Shigeya TANAKA Takashi HOTTA Fumio MURABAYASHI Hiromichi YAMADA Shoji YOSHIDA Kotaro SHIMAMURA Koyo KATSURA Tadaaki BANDOH Koichi IKEDA Kenji MATSUBARA Kouji SAITOU Tetsuo NAKANO Teruhisa SHIMIZU Ryuichi SATOMURA
A superscalar RISC processor contains 2.8 million transistors in a die size of 16.2 mm16.5 mm, and utilizes 3.3 V/0.5 µm BiCMOS technology. In order to take advantage of superscalar performance without incurring penalties from a slower clock or a longer pipeline, a tag bit is implemented in the instruction cache to indicate dependency between two instructions. A performance gain of up to 37% is obtained with only a 3.5% area overhead from our superscalar design.
Kazuhiro SAITO Hiroshi YOKOYAMA
Control of electronic states of dye molecules (organic semiconductors) by introducing appropriate substituent groups has been examined. NH2 (electron-releasing group) and NO2 (electron-withdrawing group) were introduced in thiacarbocyanine dye to modify the electronic states of the dyes. The effect of modification was examined based on the properties of photoelectric cells made by the dye derivatives. Clear increase in photocurrent, more than ten times, was observed when modified dyes were used instead of the original dye. The result shows that the introduction of substituent groups for organic semiconductors is quite effective to control the electronic states, and the introduction can be regarded as doping in molecular level.
This paper describes the general conditions for perfect signal reconstruction in adaptive blocksize MDCT. MDCT, or modified Discrete Cosine Transform, is a method in which blocks are laid to overlap each other. Because of block overlapping, some consideration must be paid to reconstructing the signals perfectly in adaptive blocksize schemes. The perfect reconstruction conditions are derived by considering the reconstruction signals, on a segment by segment basis. These conditions restrict the analysis/synthesis windows in the MDCT formula. Finally, this paper evaluates two examples of window sets, including windows used in the ISO MPEG audio coding standard.
Takashi YAHAGI Md.Kamrul HASAN
In many applications involving the processing of noisy signals, it is desired to know the noise variance. This paper proposes a new method for estimating the noise variance from the signals of autoregressive (AR) and autoregressive moving-average (ARMA) systems corrupted by additive white noise. The method proposed here uses the low-order Yule-Walker (LOYW) equations and the lattice filter (LF) algorithm for the estimation of noise variance from the noisy output measurements of AR and ARMA systems, respectively. Two techniques are proposed here: iterative technique and recursive one. The accuracy of the methods depends on SNR levels, more specifically on the inherent accuracy of the Yule-Walker and lattice filter methods for signal plus noise system. The estimated noise variance is used for the blind indentification of AR and ARMA systems. Finally, to demonstrate the effectiveness of the method proposed here many numerical results are presented.
Jun'ichi HORI Yoshiaki SAITOH Tohru KIRYU
When measuring the ejection fraction for the evaluation of the ventricular pumping function by means of the thermodilution technique, the slow response a conventional thermistor has caused it to be considered unsuitable, and fast thermistors have been proposed as an alternative. However, in this paper we propose improving the time-domain response of a conventional thermistor using a signal processing technique composed of a series of first-order high-pass filters which is known as the natural observation system. We considered the rise time of the thermistor in response to a step temperature change to effect correction for the measurement of the ejection fraction. The coefficients of the natural observation system were calculated by minimizing the square error between the step-response signal of the thermistor and the band-limited reference signal. In an experiment using a model ventricle, the thermodilution curve obtained from a conventional thermistor was improved using the proposed technique, thus enabling successful measurement of the ejection fraction of the ventricles.
Daisaburo TAKASHIMA Shigeyoshi WATANABE Hiroaki NAKANO Yukihito OOWAKI Kazunori OHUCHI
An open/folded bit-line (BL) arrangement for scaled DRAM's is proposed. This BL arrangement offers small die size and good array noise immunity. In this arrangement, one BL of an open BL pair is placed in between a folded BL pair, and the sense amplifiers (SA's) for open BL's and those for folded BL's are placed alternately between the memory arrays. This arrangement features a small 6F2 memory cell where F is the device feature size, and a relaxed SA pitch of 6F. The die size of a 64-Mb DRAM can be reduced to 81.6% compared with the one using the conventional folded BL arrangement. The BL-BL coupling noise is reduced to one-half of that of the conventional folded BL arrangement, thanks to the shield effect. Two new circuit techniques, 1) a multiplexer for connecting BL's to SA's, and 2) a binary-to-ternary code converter for the multiplexer have been developed to realize the new BL arrangement.
Kazuhiko SEKI Tetsu SAKATA Shuzo KATO
This paper proposes a digitalized quadrature modulator for burst-by-burst carrier frequency hopping in TDMA-TDD systems. It employs digital frequency synthesis and a multiplexing modulation scheme to give the frequency offset to the modulated IF signal. Moreover, to reduce the frequency settling time of the RF synthesizer below the guard time duration, a phase and frequency preset (PFP) PLL synthesizer is employed. By employing the digital modulation scheme, the proposed modulator needs only one D/A converter, as a result, the complexity of adjusting the DC offset and amplitude between analog signals of the in-phase and the quadrature phase is eliminated. The performance of the proposed modulator is analyzed theoretically and simulated by computers. Theoretical analyses show that the frequency settling time with 15MHz hopping width in the 1900MHz band is reduced by more than 75% from that of the conventional synthesizer. The settling time is less than 40µs which is shorter than the typical guard time of the burst signal format. The analyses also show that the power consumption of the proposed modulator is lower than that of the conventional modulator employing a full band digital frequency converter. Furthermore, the computer simulation confirms that the power spectra and the constellations of the proposed modulator for the coherent and the π/4-shift QPSK modulation schemes can be successfully generated.
An improved reflection wave method was described for measurement of complex permittivity of low-loss materials over 100MHz-1GHz range. The residual impedance Zr and stray admittance Ys surrounding the test sample, which terminated the transmission line, were evaluated using sapphire as a reference material. The correction by the obtained Zr and Ys gave accurate values of complex permittivities of alumina and mullite ceramics as 100MHz-1GHz.
Koji NAKAMAE Ryo NAKAGAKI Katsuyoshi MIURA Hiromu FUJIOKA
Precise matching of the SEM (secondary electron microscope) image of the DUT (device under test) interconnection pattern with the CAD layout is required in the CAD-linked electron beam test system. We propose the point pattern matching method that utilizes a corner pattern in the CAD layout. In the method, a corner pattern which consists of a small number of pixels is derived by taking into account the design rules of VLSIs. By using the corner pattern as a template, the matching points of the template are sought in both the SEM image and CAD layout. Then, the point image obtained from the SEM image of DUT is matched with that from the CAD layout. Even if the number of points obtained in the DUT pattern is different from that in the CAD layout due to the influence of noise present in the SEM image of the DUT pattern, the point matching method would be successful. The method is applied to nonpassivated and passivated LSIs. Even for the passivated LSI where the contrast in the SEM image is mainly determined by voltage contrast, matching is successful. The computing time of the proposed method is found to be shortened by a factor of 4 to 10 compared with that in a conventional correlation coefficient method.
Yue WANG Katsushi INOUE Itsuo TAKANAMI
This paper introduces a new class of machines called multihead marker finite automata, and investigates how the number of markers affects its accepting power. Let HM{0}(i, j)(NHM{0}(i, j))denote the class of languages over a one-letter alphabet accepted by two-way deterministic (nondeterminstic) i-head finite automata with j markers. We show that HM{0} (i, j) HM{0}(i, j1) and NHM{0}(i, j) NHM{0}(i, j+1) for each i2, j0.
Youji KANIE Yasushi KUBOTA Shinji TOYOYAMA Yasuaki IWASE Shuhei TSUCHIMOTO
This report describes 4-2 compressors composed of Complementary Pass-Transistor Logic (CPL). We will show that circuit designs of the 4-2 compressors can be optimized for high speed and small size using only exclusive-OR's and multiplexers. According to a circuit simulation with 0.8µm CMOS device parameters, the maximum propagation delay and the average power consumption per unit adder are 1.32 ns and 11.6 pJ, respectively.
Anthony V. W. SMITH Hiroshi SAKO
This document describes a proposal for the implementation of a new VLSI neural network technique called Parallel Propagated Targets (PPT). This technique differs from existing techniques because all layer, within a given network, can learn simultaneously and not sequentially as with the Back Propagation algorithm. the Parallel Propagated Target algorithm uses only information local to each layer and therefore there is no backward flow of information within the network. This allows a simplification in the system design and a reduction in the complexity of implementation, as well as acheiving greater efficiency in terms of computation. Since all synapses can be calculated simultaneously it is possible using the PPT neural algorithm, to parallelly compute all layers of a multi-layered network for the first time.
Yuji IWAHORI Hidekazu TANAKA Robert J. WOODHAM Naohiro ISHII
This paper proposes a new method to determine the shape of a surface by learning the mapping between three image irradiances observed under illumination from three lighting directions and the corresponding surface gradient. The method uses Phong reflectance function to describe specular reflectance. Lambertian reflectance is included as a special case. A neural network is constructed to estimate the values of reflectance parameters and the object surface gradient distribution under the assumption that the values of reflectance parameters are not known in advance. The method reconstructs the surface gradient distribution after determining the values of reflectance parameters of a test object using two step neural network which consists of one to extract two gradient parameters from three image irradiances and its inverse one. The effectiveness of this proposed neural network is confirmed by computer simulations and by experiment with a real object.
Analysis of satellite images requires classificatio of image objects. Since different categories may have almost the same brightness or feature in high dimensional remote sensing data, many object categories overlap with each other. How to segment the object categories accurately is still an open question. It is widely recognized that the assumptions required by many classification methods (maximum likelihood estimation, etc.) are suspect for textural features based on image pixel brightness. We propose an image feature based neural network approach for the segmentation of AVHRR images. The learning algoriothm is a modified backpropagation with gain and weight decay, since feedforward networks using the backpropagation algorithm have been generally successful and enjoy wide popularity. Destructive algorithms that adapt the neural architecture during the training have been developed. The classification accuracy of 100% is reached for a validation data set. Classification result is compared with that of Kohonen's LVQ and basic backpropagation algorithm based pixel-by-pixel method. Visual investigation of the result images shows that our method can not only distinguish the categories with similar signatures very well, but also is robustic to noise.
Iwao SEKITA Takio KURITA David K. Y. CHIU Hideki ASOH
The number of nodes in a hidden layer of a feed-forward layered network reflects an optimality condition of the network in coding a function. It also affects the computation time and the ability of the network to generalize. When an arbitrary number of hidden nodes is used in designing the network, redundancy of hidden nodes often can be seen. In this paper, a method of reducing hidden nodes is proposed on the condition that a reduced network maintains the performances of the original network within an accepted level of tolerance. This method can be applied to estimate the performances of a network with fewer hidden nodes. The estimated performances indicate the lower bounds of the actual performances of the network. Experiments were performed using the Fisher's IRIS data, a set of SONAR data, and the XOR data for classification. The results suggest that sufficient number of hidden nodes, fewer than the original number, can be estimated by the proposed method.
In this paper, we develop a unified synthesizing approach for the cloning templates of Cellular Neural Networks (CNNs). In particular, we shall consider the case when the signal processing problem is complex, and a multilayered CNN with time-variant templates is necessary. The method originates from the existence of correspondence between the cloning templates of Cellular Neural Network and its discrete counterpart, Discrete-Time Cellular Neural Network (DTCNN), in solving a prescribed image processing problem when time-variant templates are involved. Thus, one can start with calculating the cloning templates from DTCNN, and then translating the cloning templates to those for CNN operations. As a result, the mathematical tools being used in the synthesis of Discrete-time Cellular Neural Network can also be applied to the analog type Cellular Neural Network. This inevitably helps to simplify the design problem of CNN for signal processing. Examples akin to contour drawing and parallel thinning are shown to illustrate the merits of our proposed method.
Akihiko YAMANE Noboru OHNISHI Noboru SUGIE
A network system is proposed for segmenting and extracting multiple moving objects in 2D images. The system uses an interconnected neural network in which grouping factors, such as edge proximity, smoothness of edge orientatio, and smoothness of velocity perpendicular to an edge, are embedded. The system groups edges so that the network energy may be minimized, i.e. edges may be organized into perceptually plausible configuration. Experimantal results are provided to indicate the performance and noise robustness of the system in extracting objects in synthetic images.
Sadayuki HONGO Isamu YOROIZAWA
We propose a fast computation method of stochastic relaxation for the continuous-valued Markov random field (MRF) whose energy function is represented in the quadratic form. In the case of regularization in visual information processing, the probability density function of a state transition can be transformed to a Gaussian function, therefore, the probablistic state transition is realized with Gaussian random numbers whose mean value and variance are calculated based on the condition of the input data and the neighborhood. Early visual information processing can be represented with a coupled MRF model which consists of continuity and discontinuity processes. Each of the continuity or discontinuity processes represents a visual property, which is like an intensity pattern, or a discontinuity of the continuity process. Since most of the energy function for early visual information processing can be represented by the quadratic form in the continuity process, the probability density of local computation variables in the continuity process is equivalent to the Gaussian function. If we use this characteristic, it is not necessary for the discrimination function computation to calculate the summation of the probabilities corresponding to all possible states, therefore, the computation load for the state transition is drastically decreased. Furthermore, if the continuous-valued discontinuity process is introduced, the MRF model can directly represent the strength of discontinuity. Moreover, the discrimination function of this energy function in the discontinuity process, which is linear, can also be calculated without probability summation. In this paper, a fast method for calculating the state transition probability for the continuous-valued MRF on the visual informtion processing is theoretically explained. Next, initial condition dependency, computation time and dependency on the statistical estimation of the condition are investigated in comparison with conventional methods using the examples of the data restoration for a corrupted square wave and a corrupted one-dimensional slice of a natural image.
A new regularization cost function for generalization in real-valued function learning is proposed. This cost function is derived from the maximum likelihood method using a modified sample distribution, and consists of a sum of square errors and a stabilizer which is a function of integrated square derivatives. Each of the regularization parameters which gives the minimum estimation error can be obtained uniquely and non-empirically. The parameters are not constants and change in value during learning. Numerical simulation shows that this cost function predicts the true error accurately and is effective in neural network learning.
Kitaek KWON Hisao ISHIBUCHI Hideo TANAKA
This paper proposes an approach for approximately realizing nonlinear mappings of interval vectors by interval neural networks. Interval neural networks in this paper are characterized by interval weights and interval biases. This means that the weights and biases are given by intervals instead of real numbers. First, an architecture of interval neural networks is proposed for dealing with interval input vectors. Interval neural networks with the proposed architecture map interval input vectors to interval output vectors by interval arithmetic. Some characteristic features of the nonlinear mappings realized by the interval neural networks are described. Next, a learning algorithm is derived. In the derived learning algorithm, training data are the pairs of interval input vectors and interval target vectors. Last, using a numerical example, the proposed approach is illustrated and compared with other approaches based on the standard back-propagation neural networks with real number weights.