Towards zero-configuration for Wi-Fi Indoor Positioning System

Transcription

1 Towards zero-configuration for Wi-Fi Indoor Positioning System David JACQ, Pascal CHATONNAY, Christelle BLOCH, Philippe CANALDA, François SPIES FEMTO-ST Institute Univ. Bourgogne Franche-Comte, CNRS 1 Cours Leprince-Ringuet, Montbéliard, FRANCE pascal.chatonnay@univ-fcomte.fr christelle.bloch@univ-fcomte.fr philippe.canalda@univ-fcomte.fr francois.spies@univ-fcomte.fr French Army Logistic Academy, Bourges, FRANCE tristaury@gmail.com Abstract This paper describes the state of the art of indoor positioning. It describes the Wi-Fi auto-calibration phase of a fast and adaptive algorithm to calculate the position of access-points in a heterogeneous environment such as a building, house or flat. Our proposal can track a mobile terminal in a building with a minimum of preparation. This deployment simplicity relies on the autocalibration phase. Of course, this service must also provide good positioning accuracy. Therefore, the auto-calibration function described in this paper has two advantages: it serves to initialize and to model the environment using a set of environment parameters that the operator, using the web interface, can confirm or change. II. WIRELESS INDOOR POSITIONING SYSTEMS Many kinds of wireless indoor positioning systems have been developed and studied (Fig. 1). Keywords Positioning Algorithm, Wi-Fi Signal Strength, Autocalibration, Network-centric architecture I. INTRODUCTION Nowadays, many services and tools need to get accurate indoor positioning and indoor navigation capabilities. GPS, GLONASS and Galileo, the three GNSS (global navigation satellite systems) standards for outdoor localization do not work well in case of indoor request due to wall signal consumption and the level of precision from five to ten meters. There have been many researches or commercial efforts to develop tools to locate devices in indoor environments; it has become a key issue for many newly arriving location-based applications (LBA) and location-based services (LBS) in different fields. Different technical solutions are available; each has advantages and drawbacks. The vision-based localization needs cameras and computers, which increase the cost and the deployment time, limiting each device to a room. The preparation time consumption and the need to access the building are not compatible with military purposes and rescue operations. Accelerometer-based localization will accumulate the error made by each localization prediction. This kind of tools is not sufficient to give an accurate position but could be mixed with another solution to enhance the precision for long-term positioning. Wireless-based indoor localization gives accuracy, while remaining inexpensive. Wireless waves can pass through doors and walls and provide the full coverage of a building. Fig. 1. Hierarchy of Positioning Systems 1. Performance Criteria Wireless positioning systems must be compared using appropriate criteria. We mainly focus on six criteria among those defined by [1;2]: accuracy, responsiveness, coverage, adaptiveness, scalability, cost and complexity. Accuracy (or location error) estimates the distance error between the estimated and the actual mobile locations. We use it as the main criterion to assess the performance of the positioning system. Responsiveness is also important since we want to propose an algorithm which fast updates the estimated location of moving targets. Scalability and scalable coverage are also among the main challenges. Indeed, one of the goal is to easily make the positioning system operate with a larger number of location requests and a larger coverage, without unnecessary strain. This is based on ad hoc networking, the ability to use any kind of Wi-Fi-embedded mobile terminal, and the addition of inexpensive units. But this implies that we pay a particular attention to the actual coverage

2 which can be reached in the whole action area. We also focus on adaptiveness, in order to make the system cope with the environmental changes, and to prevent the need for repeated calibration each time it is possible. Adaptiveness and responsiveness are then used in a complementary way. Finally tradeoffs must be found between these performance requirements and the cost and the complexity of the positioning system. Regarding these objectives, we make an effort to limit the cost of the extra infrastructure, particularly by the nature of the deployed technology, and the choice of inexpensive units (Raspberry Pi model B+ units, equipped with a Wi-Fi USB device). Coupled with the self-calibration algorithm, this permits to reduce the cost induced by the installation and the survey time during the deployment period. The system complexity is mainly related to the algorithms used for self-calibration and to estimate the location. Thus, as described below, these criteria define the goals we want to pursue, and that led us to the Wi-Fi solution proposed in this paper. First, it is a multi-positioning system, which permits to reach better levels of accuracy. The infrastructure-centered architecture we propose is a decentralized system based on Raspberry Pi model B+ units, equipped with a Wi-Fi USB device. It is fault tolerant, scalable, and it permits to use load-balancing positioning algorithms. Besides, a self-calibration phase permits to make the system deployment easier and cheaper. Self-calibration can be repeated in order to improve responsiveness and adaptiveness, while simultaneously enhancing the map accuracy. Therefore the conjunction of these design choices was studied to provide good performances and scalability, while keeping quite limited costs and complexity. The following sections briefly reminds some elements required to well understand the description of the proposed architecture and auto-calibration algorithms: some of the main wireless technologies, mathematical localization techniques, and possible architectures that can be used for wireless-based indoor localization. ZigBee can be activated on 900MHz and 2.4GHz. Because ZigBee is quite similar to Bluetooth, the output power transmission and small antennas allow only small distances. The most famous Ultra-Wide Band standard is a. It utilizes mainly frequencies under 1GHz, between 3GHz and 5GHz and between 6GHz and 10GHz. The distances inside building are around 40m. The UWB functionality provides immunity to multipath interference while delivering high data rates and low power consumption. 3. Mathematical techniques Even though there are several wireless technologies used for indoor localization, the mathematical techniques that the localization is based on are limited. We categorized them into four groups: proximity, triangulation, trilateration and fingerprint. Proximity: Proximity is the simplest method for localization. The assumption this method is based on is if that the user s point is in the range of a known station, then we can approximate the location of the user point to the known station. For link-based or connection-based wireless communication, the location of the user can be approximated to the position of access point. This solution is usable inside a building for short-range-wireless technologies, but the expected accuracy is quite bad. Triangulation: Triangulation uses geometric knowledge to obtain the location of the user. The location of the user is determined by the received signal angle. If the angle of base stations to the user s point or the angle of the user s point to base stations is known (Fig. 2), we can easily obtain the location of the user s point by the intersection of three vectors. 2. Wireless Technologies Different classes of wireless technologies are used for indoor localization. [1;2] classify them in three categories according to the distance of transmission. The contribution we present is more particularly related to middle distance technologies (mainly Wi-Fi and ZigBee) and short distance technologies (in particular Bluetooth and Ultra-Wide Band or UWB). Wi-Fi is one of the most famous wireless technologies. Twolicense-free frequencies are currently used which are 2.4GHz and 5GHz on many devices such as laptops, smartphones or tablets. The maximum legal output power transmission allows to cover small building including 2 or 3 floors with a single access-point. Bluetooth is normalized into the Wireless Personal Area Network. The legal output power transmission is quite low and allows only around 10m. In order to cover small or medium building you need to deploy nearly two or more static devices by room. Fig. 2. Vector (angle) triangulation

3 Auto-calibration mechanism begins to appear in positioning systems based on Wi-Fi signal. The first mandatory information to take into account to manage auto-calibration is the positioning error measurement as it is presented in [16]. Knowing the error, some authors try to modify the propagation model to manage differences in the reception characteristics among devices [17]. Some authors have extended the propagation model with additional parameters for better propagation simulation. Their self-calibrating uses one propagation model to infer parameters of the area and the other to simulate the propagation of the signal [18]. According to the literature, no calibration mechanism is sophisticated enough to handle important changes in the location of the AP or in the reception condition like building modification. 4. Architectures Fig. 3. Trilateration and distances Trilateration: The location of the user is determined by the distance to the known fixed measurement points. Suppose we have base A, B and C, three fixed wireless beacon stations in known positions. If the distances of the user point to all three base stations is known (Fig. 3), the location of the user s point can be expressed as the intersection of three circles or spheres. The only problem left is to find out how to get the distance from the user s point to base points. Time-based trilateration is one of the methods that uses distance for calculation. The assumption under time-based trilateration is that the time used from a beacon to the user s point can be used to infer the distance between the two points. As the travel speed of the wireless signal is known, this approximately equals the speed of light in air. For time-based trilateration, there are mainly two types of methods: ToA (Time of Arrival) and TDoA (Time Difference of Arrival). Besides time, the property of the received signal is also an important way to infer the distance. In most literature, RSS is used to represent received signal property. The propagation power loss model has characterized the fading of signal strength along the distance. In reverse, the distance is deduced from the received signal strength. When triangulation or trilateration are considered, the stations and user need to be in Line-of-Sight (LoS). Otherwise, the angle or distance referred from time or RSS cannot be used to locate the user. However in a real world scenario, there might be walls and doors, rooms, and hallways in a building. Even if it is in a plaza, there might be furniture, statues, fountains or people walking that block the Line-of-Sight. Fingerprint: Fingerprint means the characteristic or feature of signals. In most literature, RSS is used with a fingerprint. The assumption underneath this fingerprint-based indoor localization is that for each position in an area there will be specific recordable features. The current location can be obtained by relying on the difference of signals in different positions. For fingerprint-based indoor localization, there are two different methods: radio-mapbased fingerprint localization, and map-free fingerprint localization [15]. With Wi-Fi, two different architecture solutions are available to build the system : mobile-centered implementation and infrastructure-centered approach. The mobile-centered implementation is based on an existing Wi-Fi infrastructure. It is simpler to implement, as the code is only in the terminal and relies on the radio network. The software depends on the terminal operating system (OS). Many researches are conducted to implement LBS and LBA on mobile devices like smartphones. Our infrastructure-centered approach allows an everywhere use and does not depend on the OS on the device. In that case, we planned to deploy the Wi-Fi infrastructure containing the code (Fig. 4). This choice is linked to the rescue and military ways to implement this system and the future ability to detect some movement without a beacon. Since 2008, our lab has been leading researches on the infrastructure-centric architecture to develop its indoor positioning and indoor navigation software. This solution does not rely on any application on the mobile device; therefore, it is able to detect any kind of Wi-Fi-embedded mobile terminal. Furthermore, the detection of human movements by monitoring prompt decreases in the Wi-Fi signal when someone is crossing the transmission line between two APs is possible. Fig. 4. First Architecture of our Open wireless Positioning System - OwlPS

4 The Simultaneous and Hierarchical Multi-Positioning System (SHMPS) is a full pack tool containing a multi-positioning system composed of GPS, OwlPS (Virtual Wi-Fi-based fingerprinting system and trilateration system) and a marker analysis system. This SHMPS competed many times in Evaluating Ambient Assisted Living solution systems through competitive benchmarking (EvAAL) [14]. The switch from one system to another enhances the accuracy from eight meters outside to some decimeters inside. Many location systems are not really cheap and simple to implement. In a self-positioning system [3], the positioning receiver takes the appropriate signal measurements from geographically distributed transmitters and uses these measurements to determine its position [4]. This system is not beacon-centered but centered on the infrastructure [8], which has proved to be a good way to detect any kind of mobile devices. The first tuning step is to build the map of the whole area including the pinpointed access-points and the ones that are not. In order to allow a quick deployment, the system is able to calculate the relative positions of access-points according to the algorithm and to load the sketch onto the web interface. There are three main categories for positioning algorithms [1]: triangulation/trilateration, scene analysis (fingerprinting) and proximity. We present a system including an auto-calibration algorithm. It calculates the positions of unknown access-points using trilateration and proximity methods on many phases. Our system utilizes the access points as positioning servers with highavailability and load-balancing configuration. This self-calibration phase can be loaded many times in order to enhance the map accuracy and to take into account the hypothetic changes in the field locations (moved access-points, landslide, environment modifications, people moving, etc.). In [13], the authors observed that the effect of environmental changes, including both time-varying effects and different hardware, can be precisely approximated by a linear relationship. 1. Architecture of OwlPS III. CONTRIBUTION Those criteria are the reasons that led us to the Wi-Fi solution, the infrastructure-centered one, the trilateration calculation based on fingerprint, as it provided the solution obtaining the best results in accordance with the goals we wanted to pursue. The Open WireLess Positioning System (OwlPS) is integrated into the network [5] (Fig. 4). The access-points report the RSSI to the aggregation server. Therefore, the mobile terminal does not need to embed any kind of specific code. The measurements are taken by sending requests while prompting the Wi-Fi device to answer. The Open WireLess Positioning System uses the IEEE and (Wi-Fi and Ethernet) connection and protocol. The infrastructure-centered architecture that we upgraded with the code presents many advantages: fault tolerance, decentralized system, load-balancing positioning algorithm and scalable architecture [6]. In this version, we use a Raspberry Pi model B+ unit equipped with a Wi-Fi USB device, D-Link DWA-127 including a Ralink RT3070 chipset. The B+ operates under a Raspbian Linux system. There is enough CPU power on a Raspberry unit to be used as a Wi-Fi Access-Point in order to take signal strength measurements and to calculate the positions as a server (Fig. 5). The readings are done by using the packet capture (pcap) library and RTAP header. The aim is to deploy the network as fast as possible depending on the purpose requiring indoor-positioning. We can connect accesspoints with a wired Ethernet or wirelessly through an ad hoc network including an adapted routing protocol (OLSR) [7]. The ad hoc mode allows the rapid deployment of access points which could last for hours with USB batteries or for unlimited duration when connected to a continuous power source. This mode matches with an operational use. The wired deployment corresponds to a more permanent installation with a reduced use of the radio channel to limit the communication overhead and to use it for red force tracking (not allowed user), therefore avoiding any detection. Wi-Fi noise is everywhere but the ad hoc protocol is easy to detect due to its rare use. Both modes could be mixed in a hybrid fashion. In this case, the combination of OLSR, for routing inside an ad hoc part and Open Shortest Path First (OSPF), for routing between ad hoc parts of the network, achieves the global interconnection [12]. The whole system does not rely on the only AP acting as a server. Each AP boards the same code and is able to replace the server in case of a breakdown, power-lack, enemy destruction. In addition, the calculation can be performed redundantly on two positioning servers in order to get the most accurate position or to distribute the load avoiding overloading. A dedicated solution has been used so as to share a single IP floating address (like VRRP (Virtual Router Redundancy Protocol) or an anycast in IPv6) among the access points in the network so the mobile terminals will communicate with a single IP address regardless of the servers which are performing the calculation. The graphical user interface gives the ability to configure the system and allows the visualizing of the results on the map using mashup tools and APIs of Google Maps (Fig. 6). This GUI is very simple to use and participates in the system preparation and tuning speed. Some tools allow to choose the server, to move the APs, to start the calibration phase, to store scenarios, to emulate stored scenarios to replay hypothesis and compare algorithms in reproducible scenarios. 2. Auto-calibration algorithm The auto-calibration algorithm aims at providing an indoor positioning service by performing the initial positioning calculation as fast as possible with the minimum possible set of input. The idea is to have seven phases of functions. The first phase is to set the reference system. The positions of the other access points are relative to this reference system, then calculated and confirmed during the other steps. Our system needs two different kinds of data to be ready, the coordinates of all the access-points and the received signal strength between APs and a mobile.

5 2017 International Conference on Indoor Positioning and Indoor Navigation (IPIN), September , Sapporo, Japan AP discovery: The second phase is to identify all the wireless MAC addresses of OwlPS beacons which are Rapsberry Pi. At this stage, all the access points will communicate with each other in order to determine the signal strength indicator (RSSI) between them and to calculate the average RSSI (Fig. 8). Fig. 8. MAC discovery Reference distance calculation: The third step aims to calculate the distances between the landmarks using the Vincenty Formula (Fig. 9). d (m) = arccos(sin(lat1)*sin(lat2) + cos(lat1)*cos(lat2)*cos(lon2-lon1))* Fig. 5. Architecture of OwlPS with Raspberry Pi devices (1) where (lat1,lon1) and (lat2,lon2) are the latitude and longitude of 2 geographical points. Fig. 9. Reference distance calculation Fig. 6. Graphical User Interface of OwlPS We developed two different algorithms; the first one takes three non-aligned points as a reference system. The second one does not need to perform the complete off-line manual configuration and gives us the ability to deploy more rapidly with less than three known points. Friis calculation: This fourth step determines the Friis environment parameter (e), which consists in modifying the square in the Friis formula, between each pair of reference APs according to the distances and the average RSSI [10] (Fig. 10). This step allows us to take into account the characteristics of the indoor environment. e = log( λ / (4π. 10 (Pr - Pt - Gt Gr)/20)) / log(d) (2) where Pt and Pr are the transmit and received power, Gt and Gr are the transmit and received antenna gain, λ is the wavelength and d is the distance between the transmit and received points. A. The partially configured Algorithm (PCA) Reference system: The system has to build a reference system with at least three different landmarks for 2D location and four for 3D [10]. For this reason, the operator has to initialize the grid vector, this is achieved by writing the known positions of each MAC address in the configuration file (Fig. 7). Fig. 10. Friis calculation Fig. 7. Reference system Friis distance calculation: The second Friis formula allows us to determine the distance between an undetermined accesspoint and a reference one. This calculation embeds the

6 average RSSI and the Friis environment parameter (e) of the nearest reference APs[11] (Fig. 11). d = ( λ / (4π. 10 (Pr - Pt - Gt Gr)/20))1/e (3) Fig. 11. Distance calculation Coordinate determination: The goal of this process is to build a mapping of the access points above the map of the experiment environment. The challenge is to deduce a geographical arrangement of the access points based on each estimated distance between them. It is divided into several steps to obtain a good and realistic positioning of the access points and a right representation of the reality in a Friis-based point of view (Fig. 12). Fig. 14. Circle junctions Distance checking: This stage focuses on calculating the distances between each pair of undefined APs trying to improve the reliability of the system. A check is carried out to make sure that all distances between an AP trio are reliable (AC AB + BC). Then we compare the circular way to calculate errors to the bisection one in order to fill the measurement with the lowest error. Fig. 12. Grid calculation There are two possible ways to achieve this step. On the one hand, we can use weighted bisections and the triangle defined by the bisection junctions (Fig. 13). On the other hand, we use a matrix based on the junctions of the circles drawn with the points and the distances (Fig. 14). In any case, the error is the radius of the circle defined by the triangle perimeter. Fig. 13. Bisection junctions Fig. 15. Distances checking After the auto-calibration phase, the drawn geometry is available for the interface in the shared memory in order to enable the operator to transform it according to his knowledge of the area. These transformations are translation, rotation around a point, axial symmetry, dilation and scaling. The operator can define the position of an access point (x, y, z coordinates) or define the exact distance between two access points and lock it. Each time the operator modifies the location of an AP, it will launch the auto-calibration process, considering the new position as a reference one. Given the small distance scale and the knowledge of this deployment, we will be able to reach a relative accuracy during this rapid loop procedure. B. Towards zero configuration Algorithm (TZCA) In case there is not enough time to perform the off-line configuration or satellites are no longer available from outdoor AP, we can tune the system according to the poor accurate knowledge of the real locations. No reference point: The system has no information about the right situation. A way to build a reference system with at least three different landmarks for 2D location is to read three different MAC addresses in the captured information. The first point is spotted as the origin with WGS-84 coordinates of the middle of the GUI map (or without lat=0 and lon=0), the second one is the abscissa reference (100; 0), the third one is

7 the coordinates (0; 100). We chose one hundred meters to match with the normal used scale between the building size and the Wi-Fi range. One reference point: The configuration with one reference point is quite similar to the one without reference. The main difference is given by the grids allocated to the origin point that are the real ones. After that, the abscissa and ordinate are determined in the same way. Two reference points: The mechanism begins to be reliable for the first two points. In order to build the third one, the system initiates the point with relative grids (0; 100) and checks if the points are not contained in the same line otherwise it will change the coordinates to new ones (100; 0). Fig. 16. Auto Calibration Process This way, the operator has an initial configuration on his interface. Thanks to his field expertise, he can move the accesspoints to their current position. Each time the operator moves something, it will restart the auto-calibration phase to depict a matching image of the field arrangement. The operator can do some geometric changes, translation, scale changing, rotation around a point, axial symmetry, homothety. Even if the grids are not 100% accurate, the important thing is the match between the GUI view and the beacon positions all around the building. There is often a light shift between the grids and the accuracy of the planar representation. In the end, you will have less accuracy but you will need less effort to deploy your network. 3. Experimentation The aim is to measure the loss of accuracy when we remove step-by-step the input data such as the coordinates of the APs deployed in the indoor area. Thanks to the real coordinates of Aps and the transmission, it is easy to get both the distances between each couple of Aps and the collected RSSI from each pair of Aps. From these two parameters, we can calculate the environment parameter by using the proposed propagation model. The environment parameter delivers information about the hardness and roughness of the indoor building environment. A high value (upper than 2) means a lot of thick walls and a low value (between 1 and 2) means nearly line of sight environment. We captured all the wireless traffic in the area during 2 minutes and we stored all the corresponding RSSI on the OwlPS server. We made an analysis of 2 scenarios. The first one corresponds to a big room (11m x 9m) with 4 APs inside. It is used to measure the loss of accuracy when we remove progressively more and more input data. The second one corresponds to a realistic deployment inside a building with 10 APs located regularly in a working building. In the first experiment, when all the 4 APs are located on known locations, the accuracy of the mobile terminal inside the room is less than 1 meter (10% of accuracy) with the PCA algorithm. The environment parameters are chosen according to the measurements between each 4 APs. Then we decrease voluntarily the configuration file with 3 APs located at known coordinates and 1 AP without any location knowledge. In this case the TZCA algorithm calculates the estimated location of the forth AP in order to use it if necessary in the calculated position of the mobile terminal. The accuracy of the mobile terminal coordinates inside the room is less than 1.5 meter, using 2 APs with known location and 1 AP with an estimated location as the repository. Finally, with a configuration file including 2 APs with known coordinates and 2 APs without known coordinates, the TZCA algorithm is run to estimate the coordinates of the 2 APs without information. In this case, the accuracy is less than 2.5 meters. Of course, this lack of accuracy is not enough in normal conditions. But this new TZCA algorithm, which continues to deliver some coordinates even in disadvantageous conditions with very little input information, can help users when there is no time to spend in initial configuration, because of an emergency situation. The second experiment was a standard deployment of an indoor location service. The size of the experiment was around 2000 square-meters. There was a great heterogeneity of the building with mainly NLoS (No Line of Sight) connections. The mobile terminal moved from the ground floor to the second floor. We first used the PCA algorithm with a fully configured input file including the coordinates of all the 10 APs. The measured accuracy of the mobile terminal locations in the building was less than 3 meters. Given the quite hard complexity of the experiment scenario described above, added to the greater distances between the mobile terminal and the Aps, the results are not so bad. Finally, the second step used a configuration file including 3 APs with known coordinates and 7 APs without any information. The NZCA algorithm estimated the coordinates of all the 7 APs and the accuracy of the mobile terminal coordinates was less than 6 meters.

8 IV. CONCLUSION The Wi-Fi infrastructure-centered system we developed is inexpensive to operate as it uses raspberry units. GPS, accelerometer, optical telemeter parts could be added to upgrade the system. The general user s interface based on Google secures a very simple and fast implementation. Our system can be deployed very quickly and, afterwards, it is able to detect any kind of mobile terminal. Therefore, the selfcalibration capabilities solve two majors issues to sketch the map : the drift which affects the accuracy in fingerprint-based algorithms which might depend on a non-updated map of fingerprints and on the other hand, the security or emergency for military and rescue operations. After the initialization, the self-calibration tools recalculate the environment parameters and then regularly take into account any changes in the Wi-Fi signal propagation environment. In the future, we would like to improve our man-detection using the shift of environment parameters. Another way to improve the code would be to input a certificate in known devices in order to discriminate between blue forces and red forces (friend vs enemy). ACKNOWLEDGMENTS The authors would like to acknowledge the 10 th French airborne commando unit for its contributions to the experimentation phase. REFERENCES [1] H. Liu, H. Darabi, P. Banerjee and J. Liu, "Survey of Wireless Indoor Positioning Techniques and Systems," in IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applications and Reviews), vol. 37, no. 6, pp , Nov doi: /TSMCC [2] Zahid Farid, Rosdiadee Nordin, and Mahamod Ismail. Recent Advances in Wireless Indoor Localization Techniques and System. Journal of Computer Networks and Communications, vol. 2013, Article ID , 12 pages, Hindawi, 2013.doi: /2013/ [3] Drane, Christopher, Malcolm Macnaughtan, and Craig Scott. "Positioning GSM telephones." Communications Magazine, IEEE 36, no. 4 (1998): [4] M. Cypriani, F. Lassabe, P. Canalda, and F. Spies, Open Wireless Positioning System: a Wi-Fi-Based Indoor Positioning System, in VTC-fall 2009, 70th IEEE Vehicular Technology Conference, Anchorage, Alaska, United States, September [5] M. Cypriani, P. Canalda, and F. Spies, OwlPS: a Self-calibrated Fingerprint- Based Wi-Fi Positioning System, in Proceedings of the 1st EvAAL competition and Workshop (EvAAL 2011), April [6] K. Chen and K. R. Vadde, Design and evaluation of an indoor positioning system framework U.C. Berkeley course project for CS262A, [7] Jacquet, Philippe, Yacine Mezali, and Georgios Rodolakis. "Indoor Positioning using the IEEE Infrastructure." Working paper INRIA (2010). [8] Yang, Zheng, Chenshu Wu, and Yunhao Liu. "Locating in fingerprint space: wireless indoor localization with little human intervention." Proceedings of the 18th annual international conference on Mobile computing and networking. ACM, 2012 [9] M. Cypriani, F. Lassabe, Ph. Canalda, and F. Spies. Open Wireless Positioning System: a Wi-Fi-Based Indoor Positioning System. In VTC-fall 2009, 70th IEEE Vehicular Technologie Conference, Anchorage, Alaska, United States, pages 1-5, September IEEE Computer Society Press [10] F. Lassabe, O. Baala, P. Canalda, P. Chatonnay, and F. Spies, A Friis- based calibrated model for WiFi terminals positioning in Proceedings of IEEE Int. Symp. on a World of Wireless, Mobile and Multimedia Networks, Taormina, Italy, Jun. 2005, pp [11] Zengshan Tian, Xiaomou Tang, Mu Zhou, Zuohong Tan: Fingerprint indoor positioning algorithm based on affinity propagation clustering. EURASIP J. Wireless Comm. and Networking 2013: 272 (2013) [12] O. Abu Oun; W. Abdou; C. Bloch; F. Spies, Broadcasting Information in Variably Dense Environment Using Connectionless Data Exchange (CoLDE), WWIC 2014, LNCS 8458, pp , Springer International Publishing France, Paris 2014 [13] A. Haeberlen, E. Flannery, A. M. Ladd, A. Rudys, D. S. Wallach, and L. E. Kavraki. "Practical robust localization over large-scale wireless networks." In Proceedings of the 10th annual international conference on Mobile computing and networking, pp ACM, [14] M. Cypriani, Ph. Canalda, and F. Spies. "OwlPS: a self-calibrated fingerprint-based Wi-Fi positioning system." In Evaluating AAL Systems Through Competitive Benchmarking. Indoor Localization and Tracking, pp Springer Berlin Heidelberg, [15] P. Mirowski, Ph. Whiting, H. Steck, R. Palaniappan, M. MacDonald, D. Hartmann and T. K. Ho. Probability kernel regression for WiFi localisation, Journal of Location Based Services Vol. 06, No. 02, June 2012, [16] Tuta, Jure and Juric, Matjaz B : A Self-Adaptive Model-Based Wi-Fi Indoor Localization Method, Sensors, Vol16, Num 12, pp , [17] Anagnostopoulos, Grigorios G and Deriaz, Michel and Konstantas, Dimitri : Online self-calibration of the propagation model for indoor positioning ranging methods ; Int. Conf. on Indoor Positioning and Indoor Navigation (IPIN), 2016 [18] Pulkkinen, Teemu and Verwijnen, Johannes : Evaluating indoor positioning errors ; IEEE Int. Conf. on Information and Communication Technology Convergence (ICTC), 2015.