Digital Twin-Assisted OWC: Towards Smart and Autonomous 6G Networks

,


I. INTRODUCTION
The rapid growth of the emerging Internet-of-Everything (IoE) services and data-hungry applications, such as haptics, holographic telepresence, collaborative robots, and autonomous systems, have motivated the development of new network architectures and innovative technologies for 6G wireless networks.These applications require stringent solutions to support enhanced mobile broadband and ultra-high-reliable and ultra-low-latency communications, which is rather challenging due to the scarcity of the spectrum.Therefore, researchers from academia and industry started exploring higher frequency bands for wireless communications in the future 6G and beyond mobile communications [1].Within this context, optical wireless communication (OWC) has been proposed as a promising energy-efficient and green wireless technology for massive connectivity of users with increased quality-of-service (QoS) requirements.This has been further supported by the rapid advancements in high resolution cameras, highly-sensitive photodetectors (PDs), high compatible Nano-photonic sensors, and energy-efficient laser/light-emitting-diodes (LD/LEDs) [2].OWC is often divided into two main categories: LD-based, Hossien B. Eldeeb, Shimaa Naser, Lina Bariah, and Sami Muhaidat are with the KU Center for Cyber-Physical Systems, Department of Electrical and Computer Engineering, Khalifa University, Abu Dhabi, UAE.Lina Bariah is with the Technology Innovation Institute, Abu Dhabi, UAE.Murat Uysal is with the Engineering Division, New York University Abu Dhabi (NYUAD), Abu Dhabi, UAE, 129188.He is also an Adjunct Professor at Özyeǧin University, Istanbul, Turkey, 34794.E-mails: hossien.eldeeb@ku.ac.ae.which stands for free space optics (FSO), and LED-based, which stands for visible light communication (VLC).VLC can be further subdivided into two subcategories based on the type of receiver used: PD and camera/image sensor.The latter subcategory is known as optical camera communication (OCC).Several research efforts have demonstrated the ability of such OWC networks to enable safe and high-speed wireless connectivity in homes, hospitals, manufacturing, airborne FSO, vehicular networks, and unmanned aerial vehicles (UAVs) [3].
Despite their undoubted capabilities and features, OWC systems are highly influenced by user mobility, random receiver orientation, weather conditions, interference from artificial light sources, and signal blockage.Consequently, a large gap exists between the physical OWC environments (with their actual objects) and the simulated/analyzed counterparts [4].Moreover, the large generated data from OWC applications poses several critical issues, including data processing, accurate system monitoring, online management, real-time control, and fast and efficient maintenance.Therefore, it is of paramount importance to develop disruptive technologies for future OWC networks to meet the requirements of self-sustaining and proactive-online learning-enabled systems.In this regard, digital twin (DT) technology has been proposed as a promising tool to provide real-time monitoring and control, predictive maintenance, risk assessment, and increased overall system efficiency.The main idea of DT is to firstly create a virtual digital representation of the actual physical system, including all physical elements and their behaviors, environmental conditions, and the dynamics and functions of the network.Then, by utilizing the collected data at the DT, historical records and real-time sensory data, advanced computing and artificial intelligence (AI) technologies are leveraged for model training and extracting insights from the DT to predict future behavior, simulate different scenarios, and other analytical tasks.Therefore, the integration of DT with OWC systems is appealing in accommodating the diverse requirements of the emerging OWC applications by enabling adaptive and efficient network optimization and analysis.
While DT technology is still new, it has received interests in several areas, including telecommunications, energy systems, Industry 4.0, military, aerospace, and smart transportation.In the context of wireless communication, an overview of some recent works [5]- [14] is summarized in Table 1.It is worth noting that, existing contributions have tackled the implementation of DT in different radio frequency (RF)-based wireless networks.To the best of our knowledge, none of the existing literature studies considers the implementation of DT in OWC networks.
Unlike existing works, this article provides an overview of DT technology and its integration with OWC networks.We start by shedding light on the main roles that DT can perform when integrated with different smart-autonomous OWC networks.Subsequently, we present some key enabling technologies and related design and deployment aspects of DT-assisted OWC systems.Finally, we discuss the impact of different system and environmental factors on the overall system performance and highlight some interesting open research problems.

II. MAIN ROLES OF DT IN SMART-AUTONOMOUS OWC NETWORK SCENARIOS
Smart and autonomous systems are rapidly evolving in different sectors including industry, healthcare, logistics, transportation, aerospace, and underwater communications, to name a few.In such systems, millions of sensors are deployed into vehicles, homes, and industrial plants to sense the surrounding environment and then analyze the collected data along with the existing knowledge to provide automatic decision-making and control of processes, devices, and systems, without human intervention.Additionally, these decisions and interactions can be further optimized based on the gained experience, knowledge, and the environmental dynamics, aiming to provide an enhanced quality of life.Maintaining such smart and end-toend network automation with constrained latency and accuracy requirements, calls for the development of a set of technologies, architectures, and paradigms for digital transformation to interact and control the virtual world that is decoupled from the physical counterpart.In this regard, DT has been recently identified as a key enabler for future smart-autonomous OWC systems.One notable example is in VLC systems, while the channel exhibits a deterministic nature, a significant obstacle arises due to user mobility, device orientation variations, and fluctuating environmental conditions.These factors make it challenging to obtain accurate and real-time Channel State Information (CSI), which is essential for system design and optimization.By leveraging DT, valuable data can be gathered from user equipment measurements, LiDAR data, or ray-tracing simulations.This data serves as prior knowledge to interpolate the CSI and minimize reporting overhead.
The main principle of DT in smart-autonomous OWC systems is to virtually represent the physical elements under different environmental conditions and network dynamics.Then, by utilizing the historical data records and real-time sensory data, fast and efficient computing and advanced AI technologies are leveraged for data processing and to facilitate efficient network configuration, optimization, design, and decision-making.Thus, the integration of DT with smart-autonomous OWC systems has the capability of designing stable and efficient OWC networks.Also, this integration plays a significant role in optimizing the network resources and improving the monitoring, diagnosing, and maintenance of autonomous OWC networks.
In Fig. 1, we present the general architecture of DT-assisted OWC system.Typically, in the physical space, indoor luminaries, optical implants, on-body LEDs, and outdoor lighting infrastructures (i.e., traffic lights, headlights, etc.) are adopted as wireless transmitters.On the other hand, optical cameras, PDs, and optical body sensors are utilized to collect real-time sensory data, including object dimensions and properties, environmental conditions, and human/system behaviors.Then, the collected, historical, and service execution data are processed and transferred to virtual models that are developed utilizing advanced tools such as twin builder, optics builder, Simulink.With the aid of the aggregated data, advanced AI, and computing techniques, the developed virtual models perform different simulations studying the performance, providing possible improvements that are applied back to the physical OWC system.Based on the feedback, real actions are performed by each smart-autonomous OWC element, depending on each particular scenario.In the following, without loss of generality, we discuss two DTassisted OWC scenarios (i.e., indoor and outdoor applications).

A. DT-Assisted OWC-Based BANs
Owning to their capability to mitigate the risk of electromagnetic interference with medical devices, OWC networks have emerged as a viable alternative to RF-based medical body-area networks (BANs) [2].This shift has been driven by advancements in bio-compatible devices, their compact size, and reduced power consumption.Nano-LEDs are employed as efficient wireless transmitters, while highly sensitive nano-PDs are utilized for wireless sensing.These sensors capture various patient information, such as blood pressure, heart rate, blood oxygen levels, glucose levels, and other vital signs.Such realtime health monitoring information plays a crucial role in detecting health emergencies and abnormalities, enabling prompt intervention.Conventionally, decisions and actions have been based on the staff's and professionals' traditional knowledge in the field and basic analysis of patients' vital signs.However, it is imperative to provide accurate and real-time decisions and actions to enhance patients' quality of life and well-being.Integrating DT with OWC-based medical BANs is expected to revolutionize precise diagnosis and enable personalized and effective treatment by developing highly accurate models of patients' bodies.Furthermore, studying diseases is anticipated to become more efficient by conducting clinical tests on these virtual models instead of directly testing on the patient's body.
Fig. 2 illustrates how DT is integrated with OWC-based medical BANs to support several potential use cases, such as efficient patient examination, disease identification, and remote monitoring/surgery.Since data is the key to unlocking the potentials of DT, optical on/off body sensors, LED hubs, and optical cameras collect simultaneous sensory data about the patient and the surrounding environment, which are then transmitted to the optical gateway.The gateway, in turn, forward the data to a control unit, which shares this data with a room server.Each room server can then share its information with the DT layer via the cloud.At the DT layer, both the real-time sensory and clinical history data are processed, and then, with the aid of AI algorithms and efficient computing, a simulation of the virtual model is carried out.It is noteworthy that, to build an accurate and realistic virtual model, it is necessary to update the model according to its performance compared to the actual patient's body.Once a holistic virtual model is fully evolved, useful insights and clinical decision-making will be fed back to the physical system.For instance, some information is directly sent to the local health service providers, such as the patient's doctor/nurse, for quick intervention and correct diagnosis in an emergency.Thus, it is anticipated that with the promising capabilities of OWCs to provide high-speed transmissions and the ability of DT to provide real-time and accurate management, precise diagnosis, real-time monitoring of various diseases, deciding on treatment plans, and studying treatment effects, can be all realized in medical BANs to enhance life quality.

B. DT-Assisted IoT-Based OWC UAV and Vehicular Networks
The proliferation of LED-based outdoor lighting, along with advancements in high-resolution cameras and highly sensitive optical sensors in vehicles, has led to the emergence of OWC as a promising communication paradigm for vehicular networks.Simultaneously, OWCs based on unmanned aerial vehicles (UAVs) have gained significant interest due to their capability to enhance coverage and capacity in rural areas.A key objective in developing vehicular and UAV networks is the realization of fully autonomous connected vehicles, where tasks are performed in real-time with minimal human interaction while ensuring road safety.This necessitates highly accurate CSI, ultra-high data rates, ultra-low latency, and efficient big data processing capabilities.Moreover, connected vehicles are susceptible to various cyber-attacks and malicious hackers, which must also be addressed.Therefore, integrating DT technology Fig. 2: DT-assisted medical Body OWC networks.into such highly mobile and complex networks is regarded as a crucial enabling solution.DT facilitates highly efficient synchronization, monitoring, and data sharing among various real-world infrastructures (using traffic light signals), vehicles (utilizing head-and taillight transmission), pedestrians, and more.It significantly improves resource allocation efficiency and optimizes system parameters.For instance, researchers have proposed the use of multi-hop relays and intelligent reflecting surfaces (IRSs) technologies to establish virtual lineof-sight (LoS) links between autonomous vehicles and road users.Here, DT, with its ability to store highly accurate CSI through historical information or generate it using ray-tracing simulators, enables timely and efficient selection of the optimal relay vehicle.This maximizes the quality of the relayed signal, minimizes energy consumption of relay nodes, and ensures fairness for road users.Furthermore, DT aids in optimizing the phase shifts, amplitude, and frequency of IRS elements installed on cars, road infrastructures, buildings, bus stops, etc. Lastly, DT considers and estimates various environmental conditions, such as noise and interference from the sun and artificial light sources, weather types, and users' illumination requirements, which have a significant impact on OWC systems.
DT possesses inherent intelligence, managerial efficiency, extensive storage capacity, fast and accurate decision-making capabilities, and comprehensive inclusivity in its rewards.As a result, its integration with O-IoT UAV/vehicular networks will revolutionize the implementation and widespread adoption of OWC-aided connected and autonomous vehicles.It also facilitates the identification of potential eavesdropper attacks in public OWC networks by concurrently monitoring their behaviors and reactions, and considers atmospheric weather conditions' impact on OWC performance through efficient simulation of virtual weather models.This comprehensive and precise perspective has the potential to expand coverage and capacity, enhance reliability and sustainability, improve secrecy and energy efficiency, and minimize latency and malfunctions.

III. KEY ENABLING TECHNOLOGIES AND DESIGN ASPECTS OF DT-ASSISTED OWC SYSTEMS
In this section, we illustrate some key enabling technologies for DT-assisted OWC networks and explore some design requirements and aspects associated with such integration which need to be considered.Figs. 3 and 4 summarize the key enabling technologies and design aspects, respectively.
A. Key Enabling Technologies for DT-Assisted OWC Systems 1) Big Data Technology: To create an accurate and holistic digital model for the considered OWC system, a wide range of data measurements pertaining to objects, their behavior, and the surrounding environment is required.The emergence of O-IoT applications and services has facilitated continuous data acquisition through connected optical sensors and cameras.Once a substantial amount of data is collected via O-IoT devices, it becomes necessary to process, analyze, and manage the data to extract valuable insights.However, conventional data processing tools struggle to handle the torrents of heterogeneous and unstructured data effectively.Hence, O-IoT-driven massive data acquisition, combined with big data analytics, plays a crucial role in enabling the successful operation of DT.Given the diverse nature, large volume, and complexity of the collected big data, traditional database technologies prove impractical.Advanced big data storage technologies such as MySQL, NoSQL, NewSQL services, distributed file storage (DFS), Mongo DB, and cloud storage become necessary for efficient data storage during the processing and analysis phases.For example, NoSQL techniques enable horizontal scaling to support massive data, while NewSQL provides scalable and robust databases with high storage and management capabilities.Additionally, DFS allows for simultaneous sharing of files and directories between hosts over the network.To extract valuable insights from incomplete, noisy, and unstructured raw data, data processing is carried-out to remove irrelevant, misleading, inconsistent, and duplicated data.This can be performed through different technologies such data cleaning, data smoothing, dimension reduction, and data Middleware.Subsequently, data fusion is initiated to produce unified and comprehensive information about an entity in a low-level category (decision level).This process involves both traditional approaches such as Bayesian estimation, weighted average method, Kalman filtering, as well as AI methods like wavelet theory, fuzzy set theory, support vector machines, and neural networks.Finally, to visualize the data, different technologies can be employed, such as geometry-based, imagebased, icon-based, and pixel-oriented approaches.These visualization techniques provide a more intuitive representation of the relationships between data, virtual models, and physical entities.
2) AI/ML/DL Technologies: OWC networks are anticipated to support a plethora of services and applications with diverse QoS requirements.These include high data-rate video gaming activities, ultra-reliable low-latency autonomous vehicles and remote surgery, as well as battery-powered machine-type applications.The presence of such varied and divergent data formats can complicate the deployment of OWC networks.However, advanced AI techniques serve as indispensable solutions due to their ability to address data diversity issues and construct a virtual representation of OWC systems based on a baseline model.The effectiveness of DT-assisted OWC networks relies on accurate representations of physical devices, their behavior, and the surrounding environment for learning, reasoning, and decision-making capabilities.While mathematical models have been extensively explored for DT, they may not accurately capture the dynamics of physical OWC networks, especially in highly dynamic scenarios and when numerous assumptions are made.Therefore, data-driven models based on AI/ML/DL play a crucial role in creating a virtual representation of the system.This is achieved by feeding processed real-time and historical data into AI/ML/DL models for training purposes.It should be noted that these models can be updated to develop generalized and mature DT models that adapt to changes in the optical physical system, enhancing operational efficiency.DT-assisted OWC networks rely on AI tools to perform various tasks, such as classification, prediction, clustering, anomaly detection, and optimization.Integrating AI in DT-assisted OWC networks enables i) Efficient network optimization and automation, ii) Rapid sharing of test results or trained models across different sectors, iii) Cost-saving when enabling new communication services with efficient troubleshooting, maintenance, and updates.Thus, the advancement in AI technologies creates a road-map toward DT-assisted OWC networks.
3) Modeling and Visualization: To achieve an accurate and realistic virtual representation of the considered OWC system, efficient tools are needed to digitally develop geometrical, physical, behavioral, and rule models.Computer-aided design (CAD) and visualization tools are essential for visualizing the geometry of various entities within the OWC system.Marketavailable tools like CNC tools, TwinCat, Aspera, RaySync, Siemens NX, etc., possess the capability to create high-fidelity virtual representations of the physical twin.In the context of DT-assisted OWC networks, significant roles are expected to be played by tools such as ANSYS Twin Builder, OpticStudio raytracing, Optics Builder, and Simulink.These tools efficiently and accurately represent the orientation of different devices, their exact size, optical characteristics of light sources (such as intensity profiles and operating spectrum), optical specifications of object surfaces (including coating materials, reflectance, and scattering factors), and precise details of physical entities.In addition to geometrical and material modeling, behavioral and rule models are required to define the actions of the virtual model in response to changes in the physical entity.These models rely on advanced algorithms and techniques that can extract rule information, analyze and optimize them, and predict the performance of objects accordingly.Various algorithms are already available, such as semantic data analytics and the extensible markup language (XML) data format.
4) Intelligent Reflecting Surface (IRS): The performance of OWC networks highly depends on the availability of LoS links, which may not always be feasible in practical scenarios.Thus, IRS technology is considered crucial for OWC networks.IRS, composed of reconfigurable metasurfaces, offers significant potential for controlling the propagation of light waves in OWC systems.This capability provides extended coverage, improved signal reliability, and enhanced energy efficiency.Moreover, deploying IRS at front-ends of transmitters in UAVs and autonomous vehicles can enhance secrecy performance by directing the beam towards intended entities.It is worth noting that achieving intelligent control of OWC environment requires efficient placement and tuning of IRS mirrors/metasurfaces. Thanks to DT capabilities and reliance on efficient AI technologies, DT can greatly aid in tuning the elements of IRS based on sensory data received from the physical twin and/or simulation data from ray-tracing of virtual models.Furthermore, DT, with its holistic view, enables the effective utilization of these techniques to optimize the number and orientation of IRS elements to optimize the energy and spectral efficiency.

B. Design Aspects of DT-Assisted OWC Systems
1) Decoupling: The efficiency and adaptability of DT in handling network dynamics and OWC channel impairments rely on the independent design and operation of DT model from Fig. 4: Design requirements for digital twin-assisted OWC system. the physical space.This requires an efficient decoupling of this model from its physical counterpart through data and network functions decoupling [8].DT models for real-time OWC applications encounter multi-temporal scale, multi-dimensional, multi-source, and heterogeneous data originating from diverse outdoor and indoor lighting infrastructures.Thus, during data collection stage, it is crucial to transform the original data into a common processable format and usable information, which is decoupled from the original data sources.Simultaneously, functions decoupling, such as resource allocation, parameter configuration, and interference management, can be accomplished through network slicing based on software-defined networking (SDN) and network function virtualization (NFV) enabling the development of efficient and adaptive DT models.While SDN separates the control plane of DT-assisted OWC network devices from the underlying data plane, facilitating centralized network orchestration, NFV leverages virtualization technology to flexibly program service functionalities as software instances known as virtual network functions.
2) Scalability and Reliability: The accelerating growth of connected optical sensors, actuators, and other equipment for data acquisition in O-IoT applications requires the development of a scalable and reliable DT architecture.It is important to note that centralized development of DT models can lead to increased latency and reduced scalability as the number of devices grows.This is attributed to the increased signaling between the centralized server and the large number of connected devices.Developing DT models in a distributed manner can reduce latency and increase system scalability.However, this approach comes with increased management complexity and implementation costs.Additionally, delays arise from communication between enddevices and distributed servers.Therefore, hybrid centralizeddistributed DT models can provide a trade-off between latency and complexity.In such hybrid architectures, a central server handles overall DT model aggregation and communication with a set of local servers.The local servers, in turn, aggregate the local DT models received from a group of end-devices that only communicate with the local servers.While DT can be used to virtually represent individual physical objects, achieving efficient control, real-time analysis, and management of OWC networks requires continuous information interaction and collaboration between physical objects and their virtual twins, as well as between various virtual twins, and physical objects.By enabling communication and interaction among interconnected twins representing multiple physical entities, a DT network for the OWC system can be established, offering enhanced reliability and efficiency [6].Furthermore, continuous updates of the DT models are necessary to consider new scenarios that reflect real-time physical situations.In this context, DT networks developed in a distributed manner using transfer learning techniques can adapt to changes in the operational environment without requiring the retraining of all the twin models.Specifically, a retrained model for a specific twin can be transferred to other twins and used as a starting point in the training process for their models.
3) Resources Allocation: To ensure real-time interaction between DTs and the corresponding physical objects in OWC networks, frequent learning model updates and control instructions need to be communicated.Also, in centralized DT models, a large amount of heterogeneous data is generated and shared with a central server for model training, draining a significant amount of the available communication resources.Thus, it is crucial to consider effective resource management, such as frequency band and access schemes, while maintaining predefined QoS requirements and latency in DT-assisted OWC.Additionally, computation offloading to distributed servers, which are close to the end devices, is another promising solution to enhance task processing in DT-assisted OWC networks.However, to preserve the computing resources and complete the computationintensive tasks within the delay constraint, there is a need to provide an efficient task offloading strategies.

IV. CHALLENGES AND FUTURE RESEARCH DIRECTIONS
The design and development of DT-assisted OWC networks involve the integration of various technologies posing several new challenges and potential future research directions: 1) Blockchain for Data Management: DT-assisted OWC networks will collect, store, process, and analyse thousands or even millions of users's data.Such a big data causes irreparable serious damage since it is vulnerable to a couple of issues, including data attack, data tampering, data loss, and data leakage.Therefore, it becomes urgent to reform the data management with more efficient manner.As a promising solution, Blockchain technology can be used, which can streamline the operations of data management and offer unprecedented data efficiency as well as compel trust [8].This is due to its unique advantages, including immutability, anonymity, privacy and security, decentralized storage, transparency, and flexibility of data access.While Blockchain can provide several opportunities to DT-assisted OWC networks, it imposes some open research directions to be considered such as scalability and limited achievable performance, i.e., throughput, computation, and storage.However, integrating ML, DL, and reinforcement learning with Blockchain can assist in optimizing its framework and maximizing the achievable throughput.
2) Design and Mobility Management: While SDN and NFV paradigms have the potential to support the decoupling of DTassisted OWC networks, they introduce significant signaling overhead in highly dynamic environments.That might affect the latency and mobility management of DT-assisted OWC networks.One solution is to consider the implementation of DT models at the network edge.Another effective solution is to offload some computational tasks toward side units (servers).Considering that and with the help of advanced AI algorithms can effectively tackle the complexity of conventional optimization methods in offloading and minimize the overheads with the aid of end devices' sensory data.
3) Security and Privacy: Another critical challenge in DTassisted public OWC networks is maintaining network security and protecting the privacy of users' data.Some solutions based on federated learning have been introduced with significant accuracy enhancement.However, there are still some open questions.One such concern is the potential inference of sensitive information about local data from shared model updates.This could be done by a potentially malicious server that observes local updates over time or by a malicious participant who monitors global parameter updates.Therefore, a paramount future direction is to develop robust mechanisms for protecting the shared model parameters against malicious attacks.Enablers such as differential privacy, homomorphic encryption, overthe-air computation, dispersed collaborative learning, mobilityaware forensics, and Blockchain can support this goal [8].
4) DT-Assisted Airborne FSO: In non-terrestrial network planning, FSO technology is used for intra-connectivity between satellites, HAPS, gateways, and the core network.Network planning and service provisioning are crucial, and DT can play a vital role.However, it's essential to design the DT system to function effectively in adverse weather conditions and obstacles like cloud blockages.To tackle these challenges, telemetry and historical weather data can be used to determine the best path and facilitate service provisioning based on transmission quality.DT can also allocate network resources reactively in case of primary path failures.By leveraging these capabilities, DT-assisted airborne FSO network ensures reliable and robust connectivity in challenging environmental conditions.5) Standardization: Efforts are underway to establish DT standards in various application domains, e.g., ISO is focusing on developing DT standards for manufacturing frameworks in the industry sector.However, there is a lack of DT standardization in communication systems, especially for OWC.Standardizing DTs for OWC is crucial for effective deployment in smart cities, enabling organizations to develop efficient APIs for data access.By adhering to such a standard, users across different physical entities like homes, vehicles, etc., can collect, store, process, manage, and exchange information with high reliability and security.Achieving that requires collaboration across both public and private sectors.This includes entities such as ITS, automotive manufacturers, home automation suppliers, smartphone manufacturers, computing software developers, and network providers.Collaboration among these stakeholders is essential to establish comprehensive standards that encompass the diverse aspects of DT-assisted OWC networks.

V. CONCLUSION
In this article, we introduced a vision of DT-assisted OWC networks.In particular, we discussed the main roles that DT can perform when integrated with different smart-autonomous OWC networks.Then, we discussed different key enabling technologies and some design and implementation aspects associated with DT-assisted OWC systems.Finally, we discussed the impact of different network and environmental factors on the overall performance, and we outlined some interesting open research directions which need to be addressed to realize the full potential of integrating DT with OWC systems.

TABLE I :
Related DT Works