PaperlayerView exact revision on arXiv

Source-linked AI summary

Internet of Underwater Things and Big Marine Data Analytics -- A Comprehensive Survey

Mohammad Jahanbakht, Wei Xiang, Lajos Hanzo, Mostafa Rahimi Azghadi

arXiv:2012.06712v2cs.NI

TL;DR

The paper addresses the combined challenges of IoUT communications and the processing of large, heterogeneous Big Marine Data in harsh underwater environments. It surveys underwater data collection, communication technologies, BMD handling, and ML/DL analytics, critically appraising research across these areas. Its supported conclusion is a comprehensive synthesis of IoUT and BMD processing research together with open issues and future directions.

  • Problem

    IoUT communication is difficult in harsh, bandwidth-limited underwater environments, while growing marine data creates storage, transport, preparation, and analysis challenges that conventional processing cannot adequately handle.

  • Method

    The paper conducts a comprehensive survey spanning underwater communications and channel modeling, BMD acquisition and processing, and ML/DL methods for marine data analytics.

  • Results

    The survey synthesizes state-of-the-art IoUT communications, BMD system components, machine intelligence applications, and practical solutions across hardware and software research.

  • Takeaways & Limitations

    The paper identifies open research issues and future directions for IoUT, BMD processing, and prospective underwater applications.

Abstract

from arXiv · show

The Internet of Underwater Things (IoUT) is an emerging communication ecosystem developed for connecting underwater objects in maritime and underwater environments. The IoUT technology is intricately linked with intelligent boats and ships, smart shores and oceans, automatic marine transportations, positioning and navigation, underwater exploration, disaster prediction and prevention, as well as with intelligent monitoring and security. The IoUT has an influence at various scales ranging from a small scientific observatory, to a midsized harbor, and to covering global oceanic trade. The network architecture of IoUT is intrinsically heterogeneous and should be sufficiently resilient to operate in harsh environments. This creates major challenges in terms of underwater communications, whilst relying on limited energy resources. Additionally, the volume, velocity, and variety of data produced by sensors, hydrophones, and cameras in IoUT is enormous, giving rise to the concept of Big Marine Data (BMD), which has its own processing challenges. Hence, conventional data processing techniques will falter, and bespoke Machine Learning (ML) solutions have to be employed for automatically learning the specific BMD behavior and features facilitating knowledge extraction and decision support. The motivation of this paper is to comprehensively survey the IoUT, BMD, and their synthesis. It also aims for exploring the nexus of BMD with ML. We set out from underwater data collection and then discuss the family of IoUT data communication techniques with an emphasis on the state-of-the-art research challenges. We then review the suite of ML solutions suitable for BMD handling and analytics. We treat the subject deductively from an educational perspective, critically appraising the material surveyed.

I. INTRODUCTION

IoUT extends IoT into underwater environments, where constrained communications, energy, and computation meet rapidly growing marine data volumes. This survey addresses these challenges by synthesizing IoUT architecture and communications, BMD acquisition and processing, and machine-learning analytics.

  • IoUT motivation: Underwater networks remain comparatively unexplored because applications are still in their infancy and monitoring technologies are underdeveloped.Sparse sensor deployment across large operational areas further limits knowledge of the underwater environment.
  • IoUT motivation: IoUT applies interconnected-device concepts to underwater environments but faces distinct communication, computational, and energy constraints.These differences motivate dedicated discussion of underwater communications, sensors, and devices.
  • Big Marine Data: IoUT ecosystems generate enormous data volumes, while conventional processing techniques struggle to generalize learned behavior and features to big-data scenarios.The paper presents machine learning as a practical approach for automatic knowledge extraction from such data.
  • Survey scope: The survey fills a stated gap by jointly covering IoUT, Big Marine Data, and machine/deep-learning analytics in a comprehensive overview.It contrasts its coverage with earlier surveys addressing IoT, big data, IoUT, or BMD separately.
  • Survey organization: The paper progresses from underwater data collection and networking to BMD analytics and processing, then reviews machine-learning techniques and research challenges.Its organization covers IoUT, BMD sensors and databases, ML-based data leveraging, and future opportunities.

2) Acoustic signal attenuation:

Underwater communications are constrained by attenuation, narrow bandwidth, and dynamic channel conditions, so link reliability depends on technology choice and operating parameters. The surveyed metrics and comparisons frame these constraints for reliable Big Marine Data delivery.

  • Acoustic channel constraints: Acoustic links are commonly used underwater because low-frequency acoustic attenuation is lower than electromagnetic attenuation, although their frequency bandwidth remains narrow.Underwater channel dynamics also affect signal amplitude, frequency, propagation speed, and delay jitter.
  • Optical signal attenuation: Optical attenuation is caused mainly by absorption and scattering, with radiative-transfer solvers offering precision at higher computational cost than simplified Monte Carlo models.The surveyed finite-difference improvement enhances the solution of the radiative-transfer equation but remains computationally expensive.
  • Reliability metrics: BER, RNP, and ETX depend on transmitted data-rate, transmitted power, and distance between consecutive transceiver pairs.These metrics support evaluation of link reliability at bit, packet, and transmission-count levels.
  • Reliability requirements: Reliable data flow requires minimum received power in random background noise, while attenuation and band-limited channels constrain achievable communication performance.The surveyed data-rate and transmission-range characteristics are summarized for underwater communication technologies in Table III.
  • Technology trade-offs: Reliable underwater communication is restricted to low-data-rate acoustic waves for long distances or high-data-rate optical rays for short distances.Short-distance, low-data-rate electromagnetic waves are substantially outperformed by acoustic and optical technologies.

1) Routing improvement:

The surveyed routing improvements target reliable and energy-conscious communication in restrictive underwater channels. They use topology, hop-count, relay, and storage strategies to improve reliability while addressing battery and bandwidth constraints.

  • Routing improvement: A multi-cluster UWSN scheme divides the network into sub-regions, with relay nodes cooperating with cluster-heads and an upper cluster coordinator.The scheme is presented as adaptable to IoUT nodes.
  • Routing improvement: Location-free, energy-based clustering aims to increase SNR reliability, while in-network data storage avoids retransmitting identical packets.The combined protocol tends to send fewer packets with higher reliability and balances routing load.
  • Routing improvement: An alternative cooperative routing scheme lets each normal node independently select a relay and cluster-head using SNR, hop-count distance, and packet arrival time.Unlike the multi-cluster approach, it does not divide the UWSN into multiple clusters.
  • Hop-count optimization: Hop-count optimization must balance reliable longer-distance transmission against reduced bandwidth and increased power consumption in battery-limited nodes.Relay deployment can reduce hop-distance and support energy-efficient, high-data-rate delivery.
  • Hop-count optimization: IoUT architecture must accommodate low bandwidth, acceptable latency, packet lengths matched to channel coherence times, signal attenuation, application-specific bit error rates, and security threats.The TCP/IP model is used as a general framework for matching IoUT architectures across underwater and overwater endpoints.

1) Underwater application layer:

The IoUT architecture adapts layered Internet protocols and communication technologies to underwater constraints, emphasizing efficient data exchange, routing, scheduling, and energy management.

  • Underwater application layer: The application layer identifies underwater objects, gathers and processes sensor data, and delivers commands.Its data-gathering functions include sensing, tracking, recording, and live-data streaming.
  • Underwater application layer: Many conventional IoT application protocols are unsuitable underwater because they rely on wideband operation or redundant headers.Protocols discussed as suitable alternatives include IMC, SLIP, XML, MQTT, DDS, Modbus, and Telnet.
  • Transport layer: The transport layer splits application data into packets while accounting for data order and potential loss.TCP provides reliability through handshaking, error detection, and correction, but its overhead increases channel-capacity requirements.
  • Network and transport layers: IoUT transport access can use multimodal acoustic, electromagnetic, and optical technologies, while the network layer handles packet exchange, protocol translation, IP, and routing.IPv4 is favored over the more verbose IPv6, and ECN is recommended with DCCP for end-to-end congestion notification.
  • Routing and energy management: Underwater routing and scheduling must extend communication range and manage scarce energy across endpoint, mid-layer, and sink nodes.Longer transmission distances reduce effective bandwidth and increase power consumption for payload delivery at minimum SNR.
  • Data link layer: MAC scheduling coordinates access to shared underwater channels, with collision-free protocols generally favored over contention-based approaches.An energy-aware scheduling method offers reasonable throughput but requires time-consuming initialization and regular probabilistic-state updates.

E. Underwater Channel Modeling

Underwater channel modeling is central to IoUT simulation and must be tailored to communication technology, environmental conditions, and application requirements. The section connects channel models with topology design, simulation, and validation constraints.

  • Channel-model design: Underwater channel behavior varies across propagation modes and channel types, requiring models appropriate to the selected technology and application.Universal accurate models are not feasible; practical models simplify environmental factors according to problem requirements.
  • Environmental effects: Channel-model studies incorporate specific environmental features such as non-isovelocity sound-speed layers, scattering particles, and seabed slopes.One scattering study is limited by its two-dimensional vertical cross-section and omission of stochastic particle sizes.
  • Acoustic channel modeling: Equation (11) models the acoustic channel as a superposition of delayed, phase-shifted propagation paths with stochastic gains and losses.The stochastic terms represent effects including scattering, Doppler shifting, node-location uncertainty, received-power changes, and dynamic seabed topology.
  • Communication technologies: Acoustic superposition can also model electromagnetic channels, while this survey focuses on line-of-sight optical and fiber-optic communications for high-data-rate underwater links.Fiber-optic channels are outside the article’s scope, whereas line-of-sight optical channel properties are discussed later.
  • Network topology: Tree and mesh topologies dominate underwater applications because bandwidth is limited and energy is difficult to harvest.Tree networks are typically used for small networks with one-way protocols, while route reliability can be assessed using overall bit error rate.
  • Network simulation: Simulation tools support network design and testing before deployment, including discrete-event modeling of underwater channels and protocols.Their support ranges from wired and wireless ad hoc networks to cellular and satellite communications, with ns-based tools widely used across institutions.

H. IoUT Network Security

IoUT security addresses eavesdropping, unauthorized access, and malicious data manipulation under mobility, computation, communication, and energy constraints. The survey covers cryptographic, covert-communication, secure-protocol, and software-defined approaches.

  • Security threats: Underwater networks are vulnerable to eavesdropping because their physical channels support broad data broadcasting, making secure communication important for harbor security and coastal defense.Threats include passive data extraction and active injection, alteration, or deletion through malicious nodes.
  • Security constraints: Authenticated data access is difficult to implement in IoUT because nodes may be mobile and constrained in communication, computation, and energy.The survey identifies cryptographic primitives and secure communications as approaches for addressing this challenge.
  • Cryptographic techniques: Symmetric encryption reduces key length and associated public-key overhead, while ECC offers lower computation complexity and smaller keys than RSA for equivalent security.Algorithms discussed for underwater use include AES, RC5, RC6, Blowfish, Twofish, and Threefish.
  • Covert communication: Covert communication reduces the average propagation-medium SNR to lower the chances of eavesdropping.Candidate techniques include acoustic phased arrays, frequency hopping, spread spectrum, code division multiplexing, analog network coding, and camouflaged transmission.
  • Secure protocols: Secure protocols span application, transport, network, data-link, and physical layers, including RADIUS, Diameter, SSL, TLS, IPSec, SeFLOOD, RPR, R-CARP, EAP, EAPOL, and MACsec.These layer-specific protocols are presented as distributed network-management methods.
  • Software-defined networking: Software-defined networking centralizes monitoring and configuration, supports spectrum-aware communication, and can detect and mitigate suspicious behavior.Its reported benefits include programmable, reconfigurable, multimodal, scalable, service-oriented, resource-sharing, and troubleshooting capabilities.

J. Edge Computing in IoUT

IoUT edge computing moves processing toward underwater or nearby edge devices to reduce data transfer across hostile, bandwidth-limited links. Its benefits are constrained by scarce and difficult-to-replenish energy, while BMD processing spans acquisition, transport, storage, specialized processing, and exploitation.

  • J. Edge Computing in IoUT: Edge computing performs some or all computations at endpoint devices, reducing required data transfer for IoUT applications.This approach is suited to hostile underwater communication conditions.
  • J. Edge Computing in IoUT: Limited energy resources and the difficulty of delivering sustainable power constrain underwater edge-processing units.These constraints create IoUT-specific challenges beyond conventional IoT edge computing.
  • BMD processing: BMD processing comprises acquisition, secure transportation, storage and privacy, special-purpose processing, and exploitation.Special-purpose processing includes searching, preprocessing, recognition, labeling, visualization, and updating.
  • BMD processing: Data acquisition proceeds through gathering, aggregation, and fusion, using vehicles equipped with primary sources such as cameras, hydrophones, and sensors.ROVs and AUVs together appeared in about 95% of relevant ML publications surveyed.
  • BMD processing: More than 50% of underwater ML research represented in the survey was published after 2014, except research involving human-occupied vehicles.The paper associates this pattern with wider adoption of automated and remote methods and reduced academic use of costly manned vehicles.
  • BMD processing: Aggregation summarizes raw data before higher-level calculations or transmission, potentially reducing bandwidth use and network-level energy consumption.It trades lower processing power against higher communication energy and may impose drawbacks in underwater applications.

3) Data fusion:

IoUT data fusion integrates heterogeneous marine data, with design choices concerning fusion location, abstraction level, and source overlap. Resource constraints favor cooperative and higher-level fusion, while observatory data require attention to measurement errors and system availability.

  • 3) Data fusion:: Data fusion combines relevant data from different sources into an integrated dataset to obtain more consistent and accurate information.IoUT fusion systems handle heterogeneous sensors, audio, video, and commands using edge devices and cloud servers.
  • 3) Data fusion:: IoUT fusion design concerns where fusion occurs, its abstraction level, and the degree of overlap among original data.Fusion may be centralized at energy-rich overwater or inland nodes or distributed hierarchically across cluster heads.
  • 3) Data fusion:: Fusion operates at low sensor, medium feature, or high decision levels, depending on whether inputs are raw data, extracted features, or model outputs.The high-level inputs are outputs of classification or clustering blocks.
  • 3) Data fusion:: Restricted underwater resources make low-level fusion generally unsuitable, whereas medium- and high-level fusion can support more sophisticated models.Raw-data fusion is excepted when performed on an edge device immediately before feature extraction.
  • 3) Data fusion:: Redundant and complementary overlap can increase confidence but is discouraged in UWSNs because underwater data transmission is expensive; cooperative fusion can increase subject knowledge.Cooperative fusion combines different data types, such as sensed parameters or associated audio and video.
  • 3) Data fusion:: Observatory data can contain environmental noise, outliers, misread values, and missing quantities, requiring dedicated measurement-error handling.Open-access active observatories provide continuously prepared and updated primary data, while some older systems are obsolete or unsupported.

3) Three-dimensional underwater video data:

Underwater video combines severe imaging and communication constraints with rapidly accumulating camera data, motivating automatic processing. Distributed and cloud platforms, especially Apache frameworks such as Spark, support high-speed BMD analytics but impose resource trade-offs.

  • 3) Three-dimensional underwater video data:: Manual annotation requires about 15 expert minutes per video minute, making automatic video processing necessary for large underwater datasets.The paper estimates approximately 10,800 man-hours to analyze one month of video from a single camera.
  • 3) Three-dimensional underwater video data:: Depth-based underwater video uses optical multi-camera systems, acoustic arrays, ToF sensors, and laser beams to provide rich 3D scene information.It is applied in underwater vehicles including ROVs and submersibles.
  • 3) Three-dimensional underwater video data:: Acoustic depth vision faces variable sound velocity, reverberation, noise, and echoes, while optical stereoscopy faces blur, haziness, unstable illumination, and refraction.Suggested responses include improved channel models or a unified opti-acoustic 3D imaging system.
  • Distributed and Cloud-based BMD Processing: Resource scarcity and limited power at collection points shift raw-data processing toward cloud-based or distributed platforms rather than local execution.These platforms address the need for high-performance processing after data gathering.
  • Distributed and Cloud-based BMD Processing: Spark is reported as about 100× faster than Hadoop, supports complex machine-learning processing and real-time batch or streaming workloads, but requires high memory.Apache frameworks including Spark, Hadoop, Storm, and Flink are presented as suitable for high-speed IoUT BMD analytics.

D. Marine Data Applications

IoUT data processing supports maritime, geographic, climate, localization, and marine-life applications. Open GIS resources can be integrated with IoUT infrastructure, while the survey adds biogeographical and georeferenced-location categories to existing marine data classifications.

  • D. Marine Data Applications: IoUT processing produces geographic data for marine-vehicle localization, weather and climate access, and recognition, counting, and distribution of underwater species.These secondary parameters are collectively termed IoUT Geographic Data.
  • D. Marine Data Applications: Open maritime GIS databases support international transportation and can provide tracking, routing, and global ecosystem-monitoring information when merged with IoUT infrastructure.The paper identifies maritime information systems as a key GIS application for accurate geographic data.
  • D. Marine Data Applications: The surveyed GIS classes cover vessel tracking, marine cartography, oceanic climate, and IoUT commerce data.Applications include monitoring vessels and accidents, mapping ports and bathymetry, accessing weather and hazards, and supporting shipping or conservation organizations.
  • D. Marine Data Applications: The paper proposes adding biogeographical data to address the geospatial distribution of underwater species.The open-access OBIS project is connected to more than 500 databases in 56 countries.
  • D. Marine Data Applications: The paper identifies georeferenced locations as a missing marine-cartographic subcategory for precisely associating marine locations with physical maps.The Marine Regions project is cited as an implementation of this navigational-assistance function.

2) Underwater localization:

Underwater localization must replace GPS with techniques suited to acoustic, inertial, and visual sensing in difficult marine environments. The survey reviews these methods, their trade-offs, and ways to combine them for improved positioning.

  • GPS signals do not penetrate seawater, so underwater systems use inertial, acoustic, image-based, and SLAM localization methods.
  • Underwater localization is complicated by long latencies, multipath fading, Doppler shifts, sparse nodes, and high mobility.
  • Blind positioning estimates location from orientation, acceleration, velocity, gravity anomalies, and integrated sensor measurements without ship or transponder support.
  • Blind positioning is power-efficient but accumulates unbounded errors, although integrated sensors can improve exact-position estimation.
  • Acoustic methods estimate position from signal time of flight, while image-based navigation matches seabed images and stereoscopic cameras recover 3D transformations.
  • A fixed geo-referenced beacon reduces an AUV’s positional uncertainty by combining uncertain prior position with uncertain beacon-distance information.
  • Image-based positioning can suffer error propagation and accumulation, motivating integration with SLAM to achieve bounded positioning error.

B. Deep Learning Frameworks and Libraries

The survey presents deep-learning frameworks and the data-preparation stages required for training ML models on underwater data. It emphasizes open-source tools, parallel GPU operation, and the heightened importance of cleansing harsh-environment measurements.

  • Open-source deep-learning frameworks and libraries support architectures such as DBNs, CNNs, autoencoders, and RNTNs for marine data processing.
  • The compared frameworks can operate on NVIDIA CUDA-supported GPUs, while only some support OpenMP shared-memory multiprocessing.
  • Data collection and cleansing precede ML training because underwater environmental factors make clean IoUT data difficult to acquire.
  • IoUT sensors continuously measure physical, chemical, and biological parameters, generating a huge volume of data.
  • Common missing-value treatments include deleting records, using constants or statistical values, and imputing likely or inter-class values.

MXNET ASF

The survey covers underwater sensory and image preparation, from noise cancellation and outlier detection to image enhancement, binning, feature extraction, and dimensionality reduction. These steps adapt general ML preprocessing to degraded underwater observations.

  • Underwater sensor data cleaning: Underwater sensory cleansing addresses missing values, contaminated measurements, noise, and outliers through imputation, filtering, regression, binning, wavelets, and detection criteria.
  • Underwater sensor data cleaning: Low-pass filtering is the most common denoising method, whereas wavelet methods are the most complex to implement; binning is more common for images.
  • Underwater image preparation: Underwater images and videos require quality assessment because absorption, scattering, chromatic aberration, and light-source noise degrade captured data.
  • Underwater image preparation: Software-based enhancement is generally cheaper than bespoke imaging hardware, and methods include dehazing, color enhancement, and contrast improvement.
  • Underwater image data binning: Image binning groups similar pixels into partitions to reduce noise and complexity, and is recommended before feature extraction to speed ML processing.
  • Underwater image data binning: No image-partitioning method is universally accepted, so selecting an appropriate clustering method remains challenging for each application.
  • Feature extraction and reduction: Feature extraction converts high-dimensional sensory vectors and image matrices into low-dimensional numerical descriptors tailored to the project.
  • Feature extraction and reduction: Dimensionality reduction evaluates feature correlation and uses reduction, selection, or aggregation methods such as PCA and LDA.

E. Hardware Platforms for ML in IoUT

ML in IoUT can run on shared-memory processors or distributed computing systems using data, model, or pipeline parallelization. The survey compares hardware choices, emphasizing GPUs for speed and compact ASIC/FPGA platforms for efficient deployment.

  • Processing architectures: ML implementations can use shared-memory multiprocessors or distributed computing systems, with throughput increased through parallel processing.
  • Processing architectures: Distributed ML supports data parallelization, model parallelization, and pipelined parallelization across networked processors.
  • Feature-oriented processing: The survey catalogs underwater ML feature sets spanning color, texture, shape, image statistics, key points, and 3D descriptors.
  • GPU platforms: GPUs provide affordable, high-speed processing for underwater ML applications but traditionally depend on bulky host computers.
  • GPU platforms: Compact NVIDIA Jetson systems provide high performance, low power, and low-latency inference suitable for underwater vehicles and platforms.
  • ASIC and FPGA platforms: ASICs and FPGAs offer small form factors, high throughput, and high power efficiency for industry-scale IoUT projects.

2) Deep NNs for dynamic IoUT data:

The paper surveys deep neural networks for dynamic underwater data, image and video processing, object recognition, and prediction. It emphasizes recurrent architectures for sequential sensor streams and compares learning approaches using average precision and mean squared error.

  • Dynamic underwater data: RNNs and variants such as LSTM and GRU model nonlinear dynamic systems using time-series and sequential underwater sensor data.RNNs can construct supervised classifiers from continuous sensor outputs.
  • Image processing: Underwater image enhancement with deep learning addresses noise, absorption, scattering, and color distortion caused by visible-light conditions.These techniques support restoration of degraded underwater imagery.
  • Object recognition: Cascaded classifiers combine complementary detectors or feature sets to improve underwater recognition precision, but some designs require additional hardware resources.Examples include shape- and texture-based classifier combinations and cascaded object-specific SVM detectors for deep-sea megafauna.
  • Object recognition: Average precision is used to compare statistical methods, traditional neural networks, and deep neural networks for underwater object recognition.The referenced comparisons include underwater species and sonar-imagery recognition using established datasets and vehicle-recorded footage.
  • Object recognition: CNN-based sonar-recognition networks can outperform traditional models without feature-extraction preprocessing when appropriately designed and trained.Statistical and traditional networks use HOG features, whereas CNNs do not require that preprocessing.
  • Video applications: Deep neural networks support visible-light and sonar video tracking, while sonar provides long-range, low-data-rate imaging in turbid environments.Tracking requires predicting an object’s next-frame location and detecting it within the predicted region.

G. Section Summary

The section summarizes the paper’s machine-learning survey and then frames IoUT research around unresolved challenges, opportunities, and future directions. It highlights underwater-specific network management, energy constraints, and diverse energy-acquisition options.

  • Section summary: The ML survey classifies methods into classic statistical techniques, traditional neural networks, and modern deep neural networks.It also reviews dynamic and static underwater processing, data cleaning, feature extraction, software frameworks, hardware platforms, and performance measures.
  • Challenges and future directions: The paper surveys IoUT, BMD, and machine intelligence before proposing solutions and future research directions for their challenges and opportunities.The discussion is presented as a synthesis of state-of-the-art research.
  • Challenges and future directions: Terrestrial technologies that perform well for IoT are often unsuitable underwater, where they encounter significant application challenges.The paper therefore treats IoUT, BMD, and ML as domains requiring dedicated investigation.
  • Underwater network management: Growing numbers of connected underwater devices increase IoUT complexity and require automatic, prompt network monitoring and control.Underwater network management must account for the distinctive communication environment and six FCAPSC management aspects.
  • Underwater network management: U-NMS research spans network functionalities and device operations, including routing, protocol assignment, security, maintenance, energy conservation, positioning, and synchronization.The paper surveys these domains and notes limitations in existing protocols such as U-SNMP and LWM2M.
  • Underwater network management: Centralized SDN management is identified as a needed direction for underwater systems, with OVSDB suggested as a design base requiring further security research.The proposed adaptation remains a research opportunity rather than an established solution.
  • Energy management: Energy conservation and harvesting are central IoUT concerns, with options including solar, tidal-wave, wireless, wired, and rechargeable-battery approaches.Cabled energy transfer is costly, while rechargeable batteries prolong network lifetime but increase maintenance and system cost.

D. Large-scale IoUT Underwater Communications

Large-scale IoUT communications are constrained by propagation, bandwidth, energy, reliability, and security limitations. The paper surveys heterogeneous networking, edge processing, routing, deep-learning traffic control, SDN, and cross-layer security as research directions.

  • Communication technologies: Underwater acoustic, electromagnetic, and optical technologies provide difficult trade-offs because they propagate poorly, have limited range or bandwidth, or are costly to deploy.Electromagnetic and optical links are short-range, while acoustic links support longer range but narrow bandwidth and cross-talk.
  • Communication technologies: Combining heterogeneous communication technologies could address IoUT communication limitations, but multi-mode gateway design and energy harvesting remain challenging.SDN and cognitive-radio concepts may help share limited underwater spectrum.
  • Edge processing: Edge processing reduces raw-data volume and bandwidth requirements, while unmanned vehicles can collect data and handle latency-sensitive computations.MEC combined with autonomous underwater or aerial vehicles is presented as an alternative for long-distance communications.
  • Routing and traffic control: Low channel capacity and concurrent transmitters make point-to-point transmission, traffic control, and QoS difficult in large-scale IoUT networks.Routing must balance short-hop delay against long-hop costs while considering low-complexity transceivers and battery charge.
  • Routing and traffic control: Deep learning and SDN are proposed for intelligent traffic control that handles concurrent transfers and avoids congestion.For wide-scale networks, reducing the action space to the next-hop destination addresses the exponential growth of possible paths.
  • SDN and intelligent management: DNNs could support SDN functions including routing, traffic and delay prediction, QoS prediction, resource allocation, spectrum sharing, and intrusion detection.The paper presents these synergies as promising for underwater applications.
  • Security: IoUT security is difficult because communication channels are unreliable, propagation delays are high, and energy resources are limited.Research gaps include protecting application-layer endpoints and distributing security functions across layers to minimize resource consumption.

I. Poor Underwater Positioning and Navigation

Underwater positioning and navigation remain inadequate because standalone techniques accumulate error, while combining heterogeneous data and building training datasets introduces cost, complexity, and precision challenges.

  • Navigation limitations: Standalone underwater positioning techniques do not provide non-accumulating error, making none of them adequate alone.The surveyed alternatives include blind positioning, acoustic transponders, ranging sonars, image-based positioning, and SLAM.
  • Navigation limitations: Combining navigation data requires balancing system cost, complexity, and precision across complementary techniques.Image-based positioning combined with SLAM is cited as one approach for improving navigation.
  • Data requirements: Training localization ML algorithms requires preliminary environmental surveys to extract positional fingerprints, but collecting these large underwater datasets is difficult.The offline survey precedes scene-analysis localization methods such as image-based positioning and acoustic transponders.
  • Navigation limitations: Gravity-aided navigation exploits differences between observed and predicted gravity, but biases and error accumulation remain unresolved.The passage identifies addressing these errors as a future research need.
  • Research opportunities: Adapting LPWAN technologies is proposed as an IoUT localization opportunity because they consume extremely low energy and provide wide reception ranges.The technologies discussed include SigFox, Lo-RaWAN, and Weightless.
Loading 2012.06712v2…