Hybrid WLAN-RFID Indoor Localization Solution Utilizing Textile Tag

Transcription

1 Tampere Universit of Technolog Hbrid WLAN- Indoor Localization Solution Utilizing Textile Tag Citation Hasan M., Talvitie, J., Lohan, E-S., Sdänheimo, L., & Ukkonen, L. (05). Hbrid WLAN- Indoor Localization Solution Utilizing Textile Tag. IEEE Antennas and Wireless Propagation Letters, 4, DOI: 0.09/LAWP Year 05 Version Earl version (pre-print) Link to publication TUTCRIS Portal ( Published in IEEE Antennas and Wireless Propagation Letters DOI 0.09/LAWP Copright 05 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other users, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works for resale or redistribution to servers or lists, or reuse of an coprighted components of this work in other works. Take down polic If ou believe that this document breaches copright, please contact and we will remove access to the work immediatel and investigate our claim. Download date:

2 > REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < Hbrid WLAN- Indoor Localization Solution Utilizing Textile Tag Masoumeh Hasan Student Member, IEEE, Jukka Talvitie, Lauri Sdänheimo, Member, IEEE, Elina-Simona Lohan, Senior Member, IEEE, Leena Ukkonen Member, IEEE, Abstract This paper presents a novel hbrid configuration for indoor positioning, utilizing the passive Radio Frequenc Identification () and Wireless Local Area Network (WLAN). Our architecture is based on a mobile device with a WLAN receiver, a textile tag and one or several readers communicating with the mobile. The proposed passive textile tag provides a ver cost effective, power efficient and eas-implemented solution for human positioning and tracking applications. In addition, the oint utilization of two technologies increases the accurac of the indoor positioning service. Our main contribution comes from the innovative - WLAN hbrid architecture based on Received Signal Strengths and able to improve the localization accurac compared to pure and pure WLAN location solutions. The proposed algorithm is tested with real-field measurements. Index Terms Backscattered power, fingerprinting, indoor positioning,, Received Signal Strength (RSS), Wireless Local Area Networks (WLAN). T I. INTRODUCTION HE indoor localization with low complexit and high accurac is one of the main challenges in toda s wireless world. Most promising technologies to be used indoors are currentl based on WLAN solutions [][][4][5][8][9]. Alternative solutions have been proposed based, for example, on infra-red light, ultra-wide-band (UWB) sstems, Bluetooth, and [6][7]. It is well understood that a low complexit can be achieved when using easil accessible information, such as Received Signal Strength (RSS) and man indoor localization solutions are based on RSS [4][6][7][8][9]. The main challenges remain in achieving the best possible accurac with the available signals, and this is the part we investigate in here, assuming that the available signals are the WLAN and signals. Indeed, the technolog has become increasingl popular over recent ears for tracking and positioning applications in indoor environments, as a costeffective and power efficient solution which can be installed easil on the different obect or people. The snergical use of WLAN and positioning has been ver scarcel addressed in the current literature so far. One example where this problem is tackled can be found in []. In [], the mobile device was assumed to be equipped with both a WLAN receiver and an reader and several passive tags were placed a room. The backscattered powers from the tags were onl used as an indicator of the zone where the mobile was placed, and the position estimation was purel based on WLAN RSS fingerprinting, applied in the zone identified b the tags. Pure -based localization was not studied, neither included in the comparisons provided in []. The results published in [] shows that the channel modeling based on the measurements are ver promising. Moreover, the standard deviation of shadowing even shows more stabilit of measured data compared to the WLAN measurement. However, utilizing the pure technolog or pure WLAN has its own limitations []. In this work, we present a hbrid solution to increase the robustness and accurac of indoor positioning sstems especiall in harsh propagation conditions. On one hand, using the textile sstem provides high accurac -within the limited distanceand eas-implemented solution for tracking people in indoor scenarios. On the other hand, a ver good coverage can be achieved utilizing the WLAN. The novelt of our paper is two-fold: first, we propose a novel hbridization architecture based on RSS collected from and WLAN sstems; and secondl, we compare the performance of the hbrid approach with the pure WLAN and pure localization solutions, based on measurement data performed in a universit room. II. MEASUREMENT SETUP To prepare the database for the fingerprinting, the RSS measurements were carried out in one universit room in Tampere, utilizing the sstem and WLAN sstem. The experiment has been done using wearable passive UHF tag installed on the human arm. The measurement procedures are explained below in more detail. A. Tag Fabrication The wearable tag is eas to be integrated with clothing due to its flexible conductive materials (electro-textile). Thus it is a great solution for monitoring human movement and tracking. In addition, it can provide a reliable and efficient wireless communication link between the bod-worn electronics and surrounding environment. The utilized sample for the measurement is a simple dipole UHF passive tag x 5 30 Fig.. The textile passive UHF tag

3 > REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 6m Ant. #4 0.5m 0.5m # # #4 #3 Ant. # Reader Ant. #3 6m (0,0) x Ant. # Fig. 3. The lef picture shows the schematic of the measurement region. The deploed experiments showed in the right pictures. fabricated b electro-textile. As it is shown in Fig., the dipole is a copper fabric which is applied on top of a polimide based thin substrate. The commercial NXP GiL chip is glued to the fabric using conductive epox.the tag has been designed for the frequenc MHz and the read range of the tag in free air measurement is about 8 meters. More detail about the electrical properties of the copper fabric and the tag design is available on [0]. B. Data Collection Using Tag Reader The RSS measurements were performed in an indoor area. As it is illustrated in Fig., a square region (6 m 6 m) has been divided to 44 smaller areas (0.5 m 0.5 m). The x and coordinate including 3 measurement points, respectivel, are marked in the picture. To have a good coverage in the room, four readear antennas were placed in each center edges of the square area. The antenna placement was done following an optimization process, while tring to maximize the overall coverage of all readers. The antennas are connected to an Speedwa Impin reader that is placed in the center of the square. The Impin reader, including four antenna ports, is capable of communicating with all four antennas simultaneousl. The reader is adusted to the ETSI frequenc. In addition, the redear is connected to the computer through the LAN connection for storing the scanned data. The maximum transmitted power of the Impin reader is about 30 dbm. To not exceed the EIRP (equivalent isotropicall radiated power), the transmitted power for the measurement is adusted to the 8 dbm with respect to the loss of the cables and gain of the antennas. The copper fabric tag has been installed on the human arm in vertical polarization since the reader anteenas are vertical polarized. However, for a practical implementation of our proposed hbrid solution, we would need a WLAN connection from the reader to the mobile device. In here, we prove that an - WLAN localization solution with wearable tags is full feasible, but we do not offer the full commercial architecture. The recommended architecture is shown later in Fig. 5 (b). The tag backscattered power measurement is performed b moving the person between all 69 measurement positions. This measurement has been repeated 4 times facing one specific antenna each time to be able to compare the results with similar orientations in the training and estimation phases and with different antenna orientations in these phases (the latter case being the most likel one in a commercial implementation). The same tpe of measurements in 4 random coordinates is also conducted for the estimation phase, several times. The block diagram of the pure location solution is shown in Fig. 3 (a). C. WLAN Data Collection The data collection for the WLAN was done similarl with [] [3]. In the same measurement room as above, the WLAN data have been collected utilizing a Nexus tablet with HERE maps. The tablet had software capable of reading the destination map and converting it to the Cartesian coordinates. During the data collection, the position on the map is manuall added and the RSS values from multiple Access Points (AP) at that point are read. The measurements were taken at the same coordinates as for to enable the hbridization. The pure WLAN positioning solution shown in Fig 3 (a) is based on traditional fingerprinting with Gaussian likelihood function [] [5]. III. Proposed Hbridization Algorithm Our proposed hbridization architecture is shown in Fig. 3 (b). It has two stages: a training stage shown in the upper part, and an estimation phase, where the actual hbridization is done. In the training (off-line) phase, received signal strengths from WLAN AP or backscattered powers from readers are collected and stored in ( x, F, F ) vectors, where, i, x, are the x and coordinates of the grid point where the i measurements were done, F is the RSS from the ap-th AP in x, point, ap,..., N, N is the total number of AP heard in the training phase, F is the backscattered power heard b the rf-th reader in x, point, rf N,..., N N, and N is the total number of readers (in our case, N 4 ). Our convention is that the last N RSS values belong to the signals, so that we can tackle ointl the WLAN and measurements. Also, if an AP is not heard in a certain point, the convention is that F i, has a dumb (NaN) value. Table illustrates the characteristics of the measured environment, in terms of the mean and maximum ranges at which a transmitter is heard, the minimum, maximum and average number of heard transmitters in each measured point the room, and the estimated shadowing standard deviation, based on the training data. The shadowing standard

4 > REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 3 deviation is computed via path-loss modeling with deconvolution estimation, based on []. In the estimation (on-line) phase, the following values are measured: ( R, R ), ap,..., N, rf N,..., N N where R is the RSS measured from the ap-th AP, and the R is the backscattered power measured at the rf-th reader. Again, a transmitter which is not heard is signaled b a NaN value. The mobile position is computed based on the following Gaussian probabilities: R F p ( R ) exp, ap,..., N () R F p ( R ) exp, rf N,..., N N. () () Above, () () p ( R ) is the probabilit to be in x, point if R the RSS value was heard from the ap-th AP, is the WLAN shadowing variance, obtained from the training data, () p ( R ) is the probabilit to be in x, point if the value R was heard at the rf-th reader, and is the shadowing variance, obtained from the training data. A decision is taken based on the oint cost function, via: ^ ^ ( x) max, with: i i, x i WLAN or Shadowing Variance Deconvolution Estimate (LS+ Weighted Centroid [][3]) (a) Estimation Data Fingerprinting via Gaussian Likelihood Deconvolution Estimate (x, ) position estimate WLAN Decon. + Weighted Centroid [][3] Fingerprinting via Gaussian Likelihood with (x i) No Estim. data point i Max ( RSS ) > -3 db Yes Fingerprinting via Gaussian Likelihood with position estimate (x i) (b) Fig. 3.(a) Pure /Pure WLAN location. (b) Proposed Hbrid - WLAN positioning algorithm. N N N () () ( ) ( ) (3) w p R w p R ap rf N and the weights defined adaptivel as: N N N max R a max R w (4) TABLE. ENVIRONMENT STATISTICS, BASED ON MEASURED DATA. Max rang e [m] Mean range [m] Mean Nr of ap rf N 0 otherwise Min Nr. of heard per point the room Max Nr. of heard per point the room heard per point the room () (4.3% cases) WLAN () Shadowing standard deviation [db] () The statistics are room-based and using the true Reader possitions (known) () The WLAN statistics are building-based and using the estimated AP positions via the weighted centroid approach [3]. a is an optimization parameter, which selects the threshold between using or WLAN-based estimates. The optimization of a is illustrated in Fig. 4, based on the distance Root Mean Square Error (RMSE) as the optimization criterion. The optimization of a was done with a different set of estimation data than that one used in Fig. 5 and Table. It was found that a db gives the best results, meaning that we trust the positioning as long as the signal strength is higher than the WLAN signal strength with a db margin. The results of the hbridization algorithm (Fig.3), compared with pure WLAN and pure positioning are shown in Fig. 5 and in terms of the cumulative distribution function of the distance error and of the percentage of estimated track points with a distance error below m, respectivel. In Fig.5 we show what happens when onl N out of the four readers are used, N=, 4. Clearl, more readers are available, the better the accurac of the location solution. The pure solution is unable to offer a position estimate in 40 to 80% of the cases (depending whether we use 4 or less readers), and this is because the reader range is limited, and there are man points the room where no backscatter power reaches the readers. On the other hand, a pure WLAN positioning solution can alwas offer a position estimate. Pure solutions are indeed slightl better at ver low distance errors (lower than m), but this kind of small errors are achieved with a low probabilit. Table illustrates the probabilities (given as percentages) of achiving a distance error less than m with an of the considered algorithms.hrid -WLAN solutions are clearl the best, even with onl one reader. The results clearl show the benefits of a hbrid -WLAN architecture, benefits which can be incorporated in a commercial architecture as illustrated in Fig 5 (b).

5 > REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 4 Fig. 4. Optimization of the hbridization parameter a. IV. CONCLUSION In this paper, we have introduced a novel hbrid WLAN- indoor localization solution. The hbrid solution is suitable for the indoor positioning in environments where both and WLAN signals are available, since it takes advantage of both tpes signals. The textile passive tag proposes ver cost effective, eas implemented and efficient configuration for human tracking in harsh environments and it can be deploed for other purpose such as access control or patient tracking in health care applications. The presented configuration based on the hbrid model can be expanded for larger indoor area utilizing more tag and readers, as illustrated in our design recommendation in Fig. 5 (b). Our experimental results show that our hbrid solution can enhance the probabilit to get errors below meters with several percents even if onl one reader is available, and that RSS coming from two different sstems can be successfull combined towards a oint hbridization solution. REFERENCES [] S. C. Spinella, A. Iera, and A. Molinaro (00, Mar.) On potentials and limitations of a hbrid WLAN- indoor positioning technique, International Journal of Navigation and Observation, Hindaw [] S. Shrestha, J. Talvitie, E. S. Lohan, "Deconvolution-based indoor localization with WLAN signals and unknown access point locations" in TABLE. PERCENTAGE OF ESTIMATED POINTS WITH ACCURACY BELOW M % of points with accurac below m Same orientation of the mobile device in training and estimation Different orientation of the mobile device in training and estimation PureWLAN 48.5% 8.57% Pure with tx 3.49% 9.5% Pure with tx 5.00% 7.85% Pure with 3 tx 33.53% 3.80% Pure with 4 tx 4.6% 8.57% tx tx 3 tx 4 tx 48.3% 39.45% 5.44% 45.06% 57.4% 49.48% 6.50% 53.06% (a) Mobile device with Hbrid location engine reader WLAN AP tag (b) Fig. 5. (a) Percentage of points where we attain an error smaller than a threshold, via pure WLAN, pure and hbrid estimates. (b) Proposed hbrid -WLAN architecture with passive tags. Proc. IEEE International Conference on Localization and GNSS (ICL- GNSS), June 03, Torino, Ital. [3] E. S. Lohan, K. Kosk J. Talvitie, and L. Ukkonen WLAN and Propagation Channels for Hbrid Indoor Positioning in Proc. IEEE International Conference on Localization and GNSS (ICL-GNSS), June 04, Helsink Finland. [4] V. Honkavira, T. Perala, S. Ali-Lott, and R. Piche, A comparative surve of WLAN fingerprint methods, in Proc. 6th Workshop Position., Navigat., Commun. 009 (WPNC 09), pp [5] N. Le Dortz, F. Gain, and P. Zetterberg, Wifi fingerprint indoor positioning sstem using probabilit distribution comparison, in Proc. IEEE Int. Conf. Acoust., Speech, Signal Process. (ICASSP), Mar. 0, pp [6] G. Yaning, A. Lo, I. Niemegeers, "A surve of indoor positioning sstems for wireless personal networks," Communications Surves & Tutorials, IEEE, vol., no., pp.3,3, First Quarter 009, doi: 0.09/SURV [7] L. Hongbo, Y. Jie, S. Sidhom, W. Yan, C. Yinging, Y. Fan, "Accurate WiFi Based Localization for Smartphones Using Peer Assistance," Mobile Computing, IEEE Transactions on, vol.3, no.0, pp.99,4, Oct. 04 [8] B. Wang, S. Zhou, W. Liu,Y. Mo, "Indoor Localization based on Curve Fitting and Location Search using Received Signal Strength," Industrial Electronics, IEEE Transactions on, vol.pp, no.99, pp.,, doi: 0.09/TIE , doi: 0.09/TMC [9] S. Guol C. Jie, G. We K. Liu, "Signal processing techniques in network-aided positioning: a surve of state-of-the-art positioning designs," Signal Processing Magazine, IEEE, vol., no.4, pp.,3, Jul 005, doi: 0.09/MSP [0] K. Kosk E. Morad A. Vena, T. Borninen, L. Sdanheimo, L. Ukkonen, and Y. Rahmat-Samii "Characterization of Electro-textiles Using Wireless Reflectometr for Optimization of Wearable UHF Tags", PIERS Proceedings, Aug. 03. [] M. Hasan E. S. Lohan, L.Sdanheimo, L.Ukkonen, "Path-Loss Model of Embroidered Passive Tag on Human Bod for Indoor Positioning Applications", IEEE -TA, Sep. 04.