A High-Performance Operational Amplifier for High-Speed High-Accuracy Switch-Capacitor Cells

A highspeed highaccuracy fully differenttial operational amplifier (opamp) is realized based on noMillercapacitor feedforward (NMCF) compensation scheme. In order to achieve a good phase margin, the NMCF compensation scheme uses the positive phase shift of lefthalfplane (LHP) zero caused by the feedforward path to counteract the negative phase shift of the nondominant pole. Compared to traditional Miller compensation method, the opamp obtains high gain and wide band synchronously without the polesplitting effect while saves significant chip area due to the absence of the Miller capacitor. Simulated by the 0.35 μm CMOS RF technology, the result shows that the openloop gain of the opamp is 118 dB with the unity gainbandwidth (UGBW)of 1 GHz, and the phase margin is 61°while the settling time is 5.8 ns when achieving 0.01% accuracy. The opamp is especially suitable for the frontend sample/hold (S/H)cell and the multiplying D/A converter(MDAC) module of the highspeed highresolution pipelined A/D converters(ADCs).     

数字自适应前馈功放线性化研究

本文研究了一种新的前馈线性化功放的结构,省去了导频信号,在经典的两环前馈法基础上引入第三环,通过第三环直接提取交调信号,同时在线性化功放系统中引入了DSP控制器,用最优化算法进行控制,实现了对线性化功放的自适应控制。此方案被用于cdma2000基站功放的设计,使三载波CDMA信号输出的ACPR低于-51dB。     

多目标规划有效解的Kuhn—Tucker必要条件

本文提出一个新的约束规格,导出可微多目标规划的有效解的Kuhn-Tucker必要条件,并证明在此条件下,有效解是Kuhn-Tucker真有效解。     

前馈神经网络的代价函数全局最小值分析

本文首次从理论上给出了三层前馈神经网络代价函数全局最小值的计算公式.这一计算公式在网络训练之前就可根据已知的目标样本和隐层神经元个数来确定网络代价函数的全局最小值.并指出代价函数全局最小值髓隐层神经元个数的增加而单调减小.当隐层神经元个数不小于样本个数时,网络的代价函数全局最小值将等于零.     

Basis Property of Chains of Eigenfunctions for Certain Boundary-Value Problems

Mathematical Notes -     

A permeameter for unsaturated soil

A permeameter for unsaturated soil was developed by observing the way in which pore water recovers hydrostatic equilibrium. It works like an hour glass that is turned upside-down everytime the state of reference (or hydrostatic equilibrium) is reached. The hydraulic conductivity is deduced from the curves of evolution of pore-water pressure and from the distribution of partial density of water at hydrostatic equilibrium. Roman Letters a is defined by (10), kg m^–3 - A _n coefficients of the analytic solution, kgm^–3 - C ₁, C ₂, C ₃, C ₄ constants and constants of integration - D diffusivity, m² s^-1 - g gravity constant, m s^-2 - g gravity vector field - K hydraulic conductivity defined by (2), m⁵ s^-1 J^-1 - K _w hydraulic conductivity defined by (5), m ^-1 - k permeability - L length of soil sample, m - n integer in (22) - n porosity - p absolute pore water pressure, Pa - p ₀ absolute pore water pressure, Pa - p _a absolute air pressure, Pa - q volume flux or Darcy's velocity, m s^-1 - r exponent defined by (13) - S _w degree of saturation, % - t time variable, sec - u _n , v _n are defined by (22b), (22c) - x(x, y, z) space variable Greek Letters , are defined by (11), (13) - _w dynamic viscosity - water partial density, kg m^–3. It is the ratio of the mass of water to total volume of a representative elementary volume - ₀, _l water partial densities, kgm^–3 - _w density of water, kgm^–3 - _s density of solid particles, kgm^–3 - differences of partial density, kgm^–3 - p differences of water pressure, Pa - pi - , · gradient operator, divergence operator - Laplacian operator - volumetric water content, % - piezometric head, m     

一种可编程高线性CMOSOTA的设计

采用SMIC 0.18 μm工艺,±0.9 V电源电压,提出了一种可编程高线性全对称伪差分 CMOS OTA 电路.提出的伪差 分输入级电路基于基本伪差分(PD)结构并采用 HD3 前馈技术消除三阶谐波失真,输出级采用可编程电流放大器实现.使用 Cadence 软件的 Spectre 仿真器进行电路仿真可得,Gm值三个...     

三级倒立摆的T-S型前馈补偿模糊神经网络控制

为解决T akag i-Sugeno型模糊神经网络在控制多变量系统时的规则组合爆炸问题,提出一种误差前馈补偿的模糊神经网络控制方案,有效实现了三级倒立摆的稳定控制。该控制方案适用对状态变量可按性质和重要程度划分的多变量系统的控制,大大减少了模糊神经网络控制器的规则数,有利于利用专家的控制经验,具有良好的鲁棒性和非线性适应能力。     

抗熵漏前馈网络研究

本文从实际出发减弱了Siegenthaler的相关免疫限制条件,实现了在不作出实质性牺牲的前提下避免了相关免疫性与线性复杂度之间的折衷(trade-off)。接着将Bent函数引入前馈网络的线性逼近熵漏现象的研究之中,得出了一些新结果。文中的研究方法也与前人不同。     

Approximation by neural networks with sigmoidal functions

In this paper, we introduce a type of approximation operators of neural networks with sigmodal functions on compact intervals, and obtain the pointwise and uniform estimates of the ap- proximation. To improve the approximation rate, we further introduce a type of combinations of neurM networks. Moreover, we show that the derivatives of functions can also be simultaneously approximated by the derivatives of the combinations. We also apply our method to construct approximation operators of neural networks with sigmodal functions on infinite intervals.