Review articleMachine learning enabled tools and methods for indoor localization using low power wireless network
Introduction
To revolutionize the user indoor experience, indoor location information is growing in importance in modern communication services and applications [1]. Health monitoring for ageing people [2], activity recognition for energy efficiency in smart homes [3] and many other Internet of Things (IoT) related applications [4] are but a few samples of examples of such applications [5] [6]. In existing active monitoring systems, where the individual has to wear a device equipped with motion sensors, the performance is shown to be around 90% for the recognition of activities such as running, walking, standing, sitting and sleeping [7]. [8] introduced sparse low-power sensor networks using passive electric potential sensors for crowd-aware smart buildings. The networks can detect human presence, track their positions, and estimate room occupancy by measuring the electric field disturbances caused by human movement. In addition, the proposed passive sensing system can remotely monitor human physiological signals of nearby stationary occupants. Unfortunately, a person may not be always wearing such sensors, which in fact would be unwieldy otherwise. On the other hand, in passive activity recognition, line-of-sight (LOS) requirement limits the use of camera-based system.
Because of their ubiquitous availability in indoor areas, wireless signals have been the focus of much research for activity recognition [1]. In fact, different indoor localization approaches in the literature can be generally classified into three main groups according to the modeling information that they are relying on. Namely, one can either rely on the received signal strength (RSS)-based methods, angle-of-arrival (AoA)-based methods or time-of-arrival (ToA)-based localization algorithms [9]. Both AoA and ToA localization schemes rely on stringent system-level requirements such as complex antenna array on the receiver side [10] and precise clock synchronization among the reference devices respectively [9]. Relying on a mutually synchronized receivers, the authors in [9] proposed an algorithm to obtain a high-resolution estimate of the time of arrival (TOA) of the long training sequence symbol at each receiver in a LoS scenario. However, unlike [9], He and al. [11] implemented a complete prototype based on asynchronous TDOA wherein the test areas are limited to a few meters’ squares.
Relying on WiFi fingerprinting techniques, the RSS has been used successfully for active localization of wireless devices [12] [13]. Tseng et al. are the first to utilize ray tracing as a channel predictor to assist indoor finger printing using channel impulse response measurements to achieve 25% localization error reduction [13]. Using ZigBee wireless sensor networks, the authors in [6] presented a novel RSS-based fingerprinting approach for room-level localization. This is a threshold algorithm based on receiver operating characteristic analysis. An alternative approach, enabled by ultra-wide band (UWB) radios, is based on measuring the time-off-flight (ToF) between two nodes and enhanced by detailed information about propagation channel characteristics [1].
The authors in [5] and [1] presented a custom UWB system and a modified WiFi system to acquire some other metrics than RSS. After transforming the wide-band 802.11 to narrowband pulses, the WiSee system [3] extracts the doppler information using the short-time Fourier Transform (STFT) which is then segmented to discriminate different patterns. These segments are then used in a simple classification scheme. On the other hand, the authors in [1] presented an approach to indoor localization using a convolutional neural network (CNN) for channel classification and ranging error regression on 1-dimensional raw CSI traces rather than on derived features. In [5] the authors used the long short-term memory (LSTM) for activity recognition where the feature extraction is not performed. Even if the LSTM approach suffers from long training time, it is shown to outperform the ML-based techniques.
Apart from WiFi signals, the expected wide spread of IoT devices using low power wide area (LPWA) technologies such as LoRaWAN, opens up new perspectives to explore the massive amount of generated data. It is therefore expected to enable various services and applications that rely on precise locations of people, goods, and assets, ranging from home automation and assisted living to increased automation of production and logistic processes and wireless network optimization. A variety of applications across several business verticals can exploit LPWA technologies to connect their end devices. These business sectors include but not limited to smart grids, smart city, and industrial monitoring, etc. [14].
LPWA networks are different as they make unique trade-offs than the traditional technologies predominant in IoT scene such as Zig-Bee and Bluetooth. These legacy wireless technologies are not ideal to connect low-power devices distributed over a large area. Therefore, the devices cannot be randomly deployed [15]. The range of these technologies is extended using a dense deployment of devices and gateways, usually connected using multi-hop mesh networking. On the other hand, legacy wireless area networks (WLANs) are categorised by shorter coverage areas and higher power consumption for machine-type communication (MTC). With a remarkable range of up to tens of kilometers [16] and battery life of ten years and more, LPWA technologies are promising for the Internet of low-power, low-cost, and low throughput devices. The authors in [17], discussed the use of a long-range (LoRa) technology, originally developed for IoT, in implementing a giant distributed measurement systems. They argued that LoRa and LoRa wide area network architectures show a good match with measurement systems.
The current dominant solutions that uses WiFi, Zig-Bee or Bluetooth are not suitable for indoor localization if the devices are not at a close reach or when the indoor propagation conditions are harsh so that the signal is so weak that it cannot be detected. In addition, these technologies are not designed to support the localization of a massive number of devices. These disadvantages are overcome by LoRaWAN that operates with extremely weak signal levels. This feature makes LoRaWAN-based localization a good candidate for outdoor environment as well. Therefore, the main contributions of this paper are:
- •
We present a measurement and localization system and framework based on an end-to-end LoRaWAN network. The measurement set-up provides access to not only the sensor data but also to the physical layer metrics such as the receiver signal strength (RSS), spreading factor (SF), and the frequency hoping signature to name a few.
- •
Exploit the RSS with the frequency hoping signature using machine and deep learning techniques for indoor localization applications. We demonstrate that the partial CSI can be efficiently exploited to predict indoor location with accuracy of more than 98% using a multilayer neural network (MNN). The idea of exploiting the frequency hoping signature stems from the failure of the current methods based on a single RSS to provide accurate localizations [28]. One can resort to use several gateways to aid in improving the indoor localization accuracy, however, this comes at extra infrastructure cost [29]. In addition, the problem of using the time difference of arrival (TDoA) for this goal is that not all devices nor all gateways can provide a proper estimation of these values, yet.
- •
Discuss the optimization methodology of the support vector machines (SVM) model-based classification using scikit-learn [27]. We particularly focus on SVM model to discuss the variance-bias trade-off [25]. Two very simple but powerful diagnostic tools that can help in improving the performance of a learning algorithm are used, namely, the learning curves and the validation curves. These learning curves aim at diagnosing whether a learning algorithm has a problem with overfitting (high variance) or underfitting (high bias). In addition, we will exploit the validation curves to address the common issues of a learning algorithm.
The paper is organized as follows: Section 2 presents the system architecture. Section 3 discusses the measurements and the dataset. Different machine and deep learning techniques are presented and leveraged to address the indoor localization application as a classification problem in Section 4. Finally, the conclusions are drawn and some future research directions are outlined in Section 5.
Section snippets
System architecture
The system architecture for indoor localization is conceptually depicted in Fig. 1. The custom end device uses Symphony-link module to transmit sensors data over-the-air to the gateway that runs the Symphony-link management application. The data are forwarded to the cloud facility. The cloud-based services enable managing the configuration of and the data used by Symphony Link networks [18]. The cloud system is currently deployed inside Amazon Web Server, though it could in principle be
Measurements and dataset
This section outlines the procedures and the measurement equipment used for collecting the indoor non-line-of-sight (NLoS) classification dataset. The measurement campaign includes the collection of NLoS data in four fixed indoor locations around the gateway. The high sensitivity of the LPWA technology such as LoRaWAN are leveraged to overcome the high path loss inherent to NLoS indoor environment equipped with a handful number of gateways (not a dense deployment).
Indoor localization
This section addresses the node localization problem as a classification task using ML models. The main novelty here is the use of the frequency value along with the RSS for incomplete CSI. Also, the importance of the separation of the data for the test group is highlighted.
Conclusion
In realizing the vision of the IoT, LPWA technologies complement and sometimes supersede the conventional cellular and short-range wireless technologies in performance for various emerging smart cities and machine-to-machine (M2M) applications. Therefore, the expected wide spread of IoT devices using LoRaWAN, opens up new perspectives to explore the massive amount of generated data to enable various services and applications that rely on precise locations of people, goods, and assets, ranging
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
This work has been funded by the Natural Sciences and Engineering Research Council of Canada grant (RGPIN-2015-03674, RDCPJ 533444-18) and the Canada Foundation for Innovation, the CMC Microsystems grant (20559). The authors would also like to thank Zine Laabidine Lobka and Nicolas Sicard for work on the IoT device design and measurements, respectively
References (32)
- et al.
Improving indoor localization using convolutional neural networks on computationally restricted devices
IEEE Access
(2018) - et al.
WiFall: device-free fall detection by wireless networks
- et al.
Whole-home gesture recognition using wireless signals
We can hear you with Wi-Fi!
IEEE Trans. Mobile Comput.
(2016)- et al.
A survey on behavior recognition using WiFi channel state information
IEEE Commun. Mag.
(2017) - et al.
RSSI-based indoor localization and identification for ZigBee wireless sensor networks in smart homes
IEEE Trans. Instrum. Meas.
(2019) - et al.
Human motion detection in daily activity tasks using wearable sensors
- et al.
Indoor occupancy awareness and localization using passive electric field sensing
IEEE Trans. Instrum. Meas.
(2019) - et al.
Indoor localization using 802.11 time differences of arrival
IEEE Trans. Instrum. Meas.
(2016) - et al.
Efficient closed-form algorithms for AOA based self-localization of sensor nodes using auxiliary variables
IEEE Trans. Signal Process.
(2014)
High-accuracy localization platform using asynchronous time difference of arrival technology
IEEE Trans. Instrum. Meas.
Chameleon: Survey-free updating of a fingerprint database for indoor localization
IEEE Pervasive Comput.
Ray-tracing-assisted fingerprinting based on channel impulse response measurement for indoor positioning
IEEE Trans. Instrum. Meas.
Low power wide area networks: an overview
IEEE Commun. Surv. Tutor.
Low power wide area machine-to-machine networks: key techniques and prototype
IEEE Commun. Mag.
On the coverage of LPWANs: range evaluation and channel attenuation model for LoRa technology
Cited by (11)
LoRa-based outdoor localization and tracking using unsupervised symbolization
2024, Internet of Things (Netherlands)A survey on optimal indoor objects localization in Internet of things based on deep learning
2023, AIP Conference ProceedingsExperimental Evaluation of Trilateration-Based Outdoor Localization with LoRaWAN
2023, Computers, Materials and ContinuaDesign and Implementation of Temperature and Humidity Monitoring System Using LPWAN Technology
2022, Ingenierie des Systemes d'Information
Laboratoire des Signaux et Systèmes Intégrés Chaire de recherche sur les signaux et l'intelligence des systèmes haute performance.