Itsuo TAKANAMI Satoru NAKAMURA Tadayoshi HORITA
Using Hopfield-type neural network model, we present an algorithm for reconstructing 3D mesh processor arrays using single-track switches where spare processors are laid on the six surfaces of a 3D array and show its effectiveness in terms of reconstruction rate and computing time by computer simulation. Next, we show how the algorithm can be realized by a digital neural circuit. It consists of subcircuits for finding candidate compensation paths, deciding whether the neural system reaches a stable state and at the time the system energy is minimum, and subcircuits for neurons. The subcircuit for each neuron including the other subcircuits can only be made with 16 gates and two flip-flops. Since the state transitions are done in parallel, the circuit will be able to find a set of compensation paths for a fault pattern very quickly within a time less than 1 µs. Furthermore, the hardware implementation of the algorithm leads to making a self-reconfigurable system without the aid of a host computer.
Tatsuo TSUJITA Yuichiro AIHARA Minoru FUJISHIMA Koichiro HOH
This paper analyzes the operations of a CMOS multivibrator-based chaos generator. The equations representing the shape of the first return map are formulated and confirmed by comparison with experimental results, and the design principles are obtained.
Masanori HARIYAMA Kazuhiro SASAKI Michitaka KAMEYAMA
High-speed collision detection is important to realize a highly-safe intelligent vehicle. In collision detection, high-computational power is required to perform matching operation between discrete points on surfaces of a vehicle and obstacles in real-world environment. To achieve the highest performance, a hierarchical matching scheme is proposed based on two representations: the coarse representation and the fine representation. A vehicle is represented as a set of rectangular solids in the fine representation (fine rectangular solids), and the coarse representation, which is also a set of rectangular solids, is produced by enlarging the fine representation. If collision occurs between an obstacle discrete point and a rectangular solid in the coarse representation (coarse rectangular solid), then it is sufficient to check the only fine rectangular solids contained in the coarse one. Consequently, checks for the other fine rectangular solids can be omitted. To perform the hierarchical matching operation in parallel, a hierarchically-content-addressable memory (HCAM) is proposed. Since there is no need to perform matching operation in parallel with fine rectangular solids contained in different coarse ones, the fine ones are mapped onto a matching unit. As a result, the number of matching units can be reduced without decreasing the performance. Under the condition of the same execution time, the area of the HCAM is reduced to 46.4% in comparison with that of the conventional CAM in which the hierarchical matching scheme is not used.
This paper shows the design of multi-stage fuzzy inference system with smaller number of rules based upon the optimization of rules by using the genetic algorithm. Since the number of rules of fuzzy inference system increases exponentially in proportion to the number of input variables powered by the number of membership function, it is preferred to divide the inference system into several stages (multi-stage fuzzy inference system) and decrease the number of rules compared to the single stage system. In each stage of inference only a portion of input variables are used as the input, and the output of the stage is treated as an input to the next stage. If we use the simplified inference scheme and assume the shape of membership function is given, the same backpropagation algorithm is available to optimize the weight of each rule as is usually used in the single stage inference system. On the other hand, the shape of the membership function is optimized by using the GA (genetic algorithm) where the characteristics of the membership function is represented as a set of string to which the crossover and mutation operation is applied. By combining the backpropagation algorithm and the GA, we have a comprehensive optimization scheme of learning for the multi-stage fuzzy inference system. The inference system is applied to the automatic bond rating based upon the financial ratios obtained from the financial statement by using the prescribed evaluation of rating published by the rating institution. As a result, we have similar performance of the multi-stage fuzzy inference system as the single stage system with remarkably smaller number of rules.
This paper gives a detailed presentation of a "vision chip" for a very fast detection of motion vectors. The chip's design consists of a parallel pixel array and column parallel block-matching processors. Each pixel of the pixel array contains a photo detector, an edge detector and 4 bits of memory. In the detection of motion vectors, first, the gray level image is binarized by the edge detector and subsequently the binary edge data is used in the block matching processor. The block-matching takes place locally in pixel and globally in column. The chip can create a dense field of motion where a vector is assigned to each pixel by overlapping 2 2 target blocks. A prototype with 16 16 pixels and four block-matching processors has been designed and implemented. Preliminary results obtained by the prototype are shown.
Dwi HANDOKO Shoji KAWAHITO Yoshiaki TADOKORO Akira MATSUZAWA
This paper presents a novel method of an on-sensor motion vector estimation. One of the key techniques is an iterative block matching algorithm using high-speed interpolated pictures. This technique allows us to estimate the video-rate (30 frame/s) motion vectors accurately from the motion vectors obtained at high-speed frames. The proposed iterative block matching reduces the computational complexity by a factor of more than one tenth compared to the conventional full search block matching algorithm. This property is particularly useful for the reduction of the power dissipation of video encoder. Another proposed technique is a high-speed non-destructive image sensing. This technique is essential to obtain high-speed interpolated pictures while maintaining high image quality in video-rate image sensing. The estimated power dissipation of the designed CMOS image sensor is sufficiently low, allowing us to achieve a totally low-power design of one-chip CMOS cameras integrating an image sensor and a video encoder.
Basabi CHAKRABORTY Yasuji SAWADA
Feature selection is an integral part of any pattern recognition system. Removal of redundant features improves the efficiency of a classifier as well as cut down the cost of future feature extraction. Recently neural network classifiers have become extremely popular compared to their counterparts from statistical theory. Some works on the use of artificial neural network as a feature selector have already been reported. In this work a simple feature selection algorithm has been proposed in which a fractal neural network, a modified version of multilayer perceptron, has been used as a feature selector. Experiments have been done with IRIS and SONAR data set by simulation. Results suggest that the algorithm with the fractal network architecture works well for removal of redundant informations as tested by classification rate. The fractal neural network takes lesser training time than the conventional multilayer perceptron for its lower connectivity while its performance is comparable to the multilayer perceptron. The ease of hardware implementation is also an attractive point in designing feature selector with fractal neural network.
Atsushi KAMO Takayuki WATANABE Hideki ASAI
This paper describes the expanded generalized method of characteristics (GMC) in order to handle large linear interconnect networks. The conventional GMC is applied to modeling each of transmission lines. Therefore, this method is not suitable to deal with large linear networks containing many transmission lines. Here, we propose the expanded GMC method to overcome this problem. This method computes a characteristic impedance and a new propagation function of the large linear networks containing many transmission lines. Furthermore the wave propagation delay is removed from the new wave propagation function using delay evaluation technique. Finally, it is shown that the present method enables the efficient and accurate simulation of the transmission line networks.
Takashi HIKIHARA Yohsuke KONDO Yoshisuke UEDA
In this paper, the stress wave propagation in a coupled pendulum system with friction force is discussed experimentally and numerically. The coupled system is analogous to the one dimensional fault dynamics model in seismicity. However, we will not intend to discuss about the geophysical feature of the system. The system has rich characteristics of the spatio-temporal stress wave propagation effected by nonlinear friction force. The relation between the wave propagation and the vibration of the pendulum is mainly discussed on the standpoint of nonlinear coupled system.
Masahiko HIRATSUKA Takafumi AOKI Tatsuo HIGUCHI
This paper explores a possibility of constructing massively parallel molecular computing systems using molecular electronic devices called enzyme transistors. The enzyme transistor is, in a sense, an artificial catalyst which selects a specific substrate molecule and transforms it into a specific product. Using this primitive function, various active continuous media for signal transfer/processing can be realized. Prominent examples discussed in this paper are: (i) Turing pattern formation and (ii) excitable wave propagation in a two-dimensional enzyme transistor array. This paper demonstrates the potential of enzyme transistors for creating reaction-diffusion dynamics that performs useful computations in a massively parallel fashion.
Ruck THAWONMAS Andrzej CICHOCKI
In this paper, we discuss a neural network approach for blind signal extraction of temporally correlated sources. Assuming autoregressive models of source signals, we propose a very simple neural network model and an efficient on-line adaptive algorithm that extract, from linear mixtures, a temporally correlated source with an arbitrary distribution, including a colored Gaussian source and a source with extremely low value (or even zero) of kurtosis. We then combine these extraction processing units with deflation processing units to extract such sources sequentially in a cascade fashion. Theory and simulations show that the proposed neural network successfully extracts all arbitrarily distributed, but temporally correlated source signals from linear mixtures.
Takaomi SHIGEHARA Hiroshi MIZOGUCHI Taketoshi MISHIMA Taksu CHEON
We propose a new method to construct a four parameter family of quantum-mechanical point interactions in one dimension, which is known as all possible self-adjoint extensions of the symmetric operator T=-Δ C0(R \{0}). It is achieved in the small distance limit of equally spaced three neighboring Dirac's δ potentials. The strength for each δ is appropriately renormalized according to the distance and it diverges, in general, in the small distance limit. The validity of our method is ensured by numerical calculations. In general cases except for usual δ, the wave function discontinuity appears around the interaction and one can observe such a tendency even at a finite distance level.
In multi-media systems, the type of interactive communication channels is found almost everywhere and plays an important role, as well as the type of unilateral communication channels. In this report, we shall construct a fluctuation theory based on the concept of set-valued mappings, suitable for evaluation, control and operation of interactive communication channels in multi-media systems, complicated and diversified on large scales. Fundamental conditions for availability of such channels are clarified in a form of fixed point theorem for system of set-valued mappings.
Yutaka HATA Naotake KAMIURA Kazuharu YAMATO
This paper describes the benefit of utilizing the unary function generators in a multiple-valued Programmable Logic Array (PLA). We will clarify the most suitable PLA structure in terms of the array size. The multiple-valued PLA considered here has a structure with two types of function generators (literal and unary function generators), a first-level array and a second-level array. On investigating the effectiveness to reduce the array size, we can pick up four form PLAs: MAX-of-TPRODUCT form, MIN-of-TSUM form, TSUM-of-TPRODUCT form and TPRODUCT-of-TSUM form PLAs among possible eight form PLAs constructing from the MAX, MIN, TSUM and TPRODUCT operators. The upper bound of the array sizes with v UGs is derived as (log2ppv + p(n-v) + 1) pn-1 to realize any n-variable p-valued function. Next, experiments to derive the smallest array sizes are done for 10000 randomly generated functions and 21 arithmetic functions. These results conclude that MAX-of-TPRODUCT form PLA is the most useful in reducing the array size among the four form PLAs.
Ashraf A. M. KHALAF Kenji NAKAYAMA
A nonlinear time series predictor was proposed, in which a nonlinear sub-predictor (NSP) and a linear sub-predictor (LSP) are combined in a cascade form. This model is called "hybrid predictor" here. The nonlinearity analysis method of the input time series was also proposed to estimate the network size. We have considered the nonlinear prediction problem as a pattern mapping one. A multi-layer neural network, which consists of sigmoidal hidden neurons and a single linear output neuron, has been employed as a nonlinear sub-predictor. Since the NSP includes nonlinear functions, it can predict the nonlinearity of the input time series. However, the prediction is not complete in some cases. Therefore, the NSP prediction error is further compensated for by employing a linear sub-predictor after the NSP. In this paper, the prediction mechanism and a role of the NSP and the LSP are theoretically and experimentally analyzed. The role of the NSP is to predict the nonlinear and some part of the linear property of the time series. The LSP works to predict the NSP prediction error. Furthermore, predictability of the hybrid predictor for noisy time series is investigated. The sigmoidal functions used in the NSP can suppress the noise effects by using their saturation regions. Computer simulations, using several kinds of nonlinear time series and other conventional predictor models, are demonstrated. The theoretical analysis of the predictor mechanism is confirmed through these simulations. Furthermore, predictability is improved by slightly expanding or shifting the input potential of the hidden neurons toward the saturation regions in the learning process.
Kazuhiko HOSOMI Masataka SHIRAI Junji SHIGETA Tomoyoshi MISHIMA Toshio KATSUYAMA
A GaAs/AlGaAs directional-coupler-type device that use polariton propagation was fabricated and its switching operation was demonstrated. The length of the switching region is as small as 300 µm. The output signal modulation under an electric field shows typical characteristics of directional-coupler type switching. The measured operation voltage is 2 V for an operation wavelength of 805 nm at 10 K. The corresponding signal extinction ratio is 8 dB. These experimental results confirm the efficient operation of the polariton devices, which can be applied to especially small optical -switching devices with low-voltage operation.
Takeshi SAKAMOTO Nobuyuki TANAKA Yasuhiro ANDO
We have developed a low-latency, error-correcting-code-(ECC-)adaptable skew-compensation technique, which is needed for high-speed and long-distance parallel optical interconnections. A new frame-coding technique called shuffled mB1C encoding, which requires no clock-rate conversion circuit and no data buffering, and a new skew-measurement method which is suitable for ECC adaptation have been developed for the compensation. Full-digital skew-compensation circuits using these new techniques were able to compensate for a two-clock-cycle skew, even when one transmission channel was removed. The maximum latency for skew compensation was only five clock cycles.
Naohide NAGATSU Satoru OKAMOTO Masafumi KOGA Ken-ichi SATO
This paper discusses global area optical transport ring networks using wavelength division multiplexing (WDM) technologies and proposes a novel optical add/drop multiplexer (OADM) architecture suitable for such an application field. Study on the requirements of a global area ring application elucidates the appropriate ring/protection architecture as the path switched bi-directional ring. The proposed OADM architecture has flexibility in terms of path provisioning and scalability. We conclude that the proposed OADM can effectively configure the large-scale path switched bi-directional rings.
Masakuni TAKI Hirotaka HATAKENAKA Toshinobu KASHIWABARA
In this paper we propose an algorithm for generating maximum weight independent sets in a circle graph, that is, for putting out all maximum weight independent sets one by one without duplication. The time complexity is O(n3 + β ), where n is the number of vertices, β output size, i. e. , the sum of the cardinalities of the output sets. It is shown that the same approach can be applied for spider graphs and for circular-arc overlap graphs.
Shu-Lin HWANG Che-Chun CHEN Feipei LAI
Modern micro-architectures employ superscalar techniques to enhance system performance. Since the superscalar microprocessors must fetch at least one instruction cache line at a time to support high issue rate and large amount speculative executions. There are cases that multiple branches are often encountered in one cycle. And in practical implementation this would cause serious problem while there are variable number of instruction addresses that look up the Branch Target Buffer simultaneously. In this paper, we propose a Range Associative Branch Target Buffer (RABTB) that can recognize and predict multiple branches in the same instruction cache line for a wide-issue micro-architecture. Several configurations of the RABTB are simulated and compared using the SPECint95 benchmarks. We show that with a reasonable size of prediction scope, branch prediction can be improved by supporting multiple / up to 8 branch predictions in one cache line in one cycle. Our simulation results show that the optimal RABTB should be 2048 entry, 8-column range-associate and 8-entry modified ring buffer architecture using PAs prediction algorithm. It has an average 5.2 IPC_f and branch penalty per branch of 0.54 cycles. This is almost two times better than a mechanism that makes prediction only on the first encountered branch.