Article Preview
TopIntroduction
The accessible distribution of Internet of Things (IoT) devices has introduced industries, organizations, and individuals to the development of significant worth including IoT applications in section of rescue monitoring. It’s a fact that IoT can certainly help in developing better solutions and real time added value to IoT devices and applications suitable to improve our lives and operational processes. In search and rescue (Bouras, Gkamas, & Katsampiris Salgado, 2021). and general healthcare area, there are several occasions such as rescue monitoring and tracking where sensors can play an important role. Even in COVID 19 era, many works have been made in order to tackle the pandemic, using IoT solutions like Vedaei et al., 2020. Vedaei et al., (2020) have implemented an IoT based system for healthcare and physical distance monitoring system. The main components are the Machine Learning (ML) fog-computing tools, biometric sensors and wireless communication technologies. One of these wireless communication technologies is Low Power Wide Area Networks (LPWAN). LPWAN comes to solve the problem of transmitting data to long distances, with very small energy consumption. Some examples of LPWAN technologies are Long Range (LoRa), Narrowband IoT (NB-IoT) (Routray, S. K., & Mohanty, S. (Eds.). (2021)), SigFox and Weightless. Each technology has its advantages and disadvantages, trying to provide energy-efficient, long-distance, low-cost solutions, sacrificing high throughput, and low latency similar to what cellular technologies provide (Buurman, Kamruzzaman, Karmakar, & Islam, 2020).
As mentioned before, IoT tries to cope with different parameters in the context of the application. In this paper, the authors study the LoRa technology for a number of reasons. Firstly, the Medium Access Control (MAC) layer of the LoRa stack is open, and in order to transmit it is not necessary to pay for a paid subscription, in contrast to the NB-IoT technology, making LoRa a more appealing solution. In a LoRaWAN network, the nodes are not related with an explicit gateway. Instead, data broadcasted by a node is usually received by many gateways. Each gateway will forward the received packet from the end-node to the cloud-based Network Server (NS) via some backhaul. NS perform complex operations including management of the network and filtering redundant received packets, performing security checks, scheduling acknowledgments through the optimal gateway, performing adaptive data rate etc.
In this paper, as far as the LoRa SF assignment is concerned, ML techniques are used, in order to export the appropriate SF that could be used by the network server for data. Authors intent to show their findings regarding the possibility of using ML techniques for SF assignment in LoRa networks. Firstly, the authors explored the data created in the process of LoRa transmissions, and then analyzed and compared four classifications algorithms for the SF assignment using the most used metrics: accuracy, precision, recall, and F1 score. After the evaluation of the models, the authors implemented the ML based system in LoRa and can be used and extended as a separate library to research or university projects. Specifically, a library was created in order to enable the communication between two very important tools, the OMNeT++ based framework called FLoRa, and one of the most well- known libraries for ML called scikit learn. The aforementioned tool uses the FLoRa simulator and python for the ML operations (namely for the training and testing phase of the classification models and for the SF prediction/assignment). Also, we formulated the process of SF assignment as a classification problem. Using the above-mentioned library, two mechanism were created based on the k-NN algorithm and Naïve Bayes classifier. Finally, we present a comparative evaluation of the two proposed mechanisms against two variants of the Adaptive Data Rate (ADR) and the random initialization of the SF. The comparative evaluation was based on delivery ratio and the energy consumption metrics, to study the energy consumption, and the trade-off with the delivery ratio.
The rest of this paper is organized as follows: The next section presents related work. In section “LoRaWAN” important aspects of LoRa are presented in order to better understand our contribution. In section “Background” related works are presented. In Section “Machine Learning Approach” the problem formulation as a ML problem is presented and our approach is presented. In Section “Simulations” the results of our approach and the comparison among other de facto approaches are presented. Finally, the last two Sections the conclusion and future work are presented respectively.