• The Journal of the Acoustical Society of Korea
• /
• v.42 no.2
• /
• pp.94-101
• /
• 2023

A deep neural network composed of encoders and decoders, such as U-Net, used for speech enhancement, concatenates the encoder to the decoder through skip-connection. Skip-connection helps reconstruct the enhanced spectrum and complement the lost information. The features of the encoder and the decoder connected by the skip-connection are incompatible with each other. In this paper, for complex-valued spectrum based speech enhancement, Self-Attention (SA) method is applied to skip-connection to transform the feature of encoder to be compatible with the features of decoder. SA is a technique in which when generating an output sequence in a sequence-to-sequence tasks the weighted average of input is used to put attention on subsets of input, showing that noise can be effectively eliminated by being applied in speech enhancement. The three models using encoder and decoder features to apply SA to skip-connection are studied. As experimental results using TIMIT database, the proposed methods show improvements in all evaluation metrics compared to the Deep Complex U-Net (DCUNET) with skip-connection only.

• ETRI Journal
• /
• v.35 no.1
• /
• pp.41-50
• /
• 2013

Network Mobility (NEMO) handles mobility of multiple nodes in an aggregate manner as a mobile network. The standard NEMO suffers from a number of limitations, such as inefficient routing and increased handoff latency. Most previous studies attempting to solve such problems have imposed an extra signaling load and/or modified the functionalities of the main entities. In this paper, we propose a more secure and lightweight route optimization (RO) mechanism based on exploiting the firewall in performing the RO services on behalf of the correspondent nodes (CNs). The proposed mechanism provides secure communications by making an authorized decision about the mobile router (MR) home of address, MR care of address, and the complete mobile network prefixes underneath the MR. In addition, it reduces the total signaling required for NEMO handoffs, especially when the number of mobile network nodes and/or CNs is increased. Moreover, our proposed mechanism can be easily deployed without modifying the mobility protocol stack of CNs. A thorough analytical model and network simulator (Ns-2) are used for evaluating the performance of the proposed mechanism compared with NEMO basic support protocol and state-of-the-art RO schemes. Numerical and simulation results demonstrate that our proposed mechanism outperforms other RO schemes in terms of handoff latency and total signaling load on wired and wireless links.

• International Journal of Control, Automation, and Systems
• /
• v.1 no.1
• /
• pp.134-141
• /
• 2003

A major research issue associated with service robots is the creation of an environment recognition system for mobile robot navigation that is robust and efficient on various environment situations. In recent years, intelligent autonomous mobile robots have received much attention as the types of service robots for serving people and industrial robots for replacing human. To help people, robots must be able to sense and recognize three dimensional space where they live or work. In this paper, we propose a three dimensional environmental modeling method based on an edge enhancement technique using a planar fitting method and a neural network technique called "Growing Neural Gas Network." Input data pre-processing provides probabilistic density to the input data of the neural network, and the neural network generates a graphical structure that reflects the topology of the input space. Using these methods, robot's surroundings are autonomously clustered into isolated objects and modeled as polygon patches with the user-selected resolution. Through a series of simulations and experiments, the proposed method is tested to recognize the environments surrounding the robot. From the experimental results, the usefulness and robustness of the proposed method are investigated and discussed in detail.in detail.

• Journal of the Korea Society of Computer and Information
• /
• v.4 no.4
• /
• pp.86-93
• /
• 1999

This paper proposes the multi-modular neural network and compensative input algorithm. The former is to reduce convergence speed which is one of the neural network's inveterate problems, and the latter is to improve the recognition performance of the neural network. This paper consists of two major parts and a simulation. First, it shows the structure of mu1ti-modular neural network, which is applied to the recognition of Korean, English characters and numbers. Second, it describes the compensative input algorithm and shows the steps that determine the compensative input. The proposed algorithm was tested and compared with the existing neural networks in the recognition of Korean and English characters and numbers. The convergence speed is three times or more faster than the existing neural network. In the case that compensative input was applied to neural network, the recognition rate was improved more than 10%.

• The KIPS Transactions:PartA
• /
• v.13A no.7 s.104
• /
• pp.573-580
• /
• 2006

The conventional transport layer protocol TCP is designed to work under condition of packet loss is due to the network congestion, so that it's suitable in the traditional wired network with fixed hosts but it's inefficient on the wireless network where the environment of fading, noise, and transmission error comes from interference. This result from the needless transmission control of the bit error is due to treats the packet loss as a packet congestion control in the wireless network. In this paper, we propose the advanced SNOOP protocol with the consecutive packet loss and TCP window control to avoid the needless congestion management algorithm in wireless network for the wireless TCP packet transmission enhancement. We verify the performance of the advanced module from the simulation experiment result.

• Industrial Engineering and Management Systems
• /
• v.15 no.3
• /
• pp.268-277
• /
• 2016

Load balancing is a significant technique to prolong a network's lifetime in sensor network. This paper introduces a hybrid approach named as Load Distributing Hybrid Routing Protocol (LDHRP) composed with a border node routing protocol (BDRP) and greedy forwarding (GF) strategy which will make the routing effective, especially in mobility scenarios. In an existing solution, because of the high network complexity, the data delivery latency increases. To overcome this limitation, a new approach is proposed in which the source node transmits the data to its respective destination via border nodes or greedily until the complete data is transmitted. In this way, the whole load of a network is evenly distributed among the participating nodes. However, border node is mainly responsible in aggregating data from the source and further forwards it to mobile sink; so there will be fewer chances of energy expenditure in the network. In addition to this, number of hop counts while transmitting the data will be reduced as compared to the existing solutions HRLBP and ZRP. From the simulation results, we conclude that proposed approach outperforms well than existing solutions in terms including end-to-end delay, packet loss rate and so on and thus guarantees enhancement in lifetime.

• Journal of Broadcast Engineering
• /
• v.13 no.1
• /
• pp.25-33
• /
• 2008

Video telephony is a representative service in 3G mobile network. And there have been efforts to improve quality of video telephony service in different fields. In Korea, the leading mobile service provider SKTelecom and KTF service the WCDMA network as 3G mobile network. Now, more than a million people is using the network. In this paper, we study about video telephony over WCDMA network. and propose error minimizing algorithm using cross-layer adaptation between physical layer and video codec. We simulated 3G-324M protocol with MPEG-4 video codec, and simulation results show suggested algorithm improve packet transmission rate for improving quality of video telephony service.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.11 no.9
• /
• pp.1679-1686
• /
• 2007

Clustering is a topology control approach often used in wireless ad hoc networks to improve scalability and prolong network lifetime. Furthermore, it is also employed to provide semi-management functionalities and capacity enhancement. The usage of clustering topology control technique can also be applied to multi-radio wireless mesh network. This would utilize the advantages of the multi-radio implementation in the network. The aggregation would result to a more stable, connected, scalable and energy-efficient network. On this paper, we design a clustering algorithm for multi-radio wireless mesh network that would use these advantages and would take into consideration both mobility and heterogeneity of the network entities. We also show that the algorithm terminates at a definite time t and the message control overhead complexity is of constant order of O(1) per node.

• Smart Structures and Systems
• /
• v.32 no.3
• /
• pp.179-193
• /
• 2023

This study explores an alternative to the existing centralized process for data anomaly detection in modern Internet of Things (IoT)-based structural health monitoring (SHM) systems. An edge intelligence framework is proposed for the early detection and classification of various data anomalies facilitating quality enhancement of acquired data before transmitting to a central system. State-of-the-art deep neural network pruning techniques are investigated and compared aiming to significantly reduce the network size so that it can run efficiently on resource-constrained edge devices such as wireless smart sensors. Further, depthwise separable convolution (DSC) is invoked, the integration of which with advanced structural pruning methods exhibited superior compression capability. Last but not least, quantization-aware training (QAT) is adopted for faster processing and lower memory and power consumption. The proposed edge intelligence framework will eventually lead to reduced network overload and latency. This will enable intelligent self-adaptation strategies to be employed to timely deal with a faulty sensor, minimizing the wasteful use of power, memory, and other resources in wireless smart sensors, increasing efficiency, and reducing maintenance costs for modern smart SHM systems. This study presents a theoretical foundation for the proposed framework, the validation of which through actual field trials is a scope for future work.

• JSTS:Journal of Semiconductor Technology and Science
• /
• v.3 no.1
• /
• pp.47-54
• /
• 2003

New efficiency enhancement techniques have been devised and implemented to InGaP/GaAs HBT MMIC power amplifiers for W-CDMA mobile terminals applications. Two different types of bias current control circuits that select the efficient quiescent currents in accordance with the required output power levels are proposed for overall power efficiency improvement. A dual chain power amplifier with single matching network composed of two different parallel-connected power amplifier is also introduced. With these efficiency enhancement techniques, the implemented MMIC power amplifiers presents power added efficiency (PAE) more than 14.8 % and adjacent channel leakage ratio(ACLR) lower than -39 dBc at 20 dBm output power and PAE more than 39.4% and ACLR lower than -33 dBc at 28 dBm output power. The average power usage efficiency of the power amplifier is improved by a factor of more than 1.415 with the bias current control circuits and even up to a factor of 3 with the dual chain power amplifier.