up opportunities to novel services and applications that do not require centralized (cloud) processing, which alleviates energy consumption and reduces the risk related to security and privacy [56].
For instance, [57] introduces a flexible toolkit that generates energy efficient and lightweight artificial neural networks for a class of microcontrollers. The authors discuss the different optimizations in energy efficiency and computational power. Besides, they test the applicability of their results in a self-sustainable wearable.
Moreover, [56] provides a comprehensive review of the TinyML ecosystem and frameworks for integrating algorithms into microcontrollers, and proposes a multi-RAT architecture suitable for MTDs comparing the performance of different classification algorithms. The results show that these algorithms achieve good accuracy and speed. However, the authors warn about careful design needed due to memory restrictions at the microcontroller. All in all, these recent findings evince a promising research area.
Heterogeneous Access and Resource Management. Conventionally, ortho-gonal access and radio resource management ensure that users connect to the network and convey information. Nonetheless, with mMTC, the number of users scales to a point beyond the availability of orthogonal resources, which may incur performance losses, extended delays, large energy consumption. In this context, non-orthogonal solutions emerge at both levels, access and resource management. To combat such losses, non-orthogonal multiple access (NOMA) multiplexes users in either of the two most prominent solutions: power [58] or code [59] domain. This strategy can allocate more users through sophisticated successive interference cancellation (SIC) techniques than the available resources, which outweighs the cost of increased complexity at the receiver. A more recent development called rate-splitting multiple access [60] promises even further improvements in spectral and energy efficiencies than other non-orthogonal counterparts [61, 62]. At the same time, [63] discusses the information-theoretical limits of massive access.
Non-orthogonality plays an essential role in other domains, such as resource management. In [64], the authors introduce and analyze the heterogeneous coexistence of different service modes, referred to as non-orthogonal slicing, from an information-theoretical perspective. Later, this idea is expanded or combined with other (non-)orthogonal access solutions [65-68]. These recent works evince the importance of non-orthogonal massive access, primarily when associated with heterogeneous resource management techniques.
Distributed Antennas and Networks. Distributed antenna system (DAS) comprises geographically separated antennas interconnected to a central processing unit, which handles the signal processing. DAS increases diversity in the network [69]. This theoretical concept permeates the wireless communications community since the early 2000s. However, only recently, it is becoming a reality in practice due to many advancements in signal processing, distributed systems, cloud and edge computing [70-73].
In a cell-free DAS5, known as cell-free massive multiple-input multiple-output (mMIMO), the access points (APs) (equipped with massive number of antennas) connect via backhaul to central processing units. These processing nodes may belong to a centralized or distributed architecture, for instance [73] evaluates pros and cons of each approach. Together these interconnected APs serve the devices under the coverage area [70], thus reducing interference and achieving higher data rates.
A pivotal distinction to previous solutions is scalability and the ability to [72] 1. provide (almost) uniformly high signal-to-noise ratio (SNR) and data rates across the network; 2. manage interference joint processing multiple APs compared to other solutions that face inter-cell interference; 3. increase SNR via coherent transmissions since APs with weaker channels contribute to joint processing.
Hardware imperfections at the user side and backhaul availability may hamper these gains and are yet open challenges. Incorporating backhaul capacity and processing ability at the edge becomes critical as well [73]. Advances in this area may lead to further improvement in network access, processing time, and energy efficiency.
Short Length Codes. Legacy communication systems design and optimization rely on channel capacity, which assumes infinite long block lengths. Thanks to the law of the large numbers, a code averages out the stochastic variations imposed by the wireless channel. The notion of capacity is a reasonable benchmark in many practical systems since packets comprise a long block of several bits (in the order of thousands of bits). This fundamental result arose in the seminal work of Claude Shannon [74].
However, in many massive IoT applications, the information payload of the message is relatively short, only a few bytes. In this case, current state-of-the-art codes perform poorly, which results in wastage of resources and poor performance.
Only recently, new information-theoretic results appeared related to the performance of short-block length communication. These results show that the achievable rate depends not only on the channel quality of the communication link but also on the actual block length error probability tolerable at the receiver [75]. Please refer to Appendix A for an overview of finite block length coding.
Cellular systems rely on low-density parity-check and polar codes, however tuned to serve broadband users. mMTC on the other hand demands many changes on, for instance, 1. the optimization of coding schemes for a short block of length - ranging from a few hundred to a thousand bits; 2. robust error detection in order to avoid the need for outer codes; and 3. novel decoding algorithms able to work with limited or even without CSI at the receiver.
In addition, in the context of WET, the message block length impacts the energy and information reliability, thus raising an interesting trade-off between energy-information reliability and resource allocation [76, 77].
CSI Free Solutions. Traditionally, CSI is used to compensate the impairments imposed by the wireless channel while consequently improving the communication performance. However, CSI acquisition becomes costly in massive IoT due to number of devices, heterogeneous traffic patterns, latency requirements [78]. In light of these constraints, novel CSI-free and CSI-limited (when, for instance, only long term statistics are available) solutions are needed taking also into account environmental conditions and side information to spare energy, time and spectrum resources.
The subsequent chapters carefully discuss promising CSI-free solutions for massive IoT.
Energy Harvesting, Zero-energy Devices and Backscatter. Incorporating energy harvesting (EH) capabilities into the MTDs will significantly impact the next generation of mMTC since MTDs will operate batteryless [78, 79]. Zero-energy devices will require a rethinking of the air interface and network architecture. Pundits forecast more than 40 years battery life and even continuous battery-less operation of zero-energy devices [35, 80].
Whenever the energy transfer happens through radio frequency (RF), the devices can either harness the ambient energy from neighboring transmissions, known as ambient energy from neighboring transmissions, known as ambient RF–EH (see Chapter 3) or be served by a dedicated carrier, know as WET. The former is discussed within Chapter 2, while the latter is the scope of this book whose objective is to promote the most recent, and in our view, the most promising solutions towards sustainable ICT in massive IoT networks.
1.5 Final Remarks and Discussions
In this chapter, we discussed the exponential growth of the IoT in the recent years, and its potential to impact complete value chains across many sectors, societal and industrial alike. We highlighted how wireless connectivity is instrumental to these changes. Moreover, we introduced and discussed selected use-cases, requirements and KPIs, key technologies and enablers for building sustainable ICT. Before concluding, let us discuss further a fundamental issue in massive IoT, and how it is addressed in this book.
The ‘Power Problem’ in Massive IoT. The energy use and demands of a MTD may be small, in the order of μW to mW. Nonetheless, powering a large number of devices in the network poses a herculean task. Even though most of the devices are battery-powered and current connectivity