Rydberg atomic receiver has recently emerged as a breakthrough technology for sensing and communications in next-generation Internet-of-Things (IoT) owing to its potential to surpass sensitivity limits of classical radio frequency (RF) receivers. In this paper, we consider a multiple-input multiple-output (MIMO) system with an RF transmitter and a Rydberg atomic receiver. Unlike prior works, a key technical innovation of our approach lies in the joint optimization of both transmit and receive processing techniques along with the design of reference signal injection according to a mean square error (MSE) criterion for signal recovery. However, the design problem is nonconvex on account of phase information loss in the received signal and nonlinearity of the objective function. To overcome this tricky challenge in an effective and intelligent manner, we propose a novel and high-performing deep learning (DL) framework calledDeepQ-MIMObased on the construction of an advanced DL network with innovative customization mechanisms. Numerical results confirm the supremacy and efficacy of the proposed DeepQ-MIMO system, and further provide useful design insights.

In this letter, we propose a novel and high-performing deep learning framework for remaining useful life (RUL) prediction, called RUL-Diff, by leveraging a generative diffusion model. It is composed of two modules that are connected in tandem: 1) a feature extractor corresponding to the encoder part of our customized U-Net and 2) a RUL predictor constructed by a multilayer perceptron. We further devise an effective two-stage training methodology for the proposed RUL-Diff, in which the feature extractor is initially pretrained for high-quality feature learning, and then, is retrained jointly with the RUL predictor for accurate RUL prediction. Extensive experimental results on NASA Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) datasets demonstrate the superiority and effectiveness of the proposed scheme.

pinching antenna system (PASS) has been demonstrated as a feasible flexible-antenna technology for upcoming 6G wireless networks and Internet of Things (IoT). In this article, we investigate a new design problem on capacity maximization for a point-to-point multiple-input–multiple-output (MIMO) PASS in a realistic IoT environment by jointly optimizing precoding matrix and antenna positioning. Unfortunately, this problem is not mathematically tractable. To break through this challenge in an effective and intelligent manner, we propose a novel and high-performing deep learning framework, named CaMPASS-Net, based on an advanced dual-stream network architecture with a residual connection, inspired by our insight into the problem. Furthermore, we present an effective unsupervised training strategy for the proposed CaMPASS-Net based on an innovative loss function design. Simulation results confirm that the proposed CaMPASS-Net exhibits remarkable performance improvements over baseline and existing schemes.

Super-resolution (SR) of remote sensing imagery based on generative AI models is vital for practical applications such as urban planning and disaster assessment. However, current approaches suffer from poor performance trade-offs among the pivotal, yet competing, objectives: perceptual quality, factual accuracy, and inference speed. To break through this limitation, we propose a novel and high-performing two-stage SR framework for the remote sensing imagery based on a generative diffusion model. First, in Stage 1, factually grounded base images are generated by employing a guidance-free diffusion process relying solely on the original low-resolution images, such that the risk of semantic hallucination can be effectively mitigated. The generated images are refined subsequently in Stage 2 such that high-frequency details for SR quality can be restored via our customized and innovative guidance mechanism with a vision–language model (VLM) and a ControlNet, and a dynamic inference acceleration technique is applied to ensure efficiency. Extensive experimental results confirm that our proposed framework excels in perceptual quality—achieving top CLIP-IQA scores—and in structural integrity while achieving robust performance. In particular, it enables reliable, high-fidelity SR for large-scale, real-world remote sensing pipelines by surpassing the conventional fidelity–hallucination trade-off at practical inference speed. Source code is available at https://github.com/Bluear7878/Remote-Sensing-Vision-Language-Diffusion-Model.

The precise and efficient extraction of buildings and road networks from aerial imagery is crucial for remote sensing applications, such as map building, urban development, and autonomous driving guidance systems. However, accurately extracting buildings and roads from remote sensing imagery is challenging due to varying resolutions, object scale variation, and diverse appearances of buildings and roads. While sophisticated models achieve high accuracy, their computational demands limit practical use on resource-constrained devices. Conversely, mainstream lightweight models often fail to generate high-quality segmentation maps for remote sensing data. To address these challenges, we introduce a dual attention selective kernel network (DASK-Net), a novel lightweight architecture for efficient pixelwise dense prediction. DASK-Net’s core features a dual attention selective kernel (DASK) module that integrates multiscale feature extraction with adaptive receptive fields and dual attention mechanisms. This design captures diverse scales and orientations of features while focusing on salient input aspects. We conducted experiments on the Massachusetts roads, DeepGlobe and WHU building datasets, comparing DASK-Net with numerous methods. The results demonstrate that DASK-Net outperforms these networks while significantly reducing computational complexity. With only 0.48M parameters, DASK-Net achieves an 89.54% IoU and a 94.48% F1 score on the WHU building dataset, setting new performance standards for lightweight methods.

Presents corrections to the paper, (Corrections to “DASK-Net: A Lightweight Dual-Attention Selective Kernel Network for Efficient Dense Prediction in Remote Sensing Imagery”).

Optical character recognition (OCR) is one of the key enabling technologies in industrial internet-of-things (IIoT) for extracting and utilizing useful textual information, but it is technically challenging due to poor environmental conditions. To deal with such challenges, in this letter, we propose a novel two-stage deep learning framework for OCR using a generative diffusion model, namely, OCR-Diff. In the first stage, our customized conditional U-Net is pre-trained jointly with a feature extractor with the aid of the forward diffusion process such that the quality of a low-resolution text image is improved via the reverse diffusion process. In the next stage, the pre-trained conditional U-Net and feature extractor are jointly fine-tuned for an off-the-shelf text recognizer to precisely recognize the texts in the image. Experimental results on TextZoom datasets substantiate the superiority and effectiveness of the proposed scheme.

One of the most critical challenges of federated learning (FL) is to send data efficiently and reliably over the noisy wireless channels between the clients and server to achieve target learning accuracy as fast as possible. To achieve this goal, we design effective error correction coded FL with managed retransmissions. Rather than using Shannon capacity as the performance measure to design the communication mechanisms for FL, our approach relies critically on learning accuracy. Our fundamental idea is based on the observation that Stochastic Gradient Decent (SGD) and its family can tolerate some errors in the course of training. Inspired by this, to reduce the communication burden without degrading the learning accuracy, our FL framework with Managed Redundancy (FL-MR) has two phases: (i) the No-Retransmission phase, where retransmissions are never performed even in case of erroneous decoding of data and (ii) the Select Retransmission phase, where only some carefully selected data packets are retransmitted. Our extensive simulation results demonstrate that the proposed coded FL system achieves target accuracies much faster than the baseline coded approach.

This work studies the cooperative inference of deep neural networks (DNNs), in which a memory-constrained end device performs a delay-constrained inference process with an aid of an edge server. Although several works considered the cooperative inference of DNNs in the literature, it was assumed in those works that the memory footprints at end devices are unlimited, which is in practice not realistic. To address this issue, in this work, a memory-aware cooperative DNN inference is proposed. Specifically, we propose to adopt knowledge distillation to obtain high-performing lightweight DNNs. To minimize the inference delay, we first analyze the end-to-end delay required for processing the proposed cooperative DNN inference, and then we minimize the delay by jointly optimizing the DNN partitioning point and the intermediate data transmission rate. Also, a dynamic DNN selection scheme is developed by fully exploiting the available memory resource in order to maximize the performance of the inference task in terms of inference accuracy. Experimental results demonstrate that the proposed cooperative DNN inference considerably outperforms the comparable schemes while satisfying both the delay constraint and the memory constraint.

This letter studies deep learning-based efficient ground vibration monitoring systems. In this work, artificial intelligence (AI) techniques are adopted to effectively deal with practical issues of data collection and classification. Specifically, we develop a novel energy-efficient data collection scheme by adopting deep Q-network-based reinforcement learning. Also, we propose an enhanced joint recurrent neural network (RNN) and convolutional neural network (CNN) approach for ground vibration classification. The performance of the proposed scheme is evaluated using real-world ground vibration data. The experimental results show that the proposed classification scheme outperforms the best existing scheme with CNN by more than 13% in terms of classification accuracy. It is also shown that the proposed energy management scheme can improve the accuracy of the proposed ground vibration monitoring system by 7.6% over the comparable scheme using equal power allocation.

Federated Learning (FL) has recently received a lot of attention for large-scale privacy-preserving machine learning. However, high communication overheads due to frequent gradient transmissions decelerate FL. To mitigate the communication overheads, two main techniques have been studied: (i) local update of weights characterizing the trade-off between communication and computation and (ii) gradient compression characterizing the trade-off between communication and precision. To the best of our knowledge, studying and balancing those two trade-offsjointly and dynamicallywhile considering their impacts on convergence has remained unresolved even though it promises significantly faster FL. In this paper, we first formulate our problem to minimize learning error with respect to two variables:local update coefficientsandsparsity budgetsof gradient compression who characterize trade-offs between communication and computation/precision, respectively. We then derive an upper bound of the learning error in a given wall-clock time considering the interdependency between the two variables. Based on this theoretical analysis, we propose an enhanced FL scheme, namely Fast FL (FFL), that jointly and dynamically adjusts the two variables to minimize the learning error. We demonstrate that FFL consistently achieves higher accuracies faster than similar schemes existing in the literature.

This paper studies a wireless powered relaying (WPR) system in which a source transmits an information signal to a destination with the help of an energy constrained relay in the presence of an eavesdropper. This work considers a scenario in the full-duplex WPR system where the information transmission/reception and the energy harvesting are performed simultaneously at the relay. We first derive an achievable secrecy rate of the system and formulate a joint optimization problem of artificial-noise-aided secure beamforming to maximize the achievable secrecy rate. Then, an efficient semi-definite relaxation (SDR) based algorithm is proposed to obtain the optimal beamforming. Furthermore, we also propose three suboptimal beamforming but low computational complexity schemes. Comparisons are carried out among the performances of the proposed optimal and suboptimal schemes.

This paper studies a secure communication over multiple-input multiple-output multi-antenna eavesdropper (MIMOME) channels which consist of legitimate parities, namely a transmitter and receiver, equipped with multi-antennas and a passive eavesdropper with multi-antennas. For securing the communication between the legitimate parties, we consider an artificial noise (AN) scheme in which the legitimate transmitter sends its secret messages and AN signal together. Meanwhile, it is assumed that the legitimate receiver has a full-duplex capability which enables it to capture the secret message from the transmitter and simultaneously generate a jamming signal to strengthen security. While there have been studies on similar setups, most of them focus on the design of the jamming signal with the assumption of full channel-state information (CSI) and/or system parameter optimization based on numerical evaluations. On the contrary, in this work, we instead introduce a tight lower bound on an achievable ergodic secrecy rate as a versatile analytic tool and derive a closed-form expressions for the bound. To confirm the analytic results, we carry out numerical evaluations of the ergodic secrecy rate which are compared with the proposed lower bound.

Biologically inspired pulse-coupled synchronization has received increasing attention as one of key techniques for developing decentralized wireless networks due to its inherent scalability and simplicity. While it has been actively studied in recent years, most of the previous works have focused on the synchronization only in fault/attack free networks. However, in reality, a network may have malfunctioning nodes and/or attackers, and thus robustness against these threats in the pulse- coupled synchronization must be an important practical issue on its path to realization. Motivated by this, we analyze the influence of attacker's behaviors in a network of pulse-coupled oscillators. The analysis shows that when the number of attackers is less than a certain number, i.e. a threshold, the network sustains its synchronization. Moreover, the threshold is proportional to both of coupling strengths among the pulse-coupled oscillators and the number of legitimate oscillators. Finally, numerical experiments are given to confirm the analytic results.

This paper introduces the system development of the ubiquitous computing environment, called easy interaction room (EIR), in which humans can get intelligent robotic services with easy interaction. The room has been built to research the roles between the ubicomp environment and service robot in it. Also EIR system collaborates on a special home service with the monolithic robot platform. In this development, cutting-edge technologies are employed on the design and implementation of hardware infrastructure for the automated environment. Intelligent Robot Software Architecture, developed by CIR for reusable and extensible architecture, is used to build EIR's software architecture for rapid development. Preliminary correlations between EIR and a service robot have been designed and implemented on the architectures.

The robots operating in hazardous environments such as nuclear plant decommissioning or terrorist bomb disposal need to show off-road capabilities due to the uneven terrain including stairs and ditches. The hazardous environment operation typically requires the use of a remotely operated robot. In this paper we propose an improved articulated track mechanism. A sensing system for haptic operation with a joystick is proposed considering the robot contacts to the stair. Additionally, we present the algorithm for the proposed robot to climb the stairs. Torque analysis is made to design electrical motors, which produce torque enough to drive the track lifting motor and track rotating motor.