Adopting Machine Learning for Images Transferred with LoRaWAN

Abstract

A R T I C L E I N F O : RECEIVED: 08 JUNE 2020 REVISED: 09 SEP 2020 ONLINE: 22 SEP 2020 K E Y W O R D S : IoT, LoRa, Raspberry PI, security, cloud, machine learning Creative Commons BY-NC 4.0 Adopting Machine Learning for Images Transferred with LoRaWAN 173 Introduction Internet of Things (IoT) denotes the concept of connected smart devices that communicate seamlessly over the Internet. As the market keeps growing, we can classify IoT solution into several major categories. The most common way to denote them is – mission-critical application and massive IoT, based on the technical and commercial requirements they prioritize. Mission-critical solutions are those which require very low latency levels on ultra-reliable networks, often combined with very high throughput. Massive IoT applications, on the other hand, refers to applications which are less latency-sensitive and have lower throughput requirements but require many low-cost, low-energy consumption devices on a network with excellent coverage. The growing popularity of IoT has driven up the demand for massive IoT technologies and the number of smart devices worldwide continues to increase at a dramatic pace. The key requirements for massive IoT are long battery life, good coverage, low cost, and performance flexibility. The technology category that addressing those requirements is low power wide area network (LPWAN) technologies. Nevertheless, modern LPWAN networks being actively analysed 3 within past years still do not have appropriate information security solutions. Probably the first complete analysis of threats and vulnerabilities was published recently. They figured out that LoRa devices have coexisting problems with other LoRa networks and devices. Devices using lower spreading factors can corrupt signals from devices using higher spreading factor in the same network. Furthermore, most LoRaWAN security measures such as the key management and frame counters need to be implemented and taken care of by developers or manufacturers. Therefore, poor implementation also may put end-devices and gateways in danger. A series of articles on key management for LoRa based networks 5,6 was published later advocating for key-management solutions. There was also proposed a way to secure communication for healthcare monitoring system. Another great post describe interference vulnerability. Information security threats should be analysed for many aspects of IoT solutions flow including but not limiting data collection by sensors, data transmission, data transformation, and analysis. All of the described aspects should be considered as sensitive. Threats rate for the given type of system can be classified with the following order for CIA triad (given order of importance from 1 to N) – 1. Integrity, 2. Availability, 3. Confidentiality. This order is driven by the fact, that majority of issues related to confidentiality has human nature, where’s integrity and availability are more technology concerns. In this paper, we adopt image transfer via LPWAN network, with an example of LoRa as hardware, to self-configure the quality of outputs and provide reliable input for Machine Learning algorithms. We advocate for a new way to define integrity within LPWAN networks. Discover and classify existing threats and limitations for media data transfer. M. Brazhenenko, V. Shevchenko et al., ISIJ 47, no. 2 (2020): 172-186 174 Methods We used the methodology of building security profiles based on existing standards, image compression/decompression methodologies, processing of statistical and experimental results methods, a theory of probability methods. Security profiles for the system Based on the existing researches on protection profiles for automated systems 8 and following the proposed way to choose protection profile it is obvious that practically possible should be implemented the following requirements 9,10 for the IoT system (Table 1). • OR Objects reuse • TI Trusted Integrity • AI Administrative Integrity • R Rollback • DEI Data exchange Integrity • RI Registration • AA Authorization and authentication • TC Trusted channel • RS Responsibilities separation • IPC Integrity of protection controls • ST Self-testing • AE Authentication on exchange. • AS Authentication of sender • AR Authentication of receiver These five profiles allow us to achieve a majority of requirements for integrity and availability and as result leaving room to evolve into profiles that address confidentiality as well. Data Exchange Integrity – is the major concern within a list since media transfer in a low-throughput network is not reasonably possible with a single message and so an intelligent approach for data transfer is necessary. Another complication of integrity is coming from the angle of stream-like data for example temperature or accelerometer sensor, because in LPWAN networks some data loss are expected. An important thing to note is that the term integrity in the scope of IoT communication is meant to be different from other more common web or desktop applications. Data integrity is the maintenance of, and the assurance of the accuracy and consistency of data over its life cycle. We definitely cannot evaluate integrity in the same way for majority IoT solutions due to the fact that expressed quality is not achievable for LPWAN. However, even losing the majority of data might not prevent Machine Learning algorithms to produce exactly the same outputs. Summarizing all of the above we propose to define that data integrity for the signal that was send over LPWAN network is achieved if a client (machinery) interpretation for initial signal and received has no difference. Adopting Machine Learning for Images Transferred with LoRaWAN 175 Table 1. Protection profiles. Requirement Profile 1 Profile 2 Profile 3 Profile 4 Profile 5 OR 1 1 TI 1 1 1 1 4