نشریه مهندسی برق
سال چهلم شماره 1 (پیاپی 59، بهار و تابستان 1389)

In this paper، a new topology based on matrix converter (MC) is proposed for dynamic voltage restorer (DVR). Matrix converters convert ac to ac with no dc link. So، using this type of converters is an acceptable way to eliminate dc link in conventional DVRs. Three independent three-phase to single-phase MCs with 6 or 8 bi-directional power switches are used in the proposed topology. The range of the DVR’s correct operation varies selecting 6 or 8 number of bi-directional power switches in matrix converters. As a result، the proposed DVR structure can be used to restore load voltage in both balanced and unbalanced voltage sag and swell conditions. Due to the fact that the proposed topology is based on matrix converters and considering that the matrix converters do not have dc link، the speed of the proposed DVR increases and in return decreases thesize of DVR، due to the dc link elimination. The proposed DVR can be presented as a package. Also، one of the most important capabilities of the proposed DVR is the voltage restoring in extremely distorted networks. It is worth noting that the proposed topology would not face any problem in long time compensation due to the fact that it provides the required energy directly through the grid. The experimental results as well as simulation results in PSCAD/EMTDC environment show the capabilities of the proposed topology.

This paper addresses the estimating of surface orientation using a set of texels. Reducing dependency to pattern، simplification of process procedure and improving accuracy are key points of the proposed method. Proposed method consists of two parts. In the first part، it is assumed that scene texels are located in rows and columns. In this part، a different approach from that used in previous shape from texture algorithms is used. Two new angles α and β are defined and thus more simple equations are obtained to estimate the normal vector of surface. The advantage of the proposed method is that the geometry of texels is not important and only repetitive property of texels is used. In addition، to improve the accuracy، texels considered as an array. In the second part، by using theory and results of the first part، laser pattern projection is proposed. It is necessary to emphasis that the assumptions of the first part، i. e. columnar and in row texels، are easily achievable because texels are projected by laser sources that are tunable in position and direction.

In this paper an approach based on hidden Markov model for recognizing online Farsi characters is presented. At first by obtaining the number of parts of a written character and recognizing its delayed strokes، the number of candidates is decreased and then the body and delayed strokes of unrecognized character are preprocessed. The stage of preprocessing is consisting of size normalization and resampling. Thus the recognition process will be robust to transition and scaling and the extracted features will be done more precisely. The extracted features are both local and structural features. The local features are consisting of the angles between the fitted vectors to some important points of unrecognized character and the structural features are consisting of cusp، left-hump and right-hump points. The training process of models is done by Baum-Welch algorithm with a post process on it. Using the mentioned stages has the advantage of doing the recognition process in an unconstrained and writer independent manner. The obtained results show the 97. 22% precision in training and 94. 9% precision in testing experiments.

Blind source separation (BSS) is the technique that anyone can separate the original signals or latent data from their mixtures without any knowledge about the mixing process، but using some statistical properties of latent or original source signals. Independent component analysis is a statistical method expressed as a set of multidimensional observations that are combinations of unknown variables. These underlying unobserved variables are called sources and they are assumed to be statistically independent with respect to each other. In this paper we will use the nonlinear autcorrelation function as an object function to separate the source signals from the mixing signals. Maximization of this object function using the LMS algorithm will be obtained the coefficients of a linear filter which separate the source signals. To calculate the performance of the proposed algorithm، two parameters of Performance Index (PI) and Signal to Interference Ratio (SIR) will be used. To test the proposed algorithm، we will use Inovation Gaussian signals، Speech signals and ECG signals. It will be shown that the proposed algorithm gives better results than the other methods such as Newton method that has been proposed by Shi.

Gabor filters are Gaussian modulated wavelets that due to their ability to extract features in different directions and frequencies in an image are used widely in different applications. However، because of their infinite support، an exact realization is not practically possible and an approximated realization results in losing performance. In this paper، a new modulated wavelet with compact support and close to Gabor wavelet is proposed. Due to its compact support، an exact and fast implementation is possible. To evaluate the performance of the proposed filters a comparative study between them and Gabor filters in two classification application containing semantic image classification and Farsi font recognition is given. The obtained results confirm the performance improvement of the proposed filters in both speed and precision.

In this paper an efficient method for solving the inverse kinematics of Hyper-Redundant Robots is presented. The method is based on a flexible curve named backbone. Three approaches are presented to solve the inverse kinematics. In the first approach the "assumed mode method" is used. In the second approach, first an optimum problem using the "calculus of variations" and minimization of bending of backbone curve is defined. Then a numerical method named "resolved rate motion" is used for solving the obtained governing equations. In the third approach, a neural network based on wavelets is used to increase accuracy while reducing the computation time. Simulation results show that through the application of this network the learning time is significantly reduced in comparison to the conventional neural networks such as back propagation type networks.