Yuichi Tanji

Cellular neural networks with output function having multiple constant regions

This paper presents a novel class of cellular neural networks (CNNs), where output of a cell in the CNN is given by the piecewise-linear (PWL) function having multiple constant regions or a quantization function. CNNs with one of these output functions allow us to extend CNNs to image processing with multiple gray levels. Since each cell of the original CNN has the PWL output function with two saturation regions, the image-processing tasks are mainly developed for black and white output images. Hence, the proposed architecture will extend the promising nature of the CNN further. Moreover, the hysteresis characteristics are introduced for these functions, which make tolerance to a noise robust. It is demonstrated mathematically that under a mild assumption, the stability of the CNN, which has an output function with hysteresis characteristics, is guaranteed, and the impressive simulation results are also presented.

Lossless image compression and reconstruction by cellular neural networks

It is clear that the characteristic of cellular neural networks (CNN) is the use of A-templates by which many kind of dynamical interpolative nonlinear effects can be generated without dependency of image scanning. This paper describes nonlinear quantization methods in a discrete-time cellular neural network (DT-CNN) which generates a high quality lossy or lossless reconstructed image. It is very important that the DT-CNN state variable image which is determined dynamically based on the minimization of the DT-CNN Lyapunov energy function to generate an optimized interpolative prediction function is a lossless interpolative DPCM image between the original input and the interpolation prediction functions. The small compression ratio for the reconstructed lossless image can be changed by the multivalue quantization and the A-template. By the DT-CNN nondependency of image scanning, the lossless image points can be extracted even in an lossy image by checking the existence of local errors.

CNN with multi-level hysteresis quantization output

This paper presents a novel class of cellular neural networks, where the output is given by the multilevel hysteresis quantization function. Since each cell of elementary CNN has bi-stable piecewise linear function, the image processing is restricted to the black-and-white case. Hence, the architecture provided in this paper would extend availability of CNN. Especially, it is extremely useful for image intensity conversion. In this paper, the Lyapunov stability of CNN with multilevel hysteresis quantization output is proven and the computer simulation shows good convergence property of the CNN.

A new order-reduction method of interconnect networks characterized by sampled data via orthogonal least square algorithm

Accurate and efficient modeling of high-speed interconnects characterized by sampled data is presented. This method is based on Cauchys rational approximation method of frequency response. Here, the dominant pole and residue pairs are determined by means of Chens adaptive orthogonalization method and its modification. Moreover, this method is applicable to analysis of interconnect networks characterized with network equations by means of a transfer function searching algorithm.

A new curve fitting technique for analysis of frequency-dependent lossy transmission lines

Analysis of frequency-dependent lossy transmission lines is very important for designing high-speed VLSI, MCMs and PCBs. The frequency-dependent parameters are always obtained as tabulated data. In this paper, a new curve fitting technique of the tabulated data for the moment matching technique in interconnect analysis is presented. This method is based on Chebyshev interpolation and enhances the efficiency of the moment matching technique.

Image intensity conversion via cellular neural networks

The image intensity conversion via CNN is presented. The intensity conversion is defined as a nonlinear optimization problem, and the templates of CNN for solving it are optimally designed. Since human visual sensitivity and linear quantization of original image are used to design the templates, it gives a smooth image preserving edge information such as character parts.

ANALYSIS OF NON-LINEAR TRANSMISSION LINES BY FREQUENCY DOMAIN PERTURBATION METHOD

SUMMARY We discuss a numerical method for solving non-linear transmission lines in the frequency domain. Such transmission lines are widely used for communications such as in GaAs integrated circuits and varactor diode circuits. The circuit equations are described by non-linear partial dierential equations, so their analysis is very complicated compared with that of linear transmission lines. In this paper we propose a frequency-domain perturbation method for weakly non-linear transmission lines where the wave-forms are approximated by Fourier expansions and each frequency component is calculated by a modied perturbation method. To improve convergence, we introduce two new techniques, the compensation method and the homotopy method, which help to make the iteration stable and can be applied to a wide class of non-linear transmission lines. We have analysed shock wave phenomena in example. ? 1997 by John Wiley & Sons, Ltd. Int. J. Circ. Theor. Appl., vol. 25, 95{105 (1997)

A method for analysis of transmission line networks using frequency response approximation

A method and apparatus for electrically biasing a substrate in an electrochemical processing system is generally provided. In one embodiment, an apparatus for electrochemical processing includes a polishing pad and a conductive element disposed therein. The polishing pad has an upper surface adapted to support a substrate thereon during processing. The conductive element disposed in the polishing pad is movable between a first position having at least a portion of the conductive element exposed above the upper surface and a second position below the upper surface, wherein the conductive element is magnetically biased towards the first position.

Cooperative and competitive cellular neural networks

The cooperative and competitive networks proposed by Amari and Arbib are a mathematical model for the winner-take-all process. If this network is implemented by realistic circuit elements, there are many problems. In this paper, we propose cooperative and competitive cellular neural networks by extending the Amari-Arbib model to achieve low power consumption, no oscillation and fast convergence by using piecewise-linear functions. Moreover, the cooperative and competitive cellular neural networks are applied to block matching using basic cellular neural networks.

Analysis of two dimensional circuits based on multi-conductor theorem

We discuss an efficient algorithm for solving 2-dimensional circuits based on a multi-conductor theorem. There have been many papers published about transmission lines, because they are very important for design of high speed VLSI chips. On the other hand, a device simulation is also very important to design ICs and to understand the qualitative behaviour. Most of the simulation techniques are based on the finite-difference time-domain method, where the devices are described by many discrete models. This is really time-consuming because, to get exact solution, the device must be divided into many sections. In this paper, we show an elegant algorithm for solving 2-dimensional circuits, which can be applied to the device simulations.