Improved AoA Based Localization of UHF RFID ... - Semantic Scholar
2011 IEEE International Conference on RFID-Technologies and Applications
Improved AoA Based Localization of UHF RFID Tags Using Spatial Diversity Salah Azzouzi, Markus Cremer, Uwe Dettmar, Thomas Knie and Rainer Kronberger Department of Telecommunication Cologne University of Applied Sciences, Germany Email: [email protected]
Abstract—In a recent paper we presented new measurement results for an angle of arrival based approach to localize UHF RFID transponders in an indoor environment with off-the-shelf IDS R901G RFID reader ICs and UPM Raflatac DogBone RFID transponders. We were able to localize the transponders with a mean accuracy of 0.21 m using three antenna array positions for triangulation in a 3 m x 3 m room. In this paper we present an RSS based algorithm to select a special subset of antenna array positions for the estimation of the transponder position in a measurement setup with five different antenna array positions. With this algorithm a reduction of the mean measurement error by 50% was achieved for the considered test grid compared to the original approach in [1].
I. I NTRODUCTION RFID technology is on the way to substitute bar codes for the identification of objects in many fields of application. With the introduction of low-cost passive UHF RFID transponders in the UHF frequency band (868 or 915 MHz) the operational range of RFID transponders was enlarged to distances of up to 10 meters. Typical applications include baggage handling, supply chain applications, or fixed asset tracking. However, the full potential of these transponders only will be exploited if the identification of objects is complemented by accurate localization. Therefore during the last years large efforts were taken to solve the problem of radio based indoor localization. For an overview see e.g. [2]–[6]. The RFID based techniques use measurements between RFID transponders and reader devices and some form of side information about the location of specific readers and/or reference transponders or statistical knowledge about the distribution of RSS values. Measurements are for the Received Signal Strength (RSS), the Angle of Arrival (AoA), the Time of Arrival (ToA), or the Time Difference of Arrival (TDoA) as well as hybrid procedures (e.g. see [7]–[11]). For the radio based RFID transponder localization in an indoor environment the time based approaches ToA and TDoA are not appropriate. Both of them require accurate time measurements. However, such measurements are quite difficult to implement with RFID systems since the defined data rates are far too low to receive a sufficient resolution. The RSS based trilateration methods show to be strongly related to the quality of the propagation channel which again is very sensitive to fading/multipath effects.
978-1-4577-0027-9/11/$26.00 ©2011 IEEE
In this paper we consider an AoA based localization approach that uses UHF RFID transponders according to the ISO standard 18000-6C (see [12], [13]) and multiple directional antenna arrays. Since the channel for RFID communication is typically small band (only a few MHz are available in the UHF band in Europe) an AoA approach is critical under severe multipath conditions, too. By applying the phase difference between incoming signals to estimate the position of the transponder the AoA approach totally depends on receiving a strong line-of-sight (LOS) path which can be isolated from other signal paths. However, even under good LOS conditions the superposition of multipath signals to the LOS signal will deteriorate the quality of the angle measurements. Therefore, for the 2D localization of an RFID tagged object the exploitation of more than the minimum two antenna positions in combination with triangulation may be beneficial. On the one hand a spatial diversity is established for the localization. On the other hand if the distance between transponder and reader antenna is small the effect of the estimation error of the incidence angle is reduced (see [14]). In environments with relaxed multipath and without shadowing the reader antennas closest to the transponder can be determined by RSSI measurements. To follow this approach low price reader modules using off-the-shelf components and self designed antenna arrays had to be implemented. The theoretical analysis of such systems in typically time variant multipath environments with strong deterministic components [15] is rather complicated and often relates to simplified signal models (see e.g. [14], [16]). This again results in significant differences between simulation and measurement results [6]. Due to the difficulty to access the IQ signals from a commercial RFID reader needed to evaluate the phase differences for AoA based approaches up to now most authors restrict to system simulations or the use of non standard compliant hardware (see [17]–[19]). The focus of this work is the evaluation of the achievable accuracy for a transponder localization under real life conditions. A simple RFID based localization approach using offthe-shelf components. A setup using IDS R901G RFID reader ICs, UPM Raflatec DogBone RFID transponders and multiple self-designed three element antenna arrays will be considered. The test environment is a seminar room in which an RFID transponder is placed at 25 different grid positions in a square plane of size 3 times 3 meters. The position of the transponder
174
TABLE I P REVIOUS MEASUREMENT RESULTS
mX = E{∆X} mY = E{∆Y } mD = E{∆D} 2 = E{(∆X − m )2 } σX X 2 = E{(∆Y − m )2 } σY Y 2 = E{(∆D − m )2 } σD D
Approach I
Approach II
0.18 m 0.17 m 0.26 m 0.05 m2 0.02 m2 0.06 m2
0.14 m 0.13 m 0.21 m 0.02 m2 0.01 m2 0.02 m2
is then determined for each position by estimating the angle of arrival measured by five antenna arrays placed outside of the plane. Application scenarios for AoA based localization of RFID transponders in relaxed multipath environments include the problem of finding RFID tagged books on a desk for sorting purposes in a library or the localization of tools or work pieces on a table during the fabrication process. The rest of the paper is organized as follows. Section II introduces our first measurement results using the same measurement setup as in this paper. Section III describes the AoA based approach to estimate the position of an object under investigation. The experimental setup and measurement results are presented in sections IV and V. Section VI concludes the paper with a summary and discussion. II. R ELATED W ORK In a recent paper [1] we described the localization of UHF RFID transponders using an AoA approach with three uniform linear antenna arrays with three patch elements each and present measurement results for a seminar room where a transponder was moved to 25 positions on a 0.5 m x 0.5 m testgrid (see Fig.1). The angle of arrival of the received tag signal was measured at each of the three antenna array positions and for each of the 25 transponder positions. Two strategies were followed to determine the transponder position in the measurement setup. In the first approach the centroid of the triangle defined by the three intercept points of the lines originating from each antenna array was used as an estimate for the actual transponder position. Fig.1 depicts the result of this approach for measurement point 25. In the second approach a subset of two out of three array positions was chosen to calculate the position of the transponder for each test point. The intercept point of the lines initiating from these two antennas was used as the estimate for the transponder position. The subset was determined as ˆ 3 of antenna array 3 as a function of the estimated angle Θ follows: ˆ 3 ≥ 0 choose the intercept point of arrays 1 and 3 else If Θ choose the intercept point of arrays 2 and 3 As a reference table I shows the measurement results of the tag position for both strategies (Approach I and Approach II). Estimating the transponder position using Approach II leads to the smallest mean positioning error of 0.21 m.
Fig. 1. Grid for localization measurements with three antenna array positions
III. L OCALIZATION M ETHOD The proposed angle of arrival based localization approach consists of two steps. In the first step the incidence angles Θn of the received transponder signal with respect to known positions of N receive antenna arrays are estimated. In the second step the position of the transponder is determined based on the knowledge of the angles Θn by using a triangulation method. To estimate the angle of arrival of the transponder signal a phased antenna array composed of M individual antenna elements is used. The single elements of the array are aligned with an equal distance d to each other along a horizontal line to form a uniform linear array (ULA, see Fig.2). The transponder signal will arrive at each of the array elements with a certain time delay resulting in a constant phase shift ∆φ with respect to a reference phase. ∆φ is related to the incidence angle Θ of a plane wave measured with respect to the normal to the linear array as 2πd ∆φ = sin Θ. λ λ is the carrier wave length [20]. To estimate the incidence angle Θ the received signals from the M = 3 antenna elements are evaluated. Applying the root-MUSIC method (see e.g. [14], [18]) to the case of one signal source and a three element ULA Θ is calculated from the measured phase shift ∆φˆ as: λ ˆ = arcsin ∆φˆ (1) Θ 2πd where ∆φˆ = 6 {E[x∗i (t)xi+1 (t)]}.
(2)
with i = 1, 2. ∗ denotes the complex conjugate, E[.] the statistic expectation value, and 6 {.} is the phase of a complex variable. xi (t) and xi+1 (t) describe the two IQ demodulated and filtered complex baseband signals from antenna i and i+1, respectively. To reduce the observation time the correlation coefficients from both pairs of antennas can be added before
175
IV. E XPERIMENTAL S ETUP A. RFID Localization System
Fig. 2.
One unit of the localization system comprises the signal processing hardware and a self designed three element phased array. Three RFID reader modules combined on one base circuit board were implemented to communicate with the transponder and process the received signals. Five of these hardware units are deployed at the four corners and in the center of the test grid to form the localization system (see Fig.6). The IQ components of the received signal from each antenna array element are sampled and read out for further processing. The estimates for Θn are calculated according to equations 1 and 2 and used to compute and plot the estimated position of the transponder. This is done on a host computer connected to the hardware units.
Phased array with ideal antennas
taking the argument. In general d < λ/2 is chosen to avoid the appearance of grating lobes. Having estimated the angles of incidence (Θ1 , ...ΘN ) for all N antenna arrays in the second step the probability of the transponder being at position (x, y) under observation of (Θ1 , ...ΘN ) P r((x, y)|(Θ1 , ...ΘN )) has to be maximized. Assuming perfect knowledge of the incidence angles (Θ1 , ...ΘN ) this can be done by constructing lines from the N antenna positions towards the direction of the corresponding Θn . One interception point results that represents the position of the transponder. In this case even a setup with only two arrays is sufficient to exactly find the position of the object under investigation. However, in reality imperfections in measuring the exact angle and perturbations like reflections or multipath lead to substantial deviations between estimated and correct position. In this paper we present two strategies to find appropriate subsets of the measured angels of incidence from N = 5 antenna positions as a simple approach for tag localization with low computational load. Both strategies make use of the assumption that the antenna array with the smallest distance to the transponder position will produce the best angle of arrival estimate. Because of the fact that the transponder position is not an available a-priori information the received signal strength (RSS) is computed during the AoA estimation process. The two algorithms for selecting appropriate antenna subsets can be formulated as follows: Approach III: Determine the subset of the two antennas with the highest RSS values and calculate the intercept point of the lines originating from these two antennas as an estimate for the transponder position. Approach IV: Determine the subset of the three antennas with the highest RSS values and compute the centroid of the triangle defined by the intercept points of the lines originating from these three antenna arrays as an estimate for the actual transponder position.
For the calculation of the AoA the transponder response containing the Electronic Product Code (EPC) and the CRC of length 16 is evaluated. From this signal 240 samples (relating to around 70 symbols) are used for computing the correlation as given by equation 2. From the estimate ∆φˆ the spatial angle ˆ is calculated. As described in [18] the phase characteristic Θ of the antenna pattern distorts the determination of the spatial ˆ To eliminate this effect a reference measurement of angle Θ. ˆ was conducted in an anechoic the characteristic Θ = f (Θ) antenna chamber (see our recent paper [1]). This characteristic ˆ was used in the sequel to correct Θ. As a measure of the received signal strength (RSS) of the incoming signal from the transponder the sum of the mean signal power of the received baseband IQ signals of the three array elements was used. First, the mean signal power Em,n for the M = 3 single elements of array n was calculated using K = 240 I and Q samples: Em,n =
K q 1 X Ik2 + Q2k K k=1
The mean received signal power of array n was then estimated as M 1 X En = Em,n . M m=1 B. Antenna The receive antenna of each hardware unit is designed as a microstrip patch antenna array with three vertically polarized elements in a horizontal line. Each single element provides a beam width of approximately 55◦ and has a size of 79 × 50 mm. The overall size of the array is 300 mm × 200 mm. The distance between two adjacent antenna elements is d = 11.5 cm leading to d/λ ≈ 0.3. Simulations and measurements have shown that a beam steering angle of approximately ±60◦ can be achieved with the array. The overall gain of the antenna in the main direction is approximately 6.5 dBi and still 4.5 dBi towards ±60◦ . The whole array is build on FR4 substrate as shown in Fig.3.
176
Fig. 3.
Fig. 4.
Realized three element antenna array
Block diagram of the signal processing hardware
Fig. 5.
C. Signal Processing Hardware Fig.4 shows the block diagram of the signal processing hardware used to communicate with the RFID transponder and to collect the I and Q samples needed for the calculation of the AoA. Each of the modules was built on a separate circuit board attached to a common base board that is used for power supply and signal distribution (see Fig.5). The antenna array introduced in section IV-B is connected to three RFID reader modules. The antenna element in the middle of the array and the reader module attached to it can be used to communicate with the RFID transponder and to read the Electronic Product Code (EPC) that is stored on every ISO18000-6C compliant tag. During this reading process the two other reader modules connected to the two patch elements at both sides of the antenna array are switched to a listening mode. The main component of each of the reader modules is the integrated ISO18000-6C RFID reader IC R901G that was developed by IDS Microchip. In addition to the complete analog and digital signal processing parts for modulation, demodulation and data encoding/decoding this IC includes two analog mixers providing the baseband I and Q components of the received transponder signal. The R901G can be controlled and configured via serial (SPI) or parallel interface. It operates in the frequency range between 860 MHz and 960 MHz. The PLL module is used to feed every RFID reader module with a synchronous 868 MHz clock signal. The integral part
Signal processing hardware and antenna array
of that module is the ADF4360-7 IC by Analog Devices. This integrated synthesizer and VCO can be configured and controlled via SPI interface. It supports a frequency range between 350 MHz and 1800 MHz. The synchronous clock is used as a reference signal for the phase difference measurement in the calculation of the AoA. The control and communication module comprises of a microcontroller (ATMEL AT91SAM7X256 with 32 bit ARM7 core) and an analog to digital converter (Maxim MAX1304). The microcontroller configures and controls the PLL module and the three RFID reader modules using two integrated SPI interfaces. This module additionally executes every step in the ISO18000-6C communication protocol that is needed to read the tag’s EPC. After reception of the EPC the analog signals from each of the antenna array elements are mixed down to the base band and distributed to the analog to digital converter on the control and communication module. The ADC simultaneously samples all six baseband signals (I1-I3 and Q1Q3, see Fig.4) with a maximum sampling rate of 680 ksps and a 12-bit resolution for each signal. Data is exchanged with the microcontroller using a 12-bit parallel interface. To send the values of the sampled analog signals to the PC for further analysis the microcontroller’s UART interface is used. In combination with a level shifter for RS232 the signal processing hardware is able to communicate with a PC software via serial port interface.
177
Fig. 6.
Fig. 7.
Grid for localization measurements
D. Measurement environment The measurements were conducted in a test grid of size 3.0 m × 3.0 m comprising of 25 measurement points with a distance of 0.5 m in x direction and 0.5 m in y direction to each other (see Fig.6). The grid was set up in a standard 7.5 m × 12 m seminar room (see Fig.7). Three sides of the room were bounded by stone walls and the other side by a stone wall with windows and three radiators. A standard passive ISO 18000-6C DogBone RFID transponder by UPM Raflatac was used for measurement purposes. It stores a 96bit EPC and is optimized for corrugated board and plastic materials. It is mainly employed in industrial environments for supply chain management. For the localization process the transponder was attached to a piece of corrugated board and placed on a 1.5 m stand above the floor. The signal processing hardware and the antenna array were put on a table 1.5 m above the floor. To avoid the need for a directional coupler a separate transmit antenna was employed during the measurements. It was attached to a 1.5 m high stand. The transmit power was 2W ERP (3,28 W EIRP). The antenna arrays were positioned at five different positions in the four corners and the middle of one side of the plane (see Fig.6). For each antenna array position the angles of arrival of the transponder positioned at the 25 measurement points were calculated on a host PC. For this purpose the PC initiated the EPC reading process and received the sampled values of the I and Q components of the transponder signal received by each of the array elements. V. E XPERIMENTAL R ESULTS To analyze and rate the experimental results various parameters are calculated in this section using the estimated position (Xe , Ye ) and the real position (Xt , Yt ) of the transponder: The deviation (∆X, ∆Y ) = (|Xe − Xt |, |Ye − Yt |) from the p exact position of the transponder, the distance ∆D = (∆X)2 + (∆Y )2 between the actual and the estimated position of the transponder and the mean values mX , mY and
Picture of the measurement setup
2 2 for all 25 grid mD as well as the variances σX , σY2 and σD positions.
A. Determination of Optimum Results As a reference for the achievable accuracy which any subset selecting algorithm in the optimum case could achieve, we computed all combinations of two or three antenna estimates, respectively, for all 25 test locations. Then we choose in each case the estimate closest to the true position of the transponder and calculated the mean estimation error. This strategy of course assumes exact a-priory information of the transponder location and therefore can not be applied in a realistic approach. This reference calculation resulted in a mean error of 0.104 m for Approach III and 0.082 m for Approach IV as an upper bound to the estimation accuracy. B. Measurement Results for the Transponder Localization The results for both approaches using the RSS values as selection criterion for the considered 3 m x 3 m test grid are shown in Fig.8. The upper numbers at each measurement point show the absolute value of the localization error (in meters) for Approach III (i.e. select the set of the two antennas with the highest RSS and calculate the intercept point), the lower numbers those for Approach IV (i.e. select the set of three antennas with the highest RSS and calculate the centroid of the triangle formed by the intercept points). Tables II and III summarize the results for the mean estimation errors for Approach III and IV. Fig.9 depicts a histogram of the 25 distances ∆D between the actual and the estimated positions of the transponder. VI. C ONCLUSIONS AND O UTLOOK In this paper the localization of UHF RFID transponders using an AoA approach was investigated by conducting a measurement campaign with off-the-shelf components for generation and processing of ISO 18000-6C compliant signals. The phase difference between pairs of antenna elements in each antenna array was used to estimate the angle of arrival.
178
Fig. 9.
Fig. 8. Localization results for both approaches (upper numbers: Approach III, lower numbers: Approach IV) TABLE II M EAN VALUES FOR A PPROACH III mX = E{∆X} mY = E{∆Y } mD = E{∆D} 2 = E{(∆X − m )2 } σX X 2 = E{(∆Y − m )2 } σY Y 2 = E{(∆D − m )2 } σD D
0.073 m 0.060 m 0.107 m 0.005 m2 0.003 m2 0.005 m2
TABLE III M EAN VALUES FOR APPROACH IV mX = E{∆X} mY = E{∆Y } mD = E{∆D} 2 = E{(∆X − m )2 } σX X 2 = E{(∆Y − m )2 } σY Y 2 = E{(∆D − m )2 } σD D
Histogram of the measurement deviation ∆D
triangle (Approach IV). The results confirm that localization by AoA estimation is a practically feasible approach for the above mentioned scenarios. Most of the bad measurement results were clustered in the center of the test environment which is due to the reduced RSS in comparison to the edge positions near the antenna array positions. Further strategies of selecting subsets have to be developed and compared to the case where all measurement results are included in the position estimate. The robustness of the different approaches in more severe multipath environments has to be investigated. In the case of moving objects the prediction of the transponder position due to the assumed direction of movement can be included in the estimation process. ACKNOWLEDGMENT
0.096 m 0.078 m 0.139 m 0.007 m2 0.004 m2 0.007 m2
This work was partially sponsored by the German Federal Ministry of Education and Research R EFERENCES
To achieve spatial diversity a new algorithm based on RSS measurements was introduced that selects a certain subset of five available antenna array positions in the test environment. The position of the transponder was calculated for 25 test points using triangulation after selecting the appropriate antenna subset for the respective test point by using the RSS value. The results of the measurements improve the results of our last paper and show that an accurate estimation of the transponder locations is possible even in typical office environments. For a grid of size 3×3 meters with 25 test locations the mean deviation figures out to be E{∆D} = 0.107 m with a standard deviation of σ∆D = 0.07 m. Triangulating the transponder position from the AoAs of just three antennas leads to a performance loss of around 50 %. Similar as for the three antenna case it shows that a subset size of two using the intercept point (Approach III) outperforms a subset size of three antenna positions using the centroid of the related
179
[1] S. Azzouzi, M. Cremer, U. Dettmar, R. Kronberger, and T. Knie, “New Measurement Results for the Localization of UHF RFID Transponders Using an Angle of Arrival (AoA) Approach,” in IEEE RFID 2011, Apr. 2011. [2] Q. Spencer, M. Rice, B. Jeffs, and M. Jensen, “A Statistical Model for Angle of Arrival in Indoor Multipath Propagation,” in IEEE Vehicular Technology Conference, 1997, pp. 1415–1419. [3] Chon, Jun, Jun, and An, “Using RFID for Accurate Positioning,” in International Symposium on GNSS/GPS, Sydney, Australia, Dec. 2004. [4] Localization using Direction of Arrival. Frauenhofer Institut f¨ur integrierte Schaltungen. [Online]. Available: www.iis.fraunhofer.de/fhg/iis/EN/bf/ec/nl/omf/lw/index.jsp [5] H. Tatsumi, Y. Murai, T. Araki, and M. Miyakawa, “RFID Localization for the Visually Impaired,” Automation Congress, 2008. WAC 2008. World, pp. 1 – 6, 2008. [6] C. Hekimian-Williams, B. Grant, X. Liu et al., “Accurate Localization of RFID Tags Using Phase Difference,” in Proceedings of the RFID 2010. IEEE, April 2010. [7] M. Bouet and A. L. Dos Santos, “RFID Tags: Positioning Principles and Localization Techniques,” 2008 1st IFIP Wireless Days, pp. 1–5, 2008. [8] C. Byoung-Suk, L. Joon-Woo, and L. Ju-Jang, “An improved Localization System With RFID Technology for a Mobile Robot,” Industrial Electronics, 2008. IECON 2008. 34th Annual Conference of IEEE, pp. 3409 – 3413, 2008. [9] G. Chandrasekaran, M. A. Ergin, J. Yang et al., “Empirical Evaluation of the Limits on Localization Using Signal Strength,” 2009 6th Annual IEEE Communications Society Conference on Sensor, Mesh and Ad Hoc Communications and Networks, pp. 1–9, 2009.
[10] Phase Difference Based RFID Navigation for Medical Applications. RFID 2011, April 2011. [11] Holographic Localization of Passive UHF RFID Transponders. RFID 2011, Apr. 2011. [12] ISO/IEC 18000-6:2004 Information Technology – Radio Frequency Identification for Item Management - Part 6: Parameters for Air Interface Communications at 860 to 960 MHz, ISO/IEC Std., 2004. [13] EPC Radio-Frequency Identity Protocols Class-1 Generation-2 UHF RFID Protocol for Communications at 860 MHz – 960 MHz, EPC global Std., Rev. 1.0.9, January 2005. [14] Y. Zhang, M. B. Amin, and S. Kaushik, “Localisation and Tracking of Passive RFID Tags Based on Direction Estimation,” International Journal of Antennas and Propagation, vol. Volume 2007, no. ID 17426, 2007. [15] P. V. Nikitin, R. Martinez et al., “Phase Based Spatial Identification of UHF RFID Tags,” in IEEE RFID 2010, April 2010. [16] M. J. Abedin and A. S. Mohan, “RFID Localization Using Planar Antenna Arrays With Arbitrary Geometry,” 2008 Asia-Pacific Microwave Conference, pp. 1–4, 2008. [17] A. Alma’aitah, K. Ali et al., “3D Passive Tag Localization Schemes for Indoor RFID Applications,” in ICC 2010, Mai 2010. [18] J. Wang, M. B. Amin, and Y. Zhang, “Signal and Array Processing Techniques for RFID Readers,” in Wireless Sensing and Processing, ser. Proc. of SPIE, vol. 6248, 2006. [19] M. Abedin and A. Mohan, “Use of Smart Antennas for the Localisation of RFID Reader,” in APMC 2009, Dec. 2009. [20] S. Haykin, Adaptive Filter Theory. Upper Saddle River, New Jersey: Prentice-Hall, 1996.
180
Improved AoA Based Localization of UHF RFID Tags Using Spatial Diversity Salah Azzouzi, Markus Cremer, Uwe Dettmar, Thomas Knie and Rainer Kronberger Department of Telecommunication Cologne University of Applied Sciences, Germany Email: [email protected]
Abstract—In a recent paper we presented new measurement results for an angle of arrival based approach to localize UHF RFID transponders in an indoor environment with off-the-shelf IDS R901G RFID reader ICs and UPM Raflatac DogBone RFID transponders. We were able to localize the transponders with a mean accuracy of 0.21 m using three antenna array positions for triangulation in a 3 m x 3 m room. In this paper we present an RSS based algorithm to select a special subset of antenna array positions for the estimation of the transponder position in a measurement setup with five different antenna array positions. With this algorithm a reduction of the mean measurement error by 50% was achieved for the considered test grid compared to the original approach in [1].
I. I NTRODUCTION RFID technology is on the way to substitute bar codes for the identification of objects in many fields of application. With the introduction of low-cost passive UHF RFID transponders in the UHF frequency band (868 or 915 MHz) the operational range of RFID transponders was enlarged to distances of up to 10 meters. Typical applications include baggage handling, supply chain applications, or fixed asset tracking. However, the full potential of these transponders only will be exploited if the identification of objects is complemented by accurate localization. Therefore during the last years large efforts were taken to solve the problem of radio based indoor localization. For an overview see e.g. [2]–[6]. The RFID based techniques use measurements between RFID transponders and reader devices and some form of side information about the location of specific readers and/or reference transponders or statistical knowledge about the distribution of RSS values. Measurements are for the Received Signal Strength (RSS), the Angle of Arrival (AoA), the Time of Arrival (ToA), or the Time Difference of Arrival (TDoA) as well as hybrid procedures (e.g. see [7]–[11]). For the radio based RFID transponder localization in an indoor environment the time based approaches ToA and TDoA are not appropriate. Both of them require accurate time measurements. However, such measurements are quite difficult to implement with RFID systems since the defined data rates are far too low to receive a sufficient resolution. The RSS based trilateration methods show to be strongly related to the quality of the propagation channel which again is very sensitive to fading/multipath effects.
978-1-4577-0027-9/11/$26.00 ©2011 IEEE
In this paper we consider an AoA based localization approach that uses UHF RFID transponders according to the ISO standard 18000-6C (see [12], [13]) and multiple directional antenna arrays. Since the channel for RFID communication is typically small band (only a few MHz are available in the UHF band in Europe) an AoA approach is critical under severe multipath conditions, too. By applying the phase difference between incoming signals to estimate the position of the transponder the AoA approach totally depends on receiving a strong line-of-sight (LOS) path which can be isolated from other signal paths. However, even under good LOS conditions the superposition of multipath signals to the LOS signal will deteriorate the quality of the angle measurements. Therefore, for the 2D localization of an RFID tagged object the exploitation of more than the minimum two antenna positions in combination with triangulation may be beneficial. On the one hand a spatial diversity is established for the localization. On the other hand if the distance between transponder and reader antenna is small the effect of the estimation error of the incidence angle is reduced (see [14]). In environments with relaxed multipath and without shadowing the reader antennas closest to the transponder can be determined by RSSI measurements. To follow this approach low price reader modules using off-the-shelf components and self designed antenna arrays had to be implemented. The theoretical analysis of such systems in typically time variant multipath environments with strong deterministic components [15] is rather complicated and often relates to simplified signal models (see e.g. [14], [16]). This again results in significant differences between simulation and measurement results [6]. Due to the difficulty to access the IQ signals from a commercial RFID reader needed to evaluate the phase differences for AoA based approaches up to now most authors restrict to system simulations or the use of non standard compliant hardware (see [17]–[19]). The focus of this work is the evaluation of the achievable accuracy for a transponder localization under real life conditions. A simple RFID based localization approach using offthe-shelf components. A setup using IDS R901G RFID reader ICs, UPM Raflatec DogBone RFID transponders and multiple self-designed three element antenna arrays will be considered. The test environment is a seminar room in which an RFID transponder is placed at 25 different grid positions in a square plane of size 3 times 3 meters. The position of the transponder
174
TABLE I P REVIOUS MEASUREMENT RESULTS
mX = E{∆X} mY = E{∆Y } mD = E{∆D} 2 = E{(∆X − m )2 } σX X 2 = E{(∆Y − m )2 } σY Y 2 = E{(∆D − m )2 } σD D
Approach I
Approach II
0.18 m 0.17 m 0.26 m 0.05 m2 0.02 m2 0.06 m2
0.14 m 0.13 m 0.21 m 0.02 m2 0.01 m2 0.02 m2
is then determined for each position by estimating the angle of arrival measured by five antenna arrays placed outside of the plane. Application scenarios for AoA based localization of RFID transponders in relaxed multipath environments include the problem of finding RFID tagged books on a desk for sorting purposes in a library or the localization of tools or work pieces on a table during the fabrication process. The rest of the paper is organized as follows. Section II introduces our first measurement results using the same measurement setup as in this paper. Section III describes the AoA based approach to estimate the position of an object under investigation. The experimental setup and measurement results are presented in sections IV and V. Section VI concludes the paper with a summary and discussion. II. R ELATED W ORK In a recent paper [1] we described the localization of UHF RFID transponders using an AoA approach with three uniform linear antenna arrays with three patch elements each and present measurement results for a seminar room where a transponder was moved to 25 positions on a 0.5 m x 0.5 m testgrid (see Fig.1). The angle of arrival of the received tag signal was measured at each of the three antenna array positions and for each of the 25 transponder positions. Two strategies were followed to determine the transponder position in the measurement setup. In the first approach the centroid of the triangle defined by the three intercept points of the lines originating from each antenna array was used as an estimate for the actual transponder position. Fig.1 depicts the result of this approach for measurement point 25. In the second approach a subset of two out of three array positions was chosen to calculate the position of the transponder for each test point. The intercept point of the lines initiating from these two antennas was used as the estimate for the transponder position. The subset was determined as ˆ 3 of antenna array 3 as a function of the estimated angle Θ follows: ˆ 3 ≥ 0 choose the intercept point of arrays 1 and 3 else If Θ choose the intercept point of arrays 2 and 3 As a reference table I shows the measurement results of the tag position for both strategies (Approach I and Approach II). Estimating the transponder position using Approach II leads to the smallest mean positioning error of 0.21 m.
Fig. 1. Grid for localization measurements with three antenna array positions
III. L OCALIZATION M ETHOD The proposed angle of arrival based localization approach consists of two steps. In the first step the incidence angles Θn of the received transponder signal with respect to known positions of N receive antenna arrays are estimated. In the second step the position of the transponder is determined based on the knowledge of the angles Θn by using a triangulation method. To estimate the angle of arrival of the transponder signal a phased antenna array composed of M individual antenna elements is used. The single elements of the array are aligned with an equal distance d to each other along a horizontal line to form a uniform linear array (ULA, see Fig.2). The transponder signal will arrive at each of the array elements with a certain time delay resulting in a constant phase shift ∆φ with respect to a reference phase. ∆φ is related to the incidence angle Θ of a plane wave measured with respect to the normal to the linear array as 2πd ∆φ = sin Θ. λ λ is the carrier wave length [20]. To estimate the incidence angle Θ the received signals from the M = 3 antenna elements are evaluated. Applying the root-MUSIC method (see e.g. [14], [18]) to the case of one signal source and a three element ULA Θ is calculated from the measured phase shift ∆φˆ as: λ ˆ = arcsin ∆φˆ (1) Θ 2πd where ∆φˆ = 6 {E[x∗i (t)xi+1 (t)]}.
(2)
with i = 1, 2. ∗ denotes the complex conjugate, E[.] the statistic expectation value, and 6 {.} is the phase of a complex variable. xi (t) and xi+1 (t) describe the two IQ demodulated and filtered complex baseband signals from antenna i and i+1, respectively. To reduce the observation time the correlation coefficients from both pairs of antennas can be added before
175
IV. E XPERIMENTAL S ETUP A. RFID Localization System
Fig. 2.
One unit of the localization system comprises the signal processing hardware and a self designed three element phased array. Three RFID reader modules combined on one base circuit board were implemented to communicate with the transponder and process the received signals. Five of these hardware units are deployed at the four corners and in the center of the test grid to form the localization system (see Fig.6). The IQ components of the received signal from each antenna array element are sampled and read out for further processing. The estimates for Θn are calculated according to equations 1 and 2 and used to compute and plot the estimated position of the transponder. This is done on a host computer connected to the hardware units.
Phased array with ideal antennas
taking the argument. In general d < λ/2 is chosen to avoid the appearance of grating lobes. Having estimated the angles of incidence (Θ1 , ...ΘN ) for all N antenna arrays in the second step the probability of the transponder being at position (x, y) under observation of (Θ1 , ...ΘN ) P r((x, y)|(Θ1 , ...ΘN )) has to be maximized. Assuming perfect knowledge of the incidence angles (Θ1 , ...ΘN ) this can be done by constructing lines from the N antenna positions towards the direction of the corresponding Θn . One interception point results that represents the position of the transponder. In this case even a setup with only two arrays is sufficient to exactly find the position of the object under investigation. However, in reality imperfections in measuring the exact angle and perturbations like reflections or multipath lead to substantial deviations between estimated and correct position. In this paper we present two strategies to find appropriate subsets of the measured angels of incidence from N = 5 antenna positions as a simple approach for tag localization with low computational load. Both strategies make use of the assumption that the antenna array with the smallest distance to the transponder position will produce the best angle of arrival estimate. Because of the fact that the transponder position is not an available a-priori information the received signal strength (RSS) is computed during the AoA estimation process. The two algorithms for selecting appropriate antenna subsets can be formulated as follows: Approach III: Determine the subset of the two antennas with the highest RSS values and calculate the intercept point of the lines originating from these two antennas as an estimate for the transponder position. Approach IV: Determine the subset of the three antennas with the highest RSS values and compute the centroid of the triangle defined by the intercept points of the lines originating from these three antenna arrays as an estimate for the actual transponder position.
For the calculation of the AoA the transponder response containing the Electronic Product Code (EPC) and the CRC of length 16 is evaluated. From this signal 240 samples (relating to around 70 symbols) are used for computing the correlation as given by equation 2. From the estimate ∆φˆ the spatial angle ˆ is calculated. As described in [18] the phase characteristic Θ of the antenna pattern distorts the determination of the spatial ˆ To eliminate this effect a reference measurement of angle Θ. ˆ was conducted in an anechoic the characteristic Θ = f (Θ) antenna chamber (see our recent paper [1]). This characteristic ˆ was used in the sequel to correct Θ. As a measure of the received signal strength (RSS) of the incoming signal from the transponder the sum of the mean signal power of the received baseband IQ signals of the three array elements was used. First, the mean signal power Em,n for the M = 3 single elements of array n was calculated using K = 240 I and Q samples: Em,n =
K q 1 X Ik2 + Q2k K k=1
The mean received signal power of array n was then estimated as M 1 X En = Em,n . M m=1 B. Antenna The receive antenna of each hardware unit is designed as a microstrip patch antenna array with three vertically polarized elements in a horizontal line. Each single element provides a beam width of approximately 55◦ and has a size of 79 × 50 mm. The overall size of the array is 300 mm × 200 mm. The distance between two adjacent antenna elements is d = 11.5 cm leading to d/λ ≈ 0.3. Simulations and measurements have shown that a beam steering angle of approximately ±60◦ can be achieved with the array. The overall gain of the antenna in the main direction is approximately 6.5 dBi and still 4.5 dBi towards ±60◦ . The whole array is build on FR4 substrate as shown in Fig.3.
176
Fig. 3.
Fig. 4.
Realized three element antenna array
Block diagram of the signal processing hardware
Fig. 5.
C. Signal Processing Hardware Fig.4 shows the block diagram of the signal processing hardware used to communicate with the RFID transponder and to collect the I and Q samples needed for the calculation of the AoA. Each of the modules was built on a separate circuit board attached to a common base board that is used for power supply and signal distribution (see Fig.5). The antenna array introduced in section IV-B is connected to three RFID reader modules. The antenna element in the middle of the array and the reader module attached to it can be used to communicate with the RFID transponder and to read the Electronic Product Code (EPC) that is stored on every ISO18000-6C compliant tag. During this reading process the two other reader modules connected to the two patch elements at both sides of the antenna array are switched to a listening mode. The main component of each of the reader modules is the integrated ISO18000-6C RFID reader IC R901G that was developed by IDS Microchip. In addition to the complete analog and digital signal processing parts for modulation, demodulation and data encoding/decoding this IC includes two analog mixers providing the baseband I and Q components of the received transponder signal. The R901G can be controlled and configured via serial (SPI) or parallel interface. It operates in the frequency range between 860 MHz and 960 MHz. The PLL module is used to feed every RFID reader module with a synchronous 868 MHz clock signal. The integral part
Signal processing hardware and antenna array
of that module is the ADF4360-7 IC by Analog Devices. This integrated synthesizer and VCO can be configured and controlled via SPI interface. It supports a frequency range between 350 MHz and 1800 MHz. The synchronous clock is used as a reference signal for the phase difference measurement in the calculation of the AoA. The control and communication module comprises of a microcontroller (ATMEL AT91SAM7X256 with 32 bit ARM7 core) and an analog to digital converter (Maxim MAX1304). The microcontroller configures and controls the PLL module and the three RFID reader modules using two integrated SPI interfaces. This module additionally executes every step in the ISO18000-6C communication protocol that is needed to read the tag’s EPC. After reception of the EPC the analog signals from each of the antenna array elements are mixed down to the base band and distributed to the analog to digital converter on the control and communication module. The ADC simultaneously samples all six baseband signals (I1-I3 and Q1Q3, see Fig.4) with a maximum sampling rate of 680 ksps and a 12-bit resolution for each signal. Data is exchanged with the microcontroller using a 12-bit parallel interface. To send the values of the sampled analog signals to the PC for further analysis the microcontroller’s UART interface is used. In combination with a level shifter for RS232 the signal processing hardware is able to communicate with a PC software via serial port interface.
177
Fig. 6.
Fig. 7.
Grid for localization measurements
D. Measurement environment The measurements were conducted in a test grid of size 3.0 m × 3.0 m comprising of 25 measurement points with a distance of 0.5 m in x direction and 0.5 m in y direction to each other (see Fig.6). The grid was set up in a standard 7.5 m × 12 m seminar room (see Fig.7). Three sides of the room were bounded by stone walls and the other side by a stone wall with windows and three radiators. A standard passive ISO 18000-6C DogBone RFID transponder by UPM Raflatac was used for measurement purposes. It stores a 96bit EPC and is optimized for corrugated board and plastic materials. It is mainly employed in industrial environments for supply chain management. For the localization process the transponder was attached to a piece of corrugated board and placed on a 1.5 m stand above the floor. The signal processing hardware and the antenna array were put on a table 1.5 m above the floor. To avoid the need for a directional coupler a separate transmit antenna was employed during the measurements. It was attached to a 1.5 m high stand. The transmit power was 2W ERP (3,28 W EIRP). The antenna arrays were positioned at five different positions in the four corners and the middle of one side of the plane (see Fig.6). For each antenna array position the angles of arrival of the transponder positioned at the 25 measurement points were calculated on a host PC. For this purpose the PC initiated the EPC reading process and received the sampled values of the I and Q components of the transponder signal received by each of the array elements. V. E XPERIMENTAL R ESULTS To analyze and rate the experimental results various parameters are calculated in this section using the estimated position (Xe , Ye ) and the real position (Xt , Yt ) of the transponder: The deviation (∆X, ∆Y ) = (|Xe − Xt |, |Ye − Yt |) from the p exact position of the transponder, the distance ∆D = (∆X)2 + (∆Y )2 between the actual and the estimated position of the transponder and the mean values mX , mY and
Picture of the measurement setup
2 2 for all 25 grid mD as well as the variances σX , σY2 and σD positions.
A. Determination of Optimum Results As a reference for the achievable accuracy which any subset selecting algorithm in the optimum case could achieve, we computed all combinations of two or three antenna estimates, respectively, for all 25 test locations. Then we choose in each case the estimate closest to the true position of the transponder and calculated the mean estimation error. This strategy of course assumes exact a-priory information of the transponder location and therefore can not be applied in a realistic approach. This reference calculation resulted in a mean error of 0.104 m for Approach III and 0.082 m for Approach IV as an upper bound to the estimation accuracy. B. Measurement Results for the Transponder Localization The results for both approaches using the RSS values as selection criterion for the considered 3 m x 3 m test grid are shown in Fig.8. The upper numbers at each measurement point show the absolute value of the localization error (in meters) for Approach III (i.e. select the set of the two antennas with the highest RSS and calculate the intercept point), the lower numbers those for Approach IV (i.e. select the set of three antennas with the highest RSS and calculate the centroid of the triangle formed by the intercept points). Tables II and III summarize the results for the mean estimation errors for Approach III and IV. Fig.9 depicts a histogram of the 25 distances ∆D between the actual and the estimated positions of the transponder. VI. C ONCLUSIONS AND O UTLOOK In this paper the localization of UHF RFID transponders using an AoA approach was investigated by conducting a measurement campaign with off-the-shelf components for generation and processing of ISO 18000-6C compliant signals. The phase difference between pairs of antenna elements in each antenna array was used to estimate the angle of arrival.
178
Fig. 9.
Fig. 8. Localization results for both approaches (upper numbers: Approach III, lower numbers: Approach IV) TABLE II M EAN VALUES FOR A PPROACH III mX = E{∆X} mY = E{∆Y } mD = E{∆D} 2 = E{(∆X − m )2 } σX X 2 = E{(∆Y − m )2 } σY Y 2 = E{(∆D − m )2 } σD D
0.073 m 0.060 m 0.107 m 0.005 m2 0.003 m2 0.005 m2
TABLE III M EAN VALUES FOR APPROACH IV mX = E{∆X} mY = E{∆Y } mD = E{∆D} 2 = E{(∆X − m )2 } σX X 2 = E{(∆Y − m )2 } σY Y 2 = E{(∆D − m )2 } σD D
Histogram of the measurement deviation ∆D
triangle (Approach IV). The results confirm that localization by AoA estimation is a practically feasible approach for the above mentioned scenarios. Most of the bad measurement results were clustered in the center of the test environment which is due to the reduced RSS in comparison to the edge positions near the antenna array positions. Further strategies of selecting subsets have to be developed and compared to the case where all measurement results are included in the position estimate. The robustness of the different approaches in more severe multipath environments has to be investigated. In the case of moving objects the prediction of the transponder position due to the assumed direction of movement can be included in the estimation process. ACKNOWLEDGMENT
0.096 m 0.078 m 0.139 m 0.007 m2 0.004 m2 0.007 m2
This work was partially sponsored by the German Federal Ministry of Education and Research R EFERENCES
To achieve spatial diversity a new algorithm based on RSS measurements was introduced that selects a certain subset of five available antenna array positions in the test environment. The position of the transponder was calculated for 25 test points using triangulation after selecting the appropriate antenna subset for the respective test point by using the RSS value. The results of the measurements improve the results of our last paper and show that an accurate estimation of the transponder locations is possible even in typical office environments. For a grid of size 3×3 meters with 25 test locations the mean deviation figures out to be E{∆D} = 0.107 m with a standard deviation of σ∆D = 0.07 m. Triangulating the transponder position from the AoAs of just three antennas leads to a performance loss of around 50 %. Similar as for the three antenna case it shows that a subset size of two using the intercept point (Approach III) outperforms a subset size of three antenna positions using the centroid of the related
179
[1] S. Azzouzi, M. Cremer, U. Dettmar, R. Kronberger, and T. Knie, “New Measurement Results for the Localization of UHF RFID Transponders Using an Angle of Arrival (AoA) Approach,” in IEEE RFID 2011, Apr. 2011. [2] Q. Spencer, M. Rice, B. Jeffs, and M. Jensen, “A Statistical Model for Angle of Arrival in Indoor Multipath Propagation,” in IEEE Vehicular Technology Conference, 1997, pp. 1415–1419. [3] Chon, Jun, Jun, and An, “Using RFID for Accurate Positioning,” in International Symposium on GNSS/GPS, Sydney, Australia, Dec. 2004. [4] Localization using Direction of Arrival. Frauenhofer Institut f¨ur integrierte Schaltungen. [Online]. Available: www.iis.fraunhofer.de/fhg/iis/EN/bf/ec/nl/omf/lw/index.jsp [5] H. Tatsumi, Y. Murai, T. Araki, and M. Miyakawa, “RFID Localization for the Visually Impaired,” Automation Congress, 2008. WAC 2008. World, pp. 1 – 6, 2008. [6] C. Hekimian-Williams, B. Grant, X. Liu et al., “Accurate Localization of RFID Tags Using Phase Difference,” in Proceedings of the RFID 2010. IEEE, April 2010. [7] M. Bouet and A. L. Dos Santos, “RFID Tags: Positioning Principles and Localization Techniques,” 2008 1st IFIP Wireless Days, pp. 1–5, 2008. [8] C. Byoung-Suk, L. Joon-Woo, and L. Ju-Jang, “An improved Localization System With RFID Technology for a Mobile Robot,” Industrial Electronics, 2008. IECON 2008. 34th Annual Conference of IEEE, pp. 3409 – 3413, 2008. [9] G. Chandrasekaran, M. A. Ergin, J. Yang et al., “Empirical Evaluation of the Limits on Localization Using Signal Strength,” 2009 6th Annual IEEE Communications Society Conference on Sensor, Mesh and Ad Hoc Communications and Networks, pp. 1–9, 2009.
[10] Phase Difference Based RFID Navigation for Medical Applications. RFID 2011, April 2011. [11] Holographic Localization of Passive UHF RFID Transponders. RFID 2011, Apr. 2011. [12] ISO/IEC 18000-6:2004 Information Technology – Radio Frequency Identification for Item Management - Part 6: Parameters for Air Interface Communications at 860 to 960 MHz, ISO/IEC Std., 2004. [13] EPC Radio-Frequency Identity Protocols Class-1 Generation-2 UHF RFID Protocol for Communications at 860 MHz – 960 MHz, EPC global Std., Rev. 1.0.9, January 2005. [14] Y. Zhang, M. B. Amin, and S. Kaushik, “Localisation and Tracking of Passive RFID Tags Based on Direction Estimation,” International Journal of Antennas and Propagation, vol. Volume 2007, no. ID 17426, 2007. [15] P. V. Nikitin, R. Martinez et al., “Phase Based Spatial Identification of UHF RFID Tags,” in IEEE RFID 2010, April 2010. [16] M. J. Abedin and A. S. Mohan, “RFID Localization Using Planar Antenna Arrays With Arbitrary Geometry,” 2008 Asia-Pacific Microwave Conference, pp. 1–4, 2008. [17] A. Alma’aitah, K. Ali et al., “3D Passive Tag Localization Schemes for Indoor RFID Applications,” in ICC 2010, Mai 2010. [18] J. Wang, M. B. Amin, and Y. Zhang, “Signal and Array Processing Techniques for RFID Readers,” in Wireless Sensing and Processing, ser. Proc. of SPIE, vol. 6248, 2006. [19] M. Abedin and A. Mohan, “Use of Smart Antennas for the Localisation of RFID Reader,” in APMC 2009, Dec. 2009. [20] S. Haykin, Adaptive Filter Theory. Upper Saddle River, New Jersey: Prentice-Hall, 1996.
180
