Novel cross-connected cascode differential amplifier

We report on a novel amplifier circuit–differential cross-connected cascode1 1.?Reported briefly on WSEAS and ICECS Conferences. . Its fundamental distinction from an ordinary differential cascode lies in simultaneous feeding of input signal both to the common emitter (source) and common base (gate) stages, which are cross-connected. Such a connection results in the creation of two loops of positive current feedback. The stability of the amplifier is achieved due to low gain and phase shift in the loops. We show that the input signal is amplified in the input circuit. The input and output impedances are gained significantly as well. The current gain is increased considerably and the bandwidth is essentially expanded. Simulation results of such a cascode designed with IBM BJT sige5am and CMOS bicmos7hp transistors are presented. When compared with the ordinary cascode, the predicted and actually obtained bandwidths proved to be more than twice as wide: 8.6–18.7 GHz and 3.4–8.7 GHz on BJT and CMOS based, respectively.     

基于跨连接卷积神经网络的绝缘子检测方法

针对电力设备图像中绝缘子所占比例较小和容易漏检的问题,提出了基于跨连接卷积神经网络的绝缘子检测方法。首先,通过在区域建议网络将网络后三层的卷积层分别和全连接层连接,使得这三层的卷积特征同时送入分类层和回归层,从而得到一系列高质量的绝缘子候选区域;将得到的候选区域映射绝缘子检测子网络,通过将得到的感兴趣区域特征送入级联的Adaboost分类器,实现对绝缘子的检测。对所提出跨连接卷积神经网络生成的候选区域进行了评估,并对不同的绝缘子检测方法进行了对比实验。实验结果表明,该方法得到的候选区域召回率高且更集中于绝缘子所在位置,绝缘子检测准确率比常规方法高出10%。所提方法能较好地对复杂背景图像中不同大小的绝缘子进行有效识别和精确定位。     

基于跨连卷积神经网络的性别分类模型

为提高性别分类准确率, 在传统卷积神经网络(Convolutional neural network, CNN)的基础上, 提出一个跨连卷积神经网络(Cross-connected CNN, CCNN)模型. 该模型是一个9层的网络结构, 包含输入层、6个由卷积层和池化层交错构成的隐含层、全连接层和输出层, 其中允许第2个池化层跨过两个层直接与全连接层相连接. 在10个人脸数据集上的性别分类实验结果表明, 跨连卷积网络的准确率均不低于传统卷积网络.     

适用于远距离大容量架空线输电的交叉型子模块拓扑

提出了一种适用于模块化多电平换流器的具有直流故障自清除能力的交叉型子模块(CCSM)。该子模块由12个绝缘栅双极型晶体管、12个反并联二极管和4个电容组成,可分别输出4电平、2电平、0电平和-2电平。同等输出条件下,CCSM除了具有较高的经济性以外,还可在保证直流故障自清除能力的基础上为换流器提供更广泛的运行范围,尤其适用于电压等级高、输送容量大、采用架空线路传输的柔性直流输电系统。此外,还研究了基于CCSM的调制方法及电容电压平衡策略,并在PSCAD/EMTDC中搭建了仿真模型,验证了所提拓扑及控制策略的有效性与可行性。