1. Introduction
Location-based services (LBS) have drawn great attention from academia and industry in recent years as they have become more frequently used and demanded. Positioning is at the core of LBS, and it remains a technical challenge for dense urban environments, indoors, and underground [
1], despite the many efforts devoted to alternatives to GNSS-based positioning [
2], and in particular to alternatives readily applicable to smartphones [
3]. Among those alternatives are magnetic signals [
4,
5], RFID [
6], LED light [
7,
8], vision-based techniques [
9], and Wi-Fi [
10] and Bluetooth Low Energy (BLE) [
11] signals, all of them often combined with Pedestrian Dead Reckoning (PDR) and Map-Matching [
3,
12]. Cellular networks can be used for positioning using techniques like UTDOA [
13], OTDOA and E-CID [
14] with 4G LTE networks. The advent of 5G NR networks will likely mean densely distributed access nodes and wide bandwidths at high frequency bands, which will likely enable very accurate positioning in GNSS-denied environments [
15,
16].
BLE beacon-based positioning is a reality [
17], and it is currently applied in museums, airports [
17,
18,
19] and applications that require fine-grained proximity or a relatively cheap positioning technique when Wi-Fi-based positioning is not appropriate [
20]. BLE beacons are usually small, advertisement-emitting, battery-powered devices designed for easy deployment in a variety of indoor environments. Currently, there is richness in the BLE beacons availability regarding vendor, price range, supported protocols, transmission powers, advertisement frequency capabilities and battery lifespan [
21]. Higher achievable accuracies, better privacy, lower phone battery drain, and lower network traffic are some of the advantages of using BLE over Wi-Fi for indoor positioning [
3,
11].
BLE was introduced in Bluetooth 4.0, and among its goals were to lower costs and to lower the power consumption in comparison to the “classical” Bluetooth [
22]. BLE shares many similarities with Wi-Fi (2.4 GHz band frequencies), and thus, it is often used in a similar way as Wi-Fi for indoor positioning, i.e., applying RSS-based techniques [
3]. However, some important differences should be considered when dealing with BLE RSS measurements [
11]. Furthermore, while most Wi-Fi positioning research works have the assumption that there are Wi-Fi routers already in place, the BLE beacon placement choice should be carefully considered to find a proper balance among goals like accuracy, cost, robustness, and battery duration.
The need for RSS databases in indoor positioning research that would foster reproducibility and comparability has been acknowledge for the case of Wi-Fi [
23], with several and variate available databases [
24]. Such need has also been referred for other radio-frequencies [
25,
26]. However, to the best of our knowledge, the number of available BLE RSS databases is currently low. Tóth and Tamás [
27] collected measurements from Bluetooth enabled devices, but their actual locations were not registered. In some databases, the measurements were collected using Raspberry Pis [
28,
29,
30,
31] while the localization target had a BLE emitting device. For Iqbal et al. [
28], Byrne et al. [
29], Sikeridis et al. [
30], the localization target was a participant describing a track, in several environments, while in [
31] the target was static. The number of BLE RSS databases is even lower for the case of RSS measurements for off-the-shelf (location or proximity intended) beacons signals collected using smartphones. The database in Mohammadi et al. [
32] provides RSS readings from 13 iBeacons (with a 30–40 feet separation) mounted on the ceiling of a floor from a university library. The measurements were collected using an iPhone 6S. The database from Lazik et al. [
33] was collected using an iPhone 5S by participants describing tracks in several environments and include measurements from several sensors. The beacons were developed by the authors, they were mounted in tripods near the ceiling, and they emitted iBeacon broadcasts at 10 Hz. In the environments, 3–4 beacons were always visible in line-of-sight to the receptor, and the number of beacons in a environment ranged from 5 to 11. Their database also provides information about the environment and scripts to load the data and to test several positioning algorithms that use measurements from several sensors. In Baronti et al. [
34], the authors setup had Raspberry Pis boards with BLE dongles acting as (fixed) emitters and receivers, also having BLE beacons and smartphones carried by subjects. Their dataset is divided into a set for remote positioning, which corresponds to data collected by fixed emitters and emitted by BLE beacons (wearable emitters), and a set for self positioning, which correspond to measurements collected by smartphones and emitted by the fixed emitters. The authors considered several rooms and three power transmission configurations. One phone model was used and there was one fixed emitter/receiver per room.
The main goal of this paper is to introduce to the research community a new BLE RSS database and its related information. The database is composed by BLE measurements collected for research in RSS and smartphone based, fine grained, indoor positioning techniques. The measurements were taken in two zones of the Universitat Jaume I (UJI): an area of its Library’s 5th floor and an area of the Geotec group office space. Both of them featured dense BLE beacon deployments that allow positioning accuracies below 3 m. The RSS measurements were carefully labeled with their collection details and they involved three smartphones of different brand and model for the Library zone, and three different transmission power levels for the Geotec zone.
To the best of our knowledge, no other freely available BLE RSS database allows testing positioning methods across several devices, environments and transmission power levels with a high density of deployed BLE beacons. The database described in this paper intents to aid the development of indoor positioning methods that may obtain high accuracies by relying in a relatively high density of emitters, namely BLE beacons. BLE-based positioning methods are able to obtain higher accuracies than those based on Wi-Fi [
3,
11]. Despite technologies like UWB and ultrasound allow higher accuracies than BLE, they are not widely supported in user devices or have cheap emitter devices that form an ecosystem that allows them a fast-paced growth in market and research [
35].
The BLE RSS data is accompanied by beacon deployment descriptions, environment information, and a set of Matlab © scripts that help in data handling and perform demonstrative analyses. The analyses are intended to show likely data usage in indoor positioning research and to highlight important details to take into account when developing BLE RSS-based positioning methods. More specifically, they show how the variations in signal strength and in the advertisement detection intervals deteriorate the positioning accuracy, while also demonstrating how the known buffering technique [
11] effectively deals with such issues. Additionally, the analyses show the difference in reported RSS and the positioning accuracy among the collection smartphones and transmission power levels. Furthermore, they present how, apart from fingerprinting, which is the star technique for Wi-Fi based positioning, the much simpler Weighted Centroid is an appropriate method for BLE-based positioning given that the beacon deployment locations are known and that its accuracy is high. In addition, the two position estimation methods are tested first under three deployment configurations and later under random beacon disconnection settings. Moreover, the collection locations match some of the positions used to collect samples for previous Wi-Fi databases [
24,
36] which may prove useful to test indoor positioning approaches that may harness both signals, as well as RSS values comparisons regarding stability and emitter detection times.
The organization of the remaining sections of the paper is as follows.
Section 2 provides insights on the the data collection process and environments.
Section 3 describes the RSS data and its related information.
Section 4 provide analyses that introduce the database usage and BLE RSS-based positioning. Finally,
Section 5 closes the paper with some concluding statements.
2. Setup and Collection Procedure
The beacons used in this work are off-the-shelf Accent Systems’ IBKS 105 [
37] with a Nordic nRF51822 core [
38]. Those beacons are capable of concurrent broadcasting iBeacon
TM and Eddystone
TM advertisements across several emission slots. The deployed beacons were configured to use only one iBeacon slot and an advertising period of 200 ms (5 Hz), which is a setup that provides a battery duration of 8 to 10 months. The collection was organized in campaigns, and performed by trained individuals (hereinafter, the subjects), that stood at predefined positions, holding their smartphones with the right hand in front of their chests in a way that resembled people following directions for an indoor environment.
The radio signals in indoor environments are known to be affected by factors like multi-path, fast fading [
39], and in the case of 2.4 GHz frequencies, strong human body absorption, which may account for up to ≈10 dBm [
40], being even more notable for devices like smartwatches [
41]. In addition, the BLE advertisements are broadcast on three narrow (2 MHz width) advertising channels in quick succession, which made the reported signal instability a higher challenge than in Wi-Fi positioning [
11].
The BLE data collection was performed using Android smartphones, due to their abundance and availability. We adapted a previous Wi-Fi RSS collection application [
24] to perform campaign-directed BLE RSS collections. Android provides two ways of notifying BLE advertisement detections: anytime an advertisement is received, or when a batch of advertisements is collected. According to our tests, the batch approach leads to fewer advertisement detections, and thus, we decided to use the former approach. Giving that the advertisement collection was not batch-driven, a fingerprint abstraction was created, so that all measurements received in a window of 1s were placed into a fingerprint. If a beacon was detected more than once in a fingerprint window time, only its mean RSS value, (and the first detection time in nanoseconds since device boot), was stored. The smartphones that the subjects used for RSS collection had the previously mentioned application, which helped the subjects (
Figure 1) in following a collection campaign and avoiding errors like wrong position tagging. A collection campaign is an ordered list of positions. The subjects had to follow that order and collect a number of consecutive samples (fingerprints) at each point facing a specific direction. The collection orientations, i.e., the facing directions, were those corresponding to the usual walking directions.
The RSS measurement collection was performed in two environments (zones) from the UJI, in Castellón, Spain: an area among bookshelves in the Library building (hereinafter, Library) and an area of the office space of the Geotec Research Group (hereinafter, Geotec).
Table 1 presents data describing the collection, the beacon deployment and the environment of each zone. In
Table 1, MDMS represents the maximum distance from the position of a collection point to another collection point that is the closest to it in the space. MDMB is described as MDMS, but considering beacon positions instead of collection points positions.
The deployment in the Library (
Figure 2a) had the goal of supporting a positioning service for users trying to find a book. The 22 beacons were placed inside the enclosed top of the (wooden) shelves for security concerns (
Figure 2b). Therefore, there were no line-of-sight situations to beacons in this environment. The dense deployment was designed so that the area covered by the beacons included all shelves and the deployed beacons were as far apart from each other as possible. The shelves height is 2.35 m, which is almost as high as the altitude of the ceiling (about 2.60 m). The collection was divided into two campaigns, one with the subjects facing the up or down directions and another one with them facing the left or right directions (blue squares and orange circles in
Figure 2b, respectively). Each collection campaign was performed three times, each time using a different smartphone and by a different individual. The smartphones were a BQ Aquaris X5 plus, a Samsung Galaxy S6 (SM-G920F) and a Samsung Galaxy A5 2017 (SM-A520F), hereinafter BQ, S6 and A5, respectively.
In the Geotec zone (
Figure 3a) there are only four tall furnitures and the beacons were attached to the ceiling tiles (
Figure 3b). As a result, every collection point had line-of-sight situations with more than three beacons at the same time, if human-body blockage is not considered. The collection was performed following one campaign (with the subject facing the up or down directions) and one smartphone of the those used for the Library zone. The campaign was performed three times, each time with the beacons configured for a different transmission power.
3. The BLE RSS Database
The provided database is openly available at Mendoza-Silva et al. [
42]. It contains a dataset for the Library zone (
) and another dataset for the Geotec zone (
). Each dataset is composed of four sets: RSS values, positions, times and identifiers sets. A set of RSS values is defined as:
where
p is the number of points (unique triplets of the 2D coordinates and the direction the subject was facing),
s is the number of samples per unique triplet collected in the zone,
a is the number of beacons deployed in the zone, and
is the BLE RSS value measured in dBm (or a non-detection value of 100) for the
j-th beacon (column) at the
i-th fingerprint (row). The operation
represent the product between two real numbers. Remember from
Section 2 that a fingerprint is composed by the RSS values associated to beacon advertisements detected in a 1-s time window. A set of times is defined as:
where
is the timestamp when a
j-th beacon RSS value was measured (or a non-detection value of 0) during the time period corresponding to the
i-th fingerprint. The timestamp is relative to the device boot and it represents the number of nanoseconds elapsed until the advertisement was detected. A set of positions is defined as:
where
,
are the (
x,
y) local coordinates where the
i-th fingerprint was collected. The local coordinates were designed taken a given position in the target zone as the coordinates origin (represented by magenta asterisks in
Figure 2a and
Figure 3a) and assuring that 1 unit of distance represents 1 m of actual distance in the zone. A set of identifiers is defined as:
where
is a number that uniquely identifies the
i-th fingerprint in the whole database. The number format contains information that allows determining membership information of the fingerprint, as shown in the example of
Figure 4. The membership information is harnessed by the supporting scripts provided along with the database for fingerprint selection.
The files in the database are organized in three folders: the RSS measurements and their labels (“rss”), the BLE beacon deployment positions (“dep”), and the geometries of obstacles (shelves and pillars) found inside the collection area (“obs”). The name of every file includes an indication of the zone its information refers to: “lib” for the Library zone and and “geo” for the Geotec zone. The RSS values, positions, times and identifiers sets of a zone are each stored in a file whose name includes “rss”, “crd”, “tms”, or “ids”, respectively. The i-th row of each of them holds the respective information of the i-th fingerprint collection in the zone.
5. Discussion and Conclusions
This paper introduced a new BLE RSS database freely available to the research community. The RSS data was taken at two distinct zones of a university: among shelves in a library and at an office space. The RSS measurements were carefully annotated with position, time, and collection details labels. The data was gathered in the library zone using three smartphones (by three individuals) and in the office zone using one of those smartphones with beacons configured at three transmission powers. The difference of zones, collection devices, and transmission powers, as well as the dense beacon deployments, make the data useful to analyze a BLE RSS-based indoor positioning system in different environments and device settings, thus, likely making it more suitable to practical tests. In addition to the RSS and their associated labels, the position where the beacons were deployed and the geometries of shelves found in the environments are also provided.
Beyond the description of the collection process and the provided data, this paper provided a starting point to the usage of the data and presented key problems in the BLE RSS-based indoor positioning. The reader is introduced to the effect of the environment on the RSS values, the signal detection sensibility differences in the smartphones, and the major hurdle that is advertisement detection delays, which is significant in some devices. To test positioning using our data, two simple but widely used methods were applied: the weighted centroid and the kNN-based fingerprinting. The paper presented the results of testing them across the range of zones, devices, and powers that our database allows, and also considered a solution to the detection delays problem. In addition, brief analyses regarding beacon placement and positioning method robustness to beacon disconnection are presented. The data, the code, the analyses and the references described in this document may prove of significant importance for further studies into challenges of BLE assisted indoor positioning.
The analyses presented in this paper are only simple examples to encourage the research community working in indoor positioning to use our dataset and to introduce some common challenges in BLE RSS-based positioning. Plenty of other experiments are possible using the dataset, like analyzing the effects of orientation of different training and test sets or beacons deployment selection on fingerprinting methods, as well as the combination of BLE and Wi-Fi, if our previously published datasets are also used, and more in deep analysis of Wi-Fi and BLE similarities and differences. We plan to create new versions of the database, which would likely include other locations and collection orientations.