Autonomous Configuration of Communication Systems for IoT Smart Nodes Supported by Machine Learning

Abstract

Machine Learning brings intelligence services to IoT systems, with Edge Computing contributing for edge nodes to be part of these services, allowing data to be processed directly in the nodes in real time. This paper introduces a new way of creating a self-configurable IoT node, in terms of communications, supported by machine learning and edge computing, in order to achieve a better efficiency in terms of power consumption, as well as a comparison between regression models and between deploying them in edge or cloud fashions, with a real case implementation. The correct choice of protocol and configuration parameters can make the difference between a device battery lasting 100 times more. The proposed method predicts the energy consumption and quality of signal using regressions based on node location, distance and obstacles and the transmission power used. With an accuracy of 99.88% and a margin of error of 1.504 mA for energy consumption and 98.68% and a margin of error of 1.9558 dBm for link quality, allowing the node to use the best transmission power values for reliability and energy efficiency. With this it is possible to achieve a network that can reduce up to 68% the energy consumption of nodes while only compromising in 7% the quality of the network. Besides that, edge computing proves to be a better solution when energy efficient nodes are needed, as less messages are exchanged, and the reduced latency allows nodes to be configured in less time.

A Review on the Application of Internet of Medical Things in Wearable Personal Health Monitoring: A Cloud-Edge Artificial Intelligence Approach

“…EFL capitalizes on the computational prowess of edge devices, orchestrating local model refinements predicated on their distinct datasets. Notably, only the model's iterative updates are relayed to a central server [126]. This collaborative learning modality facilitates the model's refinement across a heterogeneous array of data sources, all the while adhering to stringent data privacy stipulations.…”

An Efficient Wireless Sensor Network Based on the ESP-MESH Protocol for Indoor and Outdoor Air Quality Monitoring

“…Multiple sink node networks would be merged into a single network with a single sink node. In addition, setting a small number of iterations and a low voltage threshold percentage might cause network instability owing to RSSI variations [19][20][21].…”

Comparison of edge computing methods in Internet of Things architectures for efficient estimation of indoor environmental parameters with Machine Learning

“…In 2017, total spending on these technologies reached nearly $2 trillion and 20 billion of connected devices were forecasted by 2020 (Gartner Inc., 2017) and, in 2023, the global edge computing market is expected to reach $1.12 trillion (Walker et al, 2022). In addition, it is estimated that close to 50 million devices will be connected by 2030, forcing IoT architectures to be more scalable, efficient and autonomous (Gloria and Sebastiao, 2021). The IoT paradigm shift from cloud to edge has contributed to this evolution, placing increasing importance on computing close to physical data sources using resource-constrained devices (Samie et al, 2019), referred to as edge computing.…”

“…The edge computing introduces a new computation layer physically closer to the end-users that tries to overcome the problems of cloud computing, such as communication latencies, network congestion, and security issues (Harish et al, 2020) in applications of computer vision, speech recognition, natural language processing or weather forecast. In such intelligent applications, the communication and processing are the major components, but the power consumption is the most limiting factor in edge implementations that support AI processing techniques (Ai et al, 2018;Gloria and Sebastiao, 2021). These solutions are traditionally implemented using a centralised approach, in which IoT sensing nodes collect raw data to be streamed to a concentrator that acts as a gateway, and a remote cloud or another edge device with high processing capacity performs compute-intensive tasks, such as AI model training and inference (Ramos et al, 2019).…”