• Journal of the Korean Institute of Intelligent Systems
• /
• v.15 no.5
• /
• pp.552-559
• /
• 2005

Recurrent neural networks(RNNs) trained with gradient based such as real time recurrent learning(RTRL) has a drawback of slor convergence rate. This algorithm also needs the derivative calculation which is not trivialized in error back propagation process. In this paper a derivative free Kalman filter, so called the unscented Kalman filter(UKF), for training a fully connected RNN is presented in a state space formulation of the system. A derivative free Kalman filler learning algorithm makes the RNN have fast convergence speed and good tracking performance without the derivative computation. Through experiments of nonlinear channel equalization, performance of the RNNs with a derivative free Kalman filter teaming algorithm is evaluated.

• Journal of the Korea Society for Simulation
• /
• v.27 no.4
• /
• pp.27-45
• /
• 2018

The terrestrial communication network and the underwater acoustic communication network have very different communication characteristics each other in operational environments, communication media, propagation delay, frequency bandwidth, transmission speed, bit error rate, and so on. These different characteristics cause some different address schemes and different maximum transmission units and, as a result, these differences must form certainly obstacles to the intercommunication between a terrestrial communication network and an underwater acoustic communication network. In this paper, we presents a method to use the virtual addresses to resolve the interconnection problem caused by different address schemes between a terrestrial communication network and an underwater acoustic communication network, and, through a mathematical modeling, we analyze the performance on the message transceiving delay time in the underwater environment.

• Phonetics and Speech Sciences
• /
• v.13 no.1
• /
• pp.45-51
• /
• 2021

End-to-end (E2E) automatic speech recognition (ASR) has achieved promising performance gains with the introduced self-attention network, Transformer. However, due to training time and the number of hyperparameters, finding the optimal hyperparameter set is computationally expensive. This paper investigates the impact of hyperparameters in the Transformer network to answer two questions: which hyperparameter plays a critical role in the task performance and training speed. The Transformer network for training has two encoder and decoder networks combined with Connectionist Temporal Classification (CTC). We have trained the model with Wall Street Journal (WSJ) SI-284 and tested on devl93 and eval92. Seventeen hyperparameters were selected from the ESPnet training configuration, and varying ranges of values were used for experiments. The result shows that "num blocks" and "linear units" hyperparameters in the encoder and decoder networks reduce Word Error Rate (WER) significantly. However, performance gain is more prominent when they are altered in the encoder network. Training duration also linearly increased as "num blocks" and "linear units" hyperparameters' values grow. Based on the experimental results, we collected the optimal values from each hyperparameter and reduced the WER up to 2.9/1.9 from dev93 and eval93 respectively.

• Journal of the Korean Society of Manufacturing Process Engineers
• /
• v.19 no.10
• /
• pp.1-7
• /
• 2020

In this study, a multichannel gas-liquid microextraction system is designed by integrating the automatic elution of extraction line and multichannel gas-purging liquid phase microextraction. The system uses an injection pump and inert gas to push the extraction solvent to a sample bottle of a gas-phase color autosampler and then implements multichannel gas-liquid microextraction and gas chromatography-mass spectrometry. The system also employs a three-way integrated micro-high-temperature heater, syringe pump, and microflow controller to realize the simultaneous processing of multiple groups of samples, thus improving the sample pretreatment speed and reproducibility and reducing human error. Autoinjection experiments were implemented with polycyclic aromatic hydrocarbon standard samples. The experiments show that the average recovery rate of the system exceeds 70%, and the relative standard among the channels is less than 15%.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.4
• /
• pp.1095-1109
• /
• 2022

Along with the popularity of GPS system and smart cell phone, trajectories of pedestrians or vehicles are recorded at any time. The great amount of works had been carried out in order to discover traffic paradigms or other regular patterns buried in the huge trajectory dataset. The core of the mining algorithm is how to evaluate the similarity, that is, the "distance", between trajectories appropriately, then the mining results will be accordance to the reality. Euclidean distance is commonly used in the lots of existed algorithms to measure the similarity, however, the trend of trajectories is usually ignored during the measurement. In this paper, a novel segment transform distance (STD) algorithm is proposed, in which a rule system of line segment transformation is established. The similarity of two-line segments is quantified by the cost of line segment transformation. Further, an improvement of STD, named ST-DTW, is advanced with the use of the traditional method dynamic time warping algorithm (DTW), accelerating the speed of calculating STD. The experimental results show that the error rate of ST-DTW algorithm is 53.97%, which is lower than that of the LCSS algorithm. Besides, all the weights of factors could be adjusted dynamically, making the algorithm suitable for various kinds of applications.

• KIPS Transactions on Computer and Communication Systems
• /
• v.4 no.2
• /
• pp.57-64
• /
• 2015

At the case of peak threshold algorithm, 3-axes data should sample step data over 20 Hz to get sufficient accuracy. But most of the digital sensors like 3-axes accelerometer have very low sampling rate caused by low data communication speed on limited SPI or $I^2C$ bandwidth of the low-cost MPU for ubiquitous devices. If the data transfer rate of the 3-axes accelerometer is getting slow, the sampling rate also slows down and it finally degrades the data accuracy. In this study, we proved there is a distinct functional relation between the sampling rate and threshold on the peak threshold step detection algorithm under the 20Hz frequency, and made a threshold function through the experiments. As a result of experiments, when we apply threshold value from the threshold function instead of fixed threshold value, the step detection error rate can be lessen about 1.2% or under. Therefore, we can suggest a peak threshold based new step detection algorithm with threshold function and it can enhance the accuracy of step detection and step count. This algorithm not only can be applied on a digital step counter design, but also can be adopted any other low-cost ubiquitous sensor devices subjected on low sampling rate.

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.10 no.6
• /
• pp.609-616
• /
• 2017

In this paper, we propose an algorithm to obtain more stable and high recognition rate by using left and right rotation information of input image in order to obtain a stable recognition rate in face recognition. The proposed algorithm uses the facial image as the input information in the web camera environment to reduce the size of the image and normalize the information about the brightness and color to obtain the improved recognition rate. We apply Principal Component Analysis (PCA) to the detected candidate regions to obtain feature vectors and classify faces. Also, In order to reduce the error rate range of the recognition rate, a set of data with the left and right $45^{\circ}$ rotation information is constructed considering the directionality of the input face image, and each feature vector is obtained with PCA. In order to obtain a stable recognition rate with the obtained feature vector, it is after scattered in the eigenspace and the final face is recognized by comparing euclidean distant distances to each feature. The PCA-based feature vector is low-dimensional data, but there is no problem in expressing the face, and the recognition speed can be fast because of the small amount of calculation. The method proposed in this paper can improve the safety and accuracy of recognition and recognition rate faster than other algorithms, and can be used for real-time recognition system.

• The Journal of the Korea Contents Association
• /
• v.19 no.11
• /
• pp.365-374
• /
• 2019

Loss of Control in Flight(LOC-I) due to aircraft upset attitude has the highest air accident rate, and International Aviation Institute such as ICAO and FAA recommended flight crew to operate aircraft safely through UPRT(Upset Prevention & Recovery Training) program. ICAO has selected Loss of Control(LOC) as key safety indicator, and recommended to respond using TEM(Threat and Error Management). However there are not much specific treats and errors classified for UPRT programs using real TEM based on evidences. This study intends to consider the importance of UPRT through the introduction of UPRT and accident analysis using TEM. Typical upset accidents were classified to common threats as IFR, inadequate training, Automation surprise, and inexperienced copilots. The common errors were cross-check, speed and altitude deviation, callouts, communication, thrust and stall action fail. The undesired aircraft states were inadequate automation mode, Deviation of speed and vertical, stall, and crash. These suggest areas to improve UPRT.

• Journal of Broadcast Engineering
• /
• v.27 no.2
• /
• pp.244-247
• /
• 2022

In this paper, the performance of the decoding schemes using linear MMSE filters in the frequency and time domains and the reinforcement Gauss-Seidel algorithm for the orthogonal time frequency space (OTFS) system that can improve robustness under high-speed mobile environments are compared. The reinforcement Gauss-Seidel algorithm can improve the bit error rate performance by suppressing the noise enhancement. The simulation results show that the performance of the decoding scheme using the linear MMSE filter in the frequency domain is severely degraded due to the effect of Doppler shift as the mobile speed increases. In addition, the decoding scheme using the reinforcement Gauss-Seidel algorithm under the channel environment with 120 km/h and 500 km/h speeds outperforms the decoding schemes using linear MMSE filters in the frequency and time domains.

• Journal of the Institute of Electronics Engineers of Korea TC
• /
• v.48 no.1
• /
• pp.16-24
• /
• 2011

The sound speed of seawater can be calculated by the empirical formula as a function of temperature, salinity and pressure. It is little affected by salinity because the average salinity is 34 psu and varies within a few psu seasonally and spatially in the ocean. Recently, low-salinity water of 24 psu flows into the western sea area of Jeju Island due to the flood of the Yangtze River in China during summer, affecting sound speed profile. In this paper, it was analyzed how environmental changes affected to the underwater communication - the sound speed of low-salinity water was calculated, and the communication channel was estimated by the simulated acoustic rays while the transmitting and receiving depth and the range were varied with and without the low-salinity layer. And The BER (Bit error rate) was calculated by BPSK(Binary phase shift key) modulation and the effects of the low-salinity water on the BER was investigated. The sound speed profile was changed to have positive slope by the low-salinity layer at the sub-surface up to 20 m of depth, forming acoustic wave propagation channel at the sub-surface resulting in the decrease of most of the BER Consequently, this paper suggests that it is important to consider changes of the ocean environment for correctly analyzing the underwater communication and the detection capability.