Jungsik Lee

A practical radial basis function equalizer

A radial basis function (RBF) equalizer design process has been developed in which the number of basis function centers used is substantially fewer than conventionally required. The reduction of centers is accomplished in two-steps. First an algorithm is used to select a reduced set of centers that lie close to the decision boundary. Then the centers in this reduced set are grouped, and an average position is chosen to represent each group. Channel order and delay, which are determining factors in setting the initial number of centers, are estimated from regression analysis. In simulation studies, an RBF equalizer with more than 2000-to-1 reduction in centers performed as well as the RBF equalizer without reduction in centers, and better than a conventional linear equalizer.

Theoretical derivation of minimum mean square error of RBF based equalizer

In this paper, the minimum mean square error (MSE) convergence of the RBF equalizer is evaluated and compared with the linear equalizer based on the theoretical minimum MSE. The basic idea of comparing these two equalizers comes from the fact that the relationship between the hidden and output layers in the RBF equalizer is also linear. For the purpose of theoretically evaluating exact minimum MSE for both RBF and linear equalizer, a linear time dispersive channel whose order is one is selected. As extensive studies of this research, various channel models are selected, which include linearly separable channel, slightly distorted channel, and severely distorted channel models. In this work, the theoretical minimum MSE for both RBF and linear equalizers were computed, compared and the sensitivity of minimum MSE due to RBF center spreads was analyzed. It was found that RBF based equalizer always produced lower minimum MSE than linear equalizer, and that the minimum MSE value of RBF equalizer was obtained with the center spread parameter which is relatively higher (approximately 2-10 times more) than variance of AWGN. As a result of that, it leads to the better bit error rate. This work provides an analytical framework for the practical training of RBF equalizer system.

Improving ad hoc network performance using cross-layer information

Various OSI layers and network functions must be considered together while designing efficient networks to support new multimedia services. Use of cross layer approach challenging the traditional OSI layered design has currently been in vogue. Ad hoc network has emerged as an important trend of future wireless system that provide ubiquitous wireless access. In this research, we study the performance optimization challenges of ad hoc network and cross-layer processing to improve its performance. We implemented cross layer processing between physical (PHY), medium access control (MAC) and network (NET) layers using network simulator NS-2. The MAC layer adaptively selects a transmission data rate based on the channel signal strength information from physical layer. The MAC layer utilization is sent to DSR routing protocol as a congestion aware routing metric for optimal route discovery.

The Association Rule Algorithm with Missing Data in Data Mining

This paper discusses the use of an association rule algorithm in data mining and the processes of handling missing data in a distributed database environment. The investigation generated improved association rules using the model described here. The evaluations showed that more association patterns were generated in which the algorithm for missing data was used; this suggested more rules generated than by simply ignoring them. This implies that the model offer more precise and important association rules that is more valuable when applied for business decision making. With the discovery of accurate association rules or business patterns, approach for better market plans can be prepared and implemented to improve marketing schemes. One best-related application of handling missing data is for detecting fraud or devious database entries.

E-mail classification agent using category generation and dynamic category hierarchy

With e-mail use continuing to explode, the e-mail users are demanding a method that can classify e-mails more and more efficiently. The previous works on the e-mail classification problem have been focused on mainly a binary classification that filters out spam-mails. Other approaches used clustering techniques for the purpose of solving multi-category classification problem. But these approaches are only methods of grouping e-mail messages by similarities using distance measure. In this paper, we propose of e-mail classification agent combining category generation method based on the vector model and dynamic category hierarchy reconstruction method. The proposed agent classifies e-mail automatically whenever it is needed, so that a large volume of e-mails can be managed efficiently

Maritime Object Segmentation and Tracking by using Radar and Visual Camera Integration

We have proposed a method to detect and track moving ships using position from Radar and image processor. Real-time segmentation of moving regions in image sequences is a fundamental step in the radar-camera integrated system. Algorithms for segmentation of objects are implemented by composing of background subtraction, morphologic operation, connected components labeling, region growing, and minimum enclosing rectangle. Once the moving objects are detected, tracking is only performed upon pixels labeled as foreground with reduced additional computational burdens.

Fuzzy-ARTMAP based Multi-User Detection

This paper studies the application of a fuzzy-ARTMAP (FAM) neural network to multi-user detector (MUD) for direct sequence (DS)-code division multiple access (CDMA) system. This method shows new solution for solving the problems, such as complexity and long training, which is found when implementing the previously developed neural-basis MUDs. The proposed FAM based MUD is fast and easy to train and includes capabilities not found in other neural network approaches; a small number of parameters, no requirements for the choice of initial weights, automatic increase of hidden units, no risk of getting trapped in local minima, and the capabilities of adding new data without retraining previously trained data. In simulation studies, binary signals were generated at random in a linear channel with Gaussian noise. The performance of FAM based MUD is compared with other neural net based MUDs in terms of the bit error rate.

Agent-Based Approach to Distributed Ensemble Learning of Fuzzy ARTMAP Classifiers

This paper presents a parallel and distributed approach to ensemble learning of Fuzzy ARTMAP classifiers based on the multi-agent platform. Neural networks have been used successfully in a broad range of non-linear problems that are difficult to solve using traditional techniques. Training a neural network for practical applications is often time consuming thus extensive research work is being carried out to accelerate this process. Fuzzy ARTMAP (FAM) is one of the fastest neural network architectures given its ability to produce neurons on demand to represent new classification categories. FAM can adapt to the input data without having to specify an arbitrary structure. However, FAM is vulnerable to noisy data which can rapidly degrade network performance. Due to its fast learning features, FAM is sensitive to the sequence of input sample presentations. In this paper we propose a parallel and distributed approach to ensemble learning for FAM networks as a means to improve the over-all performance of the classifier and increase its resilience to noisy data. We use the multi-agent platform to distribute the computational load of the ensemble to several hosts. The multi-agent platform is a robust environment that can support large-scale neural network ensembles. Our approach also demonstrates the feasibility of large-scale ensembles. The experimental results show that ensemble learning substantially improved the performance of fuzzy ARTMAP classifiers.

Third-order moment spectrum and weighted fuzzy classifier for robust 2-D object recognition

Abstract In this paper, a robust position, scale, and rotation invariant system for the recognition of closed 2-D noise corrupted images using the bispectral features of a contour sequence and the weighted fuzzy classifier are derived. The higher-order spectrum based on third-order moment, called a bispectrum, is applied to the contour sequences of an image to extract a 15-dimensional feature vector for each of the 2-D images. This bispectral feature vector, which is invariant to shape translation, scale, and rotation transformation, can be used to represent a 2-D planar image and is fed into a weighted fuzzy classifier for the recognition process. The experiments with eight different shapes of aircraft images are presented to illustrate the high performance of the proposed system even when the image is significantly corrupted by noise.

Blind Equalization Using RBF and HOS

This paper discusses a blind equalization technique for FIR channel system, that might be minimum phase or not, in digital communication. The proposed techniques consist of two parts. One is to estimate the original channel coefficients based on fourth-order cumulants of the channel output, the other is to employ RBF neural network to model an inverse system for the original channel. Here, the estimated channel is used as a reference system to train the RBF neural network. The proposed RBF equalizer provides fast and easy learning, due to the structural efficiency and excellent recognition-capability of RBF neural network. Throughout the simulation studies, it was found that the proposed blind RBF equalizer performed favorably better than the blind MLP equalizer, while requiring the relatively smaller computation steps in training.