TECHNICAL RESEARCH REPORT

Feedback Stabilization of PWM DC-DC Converters

by C.-C. Fang, E.H. Abed

T.R. 98-51

I R
INSTITUTE FOR SYSTEMS RESEARCH

ISR develops, applies and teaches advanced methodologies of design and analysis to solve complex, hierarchical,
heterogeneous and dynamic problems of engineering technology and systems for industry and government.
ISR is a permanent institute of the University of Maryland, within the Glenn L. Martin Institute of Technol-
ogy/A. James Clark School of Engineering. It is a National Science Foundation Engineering Research Center.
Web site http://www.isr.umd.edu

Feedback Stabilization of PWM DC-DC Converters
Chung-Chieh Fang and Eyad H. Abed
Department of Electrical Engineering
and Institute for Systems Research
University of Maryland
College Park, MD 20742 USA

Manuscript: September 28, 1998

Abstract

Feedback stabilization of the nominal periodic operating condition of PWM DC-DC con-

verters is considered, using recently developed general sampled-data models. Two types of

discrete-time washout filter aided feedback stabilization scheme are proposed and studied in

detail. These are voltage reference compensation and dynamic ramp compensation. The stabi-

lization schemes preserve the nominal periodic operating condition through washout filters as

the system parameters vary.

1 Introduction

Pulse-width-modulated (PWM) DC-DC switching converters are widely used in industry and in

consumer products. These converters are designed to operate in a stable fashion. However, in

demanding applications they may be forced to operate in off-design conditions, in which case

stability may be lost. An example of such a setting is a distributed power system where large

disturbances might occur. The purpose of this paper is to develop new stabilization techniques for

PWM DC-DC converters that can be used to ensure stability in an off-design condition. The new

stabilization schemes are developed using general nonlinear sampled-data models recently developed

by the authors [1, 2].

Two discrete-time washout filter aided stabilization schemes are proposed and developed in the

paper: voltage reference compensation and dynamic ramp compensation. The nonlinear nature

1

In the diagram. 2] is given. A block diagram model for a PWM converter in continuous conduction mode [14. In Section 4. and N 2 . In Section 3. 2] is reviewed. In Section 2. and D ∈ R are constant matrices. 11. B2 ∈ RN ×1 . B1 . This includes a general block diagram model as well as associated nonlinear and linearized sampled-data models. x ∈ RN . This paper does not study these bifurcations in detail. pp. 11. A2 ∈ RN ×N . but rather focuses on removing a bifurcation by stabilizing the branch in the parameter range of interest. a general model for PWM DC-DC converters developed by the authors in [1. Several previous works have dealt with stabilization of the nominal periodic solution for a PWM DC-DC converter immediately following a bifurcation [10. Among the bifurcations found in PWM DC-DC converters are period-doubling bifurcation (subharmonic instability) [4. 7]. E1 . a summary of the sampled-data modeling of PWM converters discussed in [1. and others [9]. 13]. targeting methods are used. In [13]. E2 ∈ R1×N . 2 General Sampled-Data Model for PWM Converters In this section.of converter operation implies that the operating condition changes as parameters vary. border-collision bifurcation [8]. Loss of stability in a nonlinear system generally coincides with the occurrence of a bifurcation of the nominal operating condition. a time-delay feedback method is used. Washout filters [3] are used in order to preserve the nominal operating branch. namely voltage reference compensation and dynamic ramp compensation. Neimark-Sacker bifurcation [2]. y ∈ R are the state and the feedback signal. 6. 12. 165-168] is shown in Fig. 5. two illustrative examples are given. the two stabilization schemes proposed in this work are presented. 1. C. Conclusions are collected in Section 5. The remainder of the paper is organized as follows. respectively. For the stabilization schemes proposed here. an added benefit is robustness in the sense that the nominal operating branch is preserved. In [10. saddle-node bifurcation [2]. The contribution of the present work lies in development of new stabilization schemes that are general and widely applicable. A1 . 12].

322-336] and current mode control [14. S2 : x˙ = A2 x + B2 u vo = E2 x . Within a clock period.vo Figure 1: Block diagram model for PWM converter operation in continuous conduction mode h(t) y(t) Switch S1 S2 S1 S2 S1 S2 Figure 2: Waveforms for a PWM converter under voltage mode control 3 . The source voltage is vs . The signal h(t) is a T -periodic ramp. pp. The notation vr denotes the reference signal. although it is constant in most applications. It is used to model a compensating ramp in current mode control. typically given by the number of energy storage elements in the converter. which could be a voltage or current reference. pp.is the state dimension. 337-340]. the dynamics is switched between the two stages S1 and S2 . The clock has the same frequency fs = 1/T as the ramp.  h(t) = Vl + (Vh − Vl )( Tt mod 1) Switching  clock Decision  y = Cx + Du Switch to S1 or S2 ? ( x˙ = A1 x + B1 u S1 : vo = E1 x ( v u=( s ) vr . This frequency is called the switching frequency. 1 is applicable both to voltage mode control [14. The system is in S1 immediatedly following a clock pulse. Figure 2 illustrates the signal waveforms in the case of voltage mode control. and the output voltage is vo . The reference signal vr is allowed to be time-varying. The model in Fig. and switches to S2 at instants when y(t) = h(t).

g. Let xn = x(nT ) and von = vo (nT ). vrn ). 1 has the following sampled-data dynamics: xn+1 = f (xn . (n + 1)T ). dn ) = (x0 (0). Here we only show stabilization of the nominal T -periodic UPO. dn ) = (xn . vr ) to be constant within the cycle. Denote by nT + dn the switching instant within the cycle when y(t) and h(t) intersect. the system (1). Take u = (vs . (2) has the linearized dynamics ˆn+1 = Φˆ x xn + Γˆ un = Φˆ xn + Γ1 vˆsn + Γ2 vˆrn (3) where ((A1 − A2 )x0 (d) + (B1 − B2 )u)C A1 d Φ = eA2 (T −d) (I − )e ˙ C(A1 x0 (d) + B1 u) − h(d) (x˙ 0 (d− ) − x˙ 0 (d+ ))C A1 d = eA2 (T −d) (I − )e (4) C x˙ 0 (d− ) − h(d) ˙ Z Z Z d x˙ 0 (d− ) − x˙ 0 (d+ ) d T −d Γ = eA2 (T −d) ( eA1 σ dσB1 − (C eA1 σ dσB1 + D)) + eA2 σ dσB2 (5) 0 C x˙ 0 (d− ) − h(d) ˙ 0 0 3 Discrete-Time Washout Filter Aided Stabilization Schemes There may exist unstable periodic orbits (UPOs) with different periods when the PWM converter is unstable. x ˆn = xn − x0 (0)). [Vs . the system in Fig. dn ) Z dn Z T A2 (T −dn ) A1 (dn −σ) = e A1 dn (e xn + e dσB1 un ) + eA2 (T −σ) dσB2 un (1) 0 dn Z dn g(xn . d). Then. Let the fixed point of the system (1). un . un . Consider the cycle t ∈ [nT. un . (2) be (xn .. Using a hat ˆ to denote small perturbations (e. 1 corresponds to a point x0 (0) in the sampled data dynam- ics (1). and denote its value by un = (vsn . vrn ]0 . A similar approach 4 . [vsn . (2. dn ) = C(eA1 dn xn + eA1 (dn −σ) dσB1 un ) + Dun − h(dn ) 0 = 0 (2) A periodic solution x0 (t) in Fig. Vr ]0 .

The closed-loop system (1). vr . vr is updated at the clock time. and h(t) (or equivalently y). feedback stabilization can be achieved by adjusting vs .can be applied to stablize a general nT -periodic UPO. In steady state. The system diagram is shown in Fig. Γ2 ) is stabilizable. vˆrn =0. K2 ∈ R are the feedback gains and K2 6= 0. 1. 5 . (7) has the following linearized dynamics " # " # " # " # x ˆn+1 Φ 0 Γ2 h i x ˆn =( − K1 K2 ) (8) w ˆn+1 0 1 1 w ˆn One has the following result. feedback stabilization is achieved by adjusting vr or h(t). We will show this through an example (for n=2) in Sec. K1 ∈ R1×N . (2) is preserved. (2) is asymptotically stabilizable by the washout filter aided voltage reference control scheme if the following conditions are satisfied: (i) (Φ. then the eigenvalue at 1 (introduced by the washout filter) is controllable. In this section. " # Φ − I Γ2 (ii) is of full rank. 3. where Vr is the nominal voltage reference value. Denote it as vrn = Vr + vˆrn . The proposed washout filter aided discrete-time controller is wn+1 = −K1 xn + (1 − K2 )wn (6) vˆrn = −K1 xn − K2 wn (7) where wn ∈ R is the state of the washout filter. 4. Theorem 1 The nominal fixed point of system (1). From Fig. (2). Thus the original fixed point xn = x0 (0) in the system (1). Generally vs undergoes large variations and is therefore not a candidate as a control variable.1 Washout Filter Aided Voltage Reference (vr ) Compensation In the first stabilization scheme. 0 1 Proof: If condition (ii) holds. 3. (6).

The signal h(t) is called a dynamic ramp. The system diagram is shown in Fig.  h(t) = Vl + (Vh − Vl )( Tt mod 1) Switching  clock Decision  y = Cx + Du Switch to S1 or S2 ? vs . S2 : vo = E2 x x 6 + vrn + vˆrn ? Discrete-Time Controller  xn Sampler Figure 3: Stabilization by voltage reference (vr ) compensation by the PBH rank test [15].2 Washout Filter Aided Dynamic Ramp Compensation Next. a scheme is presented in which the slope of the ramp h(t) is changed by state feedback. Let λ be any eigenvalue of Φ which lies outside the unit circle. This " # Φ − λI 0 Γ2 eigenvalue is controllable if is of full rank. 4.vo  vrn x˙ = A2 x + B2 u Vr . to distinguish it from the traditional fixed-slope compensating ramp. The proposed discrete-time controller is wn+1 = −K1 xn + (1 − K2 )wn (9) vˆhn = −K1 xn − K2 wn (10) 6 . 2 3. . This matrix is of full rank if 0 1−λ 1 condition (i) holds. ( x˙ = A1 x + B1 u  S1 : v u=( s ) ( vo = E1 x .

where Vh is constant. (9). Let h(t) = Vl + (vh − Vl )( Tt mod 1). K1 ∈ R1×N . the original fixed point xn = x0 (0) in the system (1). Let vh be updated in each cycle and denoted as vhn = Vh + vˆhn . vhn ) = C(eA1 dn xn + eA1 (dn −σ) dσB1 un ) + Dun − Vl − 0 T = 0 (11) The closed-loop system (1). dn . now becomes Z dn (vhn − Vl )dn g(xn . The switching constraint equation. Sampler x˙ = A2 x + B2 u u=( vs vr ) . As was the case with the washout filter aided voltage reference compensation. un . S2 : vo = E2 x . (2) is preserved. (2). (11). (10) has the linearized dynamics " # " # " # " # x ˆn+1 Φ 0 Γh h i x ˆn =( − K1 K2 ) (12) w ˆn+1 0 1 1 w ˆn 7 .vo Figure 4: Stabilization by dynamic ramp compensation where wn ∈ R is the state of the washout filter. Changing the slope is equivalent to changing vh . Eq. h(t) = Vl + (Vh + vˆhn − Vl )( Tt mod 1)  Dynamic Ramp Switching Decision  clock Generator 6  vˆhn y = Cx + Du Discrete-Time Controller Switch to S1 or S2 ? 6 xn ( x˙ = A1 x + B1 u S1 : ( vo = E1 x x . K2 ∈ R are the feedback gains and K2 6= 0.

where .

∂f ∂f ∂g −1 ∂g .

.

Φ = − ( ) ∂xn ∂dn ∂dn ∂xn .

dn .vhn )=(x0 (0).Vh ) (x˙ 0 (d− ) − x˙ 0 (d+ ))C A1 d = eA2 (T −d) (I − )e (13) C x˙ 0 (d− ) − VhT−Vl .(xn .u.un .d.

∂f ∂g −1 ∂g .

.

Γh = − ( ) ∂dn ∂dn ∂vhn .

and that at least " # Φ − I Γh one of these matrices have all eigenvalues with negative real part. Vh = 8. one has Theorem 2 Assume that A1 and A2 have no eigenvalues with positive real part.1 in Appendix A. L = 20mH.8V . Thus the pair (Φ. g1 = 8. R = 22Ω. C = 47µF . and let Vs be the bifurcation parameter.2V implying h(t) = 3. Vl = 3. For this circuit. the eigenvalues of which lie inside the unit circle by Proposition A. (Discrete-time stabilization of a buck converter under voltage mode control) Consider the buck converter under voltage mode control shown in Fig.dn . Vr = 11.(xn . N =2 and the matrices in the model of 8 .4. Proof: Let K1 = − Td CeA1 d .u. then Φ − Γh K1 = eA2 (T −d) eA1 d .d.un . (2) is asymptotically stabilizable using the washout filter aided dynamic ramp scheme.8 + 4.4[ Tt mod 1]. the system (1). 2 4 Illustrative Examples Example 1.3V . Γh ) is asymptotically stabilizable.Vh ) (x˙ 0 (d− ) − x˙ 0 (d+ ))d = eA2 (T −d) (14) (C x˙ 0 (d− ) − VhT−Vl )T Analogous to Theorem 1. Let T = 400µs.vhn )=(x0 (0). The rest of the proof is similar to that of Theorem 1. 5. If the matrix is 0 1 of full rank.

6.Fig.66V . 1 with state x = (iL . ) (Eq. vC ) are given by [2] " # −1 0 L A1 = A2 = 1 −1 " # C RC " # 1 0 L B1 = B2 = 0 0 h i h i C = 0 g1 D = 0 −g1 h i E1 = E2 = 0 1 The bifurcation diagram for the circuit is shown in Fig. In the switching converter. (12)) for this example 0 1 1 0 1 1 can be shown to be controllable. assigning all of the poles of the closed-loop system to zero is called dead-beat control. the dynamics is nonlinear. the system is driven to the origin in finitely many (N ) steps. Here stabilization of this system is demonstrated by using washout filter aided voltage reference or dynamic ramp compensation. + Ramp h(t) Comparator − g1 Vr y − + L iL + + + Vs C Vc R Vo − − − Figure 5: System diagram for Example 1 " # " # " # " # Φ 0 Γ2 Φ 0 Γh The pairs ( . This is because with such a control law. so the dead-beat effect cannot be guaranteed to occur. so all of the eigenvalues of the closed-loop system can be assigned to the origin. In an N -dimensional linear time-invariant discrete system. The circuit is chaotic for Vs = 34. (8)) and ( . ) (Eq. but fast stabilization is expected if a linear dead-beat control is used. 9 .

6622. 12. Originally the system is chaotic.8 11.0160) and K2 = 0. 10 shows the bifurcation diagram with the same feedback gains. The effectiveness of washout filter aided dynamic ramp compensation in stabilizing the UPO is illustrated in Fig. Fig.4 12.2403. Next. Similar to the result of using vr as control variable. The feedback gains used in this illustration are K1 = (−21. 8. 9.4809. 5 First.6 15 20 25 30 35 Vs Figure 6: Bifurcation diagram for the circuit in Fig.3 12.5 Output Voltage Sampled at Switching Frequency 12. the chaotic trajectory is stabilized to a period-one orbit in around 3 switching periods. The stable operating range is now extended up to a source voltage of Vs = 35V .2 12. In this example. Fig. −6. with the complicated algorithm proposed in [11]. the periodic solution is stabilized in around three switching periods. dynamic ramp control is applied. not in any higher periodic mode.7 11.0048. When the control scheme is applied at t = 0. To show the flexibility of the control methods 10 . a chaotic system is stabilized by simple state feedback. It is known that in a chaotic system there are an infinite number of unstable periodic solutions embedded in the chaotic orbit. which also makes all of the closed- loop eigenvalues at 0.1 12 11. 7 shows the effectiveness of the control law.4655) and K2 = 0. Generally switching converters are designed to operate in the period-one mode.6 12. use vr as the control variable and assign all of the closed-loop eigenvalues to 0. The feedback gains achieving this are K1 = (−1.2403. −0.9 11. which can be compared. The stabilized period-one orbit is shown in Fig. for instance.

66V 11 .4 0.5 0.1 12 11.75 Inductor current (A) Figure 8: Stabilized period-one orbit in state space for Vs = 34.4 12.8 11.55 0.5 0.45 0.3 12.5 12.7 0. control turned on at t = 0.9 11.4(Vo-Vr) 12 10 (Volt) 8 6 4 2 3 4 5 6 7 8 9 −3 Control turned on Time (second) x 10 Chaotic Stabilized Figure 7: Stabilization of the buck converter in Fig.65 0.0048 12.6 11.6 0. 5 using washout filter aided voltage reference compensation. 14 Ramp h(t) Vcon(t)=8.35 0.2 Output voltage (V) 12.7 11.

control turned on at t = 0. For this circuit.006616. 13. Washout filter aided voltage reference compensation is used. The sampling rate for feedback stabilization in this case is half of the switching frequency.4(Vo-Vr) 12 10 (Volt) 8 6 4 0 1 2 3 4 5 6 7 8 −3 Control turned on Time (second) x 10 Chaotic Stabilized Figure 9: Stabilization of the buck converter in Fig. and the stabilized period-two orbit is shown in Fig. 12. Similar to stabilization of the period-one mode. 14 Dynamic ramp Vcon(t)=8. and Vr (current reference) is taken to be the bifurcation parame- ter. vC ) are given 12 . Example 2 (Discrete-time stabilization of a boost converter under current mode control) Con- sider the boost converter under current mode control shown in Fig.59) and K2 = 0.23426.0032 presented in the paper. A simulation showing stabilization of the period-two orbit is given in Fig. stabilization of a period-two orbit is demonstrated next. −0. where T = 100µs. C = 12µF . All of the closed-loop poles are assigned to zero using feedback gains K1 = (0. N =2 and the matrices in the model of Fig. 1 with state x = (iL . L = 1mH. the period-two dynamics can be derived and linearized. Vs = 10V . R = 20Ω. 11. 5 using washout filter aided dynamic ramp compensation.

1 12 11.2 12 11.025 Time (second) Figure 11: Stabilization of period-two orbit.7 11.6 11.6 12.8 12.8 11.01 0.5 Output Voltage Sampled at Switching Frequency 12.4 Output voltage (V) 12.6 15 20 25 30 35 Vs Figure 10: Bifurcation diagram for Example 1 using washout filter aided dynamic ramp compen- sation 13 12.005 0.3 12.2 11 0 0. control turned on at t = 4.9 11. 12.4 11.015 0.8 × 10−3 13 .02 0.8 11.6 12.4 12.2 12.

Here stabilization of the nominal period-one mode using washout filter aided dynamic ramp compensation for Vr = 1. Ackermann’s formula [15] can be used to schedule the gain. X Y Plot 12. The performance has been compared with traditional fixed-slope compensation with a slope of 14 .4 12.5 12.7 11.4 0. 14.1 12 11.3 12. 15.7 0.66V by [2] " # " # " # −1 1 0 0 0 L L A1 = −1 A2 = 1 −1 B1 = B2 = 0 0 h i RC h C RC i h i C = 1 0 D = 0 −1 E1 = E2 = 0 1 The bifurcation diagram for the circuit is shown in Fig. 16.6 11. The circuit is in period-two mode for Vr = 1. Therefore 0 1 1 the closed-loop eigenvalues can be placed so as to achieve dead-beat control (similar to the approach in the preceding example).65 0. The resulting bifurcation diagram is shown in Fig.85 will be demonstrated.75 Inductor current (A) Figure 12: Stabilized period-two orbit in state space for Vs = 34.5 0.6 0. If stabilization of a wide range of Vr values is desired.9 11. ) is controllable for this example. " # " # Φ 0 Γh It can be checked that the pair ( . Using dynamic ramp compensation.8 11.2 Output voltage (V) 12.35 0.45 0.85.55 0.5 0. the period-one mode can be stabilized (in around three switching periods) so that it replaces the period-two mode as shown in Fig.

5 4 4.5 5 5.5 2 1.5 2 2.5 Inductor current sampled at switching frequency 4 3. 13 15 .5 5 4. Clock − Vr Comparator + S R y Q L iL + + + Vs C Vc R Vo − − − Figure 13: System diagram for Example 2 5.5 1 1.5 3 3.5 Vr Figure 14: Bifurcation diagram for the circuit in Fig.5 3 2.5 1 0.5 0.

4 4.5 4 4.5 0.3 × 10−3 5. 13 using washout filter aided dynamic ramp compensation.5 1.1 Vr and dynamic ramp 1 3.2 t −3 x 10 Figure 15: Stabilization of the boost converter in Fig.5 1 1.8 4 4.5 3 3.2 Inductor current 1.5 3 2.7 1.8 Inductor current.6 4.5 Inductor current.5 2 1. control turned on at t = 4. 2 1.8 5 5. sampled at switching frequency 4 3.5 5 4.2 4.5 2 2.5 1 0.5 5 5.3 1.6 1.5 Vr Figure 16: Bifurcation diagram for Example 3 using dynamic ramp compensation and gain schedul- ing 16 .9 1. Vr and dynamic ramp 1.4 1.

pp. 1994. no. and R. Abed.H. 25–36. 1. pp.H. [3] E. Conditions for stabilizability have been derived. vol. 154–164. subharmonics. 1992. J. Details are in [2]. University of Maryland. Feb.” preprint. Washout filters ensure that the nominal operating branch is unaffected by the control. [5] D. 41. [2] C. 1998. 1992. 5. “Stabilization of period doubling bifurcations and implications for control of chaos.H. The same schemes can be used to stabilize an unstable period-two orbit or a higher order orbit.K. Hamill. [4] J. [7] C. 39.D.O.B. at the expense of altering the nominal operating branch. pp. “Flip bifurcation and chaos in three-state boost switching regulators. no. 17 . thesis. 680–683.C. Wang.S. 1997. 260–268. Deane. College Park. Fang and E. Chen. no.” Physica D.-C. vol. and J. pp.-C. 7. vol.” IEEE Transactions on Power Electronics. vol. The proposed stabilization schemes are simple and systematic. Jefferies.B. “Instability. the U. “Chaos in a current-mode controlled boost DC-DC converter.C. H.” IEEE Transactions on Power Electronics. A delay in the period-doubling bifurcation was achieved. 1. Fang. Abed. 1-2. and chaos in power electronics circuits. Tse. 3. 16–23.C. and by a Senior Fulbright Scholar Award. no. References [1] C.H.H. Ph. 1990. pp. Two schemes were proposed: voltage reference compensation and dynamic ramp compensation. Sampled-Data Analysis and Control of DC-DC Switching Converters. [6] J. 5 Concluding Remarks Feedback stabilization of the nominal periodic orbit in the PWM DC-DC converter has been studied. 8. Hamill. no. vol. Air Force Office of Scientific Research under Grant F49620-96-1-0161. Deane.” IEEE Trans- actions on Circuits and Systems-I: Fundamental Theory and Applications. 70. 1994. both implemented through a washout filter.value 10000 Amp/sec. “Modeling of chaotic DC-DC converters by iterated nonlinear mappings. Deane and D. “Sampled-data modeling and analysis of PWM DC-DC converters I. Closed-loop circuits. Acknowledgments This research has been supported in part by the the Office of Naval Research under Multidisci- plinary University Research Initiative (MURI) Grant N00014-96-1-1123.B. without the need for accurate knowledge of the equilibrium.” IEEE Transactions on Circuits and Systems-I: Fundamental Theory and Applications.

so all of the eigenvalues of A1 and A2 are in the open left half of the complex plane.P. A A Stability Lemma Generally the converter will dissipate energy due to the inherent resistance.” IEEE Transactions on Circuits and Systems-I: Fundamental Theory and Applications.. 44. [12] P. Control of Oscillations and Chaos Proceedings.” in Fifth Annual Applied Power Electronics Conference and Exposition. Mohan. [14] N. 1980. 1997.” Electronics Letters (IEE). Deane. Wiley. and G.” in IEEE Power Electronics Specialists Conf. Podder. A. [13] C. Ott. and Design. E. Podder. with applications to power electronics. [15] T. K. “Types of instability encountered in simple power electronic circuits: unboundedness.. pp. 42. vol. (None can be in the right half of the complex plane since the circuit is a single RLC circuit between switching instants. Rec. and W. Yuan. simultaneously. Englewood Cliffs. 590–593. Yorke. which involves P capacitor voltages and inductor currents ( 12 (Li i2Li + Ci vCi 2 )).T. 1995. 1337–1344. [8] S. pp. H.C. 1990. J. K.M. vol. Prentice-Hall. and G. no. then A1 or A2 may have some eigenvalues on the imaginary axis. t2 > 0 under the spectral assumption above. and chaos.B. Batlle. Banerjee. 18 . 25. pp. Olivar. no. J. Thus there exists a symmetric posi- tive definite matrix P ∈ RN ×N such that AT1 P + P A1 is negative semidefinite and. Chakrabarty. NJ. 3. Banerjee. [10] G. pp. pp. Robbins. Linear Systems.) If some of the resistances in the circuit are not modeled. Aston. and S. 8. Krein and R. Hamill. 100–101. [9] P.J. Undeland. 11. pp. 1997. 31. The stored energy in the circuit in stage S1 or S2 is given by the same formula. vol. 191–194. and D. vol. “Control of chaos in the boost converter.H. it will be shown that all of the eigenvalues of eA2 t2 eA1 t1 are inside the unit circle for any t1 . Fossas. E. Chakrabarty. [11] G. New York. “Targeting in systems with discontinuities. Banerjee. Kailath. Power Electronics: Converters. 10. chattering.” IEEE Transactions on Circuits and Systems-I: Fundamental Theory and Applications. “Time-delay stabilization of the buck converter. Applications. AT2 P + P A2 is negative definite. 1997. 1995. no. Bass. In this Appendix. 1034–1039. and S. Assume Re[σ(A1 )] ≤ 0 and Re[σ(A2 )] < 0.” in International Conference.M. T. “Experimental control of chaotic behavior of buck converter. “Anomalous bifurcations in DC-DC convert- ers: Borderline collisions in piecewise smooth maps. 1995.

along with Lemma A. 2 19 . where z ∈ RN is arbitrary. Then there exists a symmetric positive definite matrix P ∈ RN ×N such that for any t1 . Suppose P ∈ RN ×N is symmetric positive definite. t2 > 0. note that T T eA1 t1 eA2 t2 P eA2 t2 eA1 t1 − P T T T T = eA1 t1 eA2 t2 P eA2 t2 eA1 t1 − eA1 t1 P eA1 t1 + eA1 t1 P eA1 t1 − P T T T = eA1 t1 (eA2 t2 P eA2 t2 − P )eA1 t1 + eA1 t1 P eA1 t1 − P which is negative definite. then eA t P eAt − P is negative definite for any t > 0. all of the eigenvalues of eA2 t2 eA1 t1 lie strictly within the unit circle.1 Let the matrices A1 and A2 of the circuit model of Fig. T Proof: Let Q(t) = z T (eA t P eAt − P )z. If AT P + P A T is negative definite. T (i) eA1 t1 P eA1 t1 − P is negative semidefinite T (ii) eA2 t2 P eA2 t2 − P is negative definite (iii) Re[σ(eA2 t2 eA1 t1 )] < 0 Proof: Conclusions (i) and (ii) follow form the energy considerations discussed earlier in this Appendix.Lemma A. 1 be such that Re[σ(A1 )] ≤ 0 and Re[σ(A2 )] < 0.1 Let A ∈ RN ×N . The function Q(t) has the following properties: T ˙ Q(t) = z T eA t (AT P + P A)eAt z < 0 Q(0) = 0 Thus Q(t) < 0 for any t > 0 and the claim follows.1. Now. 2 The main result of this Appendix is Proposition A. Thus.