Two-dimensional simulation for bi-directional thyristors is carried out to clarify switching mechanisms. The switching process can be divided into four regions; (1) an avalanche breakdown region, (2) a negative resistance region, (3) a low current ON-state region and (4) a high current ON-state region. This paper considers the changes in current path and carrier distribution corresponding to the four regions for the thyristor. A way to apply the two-transistor model is proposed after analyzing inherent electron and hole currents in this thyristor. Furthermore, the effects of Auger recombination and band-gap narrowing on current-voltage characteristics are shown. We intend to apply these results to the design of semiconductor lightning surge protectors.
The lifetime of Xe flashlamp was tested by changing the geometrical shape and materials of electrode, and the fused silica tubes with and without large amount of OH base. Tungsten was best material as a electrode, because the deterioration of fused silica tube caused by electrode evaporation due to flashlamp discharge at high repetition-rate was minimum. The lifetime of Xe flashlamp was lengthened by using the fused silica tube with high thermal conductivity.
In this paper we develop a new approach to identify noncausal AR models driven by non-Gaussian i. i. d input. Under a few moderate assumptions, we derive the necessary and sufficient conditions of rebuilding the parameters of the AR models from 2nd and higher order statistics. We show that the AR model parameters are directly related to the solution of an eigenproblem. Based on the result, we present a method of AR model identification, applying eigenvector computation. The unique solution of AR parameters is guaranteed up to sign and linear phase ambiguity. Model order determination is not crucial in the method. If the order is overestimated, several equivalent AR models with different linear phases will be obtained.
In conventional image coding systems, two dimensional square blocks has been used for moving compensation and DCT. But sometimes these methods introduce great degradation, such as block distortion or mosquito noise, into objective quality. And horizontal one-dimensional DCT is reported to be more effective for residual signals suppression because of less correlation between these signals. In this report, we propose image coding method using edge adaptive one-dimensional blocks. And the effect of this method on objective image quality is described.
The generaly used noise filters are consisted of L and C, therefore their impedances are reactance. Then they may be series resonated with the connected circuits and lose filter action at the resonated frequencies. Other case, the filter connects in main lines, have many resonances. If the pulse current (such as burst noises) flows in it, the damped oscillation will be occur namely the disturbance noise. Here introduce the noise filter applied the principle of constant impedance, which construted from the bridged. T. Let the series arm is Z1, and the shunt arm is Z2, and the resistance arm are both R. If select their constants as R2=Z1•Z2 it's input impedance will be the constant R. Apply this principle to noise filter. As it's character must be low pass, than set Z1=L, and Z2=C. It's input impedance will be the constant R=√L/C for wide frequeney range (ex. 400kHz_??_30MHz), even if the condition connected any impedances between its onput terminals. The value of R should be selected same as the characteristic impedance of main lines. Constant impedance means resistance or loss element which don't resonate with any circuits. When the filter connected to main lines, the noise energy in it, will be faded away. Other case, the main lines with the filter will lose resonance action, then even if the pulse current flow in it, there are no damped oscillation or no noise. There are two kinds of noise mode, namely normal and common. Then the practical filters must be modefied to two types.
Much research has been done to apply neural networks (NNs) to pattern recognition and the results have proved that recognition methods by NNs are effective. Up to this point, we have proposed a structure reduction method for NNs and have applied the method to paper currency recognition. In the method, we have adopted slab values which are generated by random masks. Still more, we have tried to optimize these random masks using genetic algorithm. In this paper, in order to realize the neuro-paper currency recognition in the commercial products, we have developed a high-speed neuro-recognition board. This device can recognize paper currency firster than ten times compared with the conventional recognition machines. The device is constructed with digital signal processor (DSP) which is used in image processing widely. We denote its configuration and specification for paper currency recognition. Furthermore, we show its application to US dollars using this device on the prototype.
The demand for security is more and more increasing these days because of high crime rate. This paper proposes a new Indoor Observation System based on Environmental Sounds recognition (IOSES) for a security system. The IOSES is useful for various kinds of needs such as office security, home security, and so on. It uses two microphones in order to detect the direction of the sound sources. The IOSES operates as follows. First, it analyzes a background noise in order to reduce it. Spectrum subtraction method is employed for the noise reduction. The IOSES detects a sound to be recognized using some parameters even if in a noisy environment. The IOSES decides the detected sound using a recognition result by a neural network, the characteristics of the detected sound (intermittence), and the direction of the sound source. The neural network plays an important role in the decision. Neural networks are effective since they do not require modeling of sound to be recognized. After these processes, the IOSES judges the situation of the indoor environment. Computer simulations and experiments in a real environment are carried out. The results indicate the effectiveness of the proposed LOSES for a practical use.
The Traveling Salesman Problem (TSP) is well known as one of the combinatorial optimization problem. In 1988, Angeniol et al. first applied Kohonen's Self Organizing feature Map (SOM) to the TSP. They demonstrated that the practical solutions (sub-optimal tour) were obtained by their proposed method in sufficiently short time compared with conventional methods;e.g. Hopfield Network, Simulated Annealing, Genetic Algorithm (GA) and Chaos Neural Network. In this paper, we have improved Angeniol's method to reduce the calculating time by the following modifications of (1) optimization of node creation timing and (2) adding a momentum effect for the update factor. Using the best combination of the several parameters in the modified method, we confirm that the modified method can be take just one fourth time compared with original method to solve the squarely located 36 cities TSP. Finally, we illustrate that the 200 cities TSP can be solved about one minute using our method on the personal computer.
This paper presents a fuzzy estimation method for the class mixture proportion of the mixed pixel (mixel) on a remote sensing image. The training data were selected by the operator based on the subjectivity and its distribution was defined as a fuzzy set on a spectral space. It was also assumed that the spectral characteristics of the mixel was regarded as a linear function of the reflection levels of the pure pixels corresponding to the component classes. A fuzzy production rule for the estimation of the class mixture proportion on the mixel was defined in accordance with this assumption. The estimation of the class mixture proportion for the proposed method was conducted by the fuzzy simplification reasoning method. It was observed that the estimation accuracy of the fuzzy estimation method depended on the mesh interval. The mesh interval means the change percentage of the class mixture rate for the formation of the membership function. That is, the Total Root Mean Square error (TRMS) of the estimated value tended to decrease when reducing the mesh interval. The simulation results also indicated that the reasonable mesh interval (the change percentage of the class mixture rate) of the membership function might be about 0.05 (5%). It was also observed that the proposed method gave low TRMS to the simulation data which was produced by the random sampling of the training data of each class. Therefore, it was confirmed that the proposed fuzzy estimation method was an useful technique to estimate the class mixture proportion of the mixel on an actual remote sensing image.
A method by using neural networks is proposed for solving au inverse kinematics problem of redundant manipulators. When a neural network for the inverse kinematics problem is trained by supervised learning, the Jacobian matrix and the manipulability measure of a manipulator are required. Unfortunately, however, we encounter so often the difficulty that the values of these quantities are unknown. In such circumstances, we have to inevitably estimate them in an ad hoc manner. In this paper, a specific network technique is presented to estimate them skilfully. The procedure is based on the multi-layer perceptron which works as a forward model of the manipulator. Since, however, characteristics of our perceptron are expressed by an equation relating the input data to the output data, the Jacobian matrix of the perceptron is obtained readily and then the manipulability measure is calculated from the resulting matrix as well. With the help of these outcomes for the actual manipulator, the inverse kinematics learning can be carried out. The proposed method is applied to an inverse problem of the manipulator with three links. Through various computer simulation experiments, it is revealed that the Jacobian matrix and the manipulability measure can be estimated successfully by the perceptron which is originally designed in the present paper.
In this paper authors present the fuzzy algorithm of sensor fusion for discriminating material properties with touch sensors. The proposed algorithm is based on fuzzy theory, and there the notions of adaptation grade function, maximum similarity α and boundary value γ are introduced for expressing the categories of samples. For verifying the algorithm, numerical simulation is performed. It resulted in good classification for discrimination of materials such as iron, aluminium, copper, bronze, gum, wood, styrene foam and etc. The database used here is based on the experimental results with our proposed touch sensors. Our final target is to develop an artificial skin tactile sensing system.
A qualitative simulation system has been developed for understanding behavior of a structural model, which helps analysists grasp perspective. In decision making, it is necessary to understand the influence flow, namely how the model derives the behavior of the concerning node rather than the behavior of each node. This paper presents a mechanism of scenario generating for explaining the flow of influence to the concerning node. The mechanism consists of the following processes. 1) By tracing the structural model, all paths from the input node, which triggers the initial change of system, to the concerning node are derived. Using similarity of behavior, influence flow to the concerning node is selected out of all paths. 2) Important nodes are selected according to the type of nodes. 3) The scenario document is generated along the selected flow using the sentence templates and the dictionary.
Research into the brain's switching circuitry between its chaotic and non-chaotic activities is a topic of brain science. In this paper, the neuron's switching circuitry of its chaotic firing state is estimated using a neuron model. The model, which is carefully tuned up for the properties of emergence of chaos, was obtained from modifications of the McCulloch-Pitts and Nagumo-Sato models. On the modifications, two neuronal properties of the axon's refractoriness and soma's cable property were especially paid attention to. The former is crucial to cause many types of neuron's firing states, including the chaotic firing state. The other has relevance to switch the types. Computer simulation of the switching circuitry was performed under hypothesis that one kind of neuron receives steady stimulation inside the range to activate its chaotic firing. The results suggest that the neuron's chaotic firing state is stable under noisy stimulation, that synaptic transmission even of a few times triggers a bifurcation process to the periodic firing state of a higher firing-rate, and that the bifurcated periodic firing state lasts for a while stably. Such a circuitry may be not only a guide to find out brain's chaos-working circuitry but also a hint to invent a new kind of neural network.