Machine Learning for Underwater Optical Wireless Communication Systems: A Comprehensive Survey

Shaymaa Mahmoud; Ardimas Purwita; Mohamed-Slim Alouini

eess.SPSep 14, 2026

Machine Learning for Underwater Optical Wireless Communication Systems: A Comprehensive Survey

Authors: Shaymaa Mahmoud, Ardimas Purwita, Mohamed-Slim Alouini

Abstract

Underwater Optical Wireless Communication (UOWC) has emerged as a promising technology for high-speed underwater data transmission, offering significantly higher bandwidth and lower latency compared to acoustic and Radio Frequency (RF) technologies. However, the underwater medium introduces severe impairments that degrade link performance and limit communication range. The growing complexity of these challenges has increased research interest in Machine Learning (ML) and Deep Learning (DL) approaches, which offer powerful tools for channel modeling, signal processing, and system adaptation in ways that conventional analytical methods struggle to achieve. This survey provides a comprehensive review of ML methods applied across the full UOWC system pipeline, covering channel modeling, transmitter design, receiver detection and equalization, link alignment, and emerging applications, including wireless power transfer, semantic communication, object detection, optical sensing, and localization. Finally, we discuss open challenges and future research directions to motivate further work in this rapidly growing field.

Figures & tables

Abbreviation	Definition	Abbreviation	Definition
AE	Autoencoder	MLP	Multilayer Perceptron
ANN	Artificial Neural Network	NN	Neural Network
APD	Avalanche Photodiode	OAM	Orbital Angular Momentum
APT	Acquisition, Pointing, and Tracking	OFDM	Orthogonal Frequency Division Multiplexing
AUV	Autonomous Underwater Vehicle	OOK	On-Off Keying
BER	Bit Error Rate	OWC	Optical Wireless Communication

TABLE I: Principal Abbreviations Used in This Survey

Fig. 1: General architecture of an underwater optical wireless communication system. Information is encoded, modulated, and converted into optical signals by an LED /laser transmitter. The optical beam propagates through the UOWC channel, where absorption, scattering, and turbulence degrade the signal, after which photodetection, demodulation, and decoding are performed to recover the transmitted information.

Fig. 2: Organization of the survey. After the fundamentals of UOWC and ML (Section II ), Sections III – VII survey learning-based methods component by component along the UOWC pipeline, followed by open challenges and future research directions (Section VIII ) and the conclusion (Section IX ).

Panel of Table 4 — TABLE II: Comparison of Related Surveys

Pipeline stage	Reference	ML family	Data source	Evaluation
Channel modeling & estimation	Zhao et al. [ 28 ]	CNN (AlexNet)	Experimental QAM constellation diagrams	Experiment
	Al-Amodi et al. [ 29 ]	CNN	Simulated received signal samples	Simulation
	Shashikanth et al. [ 30 ]	CNN - LSTM	Bubble turbulence (15 classes)	Experiment
	Lu et al. [ 31 ]	CNN + DNN	Harbor, clear, and mixed water channels	Simulation and experiment
	Jia et al. [ 32 ]	SRCNN + DnCNN	Experimental OFDM FD- UOWC channels	Experiment
	Bo et al. [ 33 ]	CNN - AE hybrid	EGG turbulence model	Simulation

TABLE III: Classification of Surveyed Machine Learning Approaches According to UOWC System Components

Pipeline stage	Reference	ML family	Data source	Evaluation
Link alignment & adaptive control	Jia et al. [ 67 ]	YOLOv8s + RTMPose-t	ULDB image dataset	Experiment
	Lu et al. [ 68 ]	CNN	$2\times 2$ MIMO UOWC	Simulation and experiment
	Tanaka et al. [ 69 ]	CNN	Twin-beam images (2.4 m tank)	Experiment
	Ishida et al. [ 70 ]	LSTM + DQN	Wave displacement data	Simulation
	Romdhane & Kaddoum [ 71 ]	Q-learning + SARSA	4 water types	Simulation
	Shin et al. [ 72 ]	TPTA- DRL (3 agents)	USV trajectory data	Simulation

TABLE III: Classification of Surveyed Machine Learning Approaches According to UOWC System Components (Continued)

Parameter	Acoustic	RF	UOWC
Data Rate	$\sim$ kbps	$\sim$ Mbps	$\sim$ Gbps
Distance	Up to km	$\leq$ 10 m	10–100 m
Latency	High	Moderate	Low
Frequency Band	10–15 kHz	30–300 Hz (ELF) / MHz range (buoyant)	$10^{12}$ – $10^{15}$ Hz
Key Limitation	Low data rate, high latency	Severe seawater attenuation	Limited range, misalignment sensitive

TABLE IV: Comparison of Underwater Wireless Communication Technologies

Fig. 3: Unified architecture of an ML-enabled UOWC system. Environmental and hardware impairments affect the optical communication pipeline, while learning modules support channel modeling, transceiver optimization, signal recovery, link adaptation, beam control, and AUV positioning.

Fig. 4: CNN architecture operating on digitized received signal samples.

Fig. 5: RL agent-environment interaction loop for UOWC beam adaptation.

Fig. 6: CGAN framework for UOWC channel emulation: the generator synthesizes received signals conditioned on the transmitted input, while the discriminator distinguishes generated from real received signals.

Fig. 7: AE Transceiver Architecture. The transmitter learns a compact signal representation that propagates through the underwater optical channel, while the receiver reconstructs the transmitted information.

Fig. 8: Conceptual framework of semantic-aware transmission in UOWC systems. Semantic features extracted at the transmitter cross the underwater optical channel, which introduces path loss, turbulence, and noise, and are decoded into the reconstructed meaning at the receiver.

Reference	Model	Dataset	CR	Reported PSNR	Reported SSIM
[ 48 ]	ResNet + LSTM	EUVP	1/16	26.92% improvement	51.90% improvement
[ 49 ]	Swin Transformer	EUVP, UIEB, RUIE	–	2.09 dB gain (BD-PSNR)	0.0864 gain (BD-SSIM)
[ 51 ]	Swin Transformer	EUVP, UIEB, RUIE	1/6	25.462 dB	0.9417
[ 50 ]	VAE	EUVP, UIEB, LSUI	1/48	22.57 dB	0.609
[ 52 ]	ConvSC	UFO-120	1/16	$\sim$ 25.8 dB	$\sim$ 8.7 dB MS-SSIM
[ 53 ]	RQ - VAE	EUVP	–	28.89 dB	–

TABLE V: Performance Comparison of Semantic Communication Systems for Underwater Optical Wireless Communications

Fig. 9: Architecture of the DBN demodulator. Three stacked RBMs perform layer-wise unsupervised feature extraction, followed by a finetuning layer with supervised backpropagation.

Reference	Method	Signal Format	Main Result	Evaluation
[ 59 ]	Shallow DNN	$2\times$ oversampled OOK	Joint equalization and decoding; 2.3 dB and 1.1 dB received-power gains over threshold detection and maximum likelihood estimation.	Experimental
[ 62 ]	BGRU	64- APSK	Maximum Q-factor improvement of 2.4 dB and a transmission rate of 3.19 Gbps.	1.2-m experimental link
[ 64 ]	DeepESN	PAM 4 and QAM - OFDM	Outperformed recursive LS equalization and trained only the output weights using SVD .	40.5-m experimental link
[ 60 ]	TCN-LSTM-AM	PAM and QPSK - OFDM	Achieved 97.99% accuracy; the source reports a BER -performance improvement of up to 71.4% at 130 m relative to its reference results.	Simulation
[ 61 ]	CNN	Multiplexed OAM	Directly decoded intensity maps and outperformed the Gerchberg–Saxton algorithm under moderate and strong turbulence.	Numerical simulation
[ 66 ]	semiTNN	PAM 4	Reached 4.26 Gbps using only 5–10% labeled data.	14-m and 56-m experiments

TABLE VI: Neural Network Equalizers for UOWC

Fig. 10: Conceptual architecture for joint system-level optimization in ML-enabled UOWC. The cross-layer controller coordinates semantic encoding, adaptive modulation, optical transmission, beam steering, power allocation, and receiver processing across the communication pipeline.

UOWC Stage	Simulation	Experiment	Simulation and Experiment	Emulated UOWC	Total
Channel	4	6	1	0	11
Transmitter	9	2	1	3	15
Receiver	6	7	1	0	14
Link alignment	6	3	3	0	12
Emerging applications	7	9	0	0	16
Total	32	27	6	3	68

TABLE VII: Validation Modes Reported Across the Surveyed UOWC Pipeline Stages

Explore similar work

Mar 8, 2026

Machine Learning for the Internet of Underwater Things: From Fundamentals to Implementation

The Internet of Underwater Things (IoUT) is becoming a critical infrastructure for ocean observation, marine resource management, and climate science. Its development is hindered by severe acoustic attenuation, propagation delays far exceeding those of terrestrial wireless systems, strict energy constraints, and dynamic topologies shaped by ocean currents. Machine learning (ML) has emerged as a key enabler for addressing these limitations, offering data driven mechanisms that enhance performance across all layers of underwater wireless sensor networks. This tutorial survey synthesises ML methodologies supervised, unsupervised, reinforcement, and deep learning specifically contextualised for underwater communication environments. It outlines the algorithmic principles of each paradigm and examines the conditions under which particular approaches deliver superior performance. A layer wise analysis highlights physical layer gains in localisation and channel estimation, MAC layer adaptations that improve channel utilisation, network layer routing strategies that extend operational lifetime, and transport layer mechanisms capable of reducing packet loss by up to 91 percent. At the application layer, ML enables substantial data compression and object detection accuracies reaching 92 percent. Drawing on 300 studies from 2012 to 2025, the survey documents energy efficiency gains of 7 to 29 times, throughput improvements over traditional protocols, and cross layer optimisation benefits of up to 42 percent. It also identifies persistent barriers, including limited datasets, computational constraints, and the gap between theoretical models and real world deployment. The survey concludes with emerging research directions and a technology roadmap supporting ML adoption in operational underwater networks.

Kenechi Omeke, Attai Abubakar, Michael Mollel +3

James Watt School of Engineering, University of Glasgow, Glasgow, United Kingdom

Internet of Things Wireless Communications

Ask PDF

Oct 12, 2026

PRB-WM: Physics-Regularized Beta World Models for Predictive Link Adaptation on Real Time-Varying Underwater Acoustic Channels

Adaptive modulation and coding (AMC) improves the spectral efficiency of underwater acoustic (UWA) links, but its feedback is inevitably stale: sound propagates slowly, and by the time channel state information (CSI) arrives, the fast time-varying multipath channel has moved on. Predictive link adaptation therefore needs more than a point forecast: it needs a distribution over each candidate action's consequences. We propose PRB-WM, a physics-regularized Beta world model for predictive per-subcarrier link adaptation on real time-varying UWA channels. A persistence-anchored stochastic latent-space dynamics model learns how the channel evolves; a counterfactual outcome head evaluates, for every candidate modulation and coding scheme (MCS), a Beta-binomial distribution of bit errors on shared imagined futures; and a chance-constrained receding-horizon model predictive controller maximizes expected spectral efficiency subject to a bounded probability of violating the bit error rate (BER) target. To the best of our knowledge, PRB-WM is the first world model for UWA communications. In closed-loop replay of six measured sea-trial links from the public UWA Channel Library against a 24-method pool spanning the full methodological spectrum, PRB-WM attains a mean goodput 87% above the strongest baseline, ranks first on all six time-varying links at the longest lookahead, and is the only method whose realized frame risk decreases systematically with lookahead, meeting the BER-violation contract where no baseline does. Ablations confirm that rollout sampling dispersion is essential to the chance constraint and that the physics anchor is necessary for trainability.

Shengjie Zhong, Zhongliang Zhao, Jingxuan Chen +5

School of Electronic and Information Engineering, Beihang University, Beijing 100191, China · Department of Strategic and Advanced Interdisciplinary Research, Peng Cheng Laboratory · Hangzhou International Innovation Institute, Beihang University, Hangzhou +8

Wireless Communications Latent World Models

Ask PDF

Apr 18, 2026

E2E-WAVE: End-to-End Learned Waveform Generation for Underwater Video Multicasting

We present E2E-WAVE, the first end-to-end learned waveform generation system for underwater video multicasting. Acoustic channels exhibit 20--46% bit error rates where forward error correction becomes counterproductive -- LDPC increases rather than decreases errors beyond its decoding threshold. E2E-WAVE addresses this by embedding semantic similarity directly into physical layer waveforms: when decoding errors are unavoidable, the system preferentially selects semantically similar tokens rather than arbitrary corruption. Combining VideoGPT tokenization (1024x compression) with a trainable waveform bank and fully differentiable OFDM transmission, E2E-WAVE achieves +5 dB (19.26%) PSNR and +0.10 (14.28%) SSIM over the strongest FEC-protected baseline in less challenging underwater channel (NOF1) while delivering real-time 16 FPS video at 128x128 resolution over 2.3 kbps channels -- impossible for conventional digital modulation. The performance gap only increases in harsher channels (BCH1, NCS1). Trained on a single channel, E2E-WAVE generalizes to unseen underwater environments without retraining, while HEVC fails at sub-5 kbps rates and SoftCast's AWGN assumptions collapse on frequency-selective channels.

Khizar Anjum, Tingcong Jiang, Dario Pompili

Department of Electrical and Computer Engineering, Rutgers University–New Brunswick, NJ, USA

Semantic Communication Wireless Communications

Ask PDF