MPEG-4视频编码器象素压缩模块的VLSI结构设计

文章设计了一种基于MPEG-4的视频压缩编码器中象素压缩模块的VLSI结构。该设计采用分布算式结构——NEDA作为DCT变换的核心技术；应用基于LUT表结构使量化／反量化模块的设计简洁明了；同时对AC／DC预测模块还应用了新的存储策略，大大降低了FPGA中宝贵的存储空间。在满足处理速度和精度的要求下，利用了较少的晶体管数目和简洁的结构实现了象素压缩模块。     

基于卷积神经网络的HL-2A装置H模辨识研究

基于HL-2A装置的放电实验数据，利用卷积神经网络和时间窗口算法开发了高约束(H)模时段的识别算法，得到了可靠的高成功率的高约束模时段识别结果。算法中，选取206次放电实验数据中等离子体储能及氘α通道信号作为双通道原始数据进行学习，得到一个深度为21层的二分类卷积神经网络。该网络模型经过其他474次放电数据的测试集检验，高约束模识别的正确率达到了98.17%。     

Radial transport dynamics studies of SMBI with a newly developed TPSMBI code

In tokamak plasma fueling, supersonic molecule beam injection(SMBI) with a higher fueling efficiency and a deeper penetration depth than the traditional gas puffing method has been developed and widely applied to many tokamak devices.It is crucial to study the transport dynamics of SMBI to improve its fueling efficiency, especially in the high confinement regime. A new one-dimensional(1D) code of TPSMBI has also been developed recently based on a six-field SMBI model in cylindrical coordinate. It couples plasma density and heat radial transport equations together with neutral density transport equations for both molecules and atoms and momentum radial transport equations for molecules. The dominant particle collisional interactions between plasmas and neutrals, such as molecule dissociation, atom ionization and charge-exchange effects, are included in the model. The code is verified to be correct with analytical solutions and also benchmarked well with the trans-neut module of BOUT++ code. Time-dependent radial transport dynamics and mean profile evolution are studied during SMBI with the TPSMBI code in both slab and cylindrical coordinates. Along the SMBI path, plasma density increases due to particle fuelling, while plasma temperature decreases due to heat cooling. Being different from slab coordinate, the curvature effect leads to larger front densities of molecule and atom during SMBI in cylindrical coordinate simulation.     

Investigation of molecular penetration depth variation with SMBI fluxes

We study the molecular penetration depth variation with the SMBI fluxes.The molecular transport process and the penetration depth during SMBI with various injection velocities and densities are simulated and compared.It is found that the penetration depth of molecules strongly depends on the radial convective transport of SMBI and it increases with the increase of the injection velocity.The penetration depth does not vary much once the SMBI injection density is larger than a critical value due to the dramatic increase of the dissociation rate on the fueling path.An effective way to improve the SMBI penetration depth has been predicted,which is SMBI with a large radial injection velocity and a lower molecule injection density than the critical density.     

基于多帧边缘差异的视频运动对象的分割与跟踪算法

从视频场景中分割和跟踪感兴趣的视频对象对于MPEG-4等基于对象的视频编码来说是关键性的技术之一。针对目前大部分视频对象分割和追踪算法相当复杂但仍不能有效地去除背景噪声的问题,该文提出用于分割和跟踪视频运动对象的一种基于多帧边缘差异的算法。该算法利用一组帧的边缘差异来提取运动对象区域,通过聚类方法去除背景像素点,利用形态学算子得到对象分割模板,同时通过建立前帧感兴趣对象与当前帧运动对象的帧间向量跟踪当前帧的感兴趣视频对象。不同标准视频测试序列的测试结果表明,该算法能够实现对感兴趣的视频运动对象更为精确、快速和有效地分割和跟踪。     

局部运动场景中运动对象的压缩域分割算法

压缩域视频运动对象的分割技术近年来逐渐被人们所重视,并且成为较实用的技术之一.该文针对目前压缩域中运动对象分割质量不高的问题,提出一种压缩域分割的改进算法.该算法首先从MPEG压缩码流中提取运动矢量,并对矢量进行运动累加和中值滤波处理,最后利用贝叶斯风险决策对非零运动矢量区域进行校正,最终得到较好的分割结果.该文还定义了压缩域分割质量的评价参数分割率和错分率,为评价压缩域的分割结果提供了客观评价标准.     

超声分子束注入密度和宽度对托克马克装置加料深度的影响

基于HL-2A托卡马克装置的真实磁场位形,应用大型边缘等离子体湍流模拟程序BOUT++中的子程序模块trans-neut对不同的超声分子束注入(SMBI)密度和宽度进行模拟.在SMBI过程中,保持单位时间内分子注入个数和注入速度恒定,在恒定通量情况下,通过调整注入分子束密度和宽度来研究SMBI注入深度的变化.研究结果表明:在注入密度较小、注入宽度较大时,SMBI的注入深度更深,分子和原子的分解率和电离率的时空区域较宽.分子分解局域化会抑制全局分解率的增长,而分解局域化又会引发局域分解率的加速增长,进而促进全局分解率的增长,促进效果占优导致在注入速度一定的情况下,恒定通量的分子注入发散角越小,分子注入深度越浅.     

Simulations of fast component and slow component of SMBI on HL-2A tokamak

It is very important to improve the penetration depth and fueling efficiency of supersonic molecular beam injection(SMBI) especially for the next generation fusion devices such as ITER. Two components, a fast component(FC) and a slow component(SC), have been observed in the HL-2A SMBI experiments for several years, and the FC can penetrate much more deeply than the common SMBIs which draws a great deal of attention for a better fueling method. It is the first time to the FC and SC of SMBI have been simulated and interpreted in theory and simulation in this paper with the trans-neut module of the BOUT++ code. The simulation results of the FC and SC are clear and distinguishable in the same way as the observation in experiment. For the major mechanism of the FC and SC, it is found that although the difference in the injection velocity has some effect on the penetration depth difference between the FC and SC, it is mainly caused by the self-blocking effect of the first ionized SMB. We also discuss the influence of the initial plasma density on the FC and SC,and the variation of the SC penetration depth with its injection velocity.     

支持MPEG-4面向对象编码的VLSI模块结构设计

MPEG-4中一个新特性是面向对象编码。为了支持这种特性．MPEG-4标准引入了两种新的算法。重复填充和纹理填充.根据这两种算法的特点,设计了MPEG-4重复填充和纹理填充的VLSI结构。文中的方案在Xilinx ISE6．1i集成开发环境下,采用了VHDL进行了描述,并使用电子设计自动化工具进行了模拟和验证。仿真和综合结果表明,本文设计的VLSI处理器,其逻辑功能完全正确,而且可以在很低的时钟频率下满足MPEG-4 Core Profiles ＆ Level2的实时编码要求,可用于MPEG-4的VLSI实现。     

MPEG-4运动补偿的VLSI结构设计

针对MPEG．4解码中运动补偿控制复杂、数据吞吐量大、实现较困难，提出了一种适合MPEG-4的运动补偿硬件实现方案，解决了时序分配、输入输出控制等较难处理的问题。此方案已经在Xilinx ISE6．li集成开发环境下，采用了VHDL进行描述，并使用了电子设计自动化（EDA）工具进行了模拟和验证。仿真和综合结果表明，设计的运动补偿处理器逻辑功能完全正确，而且可以满足MPEG-4 Core Profiles ＆ Level 2的实时编码要求，可用于MPEG-4的VLSI实现。