Design of an Embedded UWB Hardware Platform for ... - IEEE Xplore

10 downloads 0 Views 4MB Size Report
for Navigation in GPS Denied Environments. Frank Hartmann, Felix Pistorius, Andreas Lauber, Kai Hildenbrand, Jürgen Becker and Wilhelm Stork. Institute for ...
Design of an Embedded UWB Hardware Platform for Navigation in GPS Denied Environments Frank Hartmann, Felix Pistorius, Andreas Lauber, Kai Hildenbrand, J¨urgen Becker and Wilhelm Stork Institute for Information Processing Technologies (ITIV), Karlsruhe Institute of Technology (KIT) Engesserstr. 5, 76131 Karlsruhe, Germany Email: {frank.hartmann3, felix.pistorius, andreas.lauber}@kit.edu Abstract—Autonomy of vehicles is a widely discussed topic, increasing the need for low power, inexpensive and precise localization and navigation. Due to ongoing research and standardization in last decades, ultra-wideband (UWB) based localization technology is more and more ready for large scale usage. In this paper we propose an embedded UWB localization system, consisting of an inexpensive, low power hardware platform deploying an IEEE 802.15.4a-2011 UWB-PHY standard conform RF chip, enabling communication and localization, a software framework, and localization algorithms. The key contribution of the paper is this system concept and experimental results using a developed system in an underground parking garage. Within this system evaluation we show in an accuracy analysis an achievable positioning error of around 30 cm. Furthermore we show the useage of such an UBW localization system in an use case near scenario.

I.

I NTRODUCTION AND R ELATED W ORK

In the last decades localization and navigation have become widely discussed topics in research and industry. This interest is attracted by the potential opportunities in many application fields, such as personal navigation, logistics and autonomous guided transport vehicles (AGVs), automation in farming, automobile traffic control and route planning, just to name a few. In all these applications the determination of a precise absolute position is of paramount importance to the quality of any provided Location Based Service (LBS). As an example, most scenarios in automobile traffic control and route planning, are situated in an outdoor environment where localization with a global navigation satellite system (GNSS) is a standard approach, providing sufficient accuracy. Although indoor navigation is only a corner case in automotive applications, use cases such as parking pilots in parking garages exist, just to name one. In future autonomous parking scenarios a car will drive autonomously inside the parking garage and maneuver to the parking space that was communicated to the vehicle. For such applications a high accuracy of localization in the centimeter range is needed. However in an indoor or GPS denied environment like in cities with urban canyon or parking garages, GNSS based localization is not applicable [1]. This gets even worse for underground parking garages as seen in Figure 1. In this paper we are using the aforementioned use case, however the concepts is adaptable on other use cases too. Different local positioning systems and approaches for indoor environments have been proposed recently, deploying different sensors and algorithms in order to generate a position information. State of the art of systems can be divided into

c 978-1-4673-9907-4/15/$31.00 2015 European Union

Figure 1: Parking garage as an use case environment for experimental system evaluation study.

relative and absolute localization systems, as well as hybrid systems trying to preserve the advantages of both. Relative localization systems deploying odometry or dead reckoning systems evaluate measurements from internal sensors such as accelerometers or rotationary encoders. By integrating the data over time a movement vector is resulting, which is further used to update the current relative position. Due to integrative evaluation of relative sensor data these systems show an increasing drift error with operation time leading to system failure. [2]. Furthermore, because the measurement inputs are not linked to the environment the relative position is not relatable as an absolute position on a map. Especially in scenarios like autonomous parking where an environment context exists this drawback limits the use of pure odometry systems. These problems are resolved by ranging based systems measuring their distance to reflecting environmental artifacts, like walls or posts, which make a mapping to an absolute position possible. The ranging measurement is mostly done by lidar, radar or ultrasonic based systems performing Time of Flight (ToF) or triangulation measurements. Unfortunately for high accuracies of absolute positions highly detailed maps are necessary. In addition, in most scenarios, particularly in underground parking garages, mapping results in ambiguousness because of similar building structure [3]. To overcome the inherent problems of relative positioning, infrastructure based localization systems are widely discussed. The prevalent approach is to implement a wireless sensor network (WSN) in which a mobile node can locate itself, by evaluating a measured distance related indicator to the fixed anchor nodes of the WSN. Most commonly, due to its easy availability in

nearly all radio communication systems, the received signal strength (RSS) of a radio link communication is evaluated. Yet, not only obtainable RSS values are known to be highly influenced by the environment through fading, but also the error of the distance estimation is increasing over the measurement range. This can be countered with either dense WSNs [4] resulting in cost and scalability problems, or by more complex systems fusing sensor data from different measurement inputs [5] [6]. A promising relative new technology is Ultra WideBand (UWB) communication. Its large bandwidths not only enables high rate data transmission, but also highly resolved time measurements making ranging with Time-of-Flight (ToF) measurements possible. In the past UWB systems have been very costly and applications deploying UWB suffered from high power consumptions. Due to many research contributions focusing on RF design, low-level protocols and propagation channels [7] huge progress was made in UWB technology [8] and UWB localization [9]. In the application field context of autonomous vehicles UWB technology for localization is a widely discussed topic [10]. Subject of recent research is not only the achievable accuracy, but also network architectures, cooperative localization mechanisms and V2X use case scenarios [11], [12], [13]. However the majority of the proposed work utilizes proprietary hardware [14] [15] and deploys non standardized protocols [16], leading to difficult comparability and reproducibility. Recently new integrated UWB radio communication chips implementing the IEEE 802.15.4a-2011 standard became available on the market. This is giving the opportunity for effectively evaluating UWB technology in any application scenario [17] with large measurement campaigns, obtaining comparable data. In this Paper we propose a novel embedded UWB hardware platform for precise localization and navigation in indoor environments. We explain the deployed system concept and architecture and further show the hardware design of the developed UWB nodes, as well as the implemented algorithms for positioning and localization. To demonstrate achievable accuracies and the system performance in an use case near scenario we carried out a large measurement campaign focusing in different tests on specific system characteristics. The system evaluation is based on real measurements conducted in an underground parking garage. II.

S YSTEM C ONCEPT AND A RCHITECTURE

In order to enable autonomous vehicles to locate and navigate themselves in an indoor or GPS denied environment a UWB localization system is presented and evaluated in this paper. The systems concept is visualized in a parking pilot application scenario in Figure 2, where a vehicle locates and navigates itself in an underground parking lot. The system consists of a UWB network infrastructure, being a part of the sites infrastructure service. The network is spanned by anchor nodes distributed on the setting, which enables a UWB Tag node equipped vehicle to locate itself inside this network. This localization in a temporal sequence enables the vehicle to track its past trajectory and could be used to plan and control its future trajectory. In this further section we first describe the network architecture and the deployed communication protocols, to give an overview on the setup and the interaction between single modules and functions. Thereafter we describe in detail the PCB design and the

Figure 2: System concept and use case : Autonomous vehicle locating itself in an UWB network equipped parking garage, tracking its past trajectory and planning its future drive path towards a free parking slot

scheme of the ITIV UWB LocNode. Finally we enlarge upon the issue of determination of the position by our embedded software framework and algorithms. A. Network Architecture and Protocols In order to realize localization and navigation a UWB network has to be spanned, in which a mobile tag can acquire its current position. As displayed in Figure 2 the application site has multiple, distributed anchor nodes as a fixed infrastructure. The positions and addresses are predefined and programmed into all UWB anchor nodes. If a vehicle equipped with a mobile UWB tag node enters the site, it queries the list of the existing anchor nodes in the environment with a broadcast message. Subsequently by answering on that broadcast an anchor node transmits the sites infrastructure information, like anchor nodes position and detailed map. Once this anchor node list is obtained, the tag will cyclically poll the available anchor nodes for performing the standard SDS-TWToF measurement protocol [17]. Thereby the tag is allocating the elapsed time between a transmission and the reception of its corresponding answer from the anchor and tag side and is considering the those timestamps for its ranging algorithm. Accordingly the acquired distances from the reachable anchor nodes are allocated by the mobile tag node in a dynamical measurement matrix and are further processed with algorithms to a movement information, containing the current position and current velocity vector. The algorithms deployed in this paper are further explained in section II-C. We implemented these protocols and algorithms on the ITIV UWB LocNode and its embedded micro-controller, which is described in the following subsection. B. Hardware and PCB Design The printed circuit board (PCB) design of the developed ITIV UWB LocNode and its functional parts are shown in Figure 3. The outline of the PCB layout is in the size of a check card format with the lateral dimensions of 71 mm by

Figure 3: PCB Design of the ITIV UWB LocNode: front side is shown in the middle and two parts of the back side (right and left).

46 mm. On the upper right side of Figure 3 the power supply is illustrated, it is located on the backside of the PCB. For a simple and inexpensive way to regulate output voltage of 3, 3 V from two input supply possibilities coming from battery or USB connection, we used a linear-low dropout regulator (LDO) from TI. Power consumption is strongly dependent on the operational mode, in an empirical study we determined power consumption to be 800 mW in maximum. On the upper left side of the PCB design the UWB radio is located. For this we used the IEEE 802.15.4-2011 standard conform DWM1000 UWB module from DecaWave, containing the DW1000 radio chip and an integrated chip antenna. By designing the RF PCB part around the DWM1000 module we were following RF and antenna guidelines, keeping signal and ground-planes away. For the easier use we considered two dip switches for in-use hardware configuration. One for configuring OnBoard communication like, SPI mode and GPIO selection, the other one for RF network configuration such as tag or anchor selection, address configuration, channel selection and UWB communication mode (preamble length, pulse repetition frequency, etc.). For debugging purposes we attached status LEDs to display software and network status information. The core part of the board is the STM32F105RCT6 microcontroller, controlling all functionalities, such as interfacing to the UWB module and running UWB network protocols as well as positioning and movement algorithms. In addition, it provides interfaces such as USB, CAN, SPI, GPIOs, for additional OnBoard and Off-Board or Break-Out Board communication. The micro-controller is programmed over a 10 pin JTag interface, which is also used for debugging of the firmware. In Figure 4 an overview on the interconnection and the location of the different functionalities is given.

radio packets and the time-stamping of those transmissions and receptions, the STM micro-controller picks up those timestamps through a SPI interface and hands it, for further processing, over to the positioning and movement algorithm entity. As described earlier, the mobile UWB Tag associated with a vehicle has to calculate its position out of the timestamps from each communication transaction between the transmitted poll for measurements and its final response message. The ranging algorithm considers those timestamps in order to estimate a robust propagation time needed by a pulse to reach the recipient. This propagation time between the link partners is converted into a distance by using standard light speed Equation (1). dT oF = c ∗ tT oF (1) For the experimental evaluation in this paper, described in section IV we considered speed of light to be c = 299702547.0 m/s. The estimated distances between the tag and different anchor nodes are used to estimate a position and velocity, therefore we implemented two different positioning algorithms explained in the following subsections:

C. Embedded Positioning Algorithms Figure 4 displays the application framework and maps the sub-functionalities to the hardware components and software entities of the localization and navigation system. While the UWB module undertakes UWB-PHY specific tasks, for instance transmitting and receiving IEEE 802.15.4a compliant

Figure 4: Overview on the the mapping of functionalities to the hardware components.

1) Least Squares Trilateration with Selective Filtering: In order to estimate a 2D position out of ranging data we implemented a Least Squares Estimator to solve the trilateration problem, consisting out of the euclidean distances equations between a fixed anchor position and the unknown tag node position. By considering at least 3 different anchor-tag connections the problem can be linearized and a standard Least Squares Estimator can be utilized for solving [17]. Before estimating the position out of the incoming ranges we applied a range based filtering, by taking into account that with a longer ToF the likelihood of a connection to become non line of sight and being more disturbed is higher. We took that into account by setting up a data structure saving all the distances during one tag to anchor poll round and by only choosing the distances out of that data structure with the smallest ranges.

the way segment, calculated by the multiplication of the passed time and the last estimated velocity state T ∗ [xv , yv ].

2) Kalman Filtering with Motion-Model: Due to the fact that we are reckoning a motion with the obtained distance measurements we deployed a Kalman Filter (KF) with a linearized measurement input as a second positioning and movement algorithm. For state space filtering the KF is using a standard constant motion model shown in Equation (2), in order to fuse the estimated distances to a 2D position.

(4)

s˙ = v v˙ = a = 0

(2)

The time discrete state space Kalman Filter system model is displayed in the following equations. During prediction phase the state-transition matrix Φk maps the previous state x+ k to the a priori state estimate of the state vector, and maps the previous system covariance matrix Pk+ in addition with the covariance of the process noise Qk to the a priori system covariance matrix. In order to weight the measurement inputs and the system state estimates, as well as for estimation of the a posteriori system covariance matrix, the Kalman Gain Matrix Kk is calculated in the update phase. Therefore the a priori system covariance matrix is transformed by the Jacobian matrix Ck of the observation matrix hk set into proportion to the innovation covariance including the covariance of the observation noise Rk . The a posteriori estimate of the state vector x+ k arises by correcting the a priori state estimate with the weighed difference of the a priori state estimate transformed by the observation matrix hk and the measurement inputs yk . Prediction Phase: x ˆ− ˆ+ k+1 = Φk x k − Pk+1 = Φk Pk+ ΦTk + Qk Update Phase: − Kk = Pk+1 CkT

− Ck Pk+1 CkT + Rk  x ˆ+ ˆ− ˆ− k =x k+1 − Kk hk x k+1 − yk

-1

− Pk+ = (I − Kk Ck ) Pk+1

Taking into account that a motion in the 2-D state space can be decomposed in a x- and y- direction component we translated the standard motion model in Equation (2) into the state vector xk in Equation (3) where xp and yp are the position state variables and xv and yv are the velocity state variables. With the state-transition matrix Φk in Equation (3), a next position state [xp , yp ] results out of the sum of the old position plus

x ˆ− k+1

 1 0 + = Φk x ˆk =  0 0

T 1 0 0

0 0 1 0

  ˆ + xp 0 0  xv  T   yp  1 k yv k

(3)

In case new measurements are available, the measurement equation yk = hk (xk ) is used to directly fuse those distances with the state estimates of the KF. The measurement equation is build for the case that during one filter step multiple distances are obtained dynamically as shown in Equation (5). yk = hk (xk ) q    2 2 (xs − ax1 ) + (ys − ay1 ) d1    ..    ..  .= . q  di 2 2 (xs − axi ) + (ys − ayi )

Due to the fact that the observation matrix hk as displayed in Equation (4), is of nonlinear shape and in order to process the data inside the KF we linearize the equation with the Jacobean matrix Ck shown in Equation (5) by applying the partial derivation on the measurement equation in every filter step. ∂hk (xk ) Ck = (5) ∂xk xk =ˆxk In order to parametrize the covariances in the KF we characterized our system within a measurement campaign gaining optimal values for the covariance matrix of the process noise Qk and the covariance matrix of the observation noise Rk . We evaluated the measured distances between two UWB link partners and determined the averaged variances of the error of distance measurements. Measurement data revealed that in a line of sight scenario the variances of the error of distance is nearly constant for a normal use distances in our application (0 − 50 m). Therefore we parametrized the dynamically build observation noise covariance matrix Rk as a diagonal matrix with the constant variances of the error r1,2,3,... = 0, 5. In an empirical study we determined the process noise covariance matrix Qk resulting in Equation (6), with q = 0.00325 as a value being optimal for our measurement scenario.   0 0 0 0 0 q 0 0 Qk =  (6) 0 0 0 0 0 0 0 q III.

M EASUREMENT S ETUP

In order to assess the performance of the proposed embedded UWB localization system concept presented in Section II we chose an underground parking garage as a typical evaluation site for the use case of an indoor autonomous vehicle parking system. A picture of the evaluation site is given in Figure 1. This locality, providing parking slots for around 45 cars and a pitch for motor bikes and bicycles, has a floor space of 1650 m2 . The outline of this underground garage is shown as a background of the Figures 5 and Figure 7. We distributed 12 anchors in one half of the garage, and mounted them in an height of 2 m. Each anchor node was powered

Position of Measurement Trajectory and Calculated Position

Pos. Error in [m] over Measurment Time in [ms]

25 Anchor Positions Trajectory Kalman Filter Least Squares

20

Kalman Filter: Mean Error:0.26509m Least Squares: Mean Error:0.43587m 1

0.8

10

Error [m]

y−Coordinate [m]

15

5

0.6

0.4 0 0.2 −5 0 −10 −10

0

10

20

30

40

50

x−Coordinate [m]

0

1

2

3

4

Time [ms]

5

6

7 5

x 10

Figure 5: Accuracy measurements with a robot following a trajectory.

Figure 6: Positioning error of the KF and Least Squares obtained position during the accuracy measurement

with a battery pack and starting up in reception mode waiting for a tag poll. Before installation the exact locations of the anchors were determined with a standard laser rangefinder by referencing on the zero point of the axes in Figures 5 and Figure 7. Herein the exact anchor node locations are also shown with red circles.

two deployed positioning algorithms, in blue the positioning result of the Kalman Filter and in magenta the Least Squares positioning, are displayed, together with the ground truth in red. To show the capabilities of our proposed embedded UWB localization system we calculated the positioning error along the trajectory as the euclidean distance between an estimated position and its corresponding point on the ground truth. The resulting positioning error for both estimation algorithms is plotted in Figure 6 over measurement time. The mean positioning error is around dError= 0.44 m with the Least Squares Estimator and around dError= 0.27 m with the Kalman Filter. Due to the fact that the used robot was not actively controlled to stay on the defined trajectory, he lost its optimal position during measurements, adding an approximately 20 cm error in maximum to the results.

Within this setup we performed two measurement campaigns, one aiming for an accuracy evaluation and the other aiming for performance evaluation in an use case close scenario. In both scenarios the mobile tag was connected to a tablet-PC, used for powering, controlling the measurement progress, as well as for logging and saving the measurement data. One measurement datum consists of the measured distance, the current anchor information, and the time-stamp the measurement was acquired. IV.

E XPERIMENTAL E VALUATION AND R ESULTS

In order to evaluate the performance of the proposed UWB localization system, within the use case of an autonomous vehicle parking system, we performed two different measurement campaigns with the measurement setup defined in section III. The results of those evaluations are presented and discussed in the following two subsections: 1) Accuracy Measurements: For accuracy evaluation we first defined a trajectory along the road way from the entry point of the underground garage to one of the last parking slots, through the half of the garage which was well covered with anchor nodes. We defined this trajectory, shown in Figure 5 as a thin red line, to be the ground truth of our accuracy measurement campaign. To obtain a data-set related to this ground truth we moved the mobile tag continuously with constant speed along this path, while carrying out measurements. In order to minimize the potential disturbing influence of a person on the RF signals and the ToF measurements we used a small robot driving down the predefined trajectory. The results of the

2) Dynamical Movement: In addition to the accuracy performance we wanted to evaluate our UWB localization system in an use case similar scenario including a dynamical movement of the vehicle. Therefore we mounted the mobile UWB tag together with the recording tablet PC on a bicycle and navigated by a test person several laps with varying speed in the parking garage. A lap was on the road way around the inner parking slots, one side was equipped with anchor nodes, whereas the other side was not. Due to this setting, only anchor nodes from one side could be used for estimating positions on the whole lap. The anchor node setting and the positioning results of the two deployed algorithms can be seen in Figure 7, the result trajectory obtained with the KF is again plotted in blue, the least square results in magenta. As visible from the scattered positions on the upper half of Figure 7, the estimated positions accuracy strongly vary with the anchor node distribution and quality of the communication link in terms of line of sight (LOS) condition. The positions estimated with the KF are not only obtained with a higher update rate, around 50 Hz for the KF, compared to the least squares positions estimation with around 4, 2 Hz update rate, but also

R EFERENCES

Position of Measurement Trajectory and Calculated Position 25

[1]

Anchor Positions Kalman Filter Least Squares 20

[2] y−Coordinate [m]

15

10

[3]

5

[4]

0

[5]

−5

[6] −10 −10

0

10

20

30

40

50

[7]

x−Coordinate [m]

[8]

Figure 7: Dynamical movement measurements with a test person on a bicycle tracked during 10 laps.

they are more robust against Non LOS (NLOS) conditions. V.

[9]

[10]

C ONCLUSION AND F URTHER W ORK

In this paper we have presented an embedded UWB localization system compound of a hardware platform, software framework, network architecture and protocols. In order to reveal the possibilities of standardized, integrated UWB radio chips for localization in indoor and GPS denied environments we carried out an extensive experimental study deploying two different positioning algorithms. We evaluated the systems characteristics in a scenario for vehicle navigation in an underground parking garage. Within this evaluation we determined the achievable average accuracy of positioning with a least square estimation algorithm to dError= 0.44 m and with a Kalman Filter to dError= 0.27 m. Additionally we ascertained the systems characteristics in an dynamic, use case near measurement setup in order to show its real-time behavior and robustness. Especially in disadvantageous network and signal conditions the position obtained with the Kalman Filter proved to be more precise and robust, compared to the least squares position estimation. Further experiments to determine best anchor node densities are planned. In addition we would like to improve positioning by sensor fusion and deploying additional sensors (inertial, magnetic, pressure). Moreover, we plan to analyze best network architectures and the feasibility of an cooperative localization approach, where not only fixed positioned anchors but also other mobile network participants are used as reference nodes. ACKNOWLEDGMENT This work was supported by the German Federal Ministry for Economic Affairs and Energy within the project SchiV3.0 (03SX352B) upon decision of the German Bundestag. We would like to thank the BMWi for supporting our work.

[11]

[12]

[13]

[14]

[15]

[16]

[17]

R. K¨ummerle, D. H¨ahnel, D. Dolgov, S. Thrun, and W. Burgard, “Autonomous driving in a multi-level parking structure,” in Robotics and Automation, 2009. ICRA’09. IEEE International Conference on, 2009. D. Herrero-P´erez, J. J. Alcaraz-Jim´enez, and H. Mart´ınez-Barber´a, “An accurate and robust flexible guidance system for indoor industrial environments,” Int. J. Adv. Robot. Syst, vol. 10, 2013. J. Levinson, M. Montemerlo, and S. Thrun, “Map-based precision vehicle localization in urban environments.” in Robotics: Science and Systems, vol. 4, 2007. E. DiGiampaolo and F. Martinelli, “A passive uhf-rfid system for the localization of an indoor autonomous vehicle,” Industrial Electronics, IEEE Transactions on, vol. 59, no. 10, pp. 3961–3970, 2012. F. Hartmann, T. Gaedeke, P. Leibold, L. Niestoruk, and W. Stork, “Navigation for occupational safety in harsh industrial environments,” ITG-Fachbericht-Smart SysTech 2014, 2014. S. Pittet, V. Renaudin, B. Merminod, and M. Kasser, “Uwb and mems based indoor navigation,” Journal of Navigation, vol. 61, 2008. G. R. Opshaug and P. Enge, “Gps and uwb for indoor navigation,” Department of Aeronautics and Astronautics, Stanford University, 2001. J. R. Fernandes and D. Wentzloff, “Recent advances in ir-uwb transceivers: An overview,” in Circuits and Systems (ISCAS), Proceedings of 2010 IEEE International Symposium on, 2010, pp. 3284–3287. J. Gonzalez, J. Blanco, C. Galindo, A. Ortiz-de Galisteo, J. Fern´andezMadrigal, F. Moreno, and J. L. Martinez, “Combination of uwb and gps for indoor-outdoor vehicle localization,” in Intelligent Signal Processing, 2007. WISP 2007. IEEE International Symposium on, 2007. J.-A. Fernandez-Madrigal, E. Cruz-Martin, J. Gonzalez, C. Galindo, and J.-L. Blanco, “Application of uwb and gps technologies for vehicle localization in combined indoor-outdoor environments,” in Signal Processing and Its Applications, 2007. ISSPA 2007. 9th International Symposium on, 2007. S. W. Yoon, S.-B. Park, and J. S. Kim, “Kalman filter sensor fusion for mecanum wheeled automated guided vehicle localization,” Journal of Sensors, 2015. O. Hassan, I. Adly, and K. Shehata, “Vehicle localization system based on ir-uwb for v2i applications,” in Computer Engineering & Systems (ICCES), 2013 8th International Conference on. IEEE, 2013, pp. 133– 137. F. Shen, J. W. Cheong, and A. G. Dempster, “An ultra-wide bandwidthbased range/gps tight integration approach for relative positioning in vehicular ad hoc networks,” Measurement Science and Technology, vol. 26, no. 4, p. 045003, 2015. S. Krishnan, P. Sharma, Z. Guoping, and O. H. Woon, “A uwb based localization system for indoor robot navigation,” in Ultra-Wideband, 2007. ICUWB 2007. IEEE International Conference on, 2007. M. Immaculate and V. Prithiviraj, “Test measurements of improved uwb localization technique for precision automobile parking [c],” in Proceedings of International Conference on Recent Advances in Microwave Theory and Applications, 2008. G. I. Mary and V. Prithiviraj, “Uwb localization techniques for precision automobile parking system,” in Electromagnetic Interference & Compatibility, 2008. INCEMIC 2008. 10th International Conference on, 2008. B. Silva, Z. Pang, J. Akerberg, J. Neander, and G. Hancke, “Experimental study of uwb-based high precision localization for industrial applications,” in Ultra-WideBand (ICUWB), 2014 IEEE International Conference on, 2014.