NASHRIYYAH -I MUHANDISI -I BARQ VA MUHANDISI -I KAMPYUTAR -I IRAN, A- MUHANDISI -I BARQ
Year:2010 | Volume:8 | Issue:2|Papers Count:11

In this paper the first text to speech system for Kurdish language has been introduced. Kurdish language has two standard scripts, Arabic and Latin. In the text analysis part besides treating common problems in various Kurdish texts, the problems involved in both standard scripts have been dealt with. Also, standard symbols have been introduced into which the system converts the input texts in each of the two scripts. For the first time for Kurdish language, intonation patterns for various sentence types have been determined. In the speech production part, three different synthesis systems based on allophone, syllable, and diaphone have been implemented. For quality assessment of the above mentioned systems and their comparison with each other, the four tests of MOS, Intelligibility, DRT, and MRT have been used. The test results show the high intelligibility of our systems, especially the system based on diaphone.

Stochastic games, as the generalization of Markov decision processes to the multi agent case, have long been used for modeling multi-agent system and are used as a suitable framework for Multi Agent Reinforcement Learning. Learning Automata (LA) was recently shown to be valuable tools for designing Multi-Agent Reinforcement Learning algorithms. In this paper a model based on learning automata and the concept of entropy for finding optimal policies in stochastic games is proposed. In the proposed model, for each state in the environment of the game and for each agent an S-model variable structure learning automaton is placed that tries to learn the optimal action probabilities in those states. The number of its adjacent states determines the number of actions of each learning automaton in each state and every joint action corresponds to a transition to an adjacent state. Entropy of the probability vector for the learning automaton of the next state is used to help learning process and improve the learning performance and is used a quantitative problem independent measurement for learning progress. We have also implemented a new version of the proposed algorithm that balances exploration with exploitation yielding improved performance. The experimental results show that the proposed algorithm has better learning performance than the other learning algorithms in terms of cost and the speed of reaching the optimal policy.

DGs connected to distribution systems are increasing in size and number. Intentional islanding are proper alternatives to increase the reliability of end load in the network with does not access to any other branch. In other conditions these loads experiences permanent interruption. in load restoration operation, islanding is so important in the presence of  DGs. Islanding is occurred  when a part of distribution network is supplied only from DG sources.A new algorithm for distribution load restoration in the presence of DGs is presented in this thesis. The proposed approach minimizes the load interruption cost in restoration process based on Tabu search method. In this algorithm the maximum load with the most level of importance is restored in minimum time by reconfiguration, islanding and finally load shedding. Energy storage units and the location of synchronizing devices are considered in this thesis. The algorithm is implemented on a 32-bus test system and numerical results are presented.

One of the main identity authentication methods is the use of fingerprints. The most popular biometric method is fingerprint analysis. Most of the automatic fingerprint systems are based on minutiae matching. Therefore, extraction of minutiae is a critical stage in the design of fingerprint authentication systems. Computation of direction of lines in fingerprints is a stage which affects the quality of the extracted minutiae. The existing algorithms require complex and time-consuming computations and are software-based. This paper presents a hardware implementation which has improved the current methods. The presented method is based on pipeline architecture and has proved to perform efficiently.

Existence of a high inter-channel correlation in a stereo communication system results in a considerable performance degradation in the associated stereo acoustic echo canceller and also weight misalignment of adaptive filters even after finalizing the convergence period. In this paper an approach for improving the performance of NLMS adaptive filter is developed based on reducing the correlation of input signals employing a multi-input-multi-output décor relation network. This approach has a low-complexity neural network structure and can train in a real-time manner. Simulation results show an improvement in weight convergence rate and misalignment employing the proposed method.

Image-based fingerprint verification systems have been considered as a parallel method against the minutiae-based approach. This paper proposes a training based fusion method for fingerprint verification, using likelihood ratio (L.R). In this method, the matching scores which are extracted from orientation, spectral and textural features are fused. In order to fuse these image-based features, the likelihood ratio approach has been employed. FVC2000 database has been selected to evaluate the method. Also, the proposed method has been compared to a similar one that uses the simple sum as its fusion system. The comparison results show that the proposed fusion method has made a significant improvement for the accuracy of matching system, so that the equal error rate (ERR) of proposed system has been reduced to 0.14%.

In this paper an efficient method for modeling semiconductor devices using the drift-diffusion (DD) model and neural network is presented. Unlike HD model which is complicated, time consuming with high processing cost, the proposed method has lower complexity and higher simulate speed. In our method, a RBF neural network is used to modify DD parameters. The modified DD model can generate simulate results of accurate HD model. The proposed method is first applied to a silicon n-i-n diode in one dimension, and then to a silicon thin-film MOSFET in two dimensions, both for interpolation and extrapolation. The obtained results for basic variables, i.e., electron and potential distribution for different voltages, confirm the high efficiency of the proposed method.

Sensor node operating system provides a limited number of common services for developers to construct applications for wireless sensor networks. The sensor network community selected TinyOS as the de facto standard with most existing applications, libraries and device drivers available for TinyOS. The programming model of TinyOS is event-based and is not easy to use. In this paper, we present a new task scheduler for TinyOS that includes a new computation concept, named Job. Jobs are a collaborative and non-preemptive way of multitasking. On the next step, we propose a programming model which combines the asynchronous basis of event-driven systems with a more classical programming interface for the developer. As a result, developer that uses such an interface in his application will be provided with the sequential view we wanted. This programming model is suitable for applications that have long running computations and there is a data flow dependency between different tasks.