Gyogwon Koo

Steady-state mean-square deviation analysis of improved normalized subband adaptive filter

A new minimization criterion for the normalized subband adaptive filter (NSAF), which is called improved NSAF (INSAF), was introduced recently to improve the performance of the steady-state mean-square deviation (MSD). However, the steady-state MSD analysis of the INSAF was not studied. Therefore, this paper proposes a general solution of steady-sate MSD analysis of the INSAF algorithm, which is based on the substitution of the past weight error vector in the weight error vector. The simulation shows that our theoretical results correspond closely with the computer simulation results in various environments.

Mean-Square Deviation Analysis of Multiband-Structured Subband Adaptive Filter Algorithm

A multiband-structured subband adaptive filter (MSAF) algorithm was introduced to achieve a fast convergence rate for the correlated input signal. The convergence analysis of the adaptive filter algorithm is an important concept because it provides a guideline to design the adaptive filter. However, the convergence analysis of the MSAF algorithm has not been researched as extensively as that of the normalized least-mean-square algorithm. Therefore, it needs to be researched. In this paper, we present a new approach to the mean-square deviation (MSD) analysis of the MSAF algorithm by using the persistently exciting input and the practical assumption that the stopband attenuation of the prototype filter is high. Unlike the previous analysis, the proposed analysis is possible to be applied to the long-length adaptive filter such as the acoustic echo cancellation. The proposed analysis is also applied to a non-stationary model with a random walk of the optimal weight vector. The simulation results match with the theoretical results in both the transient-state and steady-state MSD.

Fast communication: Variable regularization for normalized subband adaptive filter

To overcome the performance degradation of least mean square (LMS)-type algorithms when input signals are correlated, the normalized subband adaptive filter (NSAF) was developed. In the NSAF, the regularization parameter influences the stability and performance. In addition, there is a trade-off between convergence rate and steady-state mean square deviation (MSD) according to the change of the parameter. Therefore, to achieve both fast convergence rate and low steady-state MSD, the parameter should be varied. In this paper, a variable regularization scheme for the NSAF is derived on the basis of the orthogonality between the weight-error vector and weight vector update, and by using the calculated MSD. The performance of the variable regularization algorithm is evaluated in terms of MSD. Our simulation results exhibit fast convergence and low steady-state MSD when using the proposed algorithm.

Robust convex combination of affine projection-type algorithms using an impulsive noise indicator

A novel adaptive filter combining the affine projection algorithm (APA) and the affine projection sign algorithm (APSA) is proposed using an impulsive noise indicator. This indicator is proposed to use the APA as the component filter in impulsive noise environments, and it is easily obtained with convex combination schemes. From this, the proposed algorithm achieves robustness against impulsive noise regardless of the convergence state. In addition, the proposed algorithm exhibits a fast convergence rate of the APA for various noise environments. Simulation results verify that the proposed algorithm adequately combines the advantages of the two component filters for system identification scenarios. HighlightsThis paper is the first approach combining the APA and the APSA.A novel impulsive noise indicator is introduced to adopt APA with impulsive noise.The indicator is derived using the difference between two filter output errors.Proposed algorithm converges as fast as the APA even with impulsive noise.Proposed algorithm shows robustness in various noise environments.

Regularization parameter of normalized subband adaptive filter

The stability and performance of the normalized subband adaptive filter (NSAF) algorithm is influenced by the regularization parameter. However, in various noise environments, the regularization parameter is difficult to be determined. The basic idea of this paper is to eliminate the effects of the noise in filter estimation. Simulation results show the proposed method has valid results in various noise environment.

Sign Subband Adaptive Filter with Selection of Number of Subbands

The sign subband adaptive filter (SSAF) algorithm is introduced to reduce performance degradation of least-mean-square-type algorithms due to a correlated input signal or an impulsive noise environments. However, this algorithmh has huge computational complexity when the length of the unknown system is large. In this paper, we focus on reduce computational complexity of the conventional SSAF algorithm and propose an SSAF algorithm which selects number of subbands according to convergence state. The specific bands which contributes to decrease the mean-square deviation are used to update the adaptive filter. Thus, the proposed algorithm reduces the computational complexity compared to the conventional SSAF algorithm. The selection mehtod is derived by analysing the mean-square deviation. Through the computer simulation, simulation results are presented that demonstrate the fast convergence rate of the proposed algorithm and save the computational cost.

Variable step-size affine projection algorithm for a non-stationary system

Affine projection algorithm (APA) has advantages when input signals are highly correlated with each other. To improve convergence rate and steady-state mean square deviation (MSD) of the APA, the step size variation concept based on theoretical MSD has been researched. However, structurally, the APA based on theoretical MSD cannot track the system change without the reset algorithm. The problem is, the reset algorithm would not operate when the system change occurs at early iteration. To overcome this drawback, we propose the variable step size APA attaching the noise-error relation. We apply it to a recent variable step-size APA, which is the optimal step-size APA (OS-APA), then the simulation results show that the proposed APA tracks the system change well without the reset algorithm, and has similar performance compared to the OS-APA for all iterations.

An adjusting-block based convex combination algorithm for identifying block-sparse system

This paper is the first approach combining the BP-NLMS and the BZA-NLMS.We propose a block activeness indicator for block wise convex combination.We propose a block adjustment algorithm to overcome disadvantage of previous works.Proposed algorithm converges as fast as the BP-NLMS.Proposed algorithm shows the low steady-state misalignment as the BZA-NLMS. A novel block wise convex combination algorithm with adjusting blocks is proposed for block-sparse system identification. The proposed algorithm unifies the complementary advantages of different block-induced algorithms, which are based on block proportionate matrix and block zero attracting penalty. A mixing parameter for block wise combination is designed as a block diagonal matrix. The mixing parameter is obtained using the conventional mixing parameter, which represents convergence state, and a block activeness indicator. The indicator for each block is derived from the l0-norm measure of the block. Moreover, a block adjustment algorithm is developed using the indicator to overcome the main disadvantage of block-induced algorithms, i.e., the dependency on cluster location. The simulations for system identification are performed on several block-sparse systems including systems with single cluster and double clusters. The simulation results show that the proposed algorithm not only combines the different block-induced algorithms effectively but also improves the performance via the block adjustment algorithm.

Transfer learning of a deep convolutional neural network for localizing handwritten slab identification numbers

Most machine learning methods assume that previous and future data have same distribution in same feature space. This paper presents a real-world problem that violates the common assumption, and we propose a practical methodology to handle the problem. In the steel making industry, automated marking systems are widely used to inscribe slab identification numbers (SINs). In the previous work, a deep learning based algorithm was developed to automatically extract regions of printed SINs. However, as the marking system is outdated, few SINs are marked by hand in uncommon situations, and the existing algorithm does not work for the handwritten SINs. This paper proposes a practical method that uses very small training data (10 images) to localize handwritten SINs. The knowledge of mid-level layers or entire layers in the pre-trained deep convolutional neural network is transferred to overcome the shortage of training data in the target domain. Experiments were conducted with actual industrial data to demonstrate the effectiveness of the proposed algorithm.

Blind Decision Feedback Equalizer for Holographic Versatile Disc.

As the amount of data increases, holographic data storage (HDS) is considered as a next generation storage medium. Since HDS uses two-dimensional (2D) data, it causes intersymbol interference (ISI) between adjacent pixels not only in the horizontal direction but also in the vertical direction. Thus, studies have been carried out to reduce such 2D ISI, and especially many researches using the partial response maximum likelihood (PRML) method have been carried out. These PRML methods have good bit-error-rate (BER) performance, but also have various disadvantages. Therefore, we propose a simple blind decision feedback equalizer (blind DFE) that does not use soft output Viterbi algorithm (SOVA) for application to European standard holographic versatile disc (HVD). First, we propose a blind equalizer using simle theshold method to get information that the equalizer can refer to. In order to make it work well in any environment, the threshold value is adaptively determined using the statistical characteristics of the received image. And, in order to reduce errors due to the data that cannot be distinguished only by the blind equalizer, we add a decision feedback loop after the blind equalizer. Finally, various simulations were conducted to confirm the performance of blind DFE for HVD.