Você está na página 1de 7

IEEE Power Electronics Specialists Conference, St. Louis, Missouri, June 22-27, 1997.

17

AUTOMATED STEADY-STATE ANALYSIS OF SWITCHING POWER CONVERTERS USING A GENERAL-PURPOSE SIMULATION TOOL
Dragan Maksimovi c Power Electronics Group Department of Electrical and Computer Engineering University of Colorado, Boulder, CO 80309-0425 Phone: 303492-4863, Fax: 303492-2758, maksimov@colorado.edu

Abstract The paper describes methods for automated steady-state analysis of switching power converter systems using a general-purpose simulator such as PSpice. The convergence to steady-state is accelerated using quasi-Newton iteration in a steady-state solver implemented in Mathematica. The solver updates initial conditions in the circuit netlist description and launches the simulator to obtain the system response over a switching period. The main advantage of this approach is that steady-state converter characteristics can be obtained e ciently for any converter con guration and any model complexity supported by the general-purpose simulator. Several application examples using PSpice and PETS simulation tools are included.

Run transient simulation steady-state solver (Mathematica) x F (x)


T

file.cir file.out

simulator (PSpice or PETS)

Figure 1: Steady-state solver runs an external, generalpurpose transient simulator over an interval T with initial
conditions x to obtain the response FT x. and implementation of a computer-aided steady-state solver using two general-purpose software tools: Mathematica to implement the solver functions, and a general-purpose simulator PSpice or PETS to generate the system responses, as shown in Fig. 1. The problem of nding a steady-state solution in a switching power converter system is reviewed in Section 2. Numerical solution methods suitable for this application are described in Section 3. Several application examples are presented in Section 4.

1 Introduction

Many of design speci cations in a power electronic system are given in terms of the system's steady-state characteristics. Periodic steady-state solution is also a prerequisite for smallsignal dynamic modeling. Steady-state analysis is therefore indispensable component of the design process. Analytical steady-state solution techniques can often be supplemented with computer-aided analysis tools. A brute-force simulation method to reach a steady-state solution by running out of transients is not practical because the simulated circuit usually has time constants much longer than a switching period, which results in excessive simulation times. Methods for accelerated convergence to steady state have been proposed 1 - 4 , but the implementations have been tied to special-purpose simulation tools where only simple, piecewiselinear component models are supported. General-purpose simulators such as PSpice and other Spice derivatives support arbitrarily complex, nonlinear device models, but their application to power electronic systems is often hindered by convergence problems, excessive simulation times and the lack of a steady-state accelerator 6 . The purpose of this paper is to describe numerical methods
This work was supported in part by the National Science Foundation under grant ECS-9308764.

2 Steady-state in switching power converters

We consider a switching power converter described by _ = Gt; x ; xt0  = x0 ; t  to ; x 1 where x is a vector of n state variables capacitor voltages and inductor currents. By integrating 1 from t = t0 to t = t0 + , we can nd the vector of state variables at t = t0 + as xt0 +  = F x0 ; t0  : 2 In steady state if it exists, the state variables xt satisfy 1 and are periodic in time with period T , so that xt0 + T  = FT x0 ; t0  = x0 ; 3 where FT is a nonlinear mapping obtained by integrating 1 over the period T . Since the steady-state solution is periodic in time, we can assign t0 = 0 arbitrarily and drop the dependence on t0 in 3. Solving for steady-state reduces to nding a

IEEE Power Electronics Specialists Conference, St. Louis, Missouri, June 22-27, 1997.

27

vector of state variables x that solves the system of n nonlinear equations f x = x , FT x = 0 : 4 The approach taken in this paper is to obtain FT x numerically by running transient simulation from t = 0 to t = T , starting with initial conditions x, as shown in Fig. 1. The steady-state solver updates the circuit netlist file.cir with initial conditions x and launches the simulator. The simulator writes the waveforms xt in an output le file.out. The steady-state solver reads the output le and obtains FT x = xT . As shown in Fig. 1, the steady-state solver has been implemented in Mathematica 12 . The solver supports PETS 7, 8 and PSpice simulation tools. We consider two classes of switching converter systems: constant-frequency non-autonomous systems where the constant switching period Ts is determined by an external clock, and autonomous systems where the steady-state period T depends on circuit parameters and operating conditions values of independent input sources. In non-autonomous systems, FT x is obtained simply by setting the nal time of transient simulation to T . In most cases we are interested in a steady-state solution with the period equal to the switching period, T = Ts . In general, we may also set the simulation time to T = mTs in order to nd a sub-harmonic steady-state solution. An example of this application can be found in 9 . For autonomous systems, obtaining FT x is slightly more involved because T is not known and because in a generalpurpose simulator the nal simulation time is constant, independent of simulation results. We follow the approach described in 4 . In an autonomous switching converter system, T is determined by zero crossing of a boundary condition C x:

3.1 Fixed-point iteration

The simplest method we consider is the xed-point iteration, xk+1 = FT xk  : 6 The method is essentially equivalent to the brute force method to run out of transient by long simulation. In a switching power converter, transients can decay very slowly, and the xed-point iteration can require a prohibitively large number of simulation runs Ns = Ni . This method was used mainly to test the implementation of FT x, and to provide a reference for comparison with other methods.

3.2 Quasi-Newton methods

A wide variety of Newton-based methods are available to numerically solve the f x = 0 problem. The key idea in all of these methods is to approximate f x in the kth iteration with an a ne model: f x  f  x = Jk x , xk  + f xk  ; 7 and to obtain the next iterate xk+1 by solving f  xk+1 = 0,

xk+1 = xk , Jk

,1 , k ,

FT xk  :

8

C xT , 0 ; C xT  = 0 ; C xT + 0 :

5

Zero crossing of the boundary condition corresponds to a switching instant in the power converter. Therefore, in most cases, the boundary condition can be evaluated by monitoring a switch drive signal. To obtain FT x, the simulator is set to run transient simulation over a constant interval Tx . When the simulation is completed, T is found as the rst instant when 5 is satis ed. If no zero-crossing of the boundary condition is detected, the simulation is repeated starting with the initial conditions xTx. Because of the way the mappings FT x are de ned for non-autonomous and autonomous systems, the same solution methods can be applied in both cases.

Various Newton-based methods di er in how the matrix Jk is obtained. In the Newton-Raphson method, Jk is the Jacobian, x = I , @ FT x ; Jk = @ f@x 9 @x evaluated at xk . Assuming idealized piecewise-linear converter model, it has been shown how the Jacobian can be obtained efciently and with high numerical accuracy from the converter state-space description 1 . An alternative technique is based on the adjoint network approach 10 . The adjoint network approach to compute the Jacobian, including computation of switching time sensitivities, was developed for steady-state solvers described in 2 and 4 . All of these approaches rely on details of network description available in special-purpose simulator implementations and are limited to piecewise-linear device models. In order to apply a general-purpose simulator, we consider alternative numerical methods available for the cases where the Jacobian 9 is not available. A conceptually simple approach is to use the Newton iteration 8 and to approximate elements of the Jacobian matrix using nite di erences. The Jacobian 9 evaluated at x is approximated by xi  , FT x ; i = 1; : : : ; n 10 ji = ui , FT x + ui xi th where ji is the i column of J, and ui is the ith column of the identity matrix I. A total of n + 1 simulation runs is required to perform one iteration in the quasi-Newton nite-di erence method. Problems associated with 10 are the loss of numerical precision as xi approaches zero or excessive errors with respect

3.2.1 Finite-di erence quasi-Newton method

3 Solution methods
In this section, we brie y describe several numerical methods used to nd a periodic steady-state solution, i.e., to solve the set of n nonlinear equations 4. Ni is used to denote the total number of iterations, Ns is the total number of simulation runs evaluations of FT x, k is the iteration count, and xk is the vector of state variables initial conditions in the current iteration.

IEEE Power Electronics Specialists Conference, St. Louis, Missouri, June 22-27, 1997.

37

to true Jacobian coe cients if xi are too large. Following the guidelines in 11 pp. 94-99, the step size xi is selected as a fraction of the maximum absolute value that the state variable xi attains during the simulation interval T :
iteration failed

create netlist file.cir read file.cir, setup

xi = p Maxjxi tj; xm  ; 0  t  T ;

11

where is an estimate of the relative error in computing FT x, and xm gives a lower bound for the step xi . The nite-di erence quasi-Newton method 8 and 10 requires Ns = n + 1Ni simulation runs.

x 0 -> F T (x

) file.cir

Compute J 0 (Eq.(10))

run simulator file.out

3.2.2 Quasi-Newton method with Broyden's update


Because of the relatively high cost and the relatively high error in computing FT x by simulation, we consider a method where the Jacobian is computed using nite di erences 10 only in the initial iteration. In the subsequent iterations, the initially computed Jacobian is then updated using only one evaluation of FT x per iteration. One of the most frequently applied update methods is called the rank-one Broyden's update 11 pp. 169-189, which for the problem 4 can be written as:

Compute x k+1 (Eq.(8))

x k+1 -> F T (x

k+1

If e >= e

Compute J k+1 (Eq.(10)) or Update J k+1 (Eq.(12))

Jk+1 = Jk + x

, k+1 ,

FT xk+1  xk+1 , xk T : jjxk+1 , xk jj2

,

no

e < e

12

yes steady-state x s = x
k+1

The idea behind this update scheme is to make the least change in the previously computed Jk subject to the condition that the a ne model 7 updated with Jk+1 satis es f  xk+1  = f xk+1 . It can be shown that su ciently close to a solution the Broyden's update quasi-Newton method has superlinear convergence rate, close to the quadratic convergence rate of the Newton-Raphson iteration 11 pp. 174-186. The cost of single iteration is reduced to only one simulation run. Therefore the method requires Ns = n + Ni simulation runs.

STEADY-STATE SOLVER (Mathematica)

SIMULATOR (PSpice or PETS)

Figure 2: Implementation of the quasi-Newton methods described in Section 3.2. The steady-state solver and the simulator exchange data through the circuit netlist le file.cir and the simulator output le file.out, as shown in Fig. 1. relative errors in the state variables that are close to zero. The convergence criterion is then

3.3 Implementation
The solution methods have been implemented using two general-purpose software packages: a simulator to obtain FT x given the initial conditions x and a steady-state solver, to update the netlist file.cir, launch the simulator, read the simulation results from the output le file.out, and perform computations. The steady-state solver has been written as a Mathematica 3.0 12 package steady.m. The solver supports two simulation tools: PETS 7, 8 , and PSpice. Fig. 2 shows a owchart for the quasi-Newton methods described in Section 3.2. The convergence criterion is based on the requirement that the maximum relative error e in the state variables is smaller than a prescribed limit e0 . The error e is de ned by

e e0 ;

15

jfi xj ; i = 1; : : : ; n ei = Max xi ; xmin  e = Maxei  ; i = 1; : : : ; n

13 14

where fi is the ith components of f x, xi is the ith state variable, and xmin is a small constant used to suppress large

where e0 is the speci ed error limit. If the iteration fails to converge after a prescribed number of simulation runs, the user may choose to modify the starting point x0 , or the parameters xm , , xmin , or e0 that a ect execution of the algorithm. The achievable error limit e0 depend mainly on the relative error in computing FT x. Once the convergence criterion is satis ed, the simulator or the steady-state solver can be used to generate the steadystate waveforms or to extract other features of the steadystate solution. An advantage of the implementation using a general-purpose tool such as Mathematica is that experimenting with di erent numerical methods, generating and plotting converter characteristics etc., can be easily accomplished. For example, in addition to the methods described in this section, we have implemented and tested an extrapolation method 5 , and a method based on the built-in Mathematica function for solving a system of nonlinear equations.

IEEE Power Electronics Specialists Conference, St. Louis, Missouri, June 22-27, 1997.
Q VQ=0.7 Ron=0.1 D Vg VD=0.7 Rd=0.1 Lf 50uH, Rl=0.01 50uF Rc=0.05 Cf
IRF540 Q Lf 50uH, Rl=0.01

47

+ Cf

40V

Rload 5
40V Vg Rg 10 MUR810 50uF Rc=0.05 Rload 5

1nF C1 COMPARATOR + R14 0.1uF 2K +

C14 R13 10K + R12


+12V

40K
R10 100K 2.2K +12V R9 0.1uF C1 + +12V 1nF + C14 R11 2K R14 R13 10K + Vramp 5V Ts=10us + R8 2.2K LF411 R12 10K Vref 2.5V 40K

+
Ts=10us

5V 0 + Vramp

LM311

Vref 2.5V

Figure 3: A closed-loop DC power supply 1 using piecewise-

Ts

linear device models. Simulator: PETS; state variables: x = vCf ; vC1 ; vC14 ; iLf , at the time when the switch Q is turned on.

Figure 4: Closed-loop DC power supply using Spice li-

Ni Ns 1211 1211 3 16 4 9 Table 1: The number of iterations Ni and the number of simulation runs Ns required to converge to a steady-state in the circuit of Fig. 3. The error limit is e0 = 10,4 ; the initial point is x = 0; the steady-state solution is: xs = 12:49V; 0:8224V; 0:3mV; 1:608A.
Method Fixed point Finite-di erence quasi-Newton Broyden's update quasi-Newton

brary component models. Simulator: PSpice; state variables: x = vCf ; vC1 ; vC14 ; iLf , at the time when the ramp Vramp = 0.

4 Application examples

Ni Ns 6 31 8 13 Table 2: The number of iterations Ni and the number of simulation runs Ns required to converge to a steady-state in the circuit of Fig. 4. The error limit is e0 = 10,4 ; the initial point is x = 0; the steady-state solution is: xs = 12:47V; 1:156V; 0:8mV; 1:602A.
Method Finite-di erence quasi-Newton Broyden's update quasi-Newton cuit for PSpice simulation is shown in Fig. 4. The results of nding a steady-state solution using the two quasi-Newton methods are shown in Table 2. Again, the Broyden's update method requires more iterations, but it converges in less than one half of the number of simulation runs in the nitedi erence method. The example of Fig. 4 introduces a problem regarding the system model and the vector of state variables x used by the steady-state solver. Spice device models introduce additional reactive elements so that the complete vector of state variables x can be much larger than the considered vector of state variables x. In some cases, as in the subcircuit macro models of operational ampli ers and voltage comparators, the initial conditions on the additional state variables can be specied, although the steady-state solver would have to access and modify the library le where the models are de ned. The additional state variables associated with Spice built-in discrete device models such as the diode D and the power MOSFET Q in Fig. 4 are not accessible. In general, the complete vector of state variables can be written as x = x; xp , where x is the vector of state variables with initial conditions set by the solver, and xp is the vector of parasitic" state variables neglected by the solver. The steady-state solution obtained by the solver is then x sa = xsa ; 0, compared to the actual

Several examples are described in this section to illustrate applications and compare the numerical methods discussed in Section 3. All examples are done on a 180MHz Pentium PC running Windows 95, Mathematica 3.0, PSpice evaluation version 6.0 for DOS, and PETS.

4.1 Closed-loop DC power supply

We rst consider a closed-loop DC power supply shown in Fig. 3 1 . The system model in Fig. 3 is piecewise linear so that PETS is used as the simulation tool. The results of nding a steady-state solution using the xedpoint iteration and the two quasi-Newton methods are shown in Table 1. Using the quasi-Newton methods the number of simulation runs is reduced by two orders of magnitude compared to the xed-point iteration. In this example, the performance of the nite-di erence quasi-Newton method is about the same as the performance of the Newton-Raphson method reported in 1 . The Broyden's update quasi-Newton method reaches the steady-state in the fewest number of simulation runs although it requires one more iteration compared to the nite-di erence method. Next, we consider the same closed-loop DC power supply example, but with Spice library component models. The cir-

OPAMP

10K

R11

Eg=1

+ -

IEEE Power Electronics Specialists Conference, St. Louis, Missouri, June 22-27, 1997.

57
Lf 200uH, 0.1 D 100uF VD=0.6 Rd=0.01 Vout Cf + Rload ic 2

Ni Ns 4 21 3 8 Table 3: The number of iterations Ni and the number of simulation runs Ns required to converge to a steady-state in the circuit of Fig. 4. The error limit is e0 = 10,2 ; the initial point is x = 1; 0:1; 10; 50; the steady-state solution is: xs = 2:5A; 0:54A; 12A; 65V .
Method Finite-di erence quasi-Newton Broyden's update quasi-Newton steady-state solution x s = xs ; xsp , where in general xsp 6= 0. One can expect that the errors introduced by neglecting xp are relatively small because the time constants associated with the parasitic" state variables are usually small compared to the period T , so that xp t run out of transient during the period T . Nevertheless, the neglected state-variable waveforms can have sizable e ects on the slow" state variables x. In the example of Fig. 4, a long transient simulation was used to obtain an estimate of the actual steady-state solution x s . The solution obtained by the steady-state solver had an error of 4 with respect to the estimate of the actual solution.

Ron=0.1 Q 10V Vg

+ -

+ VT+ = 0.1V VT- = -0.1V

E1=1 H1=1 Vref 5V

Figure 6: Buck converter with sliding-mode control. Simulator: PETS; state variables: x = vCf ; iLf , at the time when the switch Q is turned on.
Ni Ns 4 25 8 13 Table 4: The number of iterations Ni and the number of simulation runs Ns required to converge to a steady-state in the circuit of Fig. 6. The error limit is e0 = 10,5 ; the initial point is x = 0; the simulation time is Tx = 25s; the steadystate solution is: xs = 5:0005V; 2:4000V ; the steady-state period is T = 15:7s.
Method Finite-di erence quasi-Newton Broyden's update quasi-Newton the system model is piecewise-linear so that PETS was used as the simulator. Table 4 shows again that the Broyden's update method outperforms the nite-di erence method. Fig. 7 shows the steady-state waveforms obtained by running simulation from the initial conditions xs found by the steady-state solver.

4.2 Phase-Shifted full-bridge converter

Fig. 5 shows an open-loop, phase-shifted full-bridge converter which was used in 6 as a benchmark example to compare several simulation tools in power electronics applications. Spice library device models are used with the PSpice simulator. In this example, running out of transient to reach a steady-state solution takes about 5ms of simulation time, or about 30 minutes on a 180MHz Pentium PC. Table 3 shows the performance of the two considered quasi-Newton methods. In this case, the Broyden's update method converged in fewer number of iterations and with the total of 80s of simulation time. This is two orders of magnitude less than in the brute-force simulation, and about three times less than with the nite-di erence method. In this case, the fewer number of iterations in the Broyden's update method can be attributed to numerical errors in repeated computations of the nite-di erence Jacobian approximation. This example has a relatively large number of neglected state variables xp that introduce high-frequency ringing in the waveforms of the considered state variables x. This, in turn, e ectively increases the error in FT x further. None of the tested methods converged when the relative error limit was reduced below 10,2 : the iterates wandered around the steady-state solution.

4.4 Steady-state characteristics of a quadratic buck converter

4.3 Buck converter with sliding-mode control

An autonomous system example is shown in Fig. 6. This is a buck converter with sliding-mode control, where the steadystate switching period depends on the parameter values and the independent input voltages Vg and Vref . With the mapping FT x de ned in Section 2, the same numerical methods can be applied to nd a steady-state solution as in nonautonomous systems. This can be compared to the method described in 4 where the boundary condition C x and its derivatives must be explicitly known in order to evaluate the Jacobian for the Newton-Raphson iteration. In this example,

Fig. 8 shows an open-loop quadratic buck converter 13 . This example is used to show how the steady-state solver can be used to generate a set of steady-state characteristics over a circuit parameter. In particular, we seek the average, steadystate output voltage VoutD and the average, steady-state capacitor C1 voltage V 1D as functions of the switch duty ratio D. The Broyden's update method was used in this example. Fig. 9 shows the results obtained by stepping D in the range 0:05  D  0:9 with a constant step size D = 0:05. For comparison, the steady-state characteristics are shown together with the ideal characteristics in the continuous-conduction mode, V 1D = DVg and VoutD = D2 Vg . Fig. 10 shows the steady-state inductor currents corresponding to the considered set of switch duty ratios. For 0 D 0:55, the converter goes through various discontinuous conduction modes. The Mathematica code used to generate the steady-state char-

+ -

COMPARATOR WITH HYSTERESIS

IEEE Power Electronics Specialists Conference, St. Louis, Missouri, June 22-27, 1997.
i1 D1 E1 0.2222 MUR1520 + + F1 i1 0.2222 F2 i2 C1 -0.2222 + Q4
IRFP460

67
22

IRFP460

IRFP460

L1 25uH RC1 20m


24

23

Q1
51 50 54

Q3

LL 15uH

10

RL 5

VIN 400V
IRFP460 53

LM 1mH

Q2
52 55

E2 -0.2222 DDR2
1N5819

+ i2

3300uF MUR1520

D2 DDR4
1N5819 55

DDR1
1N5819

DDR3
1N5819 52

VDR1 -12V 12V 5u 5n 5n 4.7u 10u

51

+ -

RD1 50

VDR2 -12V 12V 0 5n 5n 4.7u 10u

+ -

RD2 50

VDR3 -12V 12V -0.85u 5n 5n 4.7u 10u

54

+ -

RD3 50

VDR4 -12V 12V 4.15u 5n 5n 4.7u 10u

+ -

RD4 50

Figure 5: Phase-shifted full bridge converter 6 . Simulator: PSpice; state variables: x = iLl ; iLm ; iL1 ; vC1 , at the time when the drive signal V DR2 switches from ,12V to +12V.
50 53
2.6A

L1
200uH + RL1=0.6

Ron=0.85

2 C1
2uF

3 Q

L2
100uH RL2=0.3

out

2.5A

Vg
-

Rload 20

Vgate

20V
2.4A i(Lf) 5.005V

D2 D3

4
+ -

D1 C2
10uF -

5.000V

Figure 8: A quadratic buck converter 13 . Simulator: PETS; state variables: x = vC1 ; vC2 ; iL1 ; iL2 , at the time when the switch Q is turned on.
20
10us Vout 20us 30us 40us 50us

D1, D2, D3: D=0.6V, Rd=0.1

Ts=10us

4.995V 0s

Figure 7: Steady-state waveforms in the converter of Fig. 6:


V1(D), Vout(D)

17.5 15 12.5 10 7.5 5 2.5 D^2 Vg 0.2 0.4 0.6 Duty ratio D 0.8 1 D Vg Vout V1

inductor current and the output capacitor voltage.

acteristics is shown in Appendix A. In stepping the duty ratio, the previous steady-state solution is used as the initial point for the next iteration, thus speeding up the convergence, and reducing the possibility that the iteration fails because of an inadequate initial point. Obtaining a steady-state solution for 18 values of D required a total of 146 simulation runs.

5 Conclusions
The paper describes methods for automated steady-state analysis of switching power converters using PSpice or PETS simulation tools, and a steady-state solver implemented in a Mathematica package. The solver updates initial conditions in the circuit netlist description and launches the simulator to obtain the system response over a switching period. Accelerated convergence to steady state is based on quasi-Newton methods where an approximation to the Jacobian is obtained using nite di erences, or an update scheme. In a number of application examples, it was found that the Broyden's update quasi-Newton method converges in the least number of

Figure 9: Steady-state characteristics VC1 D, VoutD of

the converter in Fig. 8 obtained using the Broyden's update method. The corresponding Mathematica code is shown in Appendix A. simulation runs. The total simulation time in the Broyden's update method was smaller by a factor of 2-3 compared to the nite-di erence quasi-Newton method, and typically by two orders of magnitude compared to a brute-force simulation out of transient. The application examples described in the paper include both non-autonomous constant-frequency and

IEEE Power Electronics Specialists Conference, St. Louis, Missouri, June 22-27, 1997.
1

77

0.8

4 5

i(L1}

0.6

0.4

0.2

6
210
-6

410

-6

610

-6

810

-6

0.00001

7 8 9 10 11

0.8

i(L2)

0.6

0.4

0.2

210

-6

410

-6

610 t

-6

810

-6

0.00001

Figure 10: Steady-state inductor current waveforms, iL1  top and iL2  bottom for the switch duty ratios Di = 0:05 + 0:05i, i = 0; : : : ; 17 in the converter of Fig. 8.
autonomous power electronic systems using Spice library device models PSpice simulator or simpli ed piecewise-linear device models PETS simulator. The main advantage of the proposed approach is that a steady-state solution can be obtained e ciently for any converter con guration and any model complexity supported by the general-purpose simulator. The main limitations of the approach are the relatively large errors in the results obtained by simulation, and the inability to access all state variables in the built-in Spice device models. Another advantage of the solver implementation using Mathematica, as opposed to the methods tied to special-purpose tools, is that various extensions such as generating a set of converter steady-state characteristics can be easily accomplished. Future extensions will include globally convergent solution methods, methods to trace boundaries of converter operating modes, and methods to extract system dynamic models.

12 13

IEEE Trans. on Circuits and Systems I, Vol. 39, No. 7, pp. 520530, 1992. T. Kato, W. Tachibana, Periodic steady-state analysis of an autonomous power electronic system by a modi ed shooting method," 1996 Workshop on Computers in Power Electronics, August 11-14, Portland, Oregon. S. Skelboe, Computation of the periodic steady-state response of nonlinear networks by extrapolation methods," IEEE Trans. on Circuits and Systems, Vol. CAS-27, No. 3, March 1980. J. Thottuvelil, Comparing Spice with other circuit simulation tools for power-electronics analysis," 1996 Workshop on Computers in Power Electronics, August 11-14, Portland, Oregon. P. Pejovi c, D. Maksimovi c, A new algorithm for simulation of power electronic systems using piecewise-linear device models," IEEE Trans. on Power Electronics, Vol. 10, No. 3, May 1995, pp. 340-348. P. Pejovi c, D. Maksimovi c, PETS a simulation tool for power electronics," 1996 Workshop on Computers in Power Electronic, August 11-14, Portland, Oregon. S. Pavlja sevi c, D. Maksimovi c, Subharmonic oscillations in converters with current-mode programming under large parameter variations," IEEE PESC 1997 Record. L. O. Chua, P. M. Lin, Computer-Aided Analysis of Electronic Circuits, Algorithms and Computational Techniques, PrenticeHall, Englewood Cli s, NJ, 1975. J. E. Dennis, R. B. Schnabel, Numerical methods for unconstrained optimization and nonlinear equations, SIAM Classics in applied mathematics 16, 1996. S. Wolfram, The Mathematica Book, 3rd edition, Wolfram Media and Cambridge University Press, 1996. uk, Switching Converters With Large D. Maksimovi c and S. C Range of DC Conversion Ratios," IEEE Transactions on Power Electronics, January 1991, Vol. 6, No. 1, pp. 151-157.

A Mathematica code used to plot steady-state characteristics of the quadratic buck converter in Section 4.4
Setup "buck2" ; * use buck2.cir * x0 = 9,1,0,0 ; * initial iterate * VC1= ; Vout= ; e0 = 0.0001; * error limit * nlimit = 50; * max number of iterations allowed * method = 2; * use Broyden update method * period = 0.00001; * switching period * For Duty = 0.05, Duty = 0.9, Duty = Duty + 0.05,

References

UpdateParameter "v2", Duty period ; x0 = SteadyQuasiNewton x0, e0, nlimit, method ; VC1 = Append VC1, Duty, GetAverage 1 ; Vout = Append Vout, duty,GetAverage 2 ; ; ListPlot VC1, PlotJoined- True ; ListPlot Vout, PlotJoined- True ;

1 R. C. Wong, Accelerated convergence to the steady-state solution of closed-loop regulated switching-mode systems as obtained through simulation," IEEE PESC, 1987 Record, pp. 682-692. 2 Y. Kuroe, T. Maruhashi, N. Kanayama, Computation of sensitivities with respect to conduction time of power semiconductors and quick determination of steady state for closed-loop power electronic systems," IEEE PESC, 1988 Record, pp. 756764. 3 D. G. Bedrosian, J. Vlach, An accelerated steady-state method for networks with internally controlled switches,"

Functions: Setup, SteadyQuasiNewton, UpdateParameter and GetAverage are all parts of the Mathematica 3.0 package steady.m, which is available from the author at http: ece-www.colorado.edu ~maksimov .

Você também pode gostar