Robust Statistics - 2nd Edition
鲁棒统计，现代统计方法， Robust Statistics第二版，学习现代统计方法R○ BUST STAT|STCSSecond editionPeter j, huberProfessor of Statistics, retiredKlosters SwitzerlandEⅣ ezio m. RonchettiProfessor of StatisticsUniversity of Geneva, SwitzerlandWILEYA JOHn WileY SONS INC. PUBliCAtIONCopyrightc 2009 by John Wiley Sons, Inc. All rights reservedPublished by John Wiley sons, Inc, Hoboken, New JerseyPublished simultaneously in CanadaNo part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form orby any means, electronic, mechanical, photocopying, recording, scanning, or otherwise, except aspermitted under Section 107 or 108 of the 1976 United States Copyright Act, without either the priorwritten permission of the Publisher, or authorization through payment of the appropriate per-copy fee tothe Copyright Clearance Center, Inc, 222 Rosewood Drive, Danvers, MA 01923, (978)750-8400, fax978)750-4470,oronthewebatwww.copyrigom. requests to the publisher for permission shouldbe addressed to the permissions department John Wiley sons, Inc., 11 1 River Street, Hoboken, NJ07030,(201)748-6011,fax(201)748-6008,oronlineathttp:/www.wileycom/go/permissionLimit of Liability /Disclaimer of Warranty: While the publisher and author have used their best efforts inpreparing this book, they make no representations or warranties with respect to the accuracy orcompleteness of the contents of this book and specifically disclaim any implied warranties ofmerchantability or fitness for a particular purpose. No warranty may be created or extended by salesrepresentatives or written sales materials. The advice and strategies contained herein may not be suitablefor your situation. You should consult with a professional where appropriate. Neither the publisher norauthor shall be liable for any loss of profit or any other commercial damages, including but not limitedto special, incidental, consequential, or other damagesFor general information on our other products and services or for technical support, please contact ourCustomer Care Department within the United States at(800)762-2974, outside the United States at(317)572-3993 or fax(317)572-4002.Wiley also publishes its books in a variety of electronic formats. Some content that appears in print maynot be available in electronic format. For information about wiley products, visit our web site atwww.wileycomLibrary of Congress Cataloging-in-Publication Data:Huber Peter JRobust statistics, second edition/ Peter J. Huber, Elvezio ronchettip. cnIncludes bibliographical references and indeISBN978-0-470-12990-6( cloth)1. Robust statistics. I. Ronchetti. elvezio. II. TitleQA276.H7852009519.5-dc222008033283Printed in the United States of america10987654321To the memory o1John w. tukeyThis Page Intentionally Left BlankCONTENTSPrefacePreface to first editionGeneralities1 Why robust Procedures1. 2 What Should a robust procedure achieve?1.2.1 Robust. Nonparametric and Distribution-Free1.2.2 Adaptive procedures1.2.3 Resistant Procedures1.2. 4 Robustness versus Diagnostics1.2.5 Breakdown point1.3 Qualitative Robustness567888911. 4 Quantitative Robustness1.5 Infinitesimal Aspects141.6 Optimal Robustness171.7 Performance Comparisons18CONTENTS1.8 Computation of robust estimates181.9 Limitations to Robustness Theory202 The Weak Topology and its Metrization23eneral remarks232.2 The Weak Topology232.3 Levy and prohorov metrics272.4 The bounded Lipschitz metric322.5 Frechet and Gateaux derivatives366 Hampels Theorem413 The Basic Types of Estimates453. 1 General Remarks453.2 Maximum Likelihood Type Estimates(M-Estimates)3.2.1 Influence Function of m-estimates73.2.2 Asymptotic Properties of M-Estimates483.2.3 Quantitative and Qualitative Robustness of MEstimates3.3 Linear Combinations of Order Statistics(L-Estimates)3.3.1 Influence Function of -Estimates3.3.2 Quantitative and Qualitative robustness of l-Estimates 593. 4 Estimates Derived from Rank Tests(R-estimates3.4.1 Influence Function of R-Estimates623.4.2 Quantitative and Qualitative robustness of R-Estimates 643.5 Asymptotically Efficient M-, L,and R-Estimates674 Asymptotic Minimax Theory for Estimating Location4.1 General remarks4.2 Minimax bias4.3 Minimax Variance: Preliminaries744. 4 Distributions minimizing fisher Information764.5 Determination of Fo by Variational Methods814.6 Asymptotically Minimax M-Estimates914.7 On the minimax Property for L-and R-estimates954.8 Redescending m-estimates74.9 Questions of Asymmetric Contamination101CONTENTSScale Estimates1055.1 General remarks1055.2 M-Estimates of scale1075.3 L-Estimates of scale5.4 R-Estimates of Scale1125.5 Asymptotically efficient Scale estimates1145.6 Distributions Minimizing fisher Information for Scale5.7 Minimax Properties116 Multiparameter Problemsin Particular Joint Estimationof Location and scale1256. 1 General remarks1256.2 Consistency of M-Estimates1266.3 Asymptotic Normality of M-Estimates1306. 4 Simultaneous m-Estimates of Location and scale1336.5 M-Estimates with Preliminary Estimates of Scale1376.6 Quantitative robustness of Joint Estimates of Location and Scale 1396.7 The Computation of M-Estimates of Scale14368Studentizing1457 Regression1497. 1 General remarks1497. 2 The Classical Linear Least Squares Case1547. 2.1 Residuals and Outliers1587.3 Robustizing the Least Squares Approach1607.4 Asymptotics of robust regression Estimates163741 The Cases hp2→0 and hp→07.5 Conjectures and Empirical Results1687.5.1 Symmetric Error Distributions1687.5.2 The Question of Bias1687.6 Asymptotic Covariances and Their estimation1707. 7 Concomitant Scale estimates1727.8 Computation of Regression M-Estimates1757.8.1 The Scale Step1767.8.2 The Location Step with Modified residuals1787.8.3 The Location Step with Modified Weights179CONTENTS7.9 The Fixed Carrier Case: What Size hi?1867. 10 Analysis of Variance1907. 11 LI-estimates and Median polish1937. 12 Other Approaches to Robust Regression1958 Robust Covariance and Correlation Matrices1998. 1 General remarks8.2 Estimation of Matrix Elements Through robust Variances2038.3 Estimation of Matrix Elements Through robust Correlation2048.4 An Affinely equivariant approach2108.5 Estimates Determined by Implicit Equations2128.6 Existence and Uniqueness of Solutions2148.6. 1 The Scatter estimate v2148.6.2 The Location estimate t2198.6.3 Joint Estimation of t and y2208.7 Influence Functions and Qualitative robustness2208.8 Consistency and asymptotic normality2238.9 Breakdown Point48.10 Least informative distributions2258.1058. 10.2 Covariance2278.11 Some Notes on Computation2339 Robustness of Design2399.1 General remarks2399.2 Minimax Global Fit9.3 Minimax Slope24610 Exact Finite Sample Results24910.1 General Remarks24910.2 Lower and Upper Probabilities and Capacities25010.2.1 2-Monotone and 2-Alternating Capacities25510.2.2 Monotone and Alternating Capacities of Infinite Order 25810.3 Robust Tests25910.3. 1 Particular Cases26510.4 Sequential Tests267
- 2020-12-03下载
- 积分：1
STM32移植到GD32的注意事项
GD32的性价比和稳定性都很高。我更喜欢用GD32.。但不是100%兼容的。需要注意一些细节。2.内部 Flash1)芯片设置读保护用法描述由于GD的Fash是自己的专利技术,STM的Fash是第三方提供的,所以GD的 Flash和STM的Fash有些许差异。GD的擦除时间会长一点解决方法在写完KEY序列以后,需要读该位,确认key已生效。所以,这里应该插入While(! (FLASH->CR &0x200);//Wait OPTWRE或可简单插入两个NOPNOPONOPO在ST库中,只有FLASH Status FLaSH Erase Option Bytes(void)FLaSH Status FlaSH ProgramOption Byte Data(uint32 t Address, uint8 t DataFLASH Status FLASH_ Enable WriteProtectionuint32 t FLASH PagesFLASH Status FLASH ReadOutProtection (Functional State New State)四个函数需要修改。2)IAP在应用中编程描述GD32由于有fash的0访问时序,同SM32在Fash的Ease和 Program上存在差别,GD32的 Erase和 Program时间比STM32的稍微长些建议对 Erase和 Progran时间进行修改解决方法将宏定义井 define erase timeout(luint32 t)OX000B000O)#define ProgramTimeout(uint32_t)ox00002000修改为:#define erasetimeout((uint32_t)OX00OFFFFF#define ProgramTimeout(uint32 t)OxOOOOFFFF备注: Erase和 Program时间宏定义在stm3210 x flash. c源文件中路径: braiesSTM32F10 x Std Periph Driversr)3)用IAR下载配置解决方法在批量牛产的时候首先会烧写一个USB的boot,这个boot自动运行后在由上位机软件进行烧写应用程序。如果boot程序不能自动运行则需要重新插拔次电源。给生产造成一些麻烦。产生不能自动运行程序的原因是如果程序设置读保护的话需要等待 FLASH CR的第9[ OPTWRE]位为1.如果没有置位的话继续执行就会出错。由」ST的执行速度慢,程序执行到读 FLASH CR寄存器的时候该位已经置1,GD的执行速度比较快,程序运行到这的时候该位还没置1,因此需要在 FLASH ReadOut Protection函数里面添加一些轮询该位为1或者加延时3.ISP烧写软件1)ISP烧写,建议使用官方烧写软件性述芯片内部同有区别解决方法建以到下载最新版本的另外也有专门的烧写软件(可以到论坛下毂如果使用自制的软件或脱机编程器,实现和完全兼容,建议修改以下参数。页擦除等待超时时间增加至整片擦除等待超时时间增加至左右字编程等待超时时间增加至,臾编程等笭超时时间增加至I/0日1)I0口外部中断使用方法措述在关闭期间,如果外部引脚有电平的变化,在使用MR打开中断后会马上进入中断服务程序。理论是打开中断前,不管管脚是否有电平的变化,都不会影响到打开后的中断响应。解决方法所以解决方法就是通过禁用上升沿或者下降沿检测寄存器来开关中断,不能使用IMR屏蔽奇存器。程序如下关闭沿检测,以达到关闭中断的目的,下降沿使用寄存器,上升沿使用寄存器2)在待机模式,PA8引脚特殊设置描述在使用低功耗的情况下,PA8会被MCU在内部被设置为地PA8复用为MU内部频率输出,超低功耗设时需要悬空解决方法在待机模式,PA8悬空不用3)低功耗下必须注意描述在使用低功耗情况下,把软件全部端口(AF)时钟关掉,无论是否有该端凵。4)当有脉冲群冲击管脚摧述需要在在进入中断后关闭中断4.定时器1)定时器输入捕获模式需要软件清中断描述sTM定吋时器输入捕获模式默认能硬件清中断,GD为了更加严格要求配置,需要做软件清中断解决方法软件清除标志位2)定时器向上脉冲计数模式设置述定时器的用法差异解决方法脉冲计数模式下,装载值必须设置为比预期值大,否则不计数在ST上如果重载值不设置(初始为0)的时候,CNT可以正常计数。在GD上如果重载值不设置保持初始为0的时候,会因为重载值为零,即便是来一个脉冲也会导致所有的寄存器复位从而不能正常计数。型号GD32F1系列MCUF|ah256B8及以上的型号)3)TIM、ADC模块描述Tmer、ADC模块的触发信号宽度要求解决方法|由于内部有高速和低速两条外围总线,Tmer、ADC模块和其他外设共同使用这两个总线。GD32F103/101系列Fash128KB及以下的型号, Timer、ADC等模块识別触发信号的条件是触发信号宽度大于模块所在总线的时钟宽度5.串口 USART1) USART连续发送数据字节有空闲位带述字节间有空闲位解决方法|对于一般的通讯米说,不会有影响,只对于一般在通讯上有特殊协议的,才会产生数据不准确的情况所以,特定情况,修改程序6.I2C总线1)硬件L2C特殊配置述GD的C相对STM的来说要少一个标志位解决、宏地址定义改交方法2、硬件I2C在会在向从机发送7bits地址完成后,从札还没来得及识别。(看客户应用)我们可以在发送完7bits后加个延时,让从机完全识别I2C Send7bitAddress(I2C1, EEPROM ADDRESS, I2C Direction TransmitterintOfffhile(i --3、检测ADDR不能使用I2 C Checkevent函数,因为他会清除ADR,可以使用I2 C Get FlagStatus函数就是把while(! I2C CheckEvent(I2C1I2C EVENT MASTER TRANSMITTER MODE SELECTED))Ey while(! I2C GetFlagStatus(I2C1, I2C FLAG ADDR))4、还有个关于编程步骤的严谨性,跟STW想比,我们是先 Clcar ack,再 Clear Arrd。7.ADC采集1)ADC采样设置述ADC启动解决方法|分三个方面时写入后,需要等待一段时间,如果用库的话就在 ADC CMD后面加20us左右的延付如果采用中断获得采样数据后,需要软件清除中断。8. SDio1) SDIO DAT3pin的在1 bit bus mode和4 bit bus mode下的配置摧述1、SD|O在1 bit bus mode下,DAT3pin是低电平,这样会导致 SD Card进入SP!模式。原因:初始化失败的原因主要是因为GD32的芯片SDO的DAT3∏存在BUG2、在4位模式下,通过上面的方法,程序能止常初始化,但不能正常读写SD卡原因:因为DAT3∏在前面已经配置成推挽输出,所以在4位模式下,不危正常读下。在调用4位模式前,把DAT3的端凵配置成复用推挽输入即可解决问题解决方法「1、1 bit bus mode的解决方法:建议在使能之前,先把配置成推挽输出,)且要置成高电平,使保持高电平即可2、4 bit bus mode的解决方法:在调用4位模式前,把DAT3的端口配置成复用输出即可解决问题。2)程序在刚烧完后能正常读写SD卡,断电再上电后,SD卡初始化失败,需要手动复位一次后才正常描述在某些SD卡中,GD32断电再上电,会引起SD卡上的时钟信号不正常,导致SD卡发送命令失败。解决方法在程序中,打开时钟后,增加一小段延时,以保证下时钟信号稳定。这个延时添加的地方:在即的配置文件中,然后在这个函数中找到就在这个后面加个延时。10. USBA, USB OTG1)客户使用的原工程时需要注意几点解|1、在中,增加如下图红色字体语句for (1=0; 18; i++) EPli= GetEndPoiNT(i)for(i=0:iregs. HC REGS [num]->HCCHAR, hcchar d3 2)pdev->host hc Status =HC NAK而V2.1.0版本的NAK处理过程如下else if (hcint b nak)if(hcchar b. eptype = EP TYPE_ INTR)UNMASK HOST INT CHH(num)USB OTG HC Halt(pdev, numelse if ((hcchar. b. eptype = EP TYPE CTRL)(hcchar b eptype = EP_ TYPE BULK))A re-activate the channel *hcchar, b chen =1hcchar b chris =0USB OTG WRITE REG32(&pdev->regs. HC REGS [num]->HCCHAR, hcchar d32)pdey->host HC Status [num]=HC_NAKCLEAR HC INT(hcreg, nak)唯一的区别就是 CLEAR HC INT( here,nak)的位置,在Ⅵ1.0.0版本中对于CTRL和BUK端点的NAK中断没有清除NAK,我们的芯片会因此产生多次IN传输的请求,导致数据传输错误。改为V2.1.1的写法后传输正常。(注意 HC Status在V2.1.0是数组,在Ⅵ1.0.0是单个数据,直接拷贝的话要去掉后面的[num])B.USB外设的工作频率有限制摧述有最低工作频率的要求,也就是APB1分频后的时钟必须大于12MHz,比如HCLK为56MHz,APB1的最大分频系数为4,56/4=14MHz,可以正常工作。11 SPI1)输入与输出配置要求(STM32不需要如此要求)解决丨GD32在使用SP时,o的配置必须严格遵守主从模式下的输入与输出配置,而方法STM32无此要求,相关代码如下主机模式下|o配置(主机以SP为例):GPIO InitStructure gPio Mode gPio Mode af plGPIO_ Init Structure GPIo Speed GPlO Speed 50MHzGPIO InitStructure. GPio Pin= GPlO Pin 5 GPIO Pin_ 7;GPIO Init(GPIOA, &GPIO InitstructureGPio Init Structure gPio Mode gPio Mode IN floating:GPio InitStructure gpio Pin gpio pin 6GPIO Init(GPIOA, &GPIO InitStructure)从机模式下o配置(从机以SP2为例)GPIo Init Structure GPio Mode gPlo Mode IN FloatingGPIO InitStructure GPIO Speed= GPl Speed 50MHzGPIO_InitStructure GPIO_ Pin GPIO Pin_13 GPIO_ Pin_15GPIO Init(GPIOB, &GPIO InitStructure)gPio Initstructure gpio mode gpio mode af pp.GPIO InitStructure. GPio Pin= GPIo Pin 14:GPIO_Init(GPIOB, &GPIO_Initstructure);3)在GD32的SP的时钟信号,空闲状态需要配置成高电平,以保证数据的稳定性,具体代码如下:红色字体代码解决SPI_ InitStructure SPl Direction =SPI_ Direction_ 2Lines fullDuplex;方法SPI InitStructure SPl Mode SPi Mode master.SPI Initstructure SPl Data Size= SPl Data Size 8bSPlInitStructure SPl_CPOL= SPI CPOL HighSPl Initstructure SPl CPHA= SPI CPHA 2EdgeSPI InitStructure SPI NSS= SPI NSS SoftSPI InitStructure SPl BaudRate Prescaler =SPI Baud Rate Prescaler 256:SPI Initstructure, Spi FirstBit= SPI FirstBit MSBSPI InitStructure SPl CRCPolynomial =7;SPl Init(sPl1, &SPl Initstructure);4)当作为从机时,在GD32中,时钟信号必须为8的整数倍。例如:红色字体代码解决SPI_InitStructure SPl_ Direction =SPl_ Direction_ 2Lines_ Full Duplex;方法SPI InitStructure. SPl Mode= SPl Mode MasterSPI InitStructure SPSPi Data Size 8SPl InitStructure SPl CPOL= SPI CPOL High;SPI Init Structure. SPl CPHA SPI CPHA_ 2EdgeSPI Initstructure SPl NSS= SPI NSs SoftSPI Initstructure Spl BaudRatePrescaler= SPl BaudRatePrescaler 256SPI InitStructure. SPl FirstBit= SPl First Bit MSBSPl Initstructure SPl CRCPolynomial =7;SPI Init(SPI1, &SPl InitStructure)5)在GD32中,不能使用SPLS_FLAG_BSY该位来判断SP总线数据是否接收或发送完成12.看门狗1)进入SToP模式前打开看门狗,通过RTc的ALR唤醒后,程序会不断被复位的现像摧述WDG内部有个 Reload信号,KEY奇器写AAA会使其拉高,过段时间自动拉低。在拉底之前进入STOP状态会使 Reload信号一直为高,等到退出SToP后也保持为高,之后再写AAAA没有办法让 Reload产生上升沿,也就没办法更新计数器了解决方法「进STOP之前不要 Reload,也可以调整下程序的顺序,把WwDG的配置放到RIC配置之前,效果是一样的。
- 2021-05-06下载
- 积分：1
