ETRI Journal最新文献

We first model the channel estimation in sixth-generation (6G) systems as a super-resolution problem and adopt a deep residual attention approach to learn the nontrivial mapping from the received measurement to the reconfigurable intelligent surface (RIS) channel. Subsequently, we design a deep residual attention-based channel estimation framework (DRA-Net) to exploit the RIS channel distribution characteristics. Furthermore, to transfer the RIS channel feature maps extracted from the residual attention blocks (RABs) to the end of the estimator for accurate reconstruction, we propose a novel and effective feature fusion approach. The simulation results demonstrate that the proposed DRA-Net-based channel estimation method outperforms other deep learning-based and conventional algorithms.

A long short-term memory (LSTM) model is introduced to predict missing datapoints of dissolved oxygen (DO) in an eel (Anguilla japonica) recirculating aquaculture system. Field experiments allow to determine periodic patterns in DO data corresponding to day–night cycles and a DO decrease after feeding. To improve the accuracy of DO prediction by using a training-to-test data ratio of 5:1, training with data in sequential and reverse orders is performed and evaluated. The LSTM model used to predict DO levels in the fish tank has an error of approximately 3.25%. The proposed LSTM model trained on DO data has a high applicability and may support water quality control in aquaculture farms.

Wearable devices for augmented reality (AR) have gained considerable attention but research on contact lenses remains scarce. This study introduces an AR display that utilizes a contact lens integrated with a holographic projection display and a curved volume-holographic optical element. We designed a holographic projection display with a total FoV 15°(supported FoV 10°) and a 6.9 mm $�\times$ 3.9 mm eyebox, respectively. The computer-generated hologram was optimized to null quality degradation caused by pupil constraints and asymmetric diffraction efficiency distribution. The diffraction efficiency of the holographic optical element used in the contact lens was calculated to provide insights into the anticipated optical performance and identify issues associated with the proposed system. Excellent performance with a maximum diffraction efficiency of 95% is demonstrated under Bragg conditions. However, a substantial drop in efficiency is observed at angular deviations 4°. Conversely, the wavelength deviations had a negligible impact on diffraction efficiency. Finally, the proposed contact lens holographic display was optically reconstructed, confirming its potential as a next-generation wearable display technology.

In this study, we investigated three-dimensional configurations involving a monopole-coupled wired ring, a hollow metallic cylinder, and wired helix spring to achieve resonance lower than quarter-wave resonance. The resonance frequency of a quarter-wave monopole, originally 2.05 GHz, was significantly lowered by 54.53% through effective coupling with a wired ring, and a quarter-wave monopole properly coupled with a hollow metallic cylinder experienced a large decrease of 64.34%. The primary incentive for connecting a wired ring to a hollow metallic cylinder is to significantly reduce resonance frequency. Efforts were also made to lower resonance frequency in a wired helix spring with a length close to that of the wired ring, resulting in a substantial 78.83% reduction in resonance frequency. The −10 dB bandwidth was 14.63% for the quarter-wave monopole, 27.21% for the monopole-coupled wired ring, 33.71% for the monopole-coupled hollow metallic cylinder, and 4.69% for the monopole-coupled helix spring. The ka values for the monopole-coupled wired ring, hollow metallic cylinder, and helix spring were 0.35, 0.27, and 0.16, respectively. Hence, these antennas can be classified as electrically small antennas. The achieved resonant frequencies have applications in various wireless communication scenarios. The experimental results from the fabricated prototype agree well with the simulation results.

Function-as-a-Service (FaaS) applications suffer from cold-start latency owing to their high execution time and memory usage. We propose a FaaS orchestration framework based on a cloud-agnostic approach for application and data interoperability. The framework's performance is evaluated on the web application of a student attendance system for online learning. The test system comprises 300 records from 50 students across five courses. Through optional file elimination, entry recognition, optional function generation, and function-level rewriting, the proposed orchestration framework suitably manages the storage of the student attendance records. The framework notably contributes to the student attendance web application by ensuring scalable, reliable, and efficient operations. Various evaluation measures confirm the effectiveness of the orchestration framework. These measures include cold/warm-start execution time, cold/warm-start memory usage, and zip execution time. An empirical analysis reveals that our framework reduces the execution time by 2 s –15 s and memory usage by 1MB–3 MB for the evaluated application.

This paper introduces a framework for optimizing deep-learning models on microcontrollers (MCUs) that is crucial in today's expanding embedded device market. We focus on model optimization techniques, particularly pruning and quantization, to enhance the performance of neural networks within the limited resources of MCUs. Our approach combines automatic iterative optimization and code generation, simplifying MCU model deployment without requiring extensive hardware knowledge. Based on experiments with architectures, such as ResNet-8 and MobileNet v2, our framework substantially reduces the model size and enhances inference speed that are crucial for MCU efficiency. Compared with TensorFlow Lite for MCUs, our optimizations for MobileNet v2 reduce static random-access memory use by 51%–57% and flash use by 17%–62%, while increasing inference speed by approximately 1.55 times. These advancements highlight the impact of our method on performance and memory efficiency, demonstrating its value in embedded artificial intelligence and broad applicability in MCU-based neural network optimization.

We propose a shuffling cluster sequence technique without separate side information (SI) for multiple-input multiple-output (MIMO) orthogonal frequency-division multiplexing (OFDM) systems. In the proposed technique, the active subcarriers over two consecutive OFDM symbols are divided into $�2�V$ clusters, and each cluster (packet frame) includes a header for ID#cluster and payload for $�D�a�t�a�\#�c�l�u�s�t�e�r$. The $�2�V$ clusters are shuffled to reduce the peak-to-average power ratio (PAPR) of the time-domain OFDM signal, which includes the information data and SI signals, with a low computational complexity. At the receiver, the information data can be correctly reconstructed by ID#cluster in the header of each cluster, achieving a smaller bit error rate than the conventional MIMO-OFDM system without PAPR reduction. Moreover, our technique is comparable with the conventional partial transmit sequence technique without the impact of a separate SI signal even when increasing the number of transmitter antennas in a nonlinear multipath fading channel.

This paper describes the detection of series arc faults, which constitute the major cause of electrical fires, in a power distribution system. Because the characteristics of series arc faults change considerably depending on the load type, their accurate detection and analysis are difficult. We propose a series-arc-fault detector that uses a transfer learning (TL)-based feature fusion model. The model is trained stagewise for various features in the time and frequency domains using a one-dimensional convolutional neural network combined with a long short-term memory model that uses an attention mechanism to accurately detect arc-fault features. To enhance the reliability of the proposed model, we implement an arc-fault generator compliant with the UL1699 standard and acquire high-quality data that suitably reflect the real environment. Experimental results show that the proposed model achieves an accuracy of 99.99% in classifying series arc faults for five different loads. Hence, a performance improvement of approximately 1.7% in classification accuracy is reached compared with a feature fusion model that does not incorporate TL-based model transfer and the attention mechanism.

Stretchable electronics are gaining attention with the development of skin-compatible health monitoring sensors and robotics applications. However, existing stretchable electronics show a tradeoff between reliability and stretchability when constructing transistor arrays. We aim to construct highly stretchable thin-film transistor (TFT) arrays with high reliability for large sensor backplanes. The fabricated structure comprises oxide TFTs and serpentine-shaped molybdenum interconnects that are sandwiched between top and bottom polyimide layers, achieving robustness under repeated tensile strain. Different materials for polyimide etch masks (i.e., aluminum, silicon nitride, and indium tin oxide) are tested. Within an 83.7 mm × 61 mm area, TFT arrays with a resolution of 25 pixels per inch can be stretched up to 30% without considerable performance degradation. As our TFTs are fabricated using common materials from display manufacturing, our development and findings may substantially benefit stretchable electronic industries where high reliability is required for large-area panels.

In this paper, we study on channel eigenvalue distribution of MIMO system, namely, DAS-to-DAS system, of which transmit and receive nodes are geometrically distributed in unplanned manner. It offers key to performance upper bound analysis of collaborative transmission in unplanned network and multi-user transmission of distributed antenna system. Previous studies on stochastic geometry have analyzed performance of single typical receiver, while transmitters are geometrically distributed. And random matrix theory has unveiled probabilistic behavior of MIMO channel eigenvalues, while transmit and receive antennas are collocated. Our study integrates knowledge from the two different research fields to extend the applicability to analysis on channel eigenvalue distribution of DAS-to-DAS system. As a preliminary research on such channel eigenvalue distribution, its first few moments are derived in analytic form. And its coherency of the analysis results with Monte-Carlo simulation results has examined.