Raymond H. Kwong

A variable step size LMS algorithm

A least-mean-square (LMS) adaptive filter with a variable step size is introduced. The step size increases or decreases as the mean-square error increases or decreases, allowing the adaptive filter to track changes in the system as well as produce a small steady state error. The convergence and steady-state behavior of the algorithm are analyzed. The results reduce to well-known results when specialized to the constant-step-size case. Simulation results are presented to support the analysis and to compare the performance of the algorithm with the usual LMS algorithm and another variable-step-size algorithm. They show that its performance compares favorably with these existing algorithms. >

Analysis and design of secure watermark-based authentication systems

This paper focuses on a coding approach for effective analysis and design of secure watermark-based multimedia authentication systems. We provide a design framework for semi-fragile watermark-based authentication such that both objectives of robustness and fragility are effectively controlled and achieved. Robustness and fragility are characterized as two types of authentication errors. The authentication embedding and verification structures of the semi-fragile schemes are derived and implemented using lattice codes to minimize these errors. Based on the specific security requirements of authentication, cryptographic techniques are incorporated to design a secure authentication code structure. Using nested lattice codes, a new approach, called MSB-LSB decomposition, is proposed which we show to be more secure than previous methods. Tradeoffs between authentication distortion and implementation efficiency of the secure authentication code are also investigated. Simulations of semi-fragile authentication methods on real images demonstrate the effectiveness of the MSB-LSB approach in simultaneously achieving security, robustness, and fragility objectives.

Fault diagnosis in discrete-event systems: incorporating timing information

A framework is introduced for fault diagnosis in timed discrete-event systems. In this approach, the required estimates for system condition are updated only when the output changes or when deadlines associated with output changes expire. Thus updates at every clock tick are not required. This in many cases results in reduction in online computing requirements and in the size of the diagnosis system, at the expense of more offline design calculations. The issue of failure diagnosability is also discussed.

Analysis and design of watermarking algorithms for improved resistance to compression

We study the performance of robust digital watermarking approaches in the presence of lossy compression by introducing practical analysis methodologies. Correlation expressions between the embedded watermark and the extracted watermark are derived to determine the optimal watermarking domain to maximize data hiding rates for spread spectrum and quantization watermarking. It is determined both theoretically and through simulations that the embedding strategy, in addition to the transform used for lossy compression, dictate the optimal transform for watermarking. Through analytic comparisons, we develop a novel hybrid watermarking algorithm that exploits the best of both approaches for greater resilience to JPEG compression.

Fault diagnosis in timed discrete-event systems

A framework is introduced for passive online fault diagnosis in timed discrete-event systems (TDES). It extends the previous work of the authors (1999) on a state-based approach to fault diagnosis by incorporating timing information. This enhances the accuracy of diagnosis. In this methodology instead of directly extending the existing framework to TDES, an alternative approach is taken which, in many cases, leads to significant reduction in online computing requirements and the size of the diagnoser at the expense of more off-line design calculations.

Detection of feedback in the central nervous system using system identification techniques

An analysis method to detect the presence of feedback between biological signals, particularly those associated with the central nervous system, is presented. The technique is based on recent results in the system identification literature involving the concept of a feedback free process. It may be applied to volume conducted signals such as EEG and EMG, as well as to neuronal spike trains through the use of a data transformation procedure. The utility of the technique is then demonstrated in a study of the relationship between Parkinsonian tremor and certain tremor cells found in the thalamus of Parkinsonian patients, using data collected during thalamotomies. The results obtained suggest that feedback mechanisms may be an important factor contributing to Parkinsonian tremor.

The choice of watermark domain in the presence of compression

We determine the watermark domain that maximizes data hiding capacity. We focus on the situation in which the watermarked signal undergoes lossy compression involving quantization in a specified compression domain. A novel linear model for the process of quantization is proposed which leads to analytical results estimating the data hiding capacity for various watermarking domains. Using this framework we predict appropriate transforms for robust spread spectrum data hiding in the face of JPEG compression. Simulation results verify our theoretical observations. We find that a repetition code used in conjunction with spread spectrum watermarking in a different domain than employed for compression improves data hiding capacity.

Spectrum estimation using multirate observations

In this paper, we are interested in estimating the power spectral density of a stationary random signal x(n) when the signal itself is not available but some low-resolution measurements derived from it are observed. We consider a model where x(n) is being measured using a set of linear multirate sensors. Each sensor outputs a measurement signal v/sub i/(n) whose sampling rate is only a fraction of the sampling rate assumed for the original signal. Based on this model, we pose the following problem: Given certain autocorrelation coefficients of the observable signals v/sub i/(n), estimate the power spectral density of the original signal x(n). It turns out that this problem is ill-posed. We suggest to resolve this issue by using the principle of maximum entropy (ME). We address technical difficulties associated with the ME solution and then devise a practical algorithm for its approximate computation. We demonstrate the viability of this algorithm through simulation examples.

Fault diagnosis in discrete-event systems: framework and model reduction

A state-based approach for online passive fault diagnosis in systems modelled as finite-state automata is presented. In this framework, the system and the diagnoser (the fault detection system) do not have to be initialized at the same time. Furthermore, no information about the state or even the condition (failure status) of the system before the initiation of diagnosis is required. The design of the fault detection system, in the worst case, has exponential time complexity. A model reduction scheme with polynomial time complexity is introduced to reduce the computational complexity of the design.

Fault Diagnosis in Discrete-Event Systems: Incomplete Models and Learning

Most model-based approaches to fault diagnosis of discrete-event systems (DESs) require a complete and accurate model of the system to be diagnosed. However, the discrete-event model may have arisen from abstraction and simplification of a continuous time system or through model building from input-output data. As such, it may not capture the dynamic behavior of the system completely. In this paper, we address the problem of diagnosing faults, given an incomplete model of the discrete-event system. When the model is incomplete, discrepancies will arise between the actual output and the output predicted by the model. We introduce learning into the diagnoser construction by forming hypotheses that explain these discrepancies. We view the process of generating and evaluating hypotheses about the model of the system as an instance of the set-cover problem, which we formalize using parsimonious covering theory. We describe in detail the construction of the learning diagnoser, which not only performs fault diagnosis but also attempts to learn the missing model information. If the model is complete, the learning diagnoser reduces to the standard state-based diagnoser. Examples are provided to illustrate how learning and diagnosis can be simultaneously achieved through the learning diagnoser.