A5100331815- Advanced Wireless Communication Techniques
- Wireless Communication Networks Research
- Advanced MIMO Systems Optimization
- Cooperative Communication and Network Coding
- Error Correcting Code Techniques
- Advanced Image Processing Techniques
- PAPR reduction in OFDM
- Wireless Communication Security Techniques
- Advanced Neural Network Applications
- Software Testing and Debugging Techniques
- Context-Aware Activity Recognition Systems
- Software Engineering Research
- Advanced Wireless Network Optimization
- Advanced Vision and Imaging
- Image and Signal Denoising Methods
- Domain Adaptation and Few-Shot Learning
- Service-Oriented Architecture and Web Services
- Software Reliability and Analysis Research
- Technology and Data Analysis
- Image Processing Techniques and Applications
- Power Line Communications and Noise
- Korean Urban and Social Studies
- Speech and Audio Processing
- Green IT and Sustainability
- Software System Performance and Reliability
Animal and Plant Quarantine Agency
2021-2024
Ajou University
2014-2023
Samsung (South Korea)
2008-2023
Samsung (United States)
2013-2022
Kyungpook National University
2006-2022
Korea Advanced Institute of Science and Technology
2021
Ewha Womans University
2004-2020
Keimyung University
2020
Agency for Defense Development
2018-2020
ON Semiconductor (United States)
2016-2019
We propose a deep neural network fusion architecture for fast and robust pedestrian detection. The proposed allows parallel processing of multiple networks speed. A single shot convolutional is trained as object detector to generate all possible candidates different sizes occlusions. This outputs large variety cover the majority ground-truth pedestrians while also introducing number false positives. Next, are used in further refinement these candidates. introduce soft-rejection based method...
In this paper, we propose a novel variable-rate learned image compression framework with conditional autoencoder. Previous learning-based methods mostly require training separate networks for different rates so they can yield compressed images of varying quality. contrast, train and deploy only one network implemented We provide two rate control parameters, i.e., the Lagrange multiplier quantization bin size, which are given as conditioning variables to network. Coarse adaptation target is...
As 4G cellular systems densify their cell deployment, co-channel interference becomes a major source of obstacles to throughput improvement. In addition, edge users suffer more from interference, which may govern end users, experiences. Although some network-side solutions for management have been introduced in current standards, it turns out that most those yield only meager gains realistic environments. this article, we pay attention recent advances the network information theory and...
In this paper, a novel architecture for deep recurrent neural network, residual LSTM is introduced.A plain has an internal memory cell that can learn long term dependencies of sequential data.It also provides temporal shortcut path to avoid vanishing or exploding gradients in the domain.The additional spatial from lower layers efficient training networks with multiple layers.Compared previous work, highway LSTM, separates one by using output layers, which help conflict between and...
Transformer neural networks (TNN) demonstrated state-ofart performance on many natural language processing (NLP) tasks, replacing recurrent (RNNs), such as LSTMs or GRUs. However, TNNs did not perform well in speech enhancement, whose contextual nature is different than NLP like machine translation. Self-attention a core building block of the Transformer, which only enables parallelization sequence computation, but also provides constant path length between symbols that essential to learning...
This paper reviews the NTIRE 2020 challenge on real world super-resolution. It focuses participating methods and final results. The addresses setting, where paired true high low-resolution images are unavailable. For training, only one set of source input is therefore provided along with a unpaired high-quality target images. In Track 1: Image Processing artifacts, aim to super-resolve synthetically generated image processing artifacts. allows for quantitative benchmarking approaches w.r.t....
In this paper, we focus on constructing an accurate super resolution system based multiple Convolution Neural Networks (CNNs). Each individual CNN is trained separately with different network structure. A Context-wise Network Fusion (CNF) approach proposed to integrate the outputs of networks by additional convolution layers. With fine-tuning whole fused network, accuracy significantly improved compared networks. We also discuss other fusion schemes, including Pixel-Wise (PWF) and...
This paper defines a rate maximization power allocation game in frequency selective Gaus- sian interference channel, after assuming suboptimal but pragmatic multi-user coding scheme. We show that the Nash equilibrium always exists this game. consider distributed scheme (l, Section IV.), and provide sufficient condition for convergence of algorithm. The condi- tion is also under which unique. I. PROBLEM FORMULATION
The commercial deployment of LTE Release 8 is gaining significant momentum all over the globe, and evolving to LTE-Advanced, which offers various new features meet or exceed IMT-Advanced requirements. Since LTE-Advanced targets ambitious spectral efficiency peak throughput, it poses tremendous system design challenges operators manufacturers, especially for mobile terminals. This article discusses modem issues related carrier aggregation, enhanced ICIC HetNet, detection eight-layer...
Benefiting from the recent real image dataset, learning- based approaches have achieved good performance for real-image denoising. To further improve Bayer raw data denoising, this paper introduces two new networks, which are multi-scale residual dense network (MRDN) and cascaded U-Net with block-connection (MCU-Net). Both networks built upon a newly designed block (MRDB), MCU-Net uses MRDB to connect encoder decoder of U-Net. better exploit feature images, adds another branch atrous spatial...
This paper addresses the design of optimal and near-optimal detectors in an interference channel with fading additive white Gaussian noise (AWGN), where transmitters employ discrete modulation schemes as practical communication scenarios.The conventional typically either ignore or successively detect then cancel interference, assuming that desired signal and/or are Gaussian.This quantifies significant performance gain can be obtained if explicitly take into account formats signals.This first...
In order to meet the requirements of 4G mobile networks targeted by cellular layer IMT-advanced, next generation WiMAX devices based on IEEE 802.16m will incorporate sophisticated signal processing, seamless handover functionalities between heterogeneous technologies and advanced mobility mechanisms. This survey provides a description key projected features physical (PHY) medium access control (MAC) layers 802.16m, as major candidate for providing aggregate rates at range Gbps high-speed...
We report simultaneous multifrequency observing performance at 22 and 43 GHz of the 21 m shaped-Cassegrain radio telescopes Korean VLBI Network (KVN). KVN is first millimeter-dedicated network in Korea having a maximum baseline length 480 km. It currently operates planned to operate four frequency bands: 22, 43, 86, 129 GHz. The unique quasi optics enable observations based on efficient beam filtering accurate antenna-beam alignment found that offset beams within less than 5'' over all...
In this paper, a new deep learning architecture for stereo disparity estimation is proposed. The proposed atrous multiscale network (AMNet) adopts an efficient feature extractor with depthwise-separable convolutions and extended cost volume that deploys novel matching costs on the features. A stacked to aggregate rich contextual information from which allows estimating high accuracy at multiple scales. AMNet can be further modified foreground-background aware network, FBA-AMNet, capable of...
This article presents an open-switch fault detection method for a hybrid active neutral-point clamped (HANPC) inverter based on deep learning technology. The HANPC generates three-level output voltage with four silicon switches and two carbide per phase. probability of open in switching devices increases because the large number entire power converter. causes distortion currents. A convolution neural network (CNN) comprising several layers fully connected is used to extract features...
Deep-leaming based acoustic echo cancellation (AEC) methods have been shown to outperform the classical techniques. The main drawback of learning-based AEC is its dependency on training set, which limits practical deployment in mobile devices and unconstrained environments. This paper proposes a context- aware deep (CAD-AEC) by introducing two components. first component CAD-AEC borrows ideas from performs frequency domain adaptive filtering microphone signal, provide network with features...
Abstract. A system of nonlinear equations governing the transmission uni-axial waves in a cold collisionless plasma subject to transverse magnetic field is reduced recently proposed resonant Schrödinger (RNLS) equation. This integrable variant standard equation admits novel superposition principles associated with Bäcklund–Darboux transformations. These are used here, particular, construct analytic descriptions interaction solitonic magnetoacoustic propagating through plasma.
This letter analyzes the effect of carrier frequency offset on orthogonal division multiplexing (OFDM) systems for multipath fading channels. A simple approximate expression average signal-to-noise ratio (SNR) is derived. shown to be an upper bound SNR flat channels and exact AWGN channel. The validated using Monte Carlo simulation both frequency-selective
This paper proposes two novel knowledge transfer techniques for class-incremental learning (CIL). First, we propose data-free generative replay (DF-GR) to mitigate catastrophic forgetting in CIL by using synthetic samples from a model. In the conventional replay, model is pre-trained old data and shared extra memory later incremental learning. our proposed DF-GR, train scratch without any training data, based on classification past, so curtail cost of sharing models. Second, introduce...
When a preamble signal is repeated multiple times in OFDM systems, we derive joint maximum likelihood (ML) estimation of the frame timing (FT) and carrier frequency offset (CFO). Unlike conventional estimators which use correlation adjacent repetition patterns only or some specific sets patterns, ML exploits any pair providing optimized performance. To reduce implementation complexity involved estimation, also propose near-ML method that separates FT CFO. The performance proposed methods...
The paper presents an exact analysis of the effect carrier frequency offset on orthogonal division multiplexing (OFDM) systems for a general multipath fading channel. As is well known, attenuates desired signal and causes intercarrier interference, thus reducing signal-to-noise ratio (SNR). SNR degradation due to evaluated by deriving expression in presence offset. can be used design practical OFDM system determining how small should order maintain negligible levels.
Multiple-input multiple-output (MIMO) systems with hybrid automatic-repeat-request (HARQ) promises high throughput reliability. However, combining-scheme design for such faces challenges, the presence of interference and existence multiple signal-to-interference-and-noise power ratios (SINRs). To overcome these this paper suggests to combining schemes objective directly optimizing log-likelihood ratio (LLR) values. Using approach, proposes several then analyzes them based on three key...
A successful software development project becomes an essential part of a company's reputation. Thus, lots managers focus more on maintenance than other management processes. Previous works studied how to help the process by detecting bug duplication and predicting severity bugs. This paper continues that kind special work analyzing emotion words for bug-severity prediction. In detail, we construct words-based dictionary verifying reports' textual analyses based positive negative terms. Then,...