Download as pdf
Download as pdf
You are on page 1of 86
CONTROL SYSTEM FUNDAMENTALS edited by William S. Levine Library of Congress Cataloging-in-Publication Data vine, WS. Cone yescen fanarmereals Wil 8. Levi pcm. Includes bibliographical references, ISHN 0-1499-0053-93 cal,papert 1, Autti cated, 2 Coated try. 1 Tate TUS LAI9 1995 629d e-oceans ap ‘This book contin information obtained (rors authentic and highly reguided sources. Reprinled material is quoted with permission, and sources ae indicared. A wide variety of references ae listed, Reasonable efforts have been siode to publish reliable data and information, but the autbar and the publisher cannot assure responaitilty forthe validity ofall rterals or for the consequences of their use. [Neither this kook nor any part may be reproduced or trantmiued in any fuer or by any rans, electronic or ‘mechanical, incloding photocopying. microfilming, and recording. or by any information storage ox retrial system, without prior permission in writing from tee publishes. ‘The eansent of CRC Press LLC does not extend to copying fr general diseibusion, for promotion, for creating ew works, of for resale. Specific permission must be cbuained in waiting from CRC Press LLC far such copying. Direc all Inquiries to CRC Press LLC, 2000 NW, Corporate Blvd., Boca Ratoe, Florida 33431. ‘Trademark Notler: Product oc corporate nares may be tradermsks oc registered trademarks, aw! are used only for identification snd explanation, without intent ie infringe ‘The material in this book eas wen from, The Control Handbook Bed ty Wiliam %, Levine. ‘© 2000 by CRC Press LLC No claim to exiginal U8. Government works International Standard Back Numer 04491-0053-3 Library of Congress Card Number 98-043056 Printed in the United Sues-of America =-23436 7890 Printed on acid-free paper Contents SECTION I Mathematical Foundations 1L_Mathematical Foundations BP doth’ | 1L1_ Differential Equations _ 3 12 Difference Equations _ 2_‘The Fourier, Laplace, and 2-Teansforms Edward W. Kamen wz 21_Intzoduction r 2:2 Pundamentals ofthe Fourier, Laplace, and 2 Transforms 2 23 Applications and Examples 25 3 wat Algebra Bradley W. Dickinson... 33 Introduction st a 33 Vector Spaces 2 8 34 Linear Equations wee 35. Eigenvalues and Blgenveciors 5 8 ‘DM_The Jondan Form and Simdariyof Maices See a5 2.7__Singular Value Decomposition wena tb A Matrcesand Mulivarishle Busssions sa Complex Variables C.WiGray pe AL_Comples Numbes ese : = al 42 Comples Functions 2 43 Comples Integrals se coe woeeeeeeseeeseess $4 Conforanal Mappings ee pee SECTION II_ Models for Dynamical Systems 5._Standard Mathematical Models 00 gg 5.1 Inpot-Dutpot Models William S Levine 65 52 State Space James T Gillis a n 6_Graphical Models... pee “6.1_Block Diagrams Dean X, Frederick and Charles M4. Glest oe SS 62 Sigil Flow Graphs Norman 5. Nie wep epee Z_Determining Models... . _ we . 3B Za Medeling fom Physical Principle Francois E. Cellier, Milding Elmavist and Mastin Otter... o 7.2 System identification When Noise Is Negligible Wiliam $. Levine 18 SECTION III _Analysis and Design Methods for Continous-Time Systems 8__ Analysis Methods oe a us Time Response of Linear Timm invarlast Sites — Raymond T. Sign 5 8.2__Controllability and Observability William A. Welovich _ 12 9 Stability Tests. _ _ 2st 3.1 The Rovth:Harwiy Subilty Cerion Raber Bbup and Richard © Dory - BE 9.2 The Nyquist Stability Test Charles E Rokrs 15 9.3 Discrete-Time and Sampled:Deta Stability Tests Mohamed Mansour... ~~. ee 94_Gein Margin and Phase Mangin Raymond Z Stefani _ _ 152 10 Design Methods . 157 102, Design Using Performance Indice Bichand G Dorfand Robert H. Bho. 10.3 Nvguist, Bode, and Nichols Plots Joi. D’Aazo and Constantine H. Heupis i 104 The Root Locus Plot Wiliam 5 Ervine - 19 10.5 PID Control _ Kas! fAstrém and Tore Higa - _— ie 10.5 State Spore Pole Placement Kanihiko Ogeta _— 08 107 Internal Model Contiel fichand DB Brust 23 108 Time-Delay Compensation —Smith Predicir and its Modifications Zj. Palmer... ...-......- 224 109 Modified Predictors Riclurd DB. Simats . : a . Bi 10.10Time-Delay Compensation for Unstable Plants 2. Palmor . a : BS SECTION IV Digital Control L1_Discrete-Time Systems Mohammed S. Santina, Allen R. Stubberud and Gene H. Hostetter . 2a LL Discrete-Time Stems - oe 241 12 Sampled Data Systems A. Feuer and G.C. Gaodwin a 255 12.1 lntenduction and Mathematical Preliminaries 0 12. “Sensitivity Functions in Sampled Data Cootrol Systems 256 123 Sensitivity Consideration a i 258 124 Bxamples a 259 125 Linear Quadeatic Design of Sample® Data Controllers. = cone ee 282 13. Discrete.’ ime Equivalents to Continuous-Time Systems Mohammed $, Santina, Allen R, Stubberud and aie Viral adele 14_Design Methods for Discrete-Time, Linear Time-Invariant Systems Mohantmted S. Santina, Allen R. Stubs ‘Genudand Gone Hi Hostetter 2 141 Ano 3 142 Glasical Control Stem Design Methods se a 284 14.3 Eigenvalue Placement with State Feedback oe 289 (A.A Step- Invariant Diserete-Time Observer Dasige wees e eee eee L45 ‘Tracking Systems Design wees eee DH 146 Designing letween-Sumple Kesponse vee os oe = 2% 15 Quantization Effects _Mahanumed S. Santina, Allen R. Stubberud and Gene H. Hostetter... 303 151 Overview aa 15.2 Fixed-Point and Floating: Point Number Representations 303 15.3 Truncation and Rondafl Err 15.4 Elects of Coefficient Quantic - 206 155 Quaetlation Effects in AMD Conveision ss. esses seveeseeeeeesessereesres > 348 13 Stags Anan of Guetion Boom in Died Poe —_ 9 18a Limit Cyele and Deadband Effects voce eee ee esse esses esses SB 16 Sample-Rate Selection Mohammed 5. Santina, Allen R. Stubberud and Gene H. Hostetter 315 16.1 lntroduction ce cove 315 162 The Sampling Thearem bitin 35 16.3 ‘Control System Response and the Sampling Period 3 16A_Contrl System Response wo Exiernal Dinurbances 2 16.5. Measurement Noise and Prefiiring. — 3a 16.6- Effect of Sampling Rite on Quantiration Error — 3 17. Real-Time Software for Implementation of Feedback Control David M. Auslander, John R. Ridgely, and Jason-C. Jones . 17.1_An Application Context _ 12 The Software Hierarchy - 133 The Ground Rules... esse 124 Poruability — 17.5. Software Structure: Sean Made _ 128 Control Software Design 12 Design Formals —_ 178 Scheduling es ee 19 Tak Type Preference, eee 17.20 neta Communication - 17.4. Prototype Platform. : 17-12 Progyain Struehre: The Anatowaiy of & Conta Program 336 1213Sample Problems and Utilities _ _ 344 18 Programmable Controllers Gustaf Olsson . M7 TH.1_The Development of Programmable Controllers _— 3a 18.2 Binary Sensors and Actuators 18.3 Flementary Switching Theory 184 Ladder Dingeams 18.5 Programmable Comteollere 2.22 ee ee eee eee 18.6 PLC Programming oe 187 PLCeas Part of Automation Systerns SECTION V_ Analysis and Design Methods for Nonlinear Systems 19 Analysis Methods Derek P Atherton woe 367 TSLI_‘The Describing Function Meiod a 102 The Phase Pane Method a6 20 Design Methods... _3ai 20.1 Dealing with Rewator Saturation —R W_Middln oe ~e 20.2, Bumplese Transfer _ Stefan F Grarbe and Anders Abn 38 203 Linearization and Gain-Scheduling, Jef Sharma. - 3m SECTION VI_ Software for Control System Analysis and Design 21_Numerical and Computational Issues in Linear Control and System Theory AJ. Laub, ‘RLY Patel and PM, Van Dooren 403 Bit Inieductiog oe 43 212. Numerical Background Do - = 405 113, Fundamental Problemein Nomerical Linear Algebra 407 24 Applicationsie Systems and Control =. coeveeneeeeee 410 21S: Mathemnttical Sofware... 416. 2S Concluding Remarks . - 37 22_Software for Modeling and Simulating Control Systems Martin Orter and Frangois E. Cellier 419 1 intro {222 Special Demands of Control Engisecrs fr a Simulation Tool 223 Overview of Madeling nd Simulation Software... 224 Shooming of Gren Simlcen Sees LS Conclusions... a 23. Compune-Alded Contre Sems Design © Mayra Rival and ChrispherFoling 23.1 Introduction . . 202 A Brief History of CACSD St 333 The State ofthe Artin CACSD a 234 CACSD Block Diagram Toate 2.5 Conclusions Index I Mathematical Foundations Mathematical Foundations L.A Differential Equations soe ioonennennesisee 3 ‘Chisical Slition* Method of Commotion 12 Difference Equations ° lel Coplitions as et uive Salton * Classical Sites Method of BP, Lathi ‘Convalution Calor Stote University. Socemento References. . is 1.1 Differential Equations A fnction containing variables and their derivatives is called a differential expression, and an equation ivobving differential e&- pressions is called a differential equarion. A differential equation isan ordinary differential equation if t contains only one inde- pendent variable; ii partial differential equation if i contains ‘more than one independent variable, We shall deal here only ‘with ordinary differential equations. Inthemathematical ters, tbe independent variableis generally + whieh ean beanythingstch as time, distance, velocity presse, and so on, In most ofthe applications i contr systems, the independent variable is time. For this reason we shall use here Independent variable + for time, although it can stand far any cur variable aswell, ‘The following equation Ys osu = 433 +574 =sine i an ordinary differential equation of second arder because the highestderivativeisafthesecondorder, Anntb-orderdiferential equation is heer if it is ofthe form sat ten an Se bande + agit) ple) = ele) (Lt) where the coefficients ay) are not fusions of 9(7). these caeticenss (a) are constants, the equation is linear with can corfciens. Many engineering (ae well as nonengincering) systems can be modeled by these equations. Spsterss modeled by these equations are koh as linea tnse-ievariant (LTD ss tema. In thirchapter we aha deal exclusively wit linea der emial equations with constant coefficients. Certain other farms ‘of ifferential equations are dealt with elsewhere in this wlume (© Mi by CRE re LLC Role of Auxiliary Conditions in Sotution of Differential Equations ‘We now show thata differential equation does mot, in ge- cel, have a unique solution unless some additional constraints (or conditions) on the solution are known. This fact should ot come 35 6 surprise. A function yl¢) has 2 unique derivative yt ft fora given derivativedy there are infinite possible fonctions y(n}. Uf we are given dye, itis impossible to de- termine y(t} uniquely uniex an additional piece of information about yi@) is given. For example, the solution ofa differential equation ge ay caine by integrating bath sides the equation i wi) =a +e ua for anyvaluecfe. Equation |.2 specifies function whase slopeis 2 for all r. Any straight linewith a slope of 2 satisfiesthis equation, Cealy the solution is not unique, but if we place an additional coestainton thesolution y(t) then wespecify a unique soluton, Far example, suppose we require that y(0) = 5; then aut of all the possible solutions available, only one function hus a slope of 2Land an imercept withthe vertical area 5. By eting = On Equation 13 and substituting y(@) = Sin the same equation, we obtain yO} = 5 = cand yaya es ‘which isthe unique solution satisfying both Equation 12 and the constraint (0) = 3 Inconclusian, differentiation isan irreversible operation dur- ing which certain information i lost. To revere this operation, ‘ne piece of infarmationabout v(r) mustibe provided ta restore the original y(}. Using a sitmlar argument, we can show that, piven d! yfde?, we candetermine vit} uniquely only ifwoad tional pices of information (constraints) about y(t) are give. 4 4 In general. adetermine ylt) uniquely fromits nth derivative we ‘iced additional pieces information (constraints abou y(). ‘These constraints are also called muxiliaryconditions. When these conditions are given a ¢ = 0, they ae called ttl conditions. “We discuss ere two systematic procedures for solving lin- «ar dffeeotaleqpationsof the form in Equation 1.1. The first method isthe claical method, which i elatively simple, bat restricted ta a certain class of inputs, The second method (the convolution method) is general and isapplicabevaall types ofin- pus, A third method (Laplace tranaform)isdiscused lewhete ‘i this volume, Both the mthods discussed ere are classified as time-domain methods because with these methods we are able 1 solve the above equation directly, using fas the independent ‘variable, The method of Laplace transform (also known asthe requcocy-doraain method), om the other hand, eequires tans ‘oemation of variable # into a frequency variate 5. In engineering applications, the form of linear Aiffererstial ‘equation that accurs mast commonly i given by FR ob ann eh toot or ban vie abn Ge + an Gok +t nL tbo fs) (haa) ‘where all the coefficients ay and by are cantante. Using ap rational notation D to represent jdt, this equation can be expressed as (DP 4 eg DIN bo ha D + a9 (be D™ + bg ADO! ob tbo) (PD (Lab QD) x10) = PLD FED (te) where the polynomials QD} and PUD), respectively are OID) © DY + aD +--+ aD oH PED) = bg D® ag DOE fo byt by (Observe that this equation is ofthe form of Equation 1.1, where rf) sin the form of aTinear combination of f(t) and ite dries: tives. Je this equation, y4¢) represents an output variable, and {F(@) representsan ing variable ofan LT systesh. Thearetically, the powers m and ninthe above equationscan taken any value, Practical noise considerations, however, require |] m1. 1.1.1 Classical Solution When (0) = Equation 14 i knawn as the Rariopencaas for complementary} equation. We shall rstsalve the homogenous ‘sjuation, Lecthe solution ofthe homogeneous equation bey. (0), thats, ODrye(n (DP bg DIME poo hay D + ahye(P CONTROL SYSTEM FUNDAMENTALS ‘We first show that ifyp(@) i the solution of Equation 1.4, then y(t) + ype als is solution, This follows from the fact that ODI ° Mf yp(t) is the solution of Equation 14, then O(DIypit) = PDAS ‘Addition of these tn equations vies DY [rted+ yp(0] = POD FY ‘Ths, zo(4h + p(t) satisfies Equation 1.4 and therefore is the {generalsolotion of quation 1.4. Weecally(tthecaniplementary solution and yp(t} the particular solution. In ayster analysis parlance, these components are called the matural response and the forced response, respectively ‘Complementary Solution (The Natural Response) ‘The curaplementary solution y(t) ithe solation of OUD Iy.t0) =O (5a) (PP ead + --+a,D+aq) (0 (150) ‘A solution to this equation can be found in a systematic and formal way, However, we will take a shoet cut by using heuristi reasoning. Equation 1.5 shows that a knear combination of “yl t) and is successive derivatives fs zer, not at some values ‘oft, but fr all. This possible saad only f (0) and alts m successive derivatives are of the save foren. Oxherwise their sum an never add to zero forall values af ¢. We know that only an exponential function es this property, Soe ws assume that ae lth =e solution to Equation 1.5b. Now Delt Dri Dtytt ‘Substituting these results in Equation 1.56, we obtain (2 bay tt chek tog) eM Fora nontrivial solution of this equation, det Feber tay =O (sa) This result means that ee! is indeed a solution of Equation 1.5 provided that satisfies Equation I.6a, Note tha the polynoctial 1.1. DIFFERENTIAL EQUATIONS in Equation 16s is idlentical to che polynomial Q(D) in Equa- tia 1.56, with 3 replacing B. Therefore, Equation La can be expressed as aw) When 200) is expressed in factorized form, Equation L.6b-can be represented as (1.88) 00) = (AINA = A) =a: @ (16) (Clearly hast SoDutIONS: 2, dey oon ns Consequently, Equa ion 1.5has n possible slut esate ith, Chacon ax arbitrary constants, We can readily show that ‘general solution is given by the sum ofthese n solutions! sa that volt) = eel eye pees eget (1) where c1, ets oo. s eu ae arbitrary constants determined by sonstraints (the auniliary conditions) an the salution. The polynomial Qi} i known 28 the characteristic polyno: ‘ial. The equation 20)<4 ou is called the characteristic oe aunary equation. From Equa- sion 1c, it fe clear that Ay, 8g, ay Ay are the roots of the characteristic equation; consequently. they are called the char= scteritc roots The teens characteris ales, egies ad natural frequencies ae alsa used far characteristic roots, The exponentials ¢() = 1,2,-..4n) in the eamplementary solu tiom are the chamacteritc modes also known as made or natural modes). ‘There isa characteristic mode foreach characteristic rot, and the complementary solutizn i limenr combination of the characrerzic madet Repeated Roots “The solution of Equation 1.5 as glven in Equation 1.7 as- sums that them characterise roots Ay, are distinct. there are repeated roots (same tot occurring more than.once), the form ofthe solution is modified slightly, By direct substina- tion we can show that the solution of the equation (w~a)*rtn "Taprowe this ft assume 1 ya(?)areallsations RO of Equnson 13, Thee ayn = 2 ann = 9 Ann = 0 ‘Makiplying these equations by cive2,...-cq. respectively, and adding tem topebier yds OUD i eayeer eeapalt too tenet 8 ‘This reat howe that c(h + e2¥eht) + o> + capa) also a elution ofthe harmo geneous Equation 1.$ The teat eprom se Germ for caseteritic vo is given by ye) = de tenet Inthis case the rot repeats twice. Observe thatthe eharactr= intc modes inthis caseare eM andre, Contiaing this pattera, we can show that for the differential equation (Bayan (as) the characteristic modes are ©! 16, Pe, and the solution is rel (teat te tert 110) ‘Consequently fot a charac istic polynomial 20) =~ A)" deat the characteristic modes are ht! tei! ht, and the complementary solution is Jeff sent eh terete te tenet feito Particular Solution [The Forced Response: Method af Undetermined Coefficients ‘The particular solution yp) is the solution of QD )yp(t) = PDI aan) ‘ee a relatively simple task go determine yp(1) when the input ‘(tin soch that it yields only a finite nurnber of independent derivatives. Inputs having the foren.e€ ot” fall into this cat gory, For example et! has only ane independent derivative the repeated diferentiation of e' yields the same form, that is, ‘Similarly, the repeated differentiation of «” yields only r independent devivatives. The particular solution to such ain spot can be expressed asa linear combination af the input and its independent derivatives. Consider, for example, the input Ft) = at! + tbe. The successive derivatives ofthis input are 2at + 0 and 2a, Io this ese, the input has only two independent derivatives. Therefore the particular solotion can be assumed ta bea linear cambination of f( and its two derivatives, The soitable fore for yp) in this case is therefore nth = Bar + Bar + Bo ‘The undetermined coeificients lo, 61, and fly are determined by substituting this expression for ypi!) in Equation 1.11 and then equating coefficiests of similar terms on. both sides of the resulting expression, Although this method can be used ony for inpurs with a inite number of derivatives this class of inputs includes a wide variety of the most commonly encountered signals in practice. Table .1 shaw a variety of such inpwts and the forms ofthe particular solution corresponding to exch input. We shall demonstrate this procedure with an example Note: By definition, yp (¢bcannothave any charactriticmode terms. If ay teem pt) shove in the eight-hapd colina for the TABLE. Epo Forced Reipone Led payee pelt hom a Biett 3k (aconsane) B faconmant) 4 cas (at +8) Poolat +8} SAE arent! oe Batt + Bratt pays + ag) ef +P + Pale ‘particular solution isalso-a characteristic mode, the correct form ‘of the forced response must be medified ta! p(1}, where isthe smallest possible integer that can be used and sill can prevent ptt) from having a characteristic made term. For example, when the input is «, she forced response (right-hand coleman} haste form Be‘. Butife* happenstobeacharacteristicmode, the correct Form of the particular sobution it Bre" (see Pair 2) Ite also happens to bea characteristic mode, the correct form. of the particular solution is ref, and s0 on. EXAMPLE. Solve the differential equation (oP 43042) 40) = Dye au ifthe input andl the initial conditions are y{0*) = 2 and j(0*) = 3. “The characteristic polynomial is PeMere+NAtD ‘Therefore the characteristic mades are e~t and e“*, The com> plementary solution is a linear combination af these modes, $0 that yiseettge® 20 Here the arbitrary constants€} and cp must he determined from the given initial conditions, ‘The particularsolution tothe input? + 51 +3 isfound from Table 1.1 (Pair 3 with ¢ = 6) oe ptt) = Bal? + Bur +B Moreover, yp(t) satisfies Equation 1.21, that is, (0? +30 +2) pt) = ptr uy Now . fea : Dy) =F (mae mie tte) = 2mar tte é 2 a = Dye) =F (ui +e + to) = am CONTROL SYSTEM FUNDAMENTALS and osore £[Pesers] ames ‘Subsctuting these results in Equation 1.13 yields Ba + 32Bat + 81) + 2B + ae + BY = ES 2fat* + Bil + 6PM + fy +38) +261 =D +5 Equusting cocticents of similar powers on both sides af this ex ‘pression yields 2h 2h +662 2As+ 361 +28 ‘Solving thesethree equations for their unknowns. weabtain = 1h = 1, and fy = Therefore, yet = t+ tea ‘The total solution yi) is the sum of the complementary and particular solutions. Therefore, MO = yO) + ypttd fet hee ert oreo that Mo) = ace +1 Settings = Gand substinating thegiveninitial conditions y(0) = ‘and j(0) = 3 inthese equations, we have 2 3 atat! me 2g tl The solution these two simultaneous equations ic, = 4 and cr = 3. Thesefore Hep de! 3 tro ‘The Exponential Input ett ‘The exponential signal isthe most important signal in the study of ETL sates, Interestingly, the particular malution for an ‘exponential inp sigoal torns out to be very simple, From Tas be 1.1 ve ze thatthe particular solution for theinpute® bas the form fe‘. We mow show that @ = Q(¢)/.P(C)*. To determine This steve ont ¢ is pot a characteristic root. LLL. DIFFERENTIAL EQUATIONS the constant 8, we substitute yp(t) = Bet in Equation 1.11, which gives us OUD) [ae] = P(D yet (ata) ‘Naw observe that a 4 io De Ae) age! ae viet = F(e)ace De Consequeni. cme = eee and P(D etm Pte et! ‘Therefore, Equation 1,144 becomes POU = PrEde! (auto) ‘Thus, forthe input f(t) = et, the particular solution is given by yelth= Hee to (asa) Hi = Si This isan interesting and significant result Tt tates that for an exponential input e" the particular solutan yi) is thesame ox- pponential muliplied by 4 (€) = P(Q)/ Ce). The toa sation _y(f} to an exponential input eis then given by (1.88) vith = Figen where the arbitrary constant ci, c2 auxiliary conditions, ‘Recall that the expenental signal inchudes a large variety of signals, such as.2 constant (¢ = 0}, sinusoid (¢ = je), and an exponentially growing or decaying sinusoid ({ =o + ja). Let us consider the forced response for seine of these cases. +0 aredetermined from ‘The Constant Input ft} = © Because € = Cel, the constant input is » special case of the exponential inpat Cet! with £ = 0, The particular sotion. {otis input is then given by swith yet) = came ceo = CHM (146) ‘The Complex Exponential Input ot Here ¢ = fn and Yale) = Hane! a7) ‘The Sinusoidal Inpet ft} = cosaigt “We naw that the particular solution forthe input 2s Hc Jane®l, Since cos cot = (e+e I0) 2. the particle solution to cos aut is ptt) = Ef nual + 1 jare] ‘Because the two teems.on the right-hand side are conjugates, ott) = ne [reel] Ho) = (Hole! 4 ark) = Re urtjwyetioredervwy = WGjajcos fore Ge] 0.18) yal) = 1H iw)| cos [aw + 8 4+ LH Cieo)) a9) EXAMPLE 1.2: ‘Solve Equation 1.12 For the Following inputs: fa) We" (bY fe) e-™ _(€) Wome 3" + 30") ‘The intial conditions are (0) = 2, j(0*) = 3. ‘The complementary solotion for this ease is already found in Bxample 1 as yelthesciet+oe® peo For theexponentil input f(t) = the particular solution, as found in Equation 1.158 H (tet! where Pith £ may PO et = Guy - FER eE a) For input #0) = te, ¢ a tow(—aye 3 = 1o/— > fantom = ate ype) = roo ‘Theta solution (the sum ofthe camplementary and particular salutions) is V(t) = cre"! beget — 150 reo ‘Theinitial conditionsare y(0*) = 2and j{0*) = 3. Settings = O in the abowe equations and substituting the initial conditions yields ater i=2 md ede t4s=3 Solution ofthese equations yields c) = —8 andey = 25. There- fore, ie) ate ae 1 (b) For input fla) = Sm Se, ¢ = O,and spit SHO}=0 r>8 ‘The complee solution a y(t) = ye(t-+ yp) = ere" +epe™ ‘We then substitute the initial conditionsto determine cy and cz ‘ascxplained in Part a (cl Here ¢ = —2. which i abo a characteristic rot, Hence (see Pair2, Table 1.1, or thecommentat the bottom ofthe table}, ptt) = Be“® “aind , we eubitteyp(?) in Equation 1.18 giving os (D7 +3D-+2) yp) = DFU (8 +3042)[ne*]— pe be pla] = pa -ane* pifme2] = wu -ne™ pet 28 conse Boar = 443604 2re p(t) = ie The complete solution is (4) = pel) + yplth = cre! + ce"?! + 24e"*. We then substitute the initial conditions to determine ¢ and ¢> a8 explained in Pata. (d) Bor the input f(r) = LOcos(ir + 30°), the particular solution [see Equation 1,19) is p(t) = LM U3) oe [Be+ 30° + L27U3)} (CONTROL SYSTEM FUNDAMENTALS where a 409 = Gos UF a sD ae Bp uaja cme © eg et came Therefore, InGa| 026, LMG) = 39" and yptt) = 100.283) 008 (+ 30° — 37:9") = 2é3con (31 = 7°) ‘The complete solution is y(t) yl) + ypl) = jet + eye 42,65 cos (3t ~ 7.9"), Wethen substitute the inital con- ditions to determine ¢) and ¢3 as explained in Part a. 1.1.2. Method of Convolution Inthismethod, the input (1) is expressed asa sum of impulses. ‘Thesolution is then obtained as 2 sum of the saktions to all the impulse components. The method exploits the superposition [Property ofthe linear differential equations. From the sampling or siting) property of the impulse function, we have HO) = fo fase =x) de ‘Therighthand side expresses f(¢) assum (integral) of impulse components. Let the solution of Equation 1-4 be sé) = &e) sehen (0) = S(t) aid all the intial conditions are ter. Then ‘use of the linearity property yields the solution of Equation 14 roinput f(r) r20 (120) YG) = fo Fda = abd «my FFor this solution to be general we must add a complementary solution. Thus, the general solotion ie given by Lot's [pene —nar an) im vo where the lower limit 0s understood tobe 0” inorder to ensure thot implies. if any, in the input F(‘)at the arginareaccounted for side f (1.22) is well now i the literature asthe cna sion erga, The function ht) appearing in the integral isthe soliston of Equation 1.4 for the impulsive input (f(0) = B¢ehL ean beshown tht [1] (0) = PUDdtya(enete)] (aay Where yo(t}is a linear combination of the characteristic mades subject to initial conditions ae KO) = yA) = ov me PO) me (.24 1.2. DIFFERENCE EQUATIONS ‘The function wif) appearing on the right-hand side of Equa- tion 1.25 represents the unit step function, which is uaity for £2 Ound sO fers <0. ‘The right-hand sie of Equation 1.23 isa linear combination Of the derivatives of yo(thu(f}. Evaluating these derivatives is clumsy and inconvenient beeause of the presence of ut). The dlrivatives will generate an impulse and its derivatives at the origin [recall chat {-e(2) = #4)], Porcunately when quasioa 14, the solution simplifies to (ch = B84) + (PCD) yo (0) ]uled (125) EXAMPLE 1.3: Solve Example 1.2, Part a using method of convolution ‘We first determine hig). The characteristic modes for this case, 1s found in Example |.1,aree™! ande™™. Since yo(t)inalinear ‘combination af the characteristic modes volt) = Kiet + Ke 130 ‘Therefore, Sole Kiet ae eo ‘The initial conditions aeconding ta Equation 1.24 are Ze 1 snd yo (0) = 0. Setting # = On the above equations and using the initial conditions, we obtain Ki+Kie0 and =) -2K=1 Solution ofthese equations yields Ky = Land Ky = ~1, There: fore, wlth = ‘Also in this casethe polynomial P(D) = D is ofthe first-order, and by = 0. Therefore, from Equation 1.25 buy (PCD) yn (edulrd = [yet ]ate) [ger - | ait) fret + 20 uth [erie $2 de = Set ane — 15 and [fom ~ ar ‘The total solution is obtained by adding the complementary s0- lution ye(Q) = Ge! eye! to this component. Therefore, yO) cre tt ene = Se + 206? — 15e* Seting the conditions y(0*) = Zand y(O* ) = Jim thisequation (and its derivative), we obtainey = 3, ¢3 = $0 that ve which isidentical to the solution found by the classical method. Bet 4 ase — ase re0 Assessment of the Convolution Method ‘The convolution methad is more laborious compared to the classical method, Howevee, in system analysis, its advantages ‘ourveigh the extra work. ‘The classical method has a veriows drawback because it yields the tral response, which cannot be separated into-components arising from the interaal conditions ‘and the external input, ln the study of systems itis important to be able to express the system respanse to an input f(r) as an explicit fapction of f(t), This isnot possible in the classical method. Moreover, the elassical method is restricted toa certain lass of inputs itcannot be applied to any input *. ‘fe rast solve a particular linea differential equation find a response of a particular LT! system, the classical method may bethe best Inthe theoretical study oflinear systems, however, i practically useless. General discussion of differential equations can be found in numerous texts on the subject [3}. 1.2_Difference Equations The development of difference equations is parallel to that of diflerential equations. ‘We consider here only linear «ference equations with eoastant coefficients. An nth-order difference ‘equation can be expressed in two different forms; the first form vwss delay vermis such as k= 1], vik = 2], Uk ~ 8, 1k ~ 2], -- etc, and the alternative form uses advance teers such as YR + IyIR+ 2h... ete. Both forms ane useful. We start ‘here with a general nth-order difference equation, using advance ‘operator form vile k a] bag aylh bam 1] eos bay yl + 1] + opp EL Bm FUR | dig FL a — 1 bb fle) + bu sTA) 126) Causality Condition ‘The leftchand side of Equation 1.26 consists of values of yA] at instants k +, ktm —L. k-+a—2, and soe. The right-hand side of Equation 1.26 consists of the input at iastants Jet m, ket 1, ttm = 2, andso on, Foraeausal equation, the solution cannot depend on future imput values. This shows that when the equation isin the advance operator farm of the Equation 1.26, causality requires mv 0 ‘Time reversal ost (ce) = F (ar) sm cornplen conjugate of F feo) Mukipicationtyapowereft —lr*ft1) Fio).n=1.2,.. Mulupion expen?) 31 F(hexpGienitl Fe ~ oy) orem re mimber ae Mulipicaionbrsiieys) 317 ()sin{ens)] = (4/201 F (w+ o0) — Flas — a} Muliplcasionty coment) 3] f(r )eoutent)} = (1/2)] Flor +a) + Fle ~ om)] Diecut dona 3 fF] = Ua Fann te Mulupsion inthetimedarain [FC )gC0HT = gk F(a) « Gle)] Coovebation in the time domain =| f(N} = g(t} = Flw)G Car) uaiy arto = 2s Panera thecresn FUdgtndt = gb [2 Fl-w)Glardo Specials of Parevalstheceem J, f4u)de = gh [IFC des TABLE22 Comin Foiier aman, Fal T aredla) l -Tftsrsty2 fem { Qube Bun (Te) sin [ne omser . 0. allethr eM any > 2» eM any pene - 2xd(o =o) cos (aig +4) [et + and + 2! 8(w = wn), Sinton! +8) dt fe7!"8 (+ oon) — 9 btw — 0n9)] Given a function J), the set of all complex numbers such that the integral in Equation 2.15 exists is called the region of sonvergence ofthe Laplace transform of f(t). For example if 7 (07s the anit-ep function wf given By w(t) = fore = © and u(t = 4 fort < 0, the integral in Equation 2.15 exists for anys =o + jr with ral part @ > 0. Henée, the region af ‘convergence isthe set ofall coreplex numbers with positive eal ‘part and, for any such the transforms ofthe unit-step function suds) is equal to 1s. Given a function Fi), if the region of convergence of the Laplace transform Fs) includes al complex mamibers 2 = jie foro ranging from —oote-20, then (0) = F (Momo is well defined (ie exiss} and is given by Fw) = [5° sene™ae Thea if fe) = 0 fort « 0, the righthand side of Equation 2.6 is equal tothe FT Fo) of (7) (eee Equation 2.1). Hence, the FF of f(t) is given by Fle) = FisVenjon aan (Note that we are denoting FLjan) by F(a.) This fundarmental result shaws thatthe FT of (1 can be computed rectly from the Laplace transform F(a)if /4#) = Ofor < Oand the region (2.16 of convergence includes the naginary axis of the complex pline (all complex nusbers equal to jes ‘The Laplace transform defined by Equation 2.15.can be viewed as an operator, denoted by F(x) = LL f(s) thar maps a time Function J(7) lato the function Fs) of the complex variable s, The inverse Laplace transform operator isoften denoted by £~!, and is given by BPS xinetds 02 is evaluated along the path + = + fein the complex plane frame — foe toe + joo. where «is any real umber for which the path e+ Jo Tes in the re gion of corwergence ofthe transform F(s). {is often posible ‘tw determine f(F without having to use Equation 2.18: for ex ample, this i the case when F(s) isa rational fnction afs, TRE ‘compuration ofthe Laplacetransform ortheiverse transform i ‘often facilitated by using the properties ofthe Laplace transform, ‘which are listed in Table 23. Tn this table, f(t) and gt) are two functions with Laplace transforms F(s) and Gs), respectively and u(t) is the onit-step function defined by w(t) = 1 ford 2 0 and u(t} = O for r «<0. Using the properties in Table 2.3, itis posible to determine the Laplace transform of ouany common fonctionswithouthaving to useBquatinn 2.15. Alistofeom pion Laplace transforms is given ip Table 24, of the (One-Sied Tray + bea Life ~ au Liftan| = Lhe" fe) t[ern] L[fr0] CONTROL SYSTEM FUNDAMENTALS Teanaborm Tamera Troperey [= aFG) + BGC tr any sce, B = fo)] = F(a) expl= sty) far aap ig > © Fs (a) foray rel mame > 0 Film = 12... LL FCO} | oe FAs) bry seal ce corplen see E|¥(¢hsinteoe)] = LIF Meesteu!)] = O/2)F + joo) + FC — Jamil sFu) DIFF + fon) = Fis = jo) — fy FFU) — af) — FCO) [fre] =er0 0490-2 gyo---- Eby tft) = ‘Convolunion sn the time domain Lifer aco aeons fatness ‘Final-raive theorem IEF (FD hana finite liemie f (0) Timmy 0c Fis) TABLE2A Gammon Laplace Transforms, fa) Luger Taniem F) 14(7) = amity fancion w(n) — wl = T) forany T > 0 3} = wnieempulee 3 ~W) for any ty > 0 sro Tera tel 2.2.2 Rational Laplace Transforms ‘The Laplace transform Fs} of a function £42) is said to be a rational fenctlon of fit can be written as a ratio of polynornals, ing that in Ft) = Bi vere Ws) and (6) ae polyno in the comple arable x given by mts) 219) Dg bg nM! eevee Bus + hg (220) Diy meager has tay (221) In Equations 2.20 and 2.21, m andi are positive inwegers and the ontario Ba ad + 01, 9 a ral umber. In Equation 2.1, sts assured that W(x) and Ds} da sot have any common factors, if there are comunon factors. they should always be cancelled. Also note that the polynonsial Dés) is monic; thats, che coefficient of ¢" is equal to 1. A rational function F(s) san always be written, with a monic denominator polynamial D(s). The integer m, which is the degree of Dis), is called the onder ofthe rata! function Fs). It is assumed that = Mm, in which case F(5) i said ta be a proper rational famction. Ifin > m, Fis} is said to be strictly proper. Given rational transform F(s) = M(a)/D(s) with N(0) and (Da) defined by Equations 22nd 221, let £1, £2... m de> note the roots of the palynomial M'{s), and let pr. P2, Pr denoce the roots of D(s); thar is, Nz.) = 0 ford = 1, me and Digi = Ofori = 1,2, In general the xj and the py may be teal or comple numbers, but if any aze com: ples, they must appear in complex conjugate pair. The mure- (bers £1. 26.0.5 2m are called the zeros of the rational function Fis) since F(s) = Owhen s = 2) fori = 1,2,...,m; and the mumbers py, p.-.-. Py ar called the poles of F(x) since She cde [Fees ne er pprnces fo IF (eh isarcly proper > m) andthepoles pt, Ps. of F(x) are distinct (nocrepeated), then F(2) has the partial fraction expansion Foe Spe Septet Se am where thee; are the residues given by cree PAF Moms FB (223) For a given value of f, the residue c; is rel if and only if the coreesponding pole pis real, and ¢, i complex ifand-anlyif p, is complex. 2.2. FUNDAMENTALS OF THE FOURIER, LAPLACE, AND z-TRANSFORMS u From Equation 2.22 we seethat the inverse Laplace transforma F(t) is given by Uhefollowing:sum of exponential functions: LE) = 10 gel peo beget (2) fall the poles py pa. .-. po F(a) are real ourabersthen Fé) isa sum of real exponentials given by Equation 2.24 If Fs) has a pairofcomplex poles p =o + ja then f(r)containstheerm 4225) where ¢ isthe complerconjugateofe. Then writingeia the polar form c= |eje!*, we have cel td celtic a ele Hele- it fle ant. eset] (226) Finally, using Euler’ formula, Equation 2.26 can be written in. the form eel 10 5 ael4— os Dele costa +0) (227) From Equation 2.27 itis een that f F(e) has a pair of complex poles, then (1) contains «sinusoidal term with az exponential amplitude factor e**, Note that if @ = (so that the poles are purely imaginary), Equation 2.27 is purely sinusnidal. one of the poles (sy ps) is repented r times and the other =e poles ate distinc, Fs) has the partial fraction expansion ri) 2 flog « Goa Ga aan Pe ‘In Equation 2.28, the residves cr, crs +a recalculated ‘as in the distinet-pole case; that is, $= (= POPs FEE NEE Roo 2291 aand the residues o1,€2....4 are given by = chp [SS le-errwl} asm “Then, taking the inverse tanstoem of Equation 2.28 yields FU) = ce pepe eet erlert peretetet poo tegell a) “The absve results reveal that the analytical formofthefunction Jad) dpe directly om the poles of F(s). In particular, if F(a) has a nanrepeated real pate py then f(t) contains a real ‘exponential erm of the farm.ce” for sonve real constant c. If 2 real pole p is repeated r times, then. /(F) contains terms of the form eye", cateP™, ee" for some real constants f1.€2e +e Gre HE F{s) bias a nabrepeated complex pair o & jar of poles. then (9 costains aterm ofthe form. ce costa +8) forsome real constants ¢ and #, If che comples pair jos is repeated times, (7) comtainsterme ofthe form eye cosine + Bi deeate conlat + 8)... n6jA"!e" costa + f) for some redlcanstantscy,€3..0. Ce amd #83... 8p. These results are summarized in Table 25, Fs) is proper, bot not strictly proper {oo that = min Equations 220 and 271), then using lang division Fe) can be writen in the fates Fe) =n + where the degree of Rs) is stray less than 9. Then Ris) /Dés) ‘an be expanded via pantal fractions as wasdone in the easewhen F(s)isstricaly proper, Note hat for Fs) given by Equation 222, the inverse Laplace transform. (1) comtains the impulse il (C)- “Hence, having n = ain Fs) resus in an impulse term in the inverse wanstoem, From the relationship between the poles af F(e) and the an- lytical form of (0, it follows that 74) converges to 24r0 a8 1 => oo if and only ifall the poles p1. pa... Pe of FAs) have ‘eal partsthat ore strietly Less than vero; that i, dte( yy) < Ofor 4 = 1,2....4m, This conditions equivalent to requiring that all ‘the poles be locxted in the apen left half plane (OLMPI, which is ‘the region in the complex plane tothe lef of the imaginary axis. tals falls fev the elation ship between the poles of FS) sand the form of (that f(t) has aBnitetimit foo} sr + 98 if and only ifall the poles F (s} have rea! partethat are essthan zero, except that Fs) may have # nonreprated pole af = 0. la smathematical terms, the conuitions fac the existence of a finite limit foo) are (amy Retpa <0 fom forall poles yy #9 is apole of Fis), then py is nanrepeated Gan a3) IF the conditions in Equations 233 and 2.34 are satshed, che Himiting value f (00) given by Hoe)-= [SFO pan 235) ‘The relationship in Equation 2.89 is a restatement af the final: value theorern (given in Table 2.3) in the case when FCs) is ra tional and the poles of Fs) satisfy the conditions in Equations 2.38 and 2.44, 2.23 Irrational Transforms ‘The Laplace tansfarm F(x) ofa function f(a) is sid to be am {rational function of iis not rational; thats, Fis) cannot be expressed at ratio of polynomials ins. For example, Pts) = ¢€-% fs isitrational since the exponential fonction ¢** cannot Decspressedsa ratioof polynootialsins, Inthiscate theimeerse transform fA) is equal ta u(t — to} where u(7 isthe wnitstep function, ‘Given any function f(r) with transform. F(s) and given any real purober fy > 0, the transform of the time-shifted for time: CONTROL SYSTEM FUNDAMENTALS ‘TABLE 25 Relationship Between the Poles of F (5) andthe Fact of J (P Pole Lecten of Fa) ‘Gorvespoosing Tes in FU “Nonepened ealpalewe= pee eal poe at = p sepened tes [Nonrepeatn complex pirat = Carpe rir delayed) function £(F— tatt~ ta) inequalto F(s}e"%. Thme- dllayetsgoals arse in systems with te delays, a has irs- ‘tional transformsappear in the study of systems with time delays. ‘Alo, any functian (rf) that is af finite duration in time has a ‘transform Piz) that is irrational. For instance, suppose that He) = pte [alt es) at — 1), OS sy 1238) sothat { (2) = 1) for ip £1 < ty and f(t) = 0 for allether'. ‘Then fl) ean be written ia the form, FUE) poke = tant = 29) — pie = le a) (2.397) ‘where yoie) = y(t + to) and yy(E) = UE 1). Taking: the Laplace transform of Equation 2.37 yields Fs) = Polnde™™ — Py (elem 238) where P(e) and Ty4s} are the sransforms of yo(r) and ya), respectively. Note that by Equstion 2.38, the transform Ps) is an irrational function of ‘To strate the above constructions, suppose that fre" fue 1) wien) 239) ‘Writing £40) in the form of Equation 2.37 gives, Fy = He Mee = 1) = ele = 29 F240} ‘Thea, transforming Equation 2.40 yields Fis) = [ee a) ‘Clearly, F(s1 isan irrational function of s. 2.24 Discrete-Time Fourier Transform et {18 be a real-valued function of theintegervalued variable 4, The function £4) can be viewed asa discrete-time signal in particular, J (i) may bea sampled version of a continuous-time signal £0}. Move precisely, fk) may be equal to the sample values f(T) ofa signal 0) with ¢ evaluated at the sample ‘utnes1 = ET, where 7 iste samplinginterval, tn mathematical terms, the sampled signal is given by FUE) = Hoar = FET), k= 0 1, 2, (242) Note that weare denoting /4k7) by #(k). The FT ofa funetion (eof an integer variable is defined by FiQ= Ps, =m W, comparing Equations 2.46 and 2.43 reveals that Fa FC). Hienee, the DFT fy is equal to the values of the DTFT F(@) with © evaluated at the discrete points S2 = 2en/N for n= 8120 = ‘Thecomputation ofthe DFT Fy given by Equation 2.46.can be ‘carried out using fast algorithm called the Fast Poorier trans- form (FFT. The inverse FFT can be used to compute FE from ‘Fe, A development of the FET is beyond the scope ofthis section ‘see “Further Reading") mel .N-1 (a7 2.2. FUNDAMENTALS OF THE FOURIER, LAPLACE, AND 2-TRANSFORMS 2B 2.25 2Transform Given the function f (2), the tro-sided (acbilateral} z-transform, F(z) of fk) i defined by Fee pet Rae) =e ‘where « is a complex variable. The one-sided (or unilateral) -transdorm of f (2) is defined by Foy = YO eyen* 42.49) Nate that if (2) = 0 fork = —1.—2,.... the one-sided and fwo-sided z-transforms of f44) afe the same, Ax isthe case ‘with the Laplace transform, in sontrolsengineering the one-sided ‘version isthe moct useful, and thus the development given blow i restricted to the one-sided xtranafarm, which is referred toa the e-transfoer, Given fk, the set ofall complex numbers z such that the ssuramation ia Equation 2.08 exits called the region of come: gence of the z-transform of f(K). I the region of convergence caf the z-transform includes all complex numbers. = efor 1% ranging from —ve to 0, then F(e!) = FA2)emqit i well defined (ie exists) and i given by Fra (ety ia But (8) ep thas gation 2.50 can be rewriten Feit (250) Fe = puaye I asy Fr) ‘Then if f(E) = 0 for all ¢ < 0, the right-hand side of Equat 2.51 is equil to the DTPT F(Gi) of fh) (se Hquation 24). Therefore the DTET of f(t is piven by F(R) = Fle aoe as) ‘Thisresult shows tat the DTT off (8 canbe competed directly fom the transform F(z) if 702) = 0 forall < Oand the region Of convergence includes all complex mumibers ¢ = e! with De < @ = 65, Nate that since |e! = | for amy value of Stand Ze! 2, the set of complex numbers z = e/° ‘comprises the unit cile ofthe coroplex plane. Hence, the DTFT of Fe) sequal tothe values afthes-transform onthe unit cece of the complex plate, assuming thatthe regon of convergence of F(e) focudes the unit circle. ‘The zetransform defined by Equation 2.49 can be viewed as an operator. denoted by F(z) = 2{ f(0)|.thatmapsa discrete-time function (&) into the function Fz) of the eomplex variable i. The inverse z-transform operation is denated by FB) = 27" F(G)), Axdiscussed elo, when F(2his a rational function fof: theinverse transform. can be computed using lang division or by carrying out « partial faction expansion of Fs}, The computation of the r-transfortn. or the inverse #-transfaem is often facitated by using the properties af the xtransioem given in Table 2.6. fnthistable, /Ub)and g(t yareiwo functions with transforms F(z) and G (2), respectively, and w(k) isthe unit-step function defined by W(X) = Lfork 2 O and a(t) =i for & <0, ‘list of common z-transfarmsis given in Table 2.7. in Table2.7, ‘he fanction (2) is the unit pulse defined by 540) = 1, 5(8) = 0 fork #0. 2.2.6 Rational z-Transforms “As isthe cave for the Laplace transform, the z-transform F(z) i often a rational function of; thal is, F(c) is given by et (253) Fi) = where N(3) and D(z) are polynomials in thecomplex variable + given by Mis) = ba thy ais bob aye thy (ZY De) = taal tae tay (2.85) Itisasvumed that theardermof F(z) is greater than orequaltom, and thus F(2) isproper. The poles and zeras of F(c) are defined in the same way as gives above fr rational Laplace ransforms. Whea the transform (2) in the rational form io Equation 2.53, theinversex-tranaform f () canbe computed by expanding F(z) intoa power series in 2! by dividing Biz) into N(2) using lang division. The value of the function J (K) are then read off from the coeicients of the power seriesexpansion. The first few steps of the process are carried out below: eee Since the value af f (&) is equal us the woeticient of 2° im the power eres expansion of Fz), it follows fromthe above division process that /4— pt) = b(n m+ ED = dg — tet nde 08 “To expres the inverse transform th) in closed form, it is necessary to expand Fic) via partial fractions. It rerns out thatthe form of the inverse z-transform 2} i simplified if F(g}/z = N(cl/Diz)z is expanded by pattal fractions, Note ‘that Ftz)/2istriety proper since F(2)is assumed 1 be proper. Letting Pi. Pas Pa denote the poles Fz) ifthe py are distinct and are nonzero, then F(z}/ has the partial fraction expansion PPOs Be tt pe (28) emt ae 26 1N(0)/B(s) where M(s) and Bis) aze polyoma fas given by Equations 220 and 2.21. It was noted in the previous section that ifthe region of convergence af Fx) includes the imaginary sxis(fo-axis)ofthe complex plane, then the FT F(a) is equal 10 the Laplace transform F(s) with s = jo. When F'(s)istationl, it tras out that the region of convergence includes the ja aris ifand only iFall the poles of F¢s) le in the LMP; thus, in this ase, the FT is given by Flo = FUDeajs (269) For example, if f(t) = ce-*! fort = Owitha > 0, then Fe) = ef 4a) which bas a single pole at s = =a. Since the point — lies in the OLFIP, the FT of the exponential function Flo) = Se 2:70) In follows fram Equation 2.70 that the 3-AT bandwidth és equal ‘twa, Hence, the farther over inthe OLHP the pole —w'is (ie, the larger ais} the larger the bandwidth of the signal, Since the rate ‘af decay to zero of ce" increases as a is increased, this result agsin confirms the properry that shorter duration time signals Ihave wider bandwidths. In the case when ¢ Laplot of ‘the magnitude spectrum |F (als shown in Figure 2.2. For any veal values of a and c, the magnitude spectrum rolls off to zero athe rate of 20 dBidecede where a decade isa factor of te frequency. ‘Fo Pigure22°-Magnhwée spectrum ofthe exponential snetion —! ‘As another example, consider the signal f(s) whose Laplace ‘ransform is Fist a sarin ay where ay is ssumed tobe strictly postive (ay > 0) lathiscas, £45) hast poles Jandy piven by P= ture tom aT am Ft, = ong VET CONTROL SYSTEM FUNDAMENTALS Whee { > 1, both poles are real, nonrepested, and lie in the LHP (assuming that uty > 6). As ¢ —+ 20, the pole pj mores along the negative realaxistothe origin afthe complexplaneand the pole pz gots to ~o0 along the negative axis of the complex plane. For ¢ > 1, F(s} can be expanded by partial fractions as follows P= toaticm 7 wom lee 7 | 27) ‘Taking the inverse Laplace transform gives £10 = gop le -eP*] ar 7 and thus f(t) ie sum of two decaying real exponentials, Since bath poles lie in the OLHP, the FT F(a is given by Fol = SoS (28 Far the case whea c = ai! = 100 and ¢ = 2, the plot of the magnitude spectrum |F (ol is given in Fagore 23. tn this ease, thespectral content of the signal f(2) ols off tozero at therate of sndbdecede, sarting wit the peak magnitude of ¥ aso = 0. 70, © 2 os Figare23Magnivude cpecruenothe signal with cansiorm F(s) = 100/¢5? + Aas + 100), ‘When { = 1s the poles py and pz of #s) are both equal to any and Fs) becorses File oie 278) “king the inverse transform gives Ft) ere an) Since ay i xsuamed t be strictly positive. when { = 1 both the ‘polesare ia the OLE in this case, the Fis Fle) = gokge (a7 Acs varies from | 10 ~1, the poles of Fs) trace oat a circle ln the complex plane with radius aig. The loci of pole locations 23. APPLICATIONS AND EXAMPLES is shown in Figure 24. Note thatthe poles begin at —oy when ¢ = 1 then split apart and approach the jar-anis at fy as 5-7 and then move to wy asf > =. For=b<¢ = lithe inverse transform af F(s) can be determined by fist completing the square i the denominator of Fs): FO= Tey (279) where sig = aly T= EF BO (2.80 Note that ay is equal to the imaginary part of the pole p given bby Equation 2.22. Using Table 2,4, we have thatthe inverse trans- form of Fr) is Fit = ek sinayt cay From Equation 81, itisseen that ft) sow contains sinusoidal factor. When < { - 1,thepoleslicin the OLMP,and the signal is decaying sinusob. In this case, the FT is Fe) axe (22) Figure 24 Loco ple of Fs) 95 ¢ varie fom U9 ‘The magnitade spectrum | Ft) i plored in Figure 2.5 for the values © = of = 100 and ¢ = 0.3,0.5,0.7. Note that for {= 0.5 and 0.3, a peak appears in | F(a. This comesponds 1 she sinusoidal oscillation resulting from the sina factor in FO). Also note that as ¢ is decreased fromm 0.5 to 0.3, the ‘peak increases in. magnitude, which signifies a longer duration ‘seilation in the signal (7). "This resul is expected since the poles are approachingthe jer-axisofthecamplex plane ax =» 0- ‘As¢ + B thepeak in |F (a) approaches, so that | F(a] does fot exist (the ordinary sense) in the limit as € —+ 0, When + O,the signal £4) is porelyaxcillatory and does aot have # FFT inthe ordinary sense. In addition, when ¢ < there is no FFT Gin any sease} since there is a poe of F (8) in the open right bale plane (ORHP). wv ms) ‘Pigere2$ Magnitude spectrum ofthesignal with transform F(x) = 100/Ks? +2058 + 100) and with = 0.7, 0.3.03, The shove results lead to the foowingganeralized propertie ofthe magnitude spectrum | F(a] ofaignal fr) whose Laplace ‘rasfoer F() station with ll poles i the OLHP: + Ifchepoles of F1s) arereal the magnitude spectrum [Ficoll simply olls.off to vera as c9 —» co, starting vita peak value atc» = 0 0f F(0). © IF F(s) basa complex conjugate pais of poles ats = 0 jg and ifthe ratio ofa is sulicenly small, then | (co) will hve a peak located approximately stan oy. 2.3.2 Numerical Computation of the Fourier ‘Transform Ln many applications, the signal f (7) cannot be given in function foam; rather allonchasareactofsamplevalies fk) = OAT, ‘where kranpes over a beet of infepes and Tis the sempling interval. Wathour lot of peneraliy. we can assum that K starts with k= 0, Also, since all signals arising in practice are of finite doration intone, we cam ane that f¢) i mf all 2. N for some postive integer N-The prablem i then ta determin the FT of f(r) sing thew values FL) = (KT fork=0,1,2,....N=1 ‘One could also carry out a discrete-time analysis by taking the N-point DFT Fy of the sampled signal 7(k). in the pre- wows section, i€ was shown that Fy is equal to PERL for mL. N= 1, where FOR ithe DTFF FQ) of 1708) wit the frequency vavable revolted at 2m. Hence the discrete-time counterpart of the frequency spectrum can be determined from Fq. Por details on this, see “Further Reading.” ‘Again letting F (wo) denote the FT of f(t}, we can carry out a numerical computation of the FT as follears. First, since fie) ize for Ound = WT. fiom the definition of the FT in Bquition 2.1 we have Fane fg fine ™a1 cae 10.1. SPECIFICATION OF CONTROL SYSTEMS formula for rise time ty and delay time iy in terms of damping ratio ¢ and undamped natural frequency ty cannot be found, ‘But useful approximations ate Lisourscaa esa! us = eate zane! ‘Note that the abowe expressions are only accurate fora second. ‘order system, Many systems are more complicated than the pure second-order system. Thus, when using these expressions, the designer should be aware that they are omly rough approx: imations. The tinse-domain specifications are quite important ‘because most control systems must exhibit acceptable time re sponses ifthe values of, Fas fr» fn and P O-arespecitied, then the shape of the response curve is virtually determined. How: fever, nob all these specifications necessarily apply to any given case, For example, for an averdamped system (¢ > 1) or a ), fy aid PO are not useful specifications, Note that the time-domain specifications, such as PO, fy €, cic. can be applied to diserete-time systems with minor madifications. (Quite often the transient response requirements are described interms of pole-zere specificationsinstead of step respanse pee: fications. For example, a system may be required tahaveitspoles lying to the left of some constraint boundary in the s-plane, 2s shown in Figure 102. ' ' 1 1 ' \ 1 1 1 ' Figure102_Theshaded region istheallowable epi tbe 5plane ‘The parameters given above can be related to the pole loe tiaas of the transfer function, Far example, certain transit 159 requirements on the rise time, settling time, and percent over shoot (ef. fr.and PO less than some particular values) may restrict poles to a region of the complex plane: The poles of & second-order system with { << Late given in terms of {and ay by Pio ~bey + jay T= Pr = =bay = jay = 0 and unstable i GM. (ind) < 6. And generally, « miniomm-phase system has a positive PM. and it becomes uestable if PM. 0. For ‘n-inienusm phase systems, care must be taken in interpreting stability based on the sign of G.M.(d) and RM. tn this cas, the complete Nyquist plotorroot locus must beexamined ocrelative stability. For asysiem in which multiple ~180" crossovers osc. the simplest approach isto convert the Bode plat tothe corre sponding Nyguist plot to determine stabty andthe frequencies a which the stability margins eccur. Obviously, systems with, treater gainand phase margins.can withstand greater changesin system parameter variations before bearing Unstable Fe should be noted that neitee the gain margin alane or the phase murgin alone gives a sullicient indication of the relative stability. For instance, the G-M. does not provide complet in- formation about the system zespansc; a small G.M, indicates the [Nyquist plot Lap) 8 very close tothe —1 point, Such a sys- tem wil have alarge My (Which means an sscilatary closed-loop time response) independentaf the PM. A Nyquistplot approach: ing the =1 point iao dosely also indicates possible instability in the presence of modeling error and ubeertanty, Therefore, both GM. and PM. should be given in the determination of relative stability, Thesetwevalues bound the behavior of the closed-loop. sytem nearthe resonant frequency. Since for most systerss there is uncertainty i both the magnitude and phase of the apen-loop. transfer function £ (5), asubstantal amout of GM, and RM. is required in the control design to ussure that the possible varia tions of L(+} will nav cause instability of the osed-loop rystern, For satisfactory performance, the phase margin should le be- ‘ween 30" andl 6i?, and the gain margin should be greater than 6 8, Foran underdamped second-order system, PM.and ¢ are related by aera = 22 ‘This gives a close connection between performance and robust- ness, and allows one to fully specify such a control system by ‘means of phasemargin and bandwidth alone, Note that for Brst- and second-order systems, the Bode phase plot newer crostes the 180" line, Thus, G.M. = ec. Ia same eases the gain aad phase ‘marpins are not helpful indicators of stabiity. Far example, a high-order system may buve large G.M. and PD, (both positive); hawever, sts Nyquist plat mnay till get close enough to the —L paint to incur a large My. Such a phenomenon ean any occur in high-order systems. Thereface, the designer shauld check any results obtained by using G.M, and PM. as indicators of slative stability. 163 Sensitivity to Parameters ‘During the desiga process the engineer may want tesa sider the extent to which changes in system parameters affect the lbchavior of a system. One of the main advantages of feedback is tha it cam be sed to make the response ofa system relatively independent of certain types af changes ar inaccuracies in the plant model, Ideclly, parameter changes due to heat, burid ity, age, or other causes should not appreciably affect 2 systems performance. The degree to which changes in system parameters affect systeen transfer functions. and hence performance is called sensitivity. The greater the sensitivity, the worse isthe effect af poraroeterchange Atypicalelosed-loop control systermay be modeled as shown Sn Figure 10.8, where Gjp{s) represents the plant or process to be > 1, C/R depends almost ‘entirely aa the feedback HY alone and is virtually independent ‘of the plant and other elements in the forward path and their [Bararbete Variations. Thisis because |1 +6 ,G p4I| % |G pC pil] and |G.Gp/GcGpH| = LIM}, Therefore, the sensitivity of ‘the cloned loop performance tothe elements inthe forward path, redices asthe loop gainincreases. Thisis amajorreasonéorusing ‘feedback. With open-loopeontral (ie. H = B), CPR = GpGe ‘Choice of Gc on the basis ofan approximate plant model or 3 model for which the parameters are subjected ta variations will, ‘cause errors in € proportional ta those in Gp. With feedback, ‘the effects due to approximations and parameter variations in Gp can be greatly reduced. Note that. unlike Gp the feedback -lemvent A is usualy under the cantrel of the designe. The sensitivity of the closed-loop transfer funetion Gor 1 ‘changesin theforward path transfer function, especially the plant ‘transfer function Gp, ean be-defined by B04 Gp 1 * Gp Ga "TFL We can plot Sj) at Funetion of to and such a pl shows haw sensitivity changes with the frequency of a sinusoidal input &. Obviously, or the sensitivity to be stall over a given frequency band, theloop gain £ over that band should be large. Generally, for good sesitiviy inthe forward path of the control system, the loop gue (by definition, dhe Loup gain i 2 (je)|) i made large over as wide a band of frequencies as possible ‘The extent to whic the plat mode is lawn willbe called uncertainty. Uncertainty is ancther important issue which de signers might need to foce, ft is known that some uncertainty 5 isa is aleays present, bath in the environment ofthe system and fn the system itell. We de not knowin advance exacty-whst disturbance and noise sgaals the stem wil be subjected to ‘in the system self, we know that na mathematical expressions san exaclly model a physica system, The uncertainty may be caused by parameter changes, neglected dynsmicx (especialy, high-frequency unmadeled dynaases), or ober unspecified ef ts, which might adversely affect the performance ofa control system, Figure 10.10 shoo a posible effec of high-frequency plant uncertainty due ta dynamics in the high-frequency range ‘hein neplected inthe nocainal plant. "Lome Fiparet®.10 Effet athigh-Arequeney uasnndeled plant uacertaity, In Figure 10.10, instability can result if.an wnknown high frequency resonance causes the magnitude to rise above 1, The likelihood of an unknown resonancein theplant G prising above 1cam be reduced if we cam kecp the laop gainsmall inthe highs Frequency range. In suramary, to reduce the secsitvity we need to increase the loop pain. But in general, increasing the loop gain degrades the stability margins, Hence, we usually have a trade-off between lowe sensitivity and adequate stability margins Disturbance Rejection and Noise Suppression All physical systems are subjected to some types of extra neous signals or noise during operation. External disturbances, such as a wind gust acting on an aitera, are quite commen in controled stems. Therefore, in the design of a control system consideration should be given so that the system is insensitive to noise and disturbance, The effect of feedback on noise and disturbance depends greatly on where these exancous signals fccurinthesyatem. But in many situations, feedback can reduce the effect of noise and disturbance ae system perfornsance. To explain these effects, lets consider a closed toop unity feedback, systemasshown in Figure 10.11, where a disturbance d(1) anda CONTROL SYSTEM FUNDAMENTALS sensor noise m(r} have been added tothe system. Figure 1011 A. onitySeedback control system shemving sources of poise, N(G, abd disturbance. (3). For simplicity, we assume that the effect of external distur- bance is collected and presented a the plant output. Thesensor noise, af}, is introduced into the system viasensacs. Both disture bances and sensor noise usually include randos high-frequency Signals. Let D(s), NG), R(GD, and ¥ (5) be, respectively, the Laplace transform of the disturbance d (1), sensor noise (),sys- tem inputr(¢)and epstem output p(e). Btis eaey wa Bind that, by superposition, the total output Fs) is Gels vps) Tr GIG __GehsdG ped 1+ GG 96) aand the tracking error e(1), dened a e(1) = r() ~ (1) with its corresponding Laplace transform F(s), becomes oy = Ris) is) L Troanee wee L Fa = Teanga Por GaAanGp(ed T+6.@G,0)" 1 THeG (8) In terme of the sensitivity function 5 and cloted-loop transfer Function Gy defined in Section 10..3,theoutput ¥(s)andirack- ing ereor £ (a) become Vis) = GeulsdRs) + S{eVDIsh ~ Gents NUS) Es) = SU) Rs) = SUs)D Ox) + Gls (S) Nate thatthe transfer function Gy is alse called the commple- mentary sensitivity function because $ and Ge are related by Sis) + G(s) = | forall frequencies. Wt inclear that Sts} must be kept sell to inimize the eflets of disturbances. From the definition of S, this can be achieved if the loop gain (he, LLCJooh = 1G) (ie) is Hage, [Gx Jeo must be hep small to reduce the effects of sensor noise on the system's out pl, and this ean be achieved ifthe loop gui issinall. For good. fracking, 5(s) must be smal, which implies thatthe loop qain should be large over the frequency band of theinpur sigaalr(r). 10.1, SPECIFICATION OF CONTROL SYSTEMS: ‘Tracking and disturbance ejection require sal S, wae soise suppression requires small G.. Prom the relation between S and Gcr, clearly, we canna reduce both functions simltane- coaly. However im practice, disturbances and commands are oftenlow frequency signals whereassedsor heee ae ofteaBigh- frequency signals ‘Therefore, we can still meet both objectives by heaping $ small in the low-frequency range and Gq small in the high-frequency range Pulling together the requirments of reducing the sensitivity vo parameters, disturbance rejection, and noise supprestion Wwe arriveata general desired shape forthe open-loop transfer func- Sion, which ia shown in Figure 1.12. 9,400 ‘The general eaturesofthenpen-loop transfer function arethat ve gain in the low-frequency region should be large enough, and do the high-frequency segion, the gain should be attenuated as smock as possible. The gin t intermediate requences typically ‘controls the gain and phase margins. Near the gain crossover frequency w, the slope of the lag-magnitude curve in. the Bode plot should be close to =20 dB fdfecacte (ie. the transition from the low: to high-frequency rage aust be smG0th). Note that the phase margin (P.M,) af the feedback system is 180" +o with $= £GclJckG pho), Te the loop sransfer Fonction Lf joa} = Gi jav) (feo) dastable, proper, andminimum phase, then # istniquely determined fromthe gain plot of GG p (ee IGctJooNGpC#u)|). ade actualty showed that i given, i terms of the weighted average attenuation rate of |G. Gp|, by C dn |G Fatge NG pL FeV) ® da Unceh wheres tn (/at)- Ana ie lare fp attenuates slowly ‘and small ie attenuates rapidly, Therefore, a rapid atenation, ‘of |G 6g] a or neas crossover frequency will decrease the PM. and 4 more than ~20-dB/deeade slope near ai indiates that PM, is inadequate, Cantroiing @ is important because itis 168 ‘elated to the poten stability and performance measure. Based on the loop gain shown in Figure 10,12, therefore, the desirable shapes forthe sensitivity and comnplerpentary sent lity Fonctinns of s.dosed-loop system should be similar to those shown in Figure 40:13. That is the sensitivity fonction Sanust Ue Y Snags Sexes Figure 10.13 Desirable shape forthe smstvty and complementary seni fonctioes. be snal a ow frequencies and all offen 1 fu, 0-4) at high frequencies, whereas Gp mt be 10a a low frequencies pe sal at high fequencies. orice tha Figures 10.12 and 10.13 can he viewed ax speci fications for a control system. Such graphical specifications of the loop shape” ae suitable for todays computer-aided design packages wich have extensive graphical capabilities. wil be en shorty there apeciications on the lop sbape are particu Latly easy to-adapt to MIMO systems. Tn onder to achieve the desired performance shown in Fig: ures 10.12 and 10.13, a “loop shaping” method, which presents 4 gropical technique foe designing «contre to ache oa bust performance, may be comsidered. The idea of loop shap- ing is © design the Bode magaitade plar ofthe loop transfer fonction L(jo) = Geto) Gp) to achieve (or at leat ap proximate) the requirements shown ia Figure 1.12, and then to back soive for the contoter from the loop transfer func- 166 tion, In ather words, we dst canvert perfrrsance speciBeca- sions om [5 (Ja) and (Gj (ja) fas gven in Figure 10:13) neo specifications on [E(jw} (a shown in Figure 10.12). We then shape [L(Jai) to make it He above the frst constaint curve (ULKjwH => 1) at Jw frequencies, le below the second com sraintcurve(](jeo]] << |} andsolloffasfast axpotiblein the high-fequency range, and make a smooth transition from lve tohigh fequency i, keep the slope as gentle as possible (obout —20d 8 /decade) near the crossover fequency ai. Inloopshap- ing, the resuling controler has to be checked inthe closed-loop sytem to sce whether a satisfactory teade-ff between [Sia] and |G. (je)] has been seached. Note thai is als possible ta slestly shape [SCJu)] andor Waal Control Effort Its important ta be aveare ofthe limits on actuator sg nals i specifying comtses, Most actuators have upper Kits ‘on their magnitusles and on their rates of change, These im- its can severely constrain the performance and robustness ofthe closed-laop system. Far example, ctuntor saturation can cause instability in conditionally stablesystems and very poor tansient response because of integral windup (see Chapter 20.1), “The problem should be addrested in two places. The actuator spesificatians shauldstae the allowable limits tee magnitudes and ratesofactuatorsignals. Proper specification of theactuatare ‘greatly facilitates the rest ofthe design. Secondly, the response ‘ofthe closed-loop systens to “large” inputs should be specified. Such specifications ensure that, if tis necessary to saturate the actuators, the performance and robustness of the dosed-haap system remain satisfactory. 10.1.4 Miscellaneous Specifications There are many office aspectsof a control system that are often specified. Thete are usually consteaints on the allowable cost of the contraller. In some applications, especially spacecraft, the size, weight, and power required far the controllers operation are restricted, ‘Contra system relist is alo often specified. The simplest such specification isthe life expectancy of the contalier. This x ‘usually piven as themcan time before failure (MTBE). The allow ble ways in which a conteller may fail ate also often specified, especillyin appbictionsinvolving human, For eumple,acon- ‘el system may be required tol sale” A stronger requirement i “il sot" ‘A good example of control system that is designed #0 “fil sot” i the controller that regulates the height of the water in fa toilet tank. Sooner or later the valve that stops the law of ‘water when the tank is fll gets stuck open. This would cauce the ‘nk to overflow, creating a mess, The MTBP isa few years, not ‘very long in comparison to the life of the whole system, This is acceptable because the control system inchides an overflow tube ‘hat causes the overflow to go into thetoilet and, from there, into CONTROL SYSTEM FUNDAMENTALS the sewer, This is a sof failure, The controller is not working properly: Water ie being wasted. But, the failure is not creating a serious problem, 10.15. Performance Specifications for MIMO LTT Systems Feom the discussion given for SISO ETI systems it is dea that she open-loop transfer function (jw) plays an essential role in determining varios performance and robustness properties ‘of the closed-loop system. For MIMO systems, the inputs snd ‘outputsaregencrallysnteacting. Dic such interactions, teat ‘be dificlt ta control a MIMO system. Hovrever, the clasial Bode gainphase plas san be generalized for MIMO oysters. {n the following. we describe performance specifications for MIMO syste The responses ofa MIMO system are generally coupled. That is every inpot affects mare chan one output, and every output is influenced by more than ane inptt. [Ta centoller can be found sock that every inpot affects one and only one output, then we say the MIMO system is decoupled. Exact decoupling can be dif ful, fact impossible, to achieve in practice. There are various ways 0 specify approximate decoupling. Because decoupling is smottobeious for quare transfer fonctions. let G{ ube asrilly proper Xm transfer function and et g(j) denote the jth clement of the matrix Gf), Then the requirement that aye <8 fora ow < 05; Rom ite would force G{ja) to be approximately decoupled provided 4 were small enough, Such # specification might be defective because it allows the diagonal transfer functions, the gi (jw), (ee 12.--+,m tobearbitrary. Typically, one wants [put J2). f= 1.2,1+, mclose toone,atleastin somerangey < e <7 ‘The requirement that Irs teelient ij Lae 3 for all = ay = a = wt om Aj forcesthe gi( eo) to bessmall relative to the gi (ja). OF nurse, nothing prevents adding SISO specifications on the gyj(Ja) to the decoupling specifications. Another useful decoupling specification is diagonal domi- nance. A strictly proper square m x m transfer function Gt) lssuid to be tow diagonal dorninant if Ete Gie)! < beaded) forall ie Laem andall @ = 0 < 00 Coluran diagonal dominance is defined in the obvious way. De- coupling specifications can alsa be written in the time damain as limitson the impulse or step responses ofthe closed-loop system. 10.1, SPECIFIGATION OF CONTROL SYSTEMS Decoupling is not alrays necessary or desirable. Thus, iti ocestarptohave other ways to specifythe performaneeof MIMO controlled sptems. One effective way to do this is by means of the singular value decomposition (SD). Ic is known that the SMD ic 3 weefal tool in Hear algebra, and it has found many spplicationsin control duringshe past decade. Let G bean m xr comple (constant) matrix. Then the postive squareroos ofthe sigervalues of G*G (where G* means the comples conjugate \ranspose of) ae called the singular values of. These square roots are aleays teal numbers because the eigenvalues of G7G ae always real and 2 0. The maximums aod minimum singular values of Gate denoted by &(G) and 2G), respectively and they can also be expressed by HG) = manyyiint Gull IG) = mitgyiat Gall where wé C¥ and the vector norm | isthe Euclidean noem. ‘That is, #(G) and e6G) are the maximum and minimum gains ofthe matrix: G, Far a square G, 2(G) isa measure of how far Gis from singularity, and &(CP}/2.(G} isthe condition number whieh isa measure ofthe difficulty of iwerting G. The best way to compute the SVD is by means of an algorithm alsa kav as the singular value decomposition (SVD (see Chapter 21). Far MIMO systems, the transfer function matrices evaluated ats = jas have proven wsefal in resnlving the complexities of MIMO design. The idea is to reduce the transfer function mate ces to twa critical gas versus Frequency, that i, the maximum snd minimums singular values of the transfer fonction matrix Consider a MIM Orsystem represented by a transfer matrix G(s Similac to the constant matrix ease discussed, if we Jet 3 = jae (0 Sw < 40), then the singular values 0 (Gan will befune tionsaf, and a plotof/(G (fu) is called a singular value plot (or @-plot) which is analogous to a Bode magzinude plot of 2 SISO transfer Function. The maxiroum and eninimum singular ‘ales of G (jas) are defined, respectively. as ulGjay" GU ey) inl FLjeOUH0)] 5GUju) ecw) where A[:| denotes eigenvalues. Notethat the Hog norm of asta ble trunsfer matrix G is the maxisnum of &(G) aver all frequen: [Ginx = supwx0 6(G (Ja). For the performanceand robustness measures of MIMO systeme, we can examine & and in. manner similar to that used t0 examine the frequency= response magnitude of « SISO transfer function. 1f (6) and 1G) are wery close to each other, then we can simply treat the system like a $1SO-system. In general, they arent close and 8 is important to bound the performance requirements, ‘Without loss of generality, considerthe MIMO unity feedback system shown in Figure 10.11. Like the SISO ease, define the sensitivity and complementary sensitivity matrices as 5(s) [E+ Gls)Gcten- (where ()-! means the matrix inverse) ist and G(s) = 1 ~ Sr), respectively: Nore hat Glo) = [E+ Gpls)G4s)]" "G pls) Gcks) is the closed-loop transfer Function matrix which detribes the ssn npi-output reltenship, From Figure 10.1 L, we have Fis) = Gu(s)ls) +515)DG) ~ Guts NG) Els) = SUR) = SnD(er + Guise Similar to the arguments stated for S180 systems, for dist Dance rejection, tracking error reductn, and insensitivity to plant parameter variations. we need to make Stjos) "sonal ‘over the frequency range (ay, 0 = at = a) where the commands, disturbances, and parameter changes of Gp are significant. ‘That is, to keep 6 [GF + GyG,)7"] a8 smal as possible, or ald + -GpGel as large a8 possible Yar = oo since S[U + GpG,)"!] = fad +G Ge). The inequabticn maz (0, 216 Gc) ~ 11 220 4 GpGel = 21GpGel + | for the imply thatthe loop ain 216 6. should be mages large seposnble¥ at = oy. ‘From the equations shown, i is also clear that for sensor noite eduction, the transfer matrix Gey(jar) should be “srl” cover the Fruensy range (say, 2 > sy) where the noise i wg nificant. Using the equalities (7 XX = oe Xt and G(X)g4X7") = 1 (assumme X~! exists) this impties that SUL £6,690 ,Ge] =4[l + 1G p61) falt+ $GpGa"| showld be small, or a(/ + (GpG 1") should be large Wo 2 ay. Since ol + (G)G.)"'] > 2(GpG.)""), we should make (Cp C,)™" large, whic furter means that |G pc should be hept as smll a possible ¥ at = wo la suimioary, disturbance ejection, tracking eto ab sen- sitivity (to plant parameter varistons) reduction require large 26 6; lmbilesensor noice reduction requirersmall (6,6, | Sinee 6 [Gp Ge] = gIGpCe], a conflict between the requite- mentsof Sa) abd Gru oll exists which is exactly the sanse as shat discussed for 5150 systems, Therefore, apertormance trade ‘ff for command tracking and discurbance red sor noise reduction is unavoidable, ‘Typical specifications for “MIMO system loop gain requirements willbe smile to those of Figure 10.12, withthe lawrequeney boundary (or constraint” lacing the lever Boundary for a{GpGel when ci small, and the “high-frequency boundary” representing the upper bound: ary for &[GipG.] when a becnmes large. I other words, we need “loop shaping” of the singular valves ofthe plant transfer function mates Gp by using Gx (ies the design of cootealer ) so that the nominal closed-loop sytem isstable, of pGe} (thus, #16 6, }} stow frequencies lier above the lne-freqbency boundary, and 1G pGrI (thus, pGe]) a high fequencies les below the high-frequency hotndary. That is, we want te crease tbe low-frequency valuecl (thus, 3 }af GG. to ensure adequate attenuation of law-fequency) disturbances and better command tracking, and roll-off and o in the high-frequency range to ensure robust stabiity: Note that the forbidden?” areas shaven in Figure 10,12 are problem dependent, and they may be constructed ftom the design specifications, 168 1 is generally desirable to require that the gap between F1GpGc] and o Gp Ge] te fairly small and that thele slope be ‘dose to =20 dil/decade near the gain crossover frequencies. Here, there are a range of gain crossaver frequencies from the frequency at which a (Gp(JaMGe(fa)| = 14a the frequency Sat which 3 [GpCJw}GeLiu]] = 1. As for SISO systems, the requirements near the crossover frequencies primarily address robustness, 10.1.6 Robustness Specifications for MIMO LTT Systems Niaknawn that.in coatrolsystem desig, the plant model used is nly an approximate representation ofthe physical system. The discrepancies berween a system and is mathematical represent. tian (evodel} may lead toa violation of some performance spec~ ication, oreven to closed-loop instability. We ay the system is robust ifthe design performs satisfactartyunder variatiansin the dynamics of the plant (including parameter variations and vat ‘ous possible uncertainties). Stability and performance robust~ nessaretwo important issuesthat should be consieredin conteal design. Generally, the form ofthe plant uncertainty can be para metric, nonparametric, or bth. Typical sources of woesttainy include unmodeled high-frequency dynamics, neglected nonlin cartes, plant parameter variations (due to changes of ervizan ‘mental factors). ete. The parametric uncertainty in the plant model can be expressed ar G(s, 7), a parameterized model of the nominal plant Gp(s) with the uncertain parameter y. The three mast commonly used roedels ta represent unstructured uncertainty are additive. input multiplicative, andoutpot multi plicative types: and can be represented, respectively a5 Gps) = Gyls) + ule) Gols) = Gplsdll + Aeta)| Ep) = UF A.6NIG RG) where Gps) isthe plant transfer matrix at pervurbed from its nominal model Gp(s} due to the uncertainty A (Ay oF ‘A; oF &,) with &[ACjw)] bounded above. Tha is, we have B|ACo] © Ma), where (wp is a koown positive real salar fonction. tn general, fo) ix small at lave frequencies because we can model the plant more accurately in the lave. frequency range. The plant high-frequency dynamics are les known, 1e- sulting in large (we) at high frequencies. Note that 4; (do) assumes all the oncertainty oecurred at the plane input (owt put). Both 1h, and Ay represent relative deviation while Ay represents absolute deviation frors Gpls}. Of course, there are many possible ways to represent uncertainty, for example, Gps) = [4 do lNG As E + Ads], Gels = (MU) oe(sMILD(3) + Appts)|~! (a matrix Fractional representation with Gps) = NU)B™Madieete, We may even model the plant by combining both parametric and unstructured uncertsinties inthe form of G p(s) = [d+ Guts) IGp(s, YE + Bale. Inthe folowing, we examine the robustness of performance together with robust stability under output meltiplicative one CONTROL SYSTEM FUNDAMENTALS certainty dy. Similar resolts can be derived if other types of ‘unstructured uncertainty are wed. Vix a multivariable version of the standard Nyquist criterion, the closed-Foop system will remain stable under the uncertainty Ay if 1 GOK +G,G.0 jal] < ae Vad 1GpOU + GyGI Nad] < ey one ‘where we assume thatthe naminal closed-loop system is stable, andG As, Gp(s) have chesamee numberof unstable poles. The abowe condicion is actually necestary and sufficient for robust stability. Note that the expression inside [+] in the lefichand Side of the inequality isthe complementary sensitivity matrix ‘G.a (den), Thos, the model uncertainty imposes an upper bound fn the singular valoes of Gy (+) for sobust stability. Rewriting the inequality, we get 1 FG EMT + GG)" Ged] al + (GpG."'e)] Varo d(He) a where we assume that GpG, is invertible. Obvioushy, [+ (GpGe)"") can be used 25 a measure of the degree of sability ‘of the feedback system, This is multivariable version af SISO stability margins (i gain-and phase margins} because t allows tain and phase changes in each individual outputchannel and/or Simoktaneaus gain and phase changes in several channel. The extent ta which these changes are allowed is determined by the {inequality shown above. Therefore, we can use the singular values twdefine GM. for MIMO systems. in the high-frequency range (where the loop gain is small), from theiinequality given above, wehave BIG on cUial < ss ‘This is a constraint on the loop gain CpG, for robust stability. ‘This constraint impliesthat at high frequencies the loop gain.on 6 pGe (04 1GpGe}) should le below a certain limit for robust srability (the mame argument decribed in the previous section}. ‘Obviously. satisfaction of the high-frequency boundary shown lin Figure 10.12is mandatory for robust stability (mot just desired for sensor noise reduction For robust performance (ie. gaod carmand tracking, dis srurbance redaction, and insensitivity o plant parameter wari ations (je. structured uncertainty) under all posible Gpy(s)}, ‘we sbould keep 3 [61 + GipG.3] of the “perturbed?” sensitiv. ‘ty matrix S(s) as small as possible in the low-frequency range. Following the sameargument given in the previons section, this farther implies that 216 Gc should be large at low Frequencies Making aGp Gc! large enough ensures that 1G pG,] is large. ‘Cleary, high loop gain can compensate for adel uncertainty. “Therefore, for robust performance the low-frequency boundary given in Figure 10.12 should be satisfied, alsbough this is not ‘mandatory. 102. DESIGN USING PERFORMANCE INDIGES For MIMO systems, as forthe requirementson theloop gainin SISO control design, any violation ofthe low-frequency bound- ary in Figure 10.12 constitutes « violation of the robust perfar- rmanee specifications, while aviolation ofthe upperboundary (in the high-frequency range) leads toa violation of the robust sta bility specications. The maia distinction between MIMO and SISO design isthe use of singular values of the transfer function tmatrix to express the “size” of functions. 10.17 Conclusions Writing specifications for control systems imo eaty. "The dilfer- ent aspects of controller performance ane robustness areinter- related and, in many cases, competing, While a good controller can compensate for some deficiencies in the plant (systems to be contralled), the plant implies significant fimitzan contra per- formance. Iisa to exsy to impose unachievable specifications on a contralle, For these reasoas it important to have the plint designer, the perion responsible for the controller specications, and the contol designer work together from the beginning tothe end of any project with demanding control eequtements Further Reading Most undergraduate texs ontcontrolsystemsconaia wsefal descriptions of performance and robustness specifications for SISO IT systems. Theee examples are (1) Franklin. G.E, Powell, LD, and Emami-Naciniy A, Feedback Control of Dynamic Systems, 3rd Edition, ‘Addison-Wesley, Reading. MA, 1954. (2] Nise, N-&, Controd Systems Engineering, 2nd Edition, Benjarnin/Cumsmings, Redwood City, CA, 1995, [3] Dorf, B.C. and Bishop, RLEL, Mfadern ConeralSyztemes, 7th Edition, Addison-Wesley, Reading, MA, 1995, Several textbooks have been published recently that include a more loop shaping-oriented discussion of eontral speci- fications. These include 14] Boyd, SB and Barratt, CH, Linear Corll De sign -Limisof Peformance, Pentie Hill, Englewood lis, NH 1991 [5] Doyle, 1G, Francis, BLA. and Tannenbaum, A... ‘Feedback Control Theory, Macmillan, New York, NY, 992. [6] Belanger, PRL Control Bngnecrig « A Modern Ap proach, Saunders Cllege Publishing New York, 1995, Wolovich, WA. Automarie Cantal Syutems ~ Basic Analysis and Design, Saunders College Publishing, ‘New York, 1984 i Specifications for MIMO ET] systems are cavered in [8] Maciejowaki, 1M, Multirinble Feeback Design, ‘Adison-Wedey, Reading, MA, 1959, 19 19] Green, M. and Limebeer, DLN. Linear Robesr Con- tol, Prentice Hall Englewood Cif}, 135. Theliiations on cootol sss ndoding Bode’ ess on the relation between gain and phase, are described in (30] Sreudenberg, S. and Loon, BLP, Fraaency Boman Proper of calor ant Maiariatte Feedback Sys ten SpingstVeriag, Nee Yor, NY. 198, ‘Agood, detailed example ofthe specifications fora camplex MIMO conta syer i (11) Hoh, RB, Mitchel, DG. and Agoaso, Buby Ran dling Qualities Requirements for Military Rotorcraft Aeronautical Design Sundard, ADS-35C, US Army SC. ug. 1303 (12) Hoh, REL, Mitchel DG. Apanso, BL, Key DL tnd Hlnkes, Gy Background Iaformaon and User's Guide for Handling Quantities Requirements for biliary Rotorcraft, USAAVSCORCTRIS- A, US ‘Army ASC Dee. 1988 Finally, al tne publication ofthe IEEE Conta Stems Societe Transactions on Antomatic Control the Tans actions on Cantrel Stes Technology, andthe Conte Syoterns Maprine regulary contin ates on the design of control spams. 10.2 Design Using Performance Indices Richard C. Dorf, University of California, Devis Robert H. Bishop, The University of Texas at Austin 10.2.1 Introduction Modern contral theary axumer thatthe systems engineer can specify quanticatively the required system performance. Then 4 performance index can be caleuleed or measured and ised ro evalute the sptem’s performance. A quantitative measure of the performance ofa stem necessary for automatic parameter ‘optimization of control system and forte desig of optim stern ‘Weconsiders feedback spstem as shonin in Figure 10.14 where the closed-loop transfer Function is COSY gy x elt) 10. GeAsnGiy von ‘Whether thet is toimprove the design of system or to design 2 control system, a performance index may be chosen [2) ‘perfomance index i a quataive meuure the ‘fa ntemand is chosen sathat emphasis vento the lexpartant sytem pectic vo “The spatem is considered an eptinnam contra system when the system parameters are adjusted so thatthe index reaches an ex- ‘tremum value, commonly & iminium value. A performance index, to be wseful, must bea number that i always positive oF aero, Then the best system is defines asthe system tha ex. It i alo possible to design an optimum system to achieve a deadbeat response, which is characterized by a fast response with sinimal overshoot, The desired clased-loop system character- istic equation emeficients are elected to minim setling ime and ese time. 10.2.2 The ISE Index (One performance index isthe integral af the equate of the error (SE), vehich is defined as 7 se= [eed (102 ‘The upper limit 7 is a finite time chosen somewhat arbitrarily s0 that theintegral approaches steady-state value, 11 usualy convenient ochoose? asthe eting ie, T. This criterion wl clsrininate betwee cansivhyovercped ad nce te ‘erdarmpe! ystems, The minimum vale ofthe integral occurs for a compromise valueaf the damping, The performance index of Bqeation 102 ealy adapted for practical measurements becouse aring cra is readily obtained. Furtbermorethe sqened err ie thematically convener foc analtsl ad comptstional prpores. 10.2.3 The ITAE Index ‘To reduce the contribution of the relatively Large initial exror 10 the value of the performance integral, as well as to emphasize ertors occurring liter in the response, the following index has, been proposed [2]: irae = sewer, (103) vrhere FTAE is the integral of tise multiplied by the absolute magnitude of the err ‘The enclicients that will minimize the ETAE performance eri terion for a step input have ben determine! for the general closed-loop transer Funston (2) cia bo RG) Sot ae a” (oy “This transfer function has a steady-state err equal to zero for 42 step input, Note that the transfer funstion has m poles and no: reros. The optirnum caeFcients far the ITAE criterion are ver in Table 10.1 fora step inpot. The teansferFunctioo, Equa ‘Gon 10.4, implies the plant and controller G.(s}G(s) have ane ‘more pre integrations providezero steady-stateerror: The "espanieswsingaptimum coelivientsfora step inpurare given Figure 10.18 for TAE, The responses are provided For normalized CONTROL SYSTEM FUNDAMENTALS time, ait. Other standard forme based an different performance indices are available and can be useful in sisting the designer to determine the range of coefficients for a specific problers. afte fe y Gets) Gis) an Feedhack control stem. Figure 10.14 Figure 10.15 The step response ofa satem witha transfer function satefing the ITAL criteioa, Fora ramp input, the coefficiemts have been determined that ‘minianize the FTAL criterion forthe general closed-loop transfer fanction (2) 05) Fb inh iy “This transfer function has a steady-state error equal ta zero fora ramp input. The optimum coefficients for this transfer function are piven in Table 10.2. ‘The transfer function, Equation 10.5, implies that theplant and comtraller Gy (£}G(«) have twaarmore Pure integrations, as required 1 provide zero steady-state error, 10.2.4 Normalized Time ‘We consider the transfer function of a closed-loop system, T(s “To determine the coeflicients that yield the optimal deadbeat “response, the transfer function & fist normalized. An exarmple ‘ofthis for third-order systera is a Ti) = 1" Traut + fale tal (10.8) 102. DESIGN USING PERFORMANCE INDICES ‘TABLE 101 Ww ‘The Optima GaetBcicets of T(3) Rased onthe TTAB Cerin fo Sep np. re 8+ Laos + ait ho LPB an? 42, Kags + of S442 tans? + Stade + 2.7uks + af 254 Lika! Sande! 45 Sei? + 3a +o P+ 5DSans + Bods +B bobs + 7a? + 3.95 + of TABLETO2 The Optiewns Goethe T(s) Based the [TAB Criterion fra Ramp Input, + 32ege tom 2 iS 4 baSab + of 4+ Daluye® + 439uis? + Stoke + al 35-4 2.19qs4 + 6Sude? 46.hade? + Sc2daits ul Dividing the numerator and denominator by a yields (ior m1.) Phar 4 fel (ea Equation 10. isthe normalized third-order closed-loop transfer function. For a higher-order system, the same method is used to derive the normalized equation. When we let? = s/aiy, this has the effectin the time-domain of normalizing.time, oi. Thestep response fora normalied system is shown i Figure 10.15, 10.2.5 Deadbeat Response ‘Oken the goal for control eptem is to achieve a fast response to a step command with minimal overshoot, We define a deadbeat response as a response that proceeds rapidly to the desired level ‘and olds at that level with enisimal overshoot, We use the 2% bandat the desired level a the acceptable range of variation from the desired response; Then, ifthe response entersite band at time T,,ithas satisfied the setling time T, upoo entry to theband, as iMstratedin Figure 10.16. A deubeat respore has the following characteristics: 1, Zero steady-state error 12 Fast response —+ misimum rise time and sctling ime 3, 0.198 < percent overshoot < 2% “4, Peecent undeeshoot < 29% Characteristice 3 and 4 require that the esponse remain within the 2% band so that the entry to thebsend accuirsat the setting time, ‘A more general normalized transfer function of closed-loop system may be written a¢ t TEEPE Tw=5 Fea tos) Figure10.16 Thedeadheatnesponse, where A isthe magnitude of he ep input The coefficients ofthe denorinator equation (er, ,and soon} are then assigned the values necessary toanee! the requirement of advent response, The coefficients reconded in Table 10.3-were selected to achieve deadbeat response and to minimize seting tiie and rise time to LOWS of the desired command. The form of Equation 10,9 ismormalized since 3 = s/oy. Thus, we choose ay basen the desired settling time of tise tine. For exaenple, it we fave a third-order system with a required seting time of 1.2 secoads, wemote from Table 10-3 that the normalized setting times eon = 408 ‘Therefore, we require Once wy ischoses, the completethird-order hosed Joop teanser functog is known, having the formaf Equation 106,wWherear 1.9 and fw 2.2. When designing a system o obtain deadbeat 1, CONTROL SYSTEM FUNDAMENTALS ‘TABLE 10.3_ Coefficients und Response Measures of a Deadbeat Sjnecn, Coeficeets ay te and use sd 98220 ah 220 380 am Sh 270 490 Sa 3 Percent Percent 905% 100% Owes ‘Under= Rise Rise Setting shoot shoot = Time Time Time POM Tne Te oes 2a7 sb ua Lest basa ame onm ous kte 527 ea ee ee es response; the eampensators chosen and the closed-loop transfer function isfound. Thiscampensated transfer function isthen set ‘equal to Equation 10.9'and the required compensator parameters EXAMPLE 10.1 ‘Consider a sample plant Guys Oe (1.10) anda cantroltes Ge(s) = . The goalisto select the parameters p and K’ to yield (1) an TTAB response, and alternatively (3) 2 deadbeat response. [n addition, the settling time for a step re- sponse isspecified asless than I second. The closed-loop trasster funetion is K Tw) =~ Pyptk (eu For the ITAE systemn-we examine Figure 10.15 and note that for 2we have ain Ty = 8. Then, for fy = 0.8, We Use wy = 10, From Table 10.1, we require Pb Lone ta Pt pet KS ‘Therefore, wehave K =a = 100and p= Lang = bt ‘Mf vwe seekea deadbeat response, we use Table 10.3 and note that the normalized seing time is nT = 482 In onder wo obtain T, less than | second, we use ay = 6 and T, = O8. Thea we ose Table 10.3 to obtain a = 182. The closed-loop transfer functian is 36 = Fy +36" a TwH= a0) FF aims Pap ‘Then we require X = 36and p = 10.9, The dendbeat and the "TAR step response are shown io Figure 10.17, Figure10.17 ITAE(p = [and tnd K = 36) oymtem respon, 100} anddcadbestlp = 10.9 10.2.6 Conclusions ‘The design of a feedback system using performance indices or deadbeat control leads to predicuble system responses. This per ‘mits the designer to select system parameters to achieve desired performance. 10.2.7 Defining Terms Performance index A.quantitative measure ofthe perfor- mance of a system, SE: Imegral of the square of the error, FTAE: Iniegral of time multiplied by the absolute error. Deadbeat response: A system with rapid response, min- imalavershoot, and zero steady-state error for astep input. References [1] Bor, RiC, and Bishop, RH, Modern Control Systems, Tied, Addison-Wesley, Reading, MA, 1995, 10.3. NYQUIST, BODE, AND NICHOLS PLOTS [2] Graham, D. and Lathrop, B.C, The synthesis af pti- mum response: citeriaand standard forms I, Trans, ALEE 72, 273-288, November 1953. 10.3. Nyquist, Bode, and Nichols Plots John J. D'Azzo, Air Force institute of Technology Constantine H. Houpis, air Force Institute of Technology 10.3.1 Introduction ‘The frequency-respanse! method of analysis and design isa pow= cfu technique forthe comprehensive study ofa system by an ventional methods, Performance requirements can be readily cexpresed in termsof the frequency response, Since nose, which isabwaps present in any syteracan reset in poor overall perfor- mane, the frequency-tesponse characteristics of system permit ‘evaluation of the effect of muise. The design of a passband for the system response may result in exctoing the noe and theres fore impeeving the yptem performance a long asthe dynamic (racking performance specifications are met. The frequency r= spond lio ef in sitions or whi the transfer factions of some orall of the components in a system are unknown. The frequency response can be determined experimentally for these sivuatlons, and aa approximate expresion for the transfer fabe= tion ean be obtained fromthe graphical plot ofthe experimental dita, ‘The frequency-tespanse mthod i a a very powerfel compat (MIMO) system with structured uncertain plant param tera. In this chapter two graphical representations of tranafer fonctions ate presented the logarithmic plot and the polar plot, “These plots are used to develop Nyt’ stably 1/T this function isa straight line with a negative slope of Gf B/actave, Therefore, the asymptates of the plot cf Lm(l + JuT)~! are two straight lines, one of zere slope below w LT and ane of ~6d8/oerave slope above w = 1/T. These asyenptotes are drawn in Figure 10.19. ‘The frequency at which the wsymprotes ta the Lm curve intersect is defined asthe corner fequertyaiey. The value weg = WT 8 the corner frequency forthe function (bop foT yA! me (1+ juofeany *P ‘The exact values of Lm (+ ja )°! are given in Table 10.5 foe several foe quencies in the range a decade above ane below the comer frequency. The exact curve is alsodeawn in Figure 10.19 The error, in dl, between the exact curve and the asymptote is approximately as follows: 4 Avthe comer frequency: 3 4 # Ge octave above and below the comer frequency: 1B 1 Two octaves fram the earner frequency: 0.26 dB CONTROL SYSTEM FUNDAMENTALS TABLEIOS: Values ofLm(1 + ja)" for Sever Biactvalue dB Value ote symptor, dt v cy n 026 0 097 0 2.00 0 301 235 2.00 602 097 1204 026 209 004 Preliuinary design studies are often male by using the asyrap= totes anly. The correction to the straight-line approximation te Yield the true Lin curve is showa in Figure 10.20, ‘The phase ‘ture for this function is also plotted in Figure 10.19. At ara frequency the angle is 0"; atthe comer frequency ws way the dangle is 43°; and at infinite frequency the angle is—20", The an- she curve is symmetrical about the comer frequenry valve wher plotted against log (fax) ae fog at. Since the abscissa of the curves in Figure 10.19 is /w. the shapes of the angle and Lm curees are independent of the time constant T. Thus, when the curves are pled with the abscissa in terms of or, changing T ust sides” the Lo and the angle curves left. or right sa that the —3 dl and the —45° points oecur a the frequency ay = aie. ‘The approximation of the phase curve is a straight line drawn through the following three points: sofey OL “angle a ‘The maximum error resulting érom this approximation is about -£46°. For greater accuracy, a smwoth curve is drawn thraugh the points given in Table 10.6. TABLEIOS Angier of + fosfatey)* sor Key Frequency Pins Anal de ‘ar 57 05 258 Lo 450 20 4h 100-843 ‘The factor 1+ oT appearing in the mumerator hase Lan Lan(h + ju) = 20top- Via “This isthe same funetion asks inverse Lm (1 + jasP)“! except that itis postive The commer fequency isthe same, and the angle varies from 0 0 90" asthe frequency increases from zero to infinity, Te Land angle curves for the function (1+ jos) are syrametrical about the absciss tothe curves for (1+ joa)” 10.3 NYQUIST, BODE, AND NICHOLS PLOTS ur va Figure 1020 arg ugnitude conection fort + JorT)* Quadratic Factors Quadati fictors in the denominator of the transer function have the fren 7 i [t+ Biot Aer] for § > 2, the quadenic canbe factored ito two fore factrw thee zeros which canbe pote inthe anne sh previo, Bu for ~ | Equation 107 sarin saniugse ‘complex factors, and the entire quadratic is plotted without fas toring: 037} tn [t+ £jo+ Suan] | = -awel (2) +)']" 2[t+ Siw Suwr] = -onet Bay 1039) From Equation 10.38 itis seen that for very small values of ur, the low-frequency asymptote is represented by Lm= 0 dB. For ‘ery high vanes of frequency, the high-frequency esypote has a slope of — 80d B decade. The asymptotes cross at the cornet frequency yyw tin. 1038) From Equation 10.38 itis seenthat a resonant condition exists in thevicisityofar = wy, wheretbe peak ilueofthe Li > OB, ‘Therefore there may be a substantial deviation ofthe Lm carve fromm the straight line asyrnptotes, depending.an the value off ‘A family of Lin curves of several values af € < 1 is plotted in Figure 1021. For the appropriate the Lm cuewecan be deawn bbyselectingsulficent point rom Figure 1,23 or computed frm Equation 10.58, “The phase-angle curve for this functionals varies with €. At sero frequency the angle sO; atthe corner frequency the angle is 90°; and at infinite frequency the angle is ~180". A farily ‘of phase-angle curves for vatious values of ¢ = is plated in ‘Figure 10.21, Enough valoestodraw he appropriate phase-angle ‘curve ca be taken from Figure 1.21 or compisted from qu ‘ion 10.39, When the quadratic factor appears inthe numezatoe, ‘he magnitudes ofthe Lm and phase angle ae thesameas those in Figuce 10.21 except that they are change in sig. The Lin[} + 26 foxy + enjoy }?| 4 with ¢ < 0.707 has a peak value. The magnitide ofthis peak value andthe frequency arwhich it occursare importane terms, These values, derived in 78 Section 10.3.8 [vee 11, Sect. 9.3], are repeated here: (10.40) ave) tm = wy = 2 (10.41) [Note that the peak value Mp depends only om the damping rata £, Since Equation 10.41 is meaningful-onty for real values of ae, the curve of Mf vs,» has-a peak value geeater than unity only for < 0.707. The frequency at which the peak vale occurs depends on both the damping ratio { and the undamped natural frequency wy. This information fs used when adjusting a control system for good response characteristics, These characteristics are diccusced in Sections 10.3.8 and 10.3.9. CONTROL SYSTEM FUNDAMENTALS 10.3.3. System Type and Gain as Related to Lm Curves The steady-state ervor of a closed-loop system depends on the systeot type and the gain. The system error coeflicients are de- termined by these two characteritis (11, Chap, 6]. Far any given Lm curve, the system type and gain con be determined. Also, withthe transfer function givenso thatthe system type and ign are known, they can expedite drawing the Lm curve. This described for Type 0, |, and 2 systems. ‘Type O System Afirstorder Type 0 system has a transfer Function of the form , a We ate At law frequencies, p< 1/T.a, LmG@( ju) © 200g Ko, which isa constant. The slope of the Lan curve is zeve below the cot- ner frequency wy = 1/7 and ~20 dBidecade abowe thecorner frequency. The Lin curve is shawn in Figure 10.22 or a Typed system the characteristics ace as follows: 1, Theslapeat low frequencies is zero. 2 The magnitude at low frequencies is 20log Kp 5, Thegain dp ithe seedy state step error coeffi os Figure 1921 Logmagninude ane phase diagram for {1 + f2ewfary + Yeo/nr)*] “The Lm curves for poles and seraslying in the right-half(RE) 4—plane are the same as those for poles and zeros located inthe leit-half (LH) s—plane. However, the angle carves are different, For example, the angle forthe factor (I — jooT) varies from 0 10. 31" as cr varies from aero to infinity. Also if ¢ is negative the ‘quadratiefactoe of Equation 1037-contains RH s—planepoles oF eras, Its angle varies fram —360" atu = Ota —180° at ‘This information can be abtained from the pole-era diagram [Lt Sect 4.12] with ll angles measured in a coantersckwise (CCW) direction. Some CAD packages do not consistently use ‘OCW measurement ditection, thus resulting in inaceurateangle values Figure 1022 Log magnitude pls for Goo) = Ky/(L + feoTy) ‘Type 1 System ‘Asecond-onder Type I system has «transfer function of the form, hy foil + jet) ‘low frequencies, © < 1/Ty, Le[GCia)] ® Li( Kian [in = Lm jal, which has slope of =20 dBidceade. Ata Ki, LIK Uf} = 0, I the corner frequency 69) = Ty i (greater than A), the low-frequency partion of the curve of slope (= 10 dildecade crosses the 0=dll axis at a value of w, = Ka, 2s shows in Figure 10.23. Ifthe commer frequency it les than H, the low frequen portion ofthe curve of slope ~20 decade say be extended until it dogs crass the dB axis, The value cof the frequency at which the extension crosses the 0—dB axis is tog = Ki}. In other words, the plot Lin (Kidjo) crosses the DadB valoe at oy = Ki, axilhstrated in Figure 10.236. ‘Ate = 1, Lim jas = 0: therefore, Lm (KiJjaua) = 20 fog). For Ty -! 1 this value isa point on the slope of ~20 ABfdecade, For T, > 1 this value is @ point om the extension of Gijon 103, NYQUIST, BODE, AND NICHOLS PLOTS 179 igre 1023. Log magnitude plot for Glja) = Ky {joott + jor) the initial slope, as shown in Figure 10.236, The fiequency om it smaller larger than unity according as Ky is smaller oF than unity. Fara Type I system the characteristics are as fellows: 1. The slape at low frequencies is =20 dBdecade. 2. The intercept af the low-fequency slope of —20 4B decade (or its extension) with the 0—dB axis. oc- jours atthe feequency aty, where oly = Ki 3. Thevalue ofthe low-frequencyslope of ~20dBidecade {or its extension) at the frequency w = 1is equal vo 28 lag}. 4. The pain Ay isthe steady-state ramp error coeffi- sient, ‘Type 2 System A Type 2 system has a transfer function ofthe form ky Gia) = == Gay jel Aslow frequencies << 1/T, Lam|[G(je)} =Lm [p/(e)?] = Lal al- Lala forwich the Shope is Ad B/decade, at ‘Ky, Len [FEa/(Joo}*] = 0: therefore the intercept ofthe initial slope of —40d 8 decade (or its extension, if necessiry] with thed—dB axicaccure atthe frequency, where a = Ky At or = 1, Lin Ljool? = 0; thesefore, Lin [42/10 loot 20 log| 2]. This point eecurs on the initial slape or on its exten sion, acearding aria) = 1/7, i larger or smaller than y‘E. If Ky > I, the quantity 20og| Kis positive, and if < 1, the quantity 20log] Ka] is negative. “The Ln curve for a Type 2 teansfer function is shown in Figs ture 10.24. The determination of gain Ky from the graphis shown. For a Type 2 system the characteristic are 28 follower: 1. The slope at low Frequencies is = 4 dBi decade 2. The intercept of the low-frequency slope of ~40 2) and tangent to ane-of the major anes at « = 20. The final angle is ~90° timesthe order n ofthe denominator minus the aeder w of the numerator of Gio). ‘Type 1 Feedback Control System Atypical Type 1 system comtaining only poles és GG) = 705 = areormtermeyery (048) sok 50 as OF Note BU | ay sep" as cae 00 ) sana) Note that the ju term in the denominator contributes the angle orate langle of G (jo) for al quences ‘Thun, the ‘hse diterence between Equations 104 nnd 0.48 iathe pee sce of the rm J nthe denominator ofthe later equation Since all the (1 + foeT) terms of Equation 10.48 appear in the dbenominatoy the ngleothe polar lof {ja decenescon timuil (CM) inte ne diction from 30 to 3 aso increases rom Ot 8, TRetequeny of thecoulng pol on he ‘eegative real axis of tive Goo) function is that value of frequency sh fr which the apna part of G (ju) sequal or. The reba crouing pat sey impartant becase it determine! Closed Lop sub, at deserbed in tr sections ing with yen hy, ‘Type 2 Feedback Control System ‘The transfer function af a Type 2 system it cum a Se) = Bist = GaPTTetrTet 1050) Inspropertes are ayes ( SEE ‘Tie presence of the (jw)* term in the denominator contributes =140" to the total angle of (jw) forall frequencies, The polar Plot forthetransfer function of Equation 1040's a smooth sirre ‘whose angle lo} decreases continuously from =180 to =360" ‘The introduction of an addtional pole and a sera can alter the ore, Most a w+ toe 10.3. NYQUIST, BODE, AND NICHOLS PLOTS shape of the polar plot. It cam be shown that as w —+ O*, the polar plot of Type 2 system is below the real aris if Taumeraror — E Titemominatar {sa postive value, and above the real axi fit is « negative value. (1052 ‘Summary: Direet Polar Plots ‘To obtain thedirect polarplotofa ystems forward transfer function, the following steps are used to determine the key parts ofthe curve. Figure 10.25 shows the typical polar plot shapes for dlfferent system types. a0 Tape 3 Type | Fw 085 Asan ec ppl oe pe = stp Te wd er con athe ne a Cou) = SRST ESE ns For is acon the spre ype be ae tran derma th etch par or reg limg+o[C(fa)). The low-trequency polar-plot characteristics {ase m+ 0 of the different system types are determined by the angle at w = 0, ic, £G(/0) = m(=90?). Step 2. The high-frequeney end of the polar plat can be detes- ‘mined as follow: Timg-+ 45 [4 fo)] = Dé|(wo —m —u)90°} (10.54) Nove that since the eegree of the denominator of Equation 10.33 is always greater than the degree of the murmerator, the angular condition of the high-frequency peint (e = 60) is approached in the CW sense, The plot ends at the origin andl is tangent t0 the axis determined by Equation 10,54, Tangency may occur on either side ofthe axis Step 3. The asympiate that the low frequency end approaches, for a Type | system, is determined by taking he limit as 9 + of the teal part ofthe transfer function, ‘Step 4. The frequencies at the points of intersection of the po. lar plot with the negative real axis and the imaginary axis are 181 determined, respective by ImiGijan) =0 (1058) Re(Gijor] =0 (1058) Step 5. Ifthere are no frequency-dependent terme in the numes- ator ofthe transfer fenction, the curve isa smooth one in which the angle of Gta) continuously decresses as w goes from 0 10 co, With tle constants in the nusserator, and depending upon their values, the angle may not change continuously in the same direction, thus creating “dents” in the polar plot. Step 6. As is seen later it 8 important to know the exact shape of the polar plot of Gu in the vicinity af the —1 se #9 point and the crossing point on the negative real axis, 10.3.6 Nyquist’s Stability Criterion. The Nyquist stability cterion [2], [6], [8] provides a simple graphical pmceduce for determining cosed-lnop stability Fron the equeney-cesponse curves the open-loop tranafr Function Goo (Ja) forthe case a0 poles or zeros onthe imaginary ais etc}. The application of this method in terms ofthe polar plot is covered inthis section; application in terms of the log rmagnitude-angle (Nichols) diagram is covered in a ater section, For a sable closed-loop system, the roats of the eharacteristic squation Bis) = 1+ GUNG ° (u.s7) cannot be permitted to Lie in the RH x~plane oran the jus axis In terms of G@ = Ni/D) and H = Nz/Dz, Equation 10.37 becomes NiNa _ yD. +NiNa 56) = 145 oe = )e-B)--- 6B) = cans) = ple = pb = Pa) Notethat the numerator and denominator of B+) have thesame degree and the poles af the apensoop transfer function G (rH (s) fre the potes of B(3). The closed-loop transfer function of the system it Seah = reais = mae ‘Thedenominator of C(s)/.8(s) isthe same as the mumerator of {Bs}, The condition for sabiity may therefore be rested a Fora stable system none ofthe zeros of Bs) ca lei the REL {plane of of the imaginary axis, Nya’ sablty eiterion relates the number of zeros and pales of (2) that lie im the RH sr~plane tothe polar plot of G(s) 18) (19.39) Limitations In this analysis it i assumed that all the control systems. are inberently linear or that dhetr Limits of operation are ca fined to give linear operation. This yields a set of linear dif ferential equations that describe the dynamic performance of 182 the systems. Because of the physical nature of feedback ean. rol systems, the degree of the denominator ByD; is equal to ‘or greater than the degree ofthe numerator Ny N of the open- Joop transfer function G(s)# (s). Mathematically, this eveans (that limsco[Gts)H(2)) + 0 or a constant. These treo fae: ‘ors satisfy the necessary limitations to the generalized Nyquist stability criterion, (Generalized Nyquist's Stability Criterion Consider a clased contour @ such that the whole RH ss—plane is encicled (see Figure 10.26a with ¢ 0 ) tha en closing all zeros and poles of B(s) that have positive real parts The theory of complex variables used inthe rigorous derivation, requires that the contour @ must not pass ehrourte any poles er zeros of Bi). When these results are applied to the contour ©, the follawing properties are nated: ” “ Figure 10.36 (a) The contour (which ences te rg bal pe compete ptf GUE) © seep Strcao 1. Thetotalmumber of CW rotations of BOs} dive to is eras ixequalto the total numberof zeros Ze in the RH s~plane, 2, The total oumnber of COW rotations of B(e} due to its poles is equal tothe total number af pobes Pe in the RHs~plane. 3, The net number of rotations A of Mt) = b+ G (1H (4) aboutthe origins equal ve the Lota mun berof poles Py minus the total numberof aro Z inthe RE # plane. maybe positive (CCW), neg- ative (CW), or zero. ‘Theessence of these thiee conclusions can be represented by have change of 14010410) py — 9 (10.60) where CCW rotation is defined as being positive and CW rate- tiom is negative. In onder for B(s) to realize a net rotation N. the directed line segment representing A(z) (see Figure 10.273) ust rotate abput the origin 360§", or N complete revolutions, Solving for Zp in Equation 10.64 yields 2g = Pp — N. Since CONTROL SYSTEM FUNDAMENTALS ‘B(s}canhave no zeros Zin the RH s—plane for astable system, iz ietherefore concluded that, ora stable prem, the net muonber sf roations af B(s)aboat the origie mast be CCW and equal tv tice naraberef poles Py: that ie in the REY plane. Ln ether ward ‘B(svesperiences anet CW rotation, thisindicates that Ze = P. ‘where Px, > 0, and thus the closed-loop system is unstable, If there are zero net rotations, then 2 = Pp and the system may sor may aot be sable, acondingas Py = Dor Py = 0. ‘igure1927 A change of reference for Bs), Obtaining a Plot of 24) Figures 10:27and 10.276 show aplot of ts) anda plotof G(a) MGs), respectively. By moving the origin of Figure 10.27 sa the =] + jO poing, the curve is now equal w | ++ Gis) FCs), whichis B(s), Since G(s) (3) is know, this function plotted and then the origins moved wa the —1 point to obtain Bs} In general, the open-loop transfer functions of many physical systems.2o not have any poles Pe in the RH s—plane. In this ‘sas, Zp =. Thus, fora stable system thenet umber of retaians about the ~ 1 + j0 point must be evo when there are no poles of Gls) HAs) in the RH s=plane Analysis of Path Q. In applying Nyquist’s criterion, the whole RH s=plane rust be encircled to ensuite the inclusion of all poles zeros in thisportian ofthe plane, InFigure 10.26theentite RHs—plancis cnctosedby the closed path © whichis composed ofthe fllowing four segments 1 One seen ihe naira rom =o to 2 Theseondsepnet te smite i 3, Th hed sees enya fo Jo" 4 Thetoo segment is a semicircle of infinite radius thee ete Tee ‘The partion of the path along the imaginary axts is eepre- sented mathematically by + = ja, Thus, replacing 3 by jas in Equation 10.58 and letting ot take on all values from 60 10 +00 sivesthe portion of the B (17 plot earresponding ta that portion tf the cased contour @ that lies on the imaginary axis. 103, NYQUIST, BODE, AND NICHOLS PLOTS ‘One of the requirements of the Nyquist criterion is that fiony sg [G4s}#1(3)| ~r Cora constant. Thus im, +25[B5)] = img [1 + Gts)H GI + Lor | plus-the constant. As a ‘consequenee, the segment of the closed contour represented by the semicircle of infiniteradi, the corcesparding portion ofthe Bs) plot isa fixed point. As a result, the movement along.only the imaginary axis from —joo ta + joo results in the same pet rotation of B(s) as if the whole cantaur Q were cansidered, I other words, af the rotation of B4s) occurs while the point O, ie Figure 10.26a, goes rom — joo to + jax: along che imaginary axis. More generally. this statement appliesonly to those transfer func= tions G (2)¢8) that conform to the limitations stated eater En this section [6] Effect of Poles at the Origin on the Rotation of Bs) ‘Theamanner in which he w = 0~ and w= 0+ portionsaf the pla in Figure 10.26 are joined i now invertigated for those teansfer functions G(s} that have 5 inthe deno.inatr. Consider thetransfer function with positive values of 7, abd 7 Ge) HO) = aren The direc polar plat of GC) (fa) thie function is ob- tained by substioting s = josinco Eg Figure 1.266. The plot is deen for both postive and negative frequency values. The polar plot draw or negative fours (0~ = w > ~09) isthe conjugate of the pt drawn fr postive frequencies. This means thit the curve for negative equencies itayrametriclt the curve for psinefrequences, withthe el a asthe mis of spe ‘The closed contour Q of Figure 10.26 inthe vicinity has been modified as shown, th other words the point O is moved along the negative magnary axis from = Joo to a point where s = =j¢ = 0" é ~ m/2 becomes very small. Then the pont © roves along venlrelar path of radians = ce” inthe RF s~plane with awry smal radius ¢ unit reaches the positive imapinary aus t # = je we 0 w OFZ. From here the point © proceeds slong the posse inaginary axis fm 4:joo, The, letting the radian approach ze, «— D, for the semicircle around the orgin ensures the inchsion ofl pales and sero in the HH spline, To complete the plot of 8) in Figure 1027, the effort of moving point © on this semicele around the origin mast be investigated, Por the serccnear partion of the path represented bys = cel, where e = 0 and wn/2 20 2 n/2. Equation 1061 becernes 1061) 104, as shown in Gey) = Bm By = her : Hew (nen where Kife + couse —* 0, and ym —@ goes from m/2 to — 11/2 as the directed segment s goes CCW from ef ~ 3/2 to ot nf, Thoin Figure 10.66 the endpoints fromm a) 0” tad wm 0 ee joined by «sic of nite radi the fist aod fourth quadrants. gure 1060 shows the competed contour of G(x).A(s) asthe point O maves along the comtour @ inthe ¢plane in the CW direction. When the origin & moved tothe =1 J0 poi the curve becomes Bt). Th pla of Bt) 13 in Figure 10.26b does not encircle the ~1 + /0 point; therefore, the encirclement N is nero, From Equation 1061 there are no poles within Q; that is, Pe = 0, Thus, when Equation. 10.80 is applied, Zp = and the closed-loop system is stable. ‘Tranefer functions thar have the terms in the denominator have the general form, with s = €e!” ase — 0, Ls 7 femeln GHG) Ka ime — Hagin Ie seen fromm Equation 1063 that, 155 moves from 0 to 0, the plot of Gls) H(s) traces mi CW semiceles of infinite radius about the ecg, ‘Since the polar plots ae symmetrical about the ro ais all that needed is to determine the shape ofthe plo of Gt) (5) fr 2 range af values of © <0 00. The net rotation af the plot far the Tange of 00 << +00 tice that ofthe plot for Ue range ofa 00 10.63) When G{juJH(js} Passes through the Point -1 + j0 Whes the curve of Gio) HC) passes through the 1+ 0 point, the number of encirclements Nis indetermi mate, This corresponds to the condition where (3) has zeros ‘on the imaginary axis. A necessary condition for applying the Nyguist criterion is that the path encircling the specified are must not pass theaugh any pales or zeras of As). When this ‘condition is violated, the walue for W becomes indeterminateand the Nyquist stability ertrion.-cannat be applied. Simple emai nary zeros of B(x) mean that the closed-loop system will have a continuous stendy-state sinusoidal component im its ouput that isindependent ofthe form ofthe input. Unless otherwise staied, this condition is considered unstable. Nygquist’s Stability Criterion Apj Having Dead Time ‘The transfer fonction sepresenting transport lag dead time) is ed to Systems ier Gris =e + Gia) wr ( 10.64) Ithasa magnitude of unity anda negativeangle whose magnitude increaue directly in proportion ta frequency. The polar plot of Equation 10.64 is » unit circle that is traced indefinitely, as, shaven in Figure 10.288, The cortesponding Lm and phase-angle cdaagram shows s constant value ef 04B and a phase angle that decreases with frequency. When the contour is traversed and the polar-plot characteristicafidead time, shown in Figure 10.28, isineluded, theeffcets on the complete polar plot ate as fellows 1. In traversing the imaginary axis ofthe contour @ between O* < at < -+o0, the portion of the polar plotofG (ja) (ja) in the thin quadrants sifted (GW closer tothe —1-+ 4 point (see Figure 10-28 ‘Thos, ifthe dead time is increaed sufbciemy, the =1+ j0 point is enclosed by the polar plot and he system becomes unstable, tae Figure 10.28 (2) Polar plot characestic for transport lag, Egus- tion 10.64; (bl polar plat far Equatioe 10.644 without transport lag (er = 0); Le) desubsinng effec of transport agin Equation 106 2 Asw + 400, the magnitude of the angle can: tributed by the transport lag increases indefinitely. ‘This yields aspiraling curveas [GLa 4 (4)| —+ 0. ‘A transport lag therefore tends to make system lesstable, This is ilustrated far the transfer function GUNA) = aT foie) Figure 10.28b shows the polar plot withoot transport lag; Fig- tare 1D-2e shows the destabiliing effect of transport Tag, 10.3.7 Definitions of Phase Margin and ‘Gain Margin and Their Relation +o Stability ‘The stability and approximate degree of stability [11] can bede- termined from the Lin and phase diagram. The stability charac- teristic specified in eres of the following quantities s+Gain crossover. ‘Thisisthepointon theplotofthe trans fer Function at which the magnitude of Gt} is tity [LG (iu) = @ dB]. The frequency at gain crostover is called the pase-margin frequency ap ‘+ Phasemargin angie. ‘This angleis 180° plus the nega Live trigonometrically considered angle ofthe trane fer Function at the guin crossover point. It is des CONTROL SYSTEM FUNDAMENTALS ignated as the angle y, which can be expressed as ¥ = ISO" + @, where C{G(Jaig)] = & is negative. #Phasecrossover. This is the poiat om the plot of the transfer function at which the phase angle i — 180". The frequency at which phase crossover occurs is called the gain- margin frequency a Gain margin. The gain macgin is the factor a by Which the gain must be changed in arder to produce insta- bility. ie, [Gale = 1 (GijoN = + lme = -LmG(ja.) (06s) “These quantities ae illustrated in Figure 10.29 an both the Lm and the polar curses, Note the algebraic sign associated wi these two quantities as marked on the curves. Figuees 10-298 and 10.29b represent a stable system, and Figures 10.29 and 10.28d represent an unstablesystem. The phase margin angle is the amount of phase shift at the frequency dip that would jst produce instability: The y for minimom-phase systems must be postive fora stable system, whereas negative y mane that the system is unstable, This shown later that y isrelated tothe effective damping ratio ¢ of the system. Satisfactory response is usually obtained in the range of 40° Sy = 61°. Asan individual acquires experience, the value of 7 to be used for a particular system becomes more evident, This guideline fo those systems where beh: order system. The gain margin must be positive when expresoed in decibels (greater than unity asa numeric) fora stable system, ‘A negative gain margin means that the system is unstable, The damping ratio of the stem is also related to che gain margin, However, y gives better estimate of damping ratio, and there: fore ofthe transient overshoot of the system, than dees the gain margin. The values of ap, y. otc, and Lm a are alo readily identified onthe Nichols plot scshown in Figure 10.30 and described inthe nest section. Further information about the speed of response of the system can be ebtained from the maximum value of the comstrolratio-and the Frequency at which this maximum occurs, ‘The relationship of stability and gxin margin is modified for 2 conditionally stable system. [6] Instability can accur with both an increase or a decrease in gain, Therefore, both “upper” and “lower” gain margins must be identified, corresponding to the luppercroseover frequency ay andthelowercrassaver frequency ot Stability Characteristics of the Lm and Phase Diagram ‘The total phase angle ofthe transfer function at any fre- quency is closely related to the slope of the Li curve at that Frequency. A slope of ~20 dBfdecade is related to an angle of 20": a slope of —40 ditdecede is related ton angle of ~ 180°: 1 slope of 60 dBidecade is related ro an angle of ~270": ete Changs of dopeat higher and lower corer frequencies, around 10.3. NYQUIST, BODE, AND NICHOLS PLOTS Figure 10.29 Log magnitude and phase dlagram aed polar plots of GJa), showing gain arg and phase margin: (2) and (b) sale: {c)and id) unstable theparticular frequencybeing considered, contribute ta the tatal angle at-that frequency, The farther away the changes of slope are irom the particular frequency, theless they contribute to the total angle a that Frequency. The stability of a minimmure-phase system requires that y= 0, For this ta be true, the angle at the gain crassover [Lm G¢ju) = 0 dl] must be gremer than 186°, This places a limit on the slope ofthe Lm curve at the tain crostover. The slape atthe gain crossover shouldbe more pose tine hun ~40 decade f the ajacent corse frequencies are nat ow, A slope of —20 didecade is preferable, This is derived from the eonsideration ofa theorem by Bode. Thus the Lm and phase disgram reveals some pertinent information. Forexample, the gain can be adjusted (his raises or lowets the Lm curve to achieve the desirable range of 45° = y= 80". The shape of the 185 Pigure 1030 “hpi Lag mage angle diagram for Ge) = Felts Fiai 4 joe STIG) low-frequency portion ofthe curve detemines sytes type and therefore the degree of steady-state accuracy. The sstem type tnd the gan determine the error coeficients and therefore the stescy-state ertor, The value of ig gives a qualitative indication ofthe speed of response ofa sytem Stability from the Nichols Plot (Lm-Angle Diagram) ‘The Len-angle diagram is drawn by picking for each fhe- quency the values of Lim and angle feom the Lot and phase di agrams vs w (Bode plot). The resultant curve as frequency 3s a parameter: The curve forthe example shown in Figure 10.30, ‘Shows positive guia margin and phase margin angle therefore ‘hisrepresentsa stable syste, Changing the gain aes oF lowers the curve without changing the angle characteristics. Increasing the gen raises the curve, thereby decreasing the gain margin and phate margin angle, withthe result thatthe stability ixdecreased, Increasing the gain so that he curve has a positive Lm at 180" results in negative gain margin and phase margin angle; there- fore, an unstable system results. Decreasing the gain lowers the ‘curve and increasessability. However a large gain is desired to rediuce steady-state eetors (11), ‘The Lin-angle dingram for G(s) #(2)can be drawn for all val ves of 9 the contour Q of Figure 10-263, The resultant carve for minimum-phasesysiens is aclosed contour. Nyquist crite- rion can be opplied to this contour by deteemsining the nuenber of points (having the values 0-dB and add multepls af 180") caclosed by the curve of G(s}4(s). This number isthe value of 'N that is used in the equation 2 = N= Pg to determine the valueof Za. AneampleforG(s) = Ki/|s(1+Ts)]isshownin Figure 10.1, The Lovangle eastour far a nonsninizmum-phase system does not close [11]; thos it smote dificuls to determine the value of M Far these cases the polar plot is easier to use to determine stability Tt sat necessary to obtin the complete Len-angle contour to determine stability for mindmum-phase systems. Only that 186 portion ofthe contour isdeawa representing Glos} forthe range tf values O* < we < 60. The stability is then determined from the position of the curve of (ja) relative to the (9-48, — 180") point, In other words, the curve is traced in the direction of increasing frequency ¢, walkingalongthe curve inthedirection ‘ofincreasing frequency. The system is sable ifthe (0 dB,— 180") ppointistathe right ofthe curve. Thisisasimplifed rule of thumb, ‘which is based om Nyquist’ stability criterion for @ minimum phase sytem. ‘Acanditionally stable system is ome in which the curve exasses ‘the —180° axis at more than one point. The gain determines ‘whether the system i stable or unstable, That i, instability (oF stability} can oceur with bath an increase (ora decrease} in gin eS [et rng or re gure 1031 ‘Telog mapitade angle ga fr Gi) = ay 10.3.8 Closed-Loop Tracking Performance Based on the Frequency Response ‘Acorrelation between the frequency and time responses aa sys tem leeding to a method of gain setting in onder to achieve a specified closed-loop frequency respons, is now developed [3] ‘The closed-loop frequency response is obtained asa function of the open-loop frequency response. Although the design is pes- formed in the frequency damain, the closed-loop responses in the time domain are also obtained. Then a “best” design is se: lected by considering bot the frequency andthe time responses. Both the polar plot and the Lm-angle diagraen (Nichols plot) are used, Direct Polar Plot “The frequency control ratio Cljw)/A@jai for a unity feedback system is piven by Gy Weis Ss = |S ete = secre? (1008) CONTROL SYSTEM FUNDAMENTALS where Alas) = Go) and Bw) = 1+ GLa, Since the magnitude ofthe angle) a3 shown in Figure 10.32 is greater thas the magnitude ofthe angle (ute value of the angle @(w) ie pegatve. Remember that CCW rotation je taken ax postive. ‘The error contol rata E(w} RJ) is given by ‘teh = rere ™ aaa From Equation 10:67 anc Figure 10-33 it is seem tht the greater the distance from the =1 + 0 point to a poiat on the GLjw) locus, for a given frequency: the smaller the steady-state si8i- soidal error for a stated sinusoidal input, Thus, the usefuloest and importance of the polar poof G (jar) have been enhanced, 1067) be Bij Figure 19.32 Petar plat of Gu) fo unity-ferback aye, Determination of Mm and crm for 2 Simple Second-Order System The frequency at which the masiinum value of [CC ‘R(ju] occurs (see Figure 10.33) is referred to asthe resonant frequency tim. The maximum value is labeled My, These ro ‘quantities are figures of merit (RO.M.) of system, Compensa- tion ta improve system performance is based upon a knowledge Of ty and My. Par a simple second-order system a dinect and simple relationship can be obtained fr My ad ay in terms of the system parameters [11, Sec. 9.3], These relationships are Mas (1068) Figureta33 andims A dosedlogp fequency-tesponsecur indicating Mig 10.3, NYQUIST, BODE, AND NICHOLS PLOTS yar _ae_ ere amp 880, F igurel9.4 A plot of Mf w.£ fora simple second-order stem, From these equations itis seen that the curve of Mv. ay haga .peakvalue, otherthan at ar = 0, for omly ¢ < D707, Figure 10.34 shows plot of Meus, ¢ fora simple second-order system. tis seen forvaluesofe < 0.4 that My increases very rapidly in mag ude; the transient oscillatory response is therefore excesively Targe and might damage the physical sysiem. The correlation Ibetween the frequency and time responsesis shown qualitatively in Figure 10.35. Figure 10.35. (a) Pots of M13. w/a fora simple second-order sytem (b) corresponding time plot fora. step input. ‘The corresponding time domain FO.M (11 are The damped natural frequency ons The peak value Mp # The peaktime T, The setting time r(-42%8). Fora unit-step forcing function, these FO. forthe transient af simple second-order system are mF Mp = l+et 110.69) T= xfs (1027 oy = alte 07 hom Mn cao7 “Therefore, fora simple second-order system the follawing con clusions are obtained in correlating the frequency and time re- sponses 1, Inspection of Equation 10.88 reveals that cjg i a Function ofboth wy and ¢. Thus, for a given Z, the larger the Valo of a. the larger cn, and the faster 17 « time of response for this system piven bby Equation 10.72. 2. Laspection of Equations 10.68 and 10.69 shows that bboth My andd My are functions of¢, The smaller £ Ibeoammes, te Inger in valve Mf and My become. “Thus, itiswomcluded thatthe larger the valuecf Mu the larger the value of Mp. For values of ¢ <= 0.4, the correspondence between Mm and Mp is only qualitative for this simple case, In ather words, for {= Othe time darnain yields My = 2, whereas the ‘frequency damsin yields My = 30. When € > 04, ‘thereisa close correspondencelbeween Mx and Mf, Note that the sborter the distance between the 1+ f0 point anda particular Gta) plot (se Fip- ure 10.36), the smallertheddamping ratio, Thus, Max is lager and consequently Mf is also larger. From these characteristics, a designer can obtain a good ap proximation of the time response ofa simple second-order sys tem by knowing only the My andy of fs frequency response A carrespanding correlation for ay and My becomes tedious for mare complex systems, Therefore, a graphic procedure is wenerally used, as shawn in the following sections [13], [14] igure 10.36 Polar plots of G(fea) = Ky /Ljw6l + fw] for oe and the Mf con tourbecomesa straightline perpendicular to the real axis at x = =1/2, 3. For M+ 9, the center ofthe Af circle xy + O and the radius = 0, 4. For MT 1, the centers xp ofthe circles eto the left ofk = =1+ sOand for M < 1,.19 ofthecireles le to theright of x = 0, Allcentersare om the real axis, 2(w) Contours The eco} cantours, representing constant values of phase angle ef) for C1jan)/ R(jeo). cam also be de= termined in the samme manner as for the M contours (11]. The derivation resus in the equation ofa circle with N= tant as a parameter given by (+P +O— m= Fae (10768) shone center inlcated at xy =~}. sv fwtenyi (4) Different values of a result in & family of ccs inthe complex plane with centers on the ne represented by (—1/2, 9), seis trated in Figure 10.38, Tangents to the M Circles [11] The line drawn through the ‘origin ofthe enraplex plane and tangent to a given M ince plays sh with a radius % (00766) 10.3. NYQUIST, BODE, AND NICHOLS PLOTS a igure 1038 Constant or compart, aan important part in setting the guin of GCja). Referring: to ‘Figure 10:39 and recognizing that he = rp isthe radius and ‘ob sm ay isthe distance tothe center ofthe particular Af circle yields sin = 1/M ond 2a = 1. Figure1039 Determination asin y. 10.3.9 Gain Adjustment for a Desited My of 4 Unity-Feedback System (Direct Polar Plot} Gain adjuctrent isthe frst step in-sjsting the system for the desired performance. The procedare far adjusting the gain is outlined i this section. Figure 100 shows (jo) with its respective Mf stele in thecomplex plane, Since Gi Gas a+ jy = KG ad = Kale’ + iy) 1027 9 thea ype hay vane Pash where Gio) = Gyljoo/Ky i defined as the frequency: Figsre1040 (alPltorG, (Jl withsespectine Me cies (ide dwn tangent to both the plot of (jax) andthe ine representing the angie? main 1 /My). sersitive portion of € gs} with unity gain. Note thar changing the gin mercy changes the amplitude and not the angle ofthe locus of points of G (foi). Thus fin Figure 10.40a a change of scales made by dividing the x, y coordinate by &, so thatthe new coordinates are.x’, ¥, the following ae tre: 1. The Gx (Ja) plat becomes the G(jw) plot, 2. The My circle becomes «circle that is simultane ‘ously tangent to Gi (4a) and the line representing say = 3. The —I + jf point becomes the —1/y + j0 post. 4. The radios ry Becomes, = rif Ky. ‘Telspossbieta determine the required gain to achieves desired ‘Ma fora given system yusingthe following graphical procedare, ‘Step 1, Ifthe original system has a transfer fonction KGa) Rell JoTiN + jeTe) Gears jet + jy + 10h) (10.38) Grind = 190 with an original gain Ks, ooly the (equency- sensitive partion (ja) is plotted. Step 2. Draw a straight line throogh the origina the angle We = sin !(1/Mq). measured fram the negative real axis, Step 3. By trial and ervoe find a circle whese center liexon ‘the negative teal axis and is simultaneously tangent tw both the G,(/1) plot and the line drawn at Uke angle W Having found this ciel, locate the point of tan geneyen they angle line, Draw a vertical ine from this point of tangency perpendicular to the real axis Label the point where this line intersects the cea axis asa’, Step 5. For this circle to be an M circle represeating Mey the point a’ must be the —[ + j@ poist. Thus, the 1, ¥coordinatesmust bemuleplied bya gain factor ‘Kn in order to convert ths platinto aplot of Ga) From the graphical construction the gain value is Koi = Ho! ‘Step 6. The original gin must be changed bya factor A = Kafe Ste Note that if Ge(jew), which includes a gain Ka, i aleeody plotted, tisposibleto work directly withthe plot of the function Giri). Follewing the procedure just ourlined results in the determination of the additional gain required to produce the specifica Mr Unt i the addtional gain is Ket Ark (0.799 10.3.10 Constant af and « Curves on the Lm-Angle Diagram (Nichols Chart) ‘The transformation ofthe constant M curves (circles [11], [14] com the polar plot to the Lm-angle diagram is done sore easily by starting from the inverse polar plot since all the a! cnces have the same center at —1 4/0, Also, the constant er contours ae radial lines draven through this point [11]. Theveis a change of sign of the Lim and angle obtained, since the transformations fromthe inverse transfer function oa the inverse polar pot ta the directtransferfunction onthe Lm. plot. Thistransformation results constant Mf and a curves tha have syrametry at every 180° interval As expanded 300° sectionof theconstant Mf anda Figure 104L. This graph iscommonly refersed chart. Note that the M’ = 1 (0 4B) curve is 90? and ¢ = —270" and the curves asymptotic to & M = 1/2(~64B) are always negative, The curve far M the point at 0 4B, = 180°, andthe curves for MM > | are closed, curves inside the linits ¢ = —90" and @ = —270°. These loci for eanstant Af and er onthe Nichols Chart apply enly-for stable unity-feedback systems. The Nichols Chart har the Cartesian coordinates of dB vs. phascangle@. Standard graph paper with loci of constant M and for the closed-loop transfer function is available, The apen ‘CONTROL SYSTEM FUNDAMENTALS loop frequency sespanse GLja). with the frequencies noted along the plot, is superimposed an the Nichols Chart as shown in Figure 10-41. The intersections of the Gi fx) plocwith the Mf and a contours yield the losed-loop frequency response M da. By plotting Lo G',(jw) vx. 4G, (ja) 00 the Nichols Chart, the value of K, required to achieve a desied value of Mu can be Geermined. The amount & dB required to raise ar lower this plot of G, (ja) vs ¢ in order to make justtangentiothedesired M = My contour yields Lm, = A. The frequency value at the point of tangency, is LanG. (aim). yields the value of the resonant frequenicy a! = tm. 103.11 Correlation of Pole-Zero Diagram with Frequency and Time Responses Whenever the closed-loop contre ratio M (iu) has the cha acteristic [11] form shown in Figure 10.33, the sytern may be approximated as a simple second-order system. This usually im- plies that the poles, other than the dominant complex pair are tither far to the left ofthe darsinant camplex poles ar are close to 2eras, When these conditions are not satisfied, the frequency response may have other shapes. This can be illustrated by con sidering the following three eoatrol ratios: cw 1 = v0 ‘Risy Steel « » cor 0313 +08) = vast Ra ~ Geoeeasey Ol) cin ‘ Ro” Gasper oaet So ‘(1082 ‘The pole-tero diagram, the frequency response. and the time response to a step input for each of these equations are shown in Figure 10.42. For Equation 10-80 the following characteristics are noted from Figure 10.42 1. The control ratio has only two complex dominant poles andi no esa. 2. The frequency-response curve hae the following characteristics a, Asingle peak Muy = 1,157 atm = 0.2 bo 1 (in Figure 10.44. Note that, stricly speaking, Gai (2) = Owhen & = 0. However, the denominators ‘of Equations 10.84 and 10.85 both give the same values for the cloted-lnop poleswhen k= 0, namely —1 and ~3, Thove values are the same as the open-loop poles. By convention, all ract, Tocus plotsuse the open-loop palasasthe closed-loop poles when k=o, Gt) = Gas) = Fipme 1044 Howtloows plot for G(s) = arrivy and = 0 ‘The plot providers peat del of mscul information. Fist, it tvesthe pole locations for every posable lsed-loop system that ‘can becreated from the open-loup plant and any positive gain, Second ifthere are poinson the root locus for which the cased= loop spsem would mect the design specifications them simply ‘pplvingthe corresponding vlue of completes the design. For ‘example, if a closed-loop system with damping ratio ¢ = .707 is desired frthe ystemwhove root locus is plowed in Figare 10 4, then simply choose the value oF that pos the poles of G(s a 2 #2. Thats fom Equation 1985, choose & = 5/3. ‘The standard roat locus can be easily apptied to non-unity feedback control systems by using block diagram manipulations {opur the system ian equivalent uty feedback orm (se Chap tert), Because the standard root loess depends only 08 poper siesofpolmamialsit apples equally wellto discrete timesystems. ‘The only changeis that Gs) is replaced by G(z).the tana fonction. 194 10.4.3 Some Construction Rules Evans's procedure for plotting the Root Locus consists of col- lection of rules for determining if tex point, 4, in the cormplex plane is & pole of G(s) for some value afk, The first such rule has already been explaitsed. Rule 1: The open-top poles, iy the rents of ds points in the root lacus plot corresponding tok ‘The second rule ie also elementary. Suppote that 0, ate all dg) = aG) = Fag ay aly yg ype! pt (uae) For physical systems it is always ruc that > m, Although it 5 posible to Rave reasanable mathematical models that vialate this condition, ie will be assumed that n > m, The denominator of Equation 16.85 then (8) = "payne gm tg + (09 bg) (10.88) Rule isan obvious consequence of Equation 10.88 the asp tiom that m > m, and the fact that a polynomial of degree m has exactly n r0ots, Rule : The root locus consist of exactly w branches, ‘The remaining rulex.are derived from adifferent farm of the denominator of Gees), Equation 10.83 shaw thatthe denim nator of G(s) ¢an be written a5 1 4+kG@), Bren though 1+ £G (+) is not a polynocrial iti sill true thar the podes of Ga 6) rs satity the equation Lauer ° (10.88) Because sis 2 complex number, G(s) is generally complex and Equation 1089 is equivalent to two independent equations ‘These could be, for instance, that the real and izaginary parts ‘of Equation 19.89 mustseparately and independenty equal era, tis more convenient, and equivalent, to use the magnitude and angle of Equation 10.89. That is, Equation 10.898 equivalent to the two equations Wow = 1 LOGS) = EHF LINO", where A = 4, 1,2 (10.390) ‘The fist equation explicitly states that, for Equation 10.89 to hold, the magnitude of G(s) must leone, The second equation shows thatthe phate angle of Giz) must be £180", or 540", cc Itispossble to simplify Equations 10.90 somewhat because isareal number. Thus. for k = Equations 10.90 becoane (ios) 91,2). [Gay = Yk CGI = BNE INO", where A ‘Theform fork = @isthesame for themagnitude of Gts), except fora minus sgn }G (691 = ~}),buttheanglecondition becomes integer multiples of 360° ‘Equations 10.91 are the bass for plotting the root locus. The list step in producing the plot is to mark the kacations of the CONTROL SYSTEM FUNDAMENTALS open-loop palesand rrason graph of the complex plane. The pols are denoted by x, a8 in Figure 10.48, while the rros are denoted by os, Ifthe poles and zeros are accurately plotted it is then posse ro measure [G(s] and {4G (s,) for any given test pointy. Forexampe, suppose Gs) = and zeros of his transfer function. [Notice that the plat does not depend inary way onthe gain 10, eis generally ruc that pole-zero plots are ambiguous with reapect to pie gn, Figure MLAS oaetain a plate the complex number (f-+4) forthe specific aluesy = —1-+ 3. ts exacty ‘the same length as, and parallel 9, the vector drawn from the aera tothe points = 1+ f3, alo shown, The sameis tue of the vectors comesponding to the two poles. To sve effort, only the vectors from the poles tothe tet point = —1-+ 43 are drwwn, Once the figure is drawn, simple measurements with & ralerand a peotracier provide iea1+ p= AGR-1+ a) = flats Obras ‘Onecan then checks angle condition in Equations 10.91 to see ify = =1 4 {3 isa point on the root locus for this Gis). Figure 1045 Peter and eee of Ge) = ply. The erento exch ofthe sings theo pole se = TJ ate abo shown, ais the vector 5 + Hat ‘Ofcourse, it would betedious ts check every paint in the eam plex plane. This isos necessary. There isa collection of rules for finding points 00 the root locus plot. A few af these are-de- ‘eloped below. The others can be found in mest undergraduate textbvoks an eantrol. such as [5], [6] ale’: For & > 8, any point on the real axis that lies to the left of an odd number of singularities (poles plus zeros) on the rea axisis a point on the root locus. Anyother point on the real axis, is not, (Change “odd” to “even” for negative). 104. THE ROOT LOCUS PLOT ‘A pro follows immediately from applying the angle con tion in Equations 10.91 totest points onthe relaxis, The angular contributions due to poles and zeros that are not om the real axis cancel as result of symmetry. Poles and teros to the left ofthe testpointhaveanglesequalto zero. Polesand zeros tothe right of the test point contributeanglesof = 180" and-+ (80°, respectively In fact, 2 fourth rule follows easly from the symractry. Ralle : The root locusis symnmetrie about the eal ik ‘Wealteady knw that all branchesstart at the open-loop pales. ‘Where do they end? Rule 5: If G(s) has m poles and we finite zeros (ms =n) then cxactly me branches terminate, as & > 00, om the finite nero ‘The remaining — branches go to infinity ask -+ cc, ‘The validity ofthe rule cam be proved by taking limits ask -» oe in the magnitude part of Equations 10.91. Doing so gives : : seems fn F=0 ‘Thus, ast + oo, it must be true that [G{s)| -» 0. This is tre especie iident i 1 eal Bs bgt + + by Lim, G(s) = Finally, Hm GG) = im tee" =o (082) ‘The bm foetors out and the fact that [G(5)| —* O as 5 —* 20. veith mioltiplicity n — am is apparent. One cam think of this asa demonstration that G(e) has m — mi zeros at infinity. Equation 10.32 plays an inoportant roe inthe prof the next rule aswell Rule 6: IF GU) bas w poles and m finite neros (a > nt) and A> Othenthe n—m branches that end atinfinity asymptotically ‘appraach lines that intersect the real axis ata point ny and that sake an angle y-with the real ais, where (ot 0" a where b= 0, 1.2, oad £ retpn-E teen [A proof of the formula for y fllows from applying the angle condition of Equations 10.91 te Equation 10.92. That is Pn fa") = 2(1+2A)180" * ym eye Eta ‘A proof of the equation for eu cam be found in [5]. pp 282- (254. Mast textbouks include around a deren rules for plocting the oot locus see (5]-[7] for example. These are much lest important todry than they were just a few years ago Because toed. inexpensive software for plotting root lal is bow widely available, 195 10.4.4 Use of the Computer to Plot Root Loci ‘There are many different software packages that car be used to plot the root locus. Particularly well-known examples are MAT- LAB, Matrix-X, and Contzal-C. To same extent the software is foolproof. If the data are inpot correctly, the resulting root lo= us is calculated correctly. However, there are several possible ptfalls. For example, te software automatically scales the plot ‘Thescaling can obscure important aspects of the raat locus, a is described inthe next section. This, and other possible problems associated with computer-generated reat locus plots sdiscussed in detail in [4]. 10.4.5 Uses “The root locus pict can be an excellent 120! for designing single- {npr single-ourpur control systems. It is particularly effective when the open-loop transfer function is accurately known nnd is, at leastapprosimutely, reasonably low order. Thisis often the eae in the design of vervomechanitms. ‘The roat lorus is also ‘very useful as an aid to understanding the effec of feedback and ‘compensation on the closed-loop system poles. Some-ways 40 use the root locus ase usteated below, Design of a Proportional Feedback Gain Consider the open-loap plant G(s) = 1fs(s + IMs 1). This atypical textbook example. ‘The plant i third order and given in factored form. The plant has been narmalited so that fim G(s) = |. This is particulary helpful i. comparing Alferent candidate design. verve motor diving an inertial oad. would typically have such a decription, ‘The motor plus Joad would correspond to the poles at 0 and =1. The electrical characteristic ofthe motor adda pole thats normally rly far ‘othe lef, such asthe pote at —1in this example, The simplest controller is pure gait asin Figure 1B43. Sup ‘pose, for istration thatthe specifications on the closed top system are that £, the darpping ratio, mst be exactly 0.707 and ‘the natural frequency, ust be as lage as possible. Because the closed-loop systems atally thied order, both darsping ratio. and natural frequency are net well defined. Hawever, the open loop system fas a pair of dominant poles those at and —' “The dosed-loop syriem vill also ave « dominant pals of poles because the thifd poe willbe snare than a factor often tthe left ofthe complex conjugate pair of poles. The damping ratio and natural frequency wil then be defined bythe dominant poles ‘A roat lacus plot for this system, generated by MATLAB it shown in Figure 10.46, (Using the default scaling it was dificult tosee the exact tersetion ofthe locus with the ¢ = 0.707 lene ‘because the automatic scaling includes ll the poles in the visible Plot, Aversion ofthe same plot with better scaling was

You might also like