ENTROPY IMMUNITY OF FEEDFORWARD NETWORKS

Siegenthaler's“Correlation Immunity”concept has been improved in this paper.From practical points of view,the new concept is more powerful than the original one in avoid-ing the trade-off between“the order of correlation immunity”and“the linear complexity”of keystreams in cipher system.Bent functions are also introduced into the studies of linear approxima-tion and entropy immunity for feedforward networks.New results and new methods are presentedalso.     

Linear feedforward control of two-level quantum system by modulated external field

We propose a model of feedforward (open-loop) optical control of two-level atom in the linear form. This model allows to express the general form of solution for the atomic level populations via the arbitrary shapes of the control signal. Then we make numerical investigations of different shapes for the optical control signal.     

Boole函数的线性可分性(Ⅰ)——n-维超立方体的基本理论

Boole函数的线性可分性是前向人工神经网络理论中的一个比较困难的问题之一。目前仅对变量数n≤7的某些问题给予讨论。当n≥8时,尚无判别Boole函数线性可分的一般准则,更无线性可分Boole函数的计数公式。基于此,本文详细地研究了与Boole函数线性可分性有关的n-维超立方体的基本理论,包括n-维超立方体的基本性质、超立方体中的平行线、子超立方体的计数等,并给出了构造n-维超立方体图的一种新方法。     

Data-based spatial feedforward for over-actuated motion systems

For advanced motion systems there is an increasing demand for higher production throughput and accuracy. Traditionally, such systems are designed using a rigid-body design paradigm, which aims at designs with high stiffness. The alternative is to design a lightweight system and deal with the resulting flexibilities by over-actuation and over-sensing. This paper presents a data-based spatial feedforward method based on previous task trials, which aims at reducing the vibrations over the complete structure during motion. The proposed method is experimentally validated on an industrial prototype and compared to standard mass feedforward using rigid-body decoupling.     

A process-tolerant out-of-band blocker rejection technique for SAW-less receivers

In this paper, an active filtering technique is presented which is capable of filtering the out-of-band blockers in wireless receivers. The concept is based on the feedforward cancellation technique where a blocker replica is subtracted at the output of the low-noise amplifier (LNA). In contrast to the previously reported feedforward cancellation methods, exact gain and phase matching are easily obtained in the proposed architecture to produce a highly selective narrowband frequency response at the output of the LNA with wide rejection bandwidth. For the proof of concept, the system is implemented in a 65 nm CMOS technology. It occupies a total area of 0.8 mm² and the current consumption is 24 mA from a 1.2 V supply. The system post-layout simulations showed a blocker rejection of more than 33 dB for blocker signals 100 MHz away from the desired signal when the feedforward path is activated. The noise figure (NF) of the entire system is 3.8 dB that degrades to 5.8 dB when the feedforward path is activated.     

CDMA射频线性功率放大器

该文对前馈线性校正射频功率放大器进行了分析,给出了分析结果。并用该结果指导设计了CDMAIS-95基站射频线性功率放大器,放大器增益47dB,连续波输出功率60W,在通频带内三阶交调改善了23dB。     

前馈系统中辅助放大器非线性分析

前馈系统中由于信号对消环路失配,导致在辅助功放输入端载波信号无法完全对消,辅助放大器的再次非线性失真会影响整个系统的失真。本文研究了残余载波信号再次产生的互调失真对整个系统线性化指标的影响,并给出了系统性能与电路参数和幅度、相位失配的关系式,通过这个关系式可以计算出要达到特定载波互调比时所需器件的最低指标要求。     

前馈神经网络的一种有效学习算法

本文提出了基于混合ＧＮ－ＢＦＧＳ法进行前馈神经网络学习的新算法，该算法结合ＧＮ法与ＢＦＧＳ法的特点，既利用了问题本身的特殊结构，又能取得超线性甚至二次渐近收敛率。与ＢＰ算法相比，这种算法可取得更快和更可靠的学习特性，在学习过程中利用该方法能够区分非零残量和零残量问题的特点，提出了自动调整隐单元数的方法，从而可以保证网络学习与归纳能力，示例系统的结果表明了所提方法的有效性。     

Applications of Hybrid Monte Carlo to Bayesian Generalized Linear Models: Quasicomplete Separation and Neural Networks

Abstract

The “leapfrog” hybrid Monte Carlo algorithm is a simple and effective MCMC method for fitting Bayesian generalized linear models with canonical link. The algorithm leads to large trajectories over the posterior and a rapidly mixing Markov chain, having superior performance over conventional methods in difficult problems like logistic regression with quasicomplete separation. This method offers a very attractive solution to this common problem, providing a method for identifying datasets that are quasicomplete separated, and for identifying the covariates that are at the root of the problem. The method is also quite successful in fitting generalized linear models in which the link function is extended to include a feedforward neural network. With a large number of hidden units, however, or when the dataset becomes large, the computations required in calculating the gradient in each trajectory can become very demanding. In this case, it is best to mix the algorithm with multivariate random walk Metropolis—Hastings. However, this entails very little additional programming work.