Electromagnetic Twin Space: When Digital Twins Meet the Electromagnetic Space

Abstract

With the escalating demand for electromagnetic spectrum resources in the 5G/6G era, efficient management of the electromagnetic space has become a critical challenge. This paper proposes the concept of an Electromagnetic Digital Twin (EDT) and an innovative framework for constructing Electromagnetic Twin Space (ETS) to achieve high-fidelity dynamic mapping and real-time optimization of the electromagnetic space through digital twin technology. We elaborate on the EDT concept, introducing a three-layer architecture comprising Physical Device Twin (PDT), Propagation Environment Twin (PET), and Electromagnetic Situation Twin (EST), thereby systematically integrating digital twin technology into the electromagnetic domain. Furthermore, we designed the ETS construction framework, clarifying the four key links between ETS construction and operation and their associated technologies. Through a case study, we demonstrate the effectiveness of a GAN-based EST, which achieves significantly better prediction accuracy than traditional methods. The findings show that incorporating building information and transmitter parameters substantially enhances the accuracy of EST, as evidenced by the RMSE metrics of the constructed electromagnetic situation. Moreover, the trained GAN model can generate electromagnetic situations under various building scenarios and transmitter locations, providing a valuable experimental platform for transmitter deployment.

1. Introduction

With the advancement of 5G applications, research on 6G has progressively evolved. The vision for 6G encompasses achieving the Internet of Everything, ubiquitous coverage, and ultra-low communication latency. However, realizing these objectives remains a formidable challenge. For instance, the electromagnetic spectrum is a crucial resource for ensuring secure communications, yet the proliferation of spectrum-dependent devices leads to an increasingly congested spectrum environment. Addressing how to efficiently allocate spectrum resources to ensure adequate access for frequency-using and spectrum-dependent devices while avoiding wastage becomes imperative, [1]. In light of future communication technology tests and scenarios, establishing a highly reliable and flexible testing environment that minimizes costly test expenditures is essential. Consequently, effectively managing all aspects related to the electromagnetic space becomes a pivotal focus for future research.

Effectively managing the intangible nature of the electromagnetic space poses significant challenges. Previous studies have utilized the concept of the electromagnetic situation to depict its present state, synthetic form, and variation trends [2]. However, visualizing the electromagnetic situation solely through a radio map [3] only provides field strength distribution in a specific region at a particular time, lacking a comprehensive representation of the entire electromagnetic space. This space encompasses various elements such as spectrum-dependent devices, propagation environment, and interactions with each other, forming a complex system that complicates manual deployment, management, operation, and optimization. Furthermore, communication activities constantly impact the electromagnetic space, leading to rapid changes. Addressing these challenges effectively requires advanced technologies and efficient approaches to provide a unified management platform for electromagnetic space.

Digital Twin (DT) is the process of replicating and simulating the characteristics of physical objects, such as their structure, behavior, and status, in a virtual space using digital technology. This enables real-time monitoring, prediction, and optimization of these physical entities while providing an opportunity for comprehensive control over the electromagnetic space. Each spectrum-dependent device within this space can generate a digital twin through digital twinning techniques. These digital twins accurately represent the physical space and facilitate the twin of the electromagnetic situation by exchanging data with their corresponding physical counterparts. The virtual-reality interaction capabilities possessed by digital twins effectively capture changes in the electromagnetic situation and enable complete supervision. Ultimately, the interoperability achieved through DT establishes a unified management platform that facilitates comprehensive control over the electromagnetic space.

In recent years, DT has demonstrated remarkable capabilities in the field of communication. Specifically, extensive research has been conducted on digital twin networks (DTNs), virtual replicas of physical communication networks. DTNs provide a detailed digital platform that offers real-time information about the network’s current state and predicted outcomes [4]. DTNs facilitate real-time interaction mapping to support closed-loop logic for subsequent decision-making processes by enabling bidirectional communication between the physical network and its twin counterparts, while [5] provides a systematic description of requirement analysis, general architecture, and essential components of DTNs, it lacks elaboration on their application scenarios. To address this knowledge gap, Apostolakis et al. [6] proposes three specific application scenarios for DTNs and employs open-source tools to implement and evaluate these scenarios in detail for network devices such as virtualized Radio Access Networks (RAN) components.

Despite the extensive research on DTN, the investigation of the Electromagnetic Digital Twin (EDT) is still nascent. Gregory D. Durgin has made significant strides in advancing the concept of Digital Spectrum Twins (DST) by integrating digital twins into the domain of electromagnetic fields, thereby establishing a robust foundation for EDT [7]. However, this work solely focuses on the electromagnetic situation and overlooks other aspects within the electromagnetic space. In light of this, a novel concept called MetaRadio has been proposed to enable radios to transcend temporal and spatial limitations inherent in traditional radio systems, thus establishing an interactive framework that bridges virtuality and reality [8].

This paper aims to establish a digital replica of the electromagnetic space generated by the physical space. Firstly, we review the related work about DT, DTNs, and related subjects. Subsequently, we introduce the concepts of Electromagnetic Digital Twin (EDT) and Electromagnetic Twin Space (ETS). The electromagnetic digital twin architecture proposed in this paper is shown in Figure 1. Next, we present the construction process and framework for ETS. As a case study, we investigate electromagnetic situation twins for urban environments based on generative adversarial networks and compare the performance of the methods. Finally, several open research directions are summarised.

2. Related Work

Numerous studies have investigated the impact of DT on next-generation 6G communications. Regarding DTN, Almasan et al. [9] provides a comprehensive overview of its background, necessity, enabling technologies, and future opportunities. Within this context, Guo et al. [10] proposes innovative technologies such as Software-defined networking (SDN) and Space-Air-Ground Integrated Network (SAGIN) for future 6G to enhance the potential of DTN. Spectrum-dependent devices in the physical network play dual roles as generators and controllers of electromagnetic situations, making them closely intertwined with the physical network. Thus, DTN is an effective support system for EDT, laying a solid foundation for its implementation.

The modeling of propagation environments, a core component of the Propagation Environment Twin (PET), has been significantly advanced by recent research on near-field channel characteristics. Studies such as [11,12] provide high-efficiency analysis methods for large-scale MIMO and RIS-enabled air-ground channels, respectively, offering sophisticated models that can be integrated into the PET to enhance the fidelity of wave propagation simulation in complex scenarios [11,12]. Furthermore, the challenge of wideband channel covariance estimation in extremely large-scale MIMO (XL-MIMO) systems, as addressed by Ruan et al. [13], provides critical technical support for PET modeling in wideband and near-field conditions, effectively overcoming the beam split effect [13].

In addition to static spectrum-dependent devices, EDT twins should also consider mobile spectrum-dependent devices like vehicles and Unmanned Aerial Vehicles (UAVs). Considering the future requirements of autonomous driving scenarios, He et al. [14] presents a framework called Vehicular Digital Twin (VDT), highlighting the significance of vehicles as spectrum-dependent devices that contribute to effectively mastering electromagnetic space and supporting autonomous driving scenarios. The dynamic channel modeling for fluid antenna systems (FAS) in UAV communications, as explored in [15], is particularly relevant for the Physical Device Twin (PDT) of mobile platforms, enabling more accurate characterization of channel variations caused by the movement of antennas or platforms [15]. The (quasi-)closed-form channel statistics and parameter estimation methods for UAV-to-ground links proposed by Zeng et al. [16] further provide foundational models for constructing PDTs for UAVs, enhancing the accuracy of channel simulations in mobile scenarios [16].

For the core task of Electromagnetic Situation Twin (EST), which involves channel estimation and state prediction, advanced algorithms are crucial. The simplified learned approximate message passing (LAMP) network proposed by Ruan et al. [17] showcases an efficient deep learning framework for beamspace channel estimation in mmWave massive MIMO systems [17]. This algorithm can be directly integrated into the EST layer of the ETS to achieve rapid and accurate reconstruction of electromagnetic field strength distribution, providing a powerful data-driven tool for real-time electromagnetic situation perception.

Real-time virtual-real interaction stands out as one of the key advantages DT models offer. Hence, timeliness becomes a crucial criterion when evaluating their performance. To address this aspect comprehensively, Duran et al. [18] introduces a novel measure known as the Age of Twin (AoT), ensuring real-time reflection of physical world situations by digital twins. Another advantage lies in providing an efficient test platform that reduces field trial costs while enhancing trial-and-error capabilities. Gao et al. [19] demonstrated the benefits of DT in radio testing and conducted simulation experiments on DT environments and multiple-input multiple-output (MIMO) user devices. This paper aims to establish an ETS through EDT, providing insights into ETS construction. As a digital representation of electromagnetic space, ETS can dynamically reflect changes in real time and offer support for future 6G communications regarding the electromagnetic spectrum.

3. What Is Electromagnetic Digital Twin?

3.1. The Concept of Electromagnetic Digital Twins

The EDT is a high-fidelity, real-time dynamic mapping technology for electromagnetic space that uses DT. EDT is a specific application of digital twin technology in the electromagnetic field, which integrates digital models of physical devices, propagation conditions, and real-time electromagnetic activities. Through high-precision modeling, real-time data interaction, and dynamic simulation, a virtual mirror of the physical electromagnetic system (e.g., antenna, radar, electronic equipment) is constructed to achieve prediction, optimization, and full-life-cycle management of electromagnetic behaviors, performances, and environmental interactions. Given the rapidly evolving electromagnetic space, EDT can provide a flexible testing environment with high availability and reduce costly trial-and-error endeavors. Most importantly, it caters to varying requirements across applications and scenarios through reasonable trade-offs. Although the foundation for developing DTN and DST already exists, EDT is still nascent. Notably, EDT differs from existing counterparts in various aspects, encompassing but not limited to:

• Traditional DT lies in the physical entities existing in the tangible world, which are definite. Traditional DT involves the transition from reality to digital representation. In contrast, EDT’s main concern revolves around electromagnetic space, which lacks visibility and nondeterminacy. EDT aims to digitize electromagnetic space, presenting significant challenges to the construction of ETS.
• Traditional DT exhibits stability due to its physical entity, remaining unchanged unless subjected to significant damage. In contrast, the electromagnetic spectrum undergoes rapid fluctuations, and any communication activity within a region impacts the characteristics of electromagnetic space. It can also demonstrate unpredictability in specific complex scenarios, thereby necessitating higher demands on the timeliness of ETS.

3.2. Electromagnetic Digital Twins Fundamental Element

In essence, EDT serves as a bridge between the physical electromagnetic environment and its digital counterpart. This is achieved through a three-layer architecture:

• Physical Device Twin (PDT): PDT involves the creation of electromagnetic digital twin models for spectrum-dependent devices. Initially, a conventional digital twin is developed for the physical device, followed by establishing a PDT model based on its frequency-dependent parameters. The PDT model serves as both the physical foundation of ETS and the primary source of electromagnetic data. Subsequent processing enables real-time comprehension of the coupling between physical devices and their electromagnetic environment, facilitating convenient adjustments. The PDT framework encompasses three primary components: base stations, sensing terminals, and communication terminals. These components collectively support the comprehensive collection and integration of data from the physical electromagnetic environment. Base stations serve as the foundational infrastructure for wireless communication. They transmit wireless signals through their antenna systems to cover specific geographic areas while receiving signals from terminal devices. This dual functionality enables base stations to establish the fundamental electromagnetic situation within a given scene, providing essential data for constructing the electromagnetic digital twin. Sensing terminals focus on data collection and environmental monitoring. Their primary role is to gather data from the physical environment, capturing dynamic changes in real-time. These terminals are equipped with various sensors, such as electromagnetic, environmental, and localization, which enable the detection and uploading of data related to non-source radiation. This capability is crucial for providing real-time insights into the electromagnetic environment and supporting the dynamic mapping of the electromagnetic space. Communication terminals are responsible for data transmission and inter-device connectivity. These terminals include many devices, such as drones, vehicles, and smartphones. They facilitate the exchange of information between devices and the central system, ensuring seamless communication and data flow within the electromagnetic digital twin framework. PDT data collection relies on the collaborative work of multiple sensors and devices, and data are transmitted to the computing center through active uploads and passive monitoring. One is the active uploading of parameter data by physical devices; for example, base stations upload data such as location information, transmit frequency, transmit power, and operational status. The other is that various sensors (e.g., spectrum detection, environmental, and location sensors) passively collect and upload data to the central system. These sensors monitor non-source radiation and other environmental parameters, providing a comprehensive view of the electromagnetic environment. Recent advancements, such as dynamic channel modeling for fluid antenna systems in UAV communications [15] and closed-form channel statistics for UAV-to-ground links [16], provide sophisticated models that enhance the accuracy of PDTs for mobile devices.
• Propagation Environment Twin (PET): Electromagnetic waves are closely influenced by the propagation environment, and their characteristics vary in different environments, such as urban areas, deserts, forests, and oceans. Moreover, the ionosphere, weather conditions, temperature fluctuations, and atmospheric environment have diverse impacts on electromagnetic waves of various frequency bands. PET encompasses twin models incorporating geographical, architectural, and atmospheric environments from multiple perspectives. The integration of high-efficiency near-field channel analysis for large-scale MIMO [11] and RIS-enabled air-ground channels [12], along with robust wideband covariance estimation techniques for XL-MIMO systems [13], can significantly improve the accuracy and applicability of PET models under complex conditions.
• Electromagnetic Situation Twin (EST): EST is the twin of electromagnetic waves in diverse frequency bands, furnishing pertinent insights into their behaviors, channel conditions within the prevailing electromagnetic environment, signal intensity across the present spacetime domain, spectrum occupancy spanning various frequency ranges, and propagation effects of distinct electromagnetic waves. The application of advanced algorithms, such as the simplified learned approximate message passing network for efficient beamspace channel estimation [17], is crucial for achieving real-time and high-precision EST.

Within the three-layer architecture of the EDT, components form a closed-loop system through strict data flow and functional dependencies. Specifically, the PDT layer serves as the starting point for data acquisition. It collects real-time device parameters and electromagnetic environment data through a distributed sensor network. This raw data are transmitted via the DTN to the data processing center. The PET layer receives data uploaded from the PDT layer. It then combines geographic information systems (GIS), building models, and atmospheric environment parameters to perform high-precision simulations of electromagnetic wave propagation characteristics using near-field channel analysis models. This generates key parameters like path loss and multipath effects, outputting these environmental constraints to the EST layer. The EST layer ultimately fuses real-time PDT data with PET environmental simulation results through data-driven algorithms, dynamically generating a global electromagnetic situation map. It then feeds back optimization instructions to PDT devices at the physical layer via a virtual reality interactive interface.

Together, these three components of EDT form a cohesive framework that enables detailed modeling, analysis, and optimization of the electromagnetic space. By leveraging advanced computational techniques, such as machine learning and generative adversarial networks (GANs), EDT can dynamically adapt to changes in the physical environment and provide actionable insights for spectrum management. This innovative approach not only enhances the efficiency and reliability of electromagnetic spectrum utilization but also paves the way for future advancements in communication technologies.

4. Electromagnetic Twin Space Construction

EDT provides a flexible testing environment with high availability and reduces costly trial-and-error endeavors. We refer to the testing environment as ETS. An ETS, a digital replica depicting electromagnetic space within a specific area, serves as an online representation of the electromagnetic space, effectively monitoring current and historical radioactivity while possessing predictive capabilities for future radio behavior. ETS represents a futuristic tool for more intelligent and automated spectrum management that significantly enhances electromagnetic resource efficiency.

4.1. ETS Construction Process

In this section, we present the construction model of ETS based on urban scenarios. The model is illustrated in Figure 2. Buildings serve as the primary propagation environment in urban settings, while base stations and other sensor devices (collectively referred to as Tx and Measurable Devices or MCDs) are utilized to determine overall electromagnetic characteristics within the scenario; MCDs collect sample electromagnetic data and transmit it to ETS for analysis. PDT is applied to Tx and MCDs to construct ETS, while PET is applied to buildings to develop corresponding twin models. Subsequently, algorithms are designed for EST analyses based on these twin models and sampled data. The three EST processes are summarized below:

• Tx and Building Information: The Tx twin model incorporates the parameters of antenna direction, frequency, and position to derive fundamental electromagnetic wave propagation characteristics. However, the presence of buildings introduces complexities in the propagation process. An algorithm analyses the interaction between electromagnetic wave propagation, building information, and output electromagnetic situation. This scenario finds specific applications in base station coverage prediction, wireless network deployment, and network planning.
• Sample data and Building Information: When the Tx information is unknown, the feasibility of the above method diminishes. In such cases, many MCDs are required for sparse sampling. The twinning of MCDs involves not only twinning models of individual devices but also constructing DTN models to facilitate intra-twin communication between DTNs and inter-twin communication among DTNs to gather global sample data [14]. Furthermore, algorithms are designed to fit the distributed sample data and output the electromagnetic situation. This scenario specifically pertains to localizing illegal radiation sources.
• Tx, Building Information, and sample data: By disregarding illegal sources of radiation, Tx can establish the overall electromagnetic situation tone within the given scene, thereby serving as a foundation for incorporating sample data as supplementary features to enhance the electromagnetic situation further. This method combines both approaches and is specifically applicable in scenarios involving path planning and autonomous driving.

4.2. ETS Deployment Example

To further illustrate the role of ETS, we present three practical deployment examples.

• Base Station Coverage Prediction Scenario (corresponding to the “Tx and Building Information” method): By inputting base station parameters (including location, transmission power, and operating frequency band) alongside high-precision digital building models, the system employs ray tracing algorithms to simulate electromagnetic wave interactions with building structures (such as reflection, diffraction, and scattering). This generates coverage heatmaps to visualize signal strength distributions.
• Illegal Radiation Source Localization Scenario (corresponding to the “Sample Data and Building Information” method): The system relies on sparse sampling data from a distributed sensor network. It reconstructs the complete electromagnetic situation using an RBF interpolation algorithm and combines DTN to achieve multi-node data fusion and communication.
• Autonomous driving communication optimization scenario (corresponding to the “Tx, Building Information, and sample data” method): By integrating roadside unit parameters, building information, and real-time vehicle channel data collected through the VDT framework, the system dynamically adjusts beam steering and power allocation to counter channel abrupt changes caused by high-speed movement.

4.3. ETS Construction Framework

Based on the above construction process, we summarise the building framework of EST, which is shown in Figure 3. The ETS construction is divided into four main parts:

• Physical entity-based data collection: The ETS is a digital domain encompassing diverse data types. The collected data consists of parameters describing physical entities and sampled electromagnetic space data. These data support the foundation for the entire ETS modeling, updating, and prediction process. Physical devices will feature a dual-mode data acquisition mechanism comprising active upload mode and passive monitoring mode. In active upload mode, physical devices transmit parameters such as location, transmission power, and frequency band in real-time via the DTN network. This mode primarily maintains the dynamic nature of the ETS, enabling real-time visualization of the electromagnetic space. Passive monitoring mode involves physical devices collecting spectrum data within the space, primarily aimed at detecting the presence of unauthorized radiation sources or other anomalies in the electromagnetic space. Artificial intelligence (AI) can be used to optimize sensor network layout and data acquisition strategies. Using a graph neural network (GNN), DTN data quality and correlation are analyzed for the highest EST accuracy, and the DTN layout is optimized to reduce redundant data and improve the data acquisition efficiency. A single node in the DTN can identify anomalies and noises in the data using a self-encoder to ensure data accuracy and reliability.
• ETS modeling: In the context of electromagnetic wave propagation, numerous researchers have researched various models for wave propagation in different scenarios, laying a solid foundation for ETS modeling. However, traditional models for electromagnetic wave propagation exhibit limitations when applied in complex environments. Therefore, integrating data-driven and traditional mechanistic models is necessary to characterize the real-world electromagnetic space’s dynamic nature accurately. Furthermore, the model must possess self-updating and dynamic adjustment capabilities due to the rapid changes occurring within the electromagnetic space. AI can be used for model building and dynamic adjustment of the parameters of ETS models.GAN generates high-fidelity electromagnetic situation through adversarial training, and its Generator generates electromagnetic field distributions based on device parameters (e.g., transmit power, location) and environmental features (e.g., building layout); the Discriminator compares real sampling data with the generated results to The Discriminator dynamically optimizes the generation quality by comparing the real sampling data with the generation results to complete the dynamic scene adaptation. Physical Information Neural Network (PINN) realizes data-driven and mechanism integration. Physical constraints (e.g., electromagnetic field control equations) are introduced into the neural network loss function to ensure that the generation results conform to physical laws. Increase model interpretability by visualizing the gradient of physical constraint terms. A Long Short-Term Memory (LSTM) network analyzes time series data to predict the trend of the electromagnetic environment.
• ETS algorithm development: ETS aims to offer guidance in electromagnetics and make informed decisions in the electromagnetic domain. Following the completion of model construction, algorithms are developed encompassing fundamental data storage, processing, analysis, and mining, as well as AI-based prediction, inference, decision-making, and evaluation algorithms. Furthermore, virtual reality interaction algorithms, visualization, and display functionalities are devised alongside deduction and analysis capabilities. These algorithms and models undergo continuous iterative enhancements to effectively depict real-time data for diagnostic purposes, predictions, and informed decision-making within intelligent ETS. AI can help mine potential patterns and relationships in electromagnetic data for predictive analysis, decision support, and visualization. Random Forest, Support Vector Machines are used to mine potential patterns and relationships in electromagnetic data.YOLO, an Image Segmentation Network, realizes electromagnetic spatial information segmentation and image translation. Reinforcement Learning (RL) is used to design agents to maximize spectrum efficiency. State space (State) contains real-time spectrum occupancy, interference level, and device location, and Action space (Action) is the transmit power and frequency adjustment strategy.

  -

  Data Analysis and Mining: Specialized processing, analysis, and mining of data for the electromagnetic field are employed to enhance the refinement of ETS modeling. Furthermore, the analyzed data are utilized to update and fine-tune the model.

  -

  Visualization algorithm: The electromagnetic space can be considered a virtual space with the invisibility of electromagnetic waves. Designing versatile and multi-perspective visualization algorithms is essential to effectively display electromagnetic space in various dimensions. For instance, current visualization algorithms employ radio maps to depict the distribution of signal field strength. However, designing novel display formats that enable the representation of signal activity, multipath effects, etc., poses a significant challenge in this domain.

  -

  Diagnostics, Prediction, and Decision Making: The ETS is a comprehensive management platform that facilitates testing in controlled environments. The primary objective of establishing an ETS is to replace field testing, thereby minimizing the costs associated with trial and error. Consequently, the development of diagnostic, predictive, and decision-making algorithms ensures the efficacy of the ETS while offering valuable insights into real scenarios. For instance, AI can aid node deployment through electromagnetic data analysis.

• Real-virtual interaction: The real-virtual interaction is a prominent characteristic of digital twins and plays a fundamental role in maintaining their effectiveness. Given the high variability of electromagnetic space, the real-virtual interaction assumes great significance. The specific process involves the decision algorithm within the ETS generating behavioral instructions to guide the physical entity’s operations. Simultaneously, data collection by the physical entity facilitates closed-loop mechanisms within ETS and establishes an effective real-virtual interaction mechanism. AI enables real-time interaction between ETS and physical entities.TinyML frameworks are deployed on MCD devices, where the cloud is responsible for the global optimization of model training, and the edge performs real-time tuning for low-latency decision-making. Federation Learning to achieve privacy protection. Each device trains lightweight twin models locally and only uploads model parameters to the central server for aggregation, avoiding raw data leakage. The federated model is optimized by Meta-Learning to adapt to the differences in data distribution in different geographic regions. Reconfigurable intelligent surfaces can serve as the core component in PET systems. By leveraging the programmable units of RIS, they can adjust the phase in real time based on the electromagnetic situation predicted by PET, generating directional beam compensation signals to mitigate attenuation [20]. Furthermore, reinforcement learning can be employed to establish an adaptive beamforming interference method [21].

The Verification, Validation, and Accreditation (VV&A) are indispensable in establishing the ETS. VV&A encompasses evaluation metrics for each segment, which measure credibility and validity. The evaluation indicators for each segment are provided below:

• Data: Accuracy, age of information, and security.
• Model: Accuracy, stability, completeness, and evolutionary capability.
• Algorithm: Complexity, functional realization, degree of visualization, prediction accuracy, and decision enforceability.
• Behaviour: Implementation completion and implementation timelines.

5. Case Study: Electromagnetic Situation Twin Based on Generative Adversarial Networks

We consider well-defined scenarios for the Electromagnetic Situation Twin, where transmitter and building information are fully known. Specifically, we observe that a transmitter with identical parameters exhibits distinct electromagnetic situations across various locations within multiple building information maps. To address this, we employ a Generative Adversarial Network (GAN) in the Generative Artificial Intelligence (GAI) domain to perform EST under different situations. The overall framework diagram is depicted in Figure 4.

The primary objective of traditional AI is to perform Classification or Regression, making it a part of Discriminative Artificial Intelligence (DAI). In contrast, GAI distinguishes itself by generating novel content. Digital twins must be capable of conducting inference tests that align with the fundamental purpose of GAI. As a significant branch of GAI, GAN learns distribution information from training data and generates samples with similar distributions.

This paper selects the root mean square error (RMSE) and the normalized mean square error (NMSE) as the main evaluation indicators. RMSE directly reflects the absolute deviation between the predicted field strength values and the true values. NMSE eliminates the influence of units through normalization and is suitable for comparing the relative error levels under different power levels or frequency bands. The case study uses the RadioMapSeer dataset [22]. The RadioMapSeer dataset covers 6 cities, including Ankara, Berlin, and Glasgow, and contains 701 maps. The map resources come from OpenStreetMap [23], with a map size of $256\times 256$ m. The building density is continuously distributed from sparse suburbs to dense urban areas, effectively verifying the generalization ability of the model in heterogeneous environments. The data are generated based on the propagation model of WinProp [24], and the simulation accuracy is close to the measured values. Each map contains 80 transmitter positions, totaling 56,080 samples, meeting the data volume requirements of deep learning.

• Firstly, the transmitter parameters are selected, and multiple building information maps, transmitter position maps, sampling data, and their corresponding Electromagnetic Situations are input into the GAN for training.
• After the completion of training, when the GAN is provided with the new building information map and transmitter position map as inputs, it can generate the corresponding electromagnetic situation.
• In subsequent applications, this use case can be employed for test experiments in transmitter deployment scenarios. Analyzing the signal field strength distribution generated by the transmitter provides valuable information for optimizing transmitter placement and minimizing field experiment costs.

The results and performance evaluation of the Electromagnetic Situation Twin are depicted in Figure 5. From a visually intuitive perspective, the electromagnetic situation twin is conducted considering various building information and transmitter locations. The GAN-based approach effectively generates the electromagnetic situation under different scenarios. Furthermore, twinning accuracy is assessed by comparing several algorithms for electromagnetic situation construction, including Kriging, Radial Basis Function (RBF) interpolation, RadioUnet [22], RadioDiff [25] and RadioDiff-3D [26], which are considered classical methods in this context. The dataset used for evaluation is also derived from [22]. Root Mean Squared Error (RMSE) and normalized Mean Square Error (NMSE) are employed as metrics; RMSE represents the root mean square difference between predicted values and corresponding actual values, while NMSE provides a normalized measure of mean absolute error. Both metrics quantify the deviation between observed/predicted or actual values, reflecting prediction precision. More minor errors indicate higher accuracy of predictions. Simulation results demonstrate that GAN-based electromagnetic situation generation achieves significantly superior accuracy compared to other methods.

6. Open Challenges and Opportunities

Digital twins offer novel opportunities in the electromagnetic spectrum and establish a robust foundation for governing the electromagnetic space. In this paper, we introduce the concepts of EDT and ETS, along with their respective construction processes and frameworks. Nevertheless, specific unresolved challenges remain in realizing EDT and ETS.

Integration of Multi-Domain Interactions: The electromagnetic space is a complex interplay of physical devices, propagation environments, and electromagnetic conditions. Each element—a transmitter, the urban landscape, or atmospheric conditions—actively influences the electromagnetic situation. Integrating these multi-domain interactions into a cohesive EDT framework remains a significant challenge. Future research must focus on developing comprehensive models that capture these intricate relationships, ensuring that PDT, PET, and EST components can seamlessly interact to provide accurate and dynamic representations of the electromagnetic space.

Real-Time Data Synchronization and Processing: The dynamic nature of the electromagnetic space demands real-time data synchronization between the physical and digital domains. Achieving low-latency, high-fidelity data exchange is crucial for maintaining the accuracy and relevance of the EDT. Challenges include developing efficient data transmission protocols, optimizing sensor networks for real-time data collection, and implementing advanced processing algorithms capable of handling electromagnetic data’s high volume and velocity. Techniques such as edge computing and distributed processing may be pivotal in addressing these challenges.

Security and Privacy Concerns: As EDT relies heavily on data from various sources, ensuring the security and privacy of this information is paramount. Electromagnetic space is susceptible to cyber threats, and the integrity of the digital twin must be protected against malicious attacks. Future research should focus on developing robust security frameworks that safeguard data transmission, storage, and processing. Additionally, privacy-preserving mechanisms must be implemented to protect sensitive information, particularly in scenarios involving multiple stakeholders and cross-domain data sharing. To effectively address the security, privacy, and real-time challenges faced by the ETS framework in dynamic electromagnetic environments, a multi-tiered technical advancement plan can be designed for the future. For security and privacy protection, the system employs homomorphic encryption to enable field strength analysis of edge device data in encrypted form, ensuring “usable but not visible”. Secure multi-party computation supports collaborative analysis across multiple operators’ data without exposing raw data. The federated learning mechanism safeguards model gradients through differential privacy and secure aggregation protocols, while leveraging lightweight blockchain for data traceability verification, enhancing system credibility. For real-time data synchronization optimization, FPGA hardware acceleration reduces GAN inference latency. A lightweight algorithm employs Q-learning adaptive offloading strategies to dynamically schedule cloud-edge inference tasks, switching to compressed models during insufficient network bandwidth. This is complemented by an LSTM predictive synchronization mechanism that preemptively updates data, enhancing system robustness during unexpected events.

This article further discusses the limitations of actual deployment in dynamic electromagnetic environments.

• Real-time bottleneck and computational complexity challenge: In dynamic electromagnetic environments, the timeliness of data collection and processing is the core bottleneck. ETS needs to simultaneously process multi-source data (device parameters, environmental sensing, electromagnetic sampling). In urban scenarios, the amount of data generated per second can reach a terabyte level. Future network models need techniques such as model pruning and knowledge distillation to achieve lightweight processing.
• Limitations of uncertainty modeling in dynamic electromagnetic environments: The PET model is insufficient in adapting to instantaneous changes (such as vehicle occlusion, atmospheric changes). The electromagnetic space is a complex coupled system of “physical devices—propagation environment—electromagnetic situation”. In actual deployment, the propagation of electromagnetic waves is affected by multiple factors such as terrain, atmosphere, weather, and radio equipment. A small error in a single factor can be amplified through the PET-EST cascade, resulting in a significant reduction in prediction accuracy. Therefore, in actual deployment, it is necessary to enhance the model’s adaptability to uncertainty, i.e., to strengthen dynamic adaptability.
• Hardware deployment constraints and cost limitations: To achieve high-precision EST, theoretically, a large number of spectrum sensors need to be deployed, but the actual cost is unaffordable. Therefore, it is necessary to study efficient equipment deployment schemes to balance the relationship between actual deployment density, accuracy, and cost.
• Security and data fusion challenges: ETS relies on multi-party data sharing (such as operator base station parameters, third-party map data), which may expose sensitive information. In dynamic electromagnetic environments, attackers can inject false sensing data, causing ETS to make incorrect decisions.

7. Conclusions

This paper explores the potential of digital twin technology in addressing the challenges of electromagnetic spectrum management in the context of 5G/6G communications. By introducing the concepts of Electromagnetic Digital Twin (EDT) and Electromagnetic Twin Space (ETS), we propose a novel framework that enables high-fidelity dynamic mapping and real-time optimization of the electromagnetic space. The EDT framework, with its three-layer architecture of Physical Device Twin (PDT), Propagation Environment Twin (PET), and Electromagnetic Situation Twin (EST), offers a comprehensive approach to integrating digital twin technology into the electromagnetic domain. The ETS construction framework further clarifies the key processes and technologies involved in building and operating an electromagnetic twin space. Our case study demonstrates the effectiveness of a GAN-based approach for constructing the Electromagnetic Situation Twin (EST), achieving superior prediction accuracy compared to traditional methods. The results highlight the significant improvement in EST accuracy by incorporating building information and transmitter parameters, as evidenced by the reduced RMSE metrics. This capability allows for efficient experimentation and optimization of transmitter deployment, minimizing the need for costly field trials, while the proposed EDT and ETS frameworks show promising potential, several open challenges remain. Future work will further explore integrating model-driven and data-driven approaches to enhance the accuracy and adaptability of the ETS. Developing integrated sensing, communication, and computation frameworks and applying Reconfigurable Intelligent Surfaces (RIS) will also be investigated to optimize electromagnetic space management further. This research will advance the field of electromagnetic spectrum management and provide valuable insights for the broader application of digital twin technology in future communication systems.