This paper describes how to effectively use discriminative training based on Minimum Classification Error (MCE) criterion for a small amount of data in order to attain the highest level of recognition performance. This method is a combination of MCE training and Vector-Field-Smoothed Bayesian learning called MAP/VFS, which combines maximum a posteriori (MAP) estimation with Vector Field Smoothing (VFS). In the proposed method, MAP/VFS can significantly enhance MCE training in the robustness of acoustic modeling. In model training, MCE training is performed using the MAP/VFS-trained model as an initial model. The same data are used in both trainings. For speaker adaptation using several dozen training words, the proposed method has been experimentally proven to be very effective. For 50-word training data, recognition errors are drastically reduced by 47% compared with 16.5% when using only MCE. This high rate, in which 39% is due to MAP, an additional 4% is due to VFS, and a further improvement of 4% is due to MCE, can be attained by enhancing MCE training capability by MAP/VFS.

The multichannel distortions of direct modulated laser diode were studied from the view point of rate equations. A novel technique for compensating the composite second order distortion (CSO) was proposed. Meanwhile, the related calibration procedures were indicated. After the compensation, 10 dB improvement in CSO was obtained

A three-terminal quantum device utilizing photon-assisted tunneling in a multilayer structure is proposed and analyzed in terms of its high frequency amplification characteristics. The operation principle of this device includes photonassisted tunneling at the input, formation of a propagating charge wave due to the beat of tunneling electrons and its acceleration, and radiation of electromagnetic waves at the output. Analysis of these operations, discussion of similarities and dissimilarities to classical klystrons, and estimation of the power gain and its frequency dependence are given. A simple example demonstrates that amplification up to the terahertz frequency range is possible using this device.

We analyzed the operation speed of the resonant tunneling logic gate, MOBILE, using a simple equivalent circuit model and varying parameters of I-V characteristics and capacitance of RTTs(resonant tunneling transistors). The switching time for large peak-to-valley(P/V)current ratios is smaller at small Vbmax(maximum bias voltage), but larger at large Vbmax than that for small P/V ratios in the case of present I-V characteristics with flat valley current. It is also demonstrated that the MOBILE operation fails if the bias voltage rises too fast, when the capacitance of the load and the driver is different due to the displacement current through the capacitance. These behaviors can be explained by considering the potential diagrams of the circuit.

A subclass of context-free languages, called pure context-free languages, which is generated by context-free grammar with only one type of symbol (i.e., terminals and nonterminals are not distinguished), is introduced and the problem of identifying from positive data a restricted class of monogenic pure context-free languages (mono-PCF languages, in short) is investigated. The class of mono-PCF languages is incomparable to the class of regular languages. In this paper we show that the class of mono-PCF languages is polynomial time identifiable from positive data. That is, there is an algorithm that, given a mono-PCF language L, identifies from positive data, a grammar generating L, called a monogenic pure context-free grammar (mono-PCF grammar, in short) satisfying the property that the time for updating a conjecture is bounded by O(N3), where N is the sum of lengths of all positive data provided. This is in contrast with another result in this paper that the class of PCF languages is not identifiable in the limit from positive data.

Progress made in the field of high speed networking technology has led to the planning and prototyping of true high-bandwidth applications with very high throughput and low delay requirements. In this study we approach the problem of high throughput demand from the aspect of protocols and introduce the handling of error control in the application layer level as opposed to the transport layer since the eventual destination of data is the application itself. This scheme, called ACER (Application Conscious Error Recovery), is proposed and defined for bulk data transfers. A simple analytic throughput comparison of the sliding window scheme with go-back-N, and ACER is given later, Also, a prototype implementation of ACER for bulk data transfer and experimental measurement results are presented. Besides, we investigate the performance of the scheme by simulation for various network models. Finally, we present a discussion of extending the scheme to different traffic patterns and applications.

A graph G is called an n-channel graph at vertex r if there are n independent spanning trees rooted at r. A graph G is called an n-channel graph if G is an n-channel graph at every vertex. Independent spanning trees of a graph play an important role in fault-tolerant broadcasting in the graph. In this paper we show that if G1 is an n1-channel graph and G2 is an n2-channel graph, then G1G2 is an (n1 + n2)-channel graph. We prove this fact by a construction of n1+n2 independent spanning trees of G1G2 from n1 independent spanning trees of G1 and n2 independent spanning trees of G2. As an application we describe a fault-tolerant broadcasting scheme along independent spanning trees.

In this paper, we propose a new design algorithm for nearly linear phase IIR digital filters with prescribed log magnitude response. The error function used here is the sum of the weighted log magnitude-squared error and phase -squared error, and so it is possible to control log magnitude and phase response directly. The gradient vector of the proposed error function is easily calculated as the closed form solution because of its nature, in which the real and imaginary part of the logarithm of a complex transfer transfer function corresponds to the log magnitude and phase response, respectively. This algorithm is simple and converges quickly. Finally, we show the validity of the proposed algorithm with some examples.

We propose a method to fuse auditory information and visual information for accurate speech recognition. This method fuses two kinds of information by using Iinear combination after calculating two kinds of probabilities by HMM for each word. In addition, we use full-frame color image as visual information in order to improve the accuracy of the proposed speech recognition system. We have performed experiments comparing the proposed method with the method using either auditory information or visual information, and confirmed the validity of the proposed method.

We studied the application of precursor solutions that can be fired into oxides to form a protective layer for AC-type Plasma Display Panel (AC-PDP). Our study of alkoxide and metallic soap as MgO precursors revealed that the crystallinity of MgO films depends on the starting substance. Since the electric discharge characteristics of a panel and the lamination effect of the protective layer depend on precursors, it was confirmed that binders having higher crystallinity provide better characteristics. Our study revealed that a compound-oxide film has high crystallinity. The application of a Ba0.6Sr0.4Gd2O4 formation solution to a binder and the application of a Sr0.6Mg0.4Gd2O4 formation solution to a protective layer both are seemed promising We also found that a double-layer film, made by forming a protective layer of fine MgO powder and a Ba0.6Sr0.4Gd2O4 binder, on top of a protective layer made of fine MgO powder and a MgO binder, provides a luminous efficiency 5.3 times higher than that of sputtered MgO film which is one of candidates for the large panel, and the conventional electron beam evaporation is not suitable for the large panel. We further found that a triple-layer protective film made by forming a thin film of Sr0.6Mg0.4Gd2O4 provides low voltages of 1 V in firing voltage (Vf) and 35 V in sustaining voltage (Vs) compared to the double-layer film and provides a luminous efficiency 5.5 times higher than that of sputtered MgO film. A life test revealed the triple-layer film in particular providing a useful life of more than 10,000 hours. From these findings, we concluded that the compound-oxides which is composed of alkaline-earth-metal and rare-earth-element could be applied effectively to a protective layer for AC-PDP.

The paper deals with Kleenean functions defined as fuzzy logic functions with constants. Kleenean functions provide a means of handling conditions of indeterminate truth value (ambiguous states) which ordinary classical logic (binary logic) cannot cope with. This paper clarifies a necessary and sufficient condition for a function to be a Kleenean function. The condition is provided with a set of two conditions, and it will be shown that they are independent of each other.

For the expansion of using the integral equation methods on wave-field analysis, a new method called "Source and Radiation Field Solution" is suggested. This solution uses a couple of integral equations. One of them is the traditional integral expression giving the scattered field from the wave source, another is newly proposed one which expresses the wave source from both of the source and the scattered field, by using the conjugate Green function expression. Therefore this method can derive both of the source and the scattered field at the same time by coupled two equations. For showing the effect of this method, we analyze scattering problems for dielectrics in this paper.

In this paper, we present an analysis of microstrip line with a trapezoidal dielectric ridge in multilayered media. The method employed in this characterization is called partial-boundary element method (p-BEM) which provides an efficient technique to the analysis of the structures with multilayered media. To improve the convergence of the Green's function used in the analysis with the P-BEM, we employ a technique based on a combination of the Fourier series expansion and the method of images. Treatment on convergence for the boundary integrals is also described. After this treatment, it requires typically one tenth or one hundredth of Fourier terms to obtain the same accuracy compared with the original Green's function. Numerical results are presented for two microstrip lines that have a trapezoidal dielectric ridge placed on a one-layered substrate and a two-layered substrate. These numerical results demonstrate the effects on the characteristics of the microstrip line due to the existence of the dielectric ridge as well as the second layer between the ridge and the fundamental substrate.

We investigate bifurcations of the periodic solution observed in a phase converter circuit. The system equations can be considered as a nonlinear coupled system with Duffing's equation and an equation describing a parametric excitation circuit. In this system there are two types of solutions. One is with x = y = 0 which is the same as the solution of Duffing's equation (correspond to uncoupled case), another solution is with xy0. We obtain bifurcation sets of both solutions and discuss how does the coupling change the bifurcation structure. From numerical analysis we obtain a codimension two bifurcation which is intersection of double period-doubling bifurcations. Pericdic solutions generated by these bifurcations become chaotic states through a cascade of codimension three bifurcations which are intersections of D-type of branchings and period-doubling bifurcations.

A high-frequency approximate method for calculating the diffraction by a smooth convex surface is presented. The advantage of this method is the validity of it in the caustic region of the creeping rays where the Geometrical Theory of Diffraction (GTD) becomes invalid. The concept used in this method is based on the Method of Equivalent Edge Currents (EEC), and the equivalent line currents for creeping rays which are derived from the diffraction coefficients of the GTD are used. By evaluating the radiation integral of these equivalent line currents, the creeping ray contribution which is valid within the caustic region is obtained. In order to check the accuracy and the validity of the method, the diffraction problem by a perfectly conducting sphere of radius a is solved by applying the method, and the obtained results are compared with the exact and the GTD solutions. It is confirmed from the comparison that the failure of the GTD near the caustic is removed in this method and accurate solution is obtained in this area for high-frequency (ka8). Furthermore, it is also found that this method is valid in the backward region (0θ90, θ is an observation angle mesuered from an incident direction), whereas not in the forward region (90θ180).

Microcellular systems are suitable as personal mobile communication systems because of their high channel re-use efficiency and low transmission power. To implement a microcellular system, the antennas of base stations should be low enough, compared to the buildings around them, to reduce the interference to or from other base stations. In high-speed digital mobile radio communications, the time delay spread caused by multipath propagation is a significant factor in determining the maximum data transmission rate. In the case of a low-antenna-height microcellular system, the propagation characteristics rapidly change when the mobile terminal moves from a line-of-sight (LOS) location to a non-line-of-sight (NLOS) location. In this paper, the time dealy spread characteristics under LOS and NLOS conditions are examined using a geometrical street model which has a reflecting wall at one end of the street on which the base station is located. The RMS delay spreads are calculated using optical ray theory, taking into consideration the wedge diffraction on the street corner. If a reflecting wall exists, the RMS delay spread increases as the mobile terminal moves away from the base station under LOS conditions, or away from the street corner under NLOS conditions. The calculated results agree with the experimental results if measuring equipment noise is taken into consideration.

Personal communication systems (PCS) have more signalling traffic than conventional fixed networks and require large-scale databases to manage users' profiles, which are sets of data items, such as the location the user is currently visiting and the user's authentication key, necessary for a PCS user to be provided with PCS services. This paper focuses on inter-network roaming in PCS environments. In designing a PCS supporting roaming service, it is essential to avoid increased signalling traffic and data searching time in the database. We first identify the appropriate domains for three routing schemes-Direction Routing, Redirection Routing, and Look-ahead Routing-from the viewpoints of the number of signals for inter-network roaming and roaming probability. We do this for two kinds of PCS database network architecture, Home Location Register (HLR) and Visitor Location Register (VLR), and show that Look-ahead Routing is the best scheme for the HLR network architecture (considering the number of signals for intra-network and inter-network database access) and that in the VLR network architecture, the decreasing of the roaming probability expands domains for which Redirection Routing is appropriate. We also propose a generic PCS data model that inter-network roaming interfaces can use to search effectively for a user's profile. The data model clarifies the contents of a set of data items which share certain characteristics, data items that the contents compose, and the relationships (data structures) between sets of data items. The model is based on the X. 500 series recommendations, which are applied for an Intelligent Network. We also propose a data structure between sets of data items using the directory information tree and show the ASN. 1 notations of the data model.

Abstract machine modelling is a technique used frequently in developing the retargetable compilers. By translating the abstract machine operations into target machine instructions, we can construct retargetable compilers. However, such a technique will cause two problems. First, the code produced by the compilers is inefficient. Next, in order to emit the efficient code, the compilation time is too long. In view of these two disadvantages, we apply PO (peephole optimizer) in our retargetable compilers to do code optimization. Peephole optimizer searches for the adjacent instruction candidates in the intermediate code, and then replaces them with equivalent instructions of less cost. Furthermore, the peephole description table consists of simple tree-rewriting rules which are easily retargeted into different machines. At the same time, we have proposed a simple peephole pattern matching algorithm to reduce the peephole pattern matching time. The experiment indicates that the machine code generated by our compiler runs faster than that by GNU c compiler (gcc).

The problem of test generation for VLSI circuits computationally requires prohibitive costs. Parallel processing on a multiprocessor system is one of available methods in order to speedup the process for such time-consuming problems. In this paper, we analyze the performance of parallel test generation for combinational circuits. We present two types of parallel test generation systems in which the communication methods are different; vector broadcasting (VB) and fault broadcasting (FB) systems, and analyze the number of generated test vectors, the costs of test vector generation, fault simulation and communication, and the speedup of these parallel test generation systems, where the two types of communication factors; the communication cut-off factor and the communication period, are applied. We also present experimental results on the VB and FB systems implemented on a network of workstations using ISCAS'85 and ISCAS'89 benchmark circuits. The analytical and experimental results show that the total number of test vectors generated in the VB system is the same as that in the FB system, the speedup of the FB system is larger than that of the VB, and it is effective in reducing the communication cost to switch broadcasted data from vectors to faults.

For reducing requirement of large memory and minimizing computation complexity in a large-vocabulary continuous speech recognition system, speech segmentation plays an important role in speech recognition systems. In this paper, we formulate the speech segmentation as a two-phase problem. Phase 1 (frame labeling) involves labeling frames of speech data. Frames are classified into three types: (1) silence, (2) consonant and (3) vowel according to two segmentation features. In phase 2 (syllabic unit segmentation) we apply the concept of transition states to segment continuous speech data into syllabic units based on the labeled frames. The novel class of hyperrectangular composite neural networks (HRCNNs) is used to cluster frames. The HRCNNs integrate the rule-based approach and neural network paradigms, therefore, this special hybrid system may neutralize the disadvantages of each alternative. The parameters of the trained HRCNNs are utilized to extract both crisp and fuzzy classification rules. In our experiments, a database containing continuous reading-rate Mandarin speech recorded from newscast was utilized to illustrate the performance of the proposed speaker independent speech segmentation system. The effectiveness of the proposed segmentation system is confirmed by the experimental results.