. No. P.8. Hence. Sklar. Texas. A. Lin. Proceedings of 24th Annual Joint Conference of the IEEE Computer and Communications Societies. no.1 (2014) method could be used to find an object in a fixed location. USA. (2003) March 26-26. “WLAN Location Determination via Clustering and Probability Distributions”. Kyoto. “Developing GIS-Supported location based services”. Falcao and J. the excellence of the suggested algorithm was evaluated based on the valid data of a lineal distance. Cheonan. Martin. International Journal of IEEE Communications Magazine. Seo and H. P. pp. Proceeding of the First Annual International Conference on Mobile and Ubiquitous Systems: Networking and Services. E. 34. USA. vol. [3] D. Garmash and Y. (1997). Gibbons. 1296. he received a B. 90. A. pp. the fingerprinting method in application of the fast collection algorithm as suggested in this study to estimate the location of a moving object by means of actual wireless moving nodes prove d to be more efficient and accurate for moving objects. [9] B. no. Japan. 2. (2004) March 21-25. Ju. pp. 3.S degree in the Department of electrical engineering from Hnabat Nation university. 1. (2008). (2004) August 22-26. Proceeding of the First IEEE International Conference on Pervasice Computing and Communications. in 2008. “Data fusion methods for accuracy improvement in wireless location system”. hankar. He is currently pursuing a M. 10. 4. it has been proven that the data collecting time could be minimized in the offline step where the fingerprinting method was applied to trace the location. no. S. Proceedings of Wireless Communications and Networking Conference. V. Journal of information and communication convergence engineering. (2005) March 13-17. International Journal of Expert Systems with Applications. no. Copyright ⓒ 2014 SERSC 245 .

he was with the CDMA R&D center of Samsung Electronics Co.8. wireless sensor network. From Oct. Umsung. in 1996. channel coding and M2M network. degrees in telecommunication and information engineering from Korea Aerospace Univ. he has been with the School of Electrical. in 1988 and 1990. and an M. Korea. Since Mar. he was with Hanjin Electronics Co. Chang-Heon Oh. wireless localization. Korea University of Technology and Education.S. and wireless sensor networks with particular emphasis on wireless localization. 1999. Cheonan. and M. 1990 to Aug.S.E.International Journal of Software Engineering and Its Applications Vol. Electronics and Communication Engineering at the Korea University of Technology and Education.S degree in the Department of Game Digital Content from Far East University. in 2009. degree in Electrical. His research interests are in the areas of wireless communications. mobile communication. degree in Electrical. he received the B. From Feb. Korea. Electronics and Communication Engineering at the Korea University of Technology and Education. S.. degree in avionics engineering from Korea Aerospace Univ. Korea. 1999. He received the Ph. in 2011.1 (2014) Min-Ho Jeon. respectively.D. he received the B. where he is currently a professor. 1993 to Feb. No. 1993. Electronics and Communication Engineering. His research interests are in the context -aware. Cheonan. He is currently pursuing a Ph.D. 246 Copyright ⓒ 2014 SERSC .