نشریه مهندسی برق و مهندسی کامپیوتر ایران
سال پانزدهم شماره 3 (پیاپی 44، پاییز 1396)

Clustering is one of the important techniques for knowledge discovery in spatial databases. density-based clustering algorithms are one of the main clustering methods in data mining. DBSCAN which is the base of density-based clustering algorithms, besides its benefits suffers from some issues such as difficulty in determining appropriate values for input parameters and inability to detect clusters with different densities.
In this paper, we introduce a new clustering algorithm which unlike DBSCAN algorithm, can detect clusters with different densities. This algorithm also detects nested clusters and clusters sticking together. The idea of the proposed algorithm is as follows. First, we detect the different densities of the dataset by using a technique and Eps parameter is computed for each density. Then DBSCAN algorithm is adapted with the computed parameters to apply on the dataset. The experimental results which are obtained by running the suggested algorithm on standard and synthetic datasets by using well-known clustering assessment criteria are compared to the results of DBSCAN algorithm and some of its variants including VDBSCAN, VMDBSCAN, LDBSCAN, DVBSCAN and MDDBSCAN. All these algorithms have been introduced to solve the problem of multi-density data sets. The results show that the suggested algorithm has higher accuracy and lower error rate in comparison to the other algorithms.

Due to proliferation of smart phones, data traffic in cellular networks has been significantly increasing, which has resulted in congestions in cellular networks. Data offloading to a complementary network such as Wi-Fi has been identified as a rational and cost-effective solution to these congestions. In this paper, a multi-criteria offloading (MCO) mechanism is proposed to select the best transfer mode among: cellular delivery, delay-tolerant offloading (DTO) to a complementary network, and peer-assisted offloading (PAO). The proposed MCO mechanism utilizes TOPSIS multi-criteria decision analysis method and a prediction model for the Wi-Fi connection pattern. The decision criteria include: the fraction of total users’ request satisfied by offloading, data transfer costs of cellular operator to users, data transfer bandwidth of users in both cellular and complementary networks, and total users’ power consumption. To evaluate the proposed mechanism various scenarios have been simulated, and the results show that the MCO mechanism can successfully take into account the preferences of the cellular operator and its users. Through simulations, the MCO mechanism demonstrated superior performance in comparison with other proposed solutions in the literature in terms of balancing the load on the network, reducing the cost of the cellular operator, and reducing energy consumption of the users.

In the last decade, many researches have been done on fine-grained recognition. The main category of the object is known in this problem and the goal is to determine the subcategory or fine-grained category. Vehicle Make and Model Recognition (VMMR) is a hard fine-grained classification problem, due to the large number of classes, substantial inner-class and small inter-class distance. Furthermore, improving system accuracy leads to increasing in processing time. As we can see the state-of-the-art machine vision tool like convolutional neural networks lacks in real-time processing time. In this paper, a method has been presented briefly for VMMR firstly. Secondly, a cascading scheme for improving both speed and accuracy of this VMMR system has been proposed. In order to eliminate extra processing cost, the proposed cascading scheme applies classifiers to the input image in a sequential manner. Some effective criterions for an efficient ordering of classifiers are proposed and finally a fusion of them is used in the cascade algorithm. For evaluation purposes, a new dataset with more than 5000 vehicles of 28 different makes and models has been collected. The experimental results on this dataset and comprehensive CompCars dataset show outstanding performance of our approach. Our cascading scheme results up to 80% increase in the system processing speed.

Three-Dimensional Chips are made of stacking silicon layers which communicate with each other by Through-Silicon-Via (TSV) links. Manufacturing cost of Three-Dimensional chips is a function of the number of TSVs because the fabricating of a three-dimensional chip with fully vertical links is of high cost and high fabrication complexity.
The packet routing strategies in the 3D NoCs with partially TSVs is more complex than that in the 2D NoCs. In this paper, we proposed a routing algorithm for the 3D NoCs with partial TSVs, which provides a dynamic routing with maximum adaptivity for packets by dividing the network into three groups of layers, rows and columns. This algorithm is independent of vertical channel's position but related to layer number of the current packet and based on the layer number, odd or even, uses a special turn strategy to route packets on rows and columns with odd or even numbers. The proposed routing algorithm mitigates deadlock and livelock with only two virtual. The experiments show that average packet latency in proposed algorithm is 32.8% smaller than that in Elevator_First which is a well-known algorithm for packet routing in 3D chips. Also, this improvement on average packet latency and network throughput will be more with increasing on network size and reduction on TSV number.

Session Initiation Protocol (SIP) is an application layer protocol designed to create, manage, and terminate multimedia sessions in the IP multimedia subsystem (IMS). The widespread use of this protocol results in high traffic volume over SIP proxies, requiring delicate CPU allocation to flows. In this paper, we analyze the optimization problem of resource allocation in SIP proxies with two objective functions: maximizing total throughput and minimizing the least squares. Maximizing total throughput, prioritizes intra-domain flows over inter-domain ones, as the latter pass through two intermediate proxies. On the other hand, minimizing the least squares corresponds to a max-min fairness policy. Hence, we use round robin scheduling in proxies. In addition, we propose a SIP overload control algorithm that limits re-transmissions and prevents instability of proxies by controlling the length of SIP message backlog for each flow. This algorithm leads to better use of processing resources, in comparison with existing overload control algorithms.

Estimation of missing values is an important step in the preprocessing. In this paper, at two-step approach is proposed to fill the numeric missing values. In the first step, data is clustered. In the second step, the missing data in each cluster are estimated using a combination of weighted k nearest neighbors and linear regression methods. The correlation measure is employed to determine the appropriate method for the filling of missing data in each cluster. The quality of estimated missing values is evaluated using the root mean squared error (RMSE) criterion. Effect of different input parameters on the error of estimated values is investigated. Moreover, the performance of the proposed method for the estimation purpose is evaluated on five datasets. Finally, the efficiency of the proposed method is compared to four different estimation methods, namely, Mean estimation, multi-layer perceptron (MLP) based estimation, fuzzy C-means (FCM) based approximation method, and Class-based K-clusters nearest neighbor imputation (CKNNI) method. Experimental results show that the proposed method produces less error in comparison to other compared methods, in most of the cases.

Due to the shrinking of feature size, reduction in supply voltage and technology scaling, the sensitivity to radiation induced transient faults of digital systems has dramatically increased. Soft error causes transient distortion in circuit operation and is expected to become very important in combinational logic with increment of the circuit frequency. In this paper, we introduce an optimized method for hardening of combinational logic circuits against soft errors. In this method, first we have found the most sensitive nodes of the circuit by observability computations. Next for optimizing power-delay product and area, the reliability of the circuit has been computed and the number of the necessary nodes for hardening will be identified. In the next step, three different hardening methods including time redundancy, Schmitt trigger and transistor feedback have been carried out on standard test circuits as our vehicles. The comparison of three method results show that the hardened circuits with Schmitt trigger have the most cumulative critical charge and the least power-delay product and lead to an optimum hardening. Moreover, the simulation results approve the optimized hardening is obtained from suitable selecting the number of required nodes considering observability concepts and reliability computations together with the best node hardening method. Monte-Carlo simulations also approve the performance of the proposed method against process variations.

Due to the shrinking of feature size, reduction in supply voltage and technology scaling, the sensitivity to radiation induced transient faults of digital systems has dramatically increased. Soft error causes transient distortion in circuit operation and is expected to become very important in combinational logic with increment of the circuit frequency. In this paper, we introduce an optimized method for hardening of combinational logic circuits against soft errors. In this method, first we have found the most sensitive nodes of the circuit by observability computations. Next for optimizing power-delay product and area, the reliability of the circuit has been computed and the number of the necessary nodes for hardening will be identified. In the next step, three different hardening methods including time redundancy, Schmitt trigger and transistor feedback have been carried out on standard test circuits as our vehicles. The comparison of three method results show that the hardened circuits with Schmitt trigger have the most cumulative critical charge and the least power-delay product and lead to an optimum hardening. Moreover, the simulation results approve the optimized hardening is obtained from suitable selecting the number of required nodes considering observability concepts and reliability computations together with the best node hardening method. Monte-Carlo simulations also approve the performance of the proposed method against process variations.