Mamoru Miura

Two‐dimensional visual simulation of flames, smoke and the spread of fire

Since CG simulation of natural phenomena on the basis of their forms and motions has many applications, such as various landscape designs and special effects in films, it is very important and interesting to develop efficient techniques for their visual simulation. It is especially interesting to produce realistic images and animations of flames and smoke, on account of their complicated patterns of behaviour. Effective simulation methods for flames and smoke are expected to satisfy the following requirements: 1 The motion of flames or smoke produced in an interaction with obstacles can be simulated. 2 The motion of flames and smoke can be easily controlled according to scenarios. 3 The spread of fire can be simulated. Although several useful representation methods have been proposed so far, simulating flames and smoke still remains a challenging problem. In this paper, we first describe our basic two-dimensional visual simulation method based on particle-based simulation, but not based on exact physical simulation. Roughly speaking, our method assumes that the images of flames and smoke are basically obtained by visualizing turbulence, that is, the particles of flames and smoke play the role of tracers in the field of turbulence. Next, we present an improved method for simulating the spread of fire and the appearance of smoke produced in an incomplete combustion. Simultaneously, we show several examples of the simulation. Finally, we will touch slightly on further problems for extending the model to one which works in three-dimensional space.

Visual Simulation of Botanical Trees Based on Virtual Heliotropism and Dormancy Break

Realistic image synthesis of botanical trees has many applications. Since the generation of ‘tree skeletons’ having natural visual impressions is essential to realistic image synthesis, various methods of modelling skeletons, especially growth models, have been presented. However, no one has succeeded in simulating natural tree features which appear in a growth process, such as generation of a round tree crown, a weeping bough, an irregular branching pattern, or regeneration of a crown. This paper demonstrates, by showing several simulated examples, that a growth model having the abilities of heliotropism and dormancy break, which produces shapes of trees adapted to changes in the light environment, is effective in the CG simulation of realistic tree skeletons.

A growth model having the abilities of growth-regulations for simulating visual nature of botanical trees

Abstract Although the representation methods of botanical trees have been presented by several researchers in recent years, simulating natural tree features acquired in a growth process still remains a challenging problem. Since tree shape is significantly influenced by its growth environment, such as sunlight conditions and random “accidental” pruning of branches ( e.g. , caused by storm or a gardener), not only does no tree have a regular shape, but no two trees are identical, even if they are of the same species. In previous papers, we have shown an attractive fact that if we take into account the abilities of several growth regulations, such as heliotropism and dormancy break, we can easily simulate realistic irregular branching patterns. In this paper, we will present the improved growth model taking into account the following growth regulations: (a) withering, (b) heliotropism, (c) geotropism, and (d) apical dominance: (d-1) suppression to lateral shoots, (d-2) dormancy break, and (d-3) change in leadership. These growth regulations are implemented by employing an “imaginary plant hormone” for implementing the ability of the communication between all buds and branches of a tree. This means that any “central control unit” that keeps watch on the condition of the whole tree and issues commands for each bud and branch to control the growth is unnecessary for a tree. This point is one of the interesting features of our growth model.

New magnetic scaler used as universal arithmetic module for pulse-train signal processing

This paper describes a new magnetic decimal scaler as a universal arithmetic module (UAM) realized by expanding the function of the multi-level magnetic scaler. In order to implement pulse-train signal processing systems, three arithmetic operations (addition, multiplication and delay) are required. The proposed UAM has adder, multiplier and delay functions, all in one. So the pulse-train signal processing systems can be implemented by using only UAMs as a basic building block. This results in high reliability and simplicity with respect to the operation and construction of the systems, as compared with the case using commercially available binary-circuits.

A neural network structure based on pulse-train arithmetic

A Hopfield neural network with a simple structure based on a pulse-train signal arithmetic has been developed. It is both analog and digital in nature and involves hybrid system concepts. It is implemented by using only two types of principle elements, a neuron module and a synapse module. The neural network proposed performs pulse-train signal processing directly. The advantages of this design are a simple operating mechanism, homogeneous implementation and expandability.<<ETX>>

A magnetic neural network utilizing universal arithmetic modules for pulse-train signal processing

The authors consider two types of magnetic neural network architecture based on pulse-train signal processing with high reliability. One is realized by neuron units and synapse units, both of which use only a universal arithmetic module (UAM) having adder, multiplier, and delay (memory) functions all in one unit. This architecture thus results in an extremely homogeneous implementation. The other is realized by dividing the function of the UAM into a neuron unit and a synapse unit. The neuron unit provides functions of nonlinear adder and memory by using the properties of the magnetic core such as pulse storage, integration, and nonlinearity.<<ETX>>

Design and VLSI evaluation of a high-speed cellular array divider with a selection function

In recent years, a very high-speed divider is required in real-time applications of digital signal processing and robot control and so on. In this paper, a high-speed cellular array divider with selection function is proposed, which is based on the nonrestoring algorithm and can deal with both fixed-point and negative operands in twos complement form. This divider uses new techniques which can generate in parallel both a quotient bit of one row and a partial remainder and CLA bit of its next row. Moreover the delay time of the proposed divider is calculated in terms of a delay of one unit such as NAND gate. Finally, by using PARTHENON, a CAD (Computer Aided Design) system for VLSI, this divider is designed and evaluated. As a result, elimination of delay time for even rows becomes possible. Thus, the delay time can decrease by approximately one-half of the high-speed divider proposed by Cappa and Hamacher which uses the most general high-speed techniques of carry-save and CLA.

A simple neural network structure with good expansibility based on pulse-train arithmetic

The authors discuss the simplification, expansibility, and modularity of a neural network architecture based on pulse-train arithmetic. The simplified architecture is implemented by connecting basic modules in parallel with a single bus line of a control processor. The basic module is composed of a neuron unit and synapse unit. Thus, the expansion of the network can be easily realized by increasing the number of basic modules. The basic operations of this network architecture have been confirmed by computer simulation of the traveling-salesman problem.<<ETX>>

Implementation of simple neural network system based on pulse-train arithmetic

In this paper, implementation and operation of a simplified Hopfield type neural network system based on pulse-train arithmetic are presented. The narrow module proposed here as a basic building block is composed of a pair of neuron unit and synapse unit, which performs time-division multiple operation. The Hopfield type neural network is implemented by connecting the neuron modules in parallel with a single bus line, so that this structure can considerably reduce a mass of hardware. In particular, the network can easily be extended to more practical systems from the simple structure.<<ETX>>