Chapter 1 1 Dynamics and Control: An Overview IN Parc 1 we examined the form and function of the major families of power elec- tronic circuits. Our goal was to show how the intended power conversion function. is achieved in each case by appropriate configuration of the circuit components and by proper operation of the switches. Throughout those earlier chapters, our concem was with riominal operating conditions, that is, the ideal operating conditions in Which a circuit is designed to perform its primary conversion function. As nominal operation in most power electronic circuits involves a periodic steady state, we focused on situations in which circuit operation and behavior are the same from eyele to cycle. Now we need to deal with the consequences of the inevitable disturbances or errors that cause circuit operation to deviate from the nominal. These disturbances include variations and uncertainties in source, load, and cireuit parameters, per- turbations in switching times, and events such as startup and shutdown, We refer to the resulting evolution of the deviations from nominal behavior as the dynamic behavior of the circuit. If such deviations have a negligible effect on circuit oper- ation, the user may be content to let the circuit run without corrective action, This is rarely the case, however. Most often, departures from nominal conditions have to be counteracted through properly designed controls. We show several examples of circuits that do not recover, or whose recovery is incomplete or too slow, without such con- trols. A controller or “compensator” must first provide the user with a simple and convenient means of selecting the desired nominal operating condition. Second, it must automatically regulate the circuit at this operating condition by delaying or advancing the times at which switches are tumed on and off. Hence in this chapter we begin to build the basis for analyzing the dynamic behavior of power circuits and for designing and implementing controls that regulate these dynamics, maintaining operation in the vicinity of the nominal, despite disturbances or errors. The focus of Part If is on analysis and control design by means of appro- priate dynamic models. This approach allows us to anticipate the behavior of the * 253 ‘e \54 Chapter 11 Dynamics and Control: An Overview cireuit under diverse operating conditions, generate candidate controller structures and parameters, plan simulation studies, understand experimental results, recog- nize which regimes of operation call for further investigation, and so on. Such an approach is especially critical in power electronics because of the effort and cost involved, even in breadboarding, and the expense (not to mention the distress and, sometimes, smoke!) associated with component failure, Analytical studies must, of course, be combined with engineering experience and intuition, experimentation, and other ingredients of the design process in order to be successful 1.1 CONTROL SYSTEM CONFIGURATION Al control is based, explicitly or implicitly, on a model that describes how control actions and disturbances are expected to affect the future behavior of the system. Control of a power circuit entails specifying the desired nominal operating condition and then regulating the circuit so that it stays close to,the nominal in the face of disturbances and modeling errors that cause its operation to deviate from the nominal In simple open-toop control, the controller is not given any information about the system during operation, although the open-loop controller may be constructed ‘on the basis of prior information or models. In open-loop control with feedforward, the controler utilizes measurements of some of the disturbances affecting the sys- tem. Using feedforward, the controller can attempt to cancel the anticipated effects ‘of measured disturbances. Feedforward alone is usually insufficient, however, to ‘obtain satisfactory performance in power electronic circuits AA better strategy is for the controller to also use measurements that reveal the circuit's present behavior. The co sn thereby asses tof departure from the desired behavior and choose control,actions aimed at restoring the system rapidly and safely to nominal operation. This strategy is the essence of closed-loop or feedback control. When the model that underlies the choice of control actions is itself updated on the basis of measurements, we refer to the controller as adaptive. The block diagram in Fig. 11.1 represents the typical situation, Each connection between the blocks represents one or more actions, measurements, oF information flows affecting the block at which the arrowhead terminates—and originating in the block at the other end of the connection. We frequently refer to the feedback loop in Fig. 11.1. The output measurements fed back to the controller provide information about the behavior of the system and about the variables that we want to control The quantities directly affecting the system, namely, the control inputs and disturbances, are shown on Fig. 11.1. It also shows the feedforward of those dis- turbances that can be measured. The unmeasured disturbances, as well as modeling errors, cause the controller's actions to have unanticipated effects. The output mea- surements available to the controller are corrupt&d by what is termed measurement noise, or sensor noise. This noise, along with the unmeasured disturbances and modeling errors, causes inaccuracies in evaluating measurements. Desires nominal operation Figure 11.1. The typ EXAMPLE 11.1 Open-Loop Cont Let's consider an up built according to th ‘or with a switching (We discussed such voltage (2) within changes in the input low as 8 V. For purr ‘or uncertainties in th switches. You will s the circuit response « Figure 11.2 cu Recall from Ch with a duty ratio D ig = Vig. then (0) the inversion of polat nominal value of L under nominal condiwr structures ults, recog- on. Such an ort and cost listress and, ies must, of ‘imentation, row control the system, g condition in the face ‘e from the ation about constructed ‘edforward, rg the sys- ated effects owever, to t reveal the f departure the system ‘osed-loop actions is s adaptive. ‘connection ‘formation sinating in e feedback er provide at we want inputs and those dis- 5 modeling uutput mea sasurement fances and Section 11.1 Control System Configuration 255 a ewes eet Sacro ‘operation on Measured ew —— Feedback path _ Figure 11.1 The typical control system configuration, EXAMPLE 11.1 Open-Loop Control with Feedforward for an Up/Down Converter Let's consider an up/down (or indirect, or buek/boost) de/ée converter whose power circuit is built according to the circuit schematic in Fig. 11.2 and is operated ata frequency of 50 kt, or with a switching period of T = 20 ys. Let R =2 9, C = 220 uF, and L = 0.25 mH. (We discussed such circuits in Chapter 6.) We would like to maintain the average output voltage (x) within 5% of the nominal or reference value of Vier = —9 V, despite step changes in the input voltage vy from a nominal dc value of Viq = 12 V down to values as low as 8 V For purposes of this example, let's assume that there are no other nonidealities ‘or uncertainties in the circuit and, in particular, that the transistor and diode function as ideal switches. You will see that, even ifthe system behaves according t0 this idealized model, the ciruit response can be unsatisfactory Figure 11.2 Circuit schematic of the power stage of an up/down converter. Recall from Chapter 6 that, if the transistor is cured on periodically and operated with a duty ratio D and if the converter is operating in continuous conduction mode with Yq = Vigy then (v,) = ~VigD/D! to a good approximation, where D’ = 1 — D. (Note the inversion of polarity between input and output.) The duty ratio must therefore be set at 4 nominal value of D = Vege /(Vig: ~ Viq) = 0:43 in order to obtain the desired operation under nominal conditions ji56 Chapter 11 Dynamics and Control: An Overview o ce) ® 32 Sale Ee) 4 V) eee eens ia a 6 “Time (ms) Time (ms) @ © Figure 11.3 (a) Response of the ideal up/down converter circuit to a step from 12 V {to 8 V in the input voltage vj,.(b) Response to the same step after ‘incorporating feedforward Figure 1.3(a) shows the response of this idealized circuit toa step in the input voltage from 12 V 10 8 V. The waveform before time fp corresponds to operation in a cyclic steady state with an input voltage of 12 V and a duty ratio of 0.43. Not unexpectedly, the ideal circuit produces the correct average output voltage under nominal conditions—before to, At fp the input voltage drops to 8 V. The circuit undergoes significant oscillatory transients and settles down to an incorrect average value of output voltage, namely, —6 V. We can explain completely the steady-state behavior with the static model that we already have, but the explanation of the oscillatory behavior must await the modeling results in Section 11.3 ‘One natural idea for obtaining better responses to changes in the input voltage ty isto Use feedforward. If D is made to vary in accordance with measured variations in ty rather than being fixed by the nominal value Vig 50 that ~¥jgD/-D" is held constant at Vg, then, will atain the correct steady-state average value (v,) despite input-voltage variations. This ‘pulse-width modulation (PWM) approach requites that we select D = Viee/(Viey ~Uiq)- The resulting response to a step change in the input voltage is shown in Fig’ 11.3(b). The ideal Circuit with feedforward now settles down to the correct average output voltage, despite the step in input voltage, The peak excursions are considerably less, though sill outside the allowed 5%. The transients, however, take as long to die out as they did without feedforward We can explain why the transients in Example 11.1 appear as they do after we obtain a dynamic model for the up/down converter in Section 11.3 (see especially Example 11.7). Feedforward in Example 11.1 compensated for the steady-state effects of disturbances in the input voltage but did nothing to modify the dynamics of the transient. In the presence of nonideal components, even the steady state would not be accurately restored with feedforward alone. More than feedforward is needed to obtain significantly better behavior. We return to the’ converter in Example 11.1 several times to illustrate various aspects of dynamic modeling and * control design. + EXAMPLE 11.2 Closing the Loop The citcuit schematic phase-controlled brid (We discussed such Js denoted by 2) anv of the analysis of no assumptions that the is no commutating ree denoted by a for Nominal operat: the firing angle is ms 4] Phase conte eter @ Ler] Controller alee Figure 11.4 a} Sche: of applie the kth» from 12 V 2 step after | put voltage elie steady ly, the ideal | before to, 1 transients 5. We can ay have, but ection 11.3 age ti, isto in vj, rather Vee. then vy, ations. This —%,)- The ). The ideal age, despite ‘outside the i eedforward, fo after we especially teady-state dynamics veady state redforward voverter in deling and i Section 11.1 Control System Configuration 257 EXAMPLE 11.2 Closing the Loop on a Controlled Rectifier Drive for a de Motor ‘The circuit schematic in Fig. 11.4(a) represents a separately excited de motor, driven by a phase-controlled bridge rectifier supplied from a nominally sinusoidal voltage source, ty. (We discussed such controlled rectifiers in Chapter 5.) The voltage applied to the motor is denoted by vy and the armature current by 1g. The waveforms expected on the basis of the analysis of nominal operation in Chapter 5 are displayed in Fig. 11.4(b), with the assumptions that the armature current remains positive theoughout each cycle and that there {is no commutating reactance. Figure 11.4(b) shows the role ofthe firing angle ofthe rectifier, denoted by a for the kth cycle, in specifying the voltage waveform. Nominal operation of the drive system corresponds to a cyclic steady state in which the firing angle is maintained at a constant value 0 = and the armature current varies te 7 Be a6 é tes NS Ti : o (b) a ter—>} Controller. -® Phase-controlied |_"4] Armature = an rectifier areas @ Figure 11.4 (a) Schématic diagram of phase-controlled rectifier drive. fb) Waveforms of applied voltage », and current i,. with a rectifier fring angle of ay in the kth cyde. (c} Control block diagram of the drive.258 Chapter 11 Dynamics and Control: An Overview periodically at twice the frequency of 1c: Our objective is t0 use closed-loop control of the sensitive, slow firing angle to regulate the average armature current (i,) at any specified reference value troller reacts i in some range. This reference value, qj, is specified by a higher-level speed or torque aggravate devia controller forthe motor. The average current (ig) determines the average torque applied to opezation. In ot ‘the motor shaft, Ifthe mechanical time constant ofthe motor is considerably larger than the to design and it period ofthe soure, the speed of the motor displays very litle ripple in the steady state and depends essentially on (i) ; To represent the system entirely by a control black diagram—for comparison with the general contol configuration presented in Fig. 11-1—we ean transform Fig, I1-4() 0 11.2 MODEL $ Fig. 1.4(0). The presence of commutating reactance would intoduce a feedback connection from i tothe phase-conuolled rectfer inthis block diagram, indicating thatthe applied | The use of an a voltage rz would then depend on iy process. Differe ‘We have also chosen in Fig. 11-4) to represent the sysem ata finer level of modeling control design, detail than in Fig. 11.4{a), with the de motor mode! split into subsystems corresponding to to be several po: the armature circuit andthe electromechanical conversion process. The connections to ad domain of defin from the eletromechanical subsystem are showin in dashed lines because the back emf, E, offs are involves js most commonly treated as a slowly varying extemal disturbance acting on the armature For instance utput to 1 every reasons that we explain in Chapter 13, We assume in shat follows that m(t) varies slowly ris low initially If m(t) does not change significantly over the course of a cycle, that is, if significant variations in m(t) occur at substantially less than one half the switching frequency, then ‘m(t)/K at any time closely approximates the prevailing duty ratio. The control circuit varies ra(t) around its nominal value in accordance with the need to increase or decrease the duty ratio. This determination in tum is made by feedback signals that measure how far the power circuit is from nominal operation. Evidently, the closed loop “bandwidth” of such system, or the speed with which the controled system return to its nominal operation, is substantially Jess than one half the switching frequency under these conditions. Figure 11.9(b) shows how mit), g(t), the duty ratio d. and the continuous duty ratio. L {t) are interrelated. Note that, provided the change in duty ratio from eycle to cycle is suliciently small, dit) also closely approximates the prevailing duty ratio d,. Although we ‘ can choose fo constrain mit) to have only slow variations, d(@) is intrinsically constrained arr to vary slowly. The fastest possible variation in dy occurs when the duty rato altemates between high and low values in successive eycles. The corresponding period of dit) is | 2T. Thus dt) can never have a fundamental frequency higher than one half the switching frequency. ee [Note that variations in A’ could also be used to vary the duty ratio, Feedforward control Rar that compensates for supply voltage variations is commonly implemented this way. With the upidown converter circuit in Examples |1.1 and 11.3, for instance, the required feedforward can be achieved by making K’ proportional 10 Vig ~ Vier (recall that Vege is negative). For down converters and other circuits ofthe form shown in Fig. 11.8, we make A proportional to Vig. We leave you to verify these claims, Our averaged models typically show the dynamic dependence of averaged waveforms in the power circuit on d(t). However, of most interest in control design erties is the dependence on some real modulating control signal, such as m(¢) in Example 116, For slow variations in the modulating signal, we showed that d(t) = m(t)/K, so the averaged models are relevant. But the response of our averaged models to j copainoe Perturbations in d(t) at frequencies that exceed, or even approach, one half the » linearly to switching frequency do nor approximate the response to perturbations in m(t) at Sonia those frequencies. V at the sart of © duty ratio dy swtooth ramp in 11.3.4 Averaging a Switch ” say ratio. Although we have covered considerable ground without looking in detail at av- or by varying it raging the variables associated with a switch, it is now time to do so. Switch foeee | averaging is especially rewarding inthe case of high-frequency switching or PWM Ease | convenes, The reason is that we can approximately characterize the averaged switch in such circuits entirely in terms of the (control-dependent) constraints that270 Chapter 11 Dynamics and Control: An Overview Section 11 interval t—T < > ‘The relationships 0 characterization of 1 | imposed on the avi | ment is fully chara ‘Two circuit represe (©). Although these o sources is simpler representation invo duty ratio is held cy) © Figure 11.10 (a) The canonical switching cell for high-frequency switching con- verters. (b) Approximate averaged switching cell for continuous i ‘The high-frequ conduction with duty ratio d, using controlled sources. (c) Approx- switeing cell at th imate averaged switching cell, using ideal transformer; d’ = | — d. | Hence approximate by replacing the ca it imposes on the averaged circuit variables at its terminals. As a result, we can | replace the switch in the instantaneous circuit with a corresponding component | in the averaged circuit. The other components in the circuit are usually LTI, and | EXAMPLE 11.7 are unchanged by averaging. Hence circuit averaging for such a converter simply | Averaged Circuit involves replacement of the switch by its averaged model. Ik is natural to start with the canonical cell of Fig. 6.7, reproduced in Fig. 11.10(a) for convenience. The switching period T will again be chosen as the averaging interval. The assumptions that we make in obtaining an averaged model can be stated in different ways. We list them here as: Our development of cell structure within vy, Of Fig. 1.10 is Wwe replace the canon ideal transformer win switches of Fig. 11.2 results in Fig. 1.11 1. a small ripple assumption, namely, that the voltage v,, (=v) and current i, (Giz) at time ¢ are well approximated by their average Values, .(¢) and T,(), respectively; and 2. a slow variation assumption, namely, that these two average values do not vary significantly over any averaging interval of length T, that is, they vary substan- tially more slowly than one half the switching frequency Both assumptions are generally satisfied in well-designed high-frequency switching converters operating in continuous conduction. ‘Suppose that the switch in Fig. 11.10(a) is govemed by a 0-1 switching func- tion (), such as that described in Example 11.6, and is in position y with duty ratio d(t). As i, = qi,, we can write 1, = Ji,. Now our assumptions allow us to treat i,(7) as though it were approximately constant at Z(t) over the averaging Figure LLL AverSection 11.3 Obtaining Dynamic Models by Circutt Averaging 271 ineral ¢=T <7 € t 0 =H, © HOI) = db (115) Similarly, v,, = guyz. 80: F,4(1) = Wye * HOW, (0 = doy, (16) ‘The relationships obtained in (11.5) and (1/.6) constitute the desired approximate characterization of the averaged switch in terms of the control-dependent constraints imposed on the averaged circuit variables at its terminals, (A three-terminal ele- Ye ment is fully characterized by two constraint equations for its terminal variables.) ‘Two circuit representations of this characterization are shown in Fig. 11.10(b) and (c), Although these two are equivalent, the representation in terms of controlled ) sources is simpler for such tasks as linearization, which we consider later. The | representation involving the ideal transformer is useful for situations in which the | uty ratio is held vatching con- duty ratio is held constant Freontactactint gta) ‘The high-frequency converter topologies in Chapters 6 and 7 had the canonical 5.) Approx. switching cell at their core, with all other elements being linear and time invariant. gta Hence approximate averaged circuits"for all these converters are simply obtained by replacing the canonical cell by its approximate average, ‘eoult, we can 1g component tally LTT, and | EXAMPLE 11.7 Werte SeOPya ie Averaged Circuit for an Up/Down Converter Our development of the indirect switching converter in Section 6.4 identifies the canonical eens 3 cell structure within the up/down converter topology in Fig. 11.2. Note that in this case vy, of Fig. 11.10 is actually vi, — vc rather than vc. To obtain the averaged model, an averaged ‘we replace the canonical cell by its averaged version from Fig, 11.10(c), substituting the ideal transformer windings of d'(t) and dt) turns, respectively, for the transistor and diode switches of Fig. 11.2, We also replace all instantaneous quantities by their averages, The 1nd current i, result is in Fig, 11.11 (#) and 7, (0), do not vary vary substan- xey switching itching fune- xy with duty cons allow us be averaging Figure 11.11 Averaged circuit for an up/down converter.72 Chapter'l1 Dynamics and Control: An Overview “This averaged circuit can be wed asthe bass for generating other pes of analytical models sch , for example, state-space models. The eiruit can also be acepte directly by tmany standard circuit simulation packages. Not tha the circuit depends nonlinearly on the contol variable dt). The process of linearization that we examin in Section 1.4 provides one way of dealing wit the noninety ‘When the diy ratio dl) is constant, the averaged circuit of Fig. 11.1 is LTL, so that analysis inthis case is straightforward. For example, the circuit ean immediately be used to read off average values of voages and currents inthe nominal seagy ste obtained with Tia = Vig and lt) — D, In ths steady state, the average inductor vokage and capacitor curen ae both 20, which leads to the following expression fr the average steady-state indtorcurten, denoted by Tz, and steady-state output voltage TY, (or capacitor voltage, Yoo ain ee and ¥,= (ts) = Vay ‘These expressions are consistent with the results in Section 6.4 ‘The averaged circuit also provides a basis for understanding the results in Example 11.1, where we examined the open-loop response of v, (= tc) 10 step in the applied voltage qe With constant d(t) = D. Using the averaged circuit to compute the transfer function from the localy.averaged input voltage to the locally-averaged output voltage-—referred 10 as the audio susceptibility ransfer function of the circuit—we get PAO) =D'D/LC Ta) ~ = FU/RO\s + (DLO) 1s) With a step change in vg(t), the averaged input Zg(t) actually takes one averaging interval to cross between is initial and final values. Als, the averaging approximation in (/1.6) is poor during this interval. However, this interval Tis much smaller than the time constants ‘of the averaged model, so we can consider Tin(t itself to be a step function and can use the transfer function (11.8) of the average model to compute the response to this step. ‘The computations required to find the step response of the averaged model are outlined in Section 13.3, in the context of solving general LTT models. For now, you simply need to recall that the form of the response is largely determined by the poles of the transfer function (118). For the underdamped case, whichis what the parameter values in Example 11.1 corespond fo, the poles \; and A; afe complex and conjugate to each other: 1 1 = i= -aRg tp tl w= \ Fe - ae 9) where “ denotes complex conjugation. The transition from the steady state before the input step 10 the steady state after itis thus governed by a transient of the form: eye a chet = co MORO cinta pt +8) (11.10) where c),c, and @ are constants fixed (upon invoking the continuity of the inductor current and capacitor voltage) by the initial conditions prior tothe step. See Section 9.1 for a more Getailed analysis of a very similar situation. Section 11:3 A detailed calcula also quantitatively —for The time constant 2RC oF 146 cycles. A similar. shown in Fig. 11.3(b), 1 change as well, toa valk function tht allows ws ¢ Sometimes the ca led as an equivalent Fig. 11.10(@) may no the approximation in ( averaged-switch made to averaging (Section The switch averag (Section 6.7). The swi contacting neither 4 1 inductor is now appro switch. The style of ¢ however, Switch averaging 4uasi-resonant de/de © essary computations averaging exploits the the average current of resulting averaged-swi that approximately con 11.3.5 A Generalizatic We began this discuss electronic circuits our definition of a local av. average. In circuits suct are interested in the sw de/de converters, we a resonant variables and. ‘The component at following generalizatior We call this variable the2s of analytical ied dieecty by linearly on the 1 114 provides is LT, so that tely be used to obtained with and capacitor ge steady-state acitor voltage, a7) Example 11.1, pplied voltage asfer function e—referred to 18) aging interval on in (11.6) is ime constants ‘and can use this step. 4 are outlined simply need £ the transfer sin Example ther: tg) fore the input (1110) uctor current 1 for a more Section 11.3 Obtaining Dynamic Models by Circuit Averaging 273 AA detailed calculation based on (11.8) accounts very well—not only qualitatively but also quantitatively —for the average behavior of the open-loop step response of Fig. 11.3(a. ‘The time constant 2RC is 880 ys oF 48 switching cycles, andthe period 2=/p is 2924 ys, (or 146 cycles, A similar calculation can be carried out forthe step response with feedforward, shown in Fig. 11.3(6). However, we now have to account for the fact that D has a step change as well, to a value determined by x, after the step. Example 11.9 derives a transfer function that allows us to compute the response to (small) changes in D. Sometimes the capacitor in the canonical cell has significant resistance, mod: led as an equivalent series resistance (ESR). In this case, the voltage v,. in Fig. 11.10(@) may no longer satisfy the assumption of small ripple that underlies the approximation in (1.6). It is possible to refine the approximation and obtain an averaged-switch model that accounts for ESR. Alternatively, a state-space approach to averaging (Section 12.3) permits us to handle this situation systematically ‘The switch averaging in Fig. 11.10 does not apply to discontinuous conduction (Section 6.7). The switch of the canonical cell in this case takes a third position, contacting neither y nor z. Using the fact that the average voltage across the inductor is now approximately zero, we can still obtain an averaged model of the switch. The style of analysis demonstrated in Example 11.5 is usually simpler, however. ‘Switch averaging is effective for a related family of do/de converters, namely quasi-resonant de/de converters, typified by the circuit in Example 9.6. The nec- essary computations are somewhat more involved than the preceding ones. The averaging exploits the fact that the average voltage of the resonant inductor and the average current of the resonant capacitor are both approximately zero. The resulting averaged-switch model again displays the control-dependent constraints thet the average values of the switch variables. 11.3.5 A Generalization: The Local w-Component We began this discussion of circuit averaging by saying that for many power electronic circuits our interest is in the average values of circuit variables. Our definition of a local average then enabled us to study the dynamic behavior of the average. In circuits such as the resonant de/ac converters of Chapter 9, however, we are interested in the switching-frequency component of the output. With resonant de/de converters, we are interested in the switching-frequency component of the resonant variables and the average value of the output. ‘The component at a frequency w in a waveform (7) can be defined by the following generalization of the local average in (1/.1) =f, We call this variable the Jocal w-component of x at time t. In terms of this notation, x(r)dr aL)274 Chapter 11 Dynamics and Control: An Overview the local average is the local 0-component, given by Z(t) = Z°(t). The choices of involves circuit var T and w are usually interrelated. For the resonant converter, we would pick T to Chapters 12 and 13 be the switching period and set ws = 27/T. With this choice, Z~ in the steady state Averaged modi is the (complex) amplitude of the fundamental Fourier series component for the sveraged:-circuit mo periodic waveform z(7), The expression in (1/.11) extends the notion of a Fourier a(t), corresponds to Series component atthe frequency w tothe nonperiodie case. the high-frequency To study the dynamics of E(t), we need an expression for its derivative. It is ‘model has a steady We can now rey aH) = Fw — ju“ (inay nonlinear terms of a ing only first-order ¢ where E* is the w-component of the derivative dz(r)/dr, We leave it to you to deviations; this is th see what (11.12) implies forthe construction of the w-component circuit from the | pend on the nomtin original switching circuit and to explore the use of this generalization in circuits depend on the nomi such as resonant converters. and if the nominal linearized model tar 11.4 LINEARIZED MODELS ‘The static characteristics of power electronic circuits often depend nonlinearly on the control variables, and their dynamic characteristics are even more likely 10 do | so. With linear or nonlinear feedback control, the closed-loop system is typically | also nonlinear. Assessing stability and designing or evaluating controllers with nonlinear models are usually difficult. The most common, systematic and generally successful approach to these tasks is linearization. It yields linear models that approximately describe small deviations or perturbations from nominal operation ofa system, Linear models are, of course, far easier to analyze than nonlinear ones. Linearized models, also called small-signal models, are crucial 10 evaluating the stability of a nominal operating condition. Stability of the linearized model indicates that the nominal operating condition is stable for small perturbations at least, An initial goal for control design is therefore stabilization of the linearized ‘model. This task is much easier than direct stabilization of the nonlinear model. ‘We outline the basis of Linearization in this section and treat it in more detail in Chapter 13. We discuss the application of the resulting linear models in stability evaluation and control design in Section 11.5 and Chapters 13 and 14, 11.4.2 Linearizing < For nonlinear model earization. Using sit linearized model itse .BOUS fo those we use an instantaneous-cire signal analysis of tra We begin by rep | from the nominal, T voltage law equation | fied is that we obtai of voltazes—the pert equations, Similarly, tion from the nomins law on the given circ ‘The final step is linearized version. (T impose the same con: 1.4.1 Linearization The linearization of « The place to start when linearizing a continuous-time or discrete-time dynamic characterizing equatic ‘model is with a nominal solution, usually a steady-state solution. Even the steady linear constraints that state in a typical dynamic model for a power circuit may involve periodically- component's termina varying rather than constant quantities. For example, steady state in the instanta- the component. The r neous model of an up/down converter involves a periodic switching function and small deviations from Periodic waveforms. One way to get a constant steady sate rather than a periodic The nonlinear ave ‘Steady state in power circuit models is by working with e discrete-time model that the nominal solution cThe choices of buld pick T to he steady state ponent for the on of a Fourier lerivative. Iris (1.12) ve it to you to ircuit from the ion in circuits ronlinearly on re likely to do m is typically introllers with and generally r models that inal operation onlinear ones. to evaluating varized model srturbations at the linearized near model 1n more detail als in stability 4. ime dynamic fen the steady periodically- the instanta- function and van a periodic ‘ae model that Section 11.4 Linearized Models 275 involves circuit variables sampled once per cycle. The reason becomes clear in Chapters 12 and 13. ‘Averaged models can also yield constant steady states, Steady state in the averaged-circuit model given in Fig. 11.11, when we impose a constant duty ratio dit), corresponds to the averaged variables taking constant values. However, with the high-frequency PWM inverters described in Section 8.3, even an averaged model has a steady state that varies periodically rather than remaining constant. ‘We can now represent small deviations from the nominal by expanding all the nonlinear terms of the model into Taylor series around the nominal values. Retain- ing only first-order terms results in a linear model that approximately governs small deviations; this is the linearized model. The parameters of the linearized model de- pend on the nominal operating condition, because the Taylor series coefficients depend on the nominal solution. If the original nonlinear model is time invariant and if the nominal solution corresponds to constant values of the variables, the linearized model tums out to always be LTI. 11.4.2 Linearizing a Cireuit For nonlinear models in circuit form, we can further simplify the process of lin- earization. Using simple operations on the nonlinear circuit, we can obtain the linearized model itself in circuit form. The procedure and justification are analo- gous to those we used in Section 11.3.2 to obtain an averaged-circuit model from an instantaneous-circuit model, The arguments should also be familiar from small- signal analysis of transistor amplifier circuits, for example. ‘We begin by replacing every voltage in the nonlinear circuit by its deviation from the nominal. This step results in voltage deviations that satisfy Kirchhoff" voltage law equations on the given circuit topology. The reason KVL is satis- fied is that we obtain the deviations by taking the difference between two sets of vollages—the perturbed set and the nominal—that each satisfy the same linear ‘equations. Similarly, we replace every current in the nonlinear circuit by its devia tion from the nominal to obtain current deviations that satisfy Kirchhoff's current law on the given circuit topology. ‘The final step is to replace every nonlinear component in the circuit with its linearized version. (The linear components do not need replacement, because they impose the same constraints on the deviations as they do on the original variables.) ‘The linearization of a nonlinear component is obtained by Taylor expansion of its characterizing equations up to first-order terms. The linearized component imposes linear constraints that approximately govern small deviations of the variables at the component's terminals and small deviations of any control variables that govern the component. The result of all these manipulations is a linear circuit that governs small deviations from the nominal ‘The nonlinear averaged circuits that we described earlier are time invariant, and the nominal solution of jaterest is usually the constant steady state. The linearization6 Chapter 11 Dynamics and Control: An Overview in these cases is an LTI circuit, In the case of averaged-circuit models for the PWM waveshaping inverters in Section 8.3, the nominal steady state is periodically varying, and the corresponding linearization is a periodically varying circuit ‘XAMPLE 11.8 Linearized Circuit for an Up/Down Converter in Discontinuous Conduction ‘A noalinear averaged-circuit model for the upidown converter in discontinuous conduction, ‘was obtained in Example 11.5, Fig, 11.7(b). Our derivation of the model there allows the uty ratio and input voltage to be slowly varying rather than constant. We can therefore rewrite (11.4) with dit) instead of D and Tg instead of Vig. We reserve D and Vj, to denote the constant nominal values ‘The nominal steady-state solution corresponds to constant values of the averaged! quan: tities. In particular, ,(#) = I, and ¥,(t) = Vg, The capacitor in the averaged circuit acts as ‘an open circuit in the steady state, so we see from Fig. 11.7(b) that: ir 2EN, 13) Rly = —RfWe,Viq. D) = R Rearranging and taking the square root yields V, = —Vj,D //RIV2L. (With our sign convention, we need the negative square root) Now let the duty ratio be perturbed from D to dit) = D+ dit), but assume for simplicity that the input voltage is fixed at Vi. Correspondingly let f(t) = Ty + Ty(t) and. ¥,(0) = Vz +0,(¢). The superscript ~ denotes a perturbation from the nominal. Linearizing the current source in the averaged circuit by a Taylor series expansion up to linear terms, we find aay ‘The partial derivatives in (/.14) are evaluated at the nominal solution. Using (11.13) to simplify 17.14), we have 10 = UR) + Vg VATRDMO ats; ‘The resulting linearized averaged circuit is shown in Fig. 11.12. “Figure 11.12 Linearized averaged circuit for an up/down converter in discontinuous conduction, 7 11.4.3 Linearizing the We have already shows by averaging the switch modeled as LTT and ar the key step in lineariz: linearize the averaged are LTI, they are prese: Linearization of th out. Let’s denote nom nominal by the superse solution corresponds. to this for the lineatizatio: ae and similarly for the Fig, 11.10(b), namely, the perturbations. That the small perturbations. perturbations in the sou aut ‘The results of this calcul with an equivalent repr: Figure 11.13. (a) Linearize‘odels for the s periodically circuit. ‘onduetion us conduction allows the can therefore Vig to denote veraged quan circuit acts as 113) Vien our sign + assume for a+ T(t) and I Linearizing linear terms, (ida) g (11.13) 10 (115) Section 11.4 Linearized Models 277 11.4.3 Linearizing the Averaged Switch We have already shown how to average a high-frequency switched de/de converter by averaging the switch in the canonical cell. The nonswitch components are usually ‘modeled as LTI and are therefore not changed for the averaged model. Similarly, the key step in linearizing the nonlinear averaged circuit for stich a converter is t0 linearize the averaged model of the switch. If the remaining nonswitch components are LTI, they are preserved intact in the linearized circuit. Linearization of the averaged-switch model in Fig. 11.10(b) is easily carried ut. Let's denote nominal values by uppercase letters and deviations from the nominal by the superscript ~, as in Example 11.8. Most commonly the nominal solution corresponds to a constant or periodic steady state, but we do not require this for the linearization that follows, We can write dit)=D+dit) and a(t) =D’ -dit) (11.16) and similarly for the other variables. We now expand the source terms in Fig. 11.100), namely, dit)t_(@® and d(t)®,,(), to first-order or linear terms in the perturbations. That is, we neglect terms that involve squares or products of the small perturbations. This linearized expansion yields the following first-order perturbations in the source terms: (tii, (t) — DI, = Di,(t) + I,dit) 1147) dit), (t) — DV,. = Di, (t) + V,.dit) foe ‘The results of this calculation are represented in the linearized circuit in Fig. 11.13(a), with an equivalent representation in Fig 11.13(b).278 Chapter 11 Dynamics and Control: An Overview | ¢ be Figure 11.14 Lin with determining th provides a good star ' : in Example 113 led : Example 1. Figure 11.13 (cont.) (b) Alternative representation, The transfer F of having an LTI ¢ to apply the extens the study of deviat EXAMPLE 11.9 and techniques inc Linearized Circuit for an Up/Down Converter in Continuous Conduction ‘equivalents and the ‘We retum once more tothe up/down converter of Examples 11.1, 1.3, and 11.7 to obtain a ities, and so on. F linearized averaged model that describes small perturbations from the nominal steady state Example 11.9, we inthe continuous conduction mode. as well as various « ‘The linearized version of Fig. 11.11 is obtained by replacing all voltages and currents in describing how by their perturbations from the nominal and replacing the averaged canonical cell by its Inid the eeu linearized venion, using Fig. 1130). All other converter elements are liner and are ae therefore preserved intact in the linearized circuit. The result is shown in Fig. 11.14, with ee fs bility analysis and a representing deviations of the averaged input source from its nominal e value ae “The linearized model corresponding to the constant nominal steady state is an LTT control designs. Fe cireuit. Hence solving for the values of all circuit variables is straightforward (for example, obtained by linearis using impedance methods). For control design purposes, we must know the transfer function uous conduction. I from perturbations d in the duty cycle, which constitutes our control input, to perturbations: to zero faster if adc 2, in the output voltage, Setting 1, = 0, a straightforward computation with the circuit in Adding a physical « Fig. 11.14 shows the transfer function t0 be | is possible to repro: - control Dols) _ (Le = Wg/LIy) 7 tL inl Eta). (1118) Note that if we dis) (@)s +U/RO)s + (D?/L0} ua ae (We have used the steady-state relationships in (11-7) wo simplify the constant in the numer- replacing the curret stor.) This isthe transfer function we referred to at the end gf Example 11,7, in connection effect of replacing ¢17 to ob lal steady state and currents cal cell by its linear and are 1114, with value ate is an LTT (for example, sie funetion perturbations the circuit in (118) in the numer Section 11,4 Linearized Models. 279 Figure 11.14 Linearized averaged model of an up/down converter. ‘with determining the effect of feedforward in Example 11.1. This transfer function also provides a good starting point for an explanation of why the proportional feedback scheme in Example 11.3 led to the behavior observed there. We pursue this explanation further in Example 11.11 ‘The transfer function computation in Example 11.9 illustrates the advantages of having an LTI circuit model for power circuit dynamics. Such a model allows us to apply the extensive array of concepts and techniques available for LTI circuits to the study of deviations from nominal operation in a power circuit. These concepts and techniques include superposition, impedance methods, Norton and Thévenin equivalents and their multiport generalizations, network interconnections, sensitiv ities, and so on, For instance, in addition to the transfer function computed in Example 11.9, we could determine input and output impedances or admittances, as well as various other relevant transfer ratios. These transfer ratios are important in describing how dynamic behavior depends on characteristics of the source, the load, the circuit itself, and the control design. ‘Apart from serving as a basis for computations that support the tasks of sta bility analysis and control design, a circuit model can sometimes directly suggest control designs. For example, consider again the circuit in Fig. 11.12, which we obtained by linearizing the averaged model of the up/down converter in discontin- uuous conduction. Its damping would evidently be increased and #, would decay to zero faster if additional conductance were placed in parallel with the capacitor. ‘Adding a physical conductance is clearly ruled out on grounds of efficiency, but it is possible to reproduce the effect of additional conductance entirely by means of contro Note that if we use proportional feedback to enforce dit) = hi, (t), where h is 4 constant, the effect is the same as that of a conductance of value hV,,/2T7RL replacing the current source in Fig. 11.12. Proportional-integral conttol has the effect of replacing the current source with a parallel combination of a conductance280 Chapter 11 Dynamics and Control: An Overview and an inductance. This inductance causes the steady-state value of 2, to be zero even in the presence of parameter errors and constant disturbances (such as constant deviations of the input voltage or converter inductance from their nominal values), Reference — ‘Thus using a circuit model to approach or interpret control design can often provide * cor useful insights, = 11.5 FEEDBACK CONTROL We hive shown how fo obin both noninear and near cut models fo the | dynamics of averaged varies in cenan clases of power cuir How So we tse these mols 0 analyze and design feedback contol? | ICs generally ard to assess sabi and o design or evate feedback contol schemes drety with noninear models In this Seton we fu on i Lo modes sich a tase we obi by linearization at consent operating Po With LTT mols, we can choose fom a wide range of systema anagee tod design apraces fo feedback contol The sonvoles designed ating LT mls tre sually LTT a well Figure 11.15 Block di Controllers derived from linearized models cannot be guaranteed to provide directly applicable t satisfactory operation for large deviations from nominal operation, even if we ex. of power eicuits Peet them to funtion wel for small perturbations. Also, the contol design has to | Recent control account for the fact that the linearized model will vary with the operating condition, since the eatly 196) Nevertheless, the majority of power circuit controllers are in fact designed on the imerconnected mult basis of linearized models. A controller that performs well on the linearized model can do no more th is likely to keep the circuit operating near the nominal and to make the operation | configuration in Fig relatively insensitive to small disturbances and errors. We can therefore reasonably | have withstood the expect thatthe nonlinearities have only secondary effects, except when there are, control approactes. ‘major disturbances. We usually deal with nonlinear effects through refinements, ‘modifications, or complements to a core LTI controller. For these reasons, we confine ourselves in this overview to discussing feed- ‘back control design issues in the context of “classical” control with continuous-time ‘The quantity mi relates the output of frequency or impede all the transfer funct LTI models. This allows us to use simple transfer function computations but still functions of s. This provides ample opportunity to appreciate the potential benefits and pitfalls of feed~ We must step outsic back control. Further consideration of control design using such models, as well as | time delays; a time ¢ state-space models and discrete-time models, is left to Chapter 14, Most of Chapter | nonrational transfer { 14 is accessible directly after Chapter 11 | To keep our not | main representations 1.5.1 The Classical LTI Control Configuration | tively, whenever nece Considerable insight into what feedback can accomplish—and an understanding is an abuse of notatio | inthe expression for oof what the dangers are—may be obtained by considering the interconnection of | single-input, single-output, continuous-time LTI subsystems in Fig. 11.15. The sim- for the transform, but ple feedback configuration here is of the same form as Fig. 11.1, and lies at the averages, perturbatior hheart of classical control design. The lessons you leam feom this configuration are working in. We alway2, to be zero cch as constant ‘minal values) often provide \odels for the How do we tate feedback focus on LTT erating point, analysis and g LTI models x4 to provide ven if we ex- design has 10 ng condition. signed on the arized model the operation re reasonably ven there are refinements, cussing feed- ttinuous-time ions but still falls of feed- Is, as well as st of Chapter derstanding onnection of 15. The sim- lies at the guration are Section 11.5 Feedback Control 281 Disubance ; Reference como! vipat et _| compensator | ut] Soi E ) be oe 7 7) = 1 8 20) oie Neergeegpegarsie Figure 11.15 Block diagram of the classical LT! feedback configuration directly applicable to controlling the small-signal dynamic behavior of a wide range of power circuits, Recent control approaches are rooted in the state-space methods developed since the early 1960s and are used to tackle much more general problems, with imerconnected multiinput, multi-ourput systems. In Chapters 12, 13, and 14 we can do no more than hint at these relatively recent extensions. For the simple configuration in Fig, 11.15, however, the design approaches of classical control have withstood the test of time and provide a demanding standard for modem control approaches. The quantity marked in each block of Fig, 11.15 is the transfer function that relates the output of the block to its input—in the Laplace transform domain (or frequency or impedance domain). Except where otherwise stated, we assume that all the transfer functions we work with are ratios of polynomials in s, of rational functions of s. This case is the most important one for control design. On occasion we must step outside this category of transfer functions, as when representing time delays; a time delay of 7 between an input and an output corresponds to the nonrational transfer function e-"* To keep our notation streamlined, we denote time domain and transform do- ‘main representations of a signal such as u in Fig. 11.15 by u(t) and w(s), respec- tively, whenever necessary to make the domain explicit. However, this convention is an abuse of notation, because u(s) is not obtained through replacement of t by $ in the expression for u(t). (An unambiguous notation would, for example, be f(s) for the transform, but such symbols are awkward to combine with our notation for averages, perturbations, and so on.) The context makes clear which domain we are working in, We always jvrite transfer functions such as 9(s) with their arguments.282 Chapter 11 Dynamics and Control: An Overview Figure 11.15 represents the controlled system or plant as having a transfer function 9(s) + 9(s), which relates the output y to the control input 1, Here 9(s) constitutes the nominal model of the plant, which is the basis for the control design, and G(s) represents errors in the nominal model. These errors could, for example, reflect uncertainties regarding the load and other circuit parameters or simplifications, approximations, and compromises made during modeling. ‘The disturbances affecting the plant are represented in terms of their effects at the plant output, by means of the signal 5. For simplicity, we assume that none of the The average eur Atering the nstnan sT/2). This wernge Spe the fring an ofthe reir he "We kno fom and the avenge ny ofthe snasoial vo in amo comps a disturbances can be measured, 50 no feedforward is possible. The feedback signal Wwe have already note consists of a measurement of the system output, y, passed through a sensor whose Example 11.4, thar transfer function is p(s), and corrupted by measurement noise, n. This feedback the transition from its signal is compared with the external signal, r, which represents the reference (or These facts lead desired or commanded) value for y; we would ideally like to have r = y. The result dynamics of the phase (of the comparison is the noise corrupted error signal, , To keep things simple, we steady-state character shall omit consideration of sensor dynamics, so p(3) = 1. The control signal 10 the converter, This apy the system do not var constructed in situation continuous time mode satisfactory analysis of the plant, namely u, is produced by an LTI controller or compensator with transfer function h(s) that acts on EXAMPLE 11.10 oD Welara sede Feedback Control of Armature Current in a Controlled Rectifier Drive transfer function e~* as (I~ sT/4)/(1 +67 matched by correspond ‘The basic purpose the desired and actual a the eror signal tothe ¢ dynamic LTT section an characteristic of the co: 5.1, permits us to analy iE (of the controller shown The control structure shown in Fig. 11.16 is based on a proportional-integral (Pl) control design for the controlled rectifier drive treated in Examples 11.2 and 11.4, The plant transfer function 1/(sL + R), which relaes the averaged signals in the armature circuit, is simply the input admittance ofthe averaged circuit in Fig, 11.6(0) of Example 11.4. We could have ‘modeled the “disturbance” E as adding in directly atthe plant input. Instead, we represent it by adding an equivalent signal at the plant output, to make the configurations in Figs. HS and 11.16 look similar proportional gain and A 2 open ape Controller Controlled rectifier Jeo PL Inverse Static Armature “ a controller characteristic] | characteristic Delay circuit b and 2. to the oxiput the relationships rope - ~ hel Re Ls def a “ loop tansfer uncton Ors opel acee feedback op ame tke our assump here 1 v 1) Figure 11.16 Feedback control for a phase-controlled regtifier drive. We can write (11.19) sing a transfer u. Here 9(s) ve the control ors could, for parameters or aling. heir effects at at none of the edback signal sensor whose This feedback reference (or xy. The result 2s simple, we trol signal to with transfer Drive ‘al (PD) control = plant ransfer ‘uit, is simply We could have |, we represent ations in Figs iG x |e fd +R Section 11.5 Feedback Control 283 ‘The average current #,(¢) is usually obtained with sufficient accuracy in practice by Aitering the instantaneous current i(t through 2 low-pass filter with ransfer function 1/(1-+ sT/2), This average current is compared with J,j- Based on the discrepancy, the controller specifies the fring angle a for the phase-controlled rectifier. The output voltage waveform of the rectifier is thereby determined and hence so is its average value Ty). ‘We know from Chapter 5 that the steady-state relationship between the firing angle a and the average output voltage T,(t) is (v4) = 2V/m)cosa, where V is the amplitude Of the sinusoidal voltage source. Any transients in the average voltage, however, depend in @ more complicated way on variations in the fring angle of the rectifier. For instance, 1We have already noted when discussing the open-loop response to'8 step in fring angle in Example 11.4, that it takes one cycle of duration T for the average output voltage to make the transition from its inital to its final value ‘These facts lead to an approximate nominal model that is often used to represent the «dynamics of the phase-controlled rectifier. The model comprises a cascade ofthe converter's steady-state characteristic and a dynamic block to represent a (mean) time delay of 7/2 in the converter. This approximation of converter dynamics is reasonable when the signals in the system do not vary significantly during intervals of length T, Such models are often constructed in situations where sampling effects need to be approximately represented in a continuous time model. Sometimes, however, the approximation is too crude to permit a satisfactory analysis ofthe system, and thete is then litle choice but to go toa sampled-data model ‘We have used the rational transfer function 1/(1-+ 87/2) to approximate the nonsational transfer function e~*7/? of the time delay. Better approximations of this exponential, such as (I ~ sT/4)/(1 + 87/4), may be used, but the refinement is probably not justified unless ‘matched by corresponding refinements elsewhere in the model ‘The basic purpose ofthe controller is to vary T(t) in accordance with the eror between. the desied and actual average armature currents, In order to obtain a linear relationship from the error signal to the average voltage, the controller may be constructed as a cascade of dynamic LT! section and a static nonlinearity that represents the inverse of the steady-state