Frequency Response Analysis for Power Conversion Products Without Small-Signal Linearization 1,2 Kasemsan Siri The Aerospace Corporation 2350 E. El Segundo Blvd., M/S M4-179 El Segundo, CA 90245 310-336-2931 kasemsan.siri@aero.org Calvin Truong The Aerospace Corporation 2350 E. El Segundo Blvd., M/S M2-275 El Segundo, CA 90245 310-336-8129 calvin.h.truong@aero.org Abstract— Presented herein is an accurate approach for combining large signal modeling and frequency-domain analysis of a closed-loop DC-DC converter power system without small-signal linearization. The approach provides a high-fidelity solution for the converter's frequency response due to the inclusion of all non-linearity and parasitic effects and is applicable to both pulse-by-pulse switching and average large-signal models of DC-DC converters. In control-oriented simulators, Fast Fourier Transformation (FFT) is applied to the converter responses, being uniformly sampled for one period of each injected small signal. In circuit-oriented simulators, fundamental frequency components are extracted out of the converter time-domain responses that are usually simulated in a variable time-step mode. This simple, direct, and accurate analysis approach is critically needed for performance evaluation of the converter system frequency response and design validation of the converter closed-loop systems for which the linearizable large-signal models are not available. The "virtual network analyzer" approach provides an increase in the fidelity of non-linear and parasitic effects within the controller and converter's power stages for which the ‘as is’ non-linear pulse-by-pulse switching model is directly simulated. The frequency response analysis approach was validated with a converter power system operating in solararray voltage regulation mode. 6. ANALYSIS RESULTS .......................................... 6 7. DATA FILTERING PRIOR TO FFT ......................... 7 8. CONCLUSION.................................................... 8 REFERENCES........................................................ 8 1. INTRODUCTION Frequency response analysis techniques are among several ways of determining performance and stability robustness of the closed-loop DC-DC converter. These techniques can be effectively adapted to objectively acquire a set of frequency response functions; e.g., converter voltageregulation loop-gain, input/output impedance, and forward/reverse audio susceptibility. Frequency response analysis can be performed analytically through modeling and simulation or experimentally with a network analyzer. Currently, the ability to accurately analyze (on paper) engineering designs of power converters and controls provides a more flexible opportunity to identify fundamental design flaws and prove design solutions before proceeding to the engineering prototype phase. In such cases, modeling and analysis tools developed for designing DC-DC converters and their control laws are indispensable for design validation, cost-effectiveness, and timeliness of the power-conversion product development. Analytical modeling of DC-DC converters and their control laws has improved constantly through the use of more accurate mathematical approaches, faster processing by personal computers, and better analysis tools. Still, several converter topologies and their control laws are classified as “non-analyzable” items when dealing with their analytical frequency responses. Either their mathematics is too complex or their analytically derived small-signal average models are so overly simplified that they fail to include major non-linear effects. These modeling limitations cause design engineers to avoid analytical modeling and to depend primarily on experimental frequency/time domain measurements of the engineering prototype. Through TABLE OF CONTENTS .............................................................................. 1. INTRODUCTION ............................................... 1 2. FREQUENCY RESPONSE ANALYSIS WITH CIRCUIT ORIENTED SIMULATOR..................................... 2 3. LARGE SIGNAL MODELING OF THE CLOSED-LOOP DC-DC CONVERTER .......................................... 4 4. CONTROL-ORIENTED MODEL OF THE CLOSEDLOOP DC-DC CONVERTER ................................. 5 5. FREQUENCY RESPONSE ANALYSIS WITH FFT .... 6 1 2 0-7803-8155-6/04/$17.00© 2004 IEEE IEEEAC paper #1001, Version 3, Updated September 30, 2003 1 engineering prototyping alone, it is much more difficult to optimize the design and/or to identify potential root causes (design and/or manufacturing processes) when encountering unexpected performance deficiencies. Furthermore, limitations of test equipment may prohibit testing of engineering prototypes within their intended operating conditions, leading to potential premature failures, costly redesign, and product reliability issues. configuration of small signal acquisition, and its analysis algorithm based on fundamental-component extractions out of Fourier series expansion. Section 3 describes a specific large-signal closed-loop converter system used for testing of the analysis approach as well as its specific circuit-oriented configuration used for time-domain signal acquisition prior to the frequency-domain analysis. The section provides well-correlated results obtained from two different modeling schemes: the large-signal averaged and the nonlinear switching models. Section 4 emphasizes modeling of the same converter system in control-oriented simulators in which both of the modeling schemes are applicable for validation of the frequency response analysis approach. Section 5 outlines the FFT algorithm typically employed in control-oriented simulators for frequency response extraction. Section 6 compares frequency response results obtained from the selected control-oriented simulator (MATLAB/SIMULINK) and the selected circuit-oriented simulator (PSPICE). Section 7 presents multi-frames and window averaging techniques that are critical to accuracy of subsequent frequency response analysis. Conclusions are given in Section 8. This paper presents a frequency response analysis approach without small-signal linearization. The approach is implemented in both circuit-oriented and control-oriented simulators. The conventional small-signal linearization approach is usually applied to over-simplified approximated converter models that fail to include non-linear and parasitic effects. Furthermore, the large-signal approximated linearizable models are not available for several converter topologies (such as some resonant converters), and therefore conventional small-signal analysis is not applicable. The remaining choice of frequency response extraction is to directly analyze the time-domain response of the “as is” non-linear switching model of the converter system of interest. A high-fidelity frequency response analysis approach was first successfully developed using control-oriented simulators (such as MATLAB/SIMULINK). However, the control-oriented simulation requires a considerable analytical effort to convert the circuit-oriented converter schematic diagrams into the interconnected blocks of control-oriented models with several derived parameters. The required effort discourages most circuit-oriented engineers from applying this approach, and adds risks from errors during the derivations of the control blocks. Since the circuit-oriented simulators (such as PSPICE) allow direct input of the schematics and do not require conversion to a control system format, an opportunity has been identified to reduce the analytical overhead and associated risks through the circuit-oriented simulators instead. As a demonstration, a PSPICE-based circuit simulator will be used to produce interim frequency response data that a programmable mathematical tool (such as MATLAB) will then numerically process to create standard frequency response plots. Consequently, this analysis method yields a very high-fidelity frequency response while retaining the nonlinear effects essential for circuit operation, resulting in a more accurate measure of the system’s robustness. This technique will be validated on a converter power system operating in solar array voltage regulation mode by comparing it with the results of a linearizable approximated state-space model. The analysis approach is applicable to any non-linear switching converter. In particular, the approach must be used to get the simulated frequency response whenever a linearizable approximated model of the analyzed converter is not available. 2. FREQUENCY RESPONSE ANALYSIS WITH CIRCUIT-ORIENTED SIMULATORS Generic Simulation Set-up for Small-Signal Frequency Response Acquisition Fig. 1 represents a generic setup for frequency response data acquisition using a circuit-oriented simulator. Typically, a DC-DC converter power stage is controlled by a PWM driving signal being output from the PWM circuit. The PWM circuit converts a filtered analog input signal into the chopped two-state switching signal that commands the power switches in the converter power stage to either turn on or turn off in every switching period. The converter output voltage, Vout, is fed back to the output voltage regulation control circuit to produce a proper error signal that becomes the filtered analog signal feeding the PWM circuit. In general, Vout could be any converter feedback signal for its associated closed-loop regulation, such as converter input voltage or output current sensed signal. Prior to feeding the composite response, VFB, back into the control loop, the small-signal injection and data acquisition setup is typically inserted at the output voltage node, Vout, on which the injected AC small signal is superimposed, as depicted in the block labeled “Fourier-Series Extraction Model” in Fig. 1. For each given small-signal frequency, the acquired responses, Vout (t) and VFB (t), are taken through the process of extracting the small-signal fundamentalcomponents from their Fourier series expansions. The small-signal frequency sample points, as well as the extracted responses of the small-signal fundamental components (magnitude and phase) of the acquired signals, can then be intermediately stored in the corresponding output file (such as a PSPICE *.out file). Later, a programmable mathematical tool, such as MATLAB, may This paper is organized into seven sections. Section 2 describes small-signal frequency response analysis using circuit-oriented simulators (such as PSPICE), its generic 2 be used to read the data in the output file, post-process it into a proper format such as magnitude and phase plots (Bode plots), and store the final result as plot files. fn. The tupdate#i is the time at which the simulation updates the frequency of the injected small signal at its respective frequency fi. Another table defines a timely sequence of end-time data points that are synchronously used in conjunction with the corresponding sample-point frequencies previously defined in the former table. Each data point defined in the end-time table is given in a format of (tupdate#i, tend#i), where tupdate#i < tend#i ≤ tupdate#i+1. Each of these end-time data, tend#i, is used to compare with the simulation run-time variable called TIME (a reserved word in a PSPICE simulator) to determine its respective active time window (tend#i - tstart#i) for processing the fundamentalcomponent extraction algorithm on the acquired timedomain responses for each small-signal frequency. For precise frequency response extraction, the processing time window of each small-signal frequency is always in multiple periods of the small-signal frequency, or tend#i - tstart#i = NTi, where N is a given integer and Ti is the small-signal period 1/fi. From tupdate#i, tend#i, and fi, the starting time tstart#i for an active processing time window #i must satisfy the constraint tupdate#i < tstart#i < tend#i. In other words, tstart#i – tupdate#i is the simulation delay time after updating the small-signal frequency to fi prior to actually processing its small-signal frequency response contents. Sufficient delay time is needed for steady small-signal responses during the processing time-window of each small-signal frequency. Figure 1. Generic simulation set-up for acquisition of frequency response data using circuit-oriented and MATLAB simulator Frequency Response Extraction Algorithm Detailed modeling of the fundamental-component extraction out of the Fourier series expansion can be conceptualized with the algorithm block diagram shown in Fig. 2. Fig. 3 shows a timing diagram of the frequency samples and the respective enable/reset pulses of the processing timewindow that are determined from both of the look-up tables. The active duration of each enable/reset pulse (being asserted as “active-high” pulse-width) is always in fixed multiple periods of the injected small signal. Three additional signals are generated from the sample frequency look-up table: the small-signal Asinωt that is injected into the control loop across the two nodes, Vout and VFB, and two orthogonal sinusoidal signals, sinωt and cosωt, that are used in tandem for extracting the two corresponding orthogonal components of the acquired signals. Figure 2. Algorithm block diagram of frequency response acquisition based on fundamental-component extraction out of Fourier series using a circuitoriented simulator Two look-up tables are pre-defined in the extraction model. One look-up table defines a timely sequence of the smallsignal sample frequencies that are used during the largesignal time-domain simulation of the converter system. Each data point defined in this table is given in a format of updating-time and its associated small-signal frequency (tupdate#i, fi). These (tupdate#i, fi) data points may be listed in an ascending order in both updating-time and its respective small-signal frequency where index i=1, 2,. . n for n frequency points from start frequency f1 to stop frequency Figure 3. Timing diagram showing frequency samples and their respective enable/reset signals For any periodic signal f(t) of fundamental frequency ω, its fundamental component f1(t) can be expressed as 3 f 1 (t ) = a cos ω t + b sin ω t loop by superimposing the signal on the sensed array voltage. Through a fundamental-component extraction from Fourier series expansion, the response of loop-gain transfer function Vin./Ve can be extracted from the timedomain simulation data Vin(t) and Ve(t) being sampled over one or multiple periods of the injected small signal. For validation and comparison purposes, both the large-signal average model and the pulse-by-pulse switching model of the converter controller and its power stage operating in the continuous conduction mode are developed and tested independently. First, employing the converter-system average model, a PSPICE AC analysis with conventional small-signal linearization yields the array voltage regulation loop-gain frequency response. Subsequently, using the developed “virtual” network analyzer without small-signal linearization as depicted in Fig. 2, another set of the frequency response is extracted from the standard PSPICE transient analysis run on the converter pulse-by-pulse switching model. Finally, these two sets are plotted on the same graph to demonstrate the excellent agreement between the two different analysis approaches, as shown in Fig. 5. Note that Ve and Vin shown in Fig. 4 are respectively equivalent to VFB and Vout shown in Fig. 1. (1a) f1 (t ) = a 2 + b 2 sin(ωt + θ ) where θ = tan θ= a= 2 T a or b π (1 − sign (b) ) + tan −1 a ⋅ sign (a ) ⋅ sign (b) (1b) 2 b t0 +T ∫ −1 f (t ) ⋅ cosωt ⋅ dt, b = t0 2 T t0 +T ∫ f (t ) ⋅ sin ωt ⋅ dt (1c) t0 2π (1d) T The fundamental components of the acquired signals, Vout(t) and VFB(t) are similarly defined as 2 2 Vout1 (t ) = aout cos ωt + bout sin ωt = aout sin(ωt + θ Vout ) , (2a) + bout ω = 2 2 sin(ωt + θVFB ) + bFB VFB1 (t ) = a FB cosωt + bFB sin ωt = a FB (2b) From the above expressions (1a), (1b), (1c), (2a), and (2b), the magnitude of the processed small signals at their fundamental frequency (|Vout| and |VFB| ) can be computed as shown in expressions (3a) and (3b), respectively. 2 2 Vout = aout + bout V FB = (3a) 2 2 + b FB a FB (3b) Therefore, the voltage loop-gain in dB is computed as the ratio |Vout| to |VFB| as shown in (4) VOUT VFB = 20 log10( VOUT VFB ) dB. (4) The phase response of small signal Vout with respect to small signal VFB is therefore calculated from expressions (1b), (2a), and (2b) as θ Vout / VFB = θ Vout − θ VFB degrees Figure 4. Acquisition set-up for frequency response of array-voltage regulation loop (5) Furthermore, integral expression (1c) can be generalized for multiple small-signal periods NT as a= 2 NT t0 + NT ∫ t0 f (t ) ⋅ cosωt ⋅ dt, b = 2 NT t0 + NT ∫ f (t ) ⋅ sinωt ⋅ dt (6) t0 3.LARGE-SIGNAL MODELING OF THE CLOSED-LOOP DC-DC CONVERTER Fig. 4 shows the block diagram of the closed-loop converter system operating in the array voltage regulation mode [6], consisting of a solar-array source, line-filter, current-mode DC-DC converter power stage, output bus stabilizer, bulk output filter capacitor, load circuit, and a solar-array clamp error amplifier. Serving as a means to obtain the loop-gain frequency response of the array-voltage regulation, a small signal is injected into the array-voltage regulation control Figure 5. Array-regulation loop-gain frequency response acquired from the average and switching models implemented in a circuit-oriented simulator (PSPICE) 4 4. CONTROL-ORIENTED MODEL OF THE CLOSED-LOOP DC-DC CONVERTER capacitors, C5 and C0, within the converter power stage, the large-signal state-space averaged model of the converter output-filter can be mathematically expressed as a function of input <Vy> and state-vector ‘x’. The power stage state vector consists of three state variables <iL>, VC5, and VC0, where <f> denotes an instantaneous average over one switching period of variable f. The instantaneous average of the duty-ratio input d(t) controlling the power stage response during any switching period is also written as <d(t)>; therefore, <Vy> is the average voltage input to the converter output filter, which is defined as a product between <d(t)> and V2. Similarly, I2 is defined as the average input current of the converter power stage, which is defined as a product between <d(t)> and <iL>. Consequently, the control-oriented large-signal average model of the converter power stage has two inputs, <d(t)> and V2, and two outputs, I2 and Vo, where VC0 = Vo is the converter output voltage. As an alternative option for frequency response analysis without small-signal linearization, the control-oriented “virtual network analyzer” approach was also successfully implemented by primarily performing FFT for exactly one small-signal period on the acquired signals that must be sampled in fixed time steps during the time-domain simulation. In control-oriented simulators (MATLAB/SIMULINK), the development of a basic pulse-by-pulse non-linear switching model or a large-signal average model of a converter system requires some overhead mathematical conversion from circuits to control blocks. These control blocks represent the modeled circuits as interconnected transfer functions and/or linked sets of state and output equations, which are mathematically formatted in vectors and matrices. Once the model is set up, repetitive simulation and processing can be performed easily to obtain the frequency response. Two SIMULINK converter system models that are developed by two different modeling schemes are described in this section: large-signal state-space averaging and pulse-bypulse switching techniques. By applying the FFT to the two signals acquired from the same converter system shown in Fig. 4, the transfer function response, Vin./Ve, can be extracted from the time-domain simulation data Vin(t) and Ve(t) sampled over one period of the injected small signal. Figure 6. Figure 6 shows the detailed circuit of the line-filter being excited by the array source. The figure also shows the control-oriented model of the line-filter and array source, which is converted from the line-filter schematic. The line-filter model is represented by a control block that is derived in terms of a set of state and output equations having two inputs, Ia (array current) and I2 (line-filter output current), and two outputs, Va (array voltage) and V2 (line-filter output voltage). The state equations [3,4] are first-order differential equations with five state variables (four filter capacitor voltages and one filter inductor current). The array voltage Va, is the input excitation of the array source model that behaves like a voltage-controlled current source. Through the use of a look-up table, the behavioral model of the array I-V characteristics can be represented by a piecewise-linear function of Va. As shown in Figure 7, a circuit-oriented pulse-by-pulse switching model of the converter power stage operating in the continuous conduction mode can be converted to a large-signal control-oriented model by applying Kirchhoff’s voltage and current laws and Middlebrook’s state-space averaging technique [1]. Due to the existence of the main power stage inductor L2 and the damping and filter Conversion from circuit-oriented to controloriented models of the line-filter with an array source Figure 7. Large-signal average modeling of the switching converter power stage in a control-oriented format (SIMULINK) Figure 8 illustrates the model conversion from pulse-bypulse switching peak-current programmed control circuit to 5 5. FREQUENCY RESPONSE ANALYSIS WITH FFT the large-signal average model by applying Middlebrook’s averaging technique. After combining all the derived control-oriented models, Figs. 9 and 10 show SIMULINK models of the converter system using the large-signal average and pulse-by-pulse techniques. The SIMULINK converter system models were simulated in the time-domain and verified to function properly before proceeding to frequency response analysis. To obtain the open-loop frequency response of the SIMULINK converter circuit model, time domain simulation is performed with sinusoidal signals of various frequencies injected at the summing junction (Ve in Figs. 4 and 9, V5 in Fig. 10). The FFT technique is applied on the timedomain data to extract the fundamental frequency components (magnitude and phase). Subsequently, a singlefrequency response of Vin/Ve (gain and phase response) was stored. The process repeats for the subsequent runs with a different input frequency. The following algorithm is the converter frequency response analysis that has been coded in MATLAB [5]. Algorithm: 1) Select the frequency range of interest where f is an array containing small-signal frequencies 2) For I=1:length(f) , where length(f) is the number of frequency points or size of the array f a) Perform a time domain simulation of the model and save the vin, v5, and vsine signals. b) Pick only one small-signal cycle (N sampled points of each acquired signal) of vin, v5 and vsine and compute their respective FFTs for H = FFT (v,1,N), where N is the number of timedomain samples. Let c = H(2), where c is the complex value at fundamental frequency of vsine, then the gain and phase of the FFTs are • Gain = 2*abs(c )/N • Phase = atan (imaginary (c), real (c)) c) Save the gain and phase for vin, v5, and vsine . 3) Plot the Gain Go and phase θ of the open-loop transfer function, where Figure 8. Modeling of peak-current programmed control laws in SIMULINK Figure 9. SIMULINK model of the converter system shown in Fig. 4 using large-signal average technique Go = 20*Log [Gain(vin)/Gain(v5)] dB, and θ = (Phase(vin) - Phase(v5))*180/π degrees. 6. ANALYSIS RESULTS In this section, the FFT algorithm is applied to the DC-DC converter SIMULINK models described in Section 4, whereas the conventional AC analysis with small-signal linearization is independently applied in the PSPICE largesignal average model. Figure 11 shows a converter loop gain and phase response comparison between the PSPICE and SIMULINK large-signal average models. Note that the results are almost identical, i.e., both showing almost 5 kHz cross-over frequency with 63° phase margin. The steady-state time-domain waveforms of the acquired signals from MATLAB SIMULINK (Figure 12) were also visually inspected and confirmed the result. Figure 10. SIMULINK model of the converter system shown in Fig. 1 using pulse-by-pulse technique 6 Figure 13. Converter loop-gain response acquired from both the non-linear switching and the average models developed in SIMULINK Figure 11. Converter loop-gain response of the arrayvoltage regulation via PSPICE and SIMULINK Figure 12. Time-domain steady-state responses of the loopgain input, Ve, and output, Vin Figure 14. Time-domain steady-state responses of loop gain input, Ve, and output, Vin , from SIMULINK converter switching model Figure 13 depicts the loop-gain responses individually obtained from both the average and the non-linear switching SIMULINK models, revealing well-correlated results. Figure 14 shows a time response obtained from the same SIMULINK converter system switching model, visually confirming the same gain and phase response being extracted from the FFT. Note that the phase discrepancy between the average and the non-linear switching models at upper frequencies is caused by the failure of the large-signal average model to include non-linear and parasitic effects. At 1 MHz switching frequency, parasitic loss across the damping resistors within the converter model is greater. This causes the phase to be less steep. 7. DATA FILTERING PRIOR TO FFT In practice, the time-domain data can be corrupted by a significant content of switching noise at the converter switching frequency. This switching ripple is more pronounced at high frequency due to lower amplitude signals that lead to noisier FFTs. Therefore, it is recommended that the analysis approach without data filtering described in the previous section be applied to data less than 2/5 of the converter switching frequency. As noted previously, the “virtual network analyzer” approach is necessary for frequency response analysis of a converter with a non-linear switching model that cannot be linearized due to its time-variant structure. A technique involving filtering of the acquired data prior to FFTs is also presented in the subsequent section to reduce data noises generated from switching ripple, thereby, further improving the frequency response results and yielding smoother plots. By filtering the data using a multiple-frame data-averaging technique [7], the switching noise at the converter switching frequency can be attenuated. Therefore, the noise contribution at any frequency is significantly reduced. The SIMULINK converter switching model used for producing the frequency response result shown in Figure 13 7 [3] N. Balabanian and T. A. Bickart, “Electrical Network Theory”, New York: Wiley, 1969, Chapter 4. [4] B. C. Kuo, “Linear Networks and Systems”, New York: McGraw-Hill, 1967, Chapter 5 and 6. [5] D. Hanselman and B. Littlefiled, “Mastering MATLAB 5”, A Comprehensive Tutorial and Reference, Prentice Hall, 1998. [6] K. Siri, “Study of System Instability in Current-Mode Converter Power Systems Operating in Solar Array Voltage Regulation Mode,” APEC’2000, New Orleans, Louisiana, 228-234, Vol. 1, February, 2000. [7] H. Li and D. S. Doermann, “Text Enhancement in Digital Video Using Multiple frame Integration,” Proc. AC- Multimedia 1999, Orlando, Florida, 19-22. has a switching frequency of 1 MHz. Hence, to obtain a valid frequency response between 1 kHz and 50 kHz, the simulation time step was fixed at 2.5 ns (400 MHz.) with a decimation value of 10 (the interim record of the simulated response being sampled at 25-ns time steps). Before using the FFT algorithm, the simulation data was conditioned with a window averaging technique. We chose a windowaverage size of 25 (i.e., summing 25 consecutive data points and dividing by 25). This essentially smoothed the data within any 625 ns sampled at 40 MHz. To reduce computation time and memory storage, the data was further sampled at 10 MHz (the temporary record of the movingwindow average response being sampled at 100-ns time steps). Next, a multiple-frame average technique is used. Synchronizing with the frequency of injected signal, 4 frames of smoothed data were averaged. This further reduced numerical noise as well as system switching noise. Finally, the data was ready for the FFT algorithm. BIOGRAPHY Kasemsan Siri holds Ph.D. degrees in electrical engineering. His experience includes four years of teaching in electronics, seven years of power electronics research at the University of Illinois at Chicago, and sixteen years of industrial experience in power electronics and systems. Starting from the Associate Director of Research in Power Electronics Laboratory at the UIC, a modeling specialist at Rockwell International, Canoga Park, CA, and a senior design engineer at Hughes Aircraft Company, El Segundo, CA, Dr. Siri is presently an engineering specialist at The Aerospace Corporation, El Segundo, CA, supporting design, research, modeling and analysis of power systems for various satellite programs. He received an Aerospace President award in 1999 for technical rigor and leadership in identifying a critical design flaw in a MILSTAR satellite power system and the resolution that were adopted for four subsequent satellites. He is the author of over 55 scientific papers and holds eight U.S. patent inventions in zerovoltage switching DC-DC converters, maximum power tracking architectures, current-sharing schemes, and active power factor correction. Note that in the lower frequency region (100 Hz To 1 kHz.), a larger simulation time step of 25 ns and FFT time step resolution of 1 MHz were used. This is sufficient to maintain the accuracy of the frequency response. 8. CONCLUSION Employing both the state-space averaged model and the pulse-by-pulse switching model of the DC-DC converter power system, a simple, direct, and accurate frequency response analysis approach is demonstrated and validated based on two fundamental-component extraction techniques. One is the variable time-step extraction from Fourier series expansion in a circuit-oriented simulator, and another is the fixed time-step extraction through FFT. The developed analysis tool is critical for robust design and validation of the converter closed-loop system since more parasitic effects within the converter power stages are included in the model. Through applications of both circuitoriented and control-oriented simulators (such as PSPICE and MATLAB/SIMULINK), the approach is exercised with a converter power system operating in a solar-array voltage regulation mode. In general, different switching converter topologies and control schemes may produce different noise spectra resulting from switching modulation spreading around their switching frequencies. This may require different data filtering techniques to smooth out the noise prior to the FFT computation. Calvin Truong is a senior engineer in the Photonics and Electronics Laboratory at The Aerospace Corporation, El Segundo, CA. He holds M.S. and B.S. degrees in electrical engineering. His experience includes four years of research and development in solar power generation and conversion in electric power utility and fourteen years of space power electronics in the aerospace industry. His research interest is in the areas of analytical modeling of power conversion systems, radiation effects in power converters, Pico-satellites development, and space power system stability analysis. REFERENCES [1] R. D. Middlebrook and S. Cuk, “Advances in SwitchedMode Power Conversion”, TESLAco, Inc., Pasadena, California, 1981. [2] Rudolf P. Severns and Gordon E. Bloom, “Modern DCTO-DC Switch-mode Power Converter Circuits”, Van Nostrand Reinhold, 1985. 8