slides in PDF - People(dot)tuke(dot)sk
Transcription
slides in PDF - People(dot)tuke(dot)sk
Figure 3.2: P ID con 3.2: P IDfrom controller: points to plane. Figure 3.2: PFigure ID controller: points tofrom plane. nfractional arecommend wide class the of oP For points a wide of controlled objectsFor wethe Dδ a wide class of controlled objects we recommend fractional P Icontrolled Figure 3.2: P IDFor controller: from toclass plane. CHAPTER 3. FRACTIONAL-ORDER CONTROLLERS 13 δ which is a particular ca δ controller, controller, which is case a particular of P I λ Dwhere controller, controller, which is a particular of P I λ Dcase controller, λ = n, where n∈ λ=n n δ N and ∈ R. Integer order inte δ ∈ order R. the Integer order is for error steady-state N andobjects δ ∈N R. Integer integrator isIintegrator important forδimportant steady-state D For a wide class of controlled weand recommend fractional P cancellation on the other hand but on thethe other hand fractional integral is also cancellation but other hand integral is alsobut important for importa controller, which is a particular case ofcancellation P I λon D δthe controller, where λ fractional = n, n ∈the N and δ ∈ R. Integer order integrator is important fortransfer steady-state errorresponse obtaining aresponse Bode’s loop trans obtaining a Bode’s ideal loop function with constantideal phase obtaining a Bode’s ideal loop transfer function with constant cancellation but on the other hand for thedesired fractional integral is also important margin range (e.g. [1,range 3, for 23,(e.g. 51]).[1, 3, margin for23, desired marginfrequency for desired frequency 51]). frequency range obtaining a Bode’s ideal loop transfer function response with constant phase Figure 3.2: P ID controller: from points to plane. margin for desired frequency range (e.g. [1, 3,H.W. 23, Bode, 51]). Network Analysis and Feedback Amplifier Design. Bode’s ideal loop transfer function Fractional order systems and controllers 3.2.2 ideal shape the shape loop transfer function in his work Bodeansuggested an of ideal of the loop transfer function in his of wo Bode suggested an idealonshape Bode’s ideal Bode loopsuggested transfer function ideal loopIdeal transfer function: design of feedback amplifiers in amplifiers 1945. transfer function hasamplifiers form: designBode’s of feedback inloop 1945. Ideal of loop transfer function in has19fo design feedback Bode suggested an ideal shape of the loop transfer function in his work on ! "α ! "α s s design of feedback amplifiers in 1945. Ideal loop transfer function , = (3.6) L(s) = has form: 3.2.2 Bode’s ideal loop transfer function , L(s) L(s) ! "α ω gc ωgc Bode suggested an ideal shape s of the loop transfer function in his work on design ofL(s) feedback=amplifiers in 1945. Ideal loop transfer function has form: , (3.6) Hendrik Wade Bode frequency and α is slope the ideal where ωgc iswhere "α ωdesired gc and slopecut-off of the ideal c ω!gccrossover is isdesired crossover freq where ωof s is desired crossover frequency (1905-1982) gc α , (3.6) L(s) = ωgc characteristic. characteristic. characteristic. frequency and α is slope of the ideal cut-off where ωgc is desired crossover where ωgc is desired crossover frequency and α is slope of the ideal cut-off Phase marginPhase is Φmmargin = π(1 + all+values The amplitude isα/2) Φm =forπ(1 α/2) of forthe all gain. values of the characteristic. characteristic. Phase margin is Φgain. π(1 +amp α/ m = The o o Phase margin = infinity. π(1 + α/2) for all values of the gain. The amplitude The constant phase marginphase 60o , 45margin and 30 margin AismΦfor mis ocorrespond o Phase margin is Φm =margin π(1 + α/2) all values of the gain. The amplitude is infinity. The constant 60 , 45o and corre margin A o o o m is infinity. The30 constan margin A Am is infinity. The constant phase margin 60 , 45 and 30 correspond m the αslopes α margin =and−1.33, o and −1.66. to to the slopes = −1.33, −1.66.60o−1.5 constant phase and 30o correspond margin Am is infinity. The to−1.5 the slopes ,α45= −1.33, −1.5 and −1.66. to the slopes α = −1.33, −1.5 and The Nyquist curve for ideal Bode transfer function is simply a straight line Nyquist curveαπ/2 for(seeideal Bode transfer function is simply a straight line to the slopes α = −1.33, through −1.5 The and −1.66. the origin with arg(L(jω)) e.g. curve [1, 39]). for ideal Bode transfer function is simply a straig The= Nyquist The Nyquist curve for ideal Bod Bode’s transfer function (3.6) can be used as a reference system in the (see e.g. [1, 39]). through the origin with arg(L(jω)) = απ/2 The Nyquist curve forfollowing ideal form: Bode transfer function is simply straight line = απ/2 (see e.g. [1, through the origin witha arg(L(jω)) 39]). arg(L(jω)) through the origin4 with K function (3.6) can be used as a reference system in the Bode’s transfer (0 [1, < α <39]). 2) Gc (s) = (see through the origin with arg(L(jω)) = απ/2 e.g.transfer Bode’s function(3.7) (3.6) can be used transfer as a reference system sα + K Bode’s function (3.6) following form: K Bode’s transfer function (3.6) can be used a< αreference system(3.8)in the Go (s) = α as(0 < 2), following form: s following form: K following form: (s) is transfer function in(0 <K where Gc is transfer function of closed loop α < 2) (3.7) Gcand (s)Go= K +GKcare: (0 < α < 2) (s) = α K characteristics of Bode’s ideal transfersαfunction open loop. General Gc (s) = α (0 < α < 2) (3.7) Gc (s) = α s +K s + K s +K Go (s) = α (0 < αK< 2), (3.8) K K s Go (s) = α (0 < α < 2), Go (s) = Go (s) = α (0 < α < 2), (3.8) s s where sGc is transfer function of closed loop and Go (s) is transfer function in transfer function of closed loopGand Go (s) is function transfer funct where GcGiso (s) of loop. closedGeneral loop and is transfer function in where where Gc is transfer function is transfer of cl open characteristics of Bode’s ideal transfer are: cfunction open ideal loop.transfer General characteristics ofopen Bode’s ideal transfer function are: open loop. General characteristics of Bode’s function are: loop. General characteristics 3.2.2 1 Bode’s loopideal transfer 3.2.2 ideal Bode’s loopfunction transfer 3.2.2 function Bode’s ideal loop δ For a wide class of controlled D. objects recommend Company, the fractional Inc., P I n DNew Van we Nostrand York, 1945. controller, which is a particular case of P I λ D δ controller, where λ = n, n ∈ N and δ ∈ R. Integer order integrator is important for steady-state error cancellation but on the other hand the fractional integral is also important for obtaining a Bode’s ideal loop transfer function response with constant phase margin for desired frequency range (e.g. [1, 3, 23, 51]). Fractional order systems and controllers Fractional-order systems Bode’s ideal loop transfer function CHAPTER 3. FRACTIONAL-ORDER CONTROLLERS CHAPTER 3. FRACTIONAL-ORDER CONTROLLERS CHAPTER 3. FRACTIONAL-ORDER CONTROLLERS W(s) W(s) W(s) + E(s) U(s) Gc(s) G(s) = Y(s) CHAPTER 3. FRACTIONAL-ORDER CONTROLLERS Gs(s) βn an s an D y (t ) + an−1D • Crossover frequency: a function of K; π 2 Start !! ! β0 y (t ) + ! + a0 D y (t ) = u (t ) 14 General characteristics of the Bode’s ideal line transfer function: • Phase: horizontal of −α ; 3.3: Bode’s ideal loop. •Figure Magnitude: constant slope of −α20dB/dec.; 1 + an −1s βn −1 + ! + a1s β1 + a0 s β0 βn −1 K s! Figure 3.3: Bode’s ideal loop. (a) Open loop: βn G(s) = 14 Y(s) Y(s) K (a) Open loop: G(s) = Figure s ! 3.3: Bode’s ideal loop. • Magnitude: constant slope of −α20dB/dec.; W(s) – Gs ( s ) = Y(s) K s! 14 "" " 63 / 90 Back Full screen Close End • Nyquist: straight line at argument −α π2 . • Crossover frequency: a function of K; (a) Open loop: • Phase: horizontal line of −α π2 ; (b) Closed loop: • Magnitude: constant slope of −α20dB/dec.; Figure • Nyquist: straight line at argument −α π2 . 3.2: P ID controller: from points to plane. • Gain margin: Am = infinite; " ! • Phase margin: constant : Φm = π 1 − α2 ; • Crossover frequency: a function of K; • Phase: line of −α π2 ; (b)horizontal Closed loop: • Step response: y(t) = Ktα Eα,α+1 (−Ktα ) , where Ea,b (z) is Mittag-Leffler function of two parameters [43]. a wide class of controlled objects we recommend the fractiona = infinite; • Gain margin: AmFor " ! λ δ α controller, ;a particular case of P I D controller, where λ = • Phase margin: constant : Φmwhich = π 1 −is 2 3.2.3 Illustrative example • Gain margin:• AStep α α m = infinite; response: y(t) = Kt E (−Kt ), α,α+1 N and δ ∈ R. Integer order integrator is important for steady-sta " ! α We can illustrate the fractional order control properties by an example. As; • Phase margin: where constant : Φ = π 1 − m function of two parameters [43]. Ea,b (z) is Mittag-Leffler 2 • Nyquist: straight line at argument −α π2 . (b) Closed loop: cancellation but on the other hand the fractional integral is also impo 5 (3.9) with consta obtaining a Bode’s ideal loop transfer function response margin for desired frequency range (e.g. [1, 51]). with J being the payload inertia. Our specification is 3, the 23, constant phase We can illustrate the fractional order control properties by an example. AsIllustrative example suming that the transfer function of the DC motor is • Step response: y(t) = Ktα Eα,α+1 (−Ktα ) , Km where Ea,b (z) is Mittag-Leffler function of two parameters [43]. , G(s) = Js(s + 1) 3.2.3 Illustrative example 2 3.2.3 margin independent of the payload changes. suming that the transfer function of the DC example. motor that isAs-we would like to have a closed loop system that is insensitive We can illustrate the fractional order control properties by an Assume to gain variations with a constant phase margin of 60o. Bode’s ideal loop suming that the transfer function of the DC motor is K m (3.9) G(s) = transfer, function that gives this phase margin is Js(s + 1) Km , (3.9) G(s) = 1 Js(s + 1) . (3.10) = √ Go (s) with J being the payload inertia. Our specification is the constant phase s3s with J beingmargin the payload inertia. Our specification is the constant phase independent of the payload changes. margin independent of the payload changes. Assume that we would like to have a closed loop system that is insensitive Assume that we would like to have a closed loop system that is insensitiveo to gain variations with a constant phase margin of 60 . Bode’s ideal loop to gain variations with a constant phase margin of 60o. Bode’s ideal loop α transfer function that margin gives this transfer function that gives this phase is phase margin is 3.2.2 Bode’s ideal loop transfer function Bode suggested an ideal shape of the loop transfer function in his design of feedback amplifiers in 1945. Ideal loop transfer function has ! " s , L(s) = 1 1 ωgc (3.10) Go (s) = √ . (3.10) G (s) = √ . o s3s s3s where ωgc is desired crossover frequency and α is slope of the idea characteristic. Phase margin is Φm = π(1 + α/2) for all values of the gain. The am margin Am is infinity. The constant phase margin 60o , 45o and 30o co to the slopes α = −1.33, −1.5 and −1.66. The Nyquist curve for ideal Bode transfer function is simply a stra through the origin with arg(L(jω)) = απ/2 (see e.g. [1, 39]). transfer function can be used as a reference system Can be used as a Bode’s reference system in (3.6) the form: CHAPTER 3. FRACTIONAL-ORDER CONTROLLERS 14 following form: Y(s) W(s) K K (0 < α < 2) Gc (s) = α G(s) = s s +K K Go (s) = α (0 < α < 2), Figure 3.3: Bode’s ideal loop. CHAPTER 15 s CHAPTER 3. 3. FRACTIONAL-ORDER FRACTIONAL-ORDER CONTROLLERS CONTROLLERS 15 (a) Open loop: where Gc is transfer function of closed loop and Go (s) is transfer fun • Magnitude: constant slope of −α20dB/dec.; open loop. General characteristics of Bode’s ideal transfer function a • Crossover frequency: a function of K; PI!Dµ controllers Bode’s ideal loop transfer function ! 20 20log log1010 |G(j |G(j!)| !)| --20 20 "" dB/dec dB/dec • Phase: horizontal line of −α π2 ; • Nyquist: straight line at argument −α π2 . (b) Closed loop: log log !! arg arg((G(j G(j!)) !)) ## • Gain margin: Am = infinite; 22 " ! • Phase margin: constant : Φm = π 1 − α2 ; ## "" • Step response: y(t) = Ktα Eα,α+1 (−Ktα ) , 22 where Ea,b (z) is Mittag-Leffler function of two parameters [43]. ## log log !! 3 3.2.3 Illustrative example Figure Figure 3.4: 3.4: Bode Bode plots plots of of transfer transfer function function (3.8). (3.8). We can illustrate the fractional order control properties by an example. Assuming the transfer function of the motor transfer istransfer function Since (s) C(s)G(s), we the controller Since G Goothat (s) = = C(s)G(s), we can can find find theDC controller function in in the the following following form form ! " ! G(s) = " " Km ! !, " (3.9) 11 11 JJ 2/3 + 2/3 + Js(s + 1) ss2/3 C(s) ss2/3 = K ,, (3.11) + 1/3 + C(s) = = = K (3.11) 1/3 K ss1/3 ss1/3 Kmm with Jis abeing the payload inertia. Our specification isλ the constant phase δ which D δ controller controller (3.1), (3.1), which is a particular particular case case of of the the transfer transfer function function of of the the PPII λD marginKindependent of the payload changes. where where K = = J/K J/Kmm isis the the controller controller constant. constant. Assume that we would like to have system a closed loopa system that is insensitive The loop The phase phase margin margin of of the the controlled controlled system with with a forward forward loop controller controller o to gain isis variations with a constant phase margin of 60 . Bode’s ideal loop G (s) Gcc(s) transfer function that gives this phase margin = arg[C(jω (3.12) Φ )G(jω0 )])]is+ +π, π, (3.12) Φm = arg[C(jω0 )G(jω m 0 0 where 1 where ωω00 isis the the crossover crossover frequency. frequency. . Go (s) = √ The s3s The obtained obtained phase phase margin margin isis ## $$ 11 44ππ ππ = Φ + = .. = arg[C(jω)G(jω)] arg[C(jω)G(jω)]+ +ππ = = arg arg Φmm = +ππ = = ππ − − 4/3 (jω) 33 22 33 (jω)4/3 (3.10) (3.13) (3.13) 6 • Nyquist: straight line at argument −α . 2 α • Step response: y(t) = Ktα Eα,α+1 (−Kt ), (b) Closed loop: where Ea,b (z) is Mittag-Leffler function of two parameters [43]. 3.2.3 • Gain margin: Am = infinite; " ! • Phase margin: constant : Φm = π 1 − α2 ; • Step response: y(t) = Ktα Eα,α+1 (−Ktα ) , Illustrative example where Ea,b (z) is Mittag-Leffler function of two parameters [43]. We can illustrate the fractional order control properties by an example. Assuming that the Illustrative transfer function of the DC motor is 3.2.3 example PI!Dµ controllers Bode’s ideal loop transfer function: example We can illustrate the fractional order control Kmproperties by an example. As,is G(s) of=theofDC suming that the transfer transfer function motormotor The function a DC is (3.9) Js(s + 1) G(s) = Km J is payload inertia , (3.9) Js(s 1) Our+specification is the constant with J being the payload inertia. phase margin independent of the payload changes. with J being the payload inertia. Our specification is the constant phase Assume that we would like to have a closed loop margin independent of the payload changes. Assume that we would like to have a closed systemwith thata is insensitive that is to loop gainloop variations Assumesystem that we would likeinsensitive to have a closed system thatois insensitive . Bode’s ideal loop to gain variations with a constant phase margin of 60 o constant margin 60 . Bode’s Bode’sloop ideal loop to gain variations withphase a constant phaseofmargin of 60o. ideal transfer function that gives this phase margin is CHAPTER 3. FRACTIONAL-ORDER CONTROLLERS 15 transfer function that function gives this phase is transfer that margin gives this phase margin is 20 log 10 |G(j !)| 1 Go (s) = √ 3 1. s s - 20 " dB/dec (3.10) . Go (s) =s s√ CHAPTER 3. FRACTIONAL-ORDER 3 CONTROLLERS Fractional-order PID controllers: from points to plane CHAPTER 3. FRACTIONAL-ORDER CONTROLLERS (3.10) 15 CHAPTER 3. FRACTIONAL-ORDER CONTROLLERS 15 20 log 10 |G(j !)| Figure 3.2: P ID controller: from points to plane. 20 log 10 |G(j !)| CHAPTER 3. FRACTIONAL-ORDER 20 log 10 |G(j !)| - 20 " dB/dec CONTROLLERS 15 7 10 For a wide class of controlled objects we recommend the fractional P I n D δ controller, which is a particular case of P I λ D δ controller, where λ = n, n ∈ N and δ ∈ R. Integer order integrator is important for steady-state error cancellation but on the other hand the fractional integral is also important for obtaining a Bode’s ideal loop transfer function response with constant phase margin for desired frequency range (e.g. [1, 3, 23, 51]). - 20 " dB/dec log ! arg ( G(j !)) # 2 - 20 " dB/dec log ! arg ( G(j !)) # " # 22 log ! arg ( G(j !)) ## # " ( G(j arg !)) 2 # 2 2 log ! 3.2.2 log ! # 2 # " 2 log ! Since Go (s) = Figure C(s)G(s), can findofthe controller transfer 3.4:# we Bode plots transfer function (3.8). function in the following form # log ! " !log ! " ! Since Go (s) = C(s)G(s), find1 the controller transfer 1 function in the J we can C(s) = s2/3 + 1/3 = K s2/3 + 1/3 , (3.11) following form Km s s Figure 3.4:!plots Bodeof plots transfer function (3.8). " of function ! " Figure 3.4: Bode transfer (3.8). 1 function of Jof the2/3 2/3the P1I λ D δ controller (3.1), which is a particular case transfer C(s) = s + 1/3 = K s + 1/3 , (3.11) Kcontroller s controller transfer s function in the m where is thewe constant. Since GK (s)==J/K C(s)G(s), can find the PI!Dµ controllers Bode’s ideal loop transfer function: example Sinceo Go (s) Since =mC(s)G(s), , we cancontroller find the controller transfer function the transfer function is in the (3.1), which a particular of the transfer function of theaPforward I λ D δ controller The is phase margincase of the controlled system with loop controller following form following form ! " ! " K = J/Km is the constant. Gwhere "2/3 " 1! J controller c (s) is 2/3 ! 1 + + C(s) = s = K s , (3.11) 1 1 controller J controlled The phase margin with a forward 2/3 )G(jω )] K +s1/3 π, (3.12) (3.11) Kmof ss1/3 m = arg[C(jω 0= + 0system C(s) =Φthe s2/3 + loop , Km s1/3 s1/3 Gc (s) is which is ω a 0particular case of the transfer function of the P I λ D δ controller (3.1), where is the crossover frequency. )G(jω0 )] + π, (3.12) Φm = this a 0particular case of fractional PID (3.1), which is a particular case of arg[C(jω the transfer function of the P I λ D δ controller where K= J/K the controller constant. m is phase The obtained margin is is #constant. $ where ωK0 is the crossover frequency. The phase system where =margin J/K thecontrolled controller o a forward loop mofisthe π 1with 4 πcontroller The phase margin is arg = . loop (3.13) arg[C(jω)G(jω)] π= + π with = π −a forward Gc (s)Φis m =obtained The phase margin of +the controlled system controller 32 3 # (jω)4/3 $ Φm = arg[C(jω0 )G(jω10 )] + π, π(3.12) 4π (s) is G = . (3.13) = arg[C(jω)G(jω)] + π = arg Φ + π = π − c m The constant phase margin is not dependent changes and the 4/3 of the payload (jω) 3 2 3 where ω0 is the crossover frequency. )] + π, (3.12) Φm =isarg[C(jω 0 )G(jω system gain K and phase curve a horizontal line0at −2π/3. The constant obtained phase margin The margin isis not dependent of the payload changes and the # loop can $ be expressed as: Step response of closed control where theand crossover frequency. 0 is K system ωgain phase curve is a horizontal line at −2π/3. π 1 4π % = arg[C(jω)G(jω)] +1πmargin = arg& is 4/3 + π = π − ' = . ( (3.13) ΦmThe obtained phase Step control loop canEbe expressed as:1+1/3 (jω) 3 2−t 3 , (3.14) y(t)response = L−1 %of closed = t1+1/3 # 1+1/3, 2+1/3 $ 1+1/3 + 1) & s (s ' ( 4 πthe π The constant phase of the1 payload changes 1 not dependent −1 margin is 1+1/3 1+1/3 and , (3.14) = L = targ E1+1/3, 2+1/3+ −t = . (3.13) =Kresponse arg[C(jω)G(jω)] + π = π = π − my(t) 1+1/3 4/3 whereΦgain step is independent of the payload inertia and α = 4/3. system and phase curve is a horizontal line at −2π/3. s (s + 1) (jω) 32 3 8 Step response of closed independent control loop can be payload expressed as: where step response of the inertia α = 4/3. The constant dependent of the and payload changes and the %phaseis margin & is not ' 1+1/3 ( 1 (3.14) y(t) = L−1 K and = t1+1/3 system gain phase curve is aE1+1/3, horizontal line at, −2π/3. 2+1/3 −t s (s1+1/3 + 1) The obtained phase margin is 60 : Step response: Step response of closed control loop can be expressed as: ' ( 1 y(t) = L−1 = t1+1/3 E1+1/3, 2+1/3 −t1+1/3 , s (s1+1/3 + 1) where step response is % independent of the & payload inertia and α = 4/3. Bode’s ideal loop transfer function Bode suggested an ideal shape of the loop transfer function in his work on design of feedback amplifiers in 1945. Ideal loop transfer function has form: ! "α s , (3.6) L(s) = ωgc Figure 3.4:# Bode plots of transfer function (3.8). " 13 where ωgc is desired crossover frequency and α is slope of the ideal cut-off characteristic. Phase margin is Φm = π(1 + α/2) for all values of the gain. The amplitude o margin 60of , unit 45o and 30o correspond margin Am is infinity. The constant phaseComparison step responses to the slopes α = −1.33, −1.5 and −1.66.of the fractional-order “reality” and its integer-order “model” The Nyquist curve for ideal Bode transfer function is simply a straight line through the origin with arg(L(jω)) = απ/2 (see e.g. [1, 39]). Bode’s transfer function (3.6) can be used as a reference system in the following form: K (0 < α < 2) (3.7) Gc (s) = α s +K K Go (s) = α (0 < α < 2), (3.8) s where Gc is transfer function of closed loop and Go (s) is transfer function in open loop. General characteristics of Bode’s ideal transfer function are: (3.14) where step response is independent of the payload inertia and α = 4/3. PI!Dµ controllers PI!Dµ controllers Control of the “reality” and the “model” using a classical PD controller, which is optimal for the “model” 9 ss being developed is the method of dominant roots [37], based on the given stability measure and the damping ratio of the closed control loop. Assuming that, the desired dominant roots are a pair of complex conjugate root as follows: s1,2 = −σ ± jωd , PI!Dµ controllers designed for the damping ratio ζ and natural frequency ωn . The damping constant (stability measure) is σ = ζωn and the damped natural frequency of ! oscillation ωd = ωn 1 − ζ 2 . The design of parameters: Kp , Ti , λ, Td and δ can be computed numerically from characteristic equation. More specifically, for simple plant model P (s), this can be done by solving PI!Dµ controllers: design min Kp ,Ti ,λ,Td ,δ Control of the “reality” and the “model” using a classical PD controller, which is optimal for the “model”, after detuning (aging) of the controller (when TD = 1). ||C(s)P (s) + 1||s=−σ±jωd . The design of PI!D" controllers can be based on gain Another possible way to obtain the controller parameters is using the tuning and phase margin specifications: formula, based on gain and phase margins specifications [52]: " [C (jωp )] " [P " [C (jωp )] # [P " [C (jωg )] " [P " [C (jω )] # [P g (jωp )] − # [C (jωp )] # [P (jωp )] = − A1m , (jωp )] + # [C (jωp )] " [P (jωp )] = 0, (jωg )] − # [C (jωg )] # [P (jωg )] = − cos Φm , (jωg )] + # [C (jωg )] " [P (jωg )] = −sinΦm , where Φm is a phase margin, Am is a gain margin, ωp and ωg is 0dB (crossover) frequency. Last but not least we should mention the optimization algorithm based on the integral absolute error (IAE) minimization [44]: & t & t |e(t)|dt = |w(t) − y(t)|dt, IAE (t) = 0 0 where w(t) is the desired value of closed control loop and y(t) is the real value of closed control loop. This method does not insure the desired stability measure of the closed control loop. Measure of stability has to be checked out additionally. We can use a frequency method described in [39]. PI!Dµ controllers Control of the „reality“ using the PD controller, which is optimal for the „model“, and using the PD" controller. PI!Dµ controllers: design IV.2 A Design Approach • The plant model is assumed to be • and the fractional order PID controller is • It is expected that the gain and phase margin of the compensated systems are and • Question: how to choose parameters FOC tutorial III ©Dingyu Xue, NEU, PR China PI!Dµ controllers: design Design of fractional-order PID controller parameters is determined by given requirements. These requirements can be, for example: • the damping ratio, • the steady-state error, • dynamical properties, • etc. 27 PI!Dµ controllers: design Requiring it can be found that • For • It is found that PI!Dµ controllers: design PI!Dµ control: MATLAB I. General Description of Linear Fractional Order Systems Fourfour equations, withwith 7 variables We• have equations seven variables: I.1 The normal form FOC tutorial III ©Dingyu Xue, NEU, PR China 28 The rest of the variables can be determined by minimizing the ISE criterion • Thus compared with IO LTI’s, information on orders are also used • A FOTF model class/object can be defined in MATLAB to describe the system model FOC tutorial III ©Dingyu Xue, NEU, PR China PI!Dµ controllers: design Sample plots for combinations of ! and µ: µ, various ! combinations • For different PI!Dµ control: MATLAB II. MATLAB Class Design II.1 Create a @fotf directory • The fotf class can be defined as FOC tutorial III ©Dingyu Xue, NEU, PR China FOC tutorial III ©Dingyu Xue, NEU, PR China 30 4 Courtesy: Dingyü Xue, YangQuan Chen PI!Dµ controllers: design • A fractional order PID controller can be !Dµ controller: The designed optimal PI such that FOC tutorial III ©Dingyu Xue, NEU, PR China Courtesy: Dingyü Xue, YangQuan Chen 3 PI!Dµ control: MATLAB II.2 Other Overload Functions • Function call • A display function 31 FOC tutorial III ©Dingyu Xue, NEU, PR China 5 Courtesy: Dingyü Xue, YangQuan Chen PI!Dµ control: MATLAB PI!Dµ control: MATLAB II.3 Define a fotf Object in MATLAB • Feedback function • Example • MATLAB command • Display • These functions are suitable for interconnections of fractional order systems FOC tutorial III ©Dingyu Xue, NEU, PR China 6 Courtesy: Dingyü Xue, YangQuan Chen PI!Dµ control: MATLAB FOC tutorial III ©Dingyu Xue, NEU, PR China 9 Xue, YangQuan Chen Courtesy: Dingyü PI!Dµ control: MATLAB II.4 Overload functions for block • A common function unique • Plus function for parallel connections FOC tutorial III ©Dingyu Xue, NEU, PR China 7 Courtesy: Dingyü Xue, YangQuan Chen PI!Dµ control: MATLAB FOC tutorial III ©Dingyu Xue, NEU, PR China 10 Courtesy: Dingyü Xue, YangQuan Chen PI!Dµ control: MATLAB II.5 ofAn Example for interconnection Example interconnection: • Multiplication for series connection • Unity negative feedback system with The closed-loop system can be obtained as • Minus • The closed-loop system can be established • Uminus • The closed-loop system • Inv FOC tutorial III ©Dingyu Xue, NEU, PR China FOC tutorial III ©Dingyu Xue, NEU, PR China 8 Courtesy: Dingyü Xue, YangQuan Chen 11 • The closed-loop systems with even more complicated structures can easily be obtained with the overloaded functions FOC tutorial III ©Dingyu Xue, NEU, PR China 12 CHAPTER 4. CONTINUED FRACTION EXPANSION (CFE) + + + - - - + + 1 + + 22 + + Y2n* (s) Z2n -1(s) Y2n -2 (s) Z2n-3 (s) Y2 (s) Continued fractions (CFE) CHAPTER 4. CONTINUED FRACTION EXPANSION (CFE) CFEs and nested multiple loops Z 1 (s) 18 Figure 4.6: Nested multiple-loop control system of the first type. can be expressed in the form: G(s) ! a0 (s) + expansion, which is identical with the equation (6.1): Z(s) = Z1 (s) + b1 (s) a1 (s) + CHAPTER 4. CONTINUED FRACTION EXPANSION (CFE) b2 (s) b3 (s) a2 (s)+ a (s)+... 3 CHAPTER 4. CONTINUED FRACTION EXPANSION (CFE) b1 (s) b2 (s) b3 (s) = a0 (s) + ... a1 (s)+ a2 (s)+ a3 (s)+ + - + + 19 + - - (4.1) Z2n-3 (s) Nested loop of type I 1 Z(s) = 1 Z1 (s) + 1 CHAPTER 4. CONTINUED FRACTION EXPANSION (CFE) Y (s) + 2 Z 1 (s) 1 Z3 (s) + . . . . . . . . . . . .+. . . . . .+. . . . . . . . + ....... + + + Figure 4.6: Nested multiple-loop control system of the first type. 1 1 + + 1+ Y2n−2 (s) + 1 expansion, which is identical with the equation (6.1): Y2n* (s) Z2n−1 (s) + Y (s) 2n 1 Z(s) = Z1 (s) + 1 Z2n -1(s) Y2 (s) + Y2 (s) Z3 (s) + 1 Y2 (s) + Z3 (s) + Y2 (s) Z 1 (s) Figure 4.6: Nested multiple-loop control system of the first type. 1 1 . . . . . . . . . . . . . .expansion, . . . . . . . . . . . . .which . . . . . . is identical with the equation (6.1): 1 1 1 Z(s) = Z1 (s) + Y2n−2 (s) + 1 1 Z2n−1 (s) + Y2 (s) + 1 Y2n (s) Z3 (s) + Similarly to the above considerations, we can obtain a continued fraction expansion of the transfer function of the other interesting type of a nested multiple-loop control system, depicted in Fig. 4.7: Z(s) = Z1 (s) + CHAPTER 4. CONTINUED FRACTION EXPANSION (CFE)Y2 (s) + Figure 4.1: A control loop with From a negative + (4.4)feedback. it immediately follows that the transfer function of the circuit + + + - - 20 Y * (s)are used for creating Then this two parts of the characteristic polynomial 20 Y2n(s) - 2n = P2n−2of(s) Using the equations (4.4) and (4.5) we obtain the transfer function the = 1 system shown in Fig. 4.3: 1 + Q2n−1 (s)Y2n−2(s) system shown in Fig. 4.3: Y2n−2 (s) + Q2n−1 (s) + 1 1 Q2n−1 (s) = Z2n−1 (s) + P2n (s) = Z2n−1 (s) + 1 . (4.6) 1 Q2n−1 (s) = Z2n−1 (s) + P2n (s) = Z2n−1 (s) + Y2n (s). (4.6) Y2n (s) = (4.7) CHAPTER 4. CONTINUED FRACTION EXPANSION (CFE) 20 1 Combining the equations (4.4) and (4.5) we find the transfer function of Y2n−2 (s) + Y2n* (s) Combining the equations (4.4) and (4.5) we find the transfer function of 1 the nested multiple-loop system shown in Fig. 4.4: Letthe usnested now establish an interesting new relationship between continued fracZ2n−1 (s) + multiple-loop system shown in Fig. 4.4: + Y2n (s) 1 tions and nested multiple-loopQ2n−1 control 1 (s) systems. Figure 4.2: Nested multiple-loop control system – level 1. 1 Q2n−1 (s) = 2n−2 (s) = We firstPPrecall that2n−2 the(s)transfer function R(s) of the control 1 = = known 1 + Qfact 2n−2 (s)the 2n−1 (s)Y shown in Fig. 4.5 is then given by the Y2n−2 (s) + The1 transfer function of the system 1 + Q2n−1 (s)Y2n−2 (s) Y2n* (s) Y2n−2 + Qby (s) loop with a negative feedback shown in Fig. 4.1 is (s) given [8] + 2n−1 + relationship Q2n−1 (s) 1 1 1 + – level 1. = (4.7) Figure (4.7) 4.2: Nested multiple-loop control system G(s) 1 = 1 Q2n−3 (s)(4.4) = Z2n−3 (s) + P . Y2n−2R(s) (s) += 2n−2 (s) 1 Y2n−2 (s) + 1 + G(s)H(s) Y2n* (s) 1 Z (s) + + 2n−1 CHAPTER 4. CONTINUED FRACTION (CFE) 21 + 1 Z2n−1 (s) +EXPANSION Y2n (s) 1 = Z (s) + (4.8) Y2n (s) + 1 From (4.4) it immediately follows that the transfer function of the circuit 2n−3 Z2nY-12n−2 (s) (s) + The transfer function of the system shown in Fig. 4.5 is then given by the transfer of the system shown in Fig. 4.5 is then given by the 1 Y2n* (s) shownThe in Fig. 4.2 function is relationship Z (s) + 2n−1 relationship Y2n (s) 1 1 Figure 4.3: (4.5) Nested multiple-loop control system – level 2. Z2n -1(s) Q2n−3 (s) P = Z2n−3 + P2n−2∗(s) = , =(s)+ Q2n−3 (s) = 2nZ(s) (s)(s) 2n−3 (s) 1 +P12n−2 · Y2n Y12n (s) Continuing this process, we obtain the transfer function of the nested multiple1 = Z2n−3 (s) + (4.8) 1 control loop system shown in Fig. the form of a2. continued fraction = Z (s) + (4.8) Figure 4.3: Nested multiple-loop control system level ∗ equations (4.4) and (4.5)4.6 we in obtain the– transfer function of the 1 Using the (s) ++1. 2n−3+ where Y2n (s) = Y2n Y2n−2 (s) ++ 1 1 - Y2n−2 (s) + Z+2n−1system (s) + 1shown in Fig. 4.3: Z2n−1 (s) + Y2n (s) CHAPTER 4. CONTINUED FRACTION EXPANSION (CFE) 21 (4.4) and (4.5) we obtain the transfer function of the Using the equations Y (s) 2n Y * (s) 1 shownmultipleFig.(s) 4.3: Continuing this process, we obtain the2ntransfer function system of the nested Qin2n−1 . (4.6) = Z2n−1 (s) + P2n (s) = Z2n−1 (s) + Continuing this process, we obtain the transfer function of the nested multipleY2n (s) loop control system shown in Fig. 4.6 in the form of a continued fraction 1 loop control system shown in Fig. Z4.6 in the form of a continued fraction 2n -1(s) Q2n−1 (s) = Z2n−1 (s) + P2n (s) = Z2n−1 (s) + . (4.6) (s) transfer function of Combining the equations (4.4) and (4.5) we findY2nthe Y2n-4 (s) the nested multiple-loop system shown in we Fig. 4.4: Combining the equations (4.4) and (4.5) find the transfer function of the nested multiple-loop system shown in Fig. 4.4: 1 Q2n−1 (s) + + + Figure 4.4: Nested multiple-loop control system – level 3. (s) = = P2n−2 1 1 (s) Q2n−1 1 1+Q (s) 2n−1 (s)Y2n−2= + P2n−2 (s) = Y2n−2 (s) + 1 Q 1 + Q2n−1 (s)Y2n−2(s) 2n−1 (s) Y2n−2 (s) + Y2n* (s) Q2n−1 (s) 1 1 = 1 = (4.7) Z2n -1(s) 1 Y2n−2 Y2n−2 (s)(s) ++ 1 Z2n−1 (s) +1 Z (s) + 2n−1 Y2n -2 (s) Y2n (s) Y2n (s) - - 1 + + Y2n* (s) Z2n -1(s) Figure 4.7: Nested multiple-loop control system of the second type. 4.5 + - Z2n-3 (s) 1 (4.8) 1 Y2n (s) Continuing this process, we obtain the transfer function of the nested multiple- + + + Figure 4.5: Nested multiple-loop control system – level 4. * Y2n(s) - CFE and rational and irrational numbers Every rational and irrational number may be written in CFE form. Basically the process of finding a continued fraction development consists of two steps: if the fraction m/n is greater than 1, then divide. Otherwise, write the fraction m/n as 1/(n/m) and proceed with the first step. Continue until a numerator of 1 is obtained. Continued fractions are often used to get good rational CHAPTER 5. APPROXIMATION OF FRACTIONAL OPERATOR 25 approximations for real numbers. Let us consider the numbers written in the CHAPTER 5. APPROXIMATION OF FRACTIONAL OPERATOR 25 form: a0 + b1 /(a1 + b2 /(a2 + b3 /(a3 + . . . ))). In ”simple” continued fractions, as [a0 ;5. a1 , aAPPROXIMATION all the bi , ∀i, are 1 and the number can be re-written 2 , a3 , . . . ]. CHAPTER OF FRACTIONAL OPERATOR APPROXIMATION OF FRACTIONAL OPERATOR 5.2 + Z2n -1(s) Y2n -2 (s) Z2n-3 (s) Figure 4.5: Nested multiple-loop control system – level 4. 25 General CFE method 5.2 method General CFE method General CFE CHAPTER 5. APPROXIMATION OF FRACTIONAL OPERATOR 25 −α −α ,0<α<1 performing the CFE of the functions: fractional integral the Laplace domain) can (the fractional integral the domain) can be obtained by where the operator following idea appeared: dense of Laplace simpleby poles and In general [10],(the a rational approximation of thein function G(s) a= s−α , interlacing 0operator < αbe <in1obtained performing thesome CFE the,1functions: A rational approximation In general a rational approximation the function G(s) ==obtained sof−α 0equivalent <by α < to1 a branch (5.1) performing thezeros CFE the functions: along a of line in the sfor plane is, in way, cut; (the [10], fractional integral operator inofthe Laplace domain) can be H h (s) (1 + sT )α α 1 the negative and in s , 0the < αLaplace vieweddomain) an operator, branch "α cut along (the fractional can has be!aobtained performingintegral the can CFEoperator of functions: betheobtained by< 1,CFE ofasthe expressions: H (5.1) 1 h (s) = by 1 following α (4.7) (1 + sT ) Hl (s) 1+ (5.2) real axis for arguments but =is otherwise free of Hh (s) of = s on (−π, π) (5.1) ! poles "αand zeros. s performing the CFE of the functions: (1 + sT )α 1 1 Hl (s) = 1+ (5.2) ! "α (5.1) (ωT >>s 1), and Hl (s) for high frequencies where Hh (s) is the approximation 1 α (1Happroximation +l (s) sT )= for 1+ (5.2) the low frequencies (ω << 1). ! "α where Hsh(s) is the approximation for high frequencies (ωT >> 1), and Hl (s) h 1 α the approximation for low frequencies (ω << 1). Example Hgeneral 1rational + 5.2.1. (5.2) −α l (s) = [10], aPerforming of the(ωT function G(s) , 0we < obtain: α<1 the CFEfrequencies of the function (5.1), with 1), T =and 1,=α s= s forαapproximation approximation high >> H0.5, where Hh (s) isInthe l (s) Example 5.2.1. (the fractional integral operator in the the Laplace can+be obtained 4 1). 3 CFE ofdomain) 2 function Performing the with T = 1, by α = 0.5, we obtain: the approximation for low frequencies (ω << + 1.405s + 0.8433s + 0.1574s(5.1), 0.008995 0.3513s = functions: Hof l 1 (s) performing CFE the 4 3 2 for the high frequencies (ωT >> 1), and H (s) where Hh (s) is the approximation l s + 1.333s + 0.478s4 + 0.064s + 0.002844 0.3513s + 1.405s3 + 0.8433s2 + 0.1574s + 0.008995 Example H1 (s) = the approximation for low5.2.1. frequencies (ω << 1). 1 s4 + 1.333s3 + 0.478s2 + 0.064s + 0.002844 Example 5.2.2. Hh (s) = (4.8) 5.2 1 CFE method General H (s) = (1 + sT ) ! " 1 H (s) = 1+ Example: s (5.1) (5.2) Performing the CFE of the function H (5.1), = T = 1, αα = 0.5, we obtain: (5.1) h (s) with approximation for highPerforming frequencies (ωT >> 1), (s) where H (1 + sT ) and the CFE of the function (5.2), with TH = l1, α = 0.5, we obtain: h (s) is the Example 5.2.1. Example 5.2.2. ! "α 4 3 24 1 2of 3 CFE + 1.405s + 0.8433s 0.1574s 0.008995 0.3513s Performing the the function (5.2), with T = 1, α = 0.5, we obtain: approximation frequencies (ω << 1). += 4s +obtain: 0.448s + 0.0256 s Performing for the low CFE of the function (5.1), with T = 1, α 0.5, we+ Hl (s) =+ 1++2.4s (5.2) = H (s) H (s) = loop controlthis system shown Fig. 4.6 the form of a continued fraction Continuing process, weinobtain theintransfer function of the nested multiplethe loop control system shown in Fig. 4.6 in the form of a continued fraction Y2n -2 (s) + 1 11 = Z2n−3 (s) Y+2n−2 (s) + 1 Y2n−2 (s) + (s) + 1 Z2n−1 Y2n (s) Z2n−1 (s) + = Z2n−3 (s) + Nested loop of type II 5.2 approximation General CFE method In general [10], a rational of the function G(s) = s , 0 < α < 1 General CFEmethod method −α 5.2 5.2 General CFE In general [10],nominator a rationalpolynomials. approximation of the function G(s) = sdomain) 0< αbe< 1 in G(s) (the fractionalProbably, integral operator in the Laplace can obtained by was noted for ,the first time [17], In generalthis [10],fact a rational approximation of the function =s Thetransfer transfer function system shown in Fig. thenisgiven the by the The functionofofthethe system shown in 4.5 Fig.is 4.5 thenbygiven Figure 4.4: Nested multiple-loop control system – level 3. relationship relationship Q2n−3 (s) = Z2n−3 (s) + P2n−2 (s) Q 2n−3 (s) = Z2n−3 (s) + P2n−2 (s) 1 Z3 (s) + ................................. 1 1 Y2n−2 (s) + 1 Z2n−1 (s) + Y2n (s) 25 nominator polynomials. fact for was the noted for the firstintime nominator polynomials. Probably, this Probably, fact wasthis noted first time [17],in [17], Example 4.5.1. nominator polynomials. Probably, this fact was noted for the timeofinsimple [17], OF 5.first APPROXIMATION FRACTIONAL OPERATOR 25 where theappeared: following idea appeared: a CHAPTER dense interlacing poles and where where the following dense poles and The CFE idea for π, which gives the ”best”aapproximation of a given order, is of simple nominatorinterlacing polynomials. Probably, this fact was noted for the first time in [17], the following idea appeared: dense interlacing of simple polesinterlacing along the1,a1,swhere plane inThesome way, equivalent toand a branch cut;poles and [3, 7, 15, zeros 1, 292, 1, 1, 1, 2,a1,line 3, 1, in 14, 2, 2, 2, 2,the 2,is, ...]. very following idealarge appeared: a dense of was simple nominator polynomials. Probably, thiscut; fact noted for the first time in [17], zeros along a line in the s plane is, in some way, equivalent to a branch α the term 292 meansin a good approximation [13]. zeros along a line sα plane is, in iszeros some way, equivalent tois,ainidea branch cut; along a line in the s aplane some way, equivalent tointerlacing a branch cut; and sthat , 0the <following < 1,convergent viewed as an operator, has branch cut along the negative where the following appeared: a dense of simple poles and sα , 0 < < 1, viewed as along operator, has a negative branch cut along the negative and sαand , 0 <sαα viewed as an operator, has branch cut along the 355 and , 0< < 1, α< 1,axis viewed as an operator, has aa. .α.branch cut zeros along aanline inthe the snegative plane is, inand somezeros. way, equivalent to a branch cut; real for arguments on (−π, π) but is otherwise free of poles 3.14159292 [3, 7, 15, 1] = [3, 7, 16] = of= s αof s on (−π, π) but is otherwise free of poles and zeros. 113 real axis for arguments and s free , 0 < of α <poles 1, viewed an operator, has a branch cut along the negative realfor axisarguments for arguments s on (−π,π)π)but but is is otherwise otherwise andasand zeros. real axis of sof on (−π, free of poles zeros. real axis for arguments of s on (−π, π) but is otherwise free of poles and zeros. CFEs and nested multiple loops + 1 Both types of nested multiple-loop systems, presented in this section, can be used for simulations and realizations of arbitrary transcendental transfer functions. For this, the transfer function should be developed in a continued fraction, which after truncation can be represented by a nested multiple-loop system shown in Fig. 4.6 or Fig. 4.7. CHAPTER 5. + (4.9) 1 Z 1 (s) CFE and nested multiple-loop control systems + 1 23 Z2n-3 (s) . Y2n (s) + + + Y2n -2 (s) Z (s) Considering the continued fraction (4.3) as a tool for designing a correZ2n2n-1-1(s) equations (4.4) and (4.5) we find the transfer function of sponding LC circuit, we can conclude that stability of a Combining linear systemthe is equivFigure 4.3: Nested multiple-loop control system – level alent to realizability its test function control R(s) with the help2.2.ofmultiple-loop only passive system shown in Fig. 4.4: the Figure 4.3: of Nested multiple-loop system –nested level electric components. 1 Q2n−1 (s) 4. of CONTINUED FRACTION EXPANSION (CFE) 20 Using the equations (4.4) and (4.5) we obtain theCHAPTER transfer function the Y2n -2 (s) + + Z2n -1(s) Figure 4.2: Nested multiple-loop control system – level 1. 1 Using the equations (4.4) and (4.5) we obtain the transfer function of the + + bn−1 1 + + s + bsystem shown in Fig. 4.3: ns 1 + is +stable. If some bk is negative, If bk > 0, k = 1, . . . , n, then theY2nsystem * (s) 1 Y2n* (s) Q2n−1 (s) = Z2n−1 (s) + P2n (s) = Z2n−1 (s) + . (4.6) then the system is unstable. m(s) 1 ................................. 1 1 Y2n−2 (s) + 1 Z2n−1 (s) + Y2n (s) CFEs and nested multiple loops The rational function R(s) should be written in the form of a continuous 11 1 1 + form of Pa2nfraction, the highest power of(4.5) s is fraction: its test function ,, (4.5) (s)P2n =(s) = in which + - in1 the Z2n -1(s) ∗ ∗ = = 1 1 1 +(4.3) 11 ·+Y12n · (s) Y2n (s) Y2n Y(s) 2n (s) R(s) = 1 contained in bthe denominator: * (s) Y 2n s + 1 Y2n* (s) where Y ∗ = Y ∗ (s) 1 Y(s) +"1. 4.3: Nested multiple-loop where 2n1.Figure # control system – level 2. b2 sY +2n (s) =2n 2n (s) + . . . . . . .system . m(s) . . . . – level 1. n(s) Figure 4.2: Nested multiple-loop control n(s) Y4 (s) + 20 2n 4.4 Z2n-3 (s) (4.9) 1 Z1 (s) + CFEs and nested multiple loops 22 Y2n -2 (s) ................................. 1 1 Y2n−2 (s) + 1 Z2n−1 (s) + Y2n (s) 1 Z(s) = CHAPTER 4. CONTINUED FRACTIONbetween EXPANSION (CFE)fraccontrol systems Let us now establish an interesting new relationship continued It is also known continuous fraction expansions tionsthat and nested multiple-loop control systems. can be used for investiLet us now establish an interesting new +relationship between continued frac1 gating stabilityFRACTION of systems. Forfact this, the polynomial Q(s) Welinear first recall the known that thecharacteristic transfer function R(s) of the control CHAPTER 4. CONTINUED EXPANSION (CFE) 19 tions and nested multiple-loop control systems. loop with aWe negative feedback shown in Fig. 4.1 is *given by [8] firstofrecall the known fact that thebe transfer function the control of the differential equation the system should divided inR(s) twoof parts, the Y (s) + loop with a negative feedback shown in Fig. 4.1 is given by [8] G(s) “even” part (containing even powers of s) and G(s) the “odd” part (containing odd . (4.4) R(s) = Figure 4.2: Nested multiple-loop control system – level 1. G(s) + G(s)H(s) powers of s): H(s) . (4.4) R(s)1 = 1 + G(s)H(s) + + Q(s) = m(s) + n(s). 1 From (4.4) it immediately follows that the- transfer function of the circuit (4.9) 1 Y4 (s) + Similarly to the above considerations, we can obtain a continued fraction expansion of the transfer function of the other interesting type of a nested multiple-loop control system, depicted in Fig. 4.7: CFE and nested multiple-loop or R(s) = 1 + control 4.4 CFEsystems and nested multiple-loop CFE and stability of linear systems R(s) = 1 Similarly to the above considerations, we can obtain a continued fraction expansion of the transfer function of the other interesting type of a nested multiple-loop control system, depicted in Fig. 4.7: Y2n -2 (s) sponding LC circuit, we can conclude that stability of a linear system is equivto realizability of its. .test function R(s)i+m with)). the help of only passive the alent points (si , G(si)), . , (s i+m , G(s electric components. in Fig. 4.2 is (CFE) CHAPTER 4. CONTINUED shown FRACTION EXPANSION in Fig.(CFE) 4.2 is CHAPTER 4. CONTINUED FRACTIONshown EXPANSION 1 ................................. 1 1 Y2n−2 (s) + 1 Z2n−1 (s) + Y2n (s) Z2n -1(s) 1 bn−1 s +1 b1 s + b2 s + Pµ (s) bn s = is. . .stable. ......... k = 1,! . . . ,R n,i(i+1)...(i+m) then the system If bk > 0,G(s) Qν (s) If some bk is negative, 1 then the system is unstable. µ p0 + p1 s + . . . + pbn−1 µ ss + b s Considering the = continued fraction (4.3) as a tooln for designing a corre(4.2) ν q + q s + . . . + q s > 0, k = 1, . . . , n, then the system is stable. If some bk is negative, If b 1 ν k sponding LC circuit, we can0 conclude that stability of a linear system is equivthen the system is unstable. alent to realizability function R(s) with the help of only passive m + 1 =of its µ +test ν+ 1 Considering the continued fraction (4.3) as a tool for designing a correelectric components. 4.4 + + 1 4.3 + Y4 (s) + Y2n* (s) G(s) where ai s and bi s are rational functions of the variable s, or are constants. H(s) ! (CFE)which is19an CHAPTER 4. CONTINUED FRACTION EXPANSION The application of the method yields a rational function, G(s), approximation of the irrational Figure 4.1:function A controlG(s). loop with a negative feedback. + G(s) On the other hand, for interpolation purposes, rational functions are someThe rational function R(s) should be written in thedue formtooftheir a continuous times superior to polynomials. This is, roughlyH(s) speaking, ability fraction: to model functions with poles. (As it can be seen 1 later, branch points can be Figure (4.3) R(s)4.1: = A control loop with a negative feedback. 1 considered as accumulations of interlaced b1 s + poles and zeros). These techniques 1 The rational function R(s) should be written in the form of a continuous b s + are based on the approximations of an irrational function, G(s), by a rational 2 ............ fraction: 1 in the variable s: function defined by the quotient of R(s) two =polynomials (4.3) passing through + + 1 - 1 Y2 (s) + 22 Z3 (s) + 1 2 + 0.576s + 0.0256 s0.002844 9s40.064s + 12s3 ++4.32s s4 + 1.333s3 + 0.478s2 + s4 + 4s3 + 2.4s2 + 0.448s + 0.0256 2 H2 (s) = 4 3 2 Example 5.2.1. 3 + 4.32s2 + 0.576s + 0.0256 1.405s + 0.8433s + 0.1574s + 0.3513s4 +where 9s + 12s(ωT is the approximation for0.008995 high frequencies >> 1), and Hl (s) Hh (s) H1 (s) = 3 + 0.478s Example s4 the +5.2.2. 1.333s +Carlson’s 0.064s Performing the CFE of function (5.1), with T+=0.002844 1, α(ω=<<0.5, 5.3 2 for method the approximation low frequencies 1). we obtain: Performing the CFE of the function (5.2), with T = 1, α = 0.5, we obtain: 5.3 Carlson’s method The method Example 5.2.1. 3 2 proposed by Carlson in [4, 5], derived from a regular Newton Example 5.2.2. + 0.8433s 0.1574s + (5.1), 0.008995 0.3513s4 + 1.405s 4the CFE 3 + 2 function iterative approximation of the can bewe considered Performing of the with T α-th = 1,inroot, α[4,=5], 0.5, obtain: + 4sused +for 2.4s 0.448s + 0.0256 sprocess The + method proposed by Carlson derived from a regular Newton HPerforming 1 (s) = as(5.2), belonging to this group. starting point of the method is the statement (s) H23function thesCFE of the with T = αThe = 0.5, we obtain: 4 + 1.333s 4 2 + 30.064s 2 += 0.478s +1,used 0.002844 process for iterative approximation of the α-th root, can be considered 9sof + 12s +44.32s + 0.576s + 0.0256 the0.3513s following relationships: + 1.405s3 + 0.8433s2 + 0.1574s + 0.008995 as belonging to this group. The starting point of the method is the statement H1 (s) = 3 + 0.478s2 + 0.064s + 0.002844 2.4s2 + 0.448s +the 0.0256 s4 + 4s3 + 1/α s4 + 1.333s of following relationships: − (G(s)) = 0; H(s) = (G(s))α (H(s)) Example 5.2.2. H2 (s) = 9s4 + 12s3 + 4.32s2 + 0.576s + 0.0256 1/α (H(s)) − (G(s)) = 0; method Performing the 5.3 CFE ofCarlson’s the Example function5.2.2. (5.2), with T = 1, α = 0.5, we obtain: (5.3) H(s) = (G(s))α Performing the CFE of the function (5.2), with T = 1, α = 0.5, we obtain: 5.3 The method proposed by Carlson in [4, 5], derived from a regular Newton 0.0256 s4 + 4s3 + 2.4s2 + 0.448s + 4s + 2.4s + 0.448s + 0.0256 s + Carlson’s process for iterative approximation of the α-th root, can be considered H (s) = = usedmethod H (s) 4 2 4 3 2 2 3 2 9s4 + 12s3 + 4.32s2 + 0.576s + 0.0256 (5.3) CHAPTER 6. GENERAL APPROACH TO FRACTANCE DEVICES General Approach to Fractances 32 CHAPTER 6. GENERAL APPROACH TO FRACTANCE DEVICES 6.2 Domino ladder circuit 32 General Domino Approach ladder circuit to Fractances 6.2 Z1 Z3 Z 2n -3 Z 2n -1 CHAPTER 6. GENERAL APPROACH TO FRACTANCE DEVICES 6.2 Z1 A devices or a circuit exhibiting fractional-order behaviour is called a fractance. Y4 Y Z(s) Domino ladder circuit 2 Z3 Z1 Z3 Z 2n -3 Y2n -2 Z 2n -3 32 Y2n Z 2n -1 Z 2n -1 Figure 6.2: Finite ladder circuit. Y • domino ladder circuit network, • a tree structure of electrical elements, • transmission line circuit Y Y Y2ladder lattice 2n -2 4 Domino fractional operator Y4 can approximate Y Y2n more 2n Y2 networks 2n -2 Z(s) effectively than the lumped networks [10, 30]. Z(s) Let us consider the circuit depicted in Fig. 6.2, where Z2k−1 (s) and Y2k (s), k = 1, . . . , n, are given impedances of the circuit elements. The resulting impedance Z(s) of the entire circuit can be found easily, if we consider it in Figure 6.2: Finite ladder circuit. the right-to-left direction: 1 circuit. Figure 6.2: Finite ladder Design of fractances can be using a truncated CFE, which gives a rational approximation. (6.1) Z(s) =ladder Z1 (s)+ lattice networks can approximate fractional operator more Domino 1 Y2 (s) + 1 effectively than the lumpedZ3networks [10, 30]. (s) + Let lattice us considernetworks the circuit depicted in Fig. 6.2, where1fractional Z2k−1 (s) and Yoperator 2k (s), Y4 (s)approximate + Domino ladder can more . . . .circuit . . . . . . . . .elements. . . . . . . . . . . . . The . . . . . .resulting .. k = 1, . . . , n, are given impedances of the 1 effectively than the lumped networks [10, impedance Z(s) of the entire circuit can30]. be found easily, if1we consider it in Y2n−2 (s) + 1 the right-to-left direction: Let us consider the circuit depicted in Fig. 6.2,Zwhere 2n−1 (s) + Z2k−1 (s) and Y2k (s), Y (s) 2n 1 k = 1, . . . , n, Z(s) are =given (6.1) resulting Z1 (s)+ impedances of the circuit elements. The 1 The relationship Y2 (s) +between the finite domino ladder network, shown in Fig. 6.2, impedance Z(s) of the entire circuit can be found 1 easily, if we consider it in Z3 (s) +(6.1) provides an easy method for designing a circuit and the continued fraction 1 Y4 (s) the right-to-left direction: with a given impedance Z(s). For+ this one has to obtain a continued fraction ................................. Z(s) = Z1 (s)+ expansion for Z(s). Then the obtained particular expressions for Z2k−1 (s) and 1 1 Y2k (s), k = 1, . . . , n, will give the types 1 of necessary components of the circuit Y2n−2 (s) + and their nominal values. 1 Z2n−1 (s) + Y 1(s) 2n Y2 (s) + Example 6.2.1. 1 Z3 (s) + with the impedance To design a circuit (6.1) The relationship between the finite domino ladder network,1 shown in Fig. 6.2, Y4(6.1) (s) provides + and the continued fraction for designing a circuit + easy 4s2 +method 1 s4an Z(s) .=. . . . . . . . . ., . . . . . . . . . . . . . . . . (6.2) ...... with a given impedance Z(s). For this ones3has + s to obtain a continued fraction 1 expansion for Z(s). Then the obtained particular expressions for Z2k−1 (s) and 1 Y2k (s), k = 1, . . . , n, will give the types of necessary components of the circuit Y2n−2 (s) + 1 and their nominal values. Z2n−1 (s) + Y2n (s) Example 6.2.1. To design a circuit with the impedance CHAPTER 6. GENERAL APPROACH TO FRACTANCE DEVICES 31 General Approach to Fractances 6.1.2 Negative impedance converters It can be shown that the use of CFE for analogue realization of arbitrary transfer functions may lead to the appearance of negative impedances. This observation is not unknown. For example, in the paper [10] S. C. Dutta Roy in [19] and recalls Khovanskii’s continued fraction expansion for x1/2 found 1/2 makes a remark that S. C. Dutta Roy on Khovanskii’s CFE for x : “. . . if x is replaced by the complex frequency variable s, then the realization would require a negative resistance. Thus, the [Khovanskii’s] CFEs do not seem to be useful for realization of fractional or capacitor.” CHAPTER 6. inductor GENERAL APPROACH TO FRACTANCE DEVICES 31 Then he describes a method for circumventing this difficulty, which gives 6.1.2 Negative impedance converters a continued fraction expansion with positive coefficients. However, possibility of realization of negative However,the the possibility of realization of negative impedances in electric It can be circuits shown has thatbeen thepointed use ofoutCFE for analogue realization arbitrary by Bode [3, Chapter Later,pointed inof1970s, operaimpedances in electric circuits hasIX]. been out transfer functions may lead to the appearance of negative impedances. This tional amplifiers appeared, which significantly simplified creation of circuits exby hibiting H. W.negative Boderesistances, inFor 1945. observation is not unknown. example, in the paper [10] S. C. Dutta Roy negative capacitances, and negative inductances. Such circuitscontinued are called negative-impedance converters found in [19] and recalls Khovanskii’s fraction expansion for x1/2[9]. Thethat simplest scheme of a negative impedance converter (or current inverter) makes a remark is shown in Fig. 6.1. The circuit consists of an operational amplifier, two of replaced equal resistance and a component with the impedance “.resistors . . if x is by theR,complex frequency variable s, then Z. The entire circuit, considered a single element, has Thus, negative the realization would requireasa negative resistance. theimpedance [Kho- −Z. = V /(−Z)). This means that I in in vanskii’s] CFEs do not seem to be useful for realization of fractional For example, taking a resistor of resistance RZ instead of the element Z, we inductor or capacitor.” obtain a circuit, which behaves like a negative resistance −RZ . The negative resistance means that if such an element of negative resistance,which for instance, Then he describes a method for circumventing this difficulty, gives kΩ is connected in series with a classical 20 kΩ resistor, then the resistance a continued−10 fraction expansion with positive coefficients. of the resulting connection is 10 kΩ. However, the possibility of realization of negative impedances in electric circuits has been pointed out by Bode [3, Chapter IX]. Later, in 1970s, operaR Iin tional amplifiers appeared, which significantly +simplified creation of circuits exhibiting negative resistances, negativeVincapacitances, and negative inductances. _ Such circuits are called negative-impedance converters [9]. Z The simplest scheme of a negative impedanceRconverter (or current inverter) is shown in Fig. 6.1. The circuit consists of an operational amplifier, two resistors of equal resistance R, and component with converter. the impedance Z. The Figure 6.1: aNegative-impedance entire circuit, considered as a single element, has negative impedance −Z. This means that Iin = Vin /(−Z)). For example, taking a resistor of resistance RZ instead of the element Z, we obtain a circuit, which behaves like a negative resistance −RZ . The negative resistance means that if such an element of negative resistance, for instance, −10 kΩ isNegative connected inimpedance series with a classical 20 kΩ resistor, the resistance converters are then available: of the resulting connection is 10 kΩ. General Approach to Fractances Iin Vin R + _ Z R Figure 6.1: Negative-impedance converter. 2 The relationship between the finite domino network, shown in Fig. 6.2, +1 s4 + 4sladder Z(s) = , (6.2) s3 +easy s and the continued fraction (6.1) provides an method for designing a circuit with a given impedance Z(s). For this one has obtain a continued fraction 33 CHAPTER 6. GENERAL APPROACH TOtoFRACTANCE DEVICES expansion for Z(s). Then the obtained particular expressions for Z2k−1 (s) and Y2k we (s),have k = 1, . . , n, will giveinthe types offraction necessary components of the circuit to .develop Z(s) continued and their nominal values. CHAPTER 6. GENERAL APPROACH TO FRACTANCE DEVICES 33 4s2 + 1 1 s4 + . (6.3) =s+ Example 6.2.1. Z(s) = 1 1 s3 + s s + wedesign have to develop with Z(s) the in continued fraction To a circuit impedance 9 1 3 circuit with Example 1: design a domino ladder s+ CHAPTER 6. GENERAL APPROACH TO FRACTANCE DEVICES 2 4 GENERAL CHAPTER APPROACH DEVICES 3333 2 FRACTANCE 2 + 4s2 +4 1 4s 1TO s6. +1 .s (6.3) s+ Z(s) = Z(s)3 = s += (6.2) 1 3 1, s + s 3 +fraction s s+ we have to develop in continued we have Z(s) to develop Z(s) sin continued fraction 9 1 3 From this expansion it follows that s+ General Approach to Fractances Find a CFE: s4 1+ 4s2 + 1 s4 + 4s2 + 2 12 1 s 1. . (6.3) =s+ Z(s) = (6.3) 9 2 s3=+ ss + 1 1 1 s 3+1 s;s Y2 (s) =s +s; 3 Z3 (s) =s3 + 9Y4 (s)1= s. 2 3 3 s + 9 1 2 3 From this expansion it follows that s+2 2 3s 2 Therefore, for the analogue 9realization in the 1 form 3ofs the2 first Cauer’s Z1 (s) Y2 (s)that s; Y4 (s) = s;From Z = s; = following = s.of coils and 3 (s) expansion follows canonic LC circuit [20] this we have the values 2 to itchoose 3 3 From this expansion it follows that 9 capacitors: 1 2 Z1 (s) = s; Z(s) = Therefore, Z1 (s) = s; realization s; theY2form s. Z3 (s) = in (s) = ofs; the Yfirst Therefore, for the analogue Cauer’s 4 (s) = 2 3 1 2 3 and 9 =to9choose 2 coils canonic LC Zcircuit [20] the values of s; CY Y4 (s) s.].first Cauer’s s; LweZ=have (s)1following =[F ];s; = 3 (s)the 2= for analogue realization in the form of the [H];=Therefore, = L1 = 11 (s) [H]; C [F 3 2 4 2 3 capacitors: For the first Cauer’s LC we must 2canonic 3circuits 3 3 oftake canonic LC circuit [20] we have to choose the following values coils and Therefore,capacitors: for the analogue realization 1in the form of 2the first Cauer’s 9 Example = L3 = L16.2.2. [H]; to choose C2 = the [F ];following C4 = [F ].of coils and canonic LC1 [H]; circuit [20] we have values 9 1 2 2 3 3 [H];equation L3 = (6.2) L1 = 1by [H]; canCbe [F ]; C4 =in [F ]. form The function Z(s) given also the 2 = written capacitors: 2 3 3 Example 6.2.2. 4 29 1 2 + 4s +[H]; 1 1 C = [F1 ]; s6.2.2. Example 1Z(s) [H];= Lby L1 = Z(s) 3 = The function given equation writtenCalso the]. form .4 = in [F (6.4) = (6.2) + 2can 3be The function 1 be written3 also in the form 1 (6.2) can s3 Z(s) + s2 given bys equation + 11 s4 + 4s2 + 1 s4 +14s2 + 13s 1 1 9 Example 6.2.2. (6.4) (6.4) Z(s) = . Z(s) = = 3 + 1 = +2s1+ 2. 1 1 s3 + s s ss + s(6.2)+can The function Z(s) given by equation be written also in the form + 1 9 9 3s 1 3s 3s + 2s + 2 2 2s 1 2 +1 1 s4 + 4sthat From this expansion . 3s (6.4) = + Z(s)it=follows 3s 1 1 s s3 + s From this expansion it follows that + 1 9 1 2 1 9 3s From expansion follows that + (s) = = (s) = (s) = Z1this ; Zit3 (s) ; Y ; Y . 29 4 1 s 3s 2s 1 ;2 Y4 (s)3s= 2 . Z1 (s) = ; 2s Z3 (s) = ; 1 1Y2 (s) = 3s3s 2 s 9 2s = the = Z1 (s)for ; analogue Z3 (s) = ; Y2 (s) ; Y . 3sCauer’s 4 (s) = Therefore, realization in the form of the second s Therefore, 2s 3s 3s second for the analogue realization in the form of the Cauer’s From this expansion ithave follows that canonic LC circuit [20] we to choose the following values of coils canonic LC circuitrealization [20] we haveintothe choose theoffollowing valuesCauer’s of coilsand and Therefore, for the analogue form the second capacitors: capacitors: 1 we have to 9choose the following 1 2 and canonic LCZcircuit [20] = coils Z3 (s) = ; Y2 (s) = ; values Y4 (s)of . 1 (s) = ; s 2s 2 3s 3s 3 capacitors: C1 = 1 [F2]; C3 = [F ]; 9L2 = L = 3 [H]; L34 = [H]. Therefore, for the analogue in the form of the 29 realization 3 2second Cauer’s General Fractances [F ]; to choose L to [H].of coils and = 1 circuit [F ];Approach C we = have = 3the [H]; L =values C LC canonic [20] following 9 2 C1 = 1 [F ]; C3 = [F ]; 2 3 [H]; L4 = 2 [H]. 3 2 4 Example 6.2.3. To design a circuit with the impedance capacitors: Example 6.2.3. CHAPTER 6. GENERAL APPROACH TO FRACTANCE DEVICES 34 Example To design6.2.3. a circuit with the impedance s4 + 3s2 + 8 2 Z(s) L = = 33 [H]; , L = 3 [H]. [F ]; representation =circuit 1 [Fa];continuous 3 =impedance 2 2s + of 4 To with Cthe onedesign has C to1aobtain fraction Z(s), 4s the function 1 Example II: design a domino ladder circuit with (6.5) CHAPTER 6. GENERAL APPROACH TO2 + FRACTANCE DEVICES 34 9 s4 + 3s 2 8 2 CHAPTER 6. GENERAL DEVICES 34 1 3sAPPROACH += 8 41 TO s4 +Z(s) 2 FRACTANCE 3 s 2s + 3s + 8 , (6.6) (6.5) s+ Z(s) = = 3 + 4s 1 2 Example 6.2.3. Z(s) = , (6.5) 2s + 4s one has to obtain a continuous fraction representation of the function Z(s), 3 2s + 2srepresentation + 4s one has to obtain a continuous fraction 1 of the1function Z(s), To design a circuits4with 2the impedance − s1 + 1 4+ 3s 2+ 8 3 12 1 3s + 8 = 1 s + Z(s) == s + − s 3 (6.6)(6.6) s + Z(s) = 2 2s2s3++4s 1 1 2 22s4 +2s 4s 2s3s +++ 81, 1 Z(s) = (6.5) 1 1 From this expansion it follows that s+ −− s + 2s3 + 4s 3 3 1212 1 1 s(s)s = − 3 s. Z1 (s) = s; Z3 (s) = − s; Y2 (s) = 2s; − 2Y− 42 2 12 2 From expansionititfollows follows that that From thisthis expansion Therefore, for the analogue realization in the form of the first Cauer’s 1 1 3 3 1 1 canonic LC Z Y2 (s) = s; [20] Zwe − to s;choose (s) = −of s.coils and 1 (s) 3 (s)have Z1 (s) s;2 Z3 (s) Y2the =circuit ==− (s)=following =2s;2s; Y4values Y4 (s) =2 − s. 12 s; 2 12 2 capacitors: Therefore, for the analogue realization in the form of the first Cauer’s Therefore, for the analogue realization in the form of the 3first Cauer’s 1 1 canonic circuit [20] we−have[H]; to chooseC the following values L3 = [F ].and L1 = LC[H]; C4 = of − coils 2 = 2 [F ]; canonic LC 2circuit [20] we have 12 to choose the following values2 of coils and capacitors: capacitors: Here we see and capacitance. Such elements cannot 1 negative inductances 1 3 [H]; L3electric [H]; C2 = However, [F ]. realized L1 = = − components. 2 [F ]; C 4 = − be realized using passive they can 1 2 112 3 2 be [H]; L =inductances C2 =operating LHere − [H]; and 2 [F ]; Such C 1 = 4 = − [F ]. capacitance. elements with the help of negative active3 components, namely amplifiers. We can 2we see 12 2cannot be realized passive electric components. However, beinrealized realize this byusing negative inpedance converter which was they described previous Here we see negative inductances and capacitance. Suchcanelements cannot with theusing help of active electric components, namely operating amplifiers. besubsection. realized passive components. However, they can We be can realized realize this by negative inpedance converter which was described in previous with the help of active components, namely operating amplifiers. We can subsection. realize by negative inpedance converter which was described in previous 6.3 thisTransmission lines circuit subsection. Transmission lines circuit Let6.3 us consider the circuit depicted in Fig. 6.3, where Z2k−1(s) and Y2k (s), k = 1, . . . , n, are given impedances of the circuit elements. This structure is known Let usTransmission consider the circuit depicted in Fig. 6.3, where Z2k−1(s) and Y2k (s), k = 6.3 lines circuit as a1, transmission line or a symetrical domino ladderThis lattice network as well. . . . , n, are given impedances of the circuit elements. structure is known Theasresulting impedance Z(s) of the entire circuit canlattice be expressed as as a CFE a transmission line or a symetrical domino Let us consider the circuit depicted in Fig. 6.3,ladder where Z2k−1network (s) and Y2kwell. (s), k = described by (6.1) with symetrical distribution of elements, Z (s) = Z (s) We have: Notice negative values - can be done using OpAmps 2k−1 as a CFE 2k The resulting impedance Z(s) of the entire circuit can be expressed From this expansion it1follows that L3 = − [H]; 3 C2 = 2 [F ]; C4 = − [F ]. subsection. 1 L1 = [H]; 2 3 L1 = [H]; 1 L3 = − [H]; 2 C2 = 2 1[F ]; 312C4 = − [F1]. 1 L3 = [H]; C 2 [F ]; cannot C4 = − [F ]. 2 = − s; 2− capacitance. 1Y=(s) [H]; 2 = elements Here=we inductances and Such Z1 (s) = s; Z3 (s) Y212 (s) 2s;see Lnegative 4 2 = − s. 12 2 2 Here we see negative 12 2 components. inductances capacitance. Such elements cannot be realizedand using passive electric However, they can be realized Hereofwe negative inductances and capacitance. Such elements cannot Therefore, for the analogue realization thehelp form thesee first Cauer’s Transmission lines circuit be realized using6.3 passive electric components. However, they cannamely be realized within the of active components, operating amplifiers. We can be realized using electric components. However, they can be realized canonic LC circuit we have to choose the this following values of passive coils and with [20] the help of active components, namely operating amplifiers. can was described in previous realize by negative inpedance converterWe which with the help of active components, namely operating We can capacitors: Let us inpedance consider the circuitwhich depicted Fig. 6.3, Z2k−1 (s) and Y2k (s), amplifiers. k= realize this by negative converter was in described in where previous subsection. realize this by negative inpedance converter which was described in previous 1 subsection. 11, . . . , n, are given impedances of the 3 circuit elements. This structure is known L3 = − [H]; C2 = 2 [F L1 = [H]; ]; C4 = − [F ]. subsection. 2 12 2 as a transmission line or a symetrical dominocircuit ladder lattice network as well. 6.3 Transmission lines Here we see negative inductances and capacitance. Such cannotcircuit can be expressed as a CFE The resulting impedance Z(s)elements of the entire 6.3 Transmission lines circuit be realized using passive electric components. However, they can be realized us consider circuit depicted in Fig. 6.3, where circuit Z2k−1 (s) and k= described Let by (6.1) with the symetrical distribution oflines elements, Z2k−1 (s) Y=2k (s), Z2k (s) 6.3 Transmission with the help Let of active components, namely amplifiers. We 1, . .= . ,operating n, are given impedances thecan circuit elements. us consider the circuit depicted Fig. 6.3, where Z2k−1of (s) and Y2k (s), k = This structure is known (s) Yin and Y2k−1 2k (s), k = 1, . . . , n. as of awhich transmission line or athe symetrical domino ladder lattice network as well. realize this by negative inpedance converter was described in previous 1, . . . , n, are given impedances the circuit elements. This structure is known Let us consider circuit depicted in Fig. 6.3, where Z (s) In the case if we assume the same elements in transmision line in2k−1 series and Y2k (s), k = The resulting Z(s) of the entireof circuit can be elements. expressed as a CFE subsection. as a transmission line or a symetrical ladder lattice network as well. 1,domino . . .impedance , n, are given impedances the circuit This structure is known General Approach to Fractances (Z ) and the same in shunt (Z ), we get a structure depicted in Fig. 6.4. CHAPTER 7. R 2 3 2 4 6 2n-1 2n 3 5 3 4 5 6 4 2n-1 !5 Magnitude [dB] !10 2n 1 3 5 !20 !25 !30 !35 !40 2 3 10 4 10 Frequency [rad/sec] 10 !30 !40 !50 !60 !70 !80 !90 2 3 10 4 10 Frequency [rad/sec] 10 Figure 7.5: Bode plots of the I 1/2 controller where half order integral was approximated by transmission line depicted in Fig. 7.2. 2n-1 as a transmission line or a symetricallines dominocircuit ladder lattice network as well. 6.3 ZaTransmission Za The Za resultinglines impedance Z(s) of the entire circuit can be expressed as a CFE Transmission circuit Y6 Y 2n-1 Z6 Y 2n Z 2n us consider circuit depicted in Fig. 6.3, where Z2k−1 (s) and k= described Let by (6.1) with the symetrical distribution of elements, Z2k−1 (s) Y=2k (s), Z2k (s) Figure 6.3: General structure of transmission line. Zb us Zb Zb Zb Zb Zb depicted Zb given 1, . .= .Zb, n, are theand circuit elements. Let consider the circuit Fig. 6.3,impedances where Z2k−1of(s) Y2k (s), k = This structure is known (s) YZain and Y2k−1 2k (s), k = 1, . . . , n. Za Za Za Za line or Za Za Za as a transmission a symetrical domino ladder lattice network as well. 1, . . . , n, are given impedances of the circuit elements. is known In the case if we assume the This samestructure elements in transmision line in series The resulting impedance of the entire as circuit as a transmission line or a symetrical domino ladderZ(s) lattice well.can be expressed as a CFE (Zcomposed same in shunt (Z ), we getnetwork a structure depicted in Fig. 6.4. a ) and the Figure 6.4: Transmission line circuit of two inpedance Za and Zbb.symetrical described by (6.1) with of elements, Z2k−1 (s) = Z2k (s) Zb circuit Zb Zbcan Zb expressed Zb distribution Zb Zb The resulting impedance Z(s) of theZ(s) entire be as aZbCFE Impedance of such kind of transmision line, which can be used also as a Za Za Za Za (s) = Y2k (s), k = 1, . . . ,Zn. and Y2k−1 described by (6.1) with symetrical distribution of elements, 2k−1 (s) = Z2k (s) 6.4 Tree structure circuit model of real cable line (solved by O. Heaviside in 1887), can be as the case if we assume the same elements in transmisionexpressed line in series = Y2k (s),TO k FRACTANCE = 1, . .In . , n. and Y2k−1 (s)APPROACH CHAPTER 6. GENERAL DEVICES 36 ! Figure 6.4: Transmission line circuit of structure two inpedance Za and in Zb .Fig. 6.4. (Z same in 6.5, shunt (Z=b ),composed we Let us consider the total impedance theassume fractance circuit asthe shown in Fig. a ) and In the case ifofwe the same elements in transmision line series depicted Zaget .Zba. in (6.7) Z(s) which has a recursive structure with the combination Za such Zkind of transmision line, which can be used also as a a (Za ) and the sameZa in shunt (ZbImpedance ), of wetwo getimpedances a of structure depicted in Fig. 6.4. 6.4 Tree structure and Zb . Impedance is derived as geometrical mean of Zaof Zb ,cable If we substitute aand resistance Za(solved = circuit R and a Heaviside capacitance Zb =can 1/sC, then the real be expressed as Zb model Za Impedance of such kind of transmision line,line which canby beO.used also asina 1887), ! ! impedance is usbyconsider Za. Let the total impedance the fractance Za .Z = line model of realZ(s) cable (solved O. Heaviside in 1887),ofZ(s) can asshown in Fig. 6.5, Zb Zacircuit .Zb . as (6.7) = expressed b "structure " be ! which has a recursive with the combination of two impedances Z a Zb R −1/2 R mean .derived (6.7) Z(s) = Za .Z If Z aofcapacitance Z = 1/sC, then the andwe Zbsubstitute . Impedance a and| Zb , . b s as geometrical ω −1/2 Z(s) = aisbresistance =a = R and eZ−jπ/4 (6.8) s=jω Za = R and a capacitance Zis then If we substitute a resistance Za Za impedance !C b = 1/sC,C If we substitute a resistance Z = R and a capacitance Z = 1/sC, then the a b Zb Z(s) = Za .Z" " b. the impedance shows the as (see also (6.8)) Zb fractance characteristics R −1/2 R −1/2 −jπ/4 impedance Za " is s ω Z(s) = = e |s=jω . (6.8) Zb " " " R −1/2 R −1/2 C C Z(s) = s ωZb R = e−jπ/4 |s=jω , R a−1/2 we substitute a = R and a capacitance Zb = 1/sC, then C sIf−1/2 ω resistance Z(s)C= = e−jπ/4 |Zs=jω . (6.8) Z(s) the impedance shows the fractance characteristics as (see also (6.8)) C C Z(s) General Approach to Fractances Take Then which is a fractional order integral with absolute value of impedance propor" " A phase self-similar treeiscircuit composed of independent two inpedance Z and Z b. tional toFigure ω −1/26.5: and angle constant −π/4, ofa the frequency. R −1/2 R −1/2 −jπ/4 Z(s) = s ω = e |s=jω , However, it is impossible to construct such an infinity structure as show in C C circuit, where capacitance C was the same at every stage and resistance R Fig. 6.5.increased Therefore, it seems to be important to study the effect of the numby the ratio a at every stage of branching.which The resulting impedance is a fractional order integral with absolute value of impedance proporber of stage in the recursive self infinity (binary) structure on the impedance Z(s) had the form of a continued fraction expansion. tional to ω −1/2 and phase angle is constant −π/4, independent of the frequency. phase characteristics [31].angle is constant !"/4, independent of the frequency However, it is impossible to construct such an infinity structure as show in A similar tree structure as a fractal model ofFig. rough interface between two 6.5. Therefore, it seems to be important to study the effect of the nummaterials of very different conductivities, e.g. anber electrode an electrolyte of stageand in the recursive self infinity (binary) structure on the impedance was studied in [21]. There was described a slightly different structure of tree characteristics [31]. General Approach to Fractances CHAPTER 7. REALIZATION OF FRACTIONAL CONTROLLERS 41 CHAPTER 7. REALIZATION OF FRACTIONAL CONTROLLERS 42 Sample implementations -- transmission lines -- responses: 0.4 0.6 0.4 Output signal: U [V] Y5 0.2 0 !0.2 !0.4 0.2 0 !0.2 !0.6 !0.8 !0.4 0 0.005 CHAPTER 7. 0.01 0.015 0.02 0.025 0.03 [sec] REALIZATION OFTime FRACTIONAL CONTROLLERS 0.005 0.01 0.005 0.01 0.015 0.02 0.025 0.03 0.015 Time [sec] 0.02 0.025 0.03 [sec] REALIZATION OFTime FRACTIONAL CONTROLLERS 44 1 1 0.5 0 unit step !0.5 !1 0 CHAPTER 7. 43 Input signal: E [V] Y4 Z4 0 0.005 0.01 0.015 Time [sec] 0.02 0.025 !0.5 !1 0 0.2 0 !0.2 !0.4 0 0.005 0.01 0.015 Time [sec] 0.02 0.025 1/2 Figure 7.7: Time response of the I controller to sin input where half order integral0.4was approximated by transmission line depicted in Fig. 7.2. 0 !0.2 0.03 0 0.005 0.01 0 0.005 0.01 0.015 Time [sec] 0.02 0.025 0.03 0.015 Time [sec] 0.02 0.025 0.03 1 1 0.5 saw 0 !0.5 !1 sine 0 0.6 Figure 7.6: Time response of the I 1/2 controller to unit-step input where half 0.2 order integral was approximated by transmission line depicted in Fig. 7.2. !0.4 0.5 0.03 0.4 Output signal: U [V] Y3 Output signal: U [V] Y2 Z2 Input signal: E [V] Y1 Output signal: U [V] Za !15 2n 6 Figure 6.3: General structure of transmissionZ(s) line. 6.3 1k ! Transmission lines circuit Input signal: E [V] 2 2 1µF 1µF C 2n-1 6.3 1 1k ! Sample implementations: Figure 7.4: RC binary tree circuit. Then 1 1k ! 1µF 1µF Take 2n 6 1k ! 40 General Approach to Fractances 2n-1 5 4 1µF 1k ! Input signal: E [V] 1 5 1k ! 1µF Z F (s) Phase [deg] 3 1k ! 1µF a b described by a(6.1) with can symetrical distribution of elements, Zladder = Z2k (s) 2k−1 (s) lattice The resulting impedance Z(s) of the as entire circuit beline expressed as a CFE transmission or aline, symetrical Impedance of such ofktransmision which domino can be used also as anetwork as well. (s) = kind Y2k (s), = 1, . . . ,Zn. and Y2k−1 described by (6.1) with symetrical distribution of elements, (s)FRACTANCE =ofZthe 2k−1 2k (s) The resulting impedance Z(s) entire circuit can be expressed as a CFE CHAPTER 6. GENERAL APPROACH TO DEVICES 34 model of real cable line (solved by O. Heaviside in 1887), can be expressed as 6.3 Transmission lines circuit In the case if we assume the same elements in transmision line in series = Y2k (s), k = 1, . . . , n. described and Y2k−1 (s) CHAPTER 6. GENERAL APPROACH TO FRACTANCE DEVICES 35 by (6.1) with ! symetrical distribution of elements, Z2k−1 (s) = Z2k (s) (Z ) and the same in shunt (Z ), we get a structure depicted in Fig. 6.4. a b In the case if we assume the same elements transmision .Z . in. ,series (6.7) Z(s) = a line (s)in =fraction YY2k (s), kkZ= n. which oneZ has to 6.3, obtain aand continuous representation of the Z(s), 2k−1 Let us consider the circuit depicted in Fig. where ZYof (s) and = 1,b . . line, Z Z Z 2k such kind of(s), transmision canfunction be used also as a (Za ) and the same in shunt (ZbImpedance ), we get a2k−1 structure depicted in Fig. 6.4. In case we assume the same elements inexpressed transmision 4thestructure 1, . . . , n, are given impedances of the elements. This is and known Ifsuch we circuit substitute resistance Z2 a(solved =if8 can R a Heaviside capacitance Zb =can 1/sC, then the model ofa real cable line by O.used in 1887), be as line in series 1 + 3s + s 1 Impedance of kind of transmision line, which be also as a Z(s) Y Y Y Y Y Y Y Y ! (6.6) s + Z(s) = = (Z ) and the same in shunt (Z ), we get a structure depicted in Fig. 6.4. a b as a transmission line or a symetrical domino ladder lattice network as well. 3 impedance isZ by O. Heaviside Z 2 modelZ of realZ cable line (solved in 4s 1887), Z(s) can be expressed as 1 2s + Z .Z . (6.7) = a b of as such kind2sof+ transmision !Impedance The resulting impedance Z(s) of the entire circuit can be " expressed a"CFE 1line, which can be used also as a 1 R R Za .Z . real −1/2 (6.7) Z(s) =model If we substitute =R a by capacitance Zb = in 1/sC, then thebe expressed as − Figure 6.3: General structuredistribution of transmission line. O.s Heaviside 1887), can s cable ω −1/2 = aofbresistance =aline e−jπ/4 |+ (6.8) described by (6.1) with symetrical of Z(s) elements, Z2k−1 (s) Z Z (s)and s=jω .3 2k(solved 12 C !− s impedance is a capacitance Za Za we(s), substitute ZZa Zb = C 1/sC, then the a = R and k = Za1, . .a. ,resistance n. and Y2k−1 (s) =IfY2k CHAPTER 6. GENERAL FRACTANCE " APPROACH " TO 2 b . DEVICES 34 (6.7) Z(s) = Za .Z CHAPTER 6. GENERAL APPROACH TO FRACTANCE DEVICES 34 R in−1/2 R −1/2 −jπ/4 impedance is the same In the case if we assume elements in transmision From expansion itZ(s) follows that s series = line = . Z(s) Zb "this Zb Zb Zb Zb Zb Zb Zb If"we substitute aC resistance Zaω = Re and |of as=jω capacitance Z(6.8) b = 1/sC, then the C one a continuous fraction representation the function Z(s), (Za ) and the same in get aR structure depicted in Fig. 6.4. Za shuntZa(Zb ), we Za Zahas to obtain R −1/2 −jπ/41of the function Z(s), 1 =impedance 3 one has to obtain Z(s) a continuous fraction representation s−1/2 ω(s) = eis 4− |s=jω 2 . as Ya2 (s) = (6.8) Z1 (s) s;which = 2s; 1 Y4 (s) = − s. 3 Impedance of such kind of transmision line, be also C 2= CZcan +12 3ss; +8 " sused 1 " 2 (6.6) = = s+ 13 3sinpedance +28 Za1and ZZ(s) s4 of+two Transmission line circuit composed b. R −1/2 R1 −1/2 −jπ/4 model ofFigure real6.4:cable line (solved by= O. Heaviside in 1887), can be expressed as 2 2s + 4s (6.6) s + Z(s) = Therefore, first Cauer’s 2s +form = 1 of ωthe |s=jω . (6.8) 1Z(s) = in sthe 2 analogue realization 2s !3 + 4s for the 1 e C C 2s + 6.4 Tree structure circuit s + − Za .Zcircuit Z(s) = LC 1 (6.7) the following 1 to choose canonic [20] we have b. 3 of coils and 12 values − s+ − s Let us consider the total impedance of the fractance circuit as shown in Fig. 6.5, 3 12 2 If we substitute a resistance Zacapacitors: = combination R and aofcapacitance = 1/sC, then − s the which has a recursive structure with the two impedances ZZ a b 2 that From and Zb . Impedance is derived as geometrical mean of Z1 Zb , this expansion1it follows a and 3 impedance is From this expansion it follows that L [H]; [H]; 1 C2 = 2 [F ]; L" C4 = − [F3]. ! 1 = 3 = − "Z(s) 1 = Z .Z . 12 Z Z s; = s; Y4 (s) 2= − s. 3 (s) = − R1 −1/2a b R 2 −1/2 1 1 (s) 3Y2 (s) = 2s; −jπ/4 2 12 Z(s) =Z3 (s) e |Ys=jω . = 2s; (6.8) Z1= s; (s) = ss; Here − negative Yand = − s. we=ωsee inductances capacitance. Such elements 2cannot 2 (s) 4 (s) C2 C Therefore, 12 2 in the form of the first Cauer’s for the analogue realization Zb = 1/sC, then If we substitute a resistance Za = R and a capacitance bethe realized using passive inelectric components. However, they can be realized Therefore, for analogue realization the form of the first Cauer’s the impedance shows the fractance characteristics as canonic (see also (6.8)) LC circuit [20] we have to choose the following values of coils and " circuit " [20]the with help components, namely operating canonic LC have of to active choose the following values of coils and amplifiers. We can capacitors: R −1/2 R −1/2we Z(s) = s ω = e−jπ/4 |s=jω , C realize this by negative which was described in3 previous capacitors: C 1 inpedance converter 1 [H]; L3 = − [H]; 3 C2 = 2 [F ]; L1 = proporC4 = − [F ]. which is a fractional order integral absolute value of1impedance 1 with subsection. CHAPTER 6.−1/2GENERAL APPROACH TO FRACTANCE DEVICES 35 2 C = 2 [F 12C = − [F ]. 2 L3independent L1 = = − [H]; ]; tional to ω and phase angle is[H]; constant −π/4, of the frequency. 2 4 2 construct such an infinity12 2 capacitance. Such elements cannot Here we see innegative inductances and However, it is impossible to structure as show Z it seems toZ be important Z to study the effect Z of the numFig. 6.5. Therefore, Here we see negative inductances and capacitance. Such elements cannot be realized using passive electric components. However, they can be realized ber of stage in the recursive self infinity (binary) structure on the impedance be realized using passive electric components. However, they cannamely be realized with the help of active components, operating amplifiers. We can characteristics [31]. CHAPTER 6. GENERAL APPROACH TO FRACTANCE DEVICES 35 AZ(s) similarwith as a fractal rough interface between Ytree structure Y Y Y active Ymodel ofYcomponents, Y Ynamely the help of operating amplifiers. can was described in previous realize this by two negative inpedance converterWe which materials of very Zdifferent conductivities, e.g. Z Z an electrode and Zan electrolyte Let us inpedance consider the circuit depicted in Fig. Z which Z Z 6.3, Z Z2k−1 (s) and Y2k (s), k = realize this by negative converter was described in where previous subsection. was studied in [21]. There was described a slightly different structure of tree subsection. 1, . . . , n, are given impedances of the circuit elements. This structure is known 1 REALIZATION OF FRACTIONAL CONTROLLERS 0 0.005 0.01 0.015 Time [sec] 0.02 0.025 ramp 0 !0.5 !1 0.03 Figure 7.8: Time response of the I 1/2 controller to saw input where half order integral was approximated by transmission line depicted in Fig. 7.2. 0.5 1/2 Figure 7.9: Time response of the I controller to ramp input where half order integral was approximated by transmission line depicted in Fig. 7.2. A similar tree structure as a fractal model of rough interface between two materials of very different conductivities, e.g. an electrode and an electrolyte was studied in [21]. There was described a slightly different structure of tree III.1 Commensurate Order Systems General Approach to Fractances Sample implementations: CHAPTER 7. Stability of fractional order systems • What is a commensurate order system? ! ofExample • If one selects the greatest common divisor An Commensurate order the orders, suchsystems: that fractance REALIZATION OF FRACTIONAL CONTROLLERS 38 • A fractional order system An Example An Example where p and q are the orders of the rational approximation, P and Q are polynomials of degree p and q, respectively. Block diagram of the analogue R2 ZF Ri E _ R1 + + _ + + U CHAPTER 7. Figure 7.1: Analogue fractional-order integrator. CHAPTER 7. REALIZATION OF FRACTIONAL CONTROLLERS CHAPTER 7. REALIZATION OF FRACTIONAL CONTROLLERS 39 REALIZATION OF FRACTIONAL CONTROLLERS realization of fractional-order operator is shown in Fig. 7.1. 0.47µ F Z F (s) 0.47µ F 0.47µ F 0.47µ F 0.47µ F 0.47µ F 0.47µ F 7.2 Z (s) 0.47µ F 0.47µ F 0.47µ F 0.47µ F 0.47µ F R 7.2.1 0.47µ F 0.47µ F 0.47µ F 0.47µ F 0.47µ F 0.47µ F 1k ! 1k ! 1µF 1k ! 1k ! 1k ! 1µF 1k ! Realization of fractional I λ controller 22k ! 22k ! 22k ! 22k ! 22k ! 22k ! 22k ! 22k ! 22k ! 22k ! 22k ! 22k ! 22k ! 22k ! 22k ! 22k ! 22k ! 22k ! 22k ! 22k ! 22k ! 22k ! F 22k ! 22k ! 0.47µ F 0.47µ F 0.47µ F 0.47µmeasurement F 0.47µ F For experimental we built a fractional-order I λ controller which 0). The conis transmission a particularline casecircuit. of the P I λ D µ controller, (if Kp = 0 and Td =1µF Figure 7.2: RC Figure 7.2: RC transmission troller line wascircuit. realized in three forms, namely by: the symetrical domino ladder or 1k ! 1k ! 1µF 1µF 1k ! 1k ! 1µF 1µF 1µF 1µFZ impedance 1µF 1µFF where where Magnitude [dB] Z F (s) Transmission line approximation of the s−0.5 Transmission line approximation of the s−0.5 In Fig. 7.5 and Fig. 7.6 - 7.9 the measured characteristics of realized anaIn Fig. 7.5 and Fig. 7.6 - 7.9 the measured characteristics of realized analogue fractional-order I λ controller, where half order integral was approxilogue fractional-order I λ controller, where half order integral was approximated by transmission line depicted in Fig. 7.2, are presented. It can be seen mated by transmission line depicted in Fig. 7.2, are presented. It λcan be seen from Fig. 7.5 that the realized analogue fractional-order I controller provides from Fig. 7.5 that the realized analogue fractional-order I2λ controller provides a good approximation in the frequency range [10 rad/sec, 5 · 102rad/sec]. (For a good approximation in the frequency range [102 rad/sec, 5 · 102rad/sec]. (For comparison see expected Bode plots shown in Fig. 3.2.2.) comparison see expected Bode plots shown in Fig. 3.2.2.) Phase [deg] 2 3 4 10 !80 2 10 3 10 Frequency [rad/sec] FOC tutorial III ©Dingyu Xue, NEU, PR China FOC tutorial III ©Dingyu Xue, NEU, PR China FOC tutorial III ©Dingyu Xue, NEU, PR China !70 !90 14 ©Dingyu Xue, NEU, PR China !=1/6, thus the original order • It is obvious that It is obvious !=1/6, can be written as bethat • system can written as thus the original order system can be written as where 1µF 1k ! 1k ! 1k !line (see 1k !Fig. 7.2) 1k ! Z F (s) ladder (see Fig. 7.3) 1µF transmission for n=6, the finite domino 1k ! 1k ! 1k ! C for n = 6 and the tree structure (see Fig. 7.4) for n = 4, and Fractances with 1µF 1µFconnected 1µF to feedback 1µF were in operational amplifier (Fig. 7.1). 1µF 1µF It should be noted that the described methods work for arbitrary orders, but the circuit elements with computed values are not usually available. Because Figure 7.4: RC binary tree circuit. of this, in our experiment we proposed and realized the integrator with order Figure 7.3: RC domino ladder circuit. !5 λ =ladder 0.5. circuit. Figure 7.3: RC domino !10 It should be mentioned that this simple case of the controller order can !15 we have Ti = 1.4374. The transfer function of the realized analogue fractionalbeofrealized alsoanalogue using the methods described in [31, 33, 38, 54], which do not The transfer function the realized fractionalwe have Ti = 1.4374. !20 order I λ controller is: order I λ controller is: involve explicit rational approximations. !25 (7.2) C(s) = −0.5 1.4374 s−0.5. (7.2) C(s) = 1.4374 s In .the case, if we will use identical resistors (R-series) and identical ca!30 λ Adjustment of the integration constant T of the fractional-order I coni λ !35 pacitors (C-shunt) in the Ifractances, then the behaviour of the circuit will Adjustment of the integration constant Ti of the fractional-order controller (or half order integrator) depicted in Fig. 7.1 was done by resistor Ri . !40 half-order We used the resistor troller (or half order integrator) depictedbe in as Fig.a 7.1 was doneintegrator/differentiator. by resistor Ri . 10 values R = 10 If we change the resistor Ri , the integration constant changes the value in the If we change the resistor Ri , the integration constant thethe capacitor values C = 1µF , (Cj = C, j = Frequency [rad/sec] 1kΩ, (Rj = changes R, j = the 1, . .value . , n) inand required interval. required interval. !30 1, . . . , n). better measurement For measuremets we used frequency 100 For Hz and amplitude ±10 V. results we used two operational amplifiers For measuremets we used frequency 100 Hz and amplitude ±10connection. V. !40 TL081CN in inverting !50 The resistors R1 and R2 are R1 = R2 = 22kΩ. The integration constant Ti ! Experimental !60 7.2.2 7.2.2 Experimental resultsresults can be computed from relationship Ti = 1/ R/(Ri2 ∗ C), and for Ri = 22kΩ 1k ! 1µF Z F (s) • A fractional order system If ! can be found, commensurate order •• A fractional orderthe system • It is obvious that !=1/6, thus the origina For example: model can easily be written system can be written as FOC tutorial III 1µF Description of circuit 0.47µ F 40 39 4 10 Figure 7.5: Bode plots of the I 1/2 controller where half order integral was approximated by transmission line depicted in Fig. 7.2. 15 15 Stability of fractional order systems III.2 Stability of Fractional Order Systems • The ! curve – if , then the stable condition for the system is Unstable region with four poles in the stable region FOC tutorial III ©Dingyu Xue, NEU, PR China 17