📄 other

Channel-Aware Adaptive Hybrid RF/FSO/THz Maritime Wireless Networking

This paper proposes and evaluates a channel-aware adaptive framework for hybrid RF/FSO/THz maritime networks that utilizes physics-based modeling and machine learning to optimize channel selection, while highlighting the critical need for diverse training scenarios to ensure robustness against environmental variations and measurement errors.

Original authors: Luis Miguel Pires, Dushan Nalin Jayakody

Published 2026-07-24

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Luis Miguel Pires, Dushan Nalin Jayakody

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to send a secret message across a stormy ocean. You have three different ways to shout it out: a loud, reliable radio (Radio Frequency or RF), a super-fast laser pointer (Free-Space Optical or FSO), and a high-pitched whistle that travels incredibly fast but only for a short distance (Terahertz or THz). Each method has its own superpower and its own weakness. The radio is tough and works even when it's raining, but it's slow and crowded. The laser is blazingly fast, but if a fog bank rolls in, the beam gets scattered and the message is lost. The whistle is amazing for short bursts, but if the air is too humid, the sound gets swallowed up by the water vapor.

In the world of future wireless networks, engineers are trying to build a "smart switch" that can instantly pick the best shouting method depending on the weather. This is tricky because the ocean is unpredictable. Sometimes the fog is thick, sometimes the rain is heavy, and sometimes the air is perfectly clear. If you pick the wrong method, your message fails. This paper explores how we can teach computers to make these split-second decisions, not by following a rigid rulebook, but by learning from patterns, much like how a seasoned sailor learns to read the sky. The goal is to create a communication system that is as adaptable as a chameleon, switching between radio, laser, and whistle to keep the connection alive no matter what the sea throws at it.

The Smart Switch for the Stormy Sea

This paper presents a new way to build a "channel-aware" system for maritime (ocean-based) communications. The authors, Luis Miguel Pires and Dushan Nalin Jayakody, created a computer simulation of a hybrid network that uses Radio Frequency (RF), Free-Space Optical (FSO), and Terahertz (THz) technologies together. Think of this system as a smart traffic controller for data. Instead of just picking one road and sticking to it, this controller constantly checks the weather and the road conditions to decide which "lane" (RF, FSO, or THz) will get the message to its destination fastest and most efficiently.

To test their idea, the researchers didn't just guess; they built a detailed digital world. They programmed physics-based models that mimic how rain, fog, humidity, and atmospheric turbulence affect each type of signal. For example, they simulated how a laser beam (FSO) gets blocked by fog, or how a THz signal gets absorbed by humid air. They created a dataset with 1,600 different "scenarios," ranging from a sunny, clear day to a "worst-case" storm with heavy rain, thick fog, and high humidity.

The "Brain" of the System

The core of this research is teaching a computer to be the traffic controller. The authors used two types of "learning" methods:

Supervised Learning: Imagine showing a student a thousand flashcards where the answer is already written on the back. The computer looks at the weather (rain, fog, distance) and learns to predict the best channel based on a pre-set rule: "Pick the channel with the best Signal-to-Noise Ratio (SNR) and the lowest energy cost per bit."
Reinforcement Learning (RL): This is more like training a dog. The computer (the agent) tries different channels. If it picks the right one, it gets a "treat" (a reward). If it picks a channel that fails because of fog, it gets a "scolding" (a penalty). Over time, the agent learns to navigate the stormy seas on its own.

What They Found: The Good, The Bad, and The "What If"

The results of their simulation were revealing, but they came with some important caveats.

The Good News: When the computer was tested on weather conditions it had seen before (matched conditions), it was incredibly smart. The supervised learning models, specifically the Random Forest algorithm, got it right 98% to 99% of the time. The Reinforcement Learning agent also learned to pick the right channel, successfully mimicking the physics-based rules. For instance, in clear weather, the system correctly chose the fast laser (FSO). When fog rolled in, it smartly switched to the reliable radio (RF).

The Bad News (The Generalization Problem): Here is where the story gets interesting. The authors tested what happens if they train the computer on "Clear" weather and then suddenly throw it into a "Fog" scenario it has never seen before. The performance dropped significantly. When trained only on clear days, the system's accuracy fell to just 62.4% when facing fog and rain. This suggests that if you only teach a computer about sunny days, it will be confused when a storm hits. However, when they trained the system on all types of weather (Clear, Rain, Fog, and Worst-case), it became much more robust, maintaining an accuracy of 81.7% even when tested on the extreme "Worst-case" scenario it hadn't seen during training.

The "Noisy" Reality Check: The researchers also simulated what happens if the computer's sensors are a bit fuzzy or make mistakes (simulating measurement errors). They added "noise" to the data. The results showed that while the models were sensitive to these errors—their accuracy dropped by about 7% to 8%—they didn't crash completely. They remained functional, suggesting the system is somewhat resilient to imperfect data.

The Verdict: A Simulation, Not a Ship

It is crucial to understand the limits of this study. The authors explicitly state that these results come from a simulation, not a real ship at sea. They did not build a physical prototype or test it on actual water. The "data" they used was generated by their computer models, not measured by real instruments.

The paper argues against the idea that a simple, fixed rule (like "always switch to radio if it rains") is enough for the future. Instead, it suggests that while machine learning can work wonders, it needs to be trained on a wide variety of scenarios to be truly useful. If you only train it on one type of weather, it will fail when the weather changes.

In conclusion, this paper proposes a promising framework for a smart, adaptive communication system that can switch between radio, laser, and THz signals. It shows that machine learning can successfully learn the complex rules of the ocean, but only if it is taught with a diverse diet of weather conditions. The authors see this as a "proof-of-concept"—a strong first step that proves the idea works in a controlled digital world, but they emphasize that real-world testing with actual hardware and real ocean data is the essential next step to see if this smart switch can truly survive the real storm.

Technical Summary: Channel-Aware Adaptive Hybrid RF/FSO/THz Maritime Wireless Networking

Problem Statement
The advancement of wireless communication beyond 5G networks demands systems that offer high capacity, reliability, and energy efficiency. While hybrid systems combining Radio Frequency (RF), Free-Space Optical (FSO), and Terahertz (THz) technologies leverage the unique strengths of each, their operation in maritime environments is complicated by unpredictable atmospheric conditions. RF systems provide reliable long-distance links but suffer from limited bandwidth. FSO offers high data rates but is severely degraded by fog, turbulence, and alignment errors. THz communications enable ultra-high speeds over short distances but are highly sensitive to molecular absorption in humid conditions. Existing hybrid solutions often rely on fixed switching rules or dual-technology configurations (e.g., RF/THz only) that fail to model the complex, non-linear environmental behaviors of maritime settings. There is a lack of unified frameworks that utilize data-driven decision-making to adaptively select channels across all three technologies under dynamic conditions.

Methodology
The authors propose a channel-aware adaptive framework that integrates physics-based channel modeling with machine learning (ML) and reinforcement learning (RL) for decision-making.

Physical Channel Modeling: The study develops a unified physics-based model for RF (5 GHz), FSO (1550 nm), and THz (300 GHz) links. These models incorporate specific environmental factors:
- RF: Free-space path loss (FSPL), rain attenuation (ITU-R P.838), and log-normal shadowing.
- FSO: Geometric loss, atmospheric attenuation (Beer-Lambert law with Kruse model for visibility), and turbulence effects (Rytov variance).
- THz: FSPL, humidity-dependent molecular absorption, and directional antenna gain.
- Metrics: The models compute Signal-to-Noise Ratio (SNR), Shannon capacity, and energy-per-bit ( $E_b$ ) for each link.
System Architecture: The system simulates a maritime scenario involving an Underwater Unmanned Vehicle (UUV), a surface buoy acting as an intelligent relay, and a surface vessel. The buoy employs an Adaptive Link Switching (ALS) mechanism to select the optimal technology based on real-time environmental inputs (distance, rain rate, humidity, visibility, turbulence).
Dataset Generation: A multi-scenario synthetic dataset was created using calibrated physical ranges for four maritime conditions: Clear, Rain, Fog, and Worst-case. The dataset contains 1,600 samples (400 per scenario) with features including environmental variables and channel metrics. Labels for the "best channel" are assigned deterministically based on a rule prioritizing maximum SNR and minimum energy-per-bit.
Learning Framework:
- Supervised Learning (SL): Baseline models (Random Forest, Linear SVM, RBF SVM) were trained to predict the optimal channel label from the dataset.
- Reinforcement Learning (RL): A Deep Q-Network (DQN) agent was implemented in a Gymnasium environment. The agent observes the state (environmental and channel metrics), selects an action (RF, FSO, or THz), and receives a reward based on correct channel selection and energy efficiency, with penalties for outages.

Key Contributions

Unified Modeling Framework: Development of a physics-based simulation framework that jointly models RF, FSO, and THz channels under realistic maritime environmental factors.
Multi-Scenario Dataset: Creation of a synthetic dataset covering clear, rain, fog, and worst-case scenarios to train and evaluate adaptive algorithms.
Hybrid Decision Framework: Implementation of a system combining supervised learning and RL for context-aware channel selection.
Generalization and Robustness Analysis: Evaluation of model performance through cross-scenario generalization (training on specific conditions, testing on others) and robustness testing against noisy channel observations.

Results

Channel Dominance: Simulation results confirm distinct dominance regions for each technology. FSO is optimal for clear, mid-range distances; THz offers high bandwidth but is limited by humidity and distance; RF remains the most reliable option in fog, heavy rain, and worst-case scenarios.
Supervised Learning Performance: Under matched conditions (training and testing on the same scenarios), models achieved high accuracy (98–99% for Random Forest). However, performance degraded significantly in cross-scenario tests (e.g., training only on "Clear" and testing on "Fog" dropped accuracy to 62.4%), highlighting the dependency on training data diversity.
Reinforcement Learning: The DQN agent successfully learned policies that closely matched the physics-based optimal selection across different scenarios, demonstrating the ability to approximate complex decision boundaries.
Robustness: When Gaussian noise was introduced to SNR observations, model accuracy decreased by approximately 7–8%, indicating sensitivity to measurement errors but maintaining functionality within acceptable limits.
Generalization: Protocol C (training on all scenarios, testing on unseen "Worst" conditions) yielded better generalization (81.7% accuracy) than restrictive splits, underscoring the necessity of diverse training data.

Significance and Claims
The paper positions this work as a simulation-based proof-of-concept that establishes a controlled testing foundation for future real-world maritime communication systems. The authors claim that their findings reveal the limitations of learning-based channel selection, specifically that high accuracy under matched conditions does not guarantee performance in distribution-shifted environments. The study emphasizes that robust adaptive systems require diverse training scenarios and strong decision-making methods capable of handling measurement errors. While the framework successfully replicates physics-based policies in simulation, the authors modestly note that the results are derived from synthetic data and deterministic labeling, and future work must validate these approaches using real-world maritime measurements and dynamic learning techniques.

The Smart Switch for the Stormy Sea

The "Brain" of the System

What They Found: The Good, The Bad, and The "What If"

The Verdict: A Simulation, Not a Ship

Technical Summary: Channel-Aware Adaptive Hybrid RF/FSO/THz Maritime Wireless Networking

More like this