• 한국음향학회지
• /
• 제39권5호
• /
• pp.476-482
• /
• 2020

입력열과 출력열의 길이가 다른 경우 attention 기법을 이용한 sequence-to-sequence 심층 신경망이 우수한 성능을 보인다. 그러나, 출력열의 길이에 비해서 입력열의 길이가 너무 긴 경우, 그리고 하나의 출력값에 해당하는 입력열의 특성이 변화하는 경우, 하나의 문맥 벡터(context vector)를 사용하는 기존의 attention 방법은 적당하지 않을 수 있다. 본 논문에서는 이러한 문제를 해결하기 위해서 입력열의 왼쪽 부분과 오른쪽 부분을 각각 개별적으로 처리할 수 있는 두 개의 문맥 벡터를 사용하는 이중 attention 기법을 제안한다. 제안한 방법의 효율성은 TIMIT 데이터를 사용한 음성 인식 실험을 통하여 검증하였다.

• 한국통신학회논문지
• /
• 제6권1호
• /
• pp.38-44
• /
• 1981

본 연구는 전국 테레비죤 난시청지역 해소를 위한 TV Translator의 試作에 관한 연구이다. 본 TV Translator의 제작에 있어서 그의 이론적 근거를 제시하였으며 특히 국부 발진기의 고조파 성분에 기인되는 상호 변조를 감소시킬 수 있는 방법을 제시하였다. 試作과정에서 광대역 중간주파증폭기에 서는 全 TV찬넬을 단일 중간주파대역만의 신호에 의해서는 상호간섭을 피할 수 없음이 밝혀졌다. 본 試作品의 특성으로는 입력신호레벨 -76~-26dBm 범위내에서 +30dBm의 일정출력이 유지되며 6MHZ 의 대역내에서 95% 이상의 직선성을 얻을 수 있었다.

• KDI Journal of Economic Policy
• /
• 제39권1호
• /
• pp.19-40
• /
• 2017

This paper analyzes the interaction between monetary and macroprudential policies mainly in the context of the non-cooperation among policy authorities. Each policy authority's optimal response is to tighten its policy measures when other authorities' policy measures are loosened. This indicates that the two policies are substitutes for each other. This result still holds when an additional financial stability mandate is assigned to the central bank. The condition for the response functions to converge to a Nash equilibrium state is analyzed along with the speed of convergence, showing that they depend on the authorities' preferences and the number of mandates assigned to policy authorities. If the financial supervisory authority (FSA) assigns greater importance to the output gap or a stronger financial stability mandate is assigned to the central bank (CB), the probability of non-convergence increases and the speed of convergence declines even when the condition of convergence is satisfied. Meanwhile, if the CB considers output stability as an important task, the probability of convergence and the speed of converging to a state of equilibrium are high. Finally, when a single mandate or small number of mandates is/are assigned to each authority, stability is more quickly restored as compared to when many mandates are assigned.

• 한국결정성장학회지
• /
• 제26권4호
• /
• pp.139-144
• /
• 2016

Tonpilz 트랜스듀서의 거동에 미치는 압전 소재의 형상과 크기의 효과에 대하여 유한요소법을 통한 컴퓨터 모사를 통하여 연구하였다. 도넛 형태의 압전체의 직경과 높이 그리고 쌓아올린 stack의 수를 변수로 고려하였고 최종적으로 소재 자체를 특성이 뛰어난 단결정 소재로 변환하였을 때 그 특성인 출력과 TVR 대역폭이 어떻게 변하는지를 모사하였다. 그 결과, 리랙서 단결정을 활용함으로써 같은 구조에서 출력을 10배 정도 향상시킬 수 있음을 확인하였다.

• Journal of Electrical Engineering and Technology
• /
• 제12권5호
• /
• pp.2043-2050
• /
• 2017

This paper presents two MPC (Model Predictive Control) based charging and discharging algorithms of BESS (Battery Energy Storage System) for capacity firming of wind generation. To deal with the intermittency of the output of wind generation, a single BESS is employed. The proposed algorithms not only make the output of combined systems of wind generation and BESS track the predefined reference, but also keep the SoC (State of Charge) of BESS within its physical limitation. Since the proposed algorithms are both presented in simple if-then statements which are the optimal solutions of related optimization problems, they are both easy to implement in a real-time system. Finally, simulations of the two strategies are done using a realistic wind farm library and a BESS model. The results on both simulations show that the proposed algorithms effectively achieve capacity firming while fulfilling all physical constraints.

• Journal of Power Electronics
• /
• 제14권3호
• /
• pp.432-443
• /
• 2014

In order to improve the output efficiency of solar cells and to extend the life span of batteries, the input currents of converters are required to be continuous. If low output voltage ripple is required at the same time, it is obvious that the application of basic two-order converters (such as Buck and Boost derived converters) will not be good enough. In this paper, a lot of non-isolated push-pull converters (NIPPCs) with continuous current will be introduced due to their lower current stress, higher efficiency and better EMC performance. By decomposing the converters into push-pull cells, inductor and free-wheeling diodes, two families of NIPPCs based on single inductor and coupled inductor separately are systematically generated. Furthermore, characteristics analyses for some of the generated converters are also shown in this paper. Finally, two prototypes based on the corresponding typical topologies are built in the lab to verify the theoretical outcomes.

• 한국멀티미디어학회논문지
• /
• 제20권1호
• /
• pp.32-42
• /
• 2017

This paper proposes a new haze removal method for moving image sequence. Since the conventional dark channel prior haze removal method adjusts each color component separately in RGB color space, there can be severe color distortion in the haze removed output image. In order to resolve this problem, this paper proposes a new haze removal scheme that adjusts luminance and saturation components in HLS color space while retaining hue component. Also the conventional dark channel prior haze removal method is developed to obtain best haze removal performance for a single image. Therefore, if it is applied to a moving image sequence, the estimated parameter values change rapidly and the haze removed output image sequence shows unnatural glitter defects. To overcome this problem, a new parameter estimation method using Kalman filter is proposed for moving image sequence. Experimental results demonstrate that the haze removal performance of the proposed method is better than that of the conventional dark channel prior method.

• Journal of Multimedia Information System
• /
• 제8권4호
• /
• pp.203-210
• /
• 2021

Deep leaning convolutional neural networks (CNN) have successfully been applied to image super-resolution (SR). Despite their great performances, SR techniques tend to focus on a certain upscale factor when training a particular model. Algorithms for single model multi-scale networks can easily be constructed if images are upscaled prior to input, but sub-pixel convolution upsampling works differently for each scale factor. Recent SR methods employ multi-scale and multi-path learning as a solution. However, this causes unshared parameters and unbalanced parameter distribution across various scale factors. We present a multi-scale single-path upsample module as a solution by exploiting the advantages of sub-pixel convolution and interpolation algorithms. The proposed model employs sub-pixel convolution for the highest scale factor among the learning upscale factors, and then utilize 1-dimension interpolation, compressing the learned features on the channel axis to match the desired output image size. Experiments are performed for the single-path upsample module, and compared to the multi-path upsample module. Based on the experimental results, the proposed algorithm reduces the upsample module's parameters by 24% and presents slightly to better performance compared to the previous algorithm.

• Journal of Communications and Networks
• /
• 제13권5호
• /
• pp.456-462
• /
• 2011

This paper investigates the outage behavior of peak interference power limited cognitive radio (CR) networks with multiple transmit antennas. In CR-multi-input single-output (MISO) channel, the total transmit power is distributed over the transmitantennas. First, we use the orthogonal space-time codes (STC) to achieve the transmit diversity at CR-receiver (rx) and investigate the effect of the power distribution on the interference power received at the primary-receiver (P-rx). Then, we investigate the transmit antenna selection (TAS) scheme in which the CR system selects the best transmit antenna and allocates all the power to the selected best antenna. Two transmit antenna selection strategies are proposed depending on if feedback channel is available or not. We derive the closed form expressions of outage probability and outage capacity of all schemes with arbitrary number of transmit-antennas. We show that the proposed schemes significantly improve the outage capacity over the single antenna systems in Rayleigh fading environment. We also show that TAS based scheme outperforms the STC based scheme when peak interference power constraint is imposed on the P-rx only if a feedback channel from CR-rx to CR-transmitter is available.

• 전력전자학회:학술대회논문집
• /
• 전력전자학회 2011년도 추계학술대회
• /
• pp.219-220
• /
• 2011

In this paper, 300[W] class boost type single phase inverter system which can compensate voltage sag was designed. If the voltage sag has appeared in input voltage, the boost converter would be operated to compensate it. The system is designed for that the THD of ouput voltage is below 5[%] and steady state error of output voltage is below 1[%]. The system was verified through experiments.