Because of the increasing relevance of the Internet of Things and location-based services, researchers are evaluating wireless positioning techniques, such as fingerprinting, on Low Power Wide Area Network (LPWAN) communication. In order to evaluate fingerprinting in large outdoor environments, extensive, time-consuming measurement campaigns need to be conducted to create useful datasets. This paper presents three LPWAN datasets which are collected in large-scale urban and rural areas. The goal is to provide the research community with a tool to evaluate fingerprinting algorithms in large outdoor environments. During a period of three months, numerous mobile devices periodically obtained location data via a GPS receiver which was transmitted via a Sigfox or LoRaWAN message. Together with network information, this location data is stored in the appropriate LPWAN dataset. The first results of our basic fingerprinting implementation, which is also clarified in this paper, indicate a mean location estimation error of 214.58 m for the rural Sigfox dataset, 688.97 m for the urban Sigfox dataset and 398.40 m for the urban LoRaWAN dataset. In the future, we will enlarge our current datasets and use them to evaluate and optimize our fingerprinting methods. Also, we intend to collect additional datasets for Sigfox, LoRaWAN and NB-IoT.

Practical location estimation is never ideal, and each location estimate is burdened with a certain level of error. In many use-case scenarios, knowing the magnitude of these errors can significantly improve the usability of the location estimates. The localization errors for different localization approaches are currently assessed using static performance benchmarks. These benchmarks typically provide aggregate metrics that statistically characterize the localization errors across the entire deployment environment. Due to the potentially dynamic nature and spatial heterogeneity of the environment, this characterization can be too generic to be really useful from the point of view of an individual location estimate. To address this issue for fingerprinting-based localization, we propose a regression-based procedure for estimating the individual (i.e., per-location estimate) localization errors. We use the received signal strength (RSS) values from various locations in an environment, as well as the observed localization errors in case the location estimates are generated using these RSS values, for training a number of contemporary regression models. Using the trained models, we are able to estimate the localization error of a location estimate at a new location using only RSS values collected at that location. Both by simulation and experimentally, we demonstrate the feasibility of the proposed procedure for Wi-Fi-based indoor and LoRa-and SigFoxbased outdoor fingerprinting approaches. We do that by showing that the proposed procedure can, in the best-case scenario, yield more than 50% more accurate estimation than the reference procedure based on the average localization errors derived from the static performance benchmarks.

No abstract

The technology development in wireless sensor network (WSN) offers a sustainable solution towards precision agriculture (PA) in greenhouses. It helps to effectively use the agricultural resources and management tools and monitors different parameters to attain better quality yield and production. WSN makes use of Low-Power Wide-Area Networks (LPWANs), a wireless technology to transmit data over long distances with minimal power consumption. LoRaWAN is one of the most successful LPWAN technologies despite its low data rate and because of its low deployment and management costs. Greenhouses are susceptible to different types of interference and diversification, demanding an improved WSN design scheme. In this paper, we contemplate the viable challenges for PA in greenhouses and propose the successive steps essential for effectual WSN deployment and facilitation. We performed a real-time, end-to-end deployment of a LoRaWAN-based sensor network in a greenhouse of the ’Proefcentrum Hoogstraten’ research center in Belgium. We have designed a dashboard for better visualization and analysis of the data, analyzed the power consumption for the LoRaWAN communication, and tried three different enclosure types (commercial, simple box and airflow box, respectively). We validated the implications of real-word challenges on the end-to-end deployment and air circulation for the correct sensor readings. We found that temperature and humidity have a larger impact on the sensor readings inside the greenhouse than we initially thought, which we successfully solved through the airflow box design.

Abstract:Because of the increasing relevance of the Internet of Things and location-based services, researchers are evaluating wireless positioning techniques, such as fingerprinting, on Low Power Wide Area Network (LPWAN) communication. In order to evaluate fingerprinting in large outdoor environments, extensive, time-consuming measurement campaigns need to be conducted to create useful datasets. This paper presents three LPWAN datasets which are collected in large-scale urban and rural areas. The goal is to provide the research community with a tool to evaluate fingerprinting algorithms in large outdoor environments. During a period of three months, numerous mobile devices periodically obtained location data via a GPS receiver which was transmitted via a Sigfox or LoRaWAN message. Together with network information, this location data is stored in the appropriate LPWAN dataset. The first results of our basic fingerprinting implementation, which is also clarified in this paper, indicate a mean location estimation error of 214.58 m for the rural Sigfox dataset, 688.97 m for the urban Sigfox dataset and 398.40 m for the urban LoRaWAN dataset. In the future, we will enlarge our current datasets and use them to evaluate and optimize our fingerprinting methods. Also, we intend to collect additional datasets for Sigfox, LoRaWAN and NB-IoT.