Bartlomiej Beliczynski

Approximation of FIR by IIR digital filters: an algorithm based on balanced model reduction

An algorithm for the approximation of finite impulse response (FIR) filters by infinite impulse response (IIR) filters is presented. The algorithm is based on a concept of balanced model reduction. The matrix inversions normally associated with this procedure are notoriously error prone due to ill conditioning of the special matrix forms required. This difficulty is circumvented here by directly formulating a reduced state-space system description which is input/output equivalent to the system that would more conventionally be obtained following the explicit step of constructing an (interim) balanced realization. Examples of FIR by IIR filter approximations are included. >

Induction motor speed estimation: neural versus phenomenological model approach

Abstract Two approaches to speed estimation of an induction motor in the drive system, utilizing only easy measurable electrical signals, are presented, discussed and compared. One is based on phenomenological model of the motor and least squares solution of an overdetermined set of linear equations. Another utilizes nonlinear system modeling via neural network. These two models are complementarily treated in the paper. The phenomenological model is simple and easily interpretable, but it is very sensitive to parameter changes. The neural model requires that input variables are preprocessed. We demonstrate that the input variables selection, sampling time and neural architecture are interrelated and that time instances of sinusoidal signals such as stator voltages and currents, without preprocessing, are not convenient for speed estimation. It is proved that for the most commonly used tapped delay neural architecture, the speed estimation cannot be improved above certain level of accuracy through sampling time selection, or enlarging number of delays of the input signals. Preprocessing of the input variables may change the situation. The information obtained from phenomenological model is used to select and preprocess input variables for the neural model. Simulation examples demonstrating both approaches and very good efficiency and robustness of the neural model are included.

Critical comparison of Hankel-norm optimal approximation and balanced model truncation algorithms as vehicles for FIR-to-IIR filter order reduction

In this paper two powerful methods of FIR-to-IIR filter order reduction are presented and compared: truncation of balanced model and Hankel-norm optimal approximation. Conclusions based on very many filter order reduction runs indicate that balanced model truncation is often preferrable.<<ETX>>

Adaptive and Natural Computing Algorithms

Neural Networks.- Evolution of Multi-class Single Layer Perceptron.- Estimates of Approximation Rates by Gaussian Radial-Basis Functions.- Least Mean Square vs. Outer Bounding Ellipsoid Algorithm in Confidence Estimation of the GMDH Neural Networks.- On Feature Extraction Capabilities of Fast Orthogonal Neural Networks.- Neural Computations by Asymmetric Networks with Nonlinearities.- Properties of the Hermite Activation Functions in a Neural Approximation Scheme.- Study of the Influence of Noise in the Values of a Median Associative Memory.- Impact of Learning on the Structural Properties of Neural Networks.- Learning Using a Self-building Associative Frequent Network.- Proposal of a New Conception of an Elastic Neural Network and Its Application to the Solution of a Two-Dimensional Travelling Salesman Problem.- Robust Stability Analysis for Delayed BAM Neural Networks.- A Study into the Improvement of Binary Hopfield Networks for Map Coloring.- Automatic Diagnosis of the Footprint Pathologies Based on Neural Networks.- Mining Data from a Metallurgical Process by a Novel Neural Network Pruning Method.- Dynamic Ridge Polynomial Neural Networks in Exchange Rates Time Series Forecasting.- Neural Systems for Short-Term Forecasting of Electric Power Load.- Jet Engine Turbine and Compressor Characteristics Approximation by Means of Artificial Neural Networks.- Speech Enhancement System Based on Auditory System and Time-Delay Neural Network.- Recognition of Patterns Without Feature Extraction by GRNN.- Real-Time String Filtering of Large Databases Implemented Via a Combination of Artificial Neural Networks.- Parallel Realizations of the SAMANN Algorithm.- A POD-Based Center Selection for RBF Neural Network in Time Series Prediction Problems.- Support Vector Machines.- Support, Relevance and Spectral Learning for Time Series.- Support Vector Machine Detection of Peer-to-Peer Traffic in High-Performance Routers with Packet Sampling.- Improving SVM Performance Using a Linear Combination of Kernels.- Boosting RVM Classifiers for Large Data Sets.- Multi-class Support Vector Machines Based on Arranged Decision Graphs and Particle Swarm Optimization for Model Selection.- Applying Dynamic Fuzzy Model in Combination with Support Vector Machine to Explore Stock Market Dynamism.- Predicting Mechanical Properties of Rubber Compounds with Neural Networks and Support Vector Machines.- An Evolutionary Programming Based SVM Ensemble Model for Corporate Failure Prediction.- Biomedical Signal and Image Processing.- Novel Multi-layer Non-negative Tensor Factorization with Sparsity Constraints.- A Real-Time Adaptive Wavelet Transform-Based QRS Complex Detector.- Nucleus Classification and Recognition of Uterine Cervical Pap-Smears Using FCM Clustering Algorithm.- Rib Suppression for Enhancing Frontal Chest Radiographs Using Independent Component Analysis.- A Novel Hand-Based Personal Identification Approach.- White Blood Cell Automatic Counting System Based on Support Vector Machine.- Kernels for Chemical Compounds in Biological Screening.- A Hybrid Automated Detection System Based on Least Square Support Vector Machine Classifier and k-NN Based Weighted Pre-processing for Diagnosing of Macular Disease.- Analysis of Microscopic Mast Cell Images Based on Network of Synchronised Oscillators.- Detection of Gene Expressions in Microarrays by Applying Iteratively Elastic Neural Net.- A New Feature Selection Method for Improving the Precision of Diagnosing Abnormal Protein Sequences by Support Vector Machine and Vectorization Method.- Epileptic Seizure Prediction Using Lyapunov Exponents and Support Vector Machine.- Classification of Pathological and Normal Voice Based on Linear Discriminant Analysis.- Efficient 1D and 2D Daubechies Wavelet Transforms with Application to Signal Processing.- A Branch and Bound Algorithm for Matching Protein Structures.- Biometrics.- Multimodal Hand-Palm Biometrics.- A Study on Iris Feature Watermarking on Face Data.- Keystroke Dynamics for Biometrics Identification.- Protecting Secret Keys with Fuzzy Fingerprint Vault Based on a 3D Geometric Hash Table.- Face Recognition Based on Near-Infrared Light Using Mobile Phone.- NEU-FACES: A Neural Network-Based Face Image Analysis System.- GA-Based Iris/Sclera Boundary Detection for Biometric Iris Identification.- Neural Network Based Recognition by Using Genetic Algorithm for Feature Selection of Enhanced Fingerprints.- Computer Vision.- Why Automatic Understanding?.- Automatic Target Recognition in SAR Images Based on a SVM Classification Scheme.- Adaptive Mosaicing: Principle and Application to the Mosaicing of Large Image Data Sets.- Circular Road Signs Recognition with Affine Moment Invariants and the Probabilistic Neural Classifier.- A Context-Driven Bayesian Classification Method for Eye Location.- Computer-Aided Vision System for Surface Blemish Detection of LED Chips.- Detection of Various Defects in TFT-LCD Polarizing Film.- Dimensionality Problem in the Visualization of Correlation-Based Data.- A Segmentation Method for Digital Images Based on Cluster Analysis.- Active Shape Models and Evolution Strategies to Automatic Face Morphing.- Recognition of Shipping Container Identifiers Using ART2-Based Quantization and a Refined RBF Network.- A Local-Information-Based Blind Image Restoration Algorithm Using a MLP.- Reflective Symmetry Detection Based on Parallel Projection.- Detail-Preserving Regularization Based Removal of Impulse Noise from Highly Corrupted Images.- Fast Algorithm for Order Independent Binary Homotopic Thinning.- A Perturbation Suppressing Segmentation Technique Based on Adaptive Diffusion.- Weighted Order Statistic Filters for Pattern Detection.- Real-Time Image Segmentation for Visual Servoing.- Control and Robotics.- A Neural Framework for Robot Motor Learning Based on Memory Consolidation.- Progressive Optimisation of Organised Colonies of Ants for Robot Navigation: An Inspiration from Nature.- An Algorithm for Selecting a Group Leader in Mobile Robots Realized by Mobile Ad Hoc Networks and Object Entropy.- Robot Path Planning in Kernel Space.- A Path Finding Via VRML and VISION Overlay for Autonomous Robot.- Neural Network Control for Visual Guidance System of Mobile Robot.- Cone-Realizations of Discrete-Time Systems with Delays.- Global Stability of Neural Networks with Time-Varying Delays.- A Sensorless Initial Rotor Position Sensing Using Neural Network for Direct Torque Controlled Permanent Magnet Synchronous Motor Drive.- Postural Control of Two-Stage Inverted Pendulum Using Reinforcement Learning and Self-organizing Map.- Neural Network Mapping of Magnet Based Position Sensing System for Autonomous Robotic Vehicle.- Application of Fuzzy Integral Control for Output Regulation of Asymmetric Half-Bridge DC/DC Converter.- Obtaining an Optimum PID Controller Via Adaptive Tabu Search.

Induction motor speed estimation: limitations of certain neural schemes based on time instances of periodic signals

It is proved that certain induction motor speed estimation schemes based on tapped delay neural architecture have important limitations. If teaching data are collected as instances of periodic signals (such as voltages and currents) then accuracy of speed estimation cannot be improved above a certain level through sampling time selection, or enlarging number of delays of the input signals. Consequently frequency bandwith for which estimation can be used is also limited by accuracy of the signals amplitude measurement. For a fixed accuracy of measurement, better accuracy of speed dynamics estimation may be achieved at the expense of narrowing frequency band.

Incremental approximation by one-hidden-layer neural networks: discrete functions rapprochement

We characterize incremental approximation of discrete functions by using a one-hidden-layer neural network. The functions to be approximated are represented by a set of input/output pairs. The network consists of input, hidden and linear output layers. In a series of steps we add units to the hidden layer. In each iteration, parameters of one new hidden unit are determined and also all output weights are recalculated. We examine conditions on convergence and its rate and propose a simple algorithm of one unit parameters tuning. This algorithm uses almost exclusively analytical formulas without involving any searching method.

A method of multivariable Hermite basis function approximation

A method of multivariable (multivariate) Hermite function based approximation is presented and discussed. The multivariable basis is constructed as a product of one-variable Hermite functions with adjustable scaling parameters. Thanks basis orthonormality, the approximated function expansion coefficients are calculated by using explicit, non-search formulae. The scaling parameters are determined via a search algorithm. Initially, an excessive number of functions in the basis is calculated, then a simple pruning method is applied. Only those are taken which contribute the most to error decrease, down to a desired level. The method ensures a very good generalization property. This claim is supported by both theoretical considerations and working examples.

Properties of the Hermite Activation Functions in a Neural Approximation Scheme

The main advantage to use Hermite functions as activation functions is that they offer a chance to control high frequency components in the approximation scheme. We prove that each subsequent Hermite function extends frequency bandwidth of the approximator within limited range of well concentrated energy. By introducing a scalling parameter we may control that bandwidth.

Approximation of Functions by Multivariable Hermite Basis: A Hybrid Method

In this paper an approximation of multivariable functions by Hermite basis is presented and discussed. Considered here basis is constructed as a product of one-variable Hermite functions with adjustable scaling parameters. The approximation is calculated via hybrid method, the expansion coefficients by using an explicit, non-search formulae, and scaling parameters are determined via a search algorithm. A set of excessive number of Hermite functions is initially calculated. To constitute the approximation basis only those functions are taken which ensure the fastest error decrease down to a desired level. Working examples are presented, demonstrating a very good generalization property of this method.

On input discretisation processes for tapped delay neural architecture

For tapped delay neural architecture impact of discretisation process i.e. sampling and amplitude measurement of the input continuous signals is analysed. By using relative matrix rank and indistinguishable signals concepts we derrive upper and lower bounds of sampling time and input signal frequencies.