Wednesday.html

1996 Midwest Symposium on Circuits and Systems

August 18-21, 1996

Wednesday August 21

7:30 am - 12:20 Registration Desk Open - Scheman First Floor Lobby
7:30 - 8:30 am Continental Breakfast - Scheman First Floor Lobby
8:30 - 10:00 CONCURRENT SESSIONS

WA1.1 VLSI Hardware for FIR Filters - Chin-Long Wey, Chairperson - Scheman Room 167

A Programmable FIR Filter for TV Ghost Cancellation
Sammy Pao, Kei-Yong Khoo, Alan N. Willson Jr., University of California, Los Angeles, California, USA
A compact 64-tap programmable FIR filter suitable for TV ghost cancellation has been designed, fabricated and tested. The design uses carry-save add-shift (CSAS) multipliers to achieve area efficiency and an internally generated self-timed clock to achieve timing efficiency. The prototype chip is implemented in a die area of 12.6 mm^2 using a 0.8-um CMOS process and can operateup to 18 MHz with a 5-V supply.

A Highly-Modular Pipelined VLSI Architecture for 2-D FIR Digital Filters
Cheng-Teh Hsieh, Seung P. Kim, Polytechnic University, Brooklyn, New York, USA
A highly-modular pipelined VLSI architecture for 2-D finite impulse response (FIR) digital filters is presented. This approach decomposes the 2-D discrete convolutions into pipelined summations of parallel 1-D discrete convolutions, so that modular structure is obtained. The advantages of this architecture are: (i) regular structure with high modularity, (ii) fully pipelined operation with maximum utilization of the resources, (iii) the real-time processing with low system latency.

VLSI Voltage Mode FIR Analog Adaptive Filters
Alejandro Diaz-Sanchez, Jaime Ramirez-Angulo, New Mexico State Univesity, Las Cruces, New Mexico, USA
VLSI implementation of voltage mode transversal adaptive filter is described. The proposed approach allows operation at high frequencies. A bandpass adaptive filter with central frequency at 10 mega hertz and quality factor of ten, demonstrates the feasibility of the proposed approach.

Switched-Capacitor Finite Impulse Response Interpolators without the Input Sample-and -Hold Filtering Effect
U Seng Pan, R.P. Martins, University of Macau, Macau, Portugal; J.E. Franca, Instituto Superior Tecnico, Lisboa, Portugal
This paper presents new Switched-Capacitor (SC) Finite Impulse Response (FIR) interpolators whose frequency responses are no longer affected by the input sample-and-hold filtering effect that occurred in previous circuits. Two different types of architecture are discussed: one based on a Direct-Form Polyphase structure and another using Active-Delayed Block (ADB) Polyphase structure. Examples are given to illustrate both types of SC FIR interpolator circuits.

WA1.2 Fuzzy Logic I - Roland Priemer, Chairperson - Scheman Room 262

Recognition of Geometrical References for Mobile Robots Using Fuzzy Logic
A. Indelicato, Teodiano Freire bastos, V. Dynnikov, J.E.M. Xavier, H.A. Schneebeli, Universidada Federal do espirito Santo, Vitoria, E.S., Brazil
The geometrical references recognition task is basic to the navigation of mobile robots due the necessity of the robot calibrate the position of time in time and planning of the trajetory. Usually, ultrasonic sensors are used to collect informations about the robot's surroundings. The high grade of uncertainness associated to these informations justifies the use of fuzzy logic to carry out the geometrical references recognition task. This work develops a method that allows object recognition, based on parameters of the echo signal from target.

Applications of Fuzzy Logic in the Analysis of 57Fe Mössbauer Spectra
Paulo A. de Souza Jr., Marcelo Malini Lamego, Universidade Federal do Espirito Santo, Vitoria, E.S., Brazil; Vijayendra Kumar Garg, Universidade de Brasilia, Brasilia, D.F., Brazil
Fuzzy sets has been used to control and accelerate the fitting of typically complex Mossbauer spectra of several iron containing substances. The least square fitting of such a spectrum is difficult, requires sufficient dedicated time of a Mossbauer spectroscopist. The present study indicated that a Fuzzy Set System associated with a Mossbauer fitting program is a helpful tool to obtain parameters of a experimental Mossbauer spectra.

Speed Control of DC Motors Using a New Fuzzy Controller
A. Khoei, Kh. Hadidi, Urmia University, Urmia, Iran
The development of fuzzy controller for speed control of DC Motors based on a PC is considered. The system is capable of executing any desired speed profile for a constant torque. The performance of fuzzy controller and its capability of optimizing the time response parameters such as maximum overshoot, rise time, and dead time are also considered. Also a method named Dynamic Control Action Set (DCAS) for minimizing the chattering around the setpoint and further optimization of time response, is proposed.

WA1.3 Analog Arrays I - Sherif Embabi, Chairperson
Sherif Embabi, Session Organizer - Scheman Room 260

A Field Programmable Analog Signal Processing Array
S.H.K. Embabi, X. Quan, N. Oki, A. Manjrekar, Edgar Sanchez-Sinencio, Texas A&M University, College Station, Texas, USA
This paper presents a new approach towards Field Programmable Analog Arrays (FPAA), which are dedicated to analog signal processing. It is based on simple current-mode sub-circuits which allows for operation at high frequencies and low voltage, and can be implemented in a digital CMOS process. A novel solution to avoid using excessive number of programming devices in the signal path is proposed. We have demonstrated through simulations that the proposed FPAA can be programmed to operate over almost 2 decades (from 30 kHz to 10mhz).

Current Conveyors Field Programmable Analog array
Christopher Premont, Cimirly Insa Lyon
An approach for designing field programmable analog array is described. The analog array is based on current conveyors. They are used both as analog cells and interconnection elements. Each elementary cell is associated with tunable resistors and capacitors. The analog function is implemented by suing the elementary cells programmed as current-mode amplifier or continuous -time filters. The first step of this approach is to be able to develop basic analog functions such as low-frequency filtering and amplifying. The current conveyors and the tunable resistors have been fully simulated thanks to Spcie, and the layout design is in preparation. Future work will consist of validating this new programmable topology.

Multi-function Block for a Switched Current Field Programmable Analog Array
Simon t. ching, Barie R. Hayes-Gill, christopher J. Paul, Nottingham University, Nottingham, England
This paper presents a Field Programmable Analog Array (FPAA) based on switched-current (SI) signal processing technology. A design methodology for building complex switched current circuits using a four phase clock is proposed. A programmable building block, called the "Multi-function Block", has been designed and can be used in a FPAA to implement circuits based on this four phase tech nique. The block was fabricated using 0.7 m CMOS and test results are presented. Simulation software for circuits built using these MFBs was written in C++. Example circuits, their implementa tion using MFBs and simulation results generated by this software are also presented.

Reconfigurable Pipelined Data Converter Architecture
Edward K.F. Lee, Iowa State University, Ames, Iowa, USA
A reconfigurable pipelined data converter architecture suitable for implementing Field-Programmable Mixed analog and digital Array (FPMA) is described. The proposed architecture can be reconfigured to a number of different analog-to-digital converters and/or digital-to-analog converters for a given number of conversion stages with different resolutions.

WA1.4 Digital Filters I - B. Nowrouzian, Chairperson - Scheman 252

A Trellis search Algorithm for the design of FIR Filters with Signed-Powers-of-Two Coeffi cients
Chao-Liang Chen, Michael C. Chen, Alan N. willson, Jr., University of california, Los Angeles, California, USA
A trellis search algorithm is presented for designing FIR filters with sums of signed-powers-of-two (SPT) coefficients. Formulating the design problem as a dynamic-programming-like recursive optimization problem, our algorithm approximates its optimal solution by performing a trellis search that is similar to the Viterbi algorithm. A design example shows that our algorithm is capable of designing filters with better frequency response, using a given number of SPT terms, than the existing methods.

Design of Linear Phase Bessel IIR Digital Filters
Youssef H. Aris, Larry D. Paarmann, Wichita s tate University, Wichita, Kansas, USA
This paper reports on an algorithm that transforms from the s plane to the z plane such that phase characteristics in the s plane are preserved in the z plane. This algorithm is based on cepstral decorrelation and the Hilbert transform. The algorithm is applied to the design of digital Bessel filters in the paper.

A Computer Program for the Design and Analysis of Linear Phase FIR Raised Cosine Filters
B.M. Al-Hashimi, W.P. Somerset, M. Moniri, Staffordshire University, Stafford, United Kingdom A computer program for the design and analysis of linear phase FIR raised cosine filters is described. It allows the design of high performance traditional and equiripple FIR filters with roll-off factors in the range between 0 and 1. Furthermore, the program is capable of computing coefficients for the recently introduced raised cosine filters with roll-off factor >1. The design algorithms are solved using linear programming techniques. A number of filter examples are included.

Design of Nearly Linear Phase IIR Digital Filters with Prescribed Log Magnitude Response
Masahiro Okuda, Masaaki Ikehara, Shin-ichi Takahashi, Keio University, Yokohama, Japan
In this paper, we propose a new design algorithm for nearly linear phase IIR digital filters with prescribed log magnitude response. The error function used here is the sum of the weighted log magnitude-squared error and phase-squared error, and so it is possible to control log magnitude and phase response directly. The gradient vector of the proposed error function is easily calculated as the closed form solution because of its nature , in which the real and imaginary part of the logarithm of a complex transfer transfer function corresponds to the log magnitude and phase response, respectively. This algorithm is simple and converges quickly. Finally, we show the validity of the proposed algorithm with some examples.

WA1.5 Transforms and Spectral Analysis I - Peter Sherman, Chairperson - Scheman Room 275

Fast Computation of the Two-Dimensional Discrete Fourier Transform
D. Sundararajan, M.O. Ahmad, Concordia University, Montreal, Quebec, Canada
In this paper, a new radix-2 2-D decimation-in-time discrete Fourier transform algorithm is presented. This algorithm uses an approach employing vectorized data to reduce the structural overhead of the algorithm compared with the Cooley-Tukey radix-2 2-D algorithm. Computational complexity and run-time comparison of the algorithm are also provided.

Matrix Factorization Derivation and Analysis of Computational Complexity of a New Radix -2 DFT Algorithm
D. Sundararajan, M.O. Ahmad, Concordia University, Montreal, Quebec, Canada
Recently, a new family of discrete Fourier transform algorithms has been reported in which the structural complexity is significantly reduced without affecting the arithmetic complexity. In the present paper, a radix-2 algorithm of this family is derived using the matrix factorization approach. This approach is known to be useful in mapping algorithms to architectures. An analysis of the computational complexity of this algorithm is carried out. A run-time comparison of the proposed algorithm is made with the Cooley-Tukey radix-2 algorithm.

Refining the Digital Spectrum
Pedro A. M-S. David, Jacques Szczupak, Pontificia Universidade Catolica, Rio de Janeiro, Brazil
This paper presents a simple algorithm for the correction of the spectrum and extends it to find the true frequency of the signal components. The paper thoroughly discusses the performance of these algorithms on usual and/or critical cases. Some of them are included in this summary, for illustration. An implementation of the algorithms, as M-scripts for MATLAB 1, is appended to the paper.

Bilinear Multirate Transformation a Comparison
Carlos Augusto Duque, Jacques Szczupak, Pontificia Universidade Catolica, Rio de Janeiro, Brazil
This paper presents a modified bilinear transformation that allows magnitude linearity at a wider range of frequencies. The method is used to synthesize linear networks in the state variable form. The network is discretized using high sample rate, but simulated at lower sample rate. This is pos sible because it utilizes a multirate processing and the interaction between network and the interpola tion filter. Several examples show that distortion frequency is reduced with lower computationall effort.

WA1.6 Speech and Audio I - W.K. Jenkins, Chairperson - Scheman Room 250

Transform Domain LMS Algorithm in Fullband and Subband Based Acoustic Echo Cancella tion
Z. Zhongwei, M. Tahernezhadi, Northern Illinois University, DeKalb, Illinois, USA
The objective of this paper is to investigate the effectiveness of transform domain adaptive algorithms is suppressing the acoustic echo in fullband and subband acoustic echo cancellers (AEC). The subband AEC considered consists in main and auxiliary subbands designed to provide a numerically efficient subband structure. Comparison results between the fullband (using the transform domain LMS adaptive algorithm) and subband cancellers are provided.

Blind Adaptive Filtering of Speech from Noise of Unknown Spectrum Using a Virtual -Feedback Configuration
Daniel Graupe, Northwestern University, Evanston, Illinois, USA; Dusan Veselinovic, University of Illinois at Chicago, Chicago, Illinois, USA
The paper describes a blind adaptive filter (BAF) of speech from m noise when neither speech nor noise ae are accessible nor are their parameters known. The BAF employs prior knowledge on the stationarity structure of human speech and assumes quasistationarity of the noise over a longer interval than any speech phoneme. The system employs an identifier that is followed by a speech /noise parameter separator anf and a virual feedback loop where a cost functional is minimized to minimize the stationary part of the input while maximizing the rest, . SNR improvements for various colored and white noises range from 14 to 36 dB for -20 dB SNR inputs.

On a Cepstral Technique for Pitch Control in the High Quality Text-To-Speech Type System
MyungJin Bae, SangHyo Lee, Soongsil University, Seoul, Korea
In the area of the speech synthesis, the waveform coding methods maintain the intelligibility and naturalness of synthetic speech. But, the methods can not be easily applied to synthesis by rules, since it is difficult to change the pitch of speech signal in waveform coding. In order to apply the waveform coding or hybrid coding techniques to synthesis by rule, we must be able to alter the pitches of synthetic speech. In this paper, we propose a new pitch altering method that compensates phase distortion of the cepstrum pitch alteration method with zero inserting-pitch halving method in the time domain. This method can remove the spectrum distortion which is generated in conjunction point between the waveforms.

A Comparison Between Fourier and Biological Auditory Based Time-Frequency Distribu tions, Applied to the Speech Signals
M.N. Souza, L.P. Caloba, Universidade Federal do Rio de Janeiro, Rio de Janeiro, R.J., Brazil
The behavior of the Fourier and biological auditory based Time-Frequency Distributions will be studied in this paper using the correlative domain. Results for conventional test signals, as well as for speech signals will be presented, showing that the distribution derived from a biological auditory wavelet gives better results than the classical spectrogram.

WA1.7 Power Systems I - Jake Glower, Chairperson - Scheman Room 208

Analysis of Structural Properties Responsible for Nonlinear Modal Behavior of a Stressed Power System Using the Normal Form Technique
Y. Ni, V. Vittal, A.A. Fouad, W. Kliemann, Iowa State University, Ames, Iowa, USA
In this paper a scanning program is developed based on the normal form technique to analyze the structural properties of a stressed power system. The mathematical model and procedure are presented. Numerical testing on a 17-generator system shows that in stressed systems, more modes can be strongly excited and new modal resonance phenomenon may occur, which will cause severe nonlinear modal behavior through modal interaction.

An Algorithm for Short Term Unit Commitment Based on Heuristic Approach
Wong Tik Keong, Asrat Teshome, Wichita State University, Wichita, Kansas, USA
An alogrithm based on a heuristic approach has been developed to perform short term unit commitment. The proposed algorithm consists of several reasoning loops to generate a schedule that satisfies all given constraints. The decision from these loops are based on a partial priority list method and some appropriate heuristic rules. A prototype program written in C language is used to implement the proposed algorithm and some numerical examples are simulated to show the results. The advantages of this algorithm are featured by its simplicity of implementation and a relatively short processing time.

New Concepts About the Power Received by Ideal Energy Storage Elements: The Instanta neous Complex Power Approach
Dalgerti Lelis Milanez, Universidade Estadual Paulista, Solteira, S.P., Brazil
This paper presents some definitions and concepts of the Instantaneous Complex Power Theory which is a new approach for the Akagi's Instantaneous Reactive Power Theory. The powers received by an ideal inductor are interpreted and the knowledge of the actual nature of these powers may lead to change the conventional electrical power concepts.

Distribution Systems Load Flow Based on Radial Equivalents of Meshed Systems
Dagoberto Lourenco Ribeiro, Wellington Santos Mota, Antonio do Nascimento Epaminondas, Universidade Federal da Paraiba, Campina Grande, P.B., Brazil
A technique based on the first iteration of Newton method is proposed attempting overcome dificulties showed by the Fast Decoupled Load Flow when applied to systems with high R/X relations. A radial equivalent for the meshed system is performed during the links inclusion. Tests with ten systems were done. Results are presented and discussed.

WA1.8 Analog Integrated Circuits II - Ahmad B. Dowlatabadi, Chairperson
W. Timothy Holman, Session Organizer - Scheman Room 204

A Generic Voltage Comparator Analog Cell Produced in Standard Digital CMOS Technologies
Ahmad Baghai Dowlatabadi, National Semiconductor, Santa Clara, California, USA; J. Alvin Connelly, Georgia Institute of Technology, Atlanta, Georgia, USA
A voltage comparator has been developed for use as an analog cell in standard digital CMOS processes for a mixed-signal application, operating from single 5 Volt power supply. The design is generic to variety of CMOS technologies of various minimum feature sizes.

A Macromodel for a FLOTOX EEPROM
Ming Meng, Kenneth V. Noren, University of Idaho, Moscow, Idaho, USA
A macromodel for FLOTOX EEPROM cell is presented. The model is suitable for DC and transient simulations and programming/erasing cycles. The modeled characteristics include threshold voltage changes, drain current, tunneling current, and charge retention. Model simulations are presented and compared with measurement data available in the literature.

An Analog Gaussian Generator Circuit for Radial Basis Function Neural Networks with Non-Volatile Center Storage
John Galbraith, W. Timothy Holman, The University of Arizona, Tucson, Arizona, USA
A BiCMOS analog circuit that computes a Gaussian function suitable for use in radial basis function neural networks is proposed. The circuit accepts multiple voltage inputs and stores the basis function center in a set of non-volatile floating gate memory cells. Experimental results of the circuit are shown.

Experimental Test of a Charge Conserving Macromodel for MOSFETs
Joon-Yub Kim, Texas Instruments, Dallas, TX, USA; Randall Geiger, Iowa State University, Ames, Iowa, USA
A charge conserving macromodel convenient to use and effective for the simulation of the charge injection behavior of MOSFET switches has been presented. The macromodel is compared with the SPICE level 2 charge conserving model. The accuracy of the macromodel is tested by measuring the clock-feed-through error of a sample-and-hold circuit.

WA1.10 Image Processing, Neural Networks and Filters - Ken Noren, Chairperson - Scheman Room 171 - 179

Bayesian Motion Estimation Without Spatial and Temporal Gradients
Richard R. Schultz, University of North Dakota, Grand Forks, North Dakota, USA; Robert L. Stevenson, University of Notre Dame, Notre Dame, Indiana, USA
A Bayesian motion estimation technique is proposed which models the motion vector field with a discontinuity-preserving prior and observation noise with a Gaussian density. The method is similar to optical flow techniques, except that it is not dependent on spatial and temporal gradients which are notoriously difficult to estimate from real image sequences. Simulation results show that the Bayesian technique reduces the energy of the displaced frame difference over conventional block matching.

Image Enhancement with 2-D Block LMS Adaptive IIR Filters
Alfredo C. Tan, Fairleigh Dickinson University, Teaneck, New Jersey, USA
We introduce two-dimensional adaptive infinite impulse response filters using the block least-mean -square algorithm. In this scheme, the multiplier coefficients of the adaptive filter are updated once per block of pixels rather than per pixel. This approach leads to fast implementation and lower computational complexity. Simulation results are included to demonstrate the potential applications of the proposed adaptive filter in two-dimensional system identification and image enhancement.

MAPS: A New and Efficient Block-Matching Criterion for Motion Estimation
Ming-Der Shieh, Ming-Hwa Sheu, Yu-Chin Hsu, National Yunlin Institute of Technology, Yunlin, Taiwan
A novel block-matching criterion called Mean Absolute error of Projective Sum (MAPS) is presented to significantly reduce the intensive computation of motion estimation. It can save about 78% in the computation load comparing with the conventional MAD (Mean Absolute Difference) criterion. And, simulation results show that its prediction quality if very close to the MAD's.

A New Morphological Approach for Reducing Speckle Noise in Ultrasonic Images
Young-Won Song, Satish S. Udpa, Iowa State University, Ames, Iowa, USA
Ultrasonic images are often contaminated with speckle noise. Morphological operators have a demonstrated track record of success in minimizing speckle noise in gray level images. This paper presents a novel approach for reducing speckle noise without eliminating small features using
residual images. The validity of the approach is demonstrated using several ultrasonic nondestructive evaluation images.

Efficent Algorithm for Fast Computation of Zernike Moments
S.O. Belkasim, M. Ahmadi, M. Shridhar, University of Windsor, Windsor, Ontario, Canada
Zernike moments have been used as shape descriptors in several object recognition application.
The classical method of computing Zernike moments is dependent on the regular moments which makes them computationally expensive and inefficient. In this paper we present an efficient and fast algorithm for the direct computation of Zernike moments. This algorithm is based on using some properties of Zernike polynomials.

Asymptotic Error Bounds for Block Floating Point Digital Filters
Kamen Ralev, Peter Bauer, University of Notre Dame, Notre Dame, Indiana, USA
The paper analyzes the register length requirements for block-floating point digital filters. Both shift and delta operator sytems are considered. In particular the block mantissa length requirements are investigated, if the asymptotic response of the filter is to converge to underflow or to be bounded. Examples which illustrate the conservatism of the result are also provided.

A State-Space Realization Form of Two-Dimensional Digital Filters
Atsushi Kawakami, Kanazawa Institute of Technology, Japan
In this paper, we propose a method for obtaining the state-space realization form of two-dimensional digital filters (2DDF). By using our method, realization form can be directly obtained from the coeffi cients of the transfer functions of 2DDF, without performing canonical decomposition of the state equations and solving nonlinear equations. Consider the possiblity of reducing the realization dimension in our state-space realization form, and obtain conditions imposed on the transfer func tions of 2DDF for the reduction. Moreover, we present a class of transfer functions of 2DDF that can be realized with minimal dimension with respect to both of two variables.

Simple Explicit and Recursive Formulas for the Coefficients of FIR Digital Differentiators for Midband Frequencies
J. Le Bihan, Ecole Nationale d'Ingenieurs de Brest, France
An efficient algorithm for calculating the coefficients of FIR digital differentiators for midband frequen cies is presented. Simple closed-form explicit and recursive formulas are derived in a straight forward manner. Moreover, a simple recursive relation is obtained, relating the coefficients of two digital filters of adjacent ranks.

Some Theoretical Results Based on Passivity and Losslessness for Linear Time-Invariant Digital Filters
Ying Xiong, University-GH-Paderborn, Paderborn, Germany
In this paper a suitable definition of passivity and losslessness will be selected for linear and time -invariant digital filters described by state-space equations but without any knowledge of their struc tures. For this passivity and losslessness it is a well-known necessary condition that the digital filter has a bounded and paraunitary bounded transfer matrix, respectively. This will also become suffi cient if the digital filter is minimal. Some further results based on these concepts will be presented.

An Application of Wavelet Basis Function Neural Networks to NDE
K.T. Hwang, S. Mandayam, S. Udpa, L. Udpa, Iowa State University, Ames, Iowa, USA
This paper presents an alternate approach for training a multiresolution, hierarchical wavelet basis function neural network. Such networks can be employed for characterizing defects in gas pipelines inspected using the magnetic flux leakage method of nondestructive testing. The result indicates the accuracy of the predicted profile can be controlled by the resolution of the network.

A Neural Network-Based Detection Thresholding Scheme for Active Sonar Signal Tracking
Y. Sun, M. Farooq, Royal Military College, Kingston, Ontario, Canada
Intensity thresholding is an effective technique to cut off the low energy noises and cut down the computational load in an underwater target tracking system. A neural network based adaptive intensity thresholding scheme with CFAR for an active sonar signal tracking situation in a realistic sea environment is proposed in this paper. The proposed system has the following advantages: 1) It performs well in nonhomogenous sea environment; the false alarm rate is kept constant while the threshold changes with different sea environment; 2) It can adaptively estimate the threshold for
different range cells because the noise under estimation is strictly local so that the received intensities of noise and targets are not affected by the distance they travel; 3) The computational requirements are moderate.

A Comparison of Different Spectral Analysis Models for Speech Recognition Using Neural Networks
Ricardo S. Zebulum, Guy Perelmuter, Marley Vellasco, Marco Aurelio Pacheco, Pontificia Universidade Catolica do Rio de Janeiro, Rua Marques de Sao Vicente, R.J., Brazil
This work studies the application of Neural Networks in the problem of speech recognition. The main objective is to compare the performance of a Neural Network based Speech Recognition system using different spectral analysis models. Different types of coefficients, like the MelCepstron and the Auto Correlation coefficients are used as the Neural Network input. We focus on the particular problem of recognizing 10 isolated words in Portuguese.

Implementation of Neural Network and Genetic Algorithms for Novelty Filters for Fault Detection
Hamed Elsimary, Electronics Research Institute, Cairo, Egypt
In this paper a method for detecting faults in rotating motor/generator rotors using computational intelligence techniques (neural network and genetic algorithm) is presented. The methods of signal processing and detection of faults in operating motor/generator is discussed. The use of novelty filters for the detection of fauts is described. Genetic algorithm has been used to train the neural network to enhance the capabilities of the neural network novelty detecto. The proposed technique has been applied and the simulation results is presented to show the effectivness of the proposed technique.

Adaptive Arabic Handwritten Letter Recognition by a Layered Self-Organization Neural Network
Mazher Tayel, Hossam Shalaby, Alexandria University, Egypt; Hossam Saleh, Petroleum Pipeline Company, Alexandria, Egypt
A neural network model of visual pattern recognition, called the Necognitron, was previously proposed by Kunihiko Fukushima. After training, it can recognize input pattern without being affected by change in size or shifts in position. Unfortunately, the self-organization stability of the network is disturbed by a fundamental problem about learning g and memory that is called stability-plasticity contradiction. This article proposes a solution to the problem by developing a learning scheme for the necognitron model. The proposed scheme enhance the network capability of plasticity in order to learn about significant local features in the Arabic alphabetic, yet its memory also remains stable in response to irrelevant differences of the multishapes Arabic letters to achieve stability without rigidity. The concerned learning process defends the system fully committed memory capacity against being washed away by the incessant flux of new input features. We developed an expert system, which is written in C++ and run on a PC computer (80486-66MZ, 8MB RAM), to help in analyzing the performance of the simulated network of the proposed model. The expert system assists us to trace the proposed model behavior and then to establish relations for the vigilance parameter and for the parameters that control the selectivity for feature extraction which grantee a stable learning reliable pattern recognition model. The simulated network has 4 stages of layers. The input layer has a 2-D array of 18*18 cells, the recognition layer has 28 one-element-array and the total number of cells in the network is 40,000 cells.

Short-Term Electric Load Forecasting Using Neural Network Models
Yasser Al-Rashid, Larry D. Paarmann, Wichita State University, Wichita, Kansas, USA
A Neural Networks approach is used in this paper to construct winter and summer models for 24 hour ahead power load forecasting. After training, these two models were able to map the relationship between the input variables and output load. The models were tested by forecasting the load for Wichita, KS, area throughout 1992. Then the forecasted results were compared to the actual load and the performance was evaluated and compared with that of ARMA model.

10:00 - 10:30 am Refreshment Break - Scheman First Foor Lobby

10:30 - 12:20 am CONCURRENT SESSIONS

WA2.1 Device Design and Modeling - Ben Rodanski, Chairperson - Scheman Room 167

Modeling Non-Uniform Doping In Subthreshold MOSFETs
Richard C. Meitzler, Hewlett Packard Company, Palo Alto, California, USA; Andreas G. Andreou
For subthreshold MOSFETs, there can be large variations in depletion region width relative to above threshold operation. This paper investigates the significance of non-uniform doping effects in subthreshold conduction through numerical calculations based on measured dopant profiles. A semi-empirical expression suitable for inclusion in a public domain MOSFETmodel is also pro posed.

Improved MOSFET Model for High Frequency Analog Design
Alaa Y. El-Sherif, Southern Methodist University, Dallas, Texas, USA; Keith R. Green, Texas Instruments Co., Dallas, Texas, USA; Kenneth L. Ashley, Southern Methodist University, Dallas, Texas, USA; Thomas Vrotsos, Texas Instruments Co., Dallas, Texas, USA
An improved non-quasi-static (NQS) MOSFET model is introduced for more accurate response at high frequencies for deep-submicron and narrow devices. The model is valid over a wide range of device dimensions and different operation conditions. The evaluation of the model is based on the determi nation of the 50-Ohm scattering parameters using SPICE.

Micromachined Thermocouple Detector in CMOS Technology
Veljko Milanovic, Michael Gaitan, Mona E. Zaghloul, The George Washington University, Washington DC, USA
This paper reports fabrication of a thermocouple microwave detector through commercial CMOS to 10 Ghz, powers from -30 dBm to 10 dBm. Device has very good linerity over the 40 dB dynamic range. Input return loss is <-20 dB over the frequency range, and sensitivity is 1.3 mV/mW.

Access Paths Effects in Time-Domain Si-BJT Modelling
S. Megherbi, P. Crozat, G. Raynaud, J.F. Pone, Universite de Paris-sud, France
The aim is to subtract the small-signal high-frequency access path effects, using an equivalent model in the 45MHz-10GHz range. We propose to apply to bipolar silicon technology a de-embedding GaAs-Mesfet method, for which the effect of access path are considerably lower and therefore can be neglected. The correction will be taken in to account for each frequency value in another study, during a time domain modelisation of the BJT.

Modeling and Characterization of the IGBT Electrothermal Transistor for Circuit Applica tions
A. Amimi, R. Bouchakour, T. Maurel, Ecole Nationale Superieure des Telecommunications, Paris, France
An analytical model for the power Insulated Gated Bipolar Transistor ( I GBT ) including electrothermal interactions is developed. In this model the device's t emperature becomes an interactive variable during the simulation. The model has been implemented in Saber circuit simulator. Comparisons between simulation and measurement show that the steady-state as well as the transient analysis of the IGBT are accurately reproduced.

WA2.2 Fuzzy Logic II - Julie Dickerson, Chairperson - Scheman Room 262

Learning Optimal Fuzzy Rules Using Genetic Algorithms
Julie Dickerson, Iowa State University, Ames, Iowa, USA
Fuzzy systems can uniformly approximate continuous functions, but the number of rules increases geometrically with system dimension. Fast simulated annealing that uses alpha stable generating functions with alpha values less than 1 can find the optimal fuzzy rules that approximate a function. Smaller values of alpha give thicker tails on the probability distributions. This aids in the search for a globally optimum solution.

Fuzzy Class Learning Vector Quantizer in Image Compression
Hamed Parsiani, Eastman Kodak Company, Rochester, New York; Oscar Misla, University of Puerto Rico, Mayaguez, Puerto Rico
The Fuzzy Learning Vector Quantizer (FLVQ) has the advantage over the conventional LVQ in that every cluster center and not just the winner node is updated. However, our experimental results show that many of the edges with this method are lost, if the number of shade vectors are large. We propose a Fuzzy Class Learning Vector Quantizer which produces good image quality with very small codebook size. Typical codebook size of 50 (block size of 3X3) for a 256X256 greyscale image and PSNR=26 was obtained.

Adaptive Fuzzy Logic Based Controller for an Active Suspension System
Nizar Al-Holou, Jonathan Weaver, Dae Sung Joo, University of Detroit Mercy, Detroit, Michigan, USA
The authors have developed FLC*s for semi-active and active suspension system models. The FLC structures are based on some combinations of body velocity and wheel velocity. The models were simulated and the results demonstrate lower body accelerations (i.e., better ride quality) than achieved with Skyhook logic. We are working to develop a fuzzy neural network-based controller (FNNC) for an active suspension system. The developed system will be trainable and adaptable. The learning is accomplished by integrating interference into a feedforward neural network. A two-phase training algorithm will be employed.

Fuzzy Set Theory in Problems of Multicriteria Power and Energy Shortage Allocation
Petr Ya. Ekel, Federal University of Santa Maria, Brazil; Arthur A. Prakhovnik, Kiev Polytechnical Institute, Ukraine; Rafael E. Kempp, Marcos A. de Moraes, Andre L. Konig, Federal University of Santa Maria, Brazil
The problems of power and energy shortage (natural or associated whith the efficiency of load management) allocation are formulated as multicriteria optimization problems to consider and to minimize diverse negative consequences of consumer limitation. Their solution is based on the Bellman-Zadeth approach to descision making in fuzzy environment. The advantages and capabilities opened by its application are discussed. The result are realized as an adaptive interactive decision making system which can be used in allocating different types of resources or their shortages.

Using Fuzzy Logic Techniques to Evaluate Global Dynamic Performance of Control Sys tems
C.M. Polito, F.G. Jota, Universidade Federal de Minas Gerais, Minas Gerais, Brazil; A.R. Braga, CEFET. Minas Gerais, Brazil; R.T. Pena, Universidade Federal de Minas Gerais, Minas Gerais, Brazil
The performance assessment of control system loops is a very important task in the real plant day- to-day life. The optimal operation of multiloop process control systems depends heavily on good decisions on "where", "when", and "how much" tuning is needed. To do this job, the decision maker needs well conditioned information. Therefore, a numerical calculation of a Global Performance Assessment Index, allowing an evaluation of the whole plant unit behavior, is highly desirable. This paper proposes the determination of such an index, based on fuzzy logic techniques.

WA2.3 Analog Arrays II - Sherif Embabi, Chairperson
Sherif Embabi, Session Organizer - Scheman Room 260

The EPAC Architecture: An Expert Cell Approach to Field Programmable Analog Circuits
Hans W. Klein, IMP, Inc., Pleasanton, California, USA
This paper describes the architectural configuration of the Electrically Programmable Analog Circuit (EPAC), an expert cell approach to meeting the market need for an analog counterpart to the digital FPGA. It provides an overview of the technology and describes the internal operation of the first commercial EPAC devices.

Mixed-Mode Modeling and Simulation of a Field-Programmable Analog Array Implementa tion of a Phased-Locked Loop
Edmund Pierzchala, Charles Wilson, Portland State University, Portland, Oregon, USA
This paper describes the results of behavioral modeling of a phase-locked loop (PLL) implemented in a field-programmable analog array (FPAA) of cellular structure. The limitations of the particular FPAA architecture and cell design (as used for the PLL implementation) are explored using a mixed -mode (analog-digital) Saber simulator, and its modeling language, MAST.

Behavioral Modelling and Simulation of a Field-Programmable Analog Array (FPAA) Cell
Edmund Pierzchala, Ira Pollock, Chee How Lim, Portland State University, Portland, Oregon, USA This paper presents behavioral modeling of the analog and digital circuitry of a single cell of a field -programmable analog array (FPAA) of cellular structure. A model of the cell is developed using MAST analog hardware description language (AHDL) Mixed-mode capabilities of MAST and the Saber simulator are used to examine the combined performance of the analog and digital parts of the cell. The impact of the digital delays, D/A converter resolution, and other effects, as well as the behavior of the analog part of the cell are examined.

WA2.4 Digital Filters II - B. Nowrouzian, Chairperson - Scheman Room 252

LS-Minimax FIR Filter Design
Ashraf Alkhairy, Cambridge, Massachusetts, USA
We present a method for the design of an FIR linear-phase frequency-selective filter that minimizes the Chebyshev error in the passbands and the least squares error in the stopbands. The highly efficient procedure is characterized by desirable convergence properties and yields substantially better filters than other approaches.

QMF Filter Bank Design by a New Global Optimization Method
Benjamin w. Wah, Y. Shang, T. Wang, T. Yu, University of Illinois at Urbana-Champaign, Urbana, Illinois, USA
In this paper, we present a new global constrained optimization method for desiging QMF filter banks. We formulate the QMF filter bank design problem as a nonlinear constrained optimization problem. Our formulation uses the reconstruction error as its objective function, and constrains other perfor mance measures, such as stopband ripple, stopband energy, passband ripple and passband energy, with respect to values in the best known design. We have solved the nonlinear constrained design problem using {\em NOVEL}, a global optimization method we have developed to overcome effec tively local minima encountered in local searches. Our method has found improved designs over those found by existing methods.

Minimum Mean Square Error Design of Stack Filters
C.E. Savin, M.O. Ahmad, M.N.S. Swamy, Concordia University, Montreal, Quebec, Canada
This paper addresses the problem of minimum mean square error (MSE) design of stack filters. It is shown that the MSE between the desired signal and the estimated one can be expressed as a linear combination of the responses of positive Boolean filters to different binary input vectors. As a result, a stack filter which is optimal for signal estimation in the MSE sense can be determined as the solution of a linear program.

Optimal Design and Adaptive Implementation of Power Estimation Using the Hammerstein Model
Aiping Huang, Timo I. Laakso, Helsinki University of Technology, Espoo, Finland
Optimal power estimation problem is investigated in this paper. The method of optimizing the Hammerstein model is introduced for the power estimation of real-valued stationary signals. An adaptive algorithm for time-varying signals is developed. These enable an effective and accurate estimation of the signal power transmitted in a noisy environment like noisy fading mobile channel.

WA2.5 Transforms and Spectral Analysis II - Peter Sherman, Chairperson - Scheman Room 275

Efficient Computation of the DFT for Some Transform Lengths That Are Not Integral Powers of Two
D. Sundararajan, M.O. Ahamd, Concordia University, Montreal, Quebec, Canada
In this paper, it is shown that radix-2 discrete Fourier transform algorithms can be designed for some transform lengths that are not integral powers of two. These algorithms reduce the overhead operations significantly compared with the mixed-radix algorithms. A specific algorithm is described in detail and its computational complexity and run-time performance are compared with the corresponding mixed-radix algorithm.

Fitting of a Mössbauer Spectrum Using a Genetic Algorithm
Hannu Ahonen, Universidade Federal do Espirito Santo, Vitoria, E.S., Brazil; Paulo A. de Souza Jr., Vijayendra K. Garg, Universidade de Brasilia, Brasilia, D.F., Brazil
A genetic algorithm was implemented for finding an approximative solution for fitting a combination of Lorentzian lines of a measured Mössbauer spectrum. Mössbauer spectroscopy is applicable to large number of industrial and scientific applications. The results of the genetic algorithm are compared with those obtained by a widely used commercial program.

Fast Computation of the Discrete Walsh Transform
D. Sundararajan, M.O. Ahmad, Concordia University, Montreal, Quebec, Canada
A new discrete Walsh transform (DWT) algorithm is derived in which a modified form of the DWT relation is decomposed into smaller-sized transforms using vectorized quantities. The proposed approach results in an algorithm requiring no independent data swapping, and less number of array -index updating and bit-reversal operations. These reductions in the overhead operations readily translate into a significant reduction in the execution time compared with the existing algorithm.

Analysis of Polymer Spectra Using Wavelet Transforms and Fuzzy Algorithms
Georg Wirth, Carsten F. Ball, Dieter A. Mlynski, Institut fuer Theoretische Elektrotechnik und Messtechnik Universitaet Karlsruhe, Germany
In this paper we propose an efficient strategy for qualitative analysis of polymer spectra. The data measured by an IR-spectroscope (Infra Red) is transformed by wavelet transforms acting as a feature extractor. Thus huge data sets are enormously compressed and only a few typical features (wavelet coefficients) remain for the identification process. Afterwards a fuzzy classification algorithm assignes the different spectra to various classes represented by reference data. The calculated membership values give a qualitative interpretation of the ingredients of the polymer. For that we implemented a special rule based approach with fuzzy if-then rules, a fuzzy c-means algorithm and a fuzzy Kohonen cluster algorithm.

WA2.6 Speech and Audio II - W.K. Jenkins, Chairperson - Scheman Room 250

QR Methods of O (N) Complexity: One-Channel and Two-Channel Acoustic Echo Cancella tion
M. Dai, M. Tahernezhadi, Northern Illinois University, DeKalb, Illinois, USA
Recent attention in least squares adaptive filtering has focused on methods derived from the QR factorization of matrices, owing to the development of a family of "fast" algorithms to execute its recursive computation, combined with the apparent stability of algorithms. This paper presents a fast adaptive least squares algorithm [1] for one-channel and two-channel acoustic echo cancella tion. The algorithm is based on Householder transformations. As verified by simulation results, this algorithm exhibits good echo attenuation.

Measurement of Acoustic Parameters by Means of Pseudorandom Signals
J. Brezmes, J. Marti, Universitat Rovira i Virgili, Tarragona, Spain; E. Vidal, E. Alarcon, Universitat Politecnica de Catalunya, Barcelona, Spain
This paper describes the use of pseudo-random signals to dentify the dynamic behavior of a linear system. This technique is applied to the design of a portable acoustic parameter analyzer. Finally, the advantages of pseudo-random signals and their application in non-linear systems such as chemical plants are presented.

Cryptoanalysis of Speech Signals Ciphered by TSP Using Annealed Hopfield Neural Net works
J.A. Apolinario Jr., P.R.S. Mendoca, L.P. Caloba, R.O. Chaves, COPPE/UFRJ, Rio de Janeiro, R.J., Brazil
This paper presents a scheme of cryptanalysis of speech signals ciphered by Time Segment Permuta tion (TSP) using fixed size jumping window. Spectral distances are estimated between borders of segments and a minimum overall distance is searched with a Traveling Salesman Problem approach using Simulated Annealing, Mean-Field Annealing and Genetic Algorithms

WA2.7 Power Systems II - Vijay Vittal, Chairperson - Scheman Room 208

Pricing the Variance of Wind Energy
Jacob S. Glower, North Dakota State University, Fargo, North Dakota, USA
At present, the value of the electricity produced by wind farms is assumed to be proportional to the amount of energy produced. This is not an accurate model for determining the value of wind energy, however, since it neglects the cost involved with the variability of the wind. In this paper, a second model is proposed where the value of the energy produced includes the variance of the energy. It is shown, through a model utility, that a highly variable energy source has significantly less value than a constant source. Similarly, the value of energy from wind farms should be computed using the model proposed herein.

Pricing Distributed vs. Centralized Wind Energy Production
Jacob S. Glower, North Dakota State University, Fargo, North Dakota, USA
Several studies in recent years have looked at the merits of distributed vs. centralized wind energy production. These studies have focused on effect of each on the transmission system, the infrastructure.