Dynamic Wi-Fi RSSI normalization in unmapped locations Hanna Kavalionak Massimo Tosato National Research Council of Italy Cloud4Wi Pisa, Italy Pisa, Italy hanna.kavalionak@isti.cnr.it mtosato@cloud4wi.com Paolo Barsocchi Franco Maria Nardini National Research Council of Italy National Research Council of Italy Pisa, Italy Pisa, Italy paolo.barsocchi@isti.cnr.it francomaria.nardini@isti.cnr.it ABSTRACT of particular interest when the high localization precision is not With the growing availability of open access WLAN networks, required. we assisted to the increase of marketing services that are based There are a lot of commercial companies that are working on the data collected from the WLAN access points. The identi- with the data of commercial venues in order to provide market- fication of visitors of a commercial venue using WLAN data is ing products to the clients. These companies are buying probes one of the issues to create successful marketing products. One of data from giant wireless infrastructure providers, such as Meraki, the ways to separate visitors is to analyse the RSSI of the mobile Ruckus and Aerohive, and process this data into the marketing devices signals coming to various access points at the venue. Nev- product. The marketing product motivating the paper originates ertheless, the indoor signal distortion makes RSSI based methods from a company mission to help retailers in brick and mortars unreliable. stores to build an omni-channel communication with their cus- In this work we propose the algorithm for the WLAN based tomers. For example, one use case could be to trigger a specific RSSI normalization in uncontrolled environments. Our approach marketing action on a defined behaviour such as a push up noti- is based on the two steps, where at first based on the collected fication on the proprietary app when the customers enters the data we detect the devices whose RSSI can be taken as a basic one. store or to send a discount or promotion email to a customer At the second step the algorithm allows based on the previously who has just left the location for retaining. As many real world detected basic RSSI to normalize the received signal from mobile scenarios, the raw data contains much ’noise’ and is in need of devices. We provide the analysis of a real dataset of WLAN probes semantic attribution to be able to aim the right communication collected in several real commercial venues in Italy. to the right target: the visitors. As a matter of fact, one of the key aspect for this data pro- cessing is actually to separate the visitors of the venue from the 1 INTRODUCTION passers-by, where the visitor is usually considered a person that enters the venue and spends some time (and hopefully money) Localization is becoming a more and more important feature inside. One of the ways to separate visitors from passers-by is by in the mechanisms used for the location-based services and dif- analysing the Received Signal Strength Indicator (RSSI) of mobile ferent location-based business models [13, 14]. One of the most WiFi devices carried by nearby persons. The RSSI indicates the commonly used and precise mechanism for the localization, the power present in the received radio signal from remote mobile Global Navigation Satellite Systems (GNSS), has strong limita- devices and typically decreases with the distance between the tions in indoor environments. Hence, in order to organise the receiver and the transmitter of the signal. The RSSI parameter is localization in indoor environments, researchers and industry very volatile and highly depends on the environment. The signal approached the problem by using different signal technologies, in a propagation channel is affected by path loss and multi-path such as Radio Frequency Identification tags, Ultra Wide-Band or effects, which result in RSSI variation and attenuation [3, 11, 12]. WLAN. This is an important factor especially for commercial venues, While the technologies based on the Radio Frequency Tags in which the high fluctuation of people makes the problem of and Bluetooth Low Energy show promising results in indoor correctly distinguish between visitors and passers-by based on localization [2, 7], it is WLAN that is of our particular inter- RSSI even more challenging. est. Indoor localization based on WLAN technologies does not Most of the existing WLAN solutions are based on the knowl- show significant precision benefits in comparison with other edge of the access points (APs) position, network devices that approaches. However, in the last decades the availability and allow other Wi-Fi devices to connect to the wired network [4, 6]. distribution of free internet zones in the cities and commercial Also, the typical solutions rely on the fixed indoor layout and do venues have significantly increased. Therefore, the application not consider the people fluctuations [3]. of WLAN based localization does not require additional infras- In this paper, we focus on the characterization and possible tructure deployment and can be based on the already existing adjustments of RSSI based indoor localization for WLAN com- infrastructure. This characteristic makes this method to be one munication. More precisely, our contribution can be summarized of the most appealing for small and medium size enterprises, also as follows. We propose an algorithm that dynamically adjusts considering its flexibility and applicability. The approach can be the RSSI of the received probes from mobile devices. During the adjustment phase the RSSI of probes of selected anchor devices Copyright © 2020 for this paper by its author(s). Published in the Workshop Proceed- are used as a reference to evaluate the path loss in the propaga- ings of the EDBT/ICDT 2020 Joint Conference (March 30-April 2, 2020, Copenhagen, Denmark) on CEUR-WS.org. Use permitted under Creative Commons License At- tion channel. In general, an anchor device is (i)active during the tribution 4.0 International (CC BY 4.0). closing hours of the venue, (ii)does not change its location, and (iii)it is also present during the opening hours of the same day. the indoor environment and some controlled off-line measure- We also describe the algorithm for the anchor devices selection. ments using the existing APs. By comparison, in the scenario In our work we provide the analyses of a real dataset of WLAN described in our work we have no knowledge about the area and probes collected in several real commercial venues in Italy. This no control over the APs. dataset is provided by the company Cloud4Wi for research pur- One more aspect described in the literature is the possibility of poses. We provide the evaluation results for the anchors distri- attacks in case of fingerprinting approach. For example the work bution and presences for different types of commercial venues. of Richter et al.[10] discusses different attack types and compare We also evaluate the correlation of RSSI of anchor devices on the positioning performance of RSS-based fingerprinting under these number of incoming probes to an AP. attacks via simulations. The authors also present the simulator The benefit of described approach is that the algorithm relies for realistic RSS predictions for the simulation environments. only on the device probes received by AP and do not require Aboelnaga et al.[1] describe an algorithm to identify attacked APs deployment of additional hardware. Moreover, the adjustment and make accurate localization in the presence of attacks. These phase can be executed at runtime or offline. This important char- works are concentrated on the evaluation of different attacks acteristics are essential to follow the wireless channel variance types on the performance of RSS. Moreover, in order to detect during the day due for instance to the high fluctuation of people. and evaluate the attacks the authors rely on previously collected While the described solution does not aim for a high precision fingerprints datasets. localization, it can significantly improve the quality of marketing Our work differs from most of the state of the art because we solutions proposed on the data processing market. do not improve the precision or reliability of the indoor localiza- The remainder of this paper is organized as follows. Section tion methods. Instead, the main scope of our work is to provide 2 offers an overview of the related work. Section 3 provides the an approach that could, with minimum monetary investments, definition of the problem statement, whereas Section 4 presents improve marketing solutions provided by companies based on the algorithm for anchors selection and dynamic RSSI adjustment. the collected data. As we describe in Section3, we deal with en- In Section 5 we evaluate the anchors selection algorithm based vironment layouts that are not known and not controlled. In on the real WLAN data. Finally, we conclude the paper in Section our work we rely only on the sniffed data of probes from APs 6 by discussing the results and the future work. in a totally passive fashion. To decrease the impact of path loss effect in propagation channel on the RSSI based localization we propose to rely on a set of dynamically detected anchors. Mon- 2 RELATED WORK itoring the changes of RSSI of these anchors and applying the Indoor localization is a mature research field. Nevertheless, the correction coefficient for the sniffed RSSI on APs allows to adjust growing distribution of wireless infrastructure and open access the RSSI-based localization. internet connections raises old and brings new challenges. Re- cent overviews on technologies and techniques can be found on existing surveys [8, 13, 14]. 3 PROBLEM STATEMENT Most of the available literature in the field relies on some The wireless radio signal passing through propagation chan- known data about indoor layout. This data can include the shape nel experiences path loss and multi-pass effects influencing on of positioning measurement, points coordinates or fingerprints the RSSI sniffed by the receiver. These effects are caused by dif- [4, 5, 8], maps of the area [2, 7, 8], some known scenario [7]. ferent factors like obstacles in the propagation channel, signal For example the work of Nikoukar et al. [7] describes the study reflection on the walls and floors, etc. In order to estimate the of low-energy Bluetooth advertisement channels. The authors distance between transmitter and receiver some applications rely conduct extensive experiments in four different environments. on the strength value of the received signal RSSI. Hence, the The work describes the study of the effect of the environment deformation of the signal passing through propagation channel noise and interference on the signal propagation conditions. Our can significantly decrease the level of the provided service. work is also connected with the studies of signal deformation in Regarding the importance of the signal deformation problem a propagation channel. Nevertheless, the authors consider the for the distance measurement, various models have been pro- controlled environment in their studies, while in our case the posed to compute the strength of the signal depending on the setup layout and the indoor conditions are fully uncontrolled. distance [9]: (i) the free-space propagation model, (ii) the two-ray Another work that studies the propagation channel is the work model and (iii) the log-normal shadowing model (LNSM). For of R. Faragher and I. Papapanagiotou [2]. The authors provide a our application, these models usually do unrealistic assumptions detailed study of Bluetooth Low Energy (BLE) fingerprinting. In or have too high requirements in terms of knowledge of the their investigations the authors rely on the deployed network of layout. For example the free-space propagation model assumes 19 beacons. While the study provides the quantitative comparison the environment to be obstacle-free. A more promising models of BLE technology with the WiFi one, it still relies on a controlled based on RSSI for the indoor localization is LNSM. However, it network of hardware devices that has to be installed. heavily relies on the knowledge about the indoor environment. The work of Shrestha et al. [11] describes an approach for in- This knowledge includes positions of APs, mapping of the area, door localization with WLAN signal and unknown access point environment parameters like temperature, humidity, etc. locations. The authors formulate the problem of WLAN position- These requirements are important for correct indoor local- ing as a deconvolution problem and investigate three deconvolu- ization and cannot be ignored in case of precise modeling. Nev- tion methods with different path loss models. The work describes ertheless, in real applications it is not always possible to have the comparison of the proposed approach with the fingerprint- full knowledge about the environment and the company-owned ing one. Nevertheless, the authors relies on two stages approach data usually is very limited. Moreover, the environmental factors where on the first training stage is required the information about change over time. In our work we consider, as an use case, a company that pro- (3) the probes of the candidate must be detected in both the vides marketing products to commercial venues. The product of closing hours of the venues and the next consecutive open- the company is based on the processing data of probes sniffed ing hours. With this requirement we eliminate those de- from APs placed in some venues. As a medium size company it vices that are, for some reasons, active only during the does not provide the services of hardware deployment and relies closing hours of the venue. For example it can be spe- on the APs installation and services provided by big market play- cific security appliance that is activated when no physical ers like Meraki, Ruckus and Aerohive. The company does not guardians are presented. have the knowledge about the position of APs. Also, the com- At the end of the first step we have a list of anchor devices pany does not have the knowledge about the indoor layout of the that can be used to adjust RSSI of the mobile devices during the venue. The indoor layout of the venue is periodically changing. opening hours. Each anchor is assigned to some APs. The corre- The venues tend to have fluctuation of visitors during the day sponding RSSI detected in the closing hours corresponds to the and visitors inside the venue can significantly influence on the basic RSSI, i.e., RSSI 0 , between the anchor A and the correspond- deformation of the signal in the propagation channel. ing AP. Hence, for the following opening hours of the venue this At the same time, in order to provide services linked to its mis- RSSI 0 can be assigned as the basic RSSI for the adjustments for sion, the company has to distinguish the probes of visitors of the the propagation channel between anchor A and the assigned AP. venue from the probes of passers-by. This selection has to work Based on the detected anchors for the RSSI adjustments of under constantly changing conditions inside the propagation mobile devices we describe the second step of the proposed al- channel. To monitor the signal strength loss in the propagation gorithm, which we call dynamic RSSI adjustment. In order to channel one could organise a message exchange between the simplify the explanation we consider an example with one AP available APs in the venue. Nevertheless, as we mentioned before, and one anchor device (Figure 1): the mid-size company does not deploy the hardware itself and (1) During the opening hours, every time the anchor’s probe hence cannot influence on the protocols of messages exchange reaches the AP (RSSIa1 ), the adjustment coefficient α is between APs. re-computed by considering the RSSI of the new received In this paper we propose an approach that relies only on the probe: sniffed data of probes from APs. The proposed approach is based RSSIa1 on the monitoring of the RSSI changes between transmitter and α= RSSI 0 receiver in time. These changes correspond to the changes of (2) The adjustment coefficient α is applied to all mobile de- transmitting conditions in propagation channel between these vices probes (RSSIm1 and RSSIm2 on the Figure 1) coming two devices. to this AP until the next anchor probe arrives: One of the first questions to answer is how to detect the fixed transmitters, anchors, based on the sniffed data of probes. The RSSI = α ∗ RSSIm 1 second question is how to adjust the RSSI of sniffed probes from This algorithm permits to cope with the unavoidable channel visitors mobile devices in order to level out the influence of side variation estimating the α parameter. Indeed, α = 1 means that factors, like people fluctuation, on the visitor/passer-by detection. there are no channel variation, α < 1 means that the RSSI value We acknowledge that the proposed solution cannot be applied decreases with respect to the adjustment phase, or increases in cases when high precision is required. Nevertheless, we believe when α > 1, but in the last two cases the channel has changed. the proposed adjustment can improve the existing marketing products without additional costs for hardware deployment. No anchors detected. An additional situation we would like to discuss is the case when there are no detected anchor devices during the opening hours of the venue. For example this can 4 ALGORITHM AND MODEL happen in case of very small venues or in case of possible black outs, or for some other reasons that are difficult to estimate. The algorithm for the dynamic adjustment of RSSI is composed by Every time the algorithm does the calculation of α for AP two steps, which can be broadly summarized as the following. The based on the signals received from the anchor, it can save this first step is dedicated to detecting the devices that can possibly value together with the current number of incoming probes to play the role of anchors. The second step applies the anchors this AP. In cases when no anchors are detected, the algorithm RSSI data to adjust the RSSI data from the mobile devices. can apply the α estimation based on the current level of incoming In order to be selected as an anchor, a candidate device has to probes and the α versus number of probes collected statistics. We meet the following requirements: leave this improvement for our future work investigations. (1) the probes of the candidate device have to be presented 5 EVALUATION in the closing hours of the venue. We assume in these We have used 2.4 GBs of textual data representing a month (May hours there is minimum additional noise in the area of 2018) of raw data from 6 different locations in Italy. This data has investigation. In this case we can measure the basic level been collected by Cloud4Wi1 a SaaS company providing behav- of RSSI for the anchor device. ioral analytic and an omni-channel communication with their (2) the standard deviation of RSSI probes for the anchor can- customers to retailers worldwide. Each line, representing a single didate in the closing hours of the venue has to be low. We request-to-send/clear-to-send (RTS/CTS) handshaking exchange argue that a low standard deviation relates to those devices between a device (which has the Wi-Fi enabled) and an AP in a that have a fixed position over time. These devices can location, contains information about the time of interaction, the be represented by printers, TVs or some venue security access point id, the device id and the RSSI of the signal. appliances. 1 https://cloud4wi.com/ Closing hours Ancor Mobile 1 Mobile 2 AP detection) (Ancor RSSI_0 RSSI_0 RSSI_a1 a=RSSI_a1/RSSI_0 RSSI_m1 RSSI=a*RSSI_m1 Opening hours adjustment) RSSI_a1 (RSSI a=RSSI_a1/RSSI_0 RSSI_m2 RSSI=a*RSSI_m2 RSSI_a1 a=RSSI_a1/RSSI_0 Figure 1: An example of dynamic RSSI adjustment 5.1 Evaluation setup selected time interval also includes the preparation time In order to understand the possibility of applying the described for the staff and security checking period after a venue method, we provide an analysis of different types of venues to closing for clients. see the statistic on the perspective anchor devices. We arbitrarily The parameters described above are derived empirically and separated the locations in groups to see if any difference between are subject for evaluation and studies in a future work. the groupings could have helped in better understanding the context of the analysis. The locations can be split on the basis of Presences of anchor devices their size (both in traffic and in meters) and of their positioning. Based on the described constrains we have evaluated the pres- The two categories that we decided to implement are: ences of anchor candidates for different categories of locations (1) Single standing city venues. The venues that are located in Figure 2. more inhabited areas such as a city centre. In the following As we can see in all the venues there are devices that can be experiments we address these locations as A, B, C. Venues exploit as anchors. At the same time we can see that some venues A and C are located in seasonal touristic cities. Venue B have much more anchors then others. We explain this by the is a city shop located close to the factory building. particularity of the location of the venue. For example, the venues (2) Commercial centers. The venues are located in dedicated A and C are placed in the city center of summer touristic cities shopping areas such as malls or shopping villages. In the (Figure 2a). Since we consider the month of May, which is just following experiments we address these locations as D, E, before the start of summer season, we can see the lack of available F. anchors. At the same time, the venue B is sharing the location with the factory and so we can see an high presence of available 5.2 Data preparation and analyses anchor candidates. Figure 2b shows the evaluation of the number As we already described in Section 4 in order to select the anchor of available anchor candidates for the venues located in the places devices between the variety of available probes we apply the such as malls or shopping villages. As we can see all three venues following numerical constrains: shows availability of the anchor candidates. According to the • The anchor candidate should have at least 4 probes during processed data, the venue F has higher availability of anchors the closing hours. As closing hours of the venue we have in comparison with other venues. It can be explained by the considered deep night hours from 00 : 00AM to 04 : 00AM. particularity of the specific area organization (security devices, Since all the locations are placed in the same country, Italy, etc.). these conditions are applied to all of them. • In order to be sure the candidate device is not changing Distribution of anchors per APs its position over time we fix its standard deviation of RSSI In order to evaluate the distribution of the anchors inside the to be less then 3. Empirically we found that the maximum area we have computed the number of available anchors per of 3 − 4% standard deviation from the RSSI allows to iden- APs deployed in the venue. The results of the evaluation are tify good anchor candidates. An extensive study of this presented in Figure 3. Different dots style and color correspond parameter is out of the scope of this paper, and we leave to the different APs statistic. such study for future work. As we can see on the Figure 3 and as we have already men- • We control the presence of anchor candidates during the tioned on the previous Figure 2 the venues B, D and F have opening hours of the same day. As opening hours for the significantly higher the number of available anchor candidates. venue we have considered the time interval between 8AM This can be also explained by the number of available APs in- and 10PM. We have preferred to be more inclusive for side the location. The number of APs can be used as indirect opening hours interval in order to evaluate the maximum characterization for the venue size. An high number of APs is possible anchor candidates. Nevertheless, the real opening also connected with heterogeneity of the anchor candidates pres- hours can be different from the one mentioned here. The ences. For example on Figure 3b we can see that some of the APs 350 200 D 175 E 300 F Number of anchor devices Number of anchor devices 150 250 A 125 200 B 100 150 C 75 100 50 50 25 0 0 May 01 May 08 May 15 May 22 May 29 May 01 May 08 May 15 May 22 May 29 Date Date (a) Single standing city venue. (b) Commercial centers. Figure 2: Number of available anchor devices for two types of venue location. 60 60 60 50 50 50 Number of anchor devices Number of anchor devices Number of anchor devices 40 40 40 30 30 30 20 20 20 10 10 10 0 0 0 May 01 May 08 May 15 May 22 May 29 May 01 May 08 May 15 May 22 May 29 May 01 May 08 May 15 May 22 May 29 (a) Venue A (b) Venue B (c) Venue C 60 60 60 50 50 50 Number of anchor devices Number of anchor devices Number of anchor devices 40 40 40 30 30 30 20 20 20 10 10 10 0 0 0 May 01 May 08 May 15 May 22 May 29 May 01 May 08 May 15 May 22 May 29 May 01 May 08 May 15 May 22 May 29 (d) Venue D (e) Venue E (f) Venue F Figure 3: The distribution of the anchor devices between available APs in the venues. have a much lower number of available anchors in comparison 4). The Figure 4 shows some randomly selected days and APs with the rest of the APs. By comparison, in Figure 3f there are for the evaluation. For the sake of presentation we selected the some APs that show an higher anchors availability then the rest. anchors with the higher number of available probes during the The rest of the venues with low number of APs show a more day. uniform distribution of anchors. The heterogeneity of the APs The results show that the anchor candidates, as expected, have distribution can correlate with the size of the venues. When the a relatively stable RSSI for the closing hours of the venue and size of the venue is relatively small the available APs cover the show an high deviation during the venue opening hours. The most important zones. Instead in case of big size venue and high difference in RSSI measured in closing and opening hours is number of APs its distribution can be less uniform. Some APs higher than 50% and goes more than 60% in some cases (Figure can be placed behind obstacle or in less popular spots of the 4h) The RSSI measured on the APs reacts on the changes in the commercial venue. propagation channel, for example due to the clients presences. This is important to notice since our approach is based on this Signal strength versus number of probes assumption. In order to evaluate the impact of clients fluctuation on the Figure 5 shows the statistic for number of incoming probes transmitter-receiver propagation channel we have evaluated the to different APs. The APs on the figures are the same as the RSSI changes during the day for some venues/APs/anchors(Figure ones presented on the Figure 4. This choice is done for easing (a) Venue A. AP 88:15:44:bc:63:50 (b) Venue A. AP 88:15:44:bc:63:50 (c) Venue A. AP 88:15:44:bc:63:50 (d) Venue B. AP 88:15:44:a9:0a:85 (e) Venue B. AP 88:15:44:a9:0a:85 (f) Venue B. AP 88:15:44:a9:0a:85 (g) Venue D. AP 0c:8d:db:93:18:9b (h) Venue D. AP 0c:8d:db:93:27:82 (i) Venue D. AP 0c:8d:db:93:26:d4 Figure 4: Signal strength RSSI of anchors versus number of probes per AP. 24 hours statistics. the comparison between the statistics for the same date and AP. dataset shows the availability of anchor devices in all considered In fact by comparing both sets of figures we can see that the locations. The presence of anchor devices is homogeneous for variation of RSSI measured for anchors on the AP (Figure 3) most of the available APs. The results of data evaluation confirm correlates with the variation of the number of incoming probes that the anchor devices present a low deviation for RSSI in closed to this AP. venue hours and an high deviation in opening hours. This con- This is especially noticeable when compared Figures 4a,4b,4c firms the impact the people fluctuation does on the propagation with the corresponding Figures 5a,5b,5c. The shape of the RSSI channel properties. variation is following the shape of the incoming probes to the This work is our first attempt for the dynamic adjustment same AP. An high number incoming probes to the AP indirectly of RSSI measures in fully uncontrolled environment in case of indicates an increase of the number of clients in the area. In turn, WLAN based localization. The described evaluation is based on an high number of clients in the area increases the probability the empirically derived parameters. These parameters are subject that some of them are positioned between the transmitter and for deeper evaluation and studies in the future work. receiver devices, which leads to the variation of the incoming As future work, we plan to deploy a controlled test environ- RSSI. ment where we can evaluate the described approach and test the possible variation for the parameters. We also plan to evaluate 6 CONCLUSION AND FUTURE WORK the possibility to introduce the additional small devices that could interact with available APs. We expect such kind of devices could In this work we propose an algorithm for dynamic adjustment of significantly increase the effectiveness of the proposed approach RSSI measured on APs. The proposed approach is based on two and also can be important in case of lack of available anchor steps. The first step selects the anchor devices, that are used as a candidates. reference point for RSSI. The second step is an actual dynamic adjustment of measured RSSI values of mobile devices. We have evaluated 2.4 GBs of textual data representing a month (May 7 ACKNOWLEDGEMENTS 2018) of raw data from 6 different locations in Italy. This dataset This work was conducted under the project "Wi-Fi Analytics has been collected by Cloud4Wi company. The analysis of this for Marketing Optimization Strategies (WAMOS)" with partial 200 225 88:15:44:bc:63:50 88:15:44:bc:63:50 88:15:44:bc:63:50 180 180 200 Number of received probes Number of received probes Number of received probes 160 160 175 140 140 150 120 120 125 100 100 80 100 80 60 75 60 40 50 00:00 03:00 06:00 09:00 12:00 15:00 18:00 21:00 00:00 03:00 06:00 09:00 12:00 15:00 18:00 21:00 00:00 03:00 06:00 09:00 12:00 15:00 18:00 21:00 07-May 29-May 30-May 2018-05-07 2018-05-29 2018-05-30 (a) Venue A. AP 88:15:44:bc:63:50 (b) Venue A. AP 88:15:44:bc:63:50 (c) Venue A. AP 88:15:44:bc:63:50 900 800 88:15:44:a9:0a:85 88:15:44:a9:0a:85 88:15:44:a9:0a:85 1000 750 800 Number of received probes Number of received probes Number of received probes 900 700 800 650 700 600 700 600 550 600 500 500 500 450 400 00:00 03:00 06:00 09:00 12:00 15:00 18:00 21:00 00:00 03:00 06:00 09:00 12:00 15:00 18:00 21:00 00:00 03:00 06:00 09:00 12:00 15:00 18:00 21:00 01-May 03-May 12-May 2018-05-01 2018-05-03 2018-05-12 (d) Venue B. AP 88:15:44:a9:0a:85 (e) Venue B. AP 88:15:44:a9:0a:85 (f) Venue B. AP 88:15:44:a9:0a:85 200 220 0c:8d:db:93:18:9b 0c:8d:db:93:27:82 0c:8d:db:93:26:d4 500 180 200 180 Number of received probes Number of received probes Number of received probes 160 400 140 160 120 140 300 100 120 200 80 100 60 80 100 60 40 00:00 03:00 06:00 09:00 12:00 15:00 18:00 21:00 00:00 03:00 06:00 09:00 12:00 15:00 18:00 21:00 00:00 03:00 06:00 09:00 12:00 15:00 18:00 21:00 08-May 12-May 31-May 2018-05-08 2018-05-12 2018-05-31 (g) Venue D. AP 0c:8d:db:93:18:9b (h) Venue D. AP 0c:8d:db:93:27:82 (i) Venue D. AP 0c:8d:db:93:26:d4 Figure 5: Number of received probes per AP. 24 hours statistics. financial support of Tuscany region and in collaboration and [8] Henri Nurminen, Marzieh Dashti, and Robert Piché. 2017. A Survey on Wire- with financial support of Cloud4Wi2 less Transmitter Localization Using Signal Strength Measurements. Wireless Communications and Mobile Computing 2017 (16 2 2017). https://doi.org/10. 1155/2017/2569645 EXT="Dashti, Marzieh". REFERENCES [9] Theodore S. Rappaport. [n.d.]. Wireless communications : principles and [1] M. A. Aboelnaga, M. W. El-Kharashi, and A. Salem. 2018. Secure WiFi practice /. ([n. d.]), xxiii, 707 pages :. http://olin.tind.io/record/131417 Fingerprinting-based Localization. In 2018 13th International Conference on [10] P. Richter, M. Valkama, and E. S. Lohan. 2018. Attack tolerance of RSS- Computer Engineering and Systems (ICCES). 543–548. https://doi.org/10.1109/ based fingerprinting. In 2018 IEEE Wireless Communications and Networking ICCES.2018.8639422 Conference (WCNC). 1–6. https://doi.org/10.1109/WCNC.2018.8376977 [2] R. Faragher and R. Harle. 2015. Location Fingerprinting With Bluetooth Low [11] Shweta Shrestha, Jukka Talvitie, and Elena Simona Lohan. 2013. Energy Beacons. IEEE Journal on Selected Areas in Communications 33, 11 Deconvolution-based indoor localization with WLAN signals and unknown (Nov 2015), 2418–2428. https://doi.org/10.1109/JSAC.2015.2430281 access point locations. 1–6. https://doi.org/10.1109/ICL-GNSS.2013.6577256 [3] V. Honkavirta, T. Perala, S. Ali-Loytty, and R. Piche. 2009. A comparative [12] Jukka Talvitie, Markku Renfors, and Elena Simona Lohan. 2015. Distance- survey of WLAN location fingerprinting methods. In 2009 6th Workshop on Based Interpolation and Extrapolation Methods for RSS-Based Localization Positioning, Navigation and Communication. 243–251. https://doi.org/10.1109/ With Indoor Wireless Signals. IEEE Transactions on Vehicular Technology 64 WPNC.2009.4907834 (2015), 1340–1353. [4] Hakan Koyuncu and Shuang-Hua Yang. 2010. A Survey of Indoor Positioning [13] F. Zafari, A. Gkelias, and K. K. Leung. 2019. A Survey of Indoor Localization and Object Locating Systems. International Journal of Computer Science and Systems and Technologies. IEEE Communications Surveys Tutorials 21, 3 Network Security (IJCSNS) 10 (01 2010). (thirdquarter 2019), 2568–2599. https://doi.org/10.1109/COMST.2019.2911558 [5] Juraj Machaj, Peter Brida, and Robert Piché. 2011. Rank based fingerprinting [14] Xiaolei Zhou, Tao Chen, Deke Guo, Xiaoqiang Teng, and Bo Yuan. 2018. From algorithm for indoor positioning. 2011 International Conference on Indoor one to crowd: a survey on crowdsourcing-based wireless indoor localization. Positioning and Indoor Navigation, IPIN 2011, 1–6. https://doi.org/10.1109/ Frontiers of Computer Science 12, 3 (01 Jun 2018), 423–450. https://doi.org/10. IPIN.2011.6071929 1007/s11704-017-6520-z [6] N. Marques, F. Meneses, and A. Moreira. 2012. Combining similarity functions and majority rules for multi-building, multi-floor, WiFi positioning. In 2012 International Conference on Indoor Positioning and Indoor Navigation (IPIN). 1–9. https://doi.org/10.1109/IPIN.2012.6418937 [7] A. Nikoukar, M. Abboud, B. Samadi, M. Güneş, and B. Dezfouli. 2018. Empirical analysis and modeling of Bluetooth low-energy (BLE) advertisement channels. In 2018 17th Annual Mediterranean Ad Hoc Networking Workshop (Med-Hoc- Net). 1–6. https://doi.org/10.23919/MedHocNet.2018.8407089 2 https://cloud4wi.com/