THE OXFORD SERIES IN ELECTRICAL AND COMPUTER ENGINEERING ‘Adel S. Sedra, Series Editor, Electrical Engineering Michael R. Lightner, Series Editor, Computer Engineering Allen and Holberg, CMOS Analog Circuit Design Bobrow, Elementary Linear Circut Analysis, 2fd Ed Bobrow, Fundamentals of Electrical Engineering, nd Ed Campbell, The Science and Engineering of Microelectronic Fabrication Chen, Analog and Digital Control System Design Chen, Linear System Theory and Design, 3rd Fa Chen, System and Signal Anais, 2nd Ed Comer, Digital Logic and State Machine Design 3rd il Cooper and MeGillem, Probabilistic Methods of Signal and System Analysis, 3rd Ei Fortney, Principles of Electronics: Analog & Digital Franco, Electric Circuits Fundementals Granzow, Digital Transmission Lines Guru and Hizrogl, Eecric Machinery & Transformers, 2nd Fi Hoole and Hoote,A Modern Short Course in Engineering Electromagnetics Sones, introduction to Optical Fiber Communication Systems Kaein, Elements of Power Electronics Kuo, Digital Control Systems, 3rd Ed Lathi, Modern Digital and Analog Communications Systems, 3rd Ei MeGillem and Cooper, Continuous and Diserete Signal and System Analysis, 3rd Ed Mines. Lines and Electromagnetic Fields for Engineers Robers and Sedra, SPICE, 2nd Ea. Roulston, An Introduction othe Physics of Semiconductor Devices Sadik, Elements of Electromagnetics, 2nd Ed Santina, Stubberud, and Hostetter, Digital Control System Design, 2rd Ed Schwarz Electromagnetics for Engineers Schwarz and Oldham, Electrical Engineering: An Introduction, 2nd Ed. Seda and Smith, Microelectronic Circuits, 4 Ed. Stefani, Savant, Shahan, and Hostetter, Design of Feedback Control Systems, 3rd Ed Van Valkenburg, Analog Filter Design Wamer and Grung, Semiconductor Device Electronics Wotovich, Automatic Control Systems Yariv, Optical Electronics in Modern Communications, Sth Ed LINEAR SYSTEM THEORY AND DESIGN Third Edition Chi-Tsong Chen State University of New York at Stony Brook New York Oxford ‘OXFORD UNIVERSITY PRESS 199‘OXFORD UNIVERSITY PRESS. Oxford New York ‘Aiess Auckland Bangkok Bogoei Buenos Alves Ceuta ‘cage Town Chenaat Dares Salm Deki Florence Hong Kong. tarbul Xorchi Koala Lumpur Madrid Melbourne Mexico iy Mumbai Nici Pais Sio Plo Singapore Tipet Tokyo Toronto Waray and associ companies in Betin Thadsn Copyright © 1999 by Oxford University Pres. ne Publsted by Oxfors Univecsity Pres, ne 198 Madison Avenue, New York, New York 10016 1 is registered traderark of Oxford University Press Allright eer No prof this publication may be reprodiced, stored in a reival system, of wansited, ‘any former by any means, eleconic, mechanical hotsopying, eonting, or ebervise, without the prior ‘Permison of Oxford University Press. Library of Congress Catalgiog-in-Publicaton Data (ven, Ch-Tong Linear system theory and esign by ChiTong Chen, ede ._cm.— (The Oxford seis in electrial and computer engineering) Includes bibliographical references and inden. ISBN 0-19-511777-8 cot, L-Liesraysems, 2 System design, Tile. 1 Series. aware 1998, 29 83221 orss3s cr Pring at digit 9 #76 S45.2 1 Prise inte United States of Ameria on aches paper to Bit-JauContents Preface xi Chapter 1: Introduction 1 1.2 Introduction 1 12 Overview 2 Chapter 2; Mathematical Descriptions of Systems 5 2.1 Introduction 5 2.1.1 Causality and Lumpedness 6 22 Linear Systems 7 23 Linear Time-Invariant (Lip Systems 11 23.1 Op-Amp Ctcuit impiementation 16 24 Linearzation 17 25 Exomples 18 25.1 ALC Networks 26 26 Discrete-Time Systems 31 27 Concluding Remarks 37 Problems 38 Chapter 3: Linear Algebra 44 3.1 introduction 44 32. Basis, Representation, and Orthonermalzaton 45 33 Unear Algebraic Equations 48 34 Similarity Transformation 53 35 Diagonal Form and Jordan Form 55 36 Functions of a Square Matrix. 67 37. Lyopunov Equation 70 38 Some Useful Formulas 77 39. Quadratic Form and Positive Defniteness 73 3.10 Singular-Value Decomposition 76 3.11 Norms of Matices 78 Problems 78CONTENTS Contents ix Chooter 4: State-Space Solutions and Realizations 86 7.3 Computing Coprime Fractions 192 73.1 @R Decomposition 195 4.1 Induction 86 7.4 Balanced Realization 197 42 Solution of Ul State Equations 67 75 Reolzations ftom Markov Perameters 200 421 Dscretaotion 90 76 Degree of Transfer Matrices 205 42.2 Solution of Dscretevime Equations 92 77 Mirimal Realzations—Matrx Case 207 43 Equvaient State Equations 93 718 Matrix Polynomial Fractions 209 43.1 Cononicalforns 97 78.1 Colurm ond Row Reducedness 212 43.2 Magnitude Scalng in OprAmp Circaits 98 7.82 Computing Matix Coprime Fractions 214 44 Realzctons 100 7.9 Realzations ftom Matix Coprime Fractions 220 45. Solution of Linear Time Varying (LIV) Equations 106 710 Reotzctione from Matix Manco Poramotore 225 451 DscreteTme Cove 710 7.11 Concluding Remarks 227 46 Equivalent Time-Varying Equations 17 Leena 47 Time-Varying Realizations 115 Problems 117 ' Chapter 8: State Feedback and State Estimators 237 Chapter §: Stability 127 8.1 Inkoduction 231 82 Stote Feedback 232 8.1 Introduction 121 18211 Solving he Lyapunov Equation 239 52. Input-Output Stobiity of LN Systems 127 83 Regulation ond racking 242 52.1 Decrete-Time Case 126 £8311 Robus Tracking end Dsturbonce Rejection 243 53 Internal Stability 129 832Stobiaton 247 531 Dscrete-Time Cose_131 84 Stote Estimator 247 5.4 Lyapunov Theorem 132 8.4.1 Reduced-Dimensional state Estimator 251 541 Dicretectime Cose 135 85 Feedback ftom Estimated Stotes 253 85. Stabilty of LTV Systems 137 86 State Feedback—Multivarioble Cose 255 Problems 140 Ba Cycic Design 256 882 Lyapunov.Equation Method 259 863 CononicalForm Method 260 3.44 Effect on Transfer Motices 262 143 87. State estimators—Multwariable Cose 263 88 Feedback trom Est mated States—Mutivariabie Case 265 Problems 266 Chapter 6: Controllability and Observability 6.1 Introduction 143 62 Controlabilty 144 62.1 Controtabity Indices 150 63 Observabilty 153 | 83.1 Observabiity Indices. 157 | 64 Canonical Decompostion 158 me 8.1 Introduction 269 65. Conditions in Jordan-Form Equations 164 i estos Nene ae ee eae 66 Discrete-Time State Equations 169 f 7 92 Unily-Feedback Configuration-Pole Placement 273 ‘ontrllabilty to the Orginand Reachobiy 1 ae a eT ctr 921 Raguotion ond racking 275, Chapter 9: Pole Placement and Model Matching 269 67 Controlabiity AfterSamping 172 92.2 Roost 7 be i jobust Tracking ond Ostueonce Rejection 277 68 LIVState Equations 176 | 9.2.3 Embedding temas Models 280 Problems 180 | 9.3 implementable Transfer Functions 283 913.1 Model Matening-Two-Porometer Configuration 286 9.32 Implementation of Two-Porameter Compensators 297 Chapter 7: Minimal Realizations and Coprime Fractions 184 9.4. Multivariable Unity Feedback Systems 292 i 9.4.1 Raguaton and lacking 302 7.1 Introduction 184 | 9.42 Robust Tracking and Distubance Rejection 303 7.2 implicatione of Coprimencss 185 9.5. Multivariable Model Matching-Two-Porameter 72.1 Minimal Realeations 189 Configuration 306CONTENTS| 95.1 Decouping 317 9.6 Concluding Remarks 314 Problems 315 References 319 Answers to Selected Problems 321 Index 331 Preface ‘This text is intended for use in senir/irst-year graduate courses on linear systems and ‘multivariable system design in electrical, mechanical, chemical, and aeronautical departments. Itmay aiso be useful to practicing engineers because it contains many design procedures. The ‘mathematical background assumed is a working knowledge of linear algebra and the Laplace transform and an elementary knowledge of differential equations. Linear system theory isa vast eld. In this text, we limit our discussion to the conventional approaches of siate-space equations and the polynomial fraction method of transfer matrices. ‘The geometric approach, the abstract algebraic approach, rational fractions, and optimization axe not discussed We aim to achieve two objectives with this text. The first objective is to use simple and efficient methods to develop results and design procedures. Thus the presentation not exhaustive. For example, in introducing polynomial fractions, some polynomial theory such as the Smith-MeMillan form and Bezout identities are not discussed. The second Objective of this text is to enable the reader to employ the results to carry out design Thus most results are discussed with an eye toward numerical computation, All desiga procedures in the text can be carried out using any software package that includes singular value decomposition, QR decomposition, and the Solution of linear algebraic equations and the Lyapunov equation. We give examples using MATLAB®, as the package! seems tobe the most widely available. This edition isa complete rewriting of the book Linear System Theory and Design, which was the expanded edition of Introduction to Linear Sytem Theory published in 1970. Aside from, hopefully, a clearer presentation and a more logical development this edition differs from the book in many ways 1s The order of Chapters 2 and 3 is reversed. In this edition, we develop mathematical descriptions of systems before reviewing linear algebra. The chapter on stability is moved carler. ‘« This edition deals only with real numbers and foregoes the concept of fields. Thus itis mathematically less abstract than the original book. However, all results are still stated as theorems for easy reference. «# In Chapters 4 through 6, we discuss first the time-invariant case and then extend it to the time-varying case instead of the other way around. 1, MATLAB sa registered rademark of the MathWorks In 24 Prime Pak Way, Natick. MA 01760-1500 Phoe: s0$0517 700, fat 508-517 9001, E-mail nfo@mathworks. com, hws mauwors com. xx PREFACE + The discussion of discrete-time systems is expanded, + In sate-space design, Lyapunov equations are employed extensively and multivariable canonical forms are downplayed. This approach is not only easier for classroom presentat but also provides an attractive method for numerical computation, The presentation of the polynomial fraction method is streamlined. The method is equated with solving linear algebraic equations. We then discuss pole placement using a one-degree- of-freedom configuration, and model matching using a two-degree-of-freedom configura- Examples using MATLAB are given throughout this new edition, “This ton is geared more for classoom use and enginerng aplication: therefore, many topicsin he oial book are deleted, including srt system egtvalence. deterministic den Sean, computational issues, some multivariable canonical forms, and decoupling by state feedback The polynomial faction design inthe inpudoutpu feedback (contollefestmator) configuration is deleted. stead we discuss design inthe two-parameter configuration, Ths coniiguation seems tobe more suitable for practical application The eight appendices in the original book are ether incorporated into the text or deleted. “The logical sequence ofl chapters i 5 follows Chap. 8 chap. 15 = { oto. 6 = [ERE > Sec. 71-73 => See, 9.1-93 = Sec. 16-78 = Sec. 94-95 In addition, the material in Section 7.9 needed to study Section 8.6.4, However, Section 8.6.4, ‘may be skipped without loss of continuity. Furthermore, the concepts of controllability and bservabilty in Chapter 6 are useful, but not essential forthe material in Chapter 7. AI minimal realizacions in Chapter 7 can be checked using the concept of degrees, instead of checking controllability and observability. Therefore Chapters 6 and 7 are esseatially independent This text provides more than enough material for a one-semester course. A one-semester course at Stony Brook covers Chapters | through 6, Sections 8.1-8.5, 7.1~7.2, and 9.1-9.3, ‘Time-varying systems are not covered. Clearly, other arrangements are also possible for a fone-semester course, A solutions manual is available from the publisher. am indebted to many people in revising this text. Professor Imin Kao and Mr, Juan Ochoa helped me with MATLAB. Professor Zongli Lin and Mr. T. Anantakrishnan read the whole manuscript and made many valuable suggestions. I am grateful to Dean Yacov Shamash, College of Engineering and Applied Sciences, SUNY at Stony Brook, for his encouragement. The revised manuscript was reviewed by Professor Harold Broberg, EET Department, Indiana Purdue University; Professor Peyman Givi, Department of Mechanical and Aerospace Engineering, State University of New York at Buffalo: Professor Mustafa Khammash, Department of Electrical and Computer Engineering, Iowa State University; and Professor B. Ross Barmish, Department of Electrical and Computer Engineering, University cof Wisconsin. Their detailed and critical comments prompted me to restructure some s ang to mclude a number of mechanical vibration problems. {thank them all Preface xi am indebted to Mr. Bill Zobrist of Oxford University Press who persuaded me to undertake this revision, The people at Oxford University Press, including Krysia Bebick, Jasmine Urmeneta, Terri O°Prey, and Kristina Della Bartolomea were most helpful in this undertaking. Finally, I thank my wife, Bih-Jau, for her support during this revision, (Chi-Tsong ChenLINEAR SYSTEM THEORY AND DESIGNChapter “4 i { ok Introduction 1.1 Introduction ‘The study and design of physical systems can be cartied out using empirical methods. We can apply various signals toa physical sysem and measure its responses. If te performance fs not satisfactory, we can adjust some of is parameters or connec oi a compenstor to improve its performance. This approsch ries heavily on pastexperience andi eried out by tial and tor and has succeeded in designing many physical systems, [Empirical methods may become unworkable if physical sytem are complex or too expensive or too dangerous tobe experimented on. [nthese cases, analytical methods become indispensable. The analytical study of physical systems consists of love pars: modeling, development of mathematical descriptions, analysis, and design, We Eiefly introduce each ofthese tasks. “Thedistnction between physical systems and modelsis basic in engineering Forexample. circuits or contol systems stadied in any textbook are models of physiil systems. A resistor ‘ith a constant resistance is 8 model: it will burn out if the applied voltage is over a limit. This power limitation soften disregarded in its analytical stay. Am inductor with a constant inductance is again a model in reality, the inductance may vary with the amount of current flowing through t. Modeling is avery important problem, forthe success ofthe design depends con wheter the physical system is modeled properly. A physical system may have diferent models depending onthe questions asked, It may alsobe modeled differently in ditferentoperational ranges. Forexample, aneleconicamphiier is modeled differently at high and low frequencies. spaceship can be modeled as patcle in investigating its trajectory; however, t mist be modeled asa rigid tody in manctsering A spaceship may even be modeled asa flexible body when itis connected toa space tation. In order to develop a suitable model fora physical system, a thorough understanding of the physical system and its operational range is essential. tn this text, we wll call a move ofa physica system simply a system. Thos a physical system i a device o acollection of devices existing in the real world a system isa model ofa physical sytem Once a system (or model is selected for @ physical system, the next step i to apply various physical laws to develop mathematical equations 10 describe the system. For ex- ample, we apply Kirchhott's voltage and current laws to electrical systems and Newion's Taw to mechanical systems. The equations that describe systems may assume many forms:INTRODUCTION they may be linear equations, nonlinear equations, integral equations, difference equation, differential equations, or others. Depending on the problem under study, one form of equation ‘may be preferable to another in describing the same system, In conclusion, a system may have diferent matbematical-equation descriptions just asa physical system may have many differen models ‘After a mathematical description is obtained, we then carry out analyses—quantitative andéor qualitative. In quantitative analysis, we are interested in the responses of systems excited by certain inputs. n qualitative analysis, we ate interested in the general properties of systems, such as stability, controllability, and observabilty, Qualitative analysis is very ‘importa, because design techniques may often evolve from this svidy. IF the response of a system is unsatisfactory, the system must be modified. In some cases, this can be achieved by adjusting, some parameters of the system: in other cases, compensators must be introduced, Note thaf the design is carried out on the model of the physical system. If the model is properly chosen, then the performance ofthe physieal system should be improved by introducing the required adjustments or compensators, Ifthe model is poor, then the performance of the physical system may not improve and the design is useless. Selectrg model that is close enough toa physical system and yet simple enough tobe studied analytically isthe most difficult and important problem in system design 1.2 Overview The stuly of systems consists of four parts: modeling, setting up mathematical equations, analysis, and design. Developing models for physical systems requires knowledge of the particular field and some measuring devices. For example, to develop models for transistors requires a knowledge of quantum physics and some laboratory setup. Developing models for automobile suspension systems requires actual testing and measurements; it cannot be achieved by use of pencil and paper. Computer simulation certainly helps but cannot replace ‘actual measurements. Thus the modeling problem should be studied in connection with the specific field and cannot be properly covered in this text. In this text, we shall assume that models of physical systems are avsilable to us. “The systems to be studied in this text are limited to linear systems. Using the concept of linearity, we develop in Chapter 2 that every linear system can be described by v= f Geucyae ay ‘This equation describes the relationship between the input w and output y and is called the ‘nput-ourpue or external description. fa linear system is lumped a8 well then it ean also be described by 3(0 = A@xir) + BOQ) “a (0) = Cieyxts) + DEUCE) a3 Eaquatin (1.2)is asetof first-order differential equations and Equation (1.3) isa setof algebraic equations. They are called the internal description of linear systems. Because the vector x is called te state, the set of two equations is called the state-space or. simply, the state equation, 1.2 Overview 3 fa tinear system has, in addition, the property of time invariance, then Equations (1.1) through (1.3) reduce to yo 7 Ge Duley dr ay b and X() = AN() + Bucy as) yin = Cx(7) + Duis) 6) For this class of linear time-invariant systems, the Laplace transform is an important tool in analysis and design. Applying the Laplace transform to (1.4) yields G(syats) i) i where a variable with eirsumfex denotes the Laplace transform ofthe variable. The furtion Ge iscaled he ronsfer matric Both.) and Tareinpat-outputor extemal escripsons. “The former is said tbe in the me domain and the later inthe frequency domian Equation (1.1 through (1.5) are cll continuous-time equations bscause their ime variable ris continuum defined st every time instant in (~20. 20). H the time is dened only at discret instants then the coresponding equations are ealed discrete-time equations. ‘This text is devoted tothe analysis and design centered around (U1) through (1.7) nd bee Aiserete-time counterparts ‘We briefly discss the contents of each chapter. In Chapter 2. after introducing the aforementioned equations from the concepts of lumpednes, linearity, and time invariance. te show how these equations canbe developed to describe systems. Chapter 3 eviews eat algebraic equations the Lyapunov equation, and other pertinent topes that are sential fr thistexe. We azo trace the Jordan frm becssse it wil be wsed o establish a numberof resus, Westudy in Chapter soltion ofthe state-space equations in 2)and 15). Diterent Analyses may lead todifferent stat equation tat describe the same system. Ths weinroduce the concep of equivalent stat equations. The basi relationship between state-space equations and tansfer matrices is lo established, Chapter 5 intodces the concepts of bounds bounded-outpat (BIBO) stability, marginal stability. and asymptotic staiity. Every s}stem thust be designed tobe suble; otherwise it may burn out or disintegrate, Therefore sity isa base system concept, We ao introduce the Lyapunor theorem t cheek asympiotic stability Chapter 6 introduces the concepts of controllability and observabiliy. They are esenil in studying the internal structure of systems. A fundamental result is that the transfer matrix: desries only the contolaie and observable pat of sate equation. Chapter 7 sues ‘inimal reliations and introduces cprime polynomial fractions. We show how wo cbain Coprime fractions by solving sets of linear algebrale equations. The equivalence of contllble an observable sate eqistions and coprime polynomial factions is extabishe. “The last two chapters discuss the desigh of time-invariant systems. We use contolable and observable state equations to camry out design in Chapter 8 and use soprime pols nomial fractions in Chapter 9. We show that, under the conrlibility conon, all egemlues of a sysiem can be arbitrarily asigned by introducing state feedback. Ia slate equation fs observable, full dimensional and reduced-dimensional state estimat's, with any desired4 INTRODUCTION eigenvalues, can be constructed to generate estimates of the state. We also establish the separation property. In Chapter 9, we discuss pole placement, model matching, and their applications in tracking, disturbance rejection, and decoupling. We use the unity-feedback configuration in pole placement and the two-parameter configuration in model matching. Inour esign, no control performances such as rise time, Settling time, and overshoot are considered: neither are constraints on control signals and on the degree of compensators. Therefore this isnot a control text per se. However, all results are basic and useful in designing linear time- invariant control systems Chapter aunt Mathematical Descriptions of Systems 2.1 Introduction ‘The class of systems studied in this text is assumed to have some input terminals and output terminals as shown in Fig. 2.1, We assume that if an excitation or input is applied to the input terminals, a unique response or output signal can be measured at the output terminals. This unique relationship between the excitation and response, input and output, or eause and effect is essential in defining a system. A system with only one input terminal and only one output terminal is called a single-variable system or a single-input single-output (SISO) system. A system with two or more input terminals and/or two or more output terminals is called a multivariable system, Moze specifically, we can call a system a mult-input multi-output (MIMO) system if it has two or more input terminals and ourput terminals, a single-input rulti-output (SIMO) system if it has one input terminal and two or more output terminals. any we 0} a Back 0 at “a cS vel ah 4s ts 23458 Figure 2.1 System6 MATHEMATICAL DESCRIPTIONS OF SYSTEMS. A system is called a continuous-time system if it accepts continuous-time signals as its input and generates continuous-time signals as its ourput. The input will be denoted by lowercase italic u(t for single input or by boldface u(r) for multiple inputs. If the system has input terminals, then u(t) isa p x I vector oru = [uj 2 *-* up|’, where the prime denotes the transpose. Similarly, the output will be denoted by y(t) or y(t). The time r is assumed to range from —00 10 e. ‘A system is called a discrete-time system if it accepts discrete-time signals as its input and generates discrete-time signals as its output. All discrete-time signals in a system will be assumed to have the same sampling period T. The input and output will be denoted by u(k] := w(&T) and y{k] := y(&T), where k denotes discrete-time instant and is an integer ranging from —o0 o 2c. They become boldface for multiple inputs and multiple outputs. ' 2.1.1 Causality and Lumpedness Asystem scaled a memories ystem iit ouput y() depends only on the inp applied atin independet of the inpot applied before or after fy. This wil be stated sucinty as follows: curen ouput ofa memoryless sysiem depends only on eureninput:tis independent of pst and future inputs. A network that consists of only resistors isa memoryles system ost stems, however, have memory. By this we mean tht the ouput at depends on ula forr fy, Thats curentoutpucofasystem with memory may depend on pat, curent, and ftir inp, ' system felled causal or nonanticipatoy system fs current output depends on pastandcuret inputs but nt on futere input. Ifa system snot causal, then its curent output Will depend on future input. In other words, a aoncausa system can predict or anticipate whot Will be applied in the future-No physical system has such capability. Therefore every physical System is causal and causality isa necessary colton fora system o be builtor implemented inthe eal work. This text studies only causal systems Crrent output ofa causa syst Is affected by past np, How far back in me wil the past input affect the current output? Generally, the time should goal the way back to minus infin. In other words, the input from --20 t0 time 1 has an effect on y(1). Tracking WX) ftom: = ~o0 i ifnot impossible very inconvenient. The concep of state can deal with this problem Definition 2.1 The state xito) of a system at time ty isthe information at tp that, together withthe input u(t), fort > ty, determines uniquely the output y(t) for all > to By definition, if we know the state tf, thereis no more need to know the input u() applied before Zp in determining the output y(2) after fp. Thus in some sens, the state sumimarizes the effect of past input on future output. For the network shown in Fig. 2.2, if we know the voltages (tg) and x(t) across the two capacitors and the current xy (fa) passing through the inductor, then for any input applied on and after fo, we can determine uniquely the output for f > 1 ‘Thus the state of the network at time fy 38 2.2 Linear Systems 7 Figure 2.2. Nework with 3 sate variables. x() 9 (to) Iisa 3 x | vector. The entries of x are called state variables. Thus, in general, we may consider the initial state simply asa set of initial conditions ‘Using the state at, we can express the input and output of a system as x(i) w(t), £2 ty 2 Jar. 026 It means thatthe output is partly excited by the initial state at fo and partly by the input applied at and after‘. In using (2.1), there is no more need to know the input applied before f all the way back to ~o0. Thus (2.1) is easier to track and willbe called a state-input-output pair. ‘A system is said to be lumped if its number of state variables is finite or its state is a finite vector. The network in Fig, 2.2 is clearly a lumped system; its state consists of three rumbers. A system is called a distributed system if its state has infinitely many state variables, “The transmission line isthe most well known distributed system. We give one more example, EXAMPLE 2.1 Consider the unit-time delay system defined by 3) Sur) The output is simply the input delayed by one second, In order to determine {v(t}, 1 > to) from (u(t), £ = to. we need the information u(t), fp~ 1 < fol. The of the systemis {u(t fp ~ << fo}. There are infinitely many points in {fp = 1 = 1 < th Thus the unittime delay system is a distributed system. 2.2 Linear Systems A system is called a linear system if for every ty and any two state-input-output pairs XC) =O. 12% 4), 12% fori = 1, 2,we have8 MATHEMATICAL DESCRIPTIONS OF SYSTEMS (a) +4218) Sy) + YD, #2 6 (additivity) and axi(t) > fy (homogeneit =o pen} TM 12% (homogeneity) for eny real constant a. The first property is called the additivity propery, the second, the homogeneity property. These two properties can be combined a eax (4) + @2%2((0) avuy(t) +amur(t), 1 to | an tary, 126 is called for any real constants ay and a, ands called the superposition propery. A syste a nonlinear system ifthe superposition property doesnt hod Ifthe input wt) sdemclly er ort ty, then the output wl be exited exclusively bythe inal state (THs output sealed the Zroinpur response and wl be dened by xt) > If the initial state x(t) is zero, then the output will be excited exclusively by the input. This ‘outputs called the zero-state response and will be denoted by Ys OF xtc) uit), £210 Jone. 12% The additivity propenty implies xo) x(a) = output due to u(t), 121 a wi) =O, 12% xu) =0 40), £2 ups deo { Response ro-input response + zero-state response Thas the response of every linear system can be decomposed into the zero-state response and the zero-input response. Furthermore, the two responses can be studied separately and their sum yields the complete response. For nonlinear systems, the complete response can be very different from the sum of the zro-input response and zero-state response. Therefore we cannot sepatate the zero-input and zero-state responses in studying nonlinear systems If a system is linear. then the additivity and homogeneity properties apply to zero-state responses. To be more specific, if x(t) = 0, then the output will be excited exclusively by the input and the state-input-output equation can be simplified as (uy —> yi) If the system is Tineas, then we have {uy + us — yi + ya} and {aw, — ay;} forall @ and all w. A similar remack applies to zero-input responses of any linear system. Input-output description We develop a mathematical equation to describe the zero-state response of linear systems. In this study, the initial state is assumed implicitly tobe zero and the 22 Linear Systems 9 ‘output is excited exclusively by the input. We consider first SISO linear systems. Let S(t —t1) be the pulse shown in Fig. 2.3. Ithas width and height 1/4 and is located at time f,. Then every input u(e) can be approximated by a sequence of pulses as shown in Fig, 2.4, The pulse in Fig. 2.3 has height 1/; thus 5(7 —f,)4 has height 1 and the left-most pulse in Fig. 24 with height w(i;) can be expressed as u(i;)8,(¢ ~ f:)A. Consequently, the input u(r) can be ‘expressed symbolically as wi & ouensstt =a Let g(t be the output atime excited by the pulse u() = ba(¢~ 1) ape at ime Tren we have Bae 4) > nated Salt = t)u(t)d — gale.t)a(4)S homogeni) Tose maid > Creston ad ‘Thus the output y(1) excited by the input u(r) ean be approximated by vin = Destttyunya 2) Figure 23 Pulse at migydatt 018. Figure 2.4 Approximation of input signal10 MATHEMATICAL DESCRIPTIONS OF SYSTEMS Now if approaches zero, the pulse 8, (t ~f,) becomes an impulse at, denotedby 8(¢—1.),and the corresponding output will be denoted by (1). AS A approaches zero, the approximation in (2.2) becomes an equality, the summation becomes an integration, the diserete f, becomes ‘a continuum and can be replaced by r, and A can be written as dr. Thus (2.2) becomes yo =f ae.nuinde 23) [Note that g(t, +) is a function of two variables. The second variable denotes the time at which the impulse input is applied: the frst variable denotes the time at which the output is observed. Because g(t. r) isthe response excited by an impulse, itis called the impulse response. Ifa system is causa, the output will not appear before an input is applied. Thus we have Causal = gff.1) =0 forr fp, is excited exclusively by the input u(0) for r > tp. Thus the lower limit of the integration in (2.3) can be replaced by fo. Ifthe system is causal as well, then g(t, r) = 0 for 1 < 1. Thus the upper limit of the integration in (2.3) can be replaced by ¢. In conclusion, every linear system that is causal and relaxed at fo can be described by [lreemeoes os y@ In this derivation, the condition of lumpedness is not used, Therefore any lumped or distributed linear system has such an input-output description. This description is developed using only the additivity and homogeneity properties; therefore every linear system, be it an electrical system, a mechanical system, a chemical process, or any other system, has such a description. Ifa linear system has p input terminals and q output terminals, then (2.4) can be extended 10 yee -[ Git, nuedr es where Pane gil.) oo silt) ga(t.t) gae(t.t) oo Baplts) D galt ©) gealtst) o> aplt-) and gy (t,t) is the response at time ¢ at the ith output terminal due to an impulse applied at time r at the jth input terminal, the inputs at other terminals being identically zero. That is, _ij(t.t) isthe impulse response between the jth input terminal and the ith output terminal, Thus G is called the impulse response matrix of the system. We stress once again that if a system is described by (2.5), the system is linear, relaxed at fo, and causal ‘State-space deseription Every linear lumped system can be described by a set of equations ‘of the form 23 Linear Time-iavariant (LTH Systems X() = ACL) + BOQ) 26 yt) = Coxe) + Diu) en where X := dx/dt.! For a p-input q-output system, w isa p x 1 vector andy is aq x | vector. Ifthe system has m state variables, then x is an n x 1 veetor. In order for the matrices in (2.6) and (2.7) to be compatible, A, B, C. and D must bem x n.n x p.g xn and g % p matrices, ‘The four matrices are all functions of time or time-varying matrices. Equation (2.6) actually ‘consists of a set of first-order differential equations. Equation (2.7) consists of ¢ algebraic ‘equations. The set of «vo equations will be called an r-dimensional state-space equation oF simply, stare equation. For distributed systems, the dimension is infinity and the two equations in (2.6) and (2.7) are not used, ‘The input-output description in (2.5) was developed from the linearity condition. The ‘development of the state-space equation from the linearity condition, however, is not as simple and will not be attempted. We will simply accept it as a fact. 2.3 Linear Time-Invariant (LTI) Systems xo +7) ude) 2+ to denote tat A by definition. equals 8, We use A = 8 to denote that By definition, equal2 MATHEMATICAL DESCRIPTIONS OF SYSTEMS Input-output description The zero-state response of a linear system can be described by (24). Now ifthe system is time invariant as well, then we ha ee eee ee eet vo [ls -omeer ff gue nae em [ [ where we have replaced (@ by 0, The second equality can casily be verified by changing the variable, The integration in (2.8) is called a convolution integral, Unlike the time-varying case ‘where g isa function of two variables, g is a function of a single variable inthe time-invariant cae. By definition g(r) = g(7 — 0) is the output a time r due to an impulse input applied at time O. The condition fora linear time-invariant system to be causal is g(¢) = Ofort <0. £ EAAMPLE 2.2 The unit-time delay system studied in Example 2.1 is a device whose output 4 equals the input delayed by 1 second. If we apply the impulse 5) at the input terminal, the £ ouput is 8(r — 1), Thus the impulse response of the system is 6(¢ ~ 1). ExaMPLe 2.3 Consider the unity-feedback system shown in Fig. 2.5(a). It consists of a multiplier with gain a and a unit-time delay element. It is a SISO system. Let r(r) be the input of the feedback system. If r(1) = 4(0). then the output is the impulse response of the feedback system and equals ap) = ab = 1) 408-2) +04 - HH = Yabo) 29) Lai r(t) be any input with r(¢) sO for r < 0: then the output is given by v= f genre Ye [se-+- rea I it 70 [Because the unit-time delay system is distributed, so isthe Feedback system. a » Figure 28 Positive and negative Feedhack systems, 2 Note hat g(r) and gl +) af wo diferent functions. However. for convenience. the same symbol gue 2.3 Linear Time-tnvariaet (LT Systems 13 ‘Transfer-function matrix The Laplace transform is an important tool in the study of linear time-invariant (LTI) systems. Let §(s) be the Laplace uansform of y(t), that i, 5 -f vine"ae ‘Throughout this text, we use a variable with a circumflex to denote the Laplace transform of the variable. For causal systems, we have g(t) =O fort < Oor g(t =) =O fort > 1, Thus the upper integration limit in (2.8) can be replaced by oo, Substituting (2.8) and interchanging the order of integrations, we obtain so= (fle -pucrae) omer =f ([lse-nenan)noea Which becomes. after introducing the new variable v =r so= f° ( sto dv) ane ae Again using the causality condition to replace the lower integration limit inside the parentheses from v = —r to v = O. the integration becomes independent of ¢ and the double integrations tome s0= [goerae fo anea eta 563) = B(shi(s) (2.10) wh i) -[ eine "dr is called the transfer function ofthe system. Thus the transfer function is the Laplace trensform ‘of the impulse response and, conversely, the impulse response isthe inverse Laplace transform, of the transfer function, We see tha the Laplace transform transforms the zonvolution integral in(2.8) into the algebraic equation in 2.10). Inanalysis and design, itis sirplertouse algebraic equations than t0 use convolutions, Thus the convolution in (2.8) will rarely be used in the remainder ofthis text For a p-input q-output system, (2.10) can be extended as fu 2G) e+ Bp] Paes) BO | | dO & dap) | | ats) Sas) Byils) B25) Bap(s) J Latgts) His) = Gisiais) ey“4 MATHEMATICAL DESCRIPTIONS OF SYSTEMS saceitin uence ames cl TA A where (8) isthe wansfr function from the jth input to the th output The @ x p matrix G65) is called the transfer function matrix or,stmpy, transfer matrix of the system, EXAMPLE 24 Consider the unittime delay system studieé in Example 2.2, Its impulse resporse is 5(¢ ~ 1). Therefore its transfer function is aw =200-vi= [ B= Ded =e ‘This tansfer function isan irrational function of. EXAMPLE 2.5. Consider the feedback system shown in Fig 2.5a). The transfer function of the unittime delay elements e~. The transfer function from r toy can be computed directly ‘from the block diagram as ! 12 “This cam also be obtained by taking the Laplace transform ofthe impulse response, which was ‘computed in (2.9) a8 sto =Soa'se- Because £{5(1 ~ i)] =e“ the Laplace transform of gy(#) is as) = Ltt = Date” = 00" Dae Using which isthe same as (2.12). a ee eee: i 6D) sme M0) ad oe a cea apajen Head poe © 8G) proper e deg D(s) = deg N(s) + @(00) = zer0 or nonzero constant. 2.3 Linear Time-tnvariart (LTH Systems 15 8s) suietly proper « deg D(s) > deg N(s) © (oe) = 0. «© G(s) biproper < deg D{s) = deg N(s) & §(2e) = nonzero constant. + &(3) improper 4 deg D(s) < deg N(s) # ig(20)| = 00. Improper rational transfer functions will amplify high-frequency noise, which often exists in the real world; therefore improper rational transfer functions rarely arise in practice, A real or complex number 2 is called a pole of the proper transfer function 3(s) = N()/DI) if |8Q)| = 00% a zero if 0) = 0. 1 N(s) and D(s) are coprime, that is, have ‘no common factors of degree I or higher. then all roots of N(s) are the zeros of (5), and all roots of D(s) are the poles of f(s). In terms of poles and zeros, the transfer function can be expressed as ty ox f2DE Gt) ON pe mye Pa) Thisiscaled the zero-polegain form, Jn MATLAB, this formecan be obtained from the transfer funtion by calling (2.p.e1= t£2zp (aun, den) ‘rational mac G(s) is aid tobe proper itis every entry is proper ori (ce) sa zero or nonzero constant mais its sity poperif its every ene is stily proper orf Goo) is 44 zero matrix. Ifa rational matrix Gis) is square and if both G(s) and G~!(s) are proper, then G(s) is said to be biproper, We call 4 a pole of Gis) if itis a pole of some entry of G(s). Thus every pole of every entry of G(s) is a pole of Gla). There ae a number of ways of defining Been ray eel a se betr ee ibe ect eee am a Vara Useful dfintion isthe transmission zero, which wll be inzoduced in Chapter 9 State-space equation Every linear time-invariant lumped system can be described by a set of equations of the form X(1) = Axit) + Bul) 3) y(t) = Cx(r) + Due) For a system with p inputs, q outputs, and n state variables, A, B, C, and D are, respectively, nxn. mx p. qn. and q x p constant matrices. Applying the Laplace transform to (2.13) yields ‘which implies (6) = (61 = AY!XO) + (6 ~ AY" BKS) aay (9) = Col ~ AY*x(O) + C(sT — AY" BALS) + Dis) 2.5) ‘They are algebraic equations. Given x(0) and (s), (5) and (s) can be computed algebraically from (2.14) and (2.15). Their inverse Laplace transforms yield the time responses x(¢) and (2). The equations also reveal the fact that the response ofa linear systemean be decomposed16 MATHEMATICAL DESCRIPTIONS OF SYSTEMS asthe zero-state response and the zero-input response, Ifthe intial state x(0) is zero then (2.15) reduces to 5) = (Cl - Ay'B + DIA) ‘Comparing this with (2.11) yields Gs) = Co1-Ay"B+D e168) “This elates the input-output (or wansfer matrix) and state-space descriptions The lunctons t 268 and Sst in MATLAB compute une destiption from th other. They compute only the SISO and SIMO cases. For example, (nun, den} = 88208 (a,b,c. 2) computes the transfer matrix from the first input tall outputs or. equivalently, the ist column of G(s), If the lst argument | in gs2¢£ (2,,c, 4, 1) is replaced by 3, then the funetion generates the third column of G(s) ‘To conclude this section, we mention that the Laplace transform is aot used in studying linear time-varying systems. The Laplace ransform of g(t, x) is a function of two variables and £IA(x(O] # LIA) |LIx(e): thus the Laplace transform does not offer any advantage andis not used in studying time-varying systems 2.3.1 Op-Amp Circuit implementation Every linear time-invariant (LTD state-space equation can be implemented using an operational amplifier (op-amp) circuit. Figure 2.6 shows two standard op-amp circuit elements, Alt inputs are connected. through resistors, to the inverting terminal. Not shown are the grounded noninverting terminal and power supply. Ifthe feedback branch isa resistor as show in Fig. 2.6), then the output of the element is —(ax1 +bx>-+cxs). the Feedback branch is acapactor with capacitance C and RC = 1 as shown in Fig. 2.6(b), and if the output is assigned asx, then (avi +be:-+cu3). We call the fist element an adder: the second element. an integrator. ‘Actually. the adder functions also as multipliers andthe integrator functions also as multipliers and adder. If we use only one input, say, x3, in Fig. 2.6(a). then the output equals ~axy, and theelement can be used as an inverter with gain a. Now we use an example to show that every LT state-space equation can be implemented using the two types of elements in Fig. 2.6. ‘Consider the state equation al) 03] FO], f-2 | ute ar (iol-[ 3 ]Ee]+ Poe he ok aie at ra me ee ae Z Ti aay toe bets ® Figure 2.6 Two op-amp circuit clement. 24 Linearization 17 2.18) thas dimension 2 and we need two integrators to implement it, We have the freedom in choosing the output of each integrator as +x; or —x,. Suppose we assign the output of the Jeft-hand-side (LHS) integrator as xy and the output of the right-hand-side (RHS) integrator as xy as shown in Fig, 2.7. Them the input of the LHS integrator should be. from the first equation of (2.17), ~iy = =2xy + 0.34» + 2u and is connected as shown. The input of the HS integrator should be 2 =r) ~ 813 and is connected as shown, Ifthe output ofthe adder is chosen as y, then is input should equal —y = 2x, 3x; Su, and is connected as shown, ‘Thus the state equation n(217)and(2 18) cane implemented as shown in Fig. 27. Note hat thor ae many ways implement the same equation For example if we assign te ouput of the two integrators in a is Sierent plementation Tn actal operational amplierciruits, the ange of signals is imited, usually 1 or 2 vols below the supplied voltage If any signal grows outside the range, the circuit wil saturate or bur out and the circuit wil ot behav asthe equation dictates. There is however Wa) to dl with this problem, ase will discuss in Seton 4.31 a8 xy and xp, instead of xy and —xp, then we will obtain & 2.4 Linearization Most physical systems are nonlinear and time varying. Some of them can be described by the nonlinear differential equation of the form 2.7 Op-amp implementation of (2.17) and (2.18)MATHEMATICAL DESCRIPTIONS OF SYSTEMS H(0) = h(x), WLP). 19) yin) =fixlt), 6). ‘where hand f are nonlinear functions. The behavior of such equations can be very complicated and ils study is beyond the scope ofthis text. Some nonlinear equations, however, can be approximated by linear equations under certain conditions. Suppose for some input function w,(t) and some initial state, x4(t) is the solution of (2.19); that is Kol) = WOKE). Hol.) 220) is also [Now suppose the input is perturbed slightly to become u,(1) + i(?) andthe initial sta perturbed only slightly. For some nonlineat equations. the coresponding solution may differ from xs(¢) only slightly. In this case, the solotion can be expressed as a(£) + %(0) with (7) small foe all? Under this assumption, we can expand (2.19) as (x0) FX. UL) + 810.1) ah. ah (0), Wo. H+ IR Sgt 221) H(t) FO) where, for h-= fh ox = Ex x2 sand w= fy ay Pakulasy Bhy/@x2 Bk /Ix3 Aw) : [se lrg (9x2 Alea /9xy Bhs/Ox, Alks/Bxx dhs (9x5 Ee (: Jay wie] BC = Fes | Ohz/Ouy dha/Iu Bhs/du; Bhs/Ous “They are called Jacobians. Because A and B are computed along the two time functions X(t) and u,(0, they are, in general, functions of r. Using (2.20) and neglecting higher powers of & and @, we can reduce (2.21) to Ri) = Ankle) + BENE ‘This isa linear state-space equation. The equation y(s) = f(x(e), u(t), can be similarly linearized. This linearization technique is offen used in practice to obtain linear equations. 2.5 Examples In ths section we use examples to illustrate how to develop transfer functions and state-space ‘equations for physical systems. EXAMPLE 2.6 Consider the mechanical system shown in Fig. 2.8. It consists ofa block with mass m connected to a wall through a spring, We consider the applied force u to be the input 3. Ts is wot ee in general Fr some nonlinear equations, 2 very smal ference in inal sates wil generate ‘orally diferent selusons,yeking the phenomenon of chaos. 25 Examples 19 andaspacement omthe eqiicum ob te up The ction betwen the Nor snd the Dlosk general const of thee dst pus sta on, Coulomb conan vacos fo shoving. 29 Nee tte ol cote elt J yo The ton is cley aot linear fonction of te stoi Tp simply alge, we deen the ste and Coulomb fon an consider ely he vise fica, Ten the fon becomes linear sndcane expresedas (0, her she vinous Faston coat The characters oftiesping ate shown nF 210s etter However the space istintedt (, 3) a8 shown, then te sping canbe costed be ines nd te spe force gual ny. here the spong constant. Thsthe mechani swe can be modeled asa linen system onder neato nd Simpienon Figure 2.8 Mechanical system = Force Force iseou fiton ® fa Figure 2.9 Mechanical system.(a) Static and Coulomb frictions. (b) Viscous fection, Figure 2.10 Characteristic of spring.20. MATHEMATICAL DESCRIPTIONS OF SYSTEMS q ee A i ca a a ARR We apply Newton's law to develop an equation to describe the system. The applied force u must overcome the friction and the spring force. The remainder is used to accclerate the block. Thus we have : = ky key (2.22) y(t) (dt. Applying the Laplace transform and assuming zero where § = dy(0/de? and inital conditions, we obtain (3) — kus 96) ~ ba568) ms?) ‘which implies 1 Tate kis the ‘This he input-output description ofthe dytem, Its transfer functions 1/ima® + As ifm Ly 3. fy 2 then the impulse response ofthe system is af dopaf a eel oral er andthe convolution description of the system is 20 fee -ouinde = [oer Next we develop a state-space equation to describe the system. Let us select the displace- ment and velocity of the block as state variables: that is, xj = J, x2 = J. Then we have, using em, 508) As) u(r) de fem ombsu skin kn ‘They can be expressed in matrix form as aoO]_f 0 1 fam ° Je (i) Lie aim) l20] +L) x x SU 0] 2] ‘This state-space equation describes the system, EXAMPLE 2.7 Consider the system showa in Fig, 2.11. Ttconsists of two blocks, with masses mand ms, connected by three springs with spring constants ki = 1.2. 3. To simplify the discussion, we assume that there is no friction between the blocks and the floor. The applied He» fin Figure 2.11 Spring-mass system. by & b 25 Examples 21 force uy must overcome the spring forces and the remainder is used to accelerate the block, thus we have ue — iy ~ ke(o1 y2) = my madi + Oey + ka)yy — aya = 223) For the second block, we have mais — hay + (ky + ey a 224) ‘They can be combined as my ODP] Phtk ke If] _ fan ° Pale mh wrel lt] [2] ‘his isa standard equation ia studying vibration and sad tobe in normal form. See Reference (08}-Letus define _ 7 0 0 i oT far ° |p hs lye |*]3 o [Le] fa o| be. ot wm] Lf! © 0 07, rJ7loo 1 0. ‘This two-input two-output state equation describes the system in Fig. 2.1L ‘To develop its input-output description, we apply the Laplace transform to (2.23) and (2.24) and assume zero inital conditions to yield amis? Fils) + (kr + ke} fus) — kas) = f(s) From these two equations, we can readily obtain mst hi + where ‘This is the transfer-matrix description of the system. Thus what we will discuss in this text ean bbe applied directly to study vibration2 MATHEMATICAL DESCRIPTIONS OF SYSTEMS EXAMPLE 2.8 Consider a cart with an inverted pendulum hinged on top of it s shown in Fig 2.12 For simplicity. the cart and the pendulum are assumed to move in only one plane, and the fiction, the mass ofthe stick, and the gust of wind are disregarded, The problem is to maintain the rendulum at the vertical position. For example, ifthe inverted pendulum is falling in the direction shown, the cart moves to the right and exerts a force, trough the hinge, to push the pendulum back tothe vertical position. This simple mechanism can be used as a model of a space vehicle on takeoff. Let H and V be, respectively, the horizontal and vertical forces exerted by the carton the ‘pendulum as shown. The application of Newton's law tothe linear movements yields ue ‘The application of Newton's law tothe rotational movement of the pendulum around the hinge yields gl sin = mld 1+ milcos8 ‘They are nonlinear equations. Because the design objective is to maintain the pendulum atthe vertical position, itis reasonable to assume @ and 6 to be small. Under this assumption, ‘we can use the approximation sind = @ and cos @ = I. By retaining only the linear terms in and 4 or, equivalently, dropping the terms with 6%, (6)?, 66, and 96, we obtain V = mg and Mi =u — my — mid goat which imply — mga + 2.25) MIB = (M+ m)go —u (2.26) Figure 2.12 Car with inverted pendulum. 25 Examples 23 Using these linearized equations, we now can develop the input-ouput and state-space descriptions. Applying the Laplace transfon to (225) and (2.26) and assuming ero ial conditions, we obtain Ms? 5(s) = (s) — mg6(s) Mists (M + m)gi(s) — i(s) From these equations, we can readily compute the transfer function gy () from w to y and the transfer function gay(s) from u to 8 as Ayals bol) = Ee ‘To develop a state-space equation, we select state variables a5 x1 = y, 82 = 5. x5 and xq = 5. Then fom ths selection, 2.25), and (2.26) we can readily cbtain [os tw SPD ok] oo 0 | »| o | 0 0 (M+mg/Mi Ol Lxy ajut y=[1 000% @2n ‘This state equation has dimension 4 and describes the system when 8 and 6 are very small ExAMPLe 2.9 A communication satellite of mass m orbiting around the earth is shown in Fig. 2.13. The altitude of the satelite is specified by r(#), 6(¢). and 6(¢) as shown. The orbit «can be controlled by three orthogonal thrusts u(t), us(), and wo(t). The state, input, a 61) eo how =|? 228) cu tte syuemtechoen “ 0 wo] yo [<= | wat #0 ot), “211 42g sind amb + m/mrcsd rn) Then the system can be shown to be described by é Fr) ~Peospsing ~26/r+ueimr x)= w(t) aw) 1B cost tnd? = kit uy /m24 MATHEMATICAL DESCRIPTIONS OF sySTEMS haan Ati i aS A 4 7 Figure 213. Satelite in obit ‘One solution, which corresponds to a circular equatorial orbit, is given by Nol!) = [ro 0 ant a OOF uy =O with rod = k,a known physical constant. Once the satelite reaches the orbit, it will remain in the orbit as long as there are no disturbances. Ifthe satellite deviates from the orbit thrusts, ‘must be applied to push it back to the orbit. Define X= XO) +H) w(t) = Ul +BY YEE +901) If the perturbation is very small, then (2.28) can be linearized as o 1 0 0 0 0 Suz 00 anre 0 0 . © 9 0 3 o of w-| 5 2% 4 6 0 0 fk ° o 0 0.1 o 0 0 o ef 0 9 0 ° 1 9 o 0 0 of tho o [HO ° ° i 25 Examples 25 10 0 0 0 0 Ho={9 0 1 0 0 0 J xe) (2.29) o 0 0 0 1 oo ‘Thissix-dimensional state equation descrites the system. In this equation A, B, and Chappen to be constant. he orbits an elliptic one, then they will be ime varying. We note that the three matrices ar allblock diagonal. Thus th equation can be decomposed into wo uncoupled pais. one involving rand the other. Studying these two part independently Snalsis and design. simplify EXAMPLE 2.10 In chemical plants, itis offen necessary to maintain the levels of liquids. A simplified model of a connection of two tanks is shown in Fig. 2.14, It is assumed that under normal operation, the inflows and outflows of both tanks all equal Q and their liquid levels equal Hy and Hz, Letu be inflow perturbation of the first tank, which will cause variations in liquid level xy and outflow yy as shown. These variations will cause level variation xy and outflow variation yin the second tank. Iris assumed that Rr where R, are the flow resistances and depend on the normal height Hy ard Hs. They can also be controlled by the valves. Changes of liquid levels are governed by yds um yddt and Azdey = (y) ~ yd where A, are the eross sections of the tanks, From these equations, we can readily obtain wo ABR ‘Thus the state-space description of the system is given by [E]= [eet —ayaimeieasno lL] +[’o"]* y= (0 Rx Figure 2.14 Hydraulic tanks26 MATHEMATICAL DESCRIPTIONS OF SYSTEMS Its transfer function can be computed as L AVARR Ras? + (AiR, + AV Ra + AaRa)s +1 &) 2.5.1 RLC networks In RLC networks, capacitors and inductors can store energy and are associated with state variables. Ifa capacitor voltage is assigned as a state variable x, then its current is Cx, where C is ts capacitance. If an inductor current is assigned as a state variable x, then its voltage is Li, where Lis its inductance. Note that resistors are memoryless elements, and their currents or voltages should not be assigned as state variables, For most simple RLC networks, once state variables are assigned, their state equations can be developed by applying Kirchhoff's current and voltage laws, asthe next example illustrates, EXAMPLE 2.11 Consider the network shown in Fig. 2.15, We assign the C,-capacitor voltages, as xj, = 1,2 and the inductor current as x5. [tis important to specify their polarities. Then their currents and voltage are, respectively, Cyi;, C22, and Liy with the polarities shown From the figure, we see that the voltage across the resistor is w ~ xy with the polarity shown, ‘Thusits currents (u — x,)/R. Applying Kirchhoff’s current law at node A yields Coy = x3; at node B it yields = Cid + Coie = Crin tas ‘Thus we have CRG Appling Kirchhoff’s voltage law to the ight-hand-side loop yields Ls = x1 — x2 oF TL ‘The cutput » is given by Figure 2.15 Network 25 Examples 27 ‘They can be combined in matrix form as -U/RC, 0-1/0, 1/RC, x 0 0 cz |x+] 0 fu Lo =k 0 0 yell -1Ox+0-u ‘This three-cimensional state equation describes the network shown in Fg. 215, “The procedure used inthe preceding example can be employed o develop sate equations to describe simple RLC networks. The procedure is fairly simple: assign sate variables and then use branch characteristics and Kirehhof's laws to develop state equations. The procedure can be stated more systematically by using graph concepts, as we will introduce next. The procedure and subsequent Example 212, however, can be skipped without loss of continuity First we introduce briefly the concepts of tee. link, and cuset ofa network. We consider only connected networks. Every capacitor, inductor, esistor, voltage source, andurrent source will be considered as a branch. Branches are connected at-nodes. Thus a network can be considered to consist of only branches and nodes. Toop is a connection of branches starting from one point and coming back to the same point without passing any point twice. The algebraic sum ofall voltages along every lop i zero (Kirchhoff voltage law). The set of all branches connect to anode is called a cuset. More generally, acutst ofa connected network is any minimal set of branches so that the removal ofthe set causes the remaining network tobe unconnected, For example, removing all branches connected to a xode leaves the node unconnected to the remaining network. The algebraic sum of all branch currents in every cutser is zero (Kirchhoff’s current la) Aree ofa network s defined as any connection of branches connecting al the nodes but containing no loops. A branch is called a ree branch if itis inthe re, alink if iis not. With respect to a chosen tee, every link has a unigue loop, called the fundamental loop, in which the remaining loop branches are all tree branches. Every tre branch has aunigue cust, called the fundamental cuser. in Which the remaining cutset branches are all links. [nother words, 2 fundamental loop contains only one link anda fundamental cuset containsonly onetree branch Procedure for developing state-space equotions* 1. Consider an RLC network. We first choose a normal éree. The branches of the normal tuee are chosen in the order of voltage sources, capacitors resistors, inductors, and current 2. Assign the capacitor voltages in the normal tree and the inductor currents im the links as state variables. Capacitor voltages in the links and inductor currents in the normal tree are not assigned 3. Express the voltage and current of every branch in terms of the state variables and, if necessary, the inputs by applying Kirchhoff’s voltage law to fundamental loops and Kirchhoff's current law to fundamental cusets 4 The eader may Sp his procedue and go aitety to Example 2.1328 MATHEMATICAL DESCRIPTIONS OF SYSTEMS 4. Apply Kirchhoff’s voltage or current law to the fondamental loop or cutset of every branch that i assigned as a state variable ExamPLe 2.12 Consider the network shown in Fig. 2.16. The normal tre is chosen as shown ‘with heavy lines; it consists of the voltage source, two capacitors, and the 1-G resistor. The capacitor voltages in the normal tree and the inductor current in the link will he assigned as stace Variables. Ifthe voltage across the 3-F capacitor is assigned as.x), then its current is 3% ‘The voltage across the I-F capacitor is assigned as xy and its current is. The current through the 2-H induetor is assigned as xy and its voltage is 25. Because the 2-9 resistor isa link, we use its fundamental loop to find its voltage as u; — xy. Thus its current is (uj — #;)/2, The 1-2 resistor is a tree branch, We use its fundamental cutst to find its current as x3, Thus is voltage is 1-5 = xy. This completes Step 3, ‘The 3-F capacitor is a wee branch and its fundamental cutset is as shown. The algebraic sum of the cutset currents is 0 or 3k tu -n=0 which implies The |-F capacitor is a tree branch, and from its fundamental cutset we have iy — x5 = {_ Tee intucoriatn The vag ln is i Fundamental 1 Peusetot @ 1 Fundsmentat 1 casero © “® H A =~ |= Fandomensl ‘ser at Figure 2.16 Nexwork wien wo inputs 2.5 Examples 29 ‘They can be expressed in matrix form as [Fs al) If we consider the voltage across the 2-H inductor and the current through the 2-2 resistor as the outputs, then we have ieee ae 0 0 com Atel Je am Equations (2.30) and (2.31) are the state-space description of the network. ‘The transfer matrix of the network can be computed directly from the network or using the formula in (2.16): (s) = Cel = ay"'B+D We will use MATLAB to compute this equation. We type 0,000 0.1667 -0.0000 -0.0000 0.5000 0.2500 0.3333 -0.0000 1.0000 0.6667 0.7500 0.0833 ‘This is the frst columa of the transfer matrix. We repeat the computation for the second input ‘Thus the transfer matcix of the network is. 0.16675? 033338 DF 0666TS +0735 +0083 540.6667? + 0.755 + 0.0833 OSs? +0.25s? + 0.33335 0.16675? ~ 0.08335 ~ 0.0833, DF 0.66675 + Go)= FOT5s+0.083 + 0,66675° + 0.738 + 0.083330 MATHEMATICAL DESCRIPTIONS OF SYSTEMS EXAMPLE 2.13 Consider the network shown in Fig. 2.17(a), where T is a tunnel diode with the characteristics shown in Fig. 2.17(6). Let xy be the voltage across the capacitor and x» be the curtent through the inductor. Then we have v = x) and 22(1) = Ciu() +10) = Cin) + hGH) Lig(t) = E - Rut) — 1) ‘They can be arranged as hin), 2 c c u(t) Rn) | E pe Z ‘This st of nonlinear equations describes the network. Now if x(t) is known to lie only inside the range (a, b) shown in Fig. 2.17(b), then A(xi(t)) can be approximated by h(xi(?) -an(0)/Ri. In this case, the network can be reduced to the one in Fig. 2.17(€) and can be (er HIE Lie a 3) alt al | Aur tL. Ve © te Figure2.17 Network with a tunnel diode 26 DiscreteTime Systems 31 ‘This is an LTI state-space equation. Now if x(t) is known to lie only inside the range (c, d) shown in Fig 2.17(b), we may introduce the variables (0) = x(f)~v, and f2(2) = £210) lo and approximate f(xy (0)) as ig — 3,(0)/ Ro. Substituting these into (2.32) yields (e)-[2 EDL where E = £ ~ vp ~ Rip: This equation is obtained by shifting the operating point From (0.0) to (vss in) and by linearization at (vi). Because the two Hinearzed equations are identical if ~Rs is replaced by Ry and E by E, we can readily obtain its equivalent network shown in Fig. 2.17(d, Note that tis not ebvious how to obtain the equivalent network from the original network without fist developing the state equation. 2.6 Discrete-Time Systems ‘This section develops the discrete counterpart of continuous-time systems. Because most concepts in continuous-time systems can be applied directly to the discrete-time systems, the discussion will be brief. The input and output of every discrete-time system will be assumed to have the same sampling period T and will be denoted by ufk] := w(K), y{k]i= y(kT), where kis an integer ranging from —oc to +20, A discrete-time system is causal if current output depends on current ‘and past inputs, The state at time kp, denoted by x{ko], is the information at time instant ko, which together with u(k], & > to, determines uniquely the output y[k), k 2 ko, The entries of x are called state variables. IF the number of state variables is finite, the discrete-time system is lumped; otherwise. itis distributed, Every continuous-time system involving time delay, as the ones in Examples 2.1 and 2.3, isa distributed system. In a discrete-time system, ifthe time delay isan integer multiple of the sampling period T-, then the discrete-time systera is 2 lumped system, A discrete-time system is linear ifthe additivity and homogeneity properties hold. The response of every linear discrete-time system can be decomposed as Response = zero-state response + zero-input response and the zero-state responses Satisfy the superposition property. So do the zero-input responses. Input-output description Let 5{é] be the impulse sequence defined as {; ifk=m Mm = 19 itkam where both and m are integers. denoting Sampling instants tis the discrete countegpat of the impulse 5(1~ f). The impulse (0 ~ 1) has 2er0 width and infinite beight and cannot be generated in practice: whereas the impulse sequence {A ~ m] can easy be generated. Let ‘kl be any input sequence. Then it can be expressed as lk] = D> ulate — me} Let s{h. ml be the output at time instant & excited by the impulse sequence applied atime jstantm. Then we2 MATHEMATICAL DESCRIPTIONS OF SYSTEMS fk — m] — glk. m) 60k)» gtk. mun) omogeny) Yost — miata) > Seth man} (adv) ‘Thusthe output y{A] excited by the input u(k] equals yl = SO glk mutt OM ‘This the discrete counterpart of (2.3) and its derivation is considerably simpler. The sequence {km is called the impulse response sequence. Wa diseretetime system is causal, no output will appear before an input is applied. Thus wetave Causal = gfk,m] = 0, fork 0, then the output y(¢) depends on future input w(t + A) and the differentiator is not cats Hower if we define the iferentiaon a8, ult) — jy MO) = ult =A) (ler ee gue ae then the output y(t) does not depend on future input and the differentiator is causal. Therefore in continuous-time systems, it is open to argument whether an improper transfer function represents a noncausal system, However, improper transfer functions of s will amplify his 26 DiscreteTime Systems 35 frequency noise, which often exists in the real world. Therefore improper transfer functions are avoided in practice Statespace equations Every linear lumped discrete-time system can be described by x(k +1) = Aten] + BqUtA] lel = ClkIxik] + DEEIUTA] (2.38) where A,B,C, and D are functions of k. Ifthe system is time invariant as well, then (2.38) becomes x{k-+ 1] = Axlk) + Bute] 239) ylk] = Cxlk) + Duta) where A.B, C, nd D are constant matrices, Let &(c) be the z-transform oF x(k] oF R(S) = Zixtk I} = YO wth ‘Then we have Zink + = Soxlk + Me =e Poxtk + oe mS m [Sxaet+x-x0] = -nop which implies = Ay tex{0] + (I= Ay 2.40) Ay exf0] + C(t = A)" Bats) + DAE) esp ‘They are the discrete counterparts of (2.14) and (2.15), Note that there is an extra < in front of {0}. If x{0] = 0, then (2.41) reduces to (o) = [Cel ~ Ay'B + DIGS) 24) Comparing this with the MIMO case of (2.37) yields Gq) =CGl-ay'B+D (2.43) This isthe discrete counterpart of (2.16). Ifthe Laplace transform variable sis replaced by the :-transform variable =, then the two equations are identical36 MATHEMATICAL DESCRIPTIONS OF SYSTEMS Exampte 2.16 Consider a money market account in a brokerage firm. IF the interest rate ‘depends on the amount of money in the account, its a nonlinear system. If the interest rate is the Same no matter how much money isin the account, then itis a linear system. The account isa time-varying system if the interest rate changes wit time: a time-invariant system if the interest rate i fixed. We consider here only the LTI case with interest rate r = 0.015% per day. ‘and compounded daily. The input u{A] is the amount of money deposited into the account on. the Ath day and the ouput sk] isthe total amount of money in the account at the end of the ‘th day. If we withdraw money, then i(k] is negative. I we deposit one dollar on the first day (that is, w[0} — 1) and nothing thereafter 0. k= 1,2. ...). then yf0] = w{0] = I and yt] = 1+ 0.00015 = 1.00015, ‘Because the money is compounded daily, we have 9121 = sf] + yft] 0.0001 and. in general, EI] 1.00015 = (1.00015)? ytk] = (1.00015)! Because the input. 0. 0, response sequence or Jis an impulse sequence, the outputs. by definition, the impulse 41 = 1.00015)" andthe input-output description of the account is ' F 18) = D2 gtk mind = S.00015)"" eas ie cs ‘The discrete transfer function is the z-transform of the impulse response sequence or Same Zietklh = Soe.00015)': aS 2.5 00015=" ie Whenever we we 244) oF (2.45) the initial stake must be 2ro, of thee is niall no money in th account Next we develop a state-space equation to describe the account. Suppose sk isthe tot amnnt of money athe end ofthe kth day. Then we have MU + 1) = yfk] + 0.00015 914] + alk + 1) = LOOOIS yf] + atk +1) (2.46) If we define the state variable as xfR] := yf], then sk + 1] = 1,0001Sx(k) + ulk + 11 ean wk) = x(k) Becwuse of afk + 1]. (2.47) is not in the standae fort of (2.39). Thus we cannot select -IK1:= ylk] asa state variable. Next we select a different stae variable as 3 27 Concluding Remarks 37 x(k] 2= yfk] — alk) Substituting y[kyt 1] = afk + 1] +u{k + 1] and yf) = x[A] + (A) into (2.46) yietds xf +1] = 1.00015x{E] + 1.00015u(A) lk] = lk) + ule) ‘This is inthe standard form and describes the money market account, (2.48) ‘The linearization discussed forthe continuous-time case can alsobe agplied tothe discrete- time case with only slight modification. Therefore its discussion will notbe repeated 2.7 Concluding Remarks We introduced in this chapter the concepts of causality. tumpedness. linearity, and time invari- ‘ance. Mathematical equations were then developed to describe causal systems, as sursmarized jn the following. Internal description External description System type Ovi i or [ Gremtaee amped, tineae ‘oxeBinn y= [Gie-rucride Coos Drow Distributed, linear 29 = [Ge cmieyde time-invariant a 31s) = Gis}0(5), G1 iational Lumped, linear, x= AN+Ba se» [Gu oucyar y=Cx+Du 6) = Geordie). Gis ational Distributed systems cannot be described by finite-dimensional state-space equations. External description describes only zer-state responses thus whenever we use the description. systems are implicitly assumed tobe relaxed or tei intial condition are assumed to be 20, ‘We study in this text mainly lumped linear time-invariant systems, For this clas of systems, we use mostly the time-domain description (A. B.C.) in the internal description and the frequency-domain (Laplace-domain) desription G(s} in the external desrition. Furthermore. we will express every rational transfer matrix a a fraction of two polynomial matrices, as we will develop in the text. By so doing, all designs inthe SISO case can be extended tothe multivariable case.38 ft MATHEMATICAL DESCRIPTIONS OF SYSTEMS. ‘The class of lumped linear time-invariant systems constitutes only a very small part of nonlinear and linear systems. For this small class of systems, we are able to give a complete treatment of analyses and syntheses. This study will form a foundation for studying more general systems 21 Consider the memoryless systems with characteristics shown in Fig. 2.19, in which denotes the input and y the output. Which of them isa linear system? Is it possible to introduce a new output so that the system in Fig. 2.19(b) is linear? ‘ < " z « < " @ » © Figure 2.19 2.2. The impulse response of an ideal lowpass filters given by apsit alt — 10) ~2aatt =i) for all, where « and fy are constants. I the ideal lowpas filter causal Is it possible to build the filter inthe real world? B= 23° Consider a system whose input u and output y are related by ui) fors 0. In each case, the initial state (0) at time 1 = O's the same. The corresponding outputs are denoted by 9 Which of the following statements are correct if x(0) # 0° Problems 39 1. Thus = uy tus, then ys = yy + yx 2. Ifus = O.5(0 +12), then yy = 0.5(94 + 92) 3. Ifug = uy ~ us, then ys = yy — 93 Which are correct if x(0) = 0? 2.6 Consider a system whose input and output are related by w/a 1) ifulr— 40 wo = for all r. Show that the system satisfies the homogeneity property aut not the additivity Property. 2.7. Show that if the additivity property holds, then the homogeneity property holds for all rational numbers a. Thus if a system has some “continuity” property, then additivity implies homogeneity 28. Letg(s,r) = g(t-+a, +a) forall, r,and.a. Show that g(t, r) depends only on ¢— (Hint: Define x 21-4 and y id show that dg(?. r)/x = 0.] 2.9 Consider a system with impulse response as shown in Fig. 2.20(a), What isthe zero-state response excited by the input u(¢) shown in Fig. 2.20(b). Figure 220 2.10 Consider a system described by +2) -3y ‘What are the transfer function and the impulse response of the system? 211 Let j(¢) bethe unit-step response of alinear time-invariant sy response of the system equals d(t)/at. ‘Show that the impulse 2.12. Consider a two-input and two-output system described by Dulp)y1( + Dial p)y216) = Nu(p)ua(@) + Mie(P)ua()40 MATHEMATICAL DESCRIPTIONS OF SYSTEMS Das(piyi{t) + Daxtp)y200 pote) + Next pegte) where Nj, and D,, are polynomials of p = d/dt. What is the transfer matrix of the system? 2.13 Consider the feedback systems shown in Fig, 2.5. Show thatthe unit-step responses of the positive-feedback system are as shown in Fig. 2.21(a) for a = 1 and in Fig. 2.21(b) for a = 0.5, Show also that the unitstep responses of the negative-feedback system are as shown in Figs. 2.21(c) and 2.21(d), respectively, fora = 1 and a = 0:5. we se ors b sed omiaws sot on “A eo ° f_., ; perea sere o1?34567 Figure 221 2.14 Draw an op-amp circuit diagram for «[D s}[2)« BB 1o}x = 20 21S Find state equations to describe the pendulum systems in Fig. 2.22. The systenns are useful to model one- of two-link robotic manipulators. If. #, and & are very small, ‘can you consider the two systems as linear? 2.16 Consider the simplified model of an sircratt shown in Fig. 2.23. It is assumed that the aircraft is in an equilibrium state atthe pitched angle 8, elevator angle wo, altitude fo, and cruising speed vp. It is assumed that small deviations of @ and w from and up generate forces fi = ky and fy = kau as shown in the figure. Let m be the mass of the aireraf, [the moment of inertia about the center of gravity P. bf the aerodynamic damping, and h the deviation of the altitude from ho. Find a state equation to describe the system. Show also that the transfer function from w to h. by neglecting the effect of Lis Problems al o ” Figure 222 Figure 223, hus) Ba) ~ skids + au) 2.17 The soft landing phase of a lunar module descending on the moon can be modeled as shown in Fig. 2.24. The thrust generated is assumed to be proportional tov. where m is the mass of the module. Then the system can be described by mt = ~Arit — mg. where {¢ is the gravity constant on the lunar surface. Define state variables of the system as ty = yoy = Sats = mand u = 1h. Find a state-space equation to describe the system, 2.18 Find the transfer functions from w 10 3 and from yy 10 y of the hydraulic tank system shown in Fig. 2.25. Does the transfer function from u to y equal che product of the two transfer functions? Is this also trve for the system shown in Fig. 2.142 (Answer: No. because of the loading problem in the wo tanks in Fig. 2.14. The loading problem is an Important issue in developing mathematical equations to describe composite systems. See Reference (7}]a MATHEMATICAL DESCRIPTIONS OF SYSTEMS Fons 2.19 Find a state equation to describe the fer function. Figure 2.24 Figure 2.25 network shown in Fig. 2.26. Find also its trans- Figure 2.26 2.20 Find a state equation to describe the network shown in Fig. 2.2. Compute also ts transfer 221 Consider the mechanical system shown in Fig. 2 of the bar and block about the hinge. 1. Let f denote the moment of inertia I is assumed that the angular displacement & is very small, An extemal force u is applied to the bar as shown. Let y be the displacement | | Problems 43 of the block, with mass mz, from equilibrium. Find a state-space equation to deseribe the system, Find also the transfer function from to y Figure 227Chapter Linear Algebra 3.1 Introduction ‘This chapter reviews a number of concepts and results in linear algebra that are essential in the study ofthis text. The topics are carefully selected, and only those that willbe used subsequently are introduced. Most results are developed intuitively in order for the reader to better grasp the ideas. They are stated as theorems for easy reference in later chapters. However. no formal proofs are given, ‘As we saw in the preceding chapter, all parameters that arise in the real world are real ‘numbers. Therefore we deal only with zeal numbers, unless stated otherwise, the his text Let A, B, C. and D be, respectively. n x m.m xr, x n.andr x p real matrices. Leta bethe ith column of A, and by the jth row of B. Then we have by b; ABS [a a; -- aul eb, abs + BD by CA = Clay az ++ ag) = (Ca, Car Can] 2 and by Bay b; bp BD= 63) De bat 3.2 Basis, Representation, and Orthonormalization 48, ‘These identities can easily be verified. Note that ab, is ann x r matrix; itis the product of an nx 1 column vector and a 1 x r sow vector. The product b,a, is not defined unless = rit becomes a sealar ifn 3.2 Basis, Representation, and Orthonormalization Consider an n-dimensional real linea space, denied by , Every vector in 2 isan n-tuple ‘of real numbers such as, iE ‘To save space, we write itas x= [sy 1) --+ x]', where thesprime dens the transpose The set of vectors [x Xp) in R” is said to be linearly dependent if there exist real numbers a. a3, -... ae not al zero, such that 81 $00Ny ghy =O G4) If the only set of a for which (3.4) holds is a, = 0. vectors [X1, X Xu) is Said tobe linearfy independent If the set of vectors in (3.4) is linearly dependent, then there exists at least one a. Sty. ey. that is different from zero. Then (3.4) implies 0. ssss ety = Os then the set oF where f; = —a, /e. Such an expression is called a linear combination. The dimension of a linear space can be defined as the maximum number of linearly independent vectors in the space, Thus in R", we can find at mast n linearly independent vectors Basis and representation A set of linearly independent vectors in ‘is called a basis if every vector in " can be expressed as a unique linear combination ofthe set, In R". any set of n linearly independent vectors can be used as a basis, Let (q1. qo. +... dh) be such a set Then every vector x can be expressed uniquely as aq +e. as Define the m x m square matrix Os la @ = al Co ‘Then (3.5) can be writen as4 LINEAR ALGEBRA x=Q OY a: a (3.7) We call = fa; a «cal the representation of he vetn x with eect the bai (ai. q2. -. Ge). Fie wi oct with every Rh flowing orthonormal bass 1 0 0 o o °| 0 0 ua? (eee | es 0 1 | 0 1 Wit especto ths basis we ave eS * = nh tonb + * where Iy is the m x unit matrix. In other words, the representation of any vector x with respect to the orthonormal basis in (3.8) equals itself. EXAMPLE 3.1 Consider the vector x = {1 3]! in R? as shown in Fig. 3.1, The two vectors 41 = (3 AY’ and qz = [2 2 are clearly linearly independent and can be used as a basis. If we ‘draw from x two Tines in parallel with q> and qh, they intersect at —q, and 2q> as showa. Thus the epresentation of x with respect to {qy, Qn} is [I 2. This can also be verified from ef) e[2)-f IE] in parallel with iy and q>. They intersect at 0.5qp and 2is. Thus the representation of x with i) is 10.5 27. (Verify) respect 10 [4 Norms of vectors The concept of norm is a generalization of length or magnitude. Any real-valued function of x, denoted by [jx can be defined as a norm if it has the following properties: 1, |[xll = 0 for every x and |x| = 0 if and only if x 2, \lax| = jax), for any real a. 3. fxs +l flaalf + xa for every x} and x | 3.2 Basis, Representation, and Oxthonormalization 47 Figure 3.1 Ditlerent epresentations of ‘The last inequality is called the riangular inequality. Letx = [x1 2 --> aq]. Then the norm of x can be chosen as any ane of the following: Ili = They are called, respectively, [-norm, 2- or Euclidean norm, and infinite-norm. The 2-norm is the length of the vector from the origin, We use exclusively. unless stated otherwise, the Euclidean norm and the subscript 2 will be dropped. In MATLAB. the norms just introduced can be obtained by using the fenctions norm(x, 2), norm(x, 2) =norm(x).and norm (x, Orthonormalization A vector xis said be normalized ifits Euclidean norma is | orx'x = L Note that xx is sealar and xx’ ism x n. Two vectors x; and x: are said to be orthogonal if jay = Nix = 0. A set of vectors x), 1 = 1,2... is said to be orthonormal if wy af? HAI MTL ites j Givena setof linearly independent vectors e;, set using the procedure that follows: én, We can obtain an orthonormalLINEAR ALGEBRA, use qu := ui/|luiIl uss er (gies) uy/lius! Woe = Om — ET (Gem) dae = Ue | ‘The firstequation normalizes the vectore; tohave norm 1. The vector (q/€2)q1 isthe projection ofthe vectore; along q. Its subtraction from e: yields the vertical part ut is then normalized tol as shown in Fig. 32. Using this procedure, we can obtain an othonormal set. This is called the Schmidt orthonormalization procedure. Let A= [ay a) ++ aq) be ann x m matrix with m 0. then for every eal ai, = 1,2, ..., ky the vector 3.3 Linear Algebraic Equations — 81 X= Xp tem +o tou Gs) is asoluion of AX = y. where (my, ..., ma) is basis ofthe mall space of A. Substituting (3.15) into Ax = y yields Ax + Daan Xp bO=y ‘Thus, for every ay, (3.15) isa solution. Let & be a solution or AX = y. Subtracting this from Ax, = y yields AG x) which implies that X — x, is in the null space, Thus can be expressed as in (3.15). This establishes Theorem 3.2 ExAMPLe 33 Consider the equation ony 4 i234 twa sane[]-> 616 A 2020] ° ‘This y clearly lies in the range space of A and x, = {0 —4 0 O}'is solution. A basis ofthe null space of A was shown in (3.12). Thus the general solution of (3.16) can be expressed as, 0 ri 1 Xp term tam =| | +a, am for any real a, and a. In application, we will also encounter xA = y, where the m xn marrix A and the 1x n vector y are given, and the 1 x m vector x is to be Solved. Applying Theorems 3.1 and 3.2 0 the transpose ofthe equation, we can readily obtain the Following result. Corollary 32 1. Given an m xm matrix A. a solution x exists in XA = y, forany y, if nd only if A has full olumn 2, Givenanm xm matix A andan Ln vector, letX, bea solution ofXA = y and let ~ pia). 1k = O. the solution Xp is unique. IFk > 0, then for any a, 7 = 1, 2... the vector X= Xp taym to baa {2 solution of XA = y. where nA = O and the set (mp, ..., me linearly independent In MATLAB, the solution of Ax = y can be obtained by typing A\y. Note the use of backslash, which denotes matrix left division. For example, forthe equation in (3.16). typing52 LINEAR ALGEBRA, yields (0-4 0 01. The solution of xA = y can be obtained by typing y/A. He slash, which denotes matrix right division, Determinant and inverse of square matrices The rank of a matrix is defined as the number of linearly independent columns of rows. t can also be defined using the determinant, The ‘determinant of 4 1 > | matrix is defined ae itself. For x — 2.3, .-. the determisvant of ¥ square matrix A = (a,)] i defined recursively as, for any chosen j det A (8) where a,j denotes the entry at the ith row and jth column of A. Equation (3.18) is called the Laplace expansion. The number ¢,; is the cofactor corresponding to cy, and equals (17 det Mj. where M;; is the ( — 1) x (n ~ 1) submatrix of A by deleting its ith tow and jth column. If A is diagonal or triangular, then det A equals the product ofall diagonal ‘The determinant of any r x r submatrix of A is called a minor of order r. Then the rank ‘canbe defined a te largest order of all nonzero minors of A. Tn other words, if A has rank r, then there is at least one nonzero minor of order r. and every minor of order lager than 7 i zero. A square matrix is said to be nonsingular if ts determinantis nonzero, Thus a nonsingular square matrix has full rank and all its columns (rows) ar linearly independent ‘The inverse ofa nonsingular square matrix A = [a] is denoted by A~'. The inverse has property AA“! = A“!A = T and can be computed as, AgA 1 Geta ~ dra “1 (3.19) where ci isthe cofactor, Ifa matrix is singular, its inverse does not exist, IFA is 2 x 2. then ~ (ete Nera ote Thus the inverse of a2 x 2 matrix is very simple: interchanging diagonal entries und changing the sign of off-diagonal entries (without changing postion) and dividing the resulting matsix bs the determinant ofA. In general using (3.19) 10 compute the inverse ts complicated. If Ais srangular, itis simpler to compute its inverse by solving AA“! = I, Neve tha the inverse of a vrangularmattx is again triangular. The MATLAB function “=> computes the inverse of A. ] 320) Theorem 3.3 Consider AX = y with A square 1. IFA is ponsingula, then the equation has 8 unique solation for every ¥ and in solution In panicular. the only solution of AX = is x = 3.4 Similarity Kansformation $3 2. The homogeneous equation AX = 0 has aonzero solutions if and only if A i singular. The number of linearly independent solutions equals the nlly of A. 3.4 Similarity Transformation Consider an n x m matrix A. IC maps R" into self. If we associate with R" the orthonortnl basis iu. ip... i in (38), then the ith column of A is the representation of Ai, with re spect to the orthonormal basis. Now if we select a different set of basis (4. 4 du} then the matrix A has a different representation A. tums out that the fuk colune of is the representation 2F Ag, with respect othe basis (41. 2. =. jy}. This is illustrated by the example that follows, EXAMPLE 3.4 Consider the matsix been my [ a a Letb = (0 0 1)’. Then we have ? al ° a [ a]. esa [ ‘| weary -| | i i ; a {It can be verified that the following relation holds: A®b = 17b— 1SAb+547b G2 Because the three vectors b, Ab, and A*b are linearly independent, they ean be used asa basis. We now compute the representation of A with espect to the basis, Its clear that 0 : A(b) = (b Ab Ab] Hl ; 0 7 ACAD) ; a i an eso] : i 77 b ab wl 3 ). Thus the representation of A with respect to 00 179 Az]i 0 -15 623) lor 5 ACA) where the last equation is obtained from (3, the basis {b. Ab, Ab) is