Showing papers in "Electronics and Communications in Japan Part Iii-fundamental Electronic Science in 1992"

Proposal of chaotic steepest descent method for neural networks and analysis of their dynamics

Abstract: Most of the neural network models based on the Lyapunov stability contain various problems such as the trap of the local minimum, and the limit of their dynamic performances has been pointed out. This paper attempts to provide a new dynamic performance to such a neural network model by introducing chaos dynamics. The features of the chaotic dynamic model proposed here is that the dynamical equation describing the trajectory on the energy curve has a periodically varying nonlinear resistance in the dissipation term. By iterating the stable and unstable phases, the chaotic transitions of the state can be realized. The proposed dynamic model is applied to the error backpropagation learning and the memory recall in the Hopfield-type network, and the chaotic minimum transitions in the dynamic process are verified, it is verified further that the intermittent chaos generated by an appropriate parameter manipulation can introduce useful dynamic behaviors into the network, e.g., the flexible learning and the memory recall with a structure containing both stability and plasticity.

Dc analysis of nonlinear circuits using solution‐tracing circuits

Abstract: The dc driving-point and transfer characteristics of nonlinear circuits are the multivalued curves that arise from the nature of the circuit. These curves cannot be analyzed by general-purpose circuit simulators. One known method for analyzing these kinds of characteristic curves is the backward differentiation formula (BDF) curve-tracing algorithm proposed by Ushida and Chua. In this method, the circuit equations f(x) = 0, f(·): Rn+1 Rn, where the input voltage is assumed to be a variable, are analyzed by the predictor-corrector algorithm where the arc-length of the solution curve in n + 1-dimensional space is the parameter. However, it is not clear that this method is practical for large-scale circuits. In this paper, we extend the Ushida-Chua method from a practical method standpoint and demonstrate that the multivalued characteristic curves of large-scale circuits can easily be analyzed using general-purpose circuit simulators. In the proposed method, first, the solution curve in n + 1-dimensional space is projected into m + 1-dimensional space, where m ≤ n and the arc-length of this new curve is used as the parameter. Second, the relationship between the arc-length and the components of the curve is expressed by a function generator circuit, the solution-tracing circuit. Finally, transient analysis is performed using a general-purpose circuit simulator and the solution curve is traced. The effectiveness of this method is verified through several examples, including a bipolar analog IC with 296 nodes.

A method of two‐dimensional spectral analysis using the wigner distribution

Abstract: Wigner distribution (WD) has recently attracted attention as a powerful tool for time-varying signals analysis and several applications have been proposed. WD is calculated as the Fourier spectra of a signal with special preprocessing. The preprocessing allows the frequency resolution to be better than that of Fourier transform (FT). However, this preprocessing causes WD to generate interference terms and more complicated aliasing compared to FT. To cope with these problems, WD is usually calculated using analytic signals. In one-dimension, the method of deriving analytic signal is unique. However, in two-dimensions (2-D), several methods have been proposed. This paper presents the reason for 2D-WD to generate aliasing and interference for negative frequency components and its solution. Based on the discussion, a new analytic signal for 2D-WD is proposed together with a method of new spectral analysis using the signal. The proposed method offers higher frequency resolution and no interference term for negative frequency components, as well as no aliasing, providing characteristics similar to WD. By using this method, spectral analysis for low frequencies can be realized with small-sized windows.

Measurement of sound and vibration at the lips, nostrils, and pharynx wall in speech utterance and simulation of sound leakage from the oral cavity to the nasal cavity in nonnasal sounds

Abstract: Sounds and vibrations at the lips, the nostrils, and skin near the pharynx are measured separately in an anechoic room using a specially designed soundproof box with three microphones and three accelerometers. It is found that the radiation from the nostrils is not negligibly small even if the nonnasal sounds were spoken and that the nostril sound in a close vowel /i/ is stronger than that in an open vowel /a/. A buzz bar of voiced plosives is found to be radiated from both the nostrils and skin near the pharynx. Based on these findings, a vibrating plate model is introduced to simulate the acoustic coupling between the oral cavity and the nasal cavity through the velum. To simulate the radiation of the buzz bar of voiced plosives, an active inflation of volume of the pharynx cavity is introduced and the relationship between the inflation rate and the sound radiation is shown.

Neural network using oscillators

Abstract: This paper proposes an oscillator neural network which replaces the sigmoidal nonlinearities in a Hopfield neural network with oscillators. When oscillators are used as saturating nonlinearities and their rise and fall times are utilized, results similar to simulated annealing are obtained, and therefore the desired solution can be obtained in a relatively short time. As examples, we show that the 4-bit A/D converter problem, the 10-city traveling salesman problem, and a simple associative memory problem can be solved by the oscillator neural network.

A synthesis of switched capacitor inductance simulation circuit using voltage followers

Abstract: Voltage followers are used for the switched capacitor (SC) circuit simulating the ungrounded inductance based on the bilinear s-Z transform. The operation is confirmed by experiments. In the circuit configuration, it is pointed out that the block diagram often used in the derivation of the SC inductor complicates the circuit and the control clock and a new block diagram is proposed. The circuit derived by this method is simple since the control clock has two phases and the spreading of the capacitance value is only 2. Further, the circuit is not affected by the parasitic capacitance caused by the substrate-side electrode of the integrated capacitor which typically affects the characteristics significantly. Although two voltage followers are needed for realization of one inductor, it can be considered that effectively one inductor is constructed with one voltage follower if the adjacent SC resistance and SC inductor are shared with a voltage follower. It is described that the spread of the capacitance value in the entire filter can be made smaller than other SC inductors when this SC inductor is used in the SC filter. Simplification of the circuit with this filter is investigated.

Evaluation of coded speech quality degraded by cell loss in ATM networks

Abstract: The asynchronous transfer mode (ATM) is considered interesting as the signal transfer mode in multimedia communication which can realize flexible and efficient channel usage. In ATM, the quality of speech is degraded by cellular handling of the signal which has a property different from the quality degradation in the conventional STM. Due to the small number of studies reported so far, the characteristics of the quality degradation have not been well understood. This paper presents the design data for the speech quality in ATM network and considers the cell loss as the inherent factor degrading the quality in ATM. The complex effect of the cell loss and the coding distortion on the overall subjective quality and the articulation of the speech is measured and discussed. For several methods of speech coding and cell structures, the degradation of quality is examined when the cell loss rate increases. It is shown that the 32-kbit/s embedded ADPCM with the cell structure, where the lower two bits are selected with priority to suffer the loss, has a high immunity to the cell loss. Through the quality evaluation of the coded speech containing a cell loss, two factors are shown to be important: the speaker and the speech power at the site of cell loss. As a result of this study, new knowledge is obtained concerning the speech quality degradation characteristics in ATM network, and the data and knowledge useful in the speech quality design of ATM network are obtained.

A parameter optimization method for electronic circuit design using stochastic model function

Abstract: This paper presents an optimization procedure using the stochastic model function, aiming at the application to the optimization of the circuit constant in the integrated circuit. The optimization of the circuit constant is formulated as a minimization problem for the objective function derived from the circuit response under the specified condition. In the method presented in this paper, the model function is constructed which approximates well the objective function, and the objective function is minimized by minimizing the model function. The features of the method are that the algorithm is simple, the number of evaluations of the objective function is small, the result depends less on the initial value, and the minimum of the objective function can be determined with a high accuracy. The method is evaluated using the standard test function, and it is verified that all minimum values of the objective function can be determined with a high accuracy with fewer number of calculations of the objective function. The optimization procedure is applied to two circuit examples: the analog control circuit and the inverter chain. The result is satisfactory and the usefulness of the presented optimization procedure in the circuit design is demonstrated.

Design of layered-neural nonlinear filters using backpropagation algorithm

Abstract: The signal such as an image signal is a non-Gaussian signal with an abrupt change. When a random noise is superposed on such a signal, it is impossible to eliminate the noise effectively by using the traditional linear filter. Then it is desired to develop a nonlinear filter. From such a viewpoint, this paper proposes a nonlinear digital filter using a layered neural network, aiming at the elimination of such a noise. By thus introducing the nonlinearity and the learning ability of the layered neural network, the noise can effectively be eliminated. In this paper, the noise elimination is shown for the case of a one-dimensional test signal and a two-dimensional image signal indicating the usefulness.

Theory on extended form of interpolatory approximation of multidimensional waves

Abstract: This paper presents a comprehensive discussion of the approximation of the n-dimensional wave f(X) using the sampled values of the output wave obtained by exciting a series of time-invariant linear circuits by the wave f(X). It is assumed that the approximate wave h(X) of f(X) is given by the sum of sample values of the output wave multiplied by certain n-dimensional waves. For simplicity, n-dimensional waves to be multiplied with the sample values are called the interpolation functions. The set of sample points treated in this paper is defined as a subset obtained by sampling periodically the vertices of the n-dimensional parallelepipeds placed periodically in the space Rn. Such a set of sampling points includes the most of the typical arrangements of the sampling points, such as the hexagonal and the octagonal lattices on the two-dimensional space. It is assumed that the sample values contain statistically independent errors such as the observation error and/or the quantization error. Moreover, it is assumed that the interpolation functions have the supports which are parallel-translations of each other. First, it is assumed that the functional forms of these interpolation functions may be different. Further, a set of n-dimensional waves is considered where the corresponding spectrum has the weighted p-norms smaller than the prescribed positive constant. The standard deviations of the difference between f(X) and their approximations are considered. As the measure of the approximation error, the upper limit of the standard deviation obtained by varying the original waves over the given set of waves is adopted. In the following sections it is shown that the interpolation functions minimizing the forementioned measure of error can be expressed as the parallel-translations of a finite number of functions. Further, in special cases, the interpolation functions have the discrete orthogonality. Since the measure of error is a convex function of the interpolation functions, it is ensured that the global optimum is obtained easily by using the ordinary numerical optimization. For some special cases, the concrete expression for the optimal interpolation functions are derived. Considering the approximation system in the reverse direction, where the linear circuits first passing the input wave are exchanged with the interpolation filters, it is shown that the new interpolation functions also minimize the same measure of error. As a direct consequence, a successive approximation of the interpolation functions is presented, which is suited to the applications such as the multiplex communication of the images of both directions.

Effects of respiration on heart‐rate variability measures

Abstract: There are many studies that measure the mental workload or evaluate the autonomous nervous function by heart-rate variability measures. Heart-rate variability is influenced not only by the stress or the workload but also by respiration. Therefore, it is important to clarify the effects of respiration on heart-rate variability measures. This paper discusses the effects of the respiration frequency, the depth of respiration and the dispersion of the respiration interval on heart-rate variability. It was shown that the number of peak and trough waves in the time series of R-R intervals was influenced by the respiration frequency. Moreover, the measures obtained by the power spectral analysis of R-R intervals were found to be influenced considerably by the respiration frequency. The heart-rate variability measures tended to increase with deep respiration. The effects of the dispersion of respiration intervals on heart-rate variability measures were not so remarkable.

A synthesis of pipelineable wave digital filters by scattering matrix factorization

Abstract: This paper proposes a synthesis of low-sensitivity digital filters which are pipelineable and have simple structures. Their sensitivity characteristics are similar to those of wave digital filters. The synthesis of these filters is attained by factorizing paraunitary scattering matrices derived from given z-domain transfer functions without referring to analog filters. The synthesized structures are the T-cascade connection of fundamental sections which are constructed by using series and parallel sections of wave digital filters. The proposed synthesis method can realize any given IIR transfer function, while previous methods which synthesize T-cascaded structures can realize only the amplitude response of any given transfer function. In addition, the method can realize the transfer functions of which the order of their denominator is higher than that of their numerator including FIR transfer functions.

Student model and its recognition by hypothesis‐based reasoning in ITS

Abstract: This paper describes a framework to infer a student's misconception from observed errors during problem-solving processes. A human teacher can generate hypotheses about reasons for an error by observing a student's problem-solving process. The teacher is also able to identify a student's misconception during the process of verifying these hypotheses. Furthermore, by using these hypotheses, the teacher can generate new tasks to evaluate the student's understanding level. In this way, appropriate instructions based on the student's knowledge structure can be provided. To accomplish such a behavior within an intelligent tutoring system (ITS), the authors have defined a domain model and applied hypothesis-based reasoning to diagnose the student model. When the system finds an error in a student's problem-solving process, it attempts to generate hypotheses which explain that error in terms of the domain model.

A method for selecting the most suitable pitch from some candidates utilizing its time continuation

Abstract: Pitch frequency is a basic characteristic of the human voice, and pitch extraction is one of the most important studies for speech recognition. This paper describes a method for selecting the most suitable candidate from candidates (pitch candidates) obtained by the short-range autocorrelation function. The selection is done utilizing the time continuation which is realized by that history of pitch candidates at previous frames referred to with a referential matrix. In addition, an improved result of pitch extraction is obtained by introducing restraints of double-pitch and half-pitch. The pitch extraction was performed for three males and three females. The proposed method improves the percentage of correct pitch from 64.86 to 91.06 percent.

Numerical analysis of bifurcations in Duffing's equation with hysteretic functions

Abstract: This paper analyzes numerically the bifurcation phenomenon in Duffing's equation containing a hysteretic function and reports on the result of investigation of the effect of the hysteretic characteristics on the bifurcation. First, a large number of steady hysteresis loops is derived by numerically integrating the state equation of the system with various initial conditions, and the sets of the steady periodic solutions are observed. Then, using the method already proposed by the authors, the bifurcation of the steady solution is analyzed while varying the parameters of the system. The effects of the hysteretic loss and the ohmic loss on the bifurcation of the system are compared, and the difference is discussed. An example of the generation of chaos caused by increasing the hysteretic loss is shown.

Spectral estimation method of a multistage nonstationary process and its application

Abstract: This paper considers the time series observed through the nonstationary processes and proposes the multistage instantaneous maximum entropy method aiming at the inverse estimation of the behavior of the original process on the spectrum. The performance of the method is demonstrated and examples of useful applications are presented. To achieve the goal, the cascaded multistage AR process is considered as the model for the process from the excitation source of the time series to the observation. The excitation source is assumed as a simple white Gaussian noise. The response of the process is represented by a time-varying lattice filter. The instantaneous value of the power spectrum is determined from the prediction error estimation function defined at a point on the time axis (instantaneous maximum entropy method, IMEM). Then the nonstationarity (time-varying property) of the process is noted, and the separation of the connected processes is discussed. For this purpose, the nonstationary frequency axis of the two-dimensional spectrum is considered based on IMEM. A filter to separate the considered process from others is designed. This filter passes the time-varying component of the considered process but eliminates the time-varying components due to the interference from other processes, based on the a priori information concerning the time-varying properties of the considered process. Finally, this paper presents the results of evaluation for the test using the artificial signal as well as the results of evaluation for two application examples. Comparing the result to that of the traditional method, the novelty and the usefulness of the proposed method are demonstrated. The system features are that the accumulation of the error in the multistage prediction can efficiently be suppressed using IMEM with the excellent prediction performance, which results in the high follow-up ability to the time-varying process, and the excellent spectrum estimation.

An adaptive algorithm of a logarithmic‐type adaptive digital filter based on the LMS/Newton method

Abstract: The logarithmic adaptive digital filter (ADF) estimates the unknown system in which the transfer function is represented by a rational function of z. It consists of an ADF to estimate the logarithm of the denominator polynomial of the transfer function and a transversal ADF to estimate the numerator polynomial. Compared to the conventional transversal ADF, it is known that the number of taps can be reduced in this kind of ADF. Studies have been made concerning the adaptive algorithm of the logarithmic ADF and the convergence condition. A problem is that the correlation matrix is a function of the input and the output signals of ADF. This increases the convergence time, compared to the transversal ADF, when the input signal is white. This paper discusses the method of the improvement of the convergence speed and proposes the adaptive algorithm for the logarithmic ADF based on the LMS/Newton method. The successive computation of the inverse matrix of the correlation matrix (Jacobian) as well as the adaptive algorithm are derived. In contrast to the algorithm for the transversal ADF, it is shown that the square term of the estimation error is included in the update term of the tap vector. Finally, the convergence condition for the proposed algorithm is derived. The convergence speed, the adaptive algorithm and the convergence condition are examined by a computer simulation.

Use of a laplace-transform relaxation method for the analysis of nonlinear electric circuits

Abstract: The relaxation method is one of the methods for solving an initial value problem for a set of differential equations and is suitable for analysis of large-scale circuits such as those simulating VLSI circuits. In analysis of the relaxation method, time domain analysis is most common but complex frequency domain analysis by means of the Laplace transform also is applicable. This paper presents the Laplace-transform analysis. Semiconducting elements such as transistors of the VLSI circuits exhibit nonlinear characteristics. Thus, the differential equations to be used in the relaxation method must be nonlinear. In some cases, the properties of nonlinear elements are given not by analytic functions but by a series of discrete numerical values such as observed data. To express the nonlinearity of these elements, multidimensional interpolation formulas are used in this paper and practical computer algorithms for solving them are derived. The analysis of the conditions of convergence of the relaxation method shows that the proposed algorithm is applicable to circuits which operate in both stable and unstable domain, such as oscillators.

VLSI-oriented design method of asynchronous sequential circuits based on one-hot code and two-transistor and logic

Abstract: This paper proposes a new method of synthesis for a CMOS LSI-oriented asynchronous sequential circuit with the two-transistor AND logic structure, using the one-hot code for the internal state assignment. The assignment of the state by the one-hot code is executed uniquely, independently of the properties of the state-transition table. The CPU time for state assignment, which has been a problem, can be ignored. The critical race-free property, which is then a problem, is solved by a 4-FET set/reset flip-flop which stores the state variable and the drive circuit for the flip flop, not the internal code assignment. The set and the reset driving circuits have the AND logic circuit structure, where two transistors are placed (with the state variable and the external input as the gate inputs) on each conduction path. The synthesized circuit is hazard-free. It realizes a pseudo-CMOS operation without static power consumption. Comparing the method proposed in this paper with other methods of synthesis, it is compact and especially suited to the large-size asynchronous sequential machine (with 40 or more cells in the row-column product of the state transition table).

Image processing by two-dimensional recursive filters using the fornasini-marchesini model

Abstract: This paper considers a simple two-dimensional (2-D) IIR filter based on the Fornasini-Marchesini local state-space (LSS) model. It is shown that if image processing is carried out using such filters from four directions, smoothing, edge detection or edge enhancement can be achieved without any distortion. The proposed technique allows one to flexibly perform the forementioned filtering by choosing three parameters. Moreover, filter analysis including stability is easy due to using a well-known 2-D LSS model. Finally, some examples are given to illustrate the utility of the proposed technique.

Design of FIR fan filters using FIR hilbert transformer

Abstract: A new, more direct method for synthesizing FIR fan filters and quadrantal fan filters using a FIR Hilbert transformer as the one-dimensional (1-D) prototype filter is described. Only the case 3 transfer function of the Hilbert transformer is available. When an equiripple design is used for the transformer, the ripple amplitudes in the passbands and stopbands of the resulting fan filter are nearly the same as the ripple amplitude of the transformer. The transfer function of the fan filter has only a few nonzero coefficients because the Hilbert transformer is required to be of even order. Illustrative examples of fan filter synthesis are presented.

An efficient method for making a fault dictionary for short-circuit fault diagnosis of linear networks

Abstract: In the fault diagnosis for the breakdown fault of the analog circuit such as short- or open-circuit, the fault dictionary usually is employed which describes the input-output characteristics for the accessible nodes for all assumed faults. In the construction of the fault dictionary, generally a large number of inverse matrix calculations is required, and the computational complexity becomes tremendous with the increase of the circuit scale. Thus, how to construct and maintain the fault dictionary efficiently is important. This paper especially considers the short-circuit fault in the fault diagnosis of the analog circuit and proposes a method which efficiently constructs the fault dictionary. This method considers the difference of the input-output characteristics in normal and fault operations, for which the efficient calculation is presented. The method can be characterized as the extension of Johnson's proposal based on Householder's technique. In the diagnosis by dictionary, the fault generally is diagnosed by comparing the items of the fault dictionary and the measured result. By the proposed method, the computational complexity in constructing an item of the dictionary can be reduced to the same order as that of the comparison procedure (except for the preliminary calculations in common to the whole item).

M/G/1 system with set-up time for server replacement

Abstract: This paper proposes an M/G/1 queueing model in which the server is replaced by another one with different service time distribution when the number of customers in the system exceeds a threshold value. A random set-up time is needed to introduce the second server. This server works until the system becomes empty. This kind of system will appear, for instance, when the server is changed to a faster one as the number of waiting customers becomes too great. In data communication systems, the set-up time may correspond to the time until the new carrier frequency for the faster transmission speed becomes stable, the time needed to assign a new link and/or the time necessary to inform a partner of a change in line speed. The behavior of the customers who receive the first and second kinds of service is analyzed using the theory of finite Markov chain and the decomposition property of the M/G/1 system with server vacations, respectively. Important performance measures, such as the distribution and average of the number of customers in the system and their average time in the system, are obtained. The influence of server characteristics, the conditions under which the server is replaced, and the distribution of set-up time on the performance of the system are clarified.

Development of support facilities for arithmetical word problem exercises in intelligent tutoring system

Abstract: In the problem solving of the arithmetical word problems, the process until the solution method is identified as the most important. It is necessary to design and develop the basic functions for exercise problem support in an intelligent tutoring system (ITS), such as problem solving, problem generation, and problem explanation, aiming at the support for the identification of the solution method. In this study, those three basic functions are realized based on the problem-solving method MIPS (model of indexing in problem solving). This paper describes the problem-solving module, the problem-generation module and the problem-explanation module. The problem-solving module not only solves the arithmetical word problem but also generates the cognitive description of the identification process for the solution method. The problem-generation module can generate the problem which specifies what students should do in the identification process. The problem explanation module can indicate the equivalence and difference between any two problems concerning the identification process.

Simulation of neural network digital adder and multiplier

Abstract: A novel adder and multiplier using Hopfield-type neural networks for which a technique is introduced to avoid the false operations is proposed, and the simulation results for these operations are illustrated. The main feature of these circuits is high-speed operation which is achieved at the almost constant computation time due to their parallel calculation ability even if the number of bits in digital circuits increases. The previously proposed adder has a problem in that some false operations occur in the special cases of addition. The error between the correct and false values in the operations is always only a binary value of ±1 once the false operation occurs. This paper describes in detail the technique to avoid the false operations. It is important to determine the ratio of weight of the object function to constraint function, taking into consideration the energy function that should be chosen. Computer simulations for the adder verify that the correct operation can be achieved by choosing the special ratio of the weight for these functions. On the other hand, a novel multiplier is described using a Hopfield-type network which is composed of the proposed adder and a neural network to perform a logical product. The simulation results for the multiplier show that the correct operations of a multiplication can be achieved with high-speed performance. It is clear from these results that a high-performance digital adder and multiplier can be constructed by using the Hopfield-type neural network model.

Estimation errors of volterra kernels measured by use of nonwhite input

Abstract: It is well known that employment of nonwhite input in system identification often causes large estimation errors. While the mechanism of inducing these errors has been well understood for linear system identification, it has not been made clear yet for nonlinear systems. This paper analyzes this mechanism for nonlinear system identification with Volterra functional series models. A vector space approach gives a clear physical interpretation of the error generation from a viewpoint of the excitation intensity of the input against the identified system. Based on this analysis, new algorithms of reducing estimation errors are proposed.

A new FIR switched-capacitor circuit for high-frequencies with low-power consumption (parallel cyclic-type circuit)

Abstract: The high-frequency performance of switched-capacitor filters is bounded mainly by the finite gain bandwidth product of operational amplifiers. In this paper, the effects of finite gain bandwidth product of operational amplifiers on the performance of FIR (transversal) filters using such amplifiers as delay elements (canonical circuit) is analyzed. Next, a new low-power consumption FIR switched-capacitor circuit for high frequencies, which we call a parallel cyclic-type circuit (PCTC), is proposed. The proposed circuit does not use operational amplifiers as delay elements. Instead, switches and capacitors are used for both the delaying and the weighting operations. We present a basic circuit using subblocks, a circuit with reduced number of capacitors, and a modified circuit having plural operational amplifiers for higher-frequency use. We then show the design of a fourthorder FIR filter employing PCTC. Under the condition that sampling frequency is one-half of unity gain frequency of operational amplifier, computer simulations of fourth-order transversal switched-capacitor filters with canonical circuit and that with parallel cyclic-type circuit are performed and the experimental results of these circuits using discrete components are also given. Both demonstrate that the parallel cyclic-type circuit has significantly better performance than the canonical circuit.

Determination of the number of sinusoidal signals in AR spectral estimation

Abstract: In carrying out the AR spectral estimation of observed signals consisting of the sum of sinusoidal signal and white noise, since at low SNR many more false peaks occur than the number of sinusoids, a method which uses the eigenvalue decomposition of the correlation matrix to adopt only as many eigenvalues as the number of sinusoids, discarding the rest, is very effective. However, if the number of signals is unknown, how the discarding of eigenvalues should be determined becomes an important problem. In this paper, to select an appropriate number of eigenvalues of the correlation matrix of the observed data, a multiple number of adjustable parameters is introduced to square the prediction error and, by determining them optimally, determine the number of eigenvalues. Two new methods are proposed. The first method decides the truncation number of eigenvalues so that the Bayes information criterion with a priori distribution is minimized. The decision criterion is derived and its properties are investigated. The second method determines the number of sinusoids by computing adjustable parameters so that the AR spectrum of a given rank is the closest to the true AR model representing the sinusoidal signal model approximately, that is, the MSE criterion is minimum. Finally, the effectiveness of the authors' methods are investigated through numerical examples.

A proposal on cooperative development environment for next-generation circuit simulator

Abstract: With the recent development of large-scale, high-level functions and high-speed VLSI circuits, intensive research and development efforts have been made to obtain new circuit simulation techniques. This paper proposes a cooperative development environment, which is indispensable in developing the next-generation circuit simulation technique. Assuming UNIX and C language, a proposal for the following scheme is presented. Various circuit simulation functions are realized as independent programs (simulation commands) as far as possible. By giving specifications for those commands in a unified way, data are shared among different simulation techniques. The input file for the simulation command is constructed by filing three kinds of data: circuit connection data; the device model data; and the data for the control. The core file is used as the output file. Then the function and realization of the simulation shell, which is to execute the simulation command interactively or to efficiently share the data among the commands, are discussed. Finally, the method of sharing the circuit simulation program using the shared library function is proposed. Experimental examples are presented for the simulation commands such as shared library function and transient analysis.