Kazuo Shirakawa

An approach to determining an equivalent circuit for HEMTs

A simple way to determine a small-signal equivalent circuit of High Electron Mobility Transistors (HEMTs) is proposed. Intrinsic elements determined by a conventional analytical parameter transformation technique are described as functions of extrinsic elements. Assuming that the equivalent circuit composed of lumped elements is valid over the whole frequency range of the measurements, the extrinsic elements are iteratively determined using the variance of the intrinsic elements as an optimization criterion. Measurements of S-parameters up to 62.5 GHz at more than 100 different bias points confirmed that the HEMT equivalent circuit is consistent for all bias points. >

Structural determination of multilayered large-signal neural-network HEMT model

This paper reports on the structure of a large-signal neural-network (NN) high electron-mobility transistor (HEMT) model as determined by a pruning technique and a genetic algorithm. The bias-dependent intrinsic elements of an HEMTs equivalent circuit are described by a generalized multilayered NN whose inputs are the gate-to-source bias (V/sub gs/) and the drain-to-source bias (V/sub ds/). Using C/sub gs/ data as an example, we began by experimentally examining some of the features of the multilayered NN model to obtain rules-of-thumb on choosing training parameters and other information for succeeding studies. We then developed and studied a novel pruning technique to optimize the C/sub gs/ NN model. Excessively large NN configurations can be reduced to an appropriate size by means of a weight decay, which is based on the analysis of a synaptic connections activity. Finally, we employed a genetic algorithm for the same purpose. By representing the configuration of a standard multilayered NN as a chromosome, the optimum configuration of a C/sub gs/ model was obtained through a simulated evolution process. For this approach, the configuration of an NN that simultaneously represents seven intrinsic elements (C/sub gs/,R/sub i/,...,C/sub ds/) of an equivalent circuit was also shown for comparison to previous work. We successfully obtained simplified NN models using both approaches. The advantages and disadvantages of these two approaches are discussed in the conclusion. To our knowledge, this is the first report to clarify the general process of building an NN device model.

A 15/60 GHz one-stage MMIC frequency quadrupler

We have developed a 15/60 GHz one-stage MMIC frequency quadrupler using a 0.25-/spl mu/m AlGaAs/GaAs HEMT. The HEMT was characterized by our empirical large-signal model, in which charge conservation and dispersion are taken into consideration. We included this model in a commercially-available harmonic balance circuit simulator, and designed the one-stage quadrupler. The fabricated MMIC quadrupler has a conversion gain of -5 dBm with -5 dBm of output power for a 0 dBm input signal.

New method of analyzing BER performance of GFSK with postdetection filtering

This paper describes the theoretical performance of a frequency-hopping system using Gaussian-filtered FSK (GFSK), which is a promising candidate for 2.4-GHz wireless local area networks (WLAN). We have improved previous methods to calculate the bit-error rate (BER) performance of the GFSK system by considering the effect of a postdetection filter in the first-order approximation. The improved method is applicable when the single-sided bandwidth of the postdetection filter is wider than the bandwidth containing 99% of the GFSK signal energy. It is used to calculate the BER performance as well as the dependence of the BER on the level of interchannel interference. The calculated performance of the GFSK system suggests that the postdetection filter is very effective in reducing interchannel interference. The feasibility of a GFSK system with a 1-Mbd symbol rate and 0.32 modulation index is confirmed by the combination of a predetection filter with a 0.85-MHz bandwidth and a postdetection filter with a 0.8-MHz single-sided bandwidth.

A new empirical large-signal HEMT model

We propose an empirical large-signal model of high electron mobility transistors (HEMTs). The bias-dependent data of small-signal equivalent circuit elements are obtained from S-parameters measured at various bias settings. And C/sub gs/, C/sub gd/, g/sub m/, and g/sub ds/, are described as functions of V/sub gs/ and V/sub ds/. We included our large-signal model in a commercially available circuit simulator as a user-defined model and designed a 30/60-GHz frequency doubler. The fabricated doublers characteristics agreed well with the design calculations.

A 30-to-60-GHz Monolithic HEMT Frequency Doubler

Using 0.3-¿m AlGaAs/GaAs HEMT technology, we developed a 30-to-60-GHz monolithic frequency doubler which has a conversion loss of 5 dB and a fundamental signal suppression of 20 dB for a 0-dBm 30-GHz input signal. The doublers performance is superior to those using similarly scaled AlGaAs/GaAs HEMT. For the design, we used a nonlinear circuit simulator based on the harmonic balance method, an empirical nonlinear HEMT model, and an electromagnetic field simulator. The calculated and measured results agreed well. We developed this in order to build a 60-GHz AlGaAs/GaAs HEMT-based single-chip monolithic transceiver.

PRISM: An In-Vehicle CPU-Oriented Novel Azimuth Estimation Technique for Electronic-Scan 76-GHz Adaptive-Cruise-Control Radar System

In this paper, we have developed a novel azimuth- estimation technique for 76-GHz adaptive-cruise-control radar systems. To resolve the rank deficiency of the signal covariance matrix and to decrease the influence of dominant noise components, we used an improved SS technique. To balance computational cost with azimuth resolution for in-vehicle implementation, we introduced a projection matrix kernel, which formally is similar to that of beamforming (BF) and semantically similar to that of the propagator method. Our technique is called the PRopagator method based on an improved spatial-smoothing matrix (PRISM). Incorrectly estimating the number of targets causes the conventional BF, subspace-based (SB), or maximum- likelihood methods to produce many false spectrum peaks on the relative azimuth-to-distance plane. We omitted the process of estimating the number of targets, which requires high computational power, because our projection matrix kernel strongly suppresses these false peaks. We conducted numerical experiments using an electronic-scan radar system to examine the performance of PRISM. Compared to conventional SB methods, PRISM requires SNRs that are 2-3 dB higher but has competitive azimuth resolution at a centesimal fraction of the computational cost.

Genetic Determination of Large-Signal HEMT Model

This paper reports on a general approach to build a large-signal, neural network HEMT model using a genetic algorithm. By representing the configuration of a neural network model as the chromosome of a virtual creature, we looked for an optimum network configuration by simulating the evolution of a group of these virtual creatures (a population). We successfully designed neural networks representing bias-dependent intrinsic elements of a HEMTs equivalent circuit. We also verified the reliability of this technique by searching for the optimum model from different initial conditions.