Search

A new, direct and practical scheme is proposed for determining the model orders of a system, its signal and disturbance using key properties of Kalman filter (KF). Unlike conventional methods, it enjoys the unique property of being both necessary and sufficient. The system is described by the Box–Jenkins model, whose accessible input and output are corrupted by unknown zero-mean white Gaussian-distributed disturbances and measurement noise. The signal and disturbance are outputs of asymptotically-stable linear time-invariant systems driven by an inaccessible input and a zero-mean white Gaussian noise process, respectively. Predictive analytics is used to estimate the input by exploiting its smoothness and the randomness of the noisy input. The system, signal, and disturbance models and their associated KFs are identified for various selected model orders by minimising the KF residuals so that these become zero-mean white noise processes. The selected model-order corresponds to the minimum-variance residual. Equivalently, the minimum order is selected when the number of poles or the output estimates of the identified models are all identical for all orders equal to, or exceeding the minimal order. The scheme is successfully evaluated and shown to outperform the commonly-used but only sufficient Akaike Information Criterion.

A two-stage identification scheme is proposed for multivariable systems for applications including spectral estimation, and signal and system model estimation. The statistics of the signal and of the corrupting noise are taken as unknown, except that the signal is assumed to have a rational spectrum. First, a very high-order model is estimated and then a reduced-order model is derived from the higher-order model. An algorithm based on theory and heuristics is developed to select a set of frequencies where the signal-to-noise ratio is high. A reduced-order model is obtained from the best weighted least-squares fit at the selected frequencies.

According to the latest ITRS roadmap, the pitch of area array packages is expected to decrease to 100 /spl mu/m by 2009. Simultaneously, the electrical performance of these interconnections needs to be improved to support data rates in excess of 10 Gbps, while guaranteeing thermomechanical reliability and lowering the cost. These requirements are challenging, thus, needing innovative interconnection designs and technologies. This paper describes the development of three interconnection schemes for wafer-level packages (WLPs) at 100-/spl mu/m pitch, involving rigid, compliant, and semicompliant interconnection technologies, extending the state of the art in each. Extensive electrical and mechanical modeling was carried out to optimize the geometry of the interconnections with respect to electrical performance and thermomechanical reliability. It was found that the requirements of electrical performance often conflict with those of thermomechanical reliability and the final "optimum" design is a tradeoff between the two. For the three interconnection schemes proposed, it was found that the electrical requirements can be met fairly well but acceptable mechanical reliability may require organic boards with a coefficient of thermal expansion of 10 ppm/K or lower.

A systematic and unified scheme is proposed to monitor the performance of control systems and to tune the controller automatically so as to minimize the integral squared-error measure. The features which are necessary for monitoring and controller tuning are identified to be a feature vector and an influence matrix, and a robust and reliable estimation scheme is proposed to extract these features from the raw measurement data. The feature vector is formed from the coefficients of the sensitivity function (closed-loop transfer function relating the reference input to the tracking error signal). The influence matrix is the Jacobian matrix of the feature vector with respect to the controller parameters. The feature vector and the influence matrix are estimated from a short-time record of the tracking error signal when the reference input is excited by a known input and controller parameters are perturbed one at a time. The proposed tuning scheme is evaluated on a simulated two-area interconnected power system.

An on-line scheme for monitoring the parameters of the signal components contained in a short-time record of power system data is proposed based on a linear predictive coding algorithm (LPCA). The information is displayed in decreasing order of importance but with increasing amounts of computation, such that the essentials are known in the shortest possible time with a complete picture available after a certain time. It is shown that the measured power system signal can be generated by a linear model and the coefficients of the numerator and denominator polynomials of its transfer function, which are called linear predictive code (LPC), completely characterize the signal so that the LPC estimates are used to compute the signal parameters such as the presence or absence of critical modes, frequency, degree of damping and the amplitude of the signal components, and finally the future signal values. The proposed scheme is evaluated with a real-time simulation of various power system dynamic oscillations.

The paper presents the design of a variable structure static VAR stabilizer using a sliding mode control technique. The static VAR compensator (SVC) uses a linear voltage regulator with auxiliary stabilizing signals through a variable structure switching system. The sliding mode control constrains the system motion to a state trajectory, and provides greater robustness to plant parameter variations than classical control schemes. Transient simulations of a typical power system are performed for two alternative locations of the SVC, namely, the generator terminal and the load bus terminal. Also, a nonlinear load is connected to the SVC bus to highlight the effectiveness of variable structure control for variations in load indices.

A detection, localization and correction of failure in launch vehicle guidance and control system is proposed. The detection and localization of faults is based on monitoring an abrupt change in the error signal in each of the feedback loops, namely, the actuator, digital autopilot, navigation and guidance as well as control loops of inertial sensors such as gyroscopes, inertial platform servos and accelero-meters. Correction schemes are proposed for errors located in each of the loops. A scheme based on reshuffling of control functions to healthy actuators, in the case of a single failure in an actuator loop, is described. Failures in inertia sensors are overcome by software signal reconstruction from healthy sensors. Failures in a guidance loop cause a switchover to an open loop mode that provides vehicle turning rates that are optimal at the last valid loop instant. Retargetting is provided to account for widespread dispersions in vehicle thrust performance.

Measures of overshoot, speed of response and robustness are derived based on the state transition matrix of the closed-loop system. A measure of overshoot is defined as the maximum value of the weighted norm of the state transition matrix. An exponentially decaying envelope is defined which confines the norm of the state transition matrix. The ratio of the amplitude and decrement factor defining the envelope is shown to be a measure of robustness. Comparison with other measures is given. The importance of the proposed measures is illustrated using numerical examples.

This abstract describes an on-going research to develop bio compactable food sensor to separate mycotoxin infected barley and wheat seeds from non-infected seeds. Since 1999 research has focused on preventing and/or eliminating mycotoxin contamination in barley and wheat seeds with limited success. The mycotoxin, deoxynivalenol (DON), produced by the fungus, Fusarium graminearum, pose serious health hazards such as vomiting, dermatitis, and hemorrhagic septicemia in humans and livestock that limit the utilization of the infected grains containing high toxin concentrations. The US Food and Drug Administration have set the wheat advisory concentrations of 1 /spl mu/g/g and 5 /spl mu/g/g of DON in food for human and livestock consumption, respectively. DON-free or low DON grain is required for malting barley as DON carries through malting and brewing into finished beer. The farmers of upper Midwest greatly suffer because of fungal infection and toxin contamination as the brewing industry refuses to accept such toxin contaminated seeds. Rather the brewing industry import seeds from Canada. This research develops a solution to separate toxin contaminated seeds from non-toxin healthy seeds using biosensor technology. Over the years sensing technologies relied on silicon based inorganic substrates and electrodes to sense the presence of toxins. Toxins can be identified by understanding the required marker and interface proteins for the sensor. This research for the first time presents a method of embedding capacitor based sensor to detect DON. The capacitors uses shape memory polymers as the dielectric. The sensor process used for fabrication and data acquired from the biosensor application is presented in detail.

In this paper we present our current joint work (Georgia Tech $Northwestern University) in developing diagnosis of children's brain cancer and lays down the requirements for fabricating, integrating and testing a bio compactable embedded RF cancer sensor for in vivo monitoring of the recurrence of brain tumor after treatment. The paper presents a detailed literature search on diagnosis of brain tumor. Several techniques used for the treatment and symptoms of recurrence of brain tumor is also presented. A case study done on a group of children is also presented. Through the study the parameters required for an in vivo sensor understood. Advanced bio materials and conductive polymers are used to fabricate a shape memory biosensor. The sensors response to a change in temperature and change in PhH of fluid. The sensor fabrication and its applications is in progress. The data collected till date and required improvements are presented at the conference.

Lead free solder bumping requirements have challenged researchers to develop technologies to achieve fine pitch interconnects. ITRS has predicts that by 2017 the industry require 70 micron pitch area array lead free interconnects for flip chips. This paper describes bumping, assembly and reliability evaluation of a nano composite 15 micron pitch interconnect technology. Nanoparticles are the most common building blocks in several applications of nanotechnology. Beside, the enormous increase of their surface area, the high surface to volume ratio of the nano particles results in extraordinary high reactivity, and unusual physical properties (optical, magnetic, etc.). Several types of nanostructures (nano particles, nano-wires and nano-rods) can be fabricated. Dispersed nano particles are used in several important applications, e.g., catalysis, biomedical applications, nano coatings and nanocomposites. Nanostructured materials, prepared by consolidating nanoparticles with a very high density of grain boundary, have been shown to lead to dramatically improved mechanical and physical properties. Lead free interconnect reliability has been a concern due tin pest (whiskers), intermetallics and high temperature reflow problems. This problem becomes even more pronounced when the pitch of the flip chip reduces to 100 microns or lesser. This paper proposes a nano metal composite for fine pitch lead free interconnect as an alternative to solder for fine pitch flip chip interconnect as presented in R. R. Tummala (2001) and W. D. Brown (1999). This paper describes design, process, fabrication and reliability test for composite interconnects fabricated using Ni nano particles distributed in a tin core. A wafer level process is developed for flip chip having 15 micron pitch composite nano structured interconnect. The composition and intermetallics formed at the chip-to- interconnect-interface are evaluated when sent through 4 reflow cycles. Composition of the intermetallics is evaluated to study the distribution Ni3Sn4, Cu6Sn5 and Cu3Sn. Shear test on interconnects provide bump adhesion strength and fatigue resistance information. The failure mode associated with shear test is also presented. The chip is thermally cycled in a oven from -55/spl deg/C to 125/spl deg/C and after every 100 cycles is evaluated for defects using x-ray and Sonoscan techniques. A cross section is made and the composition at interface is also evaluated for intermetalic diffusion. The paper provides optimized data for next generation nano composite interconnect for fine pitch flip chip to achieve high reliability. The paper also describes a process to fabricate substrates with nano lines and embedded fluidic channels.

A robust scheme to estimate a set of models for a linear time-invariant system, subject to perturbations in the physical parameters, from a short-time data record is proposed. The true model as well as the disturbances affecting the system are assumed unknown. However, the physical parameters are assumed to enter the coefficients of the system transfer function multilinearly. A set of models is identified by perturbing the physical parameters one-at-time and using a two-stage identification scheme. Exploiting the assumed multilinearity, the estimated set of models is validated. The proposed scheme is evaluated on a number of simulated systems and on a physical robot manipulator arm.

A high resolution parallel algorithm is proposed for estimating the spectrum of a narrow-band signal from a short data record. The algorithm is based on combining the nonparametric and parametric approaches, where the nonparametric approach is used to decompose the measurement data into an orthogonal set of components, and the parametric approach is used to estimate the model of these components in parallel. A fast Fourier transform (FFT) is used to decompose the signal. A singular value decomposition (SVD)-based linear predictive coding algorithm (LPCA) is used to obtain an autoregressive moving average (ARMA) model of the signal components. The FFT of the signal components is translated to the low-frequency region, and their inverse FFTs are decimated before estimating the ARMA model so as to separate the closely-spaced modes. The spectra of the estimates are translated back to their original location. The proposed algorithm is evaluated using simulation. >

Embedded systems play a crucial role is achieving very high component density and functions. Embedded actives have been the focus for three most important paradigms in systems integration. The paradigms are system in a package (SIP), system on a package (SOP) and system on a chip (SOC). ITRS road map requires very high component density by the year 2007. Embedded passives play an important role in determining frequency stability and signal integrity. Their stability depends on design, choice of material and process. There are limitations in achieving accurate parameters in the passives. Embedded passives have non reversible function or reworkability. These needs require techniques and process developed to have controlled actives which can change its parameters in a combination of mechanical, electrical and material properties. Flip chips are assembled on substrates enable thermal management. The need to mount several chip in unit area drives assembly process to wafer level chip stacking. There is need for embedded techniques to achieve good process for variable capacitors, resistors and inductors along with embedded flip chips. In this paper, we characterized the performance of a novel embedded variable capacitor fabricated using a polymer dielectric material, with one metal electrode fabricated on a low strength back plane to allow the electrode to flex. Sol-gel process is selected as it is a versatile technology to produce fine powders, fibers, coatings and is expected to yield highly homogeneous thin/thick films. A variable DC current is applied on the piezoelectric material to deform the dielectric, thereby changing the capacitance thus allowing the dielectric to behave as a variable capacitor. The compression effect on the dielectric also causes a thickness variation there by making the already submicron thick dielectric vary in thickness. The thinner dielectric makes it suitable for high frequency capacitance application. By changing the thickness of the dielectric and varying the frequency, the capacitor becomes very suitable to work over a broad range of frequencies and capacitances. The characteristics including material components and construction, dielectric properties and topography at metal-dielectric interfaces are optimized for a wide frequency band width. Attention is given to frequency dependence of the different material properties. Effects of the material characteristics on electrical performance, including capacitance are examined. Finally, processing and fabrication issues are discussed.

This work presents an efficient isolation method in mixed-signal systems using a novel electromagnetic bandgap (EBG) structure called the alternating impedance EBG (AI-EBG) for isolating sensitive RF/analog circuits from noisy digital circuits. This EBG structure shows excellent isolation by suppressing almost all possible electromagnetic modes in bandgap frequencies. Measurements on a practical mixed-signal system show the feasibility of using this EBG structure to reduce noise coupling between RF/analog circuits and digital circuits, especially where a common power supply is used. To the best of our knowledge, this is the first example of a realistic mixed-signal system employing an EBG-based noise suppression scheme.

A new adaptive algorithm for system identification and adaptive signal processing has been proposed in this paper. The algorithm can be viewed as a combination of the output error type and the equation error type of identifier (i.e. a combination of IIR adaptive filter and FIR adaptive filter in signal processing application). It has been shown both mathematically and by computer simulation that the proposed algorithm can guarantee the globally asymptotical stability under certain regularity conditions on the distribution of signal modes. The transient behaviour of the algorithm is better than conventional output error type of identifier, especially in the case where the system model (signal model) has its eigenvalues on the unit circle. Under certain conditions on the system model (signal model), no moving average (MA) smoothing of the generalized error is needed for the strictly positive realness (SPR) conditions. It gives a reduced bias on the parameter estimate (compare to conventional adaptive FIR filter) when corresponding equation error is a bandlimited process.

The convergence result of least-squares equation error identification algorithm applied to an unstable linear time-invariant system is given. It is shown that in order to obtain a uniform rate of convergence for all the parameter estimates of an ARMA model, an unbounded probing signal input is necessary. However, with the bounded probing signal input, the convergence of AR parameter estimates alone is ensured. Simulation results are included to illustrate the convergence behaviours.

An adaptive linear prediction filter for speech enhancement is implemented in real time on a PC interfaced to an A/D and D/A converter board. A least means squares (LMS) algorithm is employed to update the filter weights where the learning factor is adaptively adjusted to provide faster convergence. The prediction horizon is chosen to be larger than the correlation length of the noise, thereby not restricting the noise to be white. The voice activity and noise segment of the speech waveform are detected by using the energy of the adaptive filter output. This is used to attenuate the noise-only portion. The proposed scheme is evaluated on number of speech samples.

Vector-controlled induction motor drive systems without speed sensors have been widely studied. Speed estimation can be treated as a problem of parameter estimation using the model reference adaptive system (MRAS) theory. In these cases, the convergence and robustness of the system are major issues. We discuss the stability of an MRAS that includes an adaptive pseudo-reduced-order flux observer (APFO) of the IM and is used for estimating the rotor speed and identifying the stator resistance as well. To our knowledge, the stability problem of APFO for a sensorless IM drive has not received any attention in previous research. The stability conditions of APFO are discussed. The rotor speed and stator estimation results are provided to show that the proposed scheme is hyperstable.

One highly integrated mixed-signal testbed has been developed to demonstrate the concept and realization of advanced System-on-a-Package concept. This experimental system, called Intelligent Network Communicator (INC), deals with three different status of the signals, digital, RF and optical, in a single packaging platform. The INC transmits and receives the high-speed digital signal and wireless signal over the embedded optical waveguide channel. After three years of development efforts, the system has been fabricated by utilizing advanced packaging and assembly processes and full functionality has been demonstrated successfully. Before the final test, each of the sub-blocks has been separately developed and tested. The test results clearly show that the developed system performance meets the design goals. The digital block generated up to 3.2 Gbps of data stream, the RF block had less than -1.5dB of insertion loss up to 6 GHz and the optical block achieved 10Gbps throughput over the embedded optical waveguide built on the low-cost organic substrate.

Current board technologies are inherently performance-limited (FR-4) or cost-prohibitive (Al/sub 2/O/sub 3//AlN). New package or board materials with low CTE and high elastic modulus are needed for the next generation of high-performance convergent Microsystems to be able to fabricate ultra high-density wiring without big capture pads and to assemble area-array flip-chips with minimal stress on solder joint or perhaps completely eliminate underfills around the solder joint. A novel manufacturing process has been demonstrated to yield large-area thin carbon-silicon carbide (C-SiC) based composite boards with potentially low cost and desired thermomechanical properties - ultra high modulus, Si-matched CTE and large-area manufacturability. The reliability performance of this material was evaluated with flip chip test vehicle using conventional epoxies and advanced dielectrics such as BCB and PPE. Bumped dies were assembled and liquid-liquid thermal shock tests and Shadow Moire measurements were carried out to assess the solder joint reliability of these boards. In parallel with experiments, numerical models were also developed to analyze warpage, dielectric cracking and solder fatigue failure, and provide design guidelines. Though boards with Si-matched CTE are essential to lower the solder joint strains, they result in a high CTE mismatch between the build-up dielectrics and board, generating higher stresses in the dielectric which could lead to cracking. In this work, we show that dielectric cracking can be minimized with stiffer boards and thinner build-up layers. Based on the results, it can be inferred that high board stiffness and low CTE (/spl sim/3-4 ppm//spl deg/C) are needed to ensure reliability of high-density packages without the use of underfill. The test vehicle evaluation supported by modeling results indicate that the novel low-cost large-area processable ceramic matrix composite (C-SiC) has potential to be a promising candidate substrate material for next-generation microsystems.

Twin demands on Pb-free solder interconnects for process compatibility and the decreasing pitch size of the bumps introduce enormous stresses on the joint quality of the solder-Under Bump Metallurgy (UBM) and the solder - pad interface. The joint quality is governed by the IMC formation. The objective of this study is two fold. First is to compare the near eutectic Sn-3.8Ag-0.7Cu lead-free solder with standard eutectic (63Sn-37Pb) lead based solder for IMC formation on the UBM-solder pad and solder-substrate pad interface. And second is to investigate the lead free flip chip solder joint for different surface pad finishes. The paper focuses on 4 different surface pad finishes namely, Electroless Nickel Immersion gold (ENIG), Organic Solderability Preservative (OSP), Direct Immersion Gold (Cu/Au), and Direct Immersion Silver (Cu/Ag) and its flip chip reliability. The UBM studied has Al as the adhesion layer, Ni (7% V) as the diffusion barrier layer and Cu as the wetting layer. Die shear, ball shear and electrical continuity are carried to evaluate the solder bump integrity and the failure mode. IMC formation at the interfaces is observed with optical and Electron microscopy. The results obtained in this paper will provide understanding of the fracture, shear strengths and electrical resistance of the solder bumps in flip-chips. Results correlating IMC growth and morphology for different thermal aging hours with electrical resistance and ball shear would be discussed in the paper.

The Packaging Research Center has been developing next generation system-on-a-package (SOP) technology with digital, RF, and optical system integration on a single package. SOP aims to utilize the best of on-chip SOC integration and package integration to achieve the highest system performance at the lowest cost. The micro-miniaturized multi-functional SOP package is highly integrated and fabricated on large area substrates similar to the wafer-to-IC concept. In addition to novel mixed signal design methodologies, SOP research at PRC is targeted at developing enabling technologies for package level integration including ultra-high density wiring, embedded passive components, embedded optical interconnects, wafer level packaging and fine pitch assembly. Several of these enabling technologies have been recently integrated into the first successful system level demonstration of SOP technology using the intelligent network communicator (INC) testbed. This paper reports on the latest INC and SOP testbed results at the PRC and provides an insight into the future SOP integration strategy for convergent microsystems. The focus of this paper is on integration of materials, processes and structures in a single package substrate for system-on-a-package (SOP) implementation.

A new segmentation method of multi-component noisy signals using wavelet transform is proposed, when the signal components are closely spaced and the time interval between adjacent signal components are unknown. It is shown that Morlet wavelet transform is useful for segmenting a noisy signal, when the signal components are closely spaced. The segmentation problem is formulated using the paradigm of estimating the locations and durations of noisy narrow gaps of the input noisy signals. A wavelet scale sequence comprising of the highest absolute scales for each time instant is employed as test statistics for segmentation. A number of selected local maxima obtained from the wavelet scale sequence correspond to the position of the noisy gaps. Finally, windowed approximate entropy is calculated for the masked noisy signal to estimate the locations and durations of the narrow noisy gap as well as the noisy segments. The proposed scheme is evaluated on simulated examples.

Various speech enhancement schemes are analyzed in terms of the conflicting real time requirements of computational delay, robustness and accuracy. A spectral subtraction scheme is found to be implementable in real time using the available digital processing board. The tasks such as the computation of the spectral contents of the speech plus noise and the noise, smoothing the estimates, and speech enhancement filtering are implemented in the frequency domain using FFT in view of the computational speed and robustness. A variant of a spectral subtraction scheme is implemented in real time on a DSP board and its performance is evaluated.

Current board technologies fail to meet the stringent requirements of the future high-performance microsystems either due to performance (organic) or cost (AlN, Al/sub 2/O/sub 3/) related issues. New board materials that can support multiple layers of ultra high-density wiring with small capture pads and low-cost flip chip assembly at finer pitch with reliability are needed. FE modeling was done to lay the guidelines for the material property requirements and to study the solder joint fatigue, board warpage and dielectric stresses during thermal cycling. Modeling results showed that though the Si-matched CTE of board results in low solder joint fatigue, it increases the stresses in the dielectric. In addition, apart from dielectric CTE and strength, the thickness of the build-up dielectric layer is also a vital factor in deciding dielectric reliability. A novel process to manufacture C-SiC composite boards was shown to produce large area thin panels at low cost with the required properties of Si-matched CTE and ultra-high modulus. Test vehicles were fabricated with different dielectrics by assembling flip-chips without underfill on these boards followed by reliability testing using liquid-liquid thermal shock tests and in-situ warpage measurements. Modeling and experimental results indicate that boards with low CTE (2-3 ppm//spl deg/C) and high stiffness are key factors to control the solder joint and dielectric reliability. Dielectric cracking was minimized with advanced low CTE, high-strength materials and thinner build-up layers. The C-SiC board with the novel manufacturing process meets the property requirements for high density packaging substrates in addition to being low cost and large area processable making it a promising board substrate material for future high-performance microsystems.

The trend in both the computing and electronic industries is to reduce the size of the electronic components while increasing both their performance and capability. Innovative thermal management schemes are needed in order to reduce the impact of the thermal loads. Most of the current electronic components are packaged in a plastic or ceramic container to provide chip and electrical lead protection. Both of these packaging materials have low thermal conductivity, making heat removal from the chip difficult. Thus, heat transfer through the off-chip metal interconnects offers an additional heat removal path. Ball grid array (BGA) interconnects provide an efficient means to connecting packaged high performance chips to printed circuit boards (PCB). As area array bump density increases, reducing Joule heating and electromigration will play an important role in chip and interconnect reliability. Among the many types of interconnects, solder balls offer an efficient means of connecting a chip or package to a PCB. Direct cooling of the solder balls is a new approach to removing heat from packaged chips. Jet impingement presents a unique solution for cooling the solder balls. Thermal and computational fluid dynamic (CFD) modeling of a plastic ball grid array (PBGA) package has demonstrated a significant decrease in temperature across the chip, package, and solder balls, when using jet impingement cooling.

The Design Build and Operate (DBO) course at PRC (the Packaging Research Center at Georgia Institute of Technology) provides hands-on experience in SOP (System on a Package) packaging for undergraduate and graduate students. DBO consists of a sequence of two courses: Design Build Operate I (DBO I), which focuses on substrate fabrication and Design Build Operate II (DBO II), which focuses on module assembly. This paper presents an overview of the second component of the sequence, the DBO II (module) course. A detailed presentation of the experiments conducted and their results are presented. A user friendly software which helps students to conduct virtual designs and fabrication processes is also presented.

Due to the complex nature of clinical manifestations, employing a single technique to detect different abnormalities may not necessarily yield accurate results. It is essential to develop an intelligent fusion approach to combine multiple data to arrive at an accurate final decision for clinical diagnosis. A new fusion approach is developed as an intelligent clinical data analysis technique for the cancer detection. It integrates decision-level-based fusion and multi-feature-based fusion. The results of new cases corresponding to accurate detection are fed back to refine appropriate parameters in a learning loop. The learning process is designed to estimate and reduce the error in the correct decision-making. The experimental results suggest that the newly proposed fusion approach has a better performance compared with some individual techniques, thus leading to enhanced accuracy of abnormality detection in cancer screening.

This paper proposes an efficient method of segmenting noisy audio signals using a linear binary Walsh transform, when the signal components are closely spaced and the time intervals between adjacent signal components are unknown. It is shown that the Walsh transform is appropriate for segmenting a noisy waveform. A subset of the Walsh functions is chosen to cover principally the noise subspace such that the resulting linear combination of the selected basis functions captures the features that can discriminate between signal and noise. In the absence of a priori information about the signal and noise statistics, the proposed scheme is based on the linear combination of those basis functions which must be able to identify the adjacent signal components. It is not necessary that the basis functions reconstruct the noise-free versions of the signal components. The only restriction is that the segment length should be some integer power of 2 for the most accurate segmentation. The simulation examples show effectiveness in the segmentation of narrowly separated, noisy signals by using our simple segmentation method.

A real-time integrated hardware/software approach is proposed to implement fault-tolerant control for a class of control systems that can be mathematically modeled. The detection, isolation, and accommodation (DIA) of computer faults is implemented using triple modular redundancy, and DIA of the process faults uses analytical redundancy based on an observer. The system software architecture and the design of a knowledge base for process faults are described. >

An analysis is presented of the output and parameter estimation errors of RLS (recursive least-squares) adaptive filters when they are applied to exponentially damped, undamped, and exponentially growing sinusoids. The authors show that the RLS adaptive filter guarantees zero output error for all three types of sinusoids. However, for undamped and exponentially growing sinusoids, unbiased parameter estimates of the signal models are also provided. A new method of estimating the parameters of an exponentially damped sinusoid is proposed. The method can accurately estimate the parameter values of the signal model provided the lower bound of the exponentially damping coefficient is known a priori and the measurement noise is negligible. The results are further verified through digital computer simulations. >

The general structure and design specifications of a real-time knowledge-based controller are proposed. The knowledge-based controller has been designed, implemented, and evaluated for a hydraulic turbine generator control system. The controller has been designed to maintain the system stability and performance not only during normal plant operation, but also in the presence of contingencies, such as system component failures and severe external load disturbances. The knowledge-base and the inference engine software are written in Prolog, and the information preprocessing and control algorithms are written in C language. The performance of the knowledge-based controller has been evaluated by simulating the hydraulic turbine generator system on an analog computer with the knowledge-based controller implemented in real time on a microcomputer. >

A fault detection and isolation (FDI) scheme is proposed which combines the detection filter, parity equation and parameter estimation approaches with a view to obtain a faster and a more reliable fault detection and isolation. A fault is defined to be a deviation in a parameter (termed physical parameter) associated with a device whose fault is to be detected and isolated. The feature vector which is a vector formed of the coefficients of the system transfer function is assumed to be multilinear in the physical parameters. The influence of these physical parameters on the feature vector is captured in a vector, termed the influence vector. The feature vector and the influence vector are estimated off-line, and are used to construct a detection filter and a parity equation. In real time, a fault is detected from the statistical analysis of the residual generated by the detection filter, and it is isolated using the parity equation which relates the residual, the input and the physical parameter deviations. Thanks to the multilinear relationship, the off-line parameter estimation need not be repeated every time a physical parameter varies: the initial estimates are merely updated. The proposed scheme is evaluated on a number of simulated examples.

The conventional method of fault diagnosis is to use static and dynamic models of the process. This paper starts with a brief introduction to the parametric-model-estimation approach, followed by a detailed discussion on how efficiently the idea of an influence matrix may relieve the need for the knowledge of an explicit relation between the estimated model parameters and faulty physical components. Then the nonparametric-model-estimation approach is proposed as a proper replacement to lessen the huge computational time implied by the recursive identification process. The proposed method is evaluated both on simulated examples and on an actual control system.

A real-time control system status monitoring scheme is proposed. The scheme consists of linear predictive coding algorithms, a knowledge base, and an inference engine. The scheme is capable of providing real-time information on the control system in terms of stability, settling time, percentage overshoot, and integral squared error. The proposed scheme has been applied to a DC motor position control system under various normal/abnormal operating conditions. The proposed scheme has successfully provided useful information on the overall system stability and performance. >

Estimation of unknown frequency, extraction of narrowband signals buried under noise and periodic interference are accomplished by employing existing techniques. However, the authors propose an artificial neural net based scheme together with pattern classification algorithm for narrowband signal extraction. A three layer feedforward net is trained with three different algorithms namely backpropagation, Cauchy's algorithm with Boltzmann's probability distribution feature and the combined backpropagation-Cauchy's algorithm. A constrained tangent hyperbolic function is used to activate individual neurons. Computer simulation is carried out with inadequate data to reinforce the idea of the net's generalization capability. The robustness of the proposed scheme is claimed with the results obtained by making 25% links faulty between the layers. Performance comparison of the three algorithms is made and the superiority of the combined backpropagation-Cauchy's algorithm is established over the other two algorithms. Simulation results for a wide variety of cases are presented for better appraisal. >

A real-time expert system, consisting of a knowledge base, an inference engine, and a linear predictive coding algorithm (LPCA), is proposed for monitoring the status of the control system. The ARMA (autoregressive moving average) model estimates are used to compute the performance and stability measures, predict overload, and detect faults. The control system is asserted to be in (1) normal state if the performance and stability measures are within the threshold limits, (2) alert state if the performance and/or stability measures violate the limits, and (3) emergency state if the error signal is unbounded. The LPCA is broken into a number of tasks. The expert system controls the execution of these tasks and validates the assertions using heuristic, contextual, and control-theoretic reasoning. The status information is displayed in the order of decreasing importance such that essentials are known earlier with the complete picture emerging later. The proposed scheme is implemented on a commercially available expert system shell. >

The performance of real-time intelligent reconfigurable control systems is directly related to their performance monitoring and the fault detection algorithms used. Issues in selecting suitable locations for real-time information collection are considered. Every accessible location in a typical feedback control system is analyzed and the most adequate variables in the control loop are determined for the performance monitoring and fault detection algorithms to collect the information. The entire analysis is based on system sensitivity theory. The sensitivity functions in a typical multiloop feedback control system are defined in terms of system component variations, and the optimal variables are determined on the basis of the criterion of maximal sensitivity. A three loop single input and single output system is initially used in the analysis. Results are then extended to more general systems. >

A reliable scheme based on a linear predictive coding algorithm (LPCA) is proposed for monitoring in real-time the status of the control system under both steady and transient states. An ARMA model of the error signal of the closed-loop control system is estimated using LPCA. In the transient state, using the ARMA coefficients, integral squared-error and pole-location-based unit step response measures and stability robustness using Kharitonov's theorem are computed, and a fault is predicted. In the steady state, a deviation from the nominal model is detected from the ARMA-based power spectral estimates. The monitor displays information in decreasing order of importance and with increasing amounts of computation such that the essentials are known in the shortest possible time, with complete details emerging later. The proposed scheme is evaluated on a simulated complex multivariable control system. >

An integrated approach is proposed for the design and implementation of autonomous control systems that unifies the design problems of failure detection, isolation and accommodation. The system consists of an interconnection of components that are subject to failure. A component fault occurs when one or more of the component parameters change. The paper describes a scheme for: 1) detecting when one of the components fail, 2) identifying the failed component and 3) under certain conditions accommodating the failure by controller tuning. The problem is compounded by the restriction that only the inputs and outputs of a limited number of components can be measured. Therefore it is not possible to directly determine the diagnostic parameters using parameter identification theory. The detection problem is solved using an observer. The isolation (FDI) problem is posed as a pattern classification problem that is solved using the Bayes decision strategy and a geometric analysis. Once a fault is isolated we show how a linear feedback controller may be tuned to minimize an integral squared error performance criterion using a gradient-based algorithm.

This paper presents a method to determine the presence or absence of certain polygons in a 2D image. If present the algorithm will determine the position and orientation of the polygon. The polygon may be scaled, rotated or "fipped" in the 2D image. Subpixel accuracy is possible with execution speeds suitable for real-time applications.

An artificial intelligence (AI) based robust algorithm to extract, a posteriori, the rational signal model from a noisy measurement, with little a priori information, is proposed. The spectrum and the statistics of the signal and of the corrupting noise are assumed unknown except that the signal is assumed to have a rational spectrum. An algorithm based on both system and signal theory, and on heuristics is derived to select a set of frequencies where the SNR is high from a given measurement spectrum. A relative weighting which indicates the importance of the measurement at each frequency is also obtained. An estimate of the signal model is obtained from the best weighted least squares fit to the measurement spectrum at the selected frequencies. The proposed scheme has applications to control and signal processing, and is evaluated with a number of simulated examples, and on a physical system. The results are compared with conventional adaptive filter techniques.

A novel modelling, and identification scheme for real time fault diagnosis is proposed. A diagnostic model for the system is developed which consists of a) an input-output difference equation model expressed explicitly in terms of the feature vector (which is a vector formed of the coefficients of a transfer function model) and data vector (vector formed present and past inputs and the past outputs) and b) a matrix, termed multilinearity matrix, relating the diagnostic parameter (a vector formed of the coefficients of a transfer function of a functional unit) and the feature vector. A scheme to identify the diagnostic model in the face of model uncertainties, the noise and the disturbances is proposed.

A real time scheme to track the frequencies of multiple sinusoids in a noisy waveform is proposed. The problem of estimating the frequencies is formulated as that of adapting the parameters of a notch filter so that the mean-squared prediction error is minimized. To overcome the computational complexity, an improved adaptation scheme is proposed based on estimating one frequency at a time and is compared with other state-of-the art schemes. A prototype of the proposed scheme is constructed using a commercially available digital signal processing (DSP) board.