A responsive network architecture is essential in future open distributed systems. In this paper, a framework that provides the foundations for a responsive network architecture for an internetworking environment is proposed. It is called the Virtually Separated Link (VSL) model. By incorporating this framework, communication of both data and control information can be completed in bounded time. Consequently, a protocol can initiate a recovery mechanism in bounded time, or allow an application to do the same. Its functionalities augment existing resource reservation protocols that support multimedia communication. An overview of a real-time network protocol that is based on this framework is also presented.

Space-time tradeoff is a very fundamental issue to design a fault-tolerant real-time (called responsive) system. Routing a message in large computer networks is efficient when each node knows the full topology of the whole network. However, in the hierarchical routing schemes, no node knows the full topology. In this paper, a tradeoff between an optimality of path length (message delay: time) and the amount of topology information (routing table size: space) in each node is presented. The schemes to be analyzed include K-scheme (by Kamoun and Kleinrock), G-scheme (by Garcia and Shacham), and I-scheme (by authors). The analysis is performed by simulation experiments. The results show that, with respect to average path length, I-scheme is superior to both K-scheme and G-scheme, and that K-scheme is better than G-scheme. Additionally, an average path length in I-scheme is about 20% longer than the optimal path length. On the other hand, for the routing table size, three schemes are ranked in reverse direction. However, with respect to the order of size of routing table, the schemes have the same complexity O (log n) where n is the number of nodes in a network.

The design of a high speed self-routing network switch module is described. Clock distribution and timing design to achieve high-speed operation are considered. A 88-port self-routing Benes network switch prototype on multi-chip module is fabricated using 44-port space division switch LSIs. The switch module achieves a maximum measured clock frequency of 750MHz under switching operation. Resultant total throughput of the switch module is 12Gbit/s.

This paper focuses on two areas in our effort to synthesize speech from neuromotor input using neural network models that effect transforms between cognitive intentions to speak, their physiological effects on vocal tract structures, and subsequent realization as acoustic signals. The first area concerns the biomechanical transform between motor commands to muscles and the ensuing articulator behavior. Using physiological data of muscle EMG (electromyography) and articulator movements during natural English speech utterances, three articulator-specific neural networks learn the forward dynamics that relate motor commands to the muscles and motion of the tongue, jaw, ant lips. Compared to a fully-connected network, mapping muscle EMG and motion for all three sets of articulators at once, this modular approach has improved performance by reducing network complexity and has eliminated some of the confounding influence of functional coupling among articulators. Network independence has also allowed us to identify and assess the effects of technical and empirical limitations on an articulator-by-articulator basis. This is particularly important for modeling the tongue whose complex structure is very difficult to examine empirically. The second area of progress concerns the transform between articulator motion and the speech acoustics. From the articulatory movement trajectories, a second neural network generates PARCOR (partial correlation) coefficients which are then used to synthesize the speech acoustics. In the current implementation, articulator velocities have been added as the inputs to the network. As a result, the model now follows the fast changes of the coefficients for consonants generated by relatively slow articulatory movements during natural English utterances. Although much work still needs to be done, progress in these areas brings us closer to our goal of emulating speech production processes computationally.

In our preceding paper, I-ISMA (Idle Signal Multiple Access for Integrated services), a combination of ISMA and time reservation technique, was proposed to transmit an integrated voice and data traffic in third generation wireless communication networks. There, the channel capacity of I-ISMA was evaluated by the static analysis. To fully estimate performance of contention-based channel access protocols, however, we also need dynamic analysis to evaluate stability, delay, etc. Particularly, in systems concerning real-time voice transmission, delay is one of the most important performance measures. A six-mode model to describe an I-ISMA system is set up. With some assumptions for simplification, the dynamic behavior of the system is approximated by a Markov process so that the EPA (Equilibrium Point Analysis), a fluid approximation method, can be applied to the analysis. Then, numerical and simulation results are obtained for some examples. By means of the same analysis method and under the same conditions, the performance of PRMA is evaluated and compared briefly with that of I-ISMA.

This paper describes a planning method for ATM networks. The method is based on evaluation of two design items, VC routing and VP routing, as well as on consideration of VPI constraints. In the evaluation, VC routing is compared with VP routing in separate case studies undertaken from the point of view of various parameters such as traffic volume, cost function and network scale. The results suggest the vertical relationship between VC and VP levels in optimally designed ATM networks. VC and VP network levels are then studied separately, and design methods are proposed for individual levels. In addition a perturbation method is proposed for the VC and VP routing use, whose optimum is varied as a function of the parameters described above. Evaluation results show the proposed perturbation method provides cost-effective networks.

We present an extension of the previously proposed 3-layer feedforward network called a cascaded network. Cascaded networks are trained to realize category classification employing binary input vectors and locally represented binary target output vectors. To realize a nonlinearly separable task the extended cascaded network presented here is consreucted by introducing high order cross producted inputs at the input layer. In the construction of the cascaded network, two 2-layer networks are first trained independently by delta rule and then cascaded. After cascading, the intermediate layer can be understood as a hidden layer which is trained to attain preassigned saturated outputs in response to the training set. In a cascaded network trained to categorize binary image patterns, saturation of hidden outputs reduces the effect of corrupted disturbances presented in the input. We demonstrated that the extended cascaded network was able to realize a nonlinearly separable task and yielded better generalization ability than the Backpropagation network.

Since, in a VLSI circuit, the number of transistors and the clock frequency are constantly increasing, it is important to analyze the voltage drop and current density on a full chip's power networks. We propose a new hierarchical power analysis system named XPOWER. A new reduction algorithm for the resistance and current source network is used in this system. The algorithm utilizes the design hierarchy in nature and is independent of network topology. Networks at each level are reduced into small and equivalent networks, and this reduction is performed recursively from the bottom levels of the design hierarchy. At each step of the reduction, the network under consideration consists of two kinds of objects: (1) reduced child networks, and (2) the interconnection between child networks. After all networks have been reduced, circuit equationa are solved recursively from the top. This allows to decrease the size of the matrix to be solved and to reduce the execution time. Experimental results show that the factor of reduction in matrix size is from 1/10 to 1/40 and execution is six times faster than with flat analysis. The power networks of a 16 bit digital signal processor was analyzed within 15 minutes using XPOWER.

High speed simulation of neural networks can be achieved through parallel implementations capable of exploiting their massive inherent parallelism. In this paper, we show how this inherent parallelism can be effectively exploited on parallel data-driven systems. By using these systems, the asynchronous parallelism of neural networks can be naturally specified by the functional data-driven programs, and maximally exploited by pipelined and scalable data-driven processors. We shall demonstrate the suitability of data-driven systems for the parallel simulation of neural networks through a parallel implementation of the widely used back propagation networks. The implementation is based on the exploitation of the network and training set parallelisms inherent in these networks, and is evaluated using an image data compression network.

A model for a large network with an unidirectional linear respone (ULR) is proposed in this letter. This deterministic system has powerful computing properties in very close correspondence with earlier stochastic model based on McCulloch-Pitts neurons and graded neuron model based on sigmoid input-output relation. The exclusive OR problems and other digital computation properties of the earlier models also are present in the ULR model. Furthermore, many analog and continuous signal processing can also be performed using the simple ULR neural network. Several examples of the ULR neural networks for analog and continuous signal processing are presented and show extemely promising results in terms of performance, density and potential for analog and continuous signal processing. An algorithm for the ULR neural network is also developed and used to train the ULR network for many digital and analog as well as continuous problems successfully.

In this paper, a definitce relation between the TSP's optimal solution and the attracting region in the parameters space of TSP's energy function is discovered. An many attracting region relating to the global optimal solution for TSP is founded. Then a neural network algorithm with the optimized parameters by using Orthogonal Array Table Method is proposed and used to solve the Travelling Salesman Problem (TSP) for 30, 31 and 300 cities and Map-coloring Problem (MCP). These results are very satisfactory.

Corresponding to the development of B-ISDN, integrated services for data, voice, etc. are imperatively required for the so called third generation wireless communication networks. In this paper, I-ISMA (Idle Signal Multiple Access for Integrated services) is proposed to transmit integrated voice and data traffic from dispersed terminals to a base station. In the system, data packets and the first packets of talkspurts of conversational speeches are transmitted using ISMA protocol over a shared channel while subsequent packets of talkspurts are sent with time reservation technique. The channel capacity of I-ISMA is evaluated and compared with that of PRMA. The region in which I-ISMA has larger capacity than PRMA is figured out. Generally speaking, I-ISMA has larger capacity than PRMA when the duration for transmitting and detecting an idle signal is not too long and the channel is not too congested by the reserved voice transmissions. When we concern real time voice transmission, delay is one of the most important performance measures. Only is a qualitative discussion on delay performance given here. The quantitative evaluation is obtained by the dynamic analysis in our succeeding paper.

A quantum interconnection scheme by controlling the Coulomb interaction between ballistic electrons is proposed in which 2DEG (2 dimensional electron gas) plays the role of an interconnection medium. This concept brings up new possibilities for the interconnection approach in various fields such as parallel processing, telecommunications switching, and quantum functional devices. Cross-over interconnection, address collision, and address selection in a quantum information network system were analyzed as the first step. The obtained results have shown that the interconnection probability can be controlled by the velocity and timing of the ballistic electron emission from the emitter electrode. The proposed interconnection scheme is expected to open up a new field of quantum effect integrated circuits in the 21st century.

Markov chain models are used to derive the average stationary throughput and delay performance of Idle-Signal Casting Multiple Access (ICMA), with and without Failure Detection (/FD), protocols which are suitable for use in mobile packet radio local area networks, where propagation impairments are prevalent. The models consider the effects of pure Rayleigh fading on channel access and packet transmission. Numerical results, validated by computer simulations, are obtained that enable a quantitative study of the performance of the protocols. It is found that the performance of the ICMA/FD protocol is affected more significantly by fading on the base-to-mobile channel than is the performance of the ICMA protocol. In addition, performance improves with larger packet sizes eventhough such packets are more vulnerable to failure due to fading.

An attempt to apply neural networks to the acoustic diagnosis for the reciprocating compressor is described. The proposed neural network, Hybrid Neural Network (HNN), is composed of two multi-layered neural networks, an Acoustic Feature Extraction Network (AFEN) and a Fault Discrimination Network (FDN). The AFEN has multi-layers and the number of units in the middle hidden layer is smaller than the others. The input patterns of the AFEN are the logarithmic power spectra. In the AFEN, the error back propagation method is applied as the learning algorithm and the target patterns for the output layer are the same as the input patterns. After the learning, the hidden layer acquires the compressed input information. The architecture of the AFEN appropriate for the acoustic diagnosis is examined. This includes the determination of the form of the activation function in the output layer, the number of hidden layers and the numbers of units in the hidden layers. The FDN is composed of three layers and the learning algorithm is the same as the AFEN. The appropriate number of units in the hidden layer of the FDN is examined. The input patterns of the FDN are fed from the output of the hidden layer in the learned AFEN. The task of the HNN is to discriminate the types of faults in the compressor's two elements, the valve plate and the valve spring. The performance of the FDN are compared between the different inputs; the output of the hidden layer in the AFEN, the conventional cepstral coefficients and the filterbank's outputs. Furthermore, the FDN itself is compared to the conventional pattern recognition technique based on the feature vector distance, the Euclid distance measure, where the input is taken from the AFEN. The obtained results show that the discrimination accuracy with the HNN is better than that with the other combination of the discrimination method and its input. The output criteria of network for practical use is also discussed. The discrimination accuracy with this criteria is 85.4% and there is no case which mistakes the fault condition for the normal condition. These results suggest that the proposed decision network is effective for the acoustic diagnosis.

We applied an artificial neural network approach identify possible tumors into benign and malignant ones in mammograms. A sequential-dependence technique, which calculates the degree of redundancy or patterning in a sequence, was employed to extract image features from mammographic images. The extracted vectors were then used as input to the network. Our preliminary results show that the neural network can correctly classify benign and malignant tumors at an average rate of 85%. This accuracy rate indicates that the neural network approach with the proposed feature-extraction technique has potential utility in the computer-aided diagnosis of breast cancer.

This paper describes a neural network scheduler for scheduling independent and nonpreemptable tasks with deadlines and resource requirements in critical real-time applications, in which a schedule is to be obtained within a short time span. The proposed neural network scheduler is an integrate model of two Hopfield-Tank neural network medels. To cope with deadlines, a heuristic policy which is modified from the earliest deadling policy is embodied into the proposed model. Computer simulations show that the proposed neural network scheduler has a promising performance, with regard to the probability of generating a feasible schedule, compared with a scheduler that executes a conventional algorithm performing the earliest deadline policy.

A network-topology-independent static task allocation strategy has been designed and implemented for massively parallel computers. For mapping a task graph to a processor graph, this strategy evaluates several functions that represent some intuitively feasible properties or the graphs. They include the connectivity with the allocated nodes, distance from the median of a graph, connectivity with candidate nodes, and the number of candidate nodes within a distance. Several greedy strategies are defined to guide the mapping process, utilizing the indicated function values. An allocation system has been designed and implemented based on the allocation strategy. In experiments we have defined about 1000 nodes in task graphs with regular and irregular topologies, and the same order of processors with mesh, tree, and hypercube topologies. The results are summarized as follows. 1) The system can yield 4.0 times better total communication costs than an arbitrary allocation. 2) It is difficult to select a single strategy capable of providing the best solutions for a wide range of task-processor combinations. 3) Comparison with hypercube-topology-dependent research indicates that our topology-independent allocator produces better results than the dependent ones. 4) The order of computaion time of the allocator is experimentally proved to be O (n2) where n represents the number of tasks.

The effectiveness of error-control coding in a frequency-hop radio system can be increased greatly by the use of side information that is developed in the radio receiver. The transmission of test symbols provides a simple method for the derivation of side information in a slow-frequency-hop receiver. Requirements on the reliability of the side information are presented, and their implications in determining the necessary number of test symbols are described. Other methods for developing side information are reviewed briefly, and applications of side information to routing protocols for frequency-hop packet radio networks are discussed.

Concatenated coding techniques are applied to slow frequency-hop packet radio communications for channels with partial-band interference. Binary orthogonal signaling (e.g., binary FSK) is employed with noncoherent demodulation. The outer codes are Reed-Solomon codes and the inner codes are convolutional codes. Two concatenated coding schemes are compared. The first employs an interleaver between the outer Reed-Solomon code and the inner convolutional code. The second scheme employs an additional interleaver following the convolutional code. Comparisons are made between the performance of these concatenated coding schemes and the performance of Reed-Solomon codes alone.