高维空间下求解MDA的扰动算法及其在人脸识别中的应用

本文提出了一种在高维空间下直接求MDA(Multiple discriminant analysis)最佳解的扰动算法，并把这一算法应用于人脸的识别中。在传统的MDA求解算法中，一般要求训练样本的个数足够大，以至于其类内散射矩阵为非奇异矩阵，即所谓的“小样本”问题。但是，在入脸识别中，由于人脸空间的维数非常大，而训练样本的个数有限，造成类内散射矩阵为奇异矩阵，从而使得用传统的求解方法失效。为了能在高维空间中求出MDA的最佳解，本文采用扰动的思想，巧妙地避开了矩阵的奇异性问题并找到了最佳变换矩阵。此外，把这一算法用于人脸的识别中，对ORL人脸图像库的实验显示，采用本文提出的算法达到比较低的错误率，其错误率仅为特征脸(Eigenface)方法的49．9％，为Fisher脸(Fisherface)的7914％。     

Special Apodized Fiber Bragg Grating for Flat—top Band—pass Reflectivity Filter

The characteristics of special apodized fiber Bragg grating (FBG) in flat--top pass--band as reflectivity filter are presented. This special apodized FBG was designed by the particle swarm optimization algorithm. Compared with conventional apodized FBG, the special apodized FBG presented was more robust in the flat—top pass——band characteristic even if the strength of grating is very week. This technology is very interesting in designing the filter for wavelength division multiplexing system.     

一种低功耗Cache设计技术的研究

低功耗、高性能的cache系统设计是嵌入式DSP芯片设计的关键。本文在多媒体处理DSP芯片MD32的设计实践中，提出一种利用读／写缓冲器作为零级cache，减少对数据、指令cache的读／写次数，由于缓冲器读取功耗远远小于片上cache，从而减小cache相关功耗的方法。通过多种多媒体处理测试程序的验证，该技术可减少对指令cache或者数据cache20％～40％的读取次数，以较小芯片面积的增加换取了较大的功耗降低。     

cdma1x无线子系统导频信道功率分配分析

分析了目前联通CDMA网络建设中的主要问题之一——导频功率分配问题。主要介绍了cdma 1x系统的信道类型，以及前向链路信道功率分配原则，并且对导频信道功率与无线网络覆盖和容量之间的关系进行了初步探讨。     

宏蜂窝模型的传播模型校正研究

详细介绍了 宏蜂窝传播模 型校正的基本操作流程 结果分析及注意事项 对工程 ,人员了解和开展模型校正工作有一定的指导作用 。     

软交换技术发展现状

介绍了国际软交换技术的现状，采用软交换技术的下一代网络提供的业务，从事软交换研究的国际组织及软交换技术发展中存在的问题。     

Plasma spraying of nanostructured partially yttria stabilized zirconia powders

Nanosized partially yttria stabilized zirconia particles, prepared using a co-precipitation method, were reprocessed into agglomerate powders using two methods for plasma spraying. The first method was to make micrometer-sized agglomerates directly following the grinding of the calcined yttria–zirconia agglomerates. The second method was to reconstitute the nanosized particles into micrometer agglomerates using spray drying. The deposition efficiency, porosity, microhardness and average grain size of the deposits made from these two reprocessed powders were studied. Distinct results related to the process parameters were obtained for the two types of powders. The second type of powder was more suitable for plasma spraying than the first one. Using the second type of powder, some unique results distinguished from those of the conventional partially yttria stabilized zirconia powders were observed and an optimized coating with a porosity of 3.8%, Hv_0.3 of 953 and mainly consisting of 1–3 μm columnar grains in the columnar direction and smaller than 100 nm in their cross-sections was achieved.     

Parallel multilevel fast multipole method for solving large-scale problems

The Multilevel Fast Multipole Method (FMM) is a well-established method and can be applied to solve electromagnetic (EM) scattering problems. Compared with other traditional methods, it requires less computational time and memory. However, constrained by a single processor's speed and memory limitations, the problem size that can be solved by serial implementation is still relatively small. For a million-unknown target, the computational time on a single processor is intolerable, and memory could be easily exhausted. Parallel-computing technology, which can utilize multiprocessors, provides an efficient way to solve electrically large-scale EM problems. This paper will focus on discussing the parallel methodologies applied to a multilevel FMM code, as well as demonstrating the computational efficiency of the parallel approach.     

对称非线性系统的无交互控制

本文讨论了具有对称性的非线性系统的无交互控制问题。通过定义对称系统的导出分布族的新概念,我们得出了问题可解的条件。我们取得的结果较不具对称性的情况要简单一些。从而说明,对称性不仅使得系统具有较好的结构,而且使得它的其它一些问题的研究得到简化。