Digital twin can be defined as a virtual representation of a physical asset enabled through data and simulators for real-time prediction, optimization, monitoring, controlling, and improved decision making. Recent advances in computational pipelines, multiphysics solvers, artificial intelligence, big data cybernetics, data processing and management tools bring the promise of digital twins and their impact on society closer to reality. Digital twinning is now an important and emerging trend in many applications. Also referred to as a computational megamodel, device shadow, mirrored system, avatar or a synchronized virtual prototype, there can be no doubt that a digital twin plays a transformative role not only in how we design and operate cyber-physical intelligent systems, but also in how we advance the modularity of multi-disciplinary systems to tackle fundamental barriers not addressed by the current, evolutionary modeling practices. In this work, we review the recent status of methodologies and techniques related to the construction of digital twins mostly from a modeling perspective. Our aim is to provide a detailed coverage of the current challenges and enabling technologies along with recommendations and reflections for various stakeholders.

/pdf/digital-twin-values-challenges-and-enablers-from-a-modeling-3hpoelaogn.pdf

Digital Twin: Values, Challenges and Enablers From a Modeling Perspective

[신간안내] Computational Electrodynamics (the finite difference time - domain method)

The fifth generation (5G) mobile networks are envisaged to enable a plethora of breakthrough advancements in wireless technologies, providing support of a diverse set of services over a single platform. While the deployment of 5G systems is scaling up globally, it is time to look ahead for beyond 5G systems. This is mainly driven by the emerging societal trends, calling for fully automated systems and intelligent services supported by extended reality and haptics communications. To accommodate the stringent requirements of their prospective applications, which are data-driven and defined by extremely low-latency, ultra-reliable, fast and seamless wireless connectivity, research initiatives are currently focusing on a progressive roadmap towards the sixth generation (6G) networks, which are expected to bring transformative changes to this premise. In this article, we shed light on some of the major enabling technologies for 6G, which are expected to revolutionize the fundamental architectures of cellular networks and provide multiple homogeneous artificial intelligence-empowered services, including distributed communications, control, computing, sensing, and energy, from its core to its end nodes. In particular, the present paper aims to answer several 6G framework related questions: What are the driving forces for the development of 6G? How will the enabling technologies of 6G differ from those in 5G? What kind of applications and interactions will they support which would not be supported by 5G? We address these questions by presenting a comprehensive study of the 6G vision and outlining seven of its disruptive technologies, i.e., mmWave communications, terahertz communications, optical wireless communications, programmable metasurfaces, drone-based communications, backscatter communications and tactile internet, as well as their potential applications. Then, by leveraging the state-of-the-art literature surveyed for each technology, we discuss the associated requirements, key challenges, and open research problems. These discussions are thereafter used to open up the horizon for future research directions.

A Prospective Look: Key Enabling Technologies, Applications and Open Research Topics in 6G Networks

In recent years, deep learning (DL) is becoming an increasingly important tool for solving inverse scattering problems (ISPs). This paper reviews methods, promises, and pitfalls of deep learning as applied to ISPs. More specifically, we review several state-of-the-art methods of solving ISPs with DL, and we also offer some insights on how to combine neural networks with the knowledge of the underlying physics as well as traditional non-learning techniques. Despite the successes, DL also has its own challenges and limitations in solving ISPs. These fundamental questions are discussed, and possible suitable future research directions and countermeasures will be suggested.

/pdf/a-review-of-deep-learning-approaches-for-inverse-scattering-1dpsdp7162.pdf

A review of deep learning approaches for inverse scattering problems (invited review)

In this letter, a new deep learning (DL) approach is proposed to solve the electromagnetic inverse scattering (EMIS) problems. The conventional methods for solving inverse problems face various challenges including strong ill-conditions, high contrast, expensive computation cost, and unavoidable intrinsic nonlinearity. To overcome these issues, we propose a new two-step machine learning based approach. In the first step, a complex-valued deep convolutional neural network is employed to retrieve initial contrasts (permittivities) of dielectric scatterers from measured scattering data. In the second step, the previously obtained contrasts are input into a complex-valued deep residual convolutional neural network to refine the reconstruction of images. Consequently, the EMIS problem can be solved with much higher accuracy even for high-contrast objects. Numerical examples have demonstrated the capability of the newly proposed method with the improved accuracy. The proposed DL approach for EMIS problem serves a new path for realizing real-time quantitative microwave imaging for high-contrast objects.

Two-Step Enhanced Deep Learning Approach for Electromagnetic Inverse Scattering Problems

In this communication, we propose a new physics-constrained approach to solve 2-D inverse scattering problems (ISPs) by extending physics-informed supervised residual learning (PhiSRL) with Born approximation (BA). By embedding the fixed-point iteration method in residual neural network (ResNet), PhiSRL aims to solve ISPs iteratively by applying the convolutional neural networks (CNNs) to learn the update rules of reconstructions. PhiSRL is employed to invert lossy scatterers by introducing BA to linearize ISPs and further reduce the computational burden of forward modeling. Both numerical and experimental results validate the effectiveness of the proposed approach.

Physics-Informed Supervised Residual Learning for 2-D Inverse Scattering Problems

In this paper, we propose a neural contrast source iteration (Neural CSI) model to solve 2D inverse scattering problems. Inspired by the traditional contrast source iteration (CSI) method, we construct a residual convolution neural network (CNN) to learn the gradient update process. With our dataset (generated by the method of the moment), we train the residual CNN by the objective function of the traditional CSI method in an unsupervised manner.

Neural Contrast Source Inversion Method Based on Single-frequency Data

In this paper, we propose the physics-informed supervised residual learning (PISRL) which is a general framework for electromagnetic forward modeling based on deep learning. PISRL is designed to solve a system of linear matrix equations and not limited to a specific electromagnetic problem. Stationary and non-stationary iterative PISRL neural networks (SIPISRLNN and NSIPISRLNN) are designed based on the mathematical connection between residual neural network (ResNet)[1] and stationary or non-stationary iterative methods. With convolutional neural network mapping residuals and modifications, SIPISRLNN and NSIPISRLNN can improve the approximated solutions via iterative process, which mimics the procedure of solving linear equations by stationary and non-stationary iterative methods. The generalities of SIPISRLNN and NSIPISRLNN are validated by solving different types of volume integral equations (VIEs) with non-lossy and lossy scatterers. In numerical results of non-lossy scatterers, the mean squared errors (MSEs) of SIPISRLNN and NSIPISRLNN finally converge below $3.152 \times 10^{-4}$ and $4.8925 \times 10^{-7}$; in results of lossy scatterers, the MSEs converge below $1.2775 \times 10^{-4}$ and $1.567 \times 10^{-7}$. The generalization abilities of SIPISRLNN and NSIPISRLNN are verified by testing them on the data sets of contrast shapes and different incident frequencies that are unseen during the training process.

Physics-informed Supervised Residual Learning for 2D Electromagnetic Forward Modeling

—In the study of nonlinear inverse scattering problems (ISPs), the traditional method for estimating the number of degrees of freedom (NDoF) of the scattered ﬁeld does not consider the effect of the scatterer contrast. In this work, we study the relationship between NDoF and scatterer contrast by modifying the radiation matrix. Speciﬁcally, three numerical methods for calculating the NDoF are presented. The ﬁrst NDoF is deﬁned as the number of principle singular values of the external radiation matrix, which is used as a benchmark. The second NDoF is deﬁned based on the modiﬁed radiation matrix that incorporates the contrast of domain of investigation (DoI) with the external radiation matrix. By exploring the relationship between the second NDoF and contrast, an approximate upper bound of the second NDoF is obtained, that is, the third NDoF. The accuracy of the proposed NDoF is veriﬁed by interpolation experiments on both synthetic and experimental data using frequency-domain zero-padding method. Results show that the second NDoF is more accurate than traditional NDoF for high-contrast scatterers, and the third NDoF can be easily estimated by the traditional NDoF plus a positive constant, which provides a simple way to estimate the upper bound of NDoF.

Analysis of Degrees of Freedom in Scattered Fields for Nonlinear Inverse Scattering Problems

In this paper, we study an unsupervised coding scheme for reconfigurable intelligent surfaces (RISs) based on binary neural networks (BNNs). Inspired by the difference map algorithm (DMA) that considers the beamforming of RISs as a search for the intersection of sets, the proposed coding scheme emulates the iterative procedure of the DMA by employing the BNNs to learn the parametric update rule. This scheme requires only the beam pattern masks to guide the learning process of BNNs, instead of labeled phase distributions, which enables the unsupervised training. The applications of BNNs reduce the computational load and make the proposed scheme hardware-friendly. The effectiveness of this scheme is further verified on a 1-bit RIS with 2304 unit cells.

Tao Shan

Papers

Physics-Informed Supervised Residual Learning for 2-D Inverse Scattering Problems

Neural Contrast Source Inversion Method Based on Single-frequency Data

Physics-informed Supervised Residual Learning for 2D Electromagnetic Forward Modeling

Analysis of Degrees of Freedom in Scattered Fields for Nonlinear Inverse Scattering Problems

Hardware-friendly Unsupervised Coding Scheme for Reconfigurable Intelligent Surface Based on Binary Neural Networks