A novel algorithm based on Kalman filtering is developed for the learning of a layered neural network. The problem of adjusting the weight can be regarded as that of estimating a signal state vector of a linear process. The proposed algorithm, though computationally complex, has an adaptively varying learning rate, while the back-propagation algorithm has constant learning rate. Some experiments conducted for XOR and auto-associative image compression problems have shown that the proposed learning algorithm usually converges in a few iterations and the error is comparable to that of the well-known back-propagation algorithm.

Introducing a new analytical method of nonuniform transmission line, this paper shows equivalent transformations between a circuit consisting of a cascade connection of a lumped type E section and a uniform transmission line and one consisting of a cascade connection of a class of nonuniform transmission line and a lumped type E section. Characteristic impedance distribution of these nonuniform transmission lines are expressed as hyperbolic and trigonometric functions. By using the equivalent transformation, it becomes possible to obtain exact network functions of a class of nonuniform transmission lines without solving the telegraph equation.

This letter proposes a new framework for ASIP (Application Specific Integrated Processor) development. The system is called IDEAS (Integrated Design Environment for Application Specific Integrated Processor). IDEAS accepts a set of application programs and its expected data as input, and profiles these programs both statically and dynamically. According to the profiled results, the system decides the architecture of ASIP, and synthesizes the CPU core design of the ASIP, and generates the software development tools for the ASIP such as compiler and simulator.

LDD-structure GaAs-MESFETs with WSiN bilayer gate are fabricated adopting neutral buried p-layers formed by 50-keV Be-implantation. fT of 108 GHz and fmax of over 130 GHz are obtained on 0.2-µm gate length. A direct-coupled amplifier IC with bandwidth of 10 GHz are fabricated using 0.4 µm GaAs-MESFETs and achieves a high gain of 20 dB with a minimum NF of 3.2 dB with a power consumption of 365 mW. Moreover, a bandwidth of 20 GHz is predicted for the amplifier IC using 0.2-µm GaAs-MESFETs.

In this paper, we discuss autocorrelation properties of a truncated m-sequence which is used in communication and ranging systems based on a Direct-Sequence Spread Spectrum technique such as Tracking and Data Relay Satellite System (TDRSS) of NASA. A truncated m-sequence is obtained by removing a short part or sequential chips from an original m-sequence. We derive methods to calculate periodic and partial autocorrelation functions of a truncated m-sequence with small complexity. Some constraints for truncated m-sequences which originates in system regulation of TDRSS and its related systems are described. In such systems, partial correlation of a truncated m-sequence is important to achieve fast and stable acquisition. We propose some criteria to select a proper set of truncated m-sequence, and derive the set by using the method to simply calculate correlation.

This paper proposes and analyzes experimentally an SSMA (Spread Spectrum Multiple Access) signal transmission over a high speed QPSK (Quadrature Phase Shift Keying) modulated signal to achieve higher transmission efficiency per transponder and to facilitate a lower power transmitter for SSMA signal transmission. The employment of high-coding-gain forward error correction for SSMA-QPSK signals makes it possible to transmit SSMA-QPSK signals over a non-linearly amplified QPSK signal transmission channel. Experimental results show that under the condition of a 20dB less transmission power assignment to SSMA-QPSK signals than QPSK signals, the QPSK signals achieve only 0.5dB Eb/No degradation (at Pe=110-4) by employing coding-rate seven-eighth FEC, and the SSMA-QPSK signals achieve about 7dB Eb/No degradation (at Pe=110-4) by employing coding-rate one-half FEC. The satellite link budget shows that even if the SSMA-QPSK signal transmission generates about 7dB Eb/No degradation, it requires 20 dB less transmission power and it still has an extra margin of 8.5dB compared with the high speed QPSK signals (total relative gain of 28.5dB). Thus, the proposed system makes it possible to realize one-way digital video signal transmission in QPSK mode and both-way digital voice signal transmission over one transponder for business video communication systems.

This letter describes system architecture of a measurement environment called GINGER that helps programmer productivity by automatically collecting and analyzing data about software development activities. Based on the result of analysis, the environment gives programmers feedback so that they can recognize and improve their activities. The first prototype system of GINGER is currently implemented in a UNIX environment. We expect that the proposed environment will improve the overall productivity of programmers and the quality of resulting products.

A new impedance matched film carrier is developed to package uniplanar MMICs. The carrier has an insulating polyimide film on which RF, DC bias, and ground conductive areas are formed. The areas extend into bonding windows, which are etched in the polyimide film, and the extended portions form inner and outer leads. The interconnection of the inner and outer leads to MMIC-electrode pads is highly reliable because of the gold-plated bumps formed at the distal ends of the inner and outer leads. This carrier has an insertion loss of less than 0.2 dB/mm without resonance over the frequency range from DC to 30 GHz. The electrical performance of the GaAs MMIC module with this carrier is almost equal to that of the MMIC measured directly on-wafer.

We introduce a novel neural network with a trigonometric interconnection called the T-Model neural network in this paper. A VLSI implementation of the T-Model neural network based on CMOS current-mode circuits is also presented. The circuit is completely compatible with standard VLSI technology. A set of neuron-type elements of CMOS current-mode circuits is described and a very large scale neural network is also synthesized. The feasibility and the operation principle of the synthesis of the T-Model neural network using CMOS current-mode circuits are demonstrated and confirmed by experimental results of fabricated CMOS VLSI neural chips.

The mappings from multidimension to one-dimension, or the inverse mappings, are theoretically discussed as space-filling curves, i.e., Peano curves. The Peano scan is an application of a Peano curve to the scanning of images, and it is used for analyzing, clustering, or compressing an image, and for limiting the number of the colors used in an image. The horizontal and vertical sizes of the scanned array, however, must be a power of two. To avoid such a case, we generalize the Peano scan for scanning an arbitrarily-sized array, whose horizontal and vertical sizes are possible to be different. First we propose a binary scan which is made of binarily recursive divisions of an image. As the Peano scan is characterized by the statistical property of Brownian motion, further we describe that binary scan can be also optimized to have such statistical property.

Almost half a century has passed since the first practical use of microwaves in 1940's. During this period, extensive research and development have lead us to steady progress in terms of usable frequencies, fabrication capability of active and passive devices, transmission lines, and device integration density. An overview of this half-century progress in microwaves is presented, and topics of common interest are discussed from the following points of view: (1) Exploiting frontier frequencies from VHF to millimeter-wave and optical frequencies, (2) Transition from vacuum tubes era to solid-state devices era, (3) Transition from waveguides and discrete circuits era to planar integrated circuits era, (4) Wide-ranging applications of microwaves.

Two-dimensional simulations of small-signal parameters (such as transconductance, gate capacitance and cutoff frequency) of GaAs MESFETs are performed in which impurity compensation by deep levels in the semi-insulating substrate is considered. It is shown that these are strongly affected by impurity densities in the substrate. For higher acceptor densities in the substrate, both transconductance and cutoff frequency become higher, because the substrate current is reduced. For low acceptor densities in the substrate, the gate capacitance takes a relatively large value even if the gate voltage is deeply negative, because the channel extends into the substrate and electrons there contribute to the capacitance. It is concluded that to utilize high-frequency performance of GaAs MESFETs, the acceptor densities in the semi-insulating substrate should be made high.

The performance of a digital matched filter receiver with multi-bit quantized correlators for Direct Sequence Spread Spectrum (DSSS) mobile satellite applications is evaluated with experiments. A Coherent Matched Filter (CMF) technique is used to accomplish fast code acquisition, automatic frequency control and coherent detection of data with a single matched filter circuit to satisfy the requirements for the mobile satellite applications. Loop-back test and field test results in suburban areas with the ETS-V satellite are presented, in which the use of multi-bit quantized correlators improve the initial acquisition and the bit error rate performances. The dynamic performance of the CMF receiver is confirmed to be satisfactory for the mobile satellite fading channels.

We have fabricated a 12 GHz-band monolithic microwave integrated circuit amplifier of which the noise level has been reduced down to 1.18-1.35 dB in the frequency range from 11.7 GHz to 12.7 GHz. The excellent characteristics come from the use of low-gate-resistance, low-equivalent-noise-resistance HEMT's which were obtained by using T-getes, and a hi-lo doping profile in an AlGaAs layer.

This paper shows a Direct-Sequence Spread-Spectrum (SS-DS) demodulator using block signal processing. One of the difficulties in applying SS-DS techniques to the packet radio network is that each packet needs a long initial-acquisition time for despreading. The acquisition time causes the large degradation of the data transmission efficiency. Our proposed SS-DS demodulator uses the block signal processing, unlike the conventional SS-DS demodulators using real time signal processing. Received signal demodulated quasi-coherently is once stored in memory, and after extracting matched-pulse timing and estimating carrier offset, the signal is demodulated. Incoming data, therefore, are all demodulated without being lost by the initial-acquisition time, and our proposed SS-DS demodulator can provide the higher data transmission efficiency.

This paper reviews various switch architectures for Asynchronous Transfer Mode, which have been proposed and developed so far in Japan. The switch fabrics can be classified, owing to the arrangement of switch matrices and buffer memories, into four categories, input buffer switch, output buffer switch, shared buffer switch and crosspoint buffer switch. Those switches have their own advantages and disadvantages, which require additional effort to implement the switches for the practical network. This paper introduces examples of each category switch fabric and additional technical modifications to make it practical. Other general issues to construct ATM switch fabrics, such as non-blocking characteristics and path assignment within a multi-stage switch network, are also addressed. Furthermore, future directions in the ATM switch fabric is discussed.

ATM's bandwidth flexibility and high switching speed advantages have made it the preeminent solution to the B-ISDN. However, networks based on ATM suffer from phenomena not encountered in the STM network, e.g., cell loss, and cell jitter. For ATM to support varied media, problems arising from these phenomena must be resolved. This paper discusses techniques which enable ATM to support CBR, VBR voice, VBR video, and connectionless services. We determine the problems to be resolved for each service and discuss solutions. We focus on dejittering and clock recovery for CBR service, on talk-spurt synchronization for VBR voice, and on video frame synchronization and lost-cell compensation for VBR video service. We show experimental results for a prototype HDTV codec and terminal adapter. For connectionless service, we show that connectionless service should be supplied by the B-ISDN itself using message handlers which support message routing and addressing, i.e., connectionless service functions (CLSF), as described by the CCITT. We propose a network topology of message handlers, which reduces the routing database and avoids transit message handler overload. We show an example of system configuration for connectionless service.

If π is a monotone property on graphs (that is, π is attainable by adding edges to a given graph), then the graph augmentation problem with respect to π is defined by: "Given a graph G=(V, E) and nonnegative weights w(u, v) for all pairs {u, v} VV (uv), find an edge set Eof minimum total weight such that the graph G=(V, EE) satisfies π, where we assume that w(u, v)=0 for every edge (u, v) E". The subject of the paper is to give an overview of the graph augmentation problems where π is concerned with vertex-or edge-connectivity of graphs. Also presented are basic ideas used in solving this problem.

For a given 2-edge-connected graph G and a spanning tree T of G, the graph augmentation problem 2ECA (T,G) is to find a minimum edge set AE (G) such that T A is 2-edge-connected. This note proves that 2ECA (T, G) is solvable in polynomial time if G is series-parallel.