ETRI Journal最新文献

Time series predictions are commonly used in the fields of energy, meteorology, and finance, among others. The accurate prediction of time series data is critical for making decisions and planning. In the real world, non-stationary time series data with statistical properties shift over time, making prediction more challenging. Although, conventional time series processing methods—such as multi-scale feature extraction or Transformer-based algorithms—produce superior prediction results, when dealing with data that contain more noise and outliers, the prediction ability of such methods can suffer. To address this problem, we proposed the WPFormer model, which incorporated time-frequency analysis into the Transformer architecture to increase the long-term series prediction accuracy. The model employed wavelet packet decomposition to identify and eliminate noise efficiently, increasing its immunity to interference. We evaluated WPFormer on four publicly available datasets and compared its performance against the Informer, LogTrans, Reformer, LSTMa, LSTNet, and DeepAR models using MSE and MAE metrics. On average, the WPFormer model surpassed the benchmark models by 16%.

At the forefront of digital image processing, image super-resolution has emerged as a flourishing research area. However, despite remarkable progress, current methods still encounter significant hurdles, particularly when enhancing noisy images. To overcome this limitation, this study introduces a state-of-the-art super-resolution reconstruction technique called DenoSR, which leverages a pretrained diffusion model and is categorized as a zero-shot super-resolution reconstruction methodology. DenoSR, in its engagement with noisy images, progressively refines high-frequency image features through an inverse diffusion mechanism, thereby ensuring the accurate reconstruction of fine details. An exhaustive quantitative analysis conducted on publicly available benchmark datasets demonstrated that DenoSR outperformed existing methodologies in terms of image reconstruction quality. Furthermore, qualitative assessments corroborate the superiority of DenoSR in terms of reconstruction fidelity, highlighting significant advancements in enhancing the realism and naturalness of visual perception.

We analyze the secrecy attributes of an energy-harvesting-enabled decode-and-forward relay-based cooperative nonorthogonal multiple access (NOMA) system in the presence of an eavesdropper. Information flow through the wireless channel exposes legitimate users to eavesdropping by unintended users on confidential information. We study the secrecy performance of two-user NOMA systems by calculating the secrecy outage probability (SOP) and ergodic secrecy capacity (ESC). An eavesdropper overhears the signal from a source and relays it from a relay node. We evaluate the secrecy performance of both nodes (i.e., relay node ($�{�U�}_{�1�}$) and distant node ($�{�U�}_{�2�}$)) by developing a novel closed-form expression for node $�{�U�}_{�1�}$. For node $�{�U�}_{�2�}$, we evaluate an analytical but intractable expression of SOP and tractable analytical expression of ESC. We examine the effects of network parameters on secrecy and link outages. We also observe the reliability of the link outage probability of node $�{�U�}_{�2�}$ and system outage probability. Monte Carlo simulations are conducted to verify the analytical results and evaluate the effect of SOP on imperfect channel state information and perfect channel state information.

The development of multimodal machine translation (MMT) systems has attracted significant interest due to their potential to enhance translation accuracy with visual information. However, there are two limitations: (i) scarce large-scale corpus data in the form of (text, image, text) triplets and (ii) the semantic information learned by pre-training cannot transfer to multilingual translation tasks. To address these challenges, we propose a novel multimodal retrieval-augmented generation framework for machine translation, abbreviated as MRF-MT. Specifically, using the source text as a query, we retrieve relevant (image, text) pairs to guide image generation and feed the generated images into the image encoder of Multilingual Contrastive Language-Image Pre-training (M-CLIP) for learning visual information. Subsequently, we employ a projection network to transfer visual information learned by M-CLIP as a decoder prefix to Multilingual Bidirectional and Auto-Regressive Transformers (mBART) and train the mBART decoder using a two-stage pre-training pipeline. Initially, the mBART decoder is trained for image captioning with a visual–textual decoder prefix from M-CLIP's image encoder projection network. Subsequently, it undergoes training for caption translation, using prefixes from M-CLIP's text encoder. Extensive experiments show that MFR-MT achieves promising performance compared with baselines.

A rigorous analysis of a dielectric-coated conical antenna in the presence of a loop is presented in this paper. The poor matching in return loss characteristics of an isolated loop antenna has previously been improved using a conductive cone but with a narrow bandwidth. The current approach achieves 23.10% –10 dB bandwidth, compared to the 6.56% –10 dB bandwidth of a conductive conical antenna, resulting in a 252.13% –10 dB improvement. We analyze the return loss, radiation pattern, and −10 dB bandwidth by varying parameters such as wire diameter, loop radius, loop height from the cone's vertex, and permittivity. Simulated results are compared with computed results, demonstrating the effectiveness of the proposed design.

We experimentally demonstrate 100 Gb/s bidirectional transmission over 40 km using a multi-wavelength bidirectional optical sub-assembly (BOSA) based on a single bidirectional multi-wavelength Mux/Demux. The Mux/Demux consists of an optical zig-zag glass block and thin film filters. Four wavelengths with 800 GHz spacing (that is, two wavelengths for each direction) within a 20-nm band in the O-band are utilized, and there is no four-wave mixing penalty. The multi-wavelength BOSA enables the bidirectional transmission of 2 × 50 Gb/s PAM4 signals over a 40-km single-fiber link. The multi-wavelength BOSA with multi-level modulation presents a suitable approach for future speed and transmission distance increases in 6G networks and data-center interconnections.

The advent of deep-learning technology has enhanced the performance of biometric systems, including facial and fingerprint recognition systems. Although facial recognition is now commonly used for authenticating mobile device users, it can be easily falsified as it relies on external traits. In this study, we design a user-independent model for surface electromyogram (sEMG) biometric authentication using the convolutional Siamese network with an N-pair loss function. We then implement the shift-equivariant model by exploiting the convolution and padding operations to deal with small shifts in multichannel sEMG sensors for various users. Additionally, the augmentation methods for time-series data and spectrograms are used to further improve the performances of the model. We employ the public Gesture Recognition and Biometrics electroMyogram (GRABMyo) dataset, comprising 43 subjects and 16 gestures collected over 3 days, to train and evaluate the model. The proposed model achieves equal error rates of 5.62% and 8.94% for unknown subjects while preserving and leaking the gesture code, respectively. In cross-day experiments, the model achieves rates of 4.92% and 7.56%, respectively, demonstrating robustness to intersession variations.

To meet the growing demand for higher data speeds and denser network coverage, intelligent reflecting surfaces (IRS) have gained considerable attention owing to their ability to control electromagnetic waves with accurate beam orientation. However, an optimal IRS design for maximum wireless power transfer (WPT) is essential and requires the integration of tuning features into reflect arrays. Earlier studies employed varactor and PIN diodes for the tuning of metallic reflectors; by contrast, graphene offers inherent tuning without extra components owing to its unique material properties. We propose the design of optimal graphene-enabled IRS (GIRS) to maximize WPT in Internet of Things (IoT) communications. Specifically, we formulate an optimization problem to maximize the power received by an IoT user, thereby obtaining a global optimal solution in terms of the Fermi level of GIRS. Furthermore, the closed-form expression of the Fermi level was derived as a function of the reflection amplitude controlled by GIRS, incident frequency, and other GIRS parameters. Finally, we numerically validated all investigations and provide several insights into the necessary GIRS design parameters.

Multilayer inverters (MLIs) play an important role in their efficiency and effectiveness. This study proposes a new MLI that is optimally adapted using DQZ control and a vague neurological approach for tracking the single maximum power point of a hybrid renewable energy source. This MLI has a bidirectional fixed switch, the purpose of which is to reduce harmonics and increase the voltage level. The maximum power point tracking (MPPT) method proposed here is the only MPPT method that uses neuro-fuzzy control algorithms, making it superior to other methods. The proposed inverter consists of 12 power semiconductor switches (IGBTs) connected to three DC power sources—that is, photovoltaic, wind, and tidal energy power sources. The switching angle for pulse-width modulation can be calculated using the DQZ principle in the proposed MLI. Evaluation of the effectiveness of the proposed method uses MATLAB/Simulink simulations, the results being compared to those of the prototype mechanism. We also compare the performance of the MPPT algorithm and prototype mechanism, which is connected to a single-phase microgrid. The proposed method achieves total harmonic distortion (THD) efficiency with a satisfactory performance increase.

Data collection plays an essential role in underwater acoustic sensor networks (UASNs). To address the problem of underwater information collection, autonomous underwater vehicles (AUVs), which are dynamic and easy to reprogram, are expected to provide a feasible data-gathering solution. In this study, we examined covert data collection in AUV-assisted UASNs. Specifically, an AUV gathers covert information from all underwater sensor nodes (USNs) at the planned time, while an eavesdropper attempts to eavesdrop on this secret information. To improve the performance of UASNs, we formulate a complex optimization problem to maximize secrecy capacity under the constraints of the trajectory of the AUV, USN scheduling, connection outage probability, and secrecy outage probability. To solve the nonconvex problem, an efficient iterative optimization algorithm is proposed to optimize USN scheduling and AUV trajectories. Numerical results demonstrate the effectiveness of the proposed algorithm.