Current Issue

• 2015, No. 04　Published:25 April 2015 Last issue    Next issue
  Columns in this issue: 论文

• 论文

  A 3-Dimension Monte Carlo simulator for
  semiconductor devices based on many integrated core 

  FANG Minquan1，ZHANG Weimin1，ZHANG Lilun1,2，

  2015, 37(04): 621-627. doi:

  Abstract ( 153 )   PDF (2096KB) ( 313 )   　　

  3D Monte Carlo simulation for semiconductor devices consumes long time. Especially when grids are growing and particles are increasing, the computing scale becomes very large. By analyzing the hotspots and the second level parallelism, the parallel scheme of the Effective Potential Method on Many Integrated Core is presented; Parallel schemes of particle free fighting, simulation information statistics and surface roughness scattering computation are researched. A 3-level parallel 3D Monte Carlo simulator for semiconductor devices is implemented and validated. Results show that the 3-level parallel program can get a better performance than the original single level parallel version. When the simulation accuracy is improved, the 3-level parallel program can get a greater speed-up ratio.

  A survey of the parallel solutions to
  convection-diffusion equation  

  GONG Chunye,BAO Weimin,LIU Haidong,YU Caihui,FANG Hong,ZHENG Zhaohu,WANG Li

  2015, 37(04): 628-633. doi:

  Abstract ( 469 )   PDF (528KB) ( 37687 )   　　

  The convection-diffusion equation is a typical partial differential equation. Its parallel solutions have reference significance for the parallel solutions to other equations. In this paper we conduct a survey on the parallel solutions to convection-diffusion equation. These parallel solutions are classified as four kinds of parallelism: explicit direct, implicit iterative, alternating group explicit and Monte Carlo. Computational principles and several examples are described. Finally, the prospects of future development in this field are presented.

  Acceleration of video caption
  retrieval algorithm based on Intel MIC  

  GAO Fei，LIU Yi

  2015, 37(04): 634-640. doi:

  Abstract ( 139 )   PDF (598KB) ( 261 )   　　

  Video caption retrieval is an important part of video retrieval area. With the improvement of OCR, video caption retrieval algorithm has also made a number of significant breakthroughs. However, as the retrieval results get enhanced, the fact that a video contains a large number of images and text information makes data processing the performance bottleneck. Manycore platforms develop rapidly and lay a nice hardware basis for high performance computing study. In this paper we use Intel MIC, a typical manycore platform, to accelerate video caption retrieval algorithms, and the Open MP parallel language is adopted. After an introduction of video caption retrieval, we present the parallel scheme of kernel algorithms in Intel MIC platform. Finally, we test the system on Intel Xeon Phi 7110P and obtain a desirable speedup ratio．

  Computing model and dynamic programming algorithm for
  multiple-choice hardware/software partitioning on MPSoC  

  ZHU Fengjun1，WU Jigang1,2，SHI Wenjun1，JIANG Guiyuan3

  2015, 37(04): 641-648. doi:

  Abstract ( 128 )   PDF (601KB) ( 274 )   　　

  Hardware/software (HW/SW) partitioning is one of the most crucial steps in hardware/software co-design,as the solution of partitioning directly affects the quality of the target system.For a given application,it is quite necessary to find a reasonable partitioning in order to obtain a target system which is of low cost but can be fast executed. In system design, one task has different implementation manners in hardware according to the different hardware areas.Thus,the multiple-choice HW/SW partitioning is more realistic than the traditional HW/SW partitioning which considers only a single implementation manner in hardware.It is also more challenging to solve the multiple-choice HW/SW partitioning problem. The two types of partitioning problems both have been proved to be NP-complete.For the multiple-choice HW/SW partitioning problem on Multiprocessor System-on-Chips (MPSoC),we present a computing model on MPSoC for the input controlflow graph with a binary tree.A dynamic programming algorithm is also proposed to solve this problem.Extensive simulation results show that the proposed algorithm can provide the exact solution for the problems with certain scale.It also shows the relationship between the algorithm performance and the constraints of hardware area.

  Unsupervised feature selection based on low rank and sparse score 

  YANG Guoliang,XIE Naijun,WANG Yanfang,LIANG Liming

  2015, 37(04): 649-656. doi:

  Abstract ( 133 )   PDF (4616KB) ( 333 )   　　

  Feature selection is an important reduce dimensional step in dealing with high dimensional data. Low-rank representation model shows a very good discriminative ability and can capture the global structure of the data. Sparse representation model can reveal the true intrinsic structure information with fewer connection relationships. Based on the low-rank representation model, sparse constraint items are added to construct a low-rank and sparse representation model, which is used to learn the low-rank and sparse affinity matrix between the data. Then with the obtained affinity matrix, we propose an unsupervised feature selection based on Low-Rank and Sparse Score (LRSS). After the clustering and classification of the selected features on different databases  we compare our proposal with traditional feature selection algorithms. Experimental results verify the effectiveness of our method, and show that our proposal outperforms the state-of-art feature selection approaches.

  Research and implementation of
  a Hadoop-based EPC IOT data analysis system 

  ZHAO Huiqun,HUANG Chuancong

  2015, 37(04): 657-662. doi:

  Abstract ( 122 )   PDF (660KB) ( 263 )   　　

  With the rapid growth of data which are sensed by the sensor nodes deployed in the logistics center, the current data processing methods cannot meet the demand of rapid and effective data analysis and processing. In order to improve the performance of current data processing, based on analysis of the Hadoop technique, we design a  novel EPC IOT data processing algorithm implemented by Hadoop. Experimental results show that the proposed algorithm can greatly improve the  processing capability of historical data of the logistics center when the EPC data are processed on the Hadoop-based system.

  An optimization study on earthquake data processing
  and earthquake dynamics simulation platform  

  LIAO Kaining，HAO Yongwei

  2015, 37(04): 663-669. doi:

  Abstract ( 108 )   PDF (840KB) ( 236 )   　　

  We present an application of high performance computing cluster in the field of earthquake science.  First we analyze the calculation characteristics of earthquake data processing and the earthquake dynamics simulation, and highlight the performance requirements of the earthquake science study for high performance computing platforms. Taking the earthquake data processing and earthquake dynamics simulation platform of Institute of Geology, China Earthquake Administration as an example, we introduce the platform architecture, the software environment, system operation and maintenance, system application etc. More importantly, we introduce the optimization measures for the system, and discuss the ways of improving the operating efficiency of the cluster system from the perspectives of service application, users and management.

  A framework of WAN distributed topic
  crawling system based on Hadoop   

  WANG Shufen1,GAO Junli1,ZOU Pu1,SONG Haitao2

  2015, 37(04): 670-675. doi:

  Abstract ( 211 )   PDF (921KB) ( 338 )   　　

  Comparing with LAN crawling systems, WAN distributed crawling systems have lots of advantages, however, the existing crawling systems based on Hadoop are mostly used in LAN. To achieve a high computing speed of Hadoop in WAN, we present a crawler framework based on Hadoop. To achieve an extensible storage, all data are stored on the Hadoop distributed file system and the web pages are analyzed through MapReduce in parallel. To obtain reliable communication, a message oriented middleware is used. To make the framework customizable, a template matching method is proposed. The performance simulation shows that the crawler framework can support large scale crawling work.

  A high efficient scalable mechanism for distributed storage within enterprises  

  ZHANG Jinbo

  2015, 37(04): 676-681. doi:

  Abstract ( 115 )   PDF (471KB) ( 213 )   　　

  Based on the research of the technological development and infrastructure of file storage, we present a high efficient scalable mechanism for distributed storage. Servers and storage devices that have already been used in enterprises  serve as distributed storage units and the files serve as the storage objects. We focus on establishing a distributed storage mechanism which can increase or decrease the storage units at any time with high efficiency with even zero waiting time. The more servers and storage devices a company has, the bigger role the distributed storage mechanism can play. Therefore the storage performance will be greatly enhanced, and the labor intensity of file management will be greatly reduced. More importantly, it can help enterprises make full use of the old servers and storage devices, and lessen disposable replacement investment.

  Extraction and recognition of protocol fingerprint
  based on protocol deviation  

  LI Meijian,WANG Yongjun,XIE Peidai，HUANG Zhijian

  2015, 37(04): 682-691. doi:

  Abstract ( 141 )   PDF (2070KB) ( 291 )   　　

  Since traditional protocol fingerprinting methods are usually time-consuming and cannot properly extract or recognize cryptographic protocols, we propose a novel protocol fingerprinting method based on protocol deviation. Protocol deviation describes the network behavior differentiations between different protocol implementations. Based on the dynamic binary analysis technology, the proposed method extracts protocol characteristics from the session stream level and the message level of protocol deviation. Experimental results show that the proposed method is not only feasible, but also provides a new idea for the fingerprinting of cryptographic protocol applications.

  A model for protecting location privacy
  against attacks from friends in SNS 

  CHEN Weihe,LI Wenjing,ZHU Jiang

  2015, 37(04): 692-698. doi:

  Abstract ( 138 )   PDF (654KB) ( 388 )   　　

  As the development of wireless network, more and more mobile social networking users publish their geographic locations which contain sensitive geographic location information that may lead their privacy under attack. Most existing privacy protection methods generalize  geographic location at the expense of quality of service, position attackers in LBS service, and they don't take account of the reliability degree among social network friends when location information is provided. To solve this problem, we design a privacy protection model, named Lintimacy, based on social network friends intimacy to prevent friend attackers. Theoretical analysis and experimental results show that comparied with Google Maps which is added to the Latitude services, the proposed algorithm provides privacy protection with less information loss.

  Interleaver design for RA codes based on full binary tree 

  ZHANG Hongtai,LIU Hongli,LIU Shugang

  2015, 37(04): 699-703. doi:

  Abstract ( 132 )   PDF (1552KB) ( 263 )   　　

  In order to improve the encoding efficiency of repeat accumulate codes under the circumstance of long input information, we present an optimized interleaver design for repeat accumulate codes. The input information sequence of interleaver is regrouped by the child nodes of full binary tree according to its parity arrangement, and the output sequence is obtained by using the leaf nodes to recombine the grouped sequence. Therefore, the relativity among input information is greatly reduced compared with S-random interleaver, and it prevents the class-I and classII four circles of check matrix from being produced in the encoding process while the decoding accuracy is guaranteed. Simulation results show that the optimized interleaver is faster and its BER is far less than regular interleavers under long input information; and the optimized interleaver is much faster and has a better performance(about 0.3 dB) than S-random interleavers at the same BER of 1*10-4.

  Design and implementation of a realtime
  transmission system for laboratory test data 

  XU Fuxin,XIANG Chao,LIU Yanqun,ZHANG Na

  2015, 37(04): 704-710. doi:

  Abstract ( 122 )   PDF (828KB) ( 245 )   　　

  In order to monitor the students’ measuring progress synchronously and provide effective guidance in teaching laboratories, we design a testing system to receive and send experimental data wirelessly, which consists of the single chip microcomputer, AVR ATmega 16, and the integrated wireless RF module, RF4432. Aiming at the two procedures and under the FIFOs pattern provided by wireless module, we develop a set of wireless communication protocols which can meet the needs of intercommunications among the master and slave machines. And timedivision communication is adopted to prevent data collision. We elaborate the establishment of the wireless communication circuit, the protocol and timing sequence of the system’s master and slave machines in detail, as well as the design of a visualized interface. Finally, the testing system is applied in Millikan oildrop experiment, and the results show that it not only runs stably and efficiently but also fulfills the requirements of monitoring students' experimental progress. In conclusion, it can enhance the experiment teaching efficiency significantly.

  MRNN: A novel wireless sensor network
  dynamic modeling method for fault detection
  using modified recurrent neural network  

  HUANG Xu

  2015, 37(04): 711-718. doi:

  Abstract ( 159 )   PDF (3633KB) ( 269 )   　　

  We present a novel sensor node fault detection method for wireless sensor network (WSN). Modified Recurrent Neural Network (MRNN) is used to model sensor nodes, the nodes’ dynamics, and the interconnections with other sensor network nodes. An MRNN modeling approach is used for sensor node identification and fault detection in WSN. The input to the MRNN chooses those that include previous output samples of the modeling sensor nodes, and the current and previous output samples of the neighboring sensors. The model is based on a new structure of a backpropagation type neural network. The input to the MRNN and the topology of the network are based on a general nonlinear sensor model. Simulation results demonstrate the effectiveness of the proposed scheme and the MRNN method has higher failure detection accuracy in the case with smaller confidence factors  compared with the Kalman filter method.

  Identity-based multi-recipient signcryption key
  encapsulation in the standard model  

  CHEN Ming

  2015, 37(04): 719-725. doi:

  Abstract ( 159 )   PDF (517KB) ( 188 )   　　

  Signcryption key encapsulation mechanism can simultaneously achieve the confidentiality and authentication of the key. Based on the Wa-IBE encryption and the PS-IBS signature, we propose an identitybased signcryption key encapsulation mechanism (IBSC-KEM) and an identity-based multi-recipient signcryption key encapsulation mechanism (mIBSC-KEM) in the standard model. PS-IBS signature is adopted in the new schemes as a signature algorithm and a variant of  Wa-IBE encryption is used by the key encapsulation algorithm. Therefore, the unforgeability and confidentiality of the new schemes are verified in the standard model, and are deduced toward cracking the PS-IBS signature and the Wa-IBE encryption respectively. The new schemes can be used to build secure one-to-one or many-to-one authentication and key exchange schemes.

  A realization approach of business model based on jBPM5 

  TIAN Zhaoyang1,2,KANG Guosheng1,2,YANG Liqin1,2,3,ZHANG Liang1,2,ZHANG Xiaon

  2015, 37(04): 726-733. doi:

  Abstract ( 124 )   PDF (755KB) ( 199 )   　　

  An enterprise usually evolves through mergers and acquisitions. Its bottom-up nature leads to operational process issues, such as heterogeneity,redundancy, and inconsistency. All these problems definitely hinder companies from achieving global strategy.To solve these problems,we propose a heterogeneous business process consolidation strategy based on business patterns,in which a business process is modeled as a composed business pattern.Each business pattern can be further configured to meet specific subsidiary’s requirements.To leverage the approach to real situations,here we propose a realization approach of the business pattern strategy based on jBPM5,in which business rules are introduced to configure business patterns and work as a control facility to drive the activities in patterns.We also report a case study of the proposed approach in the course of the business process consolidation in China Mobile Communication Corporation(CMCC),and the experimental results validate the effectiveness of our approach.

  A novel artificial bee colony algorithm
  based on dynamic search strategy  

  WANG Zhigang

  2015, 37(04): 734-739. doi:

  Abstract ( 125 )   PDF (628KB) ( 300 )   　　

  To prevent the traditional artificial bee colony algorithm from being trapped into premature phenomenon,we propose a novel artificial bee colony algorithm based on dynamic search strategy,in which two different search strategies are combined to produce a new one.We dynamically take the advantages of the two different search strategies,making a balance between the local search ability and the global search ability.Benchmark functions are tested, and the results demonstrate that the new algorithm has fast convergence, high accuracy and more robustness,and is more suitable to solve highdimensional complex global optimization problems.

  Application study of ant colony optimization
  in wirelesssensor network routing  

  LUO Xu,WU Xiaojun

  2015, 37(04): 740-746. doi:

  Abstract ( 137 )   PDF (701KB) ( 251 )   　　

  Though the basic ant colony algorithm can find out the optimum solution when solving optimal path problems,it is easy to fall into a local optimal solution，and meanwhile the convergence time is too long.Considering the lack of applications of the basic ant colony algorithm in wireless sensor network routing，we propose an improved ant colony algorithm，and apply it to the wireless sensor network routing. This algorithm not only adds the dynamic penalty function and the weighting factors to the state transition probability formula，but also combines the local pheromone update and the global pheromone update to update the path information.We fully take into account of the transmission distance from nodes to nodes and the residual energy of the sensor nodes as well.The simulation results obtain some different curves from the basic ant colony algorithm and the improved algorithm in the residual energy of the sensor nodes and network transmission delay in wireless sensor network routing，verifying the efficiency of the improved ant algorithm in wireless sensor network routing.

  Pose calibration and visual guidance of endmost
  tools in hand-eye system  

  WU Peiliang1，LI Pengfei2，MAO Bingzhi3

  2015, 37(04): 747-753. doi:

  Abstract ( 130 )   PDF (620KB) ( 240 )   　　

  When service robots provide daily services,different tools may be held by the gripper on the robot arm from a right angle,but it is hard to measure directly and exactly the pose relationship between hand-eye systems and endmost tools.In this paper,we calibrate this relationship by only utilizing a bifocal monocular camera focusing on the eye-in-hand system by three steps.Firstly,we propose a method to calculate the focal length after camera focusing.Secondly,taking camera focusing as the camera translation along the optical axis, we present a method to detect the position and the pose of the endmost tools in the camera and in the manipulator coordinate systems respectively according to the two images captured before and after camera focusing. Meanwhile, a method is given to calculate the direction vector from endmost tools to the expected position.

  Improved feature matching algorithm based
  on indexing sub-vector distance  

  FAN Wenting,CHEN Xiuhong

  2015, 37(04): 754-759. doi:

  Abstract ( 126 )   PDF (1065KB) ( 236 )   　　

  The indexing Sub-Vector Distance ( iSVD ) based feature matching algorithm has good search accuracy and efficiency in feature matching of high dimensional vector,but the computational complexity of this algorithm is still high and it is limited in practical applications.Aiming at this problem,we introduce a key dimension selection method to improve the iSVD based feature matching algorithm.This method first divides the feature vector into several sub vectors,and selects some sub vectors to replace the original ones;then we create the index value for the feature vector,and search the nearest neighbor point according to the index value.This method can effectively differentiate the vectors with smaller similarity,and it can further reduce the search scope of the nearest neighbor.Experimental results show that the improved iSVD based feature matching algorithm has good search accuracy,and meanwhile improves the matching accuracy and shortens the time of matching,thus having a good practicability.