1
Introducing Deep Learning for IoT Security

1.1 Introduction

Internet of Things (IoT) applications and relevant technologies are presently proliferating in all sectors of daily life, including intelligent transportation, smart buildings, smart healthcare, smart manufacturing, smart farming, irrigations, etc. Many security concerns surround the massive amounts of data being sent and received by and from smart devices as they become more widely adopted. Since many IoT applications need safety and defense, authentication and classification systems, as well as sufficient technologies to ensure integrity and confidentiality, are becoming increasingly important. An additional danger posed by a criminal IoT device use is the potential impact on internet security and robustness across the board. Mirai, an IoT-targeted malware, has demonstrated the disruptive power of malevolent operations and the need to implement proper defenses.

1.2 Internet of Things (IoT) Architecture

The purpose of this part is to emphasize the attributes of typical IoT systems and to present the most significant security risks that such systems may be subjected to in their operation. The most significant breakthrough that the IoT has brought to our world is the conversion of standard physical things into digital things that interconnect with each other through the internet using a variety of networking protocols and communication technologies such as 5G and 6G communications [1].

While it may appear to be a straightforward concept, IoT systems involve a lot of dynamic components that should operate jointly to enable them to achieve their tasks appropriately and efficiently. It is crucial for all these various cogs in the machinery or system to run collectively when the IoT solution is functioning as required. In this regard, IoT architecture could be defined as a framework that characterizes the physical elements, the operational structure, network configurations, patterns of data, as well as operational procedures to be applied. However, since IoT spans a wide range of technologies, there is no standard reference architecture. This implies that there is no unified and easy-to-follow template for all viable applications [2].

When it comes to the implementation of the IoT, there is a wide range of architectures and protocols that could be used to support many network applications. The architecture of the IoT system might vary greatly according to its implementation; thereby, it must be flexible or standardized enough to allow the development of a diversity of smart applications.

Despite the fact that there is no standard IoT architecture that is globally approved, a three-layer design seems to be the most common and broadly accepted architecture for both research and industrial communities. Then, the research efforts continue trying to improve over this architecture to cope with recent developments. The three-layer architecture of the IoT system can be displayed in Figure 1.1. As illustrated, the IoT ecosystem could be decomposed into three primary operational layers, namely the physical layer, the network layer, and the application layer. Each of these layers could be further partitioned into additional fundamental sublayers. Each level is briefly summarized in the following subsections, with special emphasis placed on the specific sublayers that can be found inside each level [3].

Figure 1.1 Illustration of the three-layered architecture of the IoT ecosystem.

Source: Geralt/Pixabay.

1.2.1 Physical Layer

The physical layer is sometimes called the perception layer and includes both perceptive activities and fundamental networking resources provided by physical IoT devices, which are all included in this layer.

For perceptive functionality, it involves the primary activities of physical things such as detecting, accumulating, and handling the data perceived from the real world to the extent that it can be done efficiently. Thus, the perception layer incorporates various sensors, such as gas sensors, proximity sensors, infrared sensors, and motion sensors, in addition to actuators, which are used to perform various activities on real-world objects. For setup purposes, a plug-and-play method is typically applied at this layer to deal with the variability of sensors and actuators. Because of their limited battery capacity and compute performance, IoT devices are demonstrated as resource-constrained devices in many ways. A significant portion of the big data volumes that are currently overflowing the cyber-physical systems originates at this IoT layer; nevertheless, these volumes of data are in their raw format, and correct interpretation of them, at this layer, is a critical stage in developing a secure, efficient, and scalable IoT system. In fact, an efficient knowledge of big data pertaining to the IoT can result in a variety of advantages. However, this is typically the responsibility of the application layer [4, 5].

On the other hand, smart devices operate in resource- and power-constrained contexts, and their communication features must be able to cope with lossy and noisy communication environments. Consequently, in order to communicate the data obtained by the sensors, low-energy physical layer connections are required.

Among the most important technologies for IoT communications at this layer are Bluetooth, wireless fidelity (Wi-Fi), Zigbee, ultra-wideband (UWB), radio frequency identification devices (RFID), low-power wide-area network (LPWAN), and near-field communication (NFC), all of which are confronting the aforementioned issues.

1.2.2 Network Layer

It includes communication facilities as well as middleware functions, which are both important for the sustainability of IoT systems. In the case of communication facilities, the resource restriction of IoT devices must however be delicately examined before implementation. One of the most difficult tasks at this layer is providing a distinctive internet protocol (IP) address to the millions of interconnected IoT devices that are internet-connected. By taking advantage of the IPv6 addressing protocol, we may gradually reduce the severity of this problem. Another communication issue in this layer is the volume of the transmitted packets, which will be addressed by the adoption of appropriate protocols, such as IPv6 over low-power wireless personal area network (6LoWPAN), that are capable of providing timely compression capabilities. A third problem pertains to transmission utilities, as transmission protocols should take into consideration the restricted resources in the physical layer, as well as the mobility and plasticity of the internet-connected things, among other things. As a remedy, a routing protocol for low-power and lossy (RPL) networks has been developed as a wireless vector-dependent transmission protocol that works on IEEE 802.15.4 channels, supporting and characterized by its power efficiency. Two modes of communications are supported by this protocol, namely one-to-one communication as well as multi-hop many-to-one communications.

On the other hand, middleware functions often relate to a software layer that sits between the application and network layers, and that is capable of addressing communication and computation concerns in a cooperative manner with the application and the network. Middleware could work as an intermediary between smart things, enabling communication among devices that could otherwise be inaccessible. As "software glue," middleware makes things simpler for developers and engineers to establish input/output operations and communications in such a way that allows them to concentrate on the specific aim of their application rather than becoming bogged down in the technical details.

A variety of useful operations are conceivable in an IoT environment using middleware functions - first, the collaboration and interoperability between diverse IoT devices. So, the varying smart objects could communicate with others seamlessly; second, the scalability to control numerous smart things at the same time; third, device and content look-ups; fourth, sustainability and responsiveness of IoT components; and fifth, IoT devices deeply integrated into our daily lives in the form of smart pillows, smartwatches, smart doors, and smart TVs, which raise the user's concerns about security and privacy of their stored or routed data. The middleware could empower the IoT with some security mechanisms such as identity management and authentication. Finally, when it comes to IoT and the cloud, a flawless and robust connection between these two realms is essential.

1.2.3 Application Layer

Big data analysis, business intelligence, as well as software applications logic are typically found in the uppermost layer. The physical layer of the IoT ecosystem collects a massive amount of valuable data, which is then analyzed using big data analytics. There is a huge amount of data, it is generated quickly, and there is a wide variety of styles. It is necessary to incorporate big data analytical methodologies into the overall IoT architecture, where ML algorithms can contribute significantly to extracting value from the abovementioned huge data and converting them into valuable information. Tasks devoted to delivering services for a particular IoT community...