PaperlayerView exact revision on arXiv

Source-linked AI summary

IoT Connectivity Technologies and Applications: A Survey

Jie Ding, Mahyar Nemati, Chathurika Ranaweera, Jinho Choi

arXiv:2002.12646v1eess.SP

TL;DR

The survey addresses how wireless IoT connectivity can support rapidly increasing numbers of devices with diverse service requirements. It synthesizes existing technologies, emerging access approaches, and application classifications, concluding that different connectivity options suit different requirements while massive-access bottlenecks remain.

  • Problem

    Existing wireless IoT technologies face open challenges in accommodating future massive connectivity from sporadic devices with small payloads, including collisions, latency, signalling overhead, and limited orthogonal resources.

  • Method

    The survey reviews existing technologies by coverage range, examines CS, NOMA, mMIMO, and ML-based random access, and classifies applications by technical requirements.

  • Results

    The survey identifies suitable connectivity options for application groups, with ZigBee, Bluetooth/BLE, WiFi HaLow, LTE-M, NB-IoT, Sigfox, and LoRa serving many medium- or low-data-rate applications.

  • Takeaways & Limitations

    CS-based connectivity and grant-free random access are presented as options for reducing signalling overhead, while emerging approaches retain limitations requiring further attention.

Abstract

from arXiv · show

The Internet of Things (IoT) is rapidly becoming an integral part of our life and also multiple industries. We expect to see the number of IoT connected devices explosively grows and will reach hundreds of billions during the next few years. To support such a massive connectivity, various wireless technologies are investigated. In this survey, we provide a broad view of the existing wireless IoT connectivity technologies and discuss several new emerging technologies and solutions that can be effectively used to enable massive connectivity for IoT. In particular, we categorize the existing wireless IoT connectivity technologies based on coverage range and review diverse types of connectivity technologies with different specifications. We also point out key technical challenges of the existing connectivity technologies for enabling massive IoT connectivity. To address the challenges, we further review and discuss some examples of promising technologies such as compressive sensing (CS) random access, non-orthogonal multiple access (NOMA), and massive multiple input multiple output (mMIMO) based random access that could be employed in future standards for supporting IoT connectivity. Finally, a classification of IoT applications is considered in terms of various service requirements. For each group of classified applications, we outline its suitable IoT connectivity options.

I. INTRODUCTION

The survey frames IoT connectivity as a growing challenge requiring technologies matched to coverage, application requirements, and massive device access. It reviews existing and emerging wireless options and classifies applications to identify suitable connectivity technologies.

  • Motivation: IoT connectivity demand is rising as connected devices and sensors proliferate across applications such as smart cities and industrial IoT.More than 8.4 billion connected devices were estimated in use worldwide in 2018, with more than 20.8 billion predicted by 2020.
  • Connectivity scope: Small-area applications can use short-range technologies, whereas environmental sensors and UAVs require long-range connectivity.Examples of short-range options include Bluetooth, Zigbee, WiFi, and OWC.
  • Survey scope: The survey reviews existing connectivity technologies, their specifications, and their fundamental bottlenecks for supporting massive IoT connectivity.It also discusses emerging technologies intended to address these bottlenecks.
  • Emerging technologies: It covers cutting-edge CS, NOMA, mMIMO, and ML-based random access technologies, including their massive-connectivity potential and limitations.These technologies may be employed with existing IoT technologies to enhance performance.
  • Application classification: The survey classifies IoT applications by general requirements and identifies feasible connectivity technologies for each group.The approach considers requirements such as data rate, latency, coverage, power, reliability, and mobility.

2) ZigBee:

This section compares short- and long-range IoT connectivity technologies, emphasizing their differing coverage, data-rate, latency, and device-support characteristics. It also identifies the coverage limitation of short-range approaches and the resource-management challenges associated with LTE and 5G.

  • 2) ZigBee:: ZigBee is a low-power WPAN technology used in home automation, industrial monitoring, and health and aging population care.It operates mainly at 2.4GHz and optionally at 868MHz or 915MHz.
  • 2) ZigBee:: ZigBee connects up to 255 devices with packets up to 128bytes and transmission ranges from a few meters to 100 meters.It supports star and peer-to-peer topologies, with coordinators and routers normally mains-powered and end devices potentially battery-powered.
  • 4) OWC:: OWC uses visible, infrared, or ultraviolet light as a propagation medium for indoor IoT connectivity with high bandwidth and low latency.Its two major categories are visible light communication and beam-steered infrared light communication.
  • 4) OWC:: VLC can provide multi-gigabit wireless connectivity through LED illumination infrastructure and modulation schemes including OOK, QAM, and OFDM.VLC was initially standardized as IEEE 802.15.7 in 2011.
  • Coverage boundary: Short-range technologies support varied data-rate use cases but are unsuitable for applications requiring wide coverage.Long-range technologies are introduced as their counterpart.
  • B. LTE and 5G: Compared with LTE, 5G is described as offering 100x higher data rate, 10x lower latency, and support for 100x more connected devices.The survey also notes the need for flexible resource management and accurate, fast traffic prediction for diverse applications.

C. LPWAN Technologies

LPWAN technologies support wide-area, large-scale connectivity for low-power, low-cost, low-data-rate devices, using either unlicensed or licensed spectrum. The survey reviews LoRa and Sigfox alongside LTE-M and NB-IoT, including their access procedures, deployment trade-offs, and application fit.

  • LPWAN targets wide-area communications and large-scale connectivity for low-power, low-cost, low-data-rate devices with certain delay tolerance.
  • Unlicensed LPWAN uses ISM spectrum and includes LoRa and Sigfox, while licensed LPWAN uses licensed spectrum and includes LTE-M and NB-IoT standardized by 3GPP.Unlicensed deployment reduces spectrum-licensing costs, whereas licensed systems can provide more guaranteed performance through cellular resource allocation.
  • 1) Unlicensed LPWAN: LoRa uses chirp spread spectrum modulation, with spreading factors from 7 to 12 that trade data rate against coverage range, link robustness, and energy consumption.Depending on spreading factor and channel bandwidth, LoRa data rates range from 50bps to 300kbps.
  • 2) Licensed LPWAN: LTE-M is a simplified, LTE-compatible cellular technology for low-cost and low-power IoT applications, supporting mobile MTC use-cases and voice over networks.
  • 2) Licensed LPWAN: NB-IoT uses a 200kHz narrowband built on LTE functionality to support wider coverage, lower device cost, longer battery life, and higher connection density.NB-IoT supports OFDMA downlink and SC-FDMA uplink, half-duplex operation, and a 164dB link budget compared with LTE's 142dB.
  • Short-range technologies serve local IoT applications, while licensed LPWAN technologies support nationwide applications requiring unified connectivity, such as smart-meter deployments.

III. EMERGING WIRELESS TECHNOLOGIES FOR MASSIVE CONNECTIVITY

Emerging connectivity approaches address massive IoT access through compressive sensing, NOMA, and mMIMO, while each introduces implementation or scalability challenges.

  • A. CS based IoT Connectivity: Grant-free compressive random access exploits sparse device activity to detect multiple signals while reducing signaling overhead.Devices transmit preambles and payloads directly, and CS-based schemes use spreading sequences or multiple resource blocks to reduce collisions and support device identification.
  • A. CS based IoT Connectivity: cGFRA complexity grows with the total number of MTC devices, and bandwidth expansion may be required to support more simultaneous devices.These constraints motivate low-complexity designs and complementary technologies such as NOMA and mMIMO.
  • B. NOMA based IoT Connectivity: NOMA overlaps signals on the same time-frequency resource and uses successive interference cancellation to decode devices separately.Power-domain NOMA enables simultaneous grant-free access without bandwidth spreading, while code-domain NOMA uses code-domain multiplexing.
  • B. NOMA based IoT Connectivity: NOMA implementation still requires detection and decoding strategies that increase device-pair capacity while suppressing error propagation.Code-domain designs also require factor-graph optimization to balance overloading factor and receiver complexity.
  • C. mMIMO based IoT Connectivity: mMIMO uses many base-station antennas to create near-orthogonal channel responses and accommodate many devices in the same time-frequency resource.However, supporting massive access may require hundreds or thousands of antennas, making centralized deployment potentially impractical and motivating distributed mMIMO.
  • C. mMIMO based IoT Connectivity: Distributed mMIMO has demonstrated performance superiority over centralized mMIMO in several perspectives, but its potential for massive-access MTC remains lightly studied.Existing work provides only preliminary analyses of grant-free random access in distributed mMIMO.

D. Machine Learning-assisted IoT Connectivity

The survey discusses machine learning for dynamic IoT resource management and classifies applications by user type and technical requirements. It presents these technologies as promising but notes implementation limitations and the need for coordinated future development.

  • D. Machine Learning-assisted IoT Connectivity: Machine learning can address wireless resource-allocation problems that are difficult to model in dynamic IoT environments.The survey describes ML applications including link adaptation, traffic control, resource allocation, and massive-access management.
  • D. Machine Learning-assisted IoT Connectivity: Reinforcement-learning schemes have been applied to base-station selection, congestion avoidance, packet-delay reduction, traffic control, and adaptive access control.Q-learning enables devices to select base stations according to changing traffic conditions and QoS parameters.
  • D. Machine Learning-assisted IoT Connectivity: RL designs must trade computational requirements and energy consumption against learned-model accuracy.Temporally correlated observations can also make convergence time-consuming and unsuitable for highly dynamic environments.
  • IV. CLASSIFICATION OF IOT APPLICATIONS: The paper maps application categories to feasible connectivity technologies based on their service requirements.It presents this approach as an alternative to utilization-domain classifications, which can create overlapping categories.
  • IV. CLASSIFICATION OF IOT APPLICATIONS: The survey classifies IoT applications by end-user type as human-oriented or machine-oriented, then considers data rate, latency, coverage, power, reliability, and mobility.Machine-oriented applications dominate the aggregation and therefore require high connectivity density.
  • D. Machine Learning-assisted IoT Connectivity: CS-based connectivity, NOMA, and mMIMO are identified as emerging technologies for sharing bandwidth among massive numbers of IoT devices.The survey concludes that these technologies may support massive connectivity, high reliability, and low latency, while requiring further development and integration.

A. Human-Oriented IoT Applications

IoT applications are classified by end-user type and technical requirements to map them to suitable connectivity technologies. Human-oriented applications involve human interaction, while machine-oriented applications communicate with minimal human involvement.

  • Human-oriented applications: Human-oriented applications require human interaction with a network and include smartphones, security cameras, and patient surveillance systems.They commonly visualize information intuitively or accept human interaction.
  • Machine-oriented applications: Machine-oriented applications automatically communicate with other devices or remote servers with minimal human involvement.Traditional examples include monitoring sensors, wireless sensor networks, and RFID systems.
  • Overlapping categories: Some applications can be classified as either human-oriented or machine-oriented depending on whether they report risks to people or control medical instruments automatically.Health risk detection sensors illustrate this dual classification.
  • Classification method: The survey maps applications using end-user type together with data rate, latency, coverage, power, reliability, and mobility requirements.These requirements can overlap and create performance trade-offs.
  • Data-rate mapping: High data-rate applications are typically supported by 4G, 5G, optical wireless communication, WiFi, and short-range mmWave technologies.Examples include streaming video, web applications, and smartphones; mmWave systems can provide up to tens of Gbps.
  • Data-rate mapping: Medium- and low-data-rate applications are expected to rely on ZigBee, Bluetooth/BLE, WiFi HaLow, LTE-M, NB-IoT, Sigfox, and LoRa.Low-power consumption is critical for these applications, which are estimated to represent 60% of the market.

2) Latency :

IoT applications are differentiated by latency sensitivity and other service requirements, then mapped to connectivity technologies. The survey emphasizes that technology selection depends jointly on requirements such as coverage, power, reliability, and mobility.

  • Latency: Delay-sensitive applications include autonomous vehicles and healthcare systems, where low latency supports rapid decisions or early treatment.Current 4G and WiFi can provide latency up to tens of milliseconds; current 4G round-trip latency is about 15ms.
  • Coverage: IoT applications span short-range links of up to tens of meters and longer-range links extending to tens of kilometres.Smart-home and smart-retail applications commonly involve connected objects within about 100m.
  • Power: Power efficiency affects IoT device cost, while LPWAN applications may require batteries with service lives of up to 10 years.Agricultural metering sensors exemplify ultra-low-power deployments that cannot be regularly recharged.
  • Reliability: Mission-critical applications include smart grids, manufacturing robots, autonomous vehicles, and mobile healthcare, but they are forecast to form only a small portion of IoT applications by 2024.Most IoT applications are mission non-critical.
  • Technology selection: The mapping is multidimensional: applications can belong to several categories simultaneously, and technology assignments are not always unilateral.Smart agricultural sensors are characterized as machine-oriented, low-data-rate, delay-tolerant, long-range, low-power, non-critical, and low-mobility.
  • Technology selection: For smart agricultural sensors, LPWAN technologies such as LoRa and NB-IoT are identified as suitable connectivity options.The classification is summarized in Table VII to support application categorization and technology selection.
  • Emerging connectivity: Emerging CS-based and grant-free random-access protocols can reduce signaling overhead, but their complexity in high-density machine-type communications remains open.NOMA detection and interference cancellation also remain challenging in high-density MTC, while mMIMO is discussed as an interference-mitigation approach.

APPENDIX A ALOHA

Slotted ALOHA lets synchronized nodes probabilistically share a common channel, with successful transmission requiring no simultaneous competitors. Its throughput peaks under a specific offered-load condition, but stability remains a concern.

  • Operation: Slotted ALOHA divides time into packet-length slots and assumes synchronized nodes communicating with a receiver station.Each node attempts transmission in a slot with access probability p.
  • Transmission model: A transmitted packet succeeds only when no other node transmits simultaneously; two or more simultaneous transmissions cause a collision and no successful packet.Throughput denotes the number of nodes that successfully transmit packets.
  • Throughput: For large K, throughput is approximated by xe^-x, where x = Kp, and reaches its maximum e^-1 at x = 1.The corresponding access probability is p = 1/K.
  • Coordination: Slotted ALOHA requires synchronization beacons and feedback signals so nodes learn whether transmissions succeeded.The receiver periodically broadcasts beacon signals and ACK or NAK feedback at the end of a slot.
  • Stability: Distributed slotted ALOHA can suffer buffer overflow from frequent collisions, requiring access-control and retransmission strategies to stabilize buffer length and access delay.Buffer length is proportional to access delay in the described model.

APPENDIX B CSMA

CSMA protocols sense a shared channel before transmission, while CSMA/CD detects collisions and CSMA/CA avoids them through reservation exchanges. Wireless operation favors CSMA/CA because a transmitting node cannot sense the channel simultaneously.

  • CSMA variants: CSMA requires a node to verify that a common channel is free before transmitting, with variants including CSMA/CD and CSMA/CA.The protocols differ in how they handle simultaneous transmissions and collisions.
  • CSMA/CD: CSMA/CD aborts simultaneous transmissions after collision detection and a jamming signal, shortening collision duration and potentially improving throughput.It is generally used in wired networks where nodes can sense while transmitting.
  • CSMA/CD: CSMA/CD retransmission strategies include nonpersistent, 1-persistent, and p-persistent operation.These strategies differ in whether nodes wait randomly, transmit immediately when idle, or transmit with probability p.
  • CSMA/CA: Wireless nodes cannot sense the channel while transmitting, so CSMA/CA uses strategies intended to avoid collisions.An inter-frame space helps account for signals that have not yet reached the sensing node.
  • CSMA/CA: CSMA/CA reserves transmission using RTS and CTS exchanges before sending data.RTS requests the right to send, while CTS grants that right when the receiver is ready.
  • CSMA/CA: The standard CSMA/CA sequence is channel sensing, RTS, CTS, data transmission, and ACK, with exposed nodes deferring during NAV periods.The described mechanism is shown in Fig. 11 and is associated with collision-free transmission in the passage.

APPENDIX C CS

The appendix explains how compressive sensing recovers sparse signals and applies that framework to random access, where sparse user activity enables active-user detection with non-orthogonal signatures. It also highlights computational and bandwidth constraints that limit straightforward approaches.

  • Compressive sensing addresses sparse-signal recovery, including its application to random access.The appendix introduces sparse-signal recovery before connecting compressive sensing to random access.
  • M < L measurements provide dimensionality reduction but make recovery impossible without suitable conditions on the signal and measurement matrix.C is an M × L measurement matrix, and the resulting system is underdetermined.
  • The maximum-likelihood approach enumerates Lq possible supports, making its computational complexity potentially prohibitive for large L.The appendix notes that exhaustive support consideration yields complexity proportional to Lq.
  • Because the ℓ0-norm formulation is nonconvex, p-norm generalizations and relaxed constraints provide convex alternatives when p ≥1 or noise is present.For sparse solutions, p ≤1 is desirable; p = 1 remains convex, whereas p = 2 is not sparse in the underdetermined setting.
  • In compressive random access, sparse activity among L users is detected from non-orthogonal signature sequences of length M, supporting large L for fixed M.When only a few users are active, the activity vector is sparse; this enables detection when L > M.

APPENDIX D NOMA

The appendix describes power-domain NOMA as shared-channel transmission with non-orthogonal access, power allocation, and successive interference cancellation. It covers downlink decoding constraints, uplink superposition, and the resulting sum-rate property.

  • NOMA uses non-orthogonal multiple-access channels, with power-domain NOMA as the most popular form.Unlike TDMA and FDMA, multiple users share non-orthogonal access resources.
  • Power-domain NOMA superposes signals at different power levels and commonly pairs a near strong user with a far user.The near user can decode and remove the far-user signal through successive interference cancellation.
  • Downlink rate constraints combine successful far-user and near-user decoding requirements and guide power allocation.The combined constraint is identified as playing a key role in power allocation.
  • In uplink NOMA, the base station receives a superposition of user signals and can reduce joint-decoding complexity using successive interference cancellation.When one received signal power is much larger than the other, the stronger signal is decoded first and removed.
  • Power-domain NOMA is also optimal in sum rate because its sum rate equals the achievable multiple-access-channel rate.

APPENDIX E MMIMO

The appendix presents massive MIMO as a many-antenna system serving many users on shared time-frequency resources, with favorable propagation simplifying processing. As antenna and user dimensions grow, interference can vanish while spectral and energy efficiency improve.

  • Massive MIMO uses hundreds or thousands of base-station antennas to serve tens or hundreds of users over the same wireless resource.Time-division duplexing is favored because channel reciprocity can be exploited within a coherence interval.
  • When M ≫K, favorable propagation makes user channel vectors mutually orthogonal or quasiorthogonal, enabling simple linear processing.Conjugate beamforming and zero-forcing beamforming are cited as examples.
  • As M →∞, multiuser interference and noise can be eliminated in massive MIMO under the law of large numbers.
  • The asymptotic SINR and sum-rate expressions characterize massive-MIMO performance when M and K grow with a fixed ratio.
  • Increasing M without increasing transmitted power can raise per-user throughput and serve more users simultaneously, while meeting a target throughput with less power.The appendix associates large M and K with high spectral and energy efficiency.
Loading 2002.12646v1…