Inter national J our nal of P o wer Electr onics and Dri v e Systems (IJPEDS) V ol. 6, No. 4, December 2015, pp. 665 682 ISSN: 2088-8694 665 Co-Simulation Interfacing Capabilities in De vice-Le v el P o wer Electr onic Cir cuit Simulation T ools: An Ov er view W entao W ang * , W enying Y ang ** , and V enkata Dina v ahi * * Department of Electrical & Computer Engineering, Uni v ersity of Alberta, Edmonton, AB T6G 2V4 Canada. ** School of Electrical Engineering & Automation, Harbin Institute of T echnology , Harbin 150001 China. Article Inf o Article history: Recei v ed Jun 20, 2015 Re vised Aug 13, 2015 Accepted Aug 29, 2015 K eyw ord: Co-simulation Circuit simulators De vice-le v el modeling Interf acing, Po wer electronics System-le v el simulation. ABSTRA CT Po wer electronic circuit simulation t oday has become increasingly more demanding in both the speed and accurac y . Whilst almost e v ery simulator has its o wn adv antages and disad- v antages, co-simulations are becoming more pre v alent. This paper pro vides an o v ervie w of the co-simulation capabilities of de vice-le v el circuit simulators. More specifically , a list- ing of de vice- le v el simulators with their salient features are compared and contrasted. The co-simulation interf aces between se v eral simulation tools are discussed. A c ase study is presented to de monstrate the co-simulation between a de vice-le v el simulator (PSIM) inter - f acing a system-le v el simulator (Simulink), a nd a finite element simulation tool (FLUX). Results demonstrate the necessity and con v enience as well as the dra wbacks of such a com- prehensi v e simulation. Copyright c 2015 Institute of Advanced Engineering and Science . All rights r eserved. Corresponding A uthor: V enkata Dina v ahi Dept. of Electrical & Computer Engineering Uni v ersity of Alberta 9107 116 St., Edmonton, AB T6G 2V4 Canada Ph: (780) 492-1003 Email: dina v ahi@ualberta.ca 1. INTR ODUCTION De vice-le v el circuit simulators are essential to tackle the increasing demand of the accurac y and precision in the design and implementation of comple x systems. The special adv antage of such simulators is based on their conformity to fundamental ph ysical principles, thereby achie ving more accurate simulation results [1],[2]. Ho we v er , since de vice-le v el simulators ha v e to depict the w a v eform of a design node in great detail, the y are relati v ely slo w in e x ecution speed, which is the k e y bottleneck in lar ge desi gn projects. Besides accurac y and speed, man y system- le v el simulation projects e v en require specific analyses (e.g. transient, statistical analysis) in certai n parts of the systems, which could only be achie v ed using ci rcuit simulators [3, 4]. In such cases, hierarchical design structure strate gies become necessary . F or e xample, some lar ge po wer system simulation tasks may require implementation in both system-le v el lik e Matlab/Simulink R and de vice-le v el circuit simulators to e xploit the modeling strengths in their respecti v e areas of simulation [5],[6]. Meanwhile, man y of the design projects today may include analog, digital and mix ed-signal simulations, which mak e it necessary for analog and mix ed-signal simulators to support the function of co-simulation within standard digital simulators [7]. Furthermore, researchers ha v e endea v ored to de vise methods to enable de vice-le v el circuit simulators to co-simulate with programming languages lik e SystemC TM , which are especially u s eful during hardw are realization [ 8 ] ,[9]. As a result of all aforementioned f actors, the discussi on of co-simulation issue for de vice-le v el circuit simulators becomes paramount. Another important issue about de vice-le v el circuit simulators ob viously is the numerical solution method. F or nonlinear dc, transient and linear ac circuit simulations, the circuit may comprise of semiconductor de vices such as diodes, BJTs, JFETs, MOSFETs and IGBTs; linear/nonlinear resistors, capacit ors, inductors, independent/dependent v oltage and current sources, etc. [1] All these actual de vices can be approximated by b uilt-in or user -defined models, where the complete system of models can be highly coupled. Standard methods e xist to solv e nonlinear dif ferential Evaluation Warning : The document was created with Spire.PDF for Python.
666 ISSN: 2088-8694 algebraic equations. T ypical ly , numerical inte gration, the Ne wton-Raphson method, and sparse-matri x techniques are applied [10]. One of the circuit simulators, that applies the standard met hod is SPICE [11]. This is also true of other simulators including those among the SPICE f amily which are essentially based on the k ernel of SPICE [12], [13], [14]. Although the standard methods are ef ficacious in man y applications, the y were in v ented se v eral decades ago, and the y ha v e the dra wback of slo w computing speed, especially when used in conjunction with lar ger circuit simulations. Therefore, there is an ur gent need to de vise more ef ficient methods of solution. Actually , man y of these kinds of no v el simulators ha v e been de v eloped, which include MEDUSA and CODECS [15],[16]. The purpose of this paper is to enumerate co-simulation interf aces in de vice-le v el circuit simulators, and to identify their specific application in the literature. A PSIM R -Simulink R -FLUX R case study is pro vided to demonstrate the ef fecti v eness of the co-simulation interf acing in de vice-le v el circuit simulators. 2. GENERIC NUMERICAL SOLUTION APPR O A CH IN DEVICE-LEVEL CIRCUIT SIMULA T ORS While there are a handful of commonly used and popular de vice-le v el circuit simulators, this study has re v ealed a host of other tools that are less well kno wn b ut nonetheless unique in t heir modeling and simulation ca- pabilities. T able 1 enumerates the de vice-le v el circuit simulators found in this study with their salient features and rele v ant references, with a special emphasis on their adv antages and disadv antages from a user’ s point of vie w . These simulators can be broadly cate gorized into research-oriented simulators (e.g., SPICE, Ngspice, Q UCS, MEDUSA, and CODECS) and commercial simulators (e.g., PSpice R , Saber R , and PSIM R ). In the tables, research-oriented simulators are mark ed by R and commercial ones are mark ed by C . Furthermore, a group of simulators, includ- ing SPICE, S pice2, Spice3, Ngspice, PSpice, HomSPICE, MultiSim TM and so on can be cate gorized as SPICE-lik e simulators, which are essentially based on the k ernel of SPICE [12], [13], [14], and their transient analysis operations are basically the same. In this section, the numerical solution approach for transient analysis used by a majority of de vice-le v el simulators is described. The first step is the system matrix for mulation. The modified nodal analysis (MN A) approach is tak en by Saber and SPICE-lik e sim u l ators, where the system is represented by a group of nonlinear first-order dif ferential algebraic equations (D AEs): N ( x ( t ) ; d x ( t ) dt ; t ) = 0 (1) where x ( t ) is the v ector of unkno wn circuit v ariables, and N ( ) are the nonlinear v ector functions [1], [11]. In contrast, other circuit simulators such as PECS, utilize a state-space approach to undertak e the system matrix formulation step [30]. The nonlinear D AE group contain nonlinear ordinary dif ferential equations (ODEs) of the form [2]: d x ( t ) dt = o ( x ( t ) ; t ) (2) T o solv e (2) at the ne xt time-step t n +1 , numerical inte gration is applied. F or instance, SPICE-lik e simulators and Saber use Gear or T rapezoidal methods. F or the T rapezoidal method, x n +1 = x n + h 2 [ o ( x n +1 ; t n +1 ) + o ( x n ; t n )] (3) where x n +1 is the solution of ne xt time-step t n +1 , and x n is the solution at the current time-step t n [2]. F or the second-order Gear method (the def ault method for Saber), x n +1 = 4 3 x n 1 3 x n 1 + 2 3 h o ( x n +1 ; t n +1 ) : (4) Thus (2) is successfully transformed into nonlinear algebraic equations (N AEs), which ha v e the general form of F ( x ) = 0 ; (5) where x x n +1 , and F ( ) is the general nonlinear operator [1], [11]. T o solv e these N AEs, (5) needs to be linearized in this step. SPICE-lik e simulators and Saber use Ne wton- Raphson or the Katzenelson algorithms. (5) can be further written as F ( x ) = [ f 1 ( x ) ; f 2 ( x ) ; : : : ; f m ( x )] T ; (6) IJPEDS V ol. 6, No. 4, December 2015: 665 682 Evaluation Warning : The document was created with Spire.PDF for Python.
IJPEDS ISSN: 2088-8694 667 T able 1. Comparison of De vice-Le v el Circuit Simulators Simulators F : Salient F eatur es; N : Adv antages; H : Disadv antages SPICE (Spice2, F : Simulation Program with Inte grated Circuit Emphasis (SPICE), the most popular Spice3) general purpose and open source analog circuit simulator[12]; N : i) Lar ge model library; R ii) Ef ficient solution method: finite dif ference numerical method; iii) Three time-step strate- gies: iteration count, dV/ d t cont rol, and local truncation error (L TE); i v) Simulation studies: DC and A C analysis, DC transfer curv e analysis, Noise analysis, T ransient analysis. H : i) Less simulation accurac y in a comple x circuit en vironment; ii) Limited interf acing capability . HomSPICE F : A member of SPICE-f amily circuit simulators, uses three homotop y algorithms: R FIXPDF , FIXPNF , and FIXPQF , which is in f a v or of calculating a circuit’ s dc operating points and periodic steady-state response [13]. N : P arameter embedding methods are rob ust and accurate numerical techniques for solving nonlinear algebraic equations; H : Computational intensity is high. JSPICE F : Based on Spice2, designed for superconductor and semiconductor circuits [17]; N : R i) Can be incorporated with the Josephson junction model; ii) Supports the same Spice2 format and runs in the batch mode. H : i) Only v alid for transient simulations; ii) DC operating point and A C small-signal analyses are not allo wed. MacSpice F : A Mac v ersion of SPICE, open source and free for non-commercial use only [18]. R N : i) An adapti v e step-size algorithm is used to g ain better con v er gence during a transient analysis; ii) Pro vides a rob ust multi-parameter optimizer and f acility for interprocess commu- nication with other applications. H : Limited interf acing capability . XSpice F : Based on SPICE, b ut further incorporates arbitrary user models [19]. N : i) Sup- R ports for adding “code models” written in the C programming language and contain o v er 40 ne w functional blocks; ii) An embedded e v ent-dri v en algorithm is coordinated with the analog simulation algorithm to pro vide f ast and accurate simulation of mix ed-signal circuits and sys- tems; iii) An interprocess communicati on interf ace for connection to CAE system softw are. H : A C analysis is not supported for circuits with e v ent-dri v en nodes. Ngspice F : A mix ed-le v el/mix ed-signal circuit simulator [20]; Based on three softw are pack- R ages: Spice3f5, Xspice and Cider1b1; Ngspice is part of gED A project, de v elopment from its users contrib ution; N : i) Pro vides additional C language code models to support analog beha v- ioral modeling and co-simulation of digital components through a f ast e v ent dri v en algorithm; ii) In v olv es a numerical de vice sim u l ator to couple the circuit le v el simulator to the de vice simulator to pro vide enhanced simulation accurac y (at the e xpense of increase d simulation time); iii ) Mi grates to ot her com mercial SPICE simulator fla wlessly; i v) Supports both W in- do ws and Linux platforms; H : i) Man y SPICE param eters will not be supported by Ngspice and simulation results can be inaccurate; ii) Does not pro vide schematic capture function. Q UCS F : An inte grated circuit simulator with a graphical user interf ace[21]; Simulates the R lar ge-signal, s mall-signal and noise beha vior of the circuit. Pure digital simulations are also supported using VHDL and/or V erilog; N : i) A graphical interf ace for schematic capture. Sim- ulation data can be represented in v arious types of diagrams, including Smith-Chart, Cartesian, T ab ular and so on; ii) Existing SPICE models can be imported for use; H : Only supports the GNU/Linux OS. HSpice R F :An analog circuit simulator similar to Spice3 b ut has better con v er gence, commercial C product from Synopsys R [22]. N : i) Performs transient, steady-state, and frequenc y do- main analyses; ii) Better con v er gence than SPICE3. H : i) GUI (graphic user interf ace) is n ot friendly; ii) T ime-step setting is a littl e complicated when performing high frequenc y analysis. Orcad R /PSpice F : As a PC v ersion of SPICE, PSpice R is a dominant industrial standard for circuit C and system analysis, w orks in analog and mix ed signal en vironments, supports the functions for analog beha vioral modeli ng [14] [23]; N : i) Comprehensi v e model library (about 30000); ii) Po werful w a v eform/simulation result analysis interf ace; iii) Of fers models for all kinds of de vices lik e Electromechanical systems (Resolv ers, Brushless Co-Simulation Interfacing Capabilities in De vice-Le vel Cir cuit Simulation T ools () Evaluation Warning : The document was created with Spire.PDF for Python.
668 ISSN: 2088-8694 T able 1. Comparison of De vice-Le v el Circuit Simulators (Continued) Simulators F : Salient F eatur es; N : Adv antages; H : Disadv antages DC motors etc.), Mechanical systems (Flywheel etc.); i v) Support co-simulation with Matlab; v) W ith se v eral Adv ance Analysis modes - check for reliability /stress (SMOKE Analysis), calculates Y ield for m ultiple goals (Monte Carlo), Optimization Module with multiple optimization engines (Optimizer), Design Space e xplorations (P arametric Plotter); vi) W ell inte grated with complete system design and analysis tools (PCB Editor). H : i) Allo w user to select specific components with industry standard part numbers and specifications. But searching for these components is time-consuming; ii) Comple x circuit simulator; iii) The setting of simulation parameters is critical and dif ficult to set in order to a v oid numerical con v er gence problems. i v) No data visualization during simulation. Saber R F : A comprehensi v e mix ed-signal simulator , pro vides a v ersatile modeling language named C MAST , which mak es i t possible t o di vide specific models from simulation al go r ithms; applied to electrical, optical, thermal, mechanical systems [10], [24]. N : i) Comprehensi v e model library: 30,000+ models; ii) Accurate and ef ficient solution method: adopts the piece wise linear e v aluation technique and subdi vision; iii) Fle xible time-step strate gy: fix ed and v ariable; i v) V ery friendly GUI: for generating virtual prototypes of po wer system netw orks; v) Po werful interf acing capabil- ity: with popular 3D CAD tools (Catia V5, Siemens (UGS), Pro/E), MA TLAB/Simulink, Zuk en, Mentor Graphics, Cadence, Synopsys VCS; vi) V arious simulation studies: DC and A C static (steady) or transient solution, r o b ust design methods (e.g., Stress, Sensiti vity , Monte Carlo, etc.) ; vii) W ide range of industrial applications, for design v alidation and optimization for automo- ti v e, aerospace, industrial po wer and ener gy systems; viii) Supports for MAST and VHDL-AMS language standards; ix) V erify the beha vior of ph ysical systems (i.e., electrical, mechanical, h y- draulic etc.); x) Use grid computing to minimize time for compute intensi v e statistical analyses. H : i) No front end for algorithmic modeling; ii) No links to rapid prototyping or to operation with hardw are-in-the-loop Simulation; iii) Co-simulation with Simulink requires that Saber is running. PSIM R F : A strong simulation platform for po wer electronics and motor dri v e control [25]. N : i) C W ith strong algorithm dedicated to electrical circuits (piece wise method, generic models and a fix ed time-step) and simulation times are significantly reduced; ii) Friendly user interf ace; iii) Po werful interf acing capability: with Matlab/Simulink, JMag and Magnet; Link to e xternal C/C++ Code via DLL. i v) V arious simulation studies: A C Analysis, A C Sweep, Harmonic Analysis, Motor Dri v e Analysis, Switch Losses Calculation and Thermal Analysis. H : Comple xity of the block diagrams used to simulate the po wer circuits can increase drastically with the number of semi-conductors in the circuit. PLECS R F : A Piece-wise Linear Electrical Circuit Simulator (PLECS) based on the state-v ariable C formulation w orking within Matlab/Simulink en vironment and inte grating circuits entered into Simulink as S-functions [26]. N : i) Direct access to Matlab/Simulink en vironment. ii) Nonlinear component and thermal modeling library are included. H : i) Semiconductor models are not includ- ing all ef fects. ii) Onl y electrical domain simulation (some thermal domain components e xist). iii) Limited to place electric components in restricted modeling area. HSIM R F : Designed to meet the requirement of nanometer circuit analysis, able to perform C hierarchical simulation, commercial product de v eloped by Synopsys R [27]; N : i) T echniques of hierarchical storage and isomorphic matching in HSIM speed up the simulation of lar ge circuits; ii) Full SPICE functionality: A C, DC, transient, Monte Carlo and FFT analyses. H : Needs to set man y options to get the correct simulation. XSIM F : An ef ficient crosstalk simulator , based on indigenous methodology in V isual C++[28]. R N : i) Friendly graphical user interf ace; ii) Ex ecutes a parallel application with a virtual w all clock time. H : i) Simulation studies are limited; ii) Needs to enlar ge the model library . Multisim TM F : A updated v ersion of SPICE; softw are simulation kit pro vides dynamic simulation C models, with po werful interacti vity; has po werful Design-Rules-Check and Connect i vity Check with the breadboard tool [12], [29]. N : i) Comprehensi v e model library; ii) Friendly graphical user interf ace; iii) V arious simulation studies. H : Interf acing capability should be further impro v ed. IJPEDS V ol. 6, No. 4, December 2015: 665 682 Evaluation Warning : The document was created with Spire.PDF for Python.
IJPEDS ISSN: 2088-8694 669 T able 1. Comparison of De vice-Le v el Circuit Simulators (Continued) Simulators F : Salient F eatur es; N : Adv antages; H : Disadv antages PECS F : Po wer Electronics Circuit Simulator (PECS), e xcels in time-domain simulation of R switched netw orks with nonlinearity[30]. N : i) Friendly graphical user interf ace; ii) State-space method for the analysis of switched netw orks is adopted to achie v e both high speed and accurac y . H : i) Model library is small; ii) Simulation studies are limited. TIT AN F : A complete customizable simulator , which pro vides the freedom to di vide the whole C circuit into arbitrary subcircuits [31]. N : i) V arious simulation studies; ii) Circuit equations are solv ed by special v ectorized solv er . H : Interf acing capability should be impro v ed. PETS F : Po wer Electronics T ransient Simulator (PETS), used for time-domain analysis, pro- R vides freedom to choose dif ferent de grees of comple xity for piece wise-linear models[32]. N : i) Supports continuously dif ferentiable nonlinear models by applying a delay approximation method with Ne wton-Raphson iteration. ii) Automatic time-step control. H : i) Model library is small; i i) Simulation studies are limited. Spectre R F : An impro v ed SPICE-lik e analog simulator from Cadence R [33]. N : i) Pro vides an C adapti v e time-step control algorithm that reliably follo ws rapid changes in the solution w a v eforms; ii) Impro v es simulation speed by increasing the ef ficienc y of the simulator , typically tw o to three times f aster than SPICE; iii) V arious simulation studies: DC and A C analysis, Monte Carlo analy- sis, S-P arameter analysis, etc. H : Limited interf acing capability . MEDUSA F : A user -oriented simulator for modular circuits, satisfies the needs for de vice and circuit R simulations at the same time [15]. N : Solv es the lar ge circuits ef fecti v ely by utilizing system modularity . H : i) Simulation studies are limited; ii) Interf acing capability could be impro v ed. CODECS F : A mix ed-le v el circuit simulator , based on Spice3, while incorporates a set of numer - R ical models under its main structure without changi ng its Spice3 core [16]. N : i) Supports v arious simulation studies: dc, transient, small-signal ac and pole-zero analysis; ii) Numeri cal models in CODECS include ph ysical ef fects, such as bandg ap narro wing, Shockle y-Hall-Read and Auger recombination, etc. H : Simulation ef ficienc y could be impro v ed. CASPOC R F : A multi-le v el simulator de v eloped for the simulation/animation of po wer electronics C and electrical dri v es, especially suited for the simulation of switching ci rcuits with highly nonlin- ear models [34, 35]. N : i) Adds fle xible rob ust non-linear function solv er for modeling non-linear components; ii) Friendly user interf a ce; iii) Fle xible interf acing capability: coupling to Simulink, T esla (a machine design tool) and ANSYS R . H : Simulation studies can be added more; ii) Simu- lation accurac y should be impro v ed. A WEswit F : A mix ed analog and digital circuit simulator , special for the switched capacitor circuits C [36]. N : i) Emplo ys asymptotic w a v eform e v aluation (A WE) technique to ef ficientl y e v aluate more detailed circuit models; ii) Pro vides fle xibility in circuit formulations. H : i) Simulation studies are limited; ii) Interf acing capability should be impro v ed. DesignLab F : A W indo ws v ersion of PSpice R , de v eloped in the form of web pages with multimedia C ef fect for analysis and design of circuits and electronics [37]. N : i) Comprehensi v e model library; ii) V arious simulation studies: DC and A C, transient, Monte Carlo analysis, etc.; iii) Good inter - f acing capability . H : i) User interf ace still requires refinement; ii) No data visualization during simulation. Eldo TM F : An analog, digital and mix ed circuit simulator with a VHDL-based Analog Hardw are C Language [38]. N : i) Extensi v e de vice model libraries including lea d i ng MOS, bipolar and MES- FET transistor; ii) Of fers a unique partitioning scheme allo wing the use of dif ferent algorithms on dif fering portions of design; iii) 3 to 10 g ain in simulation speed o v er other commercial SPICE simulators, while maintaining same accurac y; i v) V arious simulation studies: pole-zero, enhanced Monte-Carlo analys is and reliability simulation. H : W eak interf ace capability with other simulators. IsSPICE4 F : An impro v ed v ersion from Spice3f5 and XSpice, adding some strong interacti v e features C and e xtensions[39]. N : i) Adds a v ariety of ne w models: lossy transmission line model, GaAs Mesfet models, JFET model, etc.; ii) Friendly user interf ace; iii) V arious simulation Co-Simulation Interfacing Capabilities in De vice-Le vel Cir cuit Simulation T ools () Evaluation Warning : The document was created with Spire.PDF for Python.
670 ISSN: 2088-8694 T able 1. Comparison of De vice-Le v el Circuit Simulators (Continued) Simulators F : Salient F eatur es; N : Adv antages; H : Disadv antages studies: DC and A C sensiti vity , transient, pole-zero analysis, etc. H : Limited Interf acing capabil- ity . Gnucap F : A general purpose mix ed analog and digital circuit simulator , fully interacti v e, compat- R ible to SPICE, containing a s imple beha vioral modeling language [40]. N : i) Emplo ys plugins to mak e the simulation e xtremely fle xible; ii) Fle xible interf ace to other softw are; iii) Impro v es ac- curac y through rigorous error control method; i v) Multiple simulation languages, including Spice, V erilog and Spectre. H : User interf ace requires w ork. L TSpice F : A high performance simulator , with man y enhancements based on traditional SPICE R simulator [41]. N : i) Stable SPICE circuit simulation with unlimited number of nodes; ii) F ast simulation of switching mode po wer supplies (SMPS); iii) Lar ge model library: o v er 1100 macro- models of Linear T echnology products and 500+ SMPS. H : i) More simulation studies are better to add; ii) Interf acing capability with other simulators can be impro v ed. TIN A-TI TM F : A user -friendly , po werful circuit simulator based on a v ersion of SPICE [42]. N : i) C Includes more SPICE models and e xample circuits; ii) Supports multi-core processor and mak e simulations run 2-20 times f aster; iii) Friendly user graphic interf ace; i v) V arious simulation stud- ies: DC, A C, transient, noise analysis. H : Interf acing capability with other simulators can be impro v ed. where f 1 ( ) ; f 2 ( ) ; : : : ; f m ( ) are all nonlinear operators [1]. F or the Ne wton-Raphson algorithm, the Jacobian matrix must be formulated, which is gi v en as J ( x k ) = 0 B @ @ f 1 @ x 1 j x k @ f 1 @ x 2 j x k @ f 1 @ x m j x k . . . . . . @ f m @ x 1 j x k @ f m @ x 2 j x k @ f m @ x m j x k 1 C A (7) where x k is the solution at the k th Ne wton iteration [1]. Then, a linearized system of equations is obtained as: x k +1 = x k J ( x k ) 1 F ( x k ) (8) where x k +1 is the solution at the ( k + 1 )th iteration [1]. Notably , when e v aluating the Jacobian matrix, Saber applies a simplified subdi vision technique to calcu- late the first deri v ati v es thereby reducing the computational b urden at e v ery iteration, which is dif ferent from SPICE [10]. When applying the Ne wton-Raphson method, the SPICE-li k e simulators apply pre-specified tolerances (e.g. ABST OL , REL T OL , and CHGT OL ) to determine con v er gence to a v alid solution. In contrast, Saber uses no tolerance to determine con v er gence, since the system of equations are e v aluated piece wise linearly and solv ed e xactly [10]. The CODECS simulator , ho we v er , use a modified tw o-le v el Ne wton algorithm [16] in this step. Apart from Ne wton- Raphson, the Katzenelson algorithm is based on piece wise linear systems. Sample points are used to find the linear re gions for e v ery nonlinea r de vice [2]. Finally , in order to solv e the linear algebraic equations (8), SPICE-lik e simula- tors and Saber use the methods of Gaussian elimination or LU decomposition techniques with forw ard and backw ard substitutions [1], [11], whilst the TIT AN simulator uses a v ectorized solv er method [31]. The aforementioned transient analysis procedure (for e.g., in Saber) can be illustrated by the flo wchart in Fig. 1. F or formulating the Jacobian matrix other circuit simulators mostly utilize the standard techniques mentioned abo v e, ho we v er , b ut the y also use alternate methods. F or e xample, PETS uses a no v el algorithm to decide the accurate element states of its piece wise-linear netw orks as well as an ef ficient w ay to a v oid its piece wise-linear and reacti v e elements from changing v alues with each time-step, thereby k eeping the system matrix constant [32]. 3. CO-SIMULA TION TECHNIQ UES IN DEVICE-LEVEL CIRCUIT SIMULA T ORS There are three main co-simulation patterns: co-simulation of de vice-le v el circuit simulators and system- le v el simulators, analog and digital co-simulation of circuit simulators, and co-simulation of circuit simulators with programming languages. IJPEDS V ol. 6, No. 4, December 2015: 665 682 Evaluation Warning : The document was created with Spire.PDF for Python.
IJPEDS ISSN: 2088-8694 671 Read initial point f ile Ordinary differential equations Gear or Trapezoidal numerical Integration Linearization (Newton-Raphson or Katzenelson Algorithm) Linear algebraic equations Direct matrix t echnique e.g. LU decomposition Solution converged? Error within bound? Modify time-step Process events Finish ? Update time vector Write results System matrix f ormulation e.g. Modified Nodal Approach Initial guess of the next time-step Nonlinear algebraic equations N Y N Y N Y Figure 1. Flo wchart of transient analysis operation [2]. 3.1. Co-Simulation of De vice-Le v el Cir cuit Simulators and System-Le v el Simulators In this section, co-simulation e xamples for de vice-le v el circuit simulators are discussed. 3.1.1. Saber R and Matlab/Simulink R Saber is a de vice-le v el circuit simulator which specializes in po wer electronic simulation, while Simulink is v ersatile in b uilding control systems. Lik e other m ulti-domain designs, the co-simul ation between Saber and Mat- lab/Simulink can be v ery ef fecti v e in man y circumstances. F or instance, the Saber solution could calculate de vice losses which Simulink by itself may not be able to do. The procedure of using the Saber -Simulink co-simulation tool is discussed in [43]. The principle of Saber -Simulink interf acing is illustrated in Fig. 2. Notably , the running processes of the tw o simulators are fully independent e xcept when the y need to e xchange data at fix ed period of interv als. This communication mechanism is realized by using an S-function. Additionally , a Saber Co-simulation block (SaberCosim.mdl) should be inte grated into the Simulink model, which is then imported in to Saber interf ace using Saber -Simulink co-simulation tool. This tool is responsible for producing the required co-simulation interf ace symbol and the MAST template. This e xpressi v e user interf ace mak es it possible for the co-simulation to run completely in the Saber interf ace. Ref. [6] presented a method to construct a high-v oltage source circuit system by the h ybrid modeling of Saber and Simulink. In their designs, the Saber softw are w as responsible for the po wer electronic circuit, while Simulink w as in char ge of b uilding a fuzzy PID controller , because Saber is more specialized and po werful in switched-circuit analysis, while Matlab/Simulink is more po werful in control sys tem design. The results sho wed that the co-simulation of Saber and Simulink w as a highly ef ficient w ay to analyze and design a switching circuit system which included an intelligent control system. Co-Simulation Interfacing Capabilities in De vice-Le vel Cir cuit Simulation T ools () Evaluation Warning : The document was created with Spire.PDF for Python.
672 ISSN: 2088-8694 MATLAB ϟ ϟ Simulink Diagram Saber ϟ   Sketch SaberCosim.mdl File inserted Interface Symbol MAST Template S-function Exchange data at fixed period Figure 2. Illustration of Saber R Simulink R co-simulation interf ace [43]. PSpice ϟ ϟ MATLAB SIMULINK ϟ SLPS Block A CIR file Analysis settings Net list information Figure 3. Illustration of PSpice R Simulink R co-simulation interf ace [47]. Similarly , [44] applied the Saber and Simulink co-simulation for the modeling of a high pulse po wer circuit. In their studies, the ph ysical model of Re v ersely Switched Dynistor (RSD) w as constructed using Simulink, and Saber w as used to model the pulse po wer circuit and the magnetic switch. In f act, RSD ph ysical model and circuit module can both be realized in M file form of Matl ab and then the whole simulation can be finished just in only one softw are. But there are tw o disadv antages of this methodology , one is that a little change of the circuit structure leads to re write the code; the other is that the poor visualization of circuit structure. The Saber -Simulink Co-Simulation en vironment (SSCSE) allo ws the algorithm realized in Simulink and po wer circuit modeled in SABER, respecti v ely , and the co- simulation method can o v ercome the tw o dra wbacks abo v e easily . 3.1.2. PSpice R and Matlab/Simulink R PSpice is a simulation tool which can tackle models in analog and mix ed-signal en vironments, whereas Matlab/Simulink, as a platform for multi-domain simulation, is mainly based on approximate c on t inuous-time and discrete-time models of dynamic systems. Ob viously , the y both ha v e adv antages in a simulation project, since the former can mak e it possible for designers to perform simulation which includes realistic electrical models, such as the electrical circuits in switched reluctance motor in [45] and the zero v oltage switch (ZVS) in v erter in [46], while the latter is mainly focused on b uilding the whole system. The co-simulation between Matlab/Simulink and PSpice is realized by an interf a ce tool named the PSpice SLPS Interf ace, which enables electro-mechanical system designers to perform system-le v el simulation which include specific de vice-le v el circuit simulation. More specifically , the PSpice SLPS Interf ace mak es it possible for the designer to include realistic electrical PSpice models of actual components when per forming system-le v el simulation. The detailed procedure to use the SLPS interf ace is a v ailable in [47] and the co-simulation is illustrated in Fig. 3. Notably , the co-simulation of PSpice and Simulink is initialized by creating a CIR file, which specifies the PSpice analysis settings (e.g. the analysis time) and the net list information, the reby assigning the circuit b uilt in PSpice to the SLPS block, which, in turn, is inte grated into Simulink models [47]. The co-simulation i s dominated by Simulink, which e xchanges data with PSpice at its o wn time-step [48]. In general, the internal time steps of PSpice are much smaller than the Simulink time steps. If PSpice uses the minimum/maximum time step determined by Simulink, a con v er gence error will be lik ely to occur; that is, no result can be found. T o solv e this problem, SLPS calculates v alues suited to PSpice based on Simulink parameters, and then passing these to PSpice. Fig. 4 sho ws the mechanism clearly . System-le v el simulation can be finished in a short time, b ut it can only v erify the function of design and can not check the performance and non-idealities of the real circuits. De vice le v el simulation can solv e such problems b ut it usually needs a much l on ge r simulation run time. So through the combination of the meri ts of the Simulink and PSpice, the simulation run time can be diminished greatly while the simulation precision can be well ensured and a satisf actory simulation result is generated. The PSpice SLPS Interf ace has been used in man y electro-mechanical designs. These applications include the design presented in [5], which successfully simulates pipeline ADC circuits IJPEDS V ol. 6, No. 4, December 2015: 665 682 Evaluation Warning : The document was created with Spire.PDF for Python.
IJPEDS ISSN: 2088-8694 673 Figure 4. Data e xchange mechanism of PSpice R and Simulink R [48] MATLAB Simulink ϟ ϟ PSIM ϟ In Link Node Out Link Node SimCoupler Module Figure 5. Illustration of PSIM R Simulink R co-simulation interf ace [51]. and obtains satisf actory results. Also, [49] presents a method for the simulation of solar photo v oltaic (PV) cell by applying the PSpice and Simulink co-simulation interf ace; it b uilds a h ybrid model of the PV module using PSpice and Simulink. Additionally , Ref. [50] presents a co-simulation solution for a high ef ficienc y full-bridge DC-DC con v erter for fuel cell; because the Simulink has merits in b uilding the feedback controller of the con v erter while PSpice e xcels in modeling the electronic circuits. 3.1.3. PSIM R and Matlab/Simulink R Ref. [51] introduced the SimCoupler Module by which the co-simulation between PSIM and Matlab/Simulink w as made possible and pro vided detailed procedures to use this module. The principle of PSIM-Simulink interf acing is illustrated in Fig. 5. As a de vice-le v el circuit simulator which has special adv antages in po wer electronics simulation [52, 53], PSIM has disadv antages to b uild a control subsystem. Ho we v er , Matlab/Simulink is good at constructing control circuits. Therefore, the co-simulation between PSIM and Simulink has its unique v al ue to be studied and applied in man y circumstances. F or e xample, [54] b uilt a simulation platform for a three-le v el adjustable speed dri v e. In vie w of the merits of PSIM, the designer b uilt the main circuit of the three-le v el adjustable speed dri v e using PSIM, whereas the y constructed the control system, observ ed the output v oltage and performed F ourier analysis in Matlab/Simulink. Another case for the application of the co-simulation between PSIM and Simulink w as presented in [55], where the designer first compared PSIM and Simulink separately in simulating single-phase uncontrolled rectifier and three-phase controlled rectifier . After discuss ing the pros and cons of both simulators, the y concluded that the co-simulation between them w as a better solution. The simulation results states that Matlab/Simulink is a suitable platform for control and re gulation of the simulation processes, in additional to its dominant role in conducting research tasks. Con v ersely , PSIM is dedicated to po wer electronic circuits and machine simulation tasks wit h f ast and rob ust algorithms. So the authors concluded that the co-simulation between them w as a better solution. 3.1.4. SPICE and Matlab/Simulink R Ref. [56] introduces a co-simulation interf ace between Simulink and SPICE, which is realized by a ne w Simulink block SLSP . The SLSP block is written in C MEX S-function, which is responsible for reading in the circuit file, initializing the simulation, performing the time-domain numerical inte gration and manipulating the SPICE- Simulink communication. The co-simulation setting is simple to b uild up as an SLSP block in the Matlab/Simulink en vironment with a parameter for the name of a SPICE circuit file. This mechanism is described in Fig. 6. Ref. [56] also elaborates an application to use Simulink-SPICE Interf ace to simulate a speed control system of a dc motor . Specifically , the whole system is modeled in the Simulink en vironment, e xcept for the PI controller , the machine is modeled in SPICE. Co-Simulation Interfacing Capabilities in De vice-Le vel Cir cuit Simulation T ools () Evaluation Warning : The document was created with Spire.PDF for Python.
674 ISSN: 2088-8694 MATLAB SIMULINK ϟ ϟ SLSP Block Spice Netlist file Integrated to Simulink ϟ Based on S-function Name specified in SLSP Figure 6. Simulink R -SPICE interf ace mechanism [56]. MATLAB Simulink ϟ CASPOC ϟ From SLNK To SLNK S-function Figure 7. CASPOC R Simulink R co-simulation interf ace [35]. 3.1.5. PLECS R and Matlab/Simulink R PLECS, is a piece-wise linear electrical simulator that enters the circuit information as netlists, which are in turn inte grated into Matlab/Simulink using S-functions. Compared to the Po wer System Blockset models of Simulink, the PLECS impro v es the performance greatly using the ideal switch models [26]. Ref. [57] applies Simulink and PLECS co-simulation in a photo v oltaic ener gy con v ersion system. The control subsystem is modeled using Simulink, whilst the plant subsystem, including DC supply , in v erter , LCL filter and utility grid, is modeled using PLECS. The simulation results sho w that the co-simulation between Simulink and PLECS is much f aster than using Simulink transfer functions. 3.1.6. CASPOC R and Matlab/Simulink R CASPOC is a circuit simulator dedicated to po wer electronics and electrical dri v es simulation [34]. Ref. [35] describes that simulations performed by coupling CASPOC and Matlab/Simul ink of fer the adv antage of modeling of the control of Matlab/Simulink while connecting with CASPOC modeling the po wer electronics de vices and circuits. Fig. 7 illustrates CASPOC and Sim ulink co-simulation interf ace. The co-simulation is performed through data e xchange between CASPOC and Simulink at e v ery time step. Simulati o ns in Simulink are controlled from CASPOC and the control then can be implemented in Simulink and co-simulated with a simulation in CASPOC. The co-simulation be gins with creating tw o function blocks in CASPOC before starti ng the simulation in Simulink to communicate data. Then CASPOC is running in the background while data can be e xchanged between the tw o simulators with the use of S-function block in Simulink. The S-function block represents the complete CASPOC simulation. It is required that the step-size in CASPOC and Simulink must be equal during the simulation. 3.2. Analog and Digital Co-Simulation of Cir cuit Simulators Although co-simulation with some system-le v el simulators may be ef fecti v e in handling lar ge and comple x simulation tasks, for those applications such as po wer and mechatronic system simulations, which may include a great man y analog and digital subsystems, no v el co-simulation platforms ha v e become necessary [65]. F or e xample, simulators such as Saber w ork in analog and mix ed-signal en vironments, whereas man y of the control subsystems are performed in digital logic. Therefore, co-simulations with other digital simulators such as ModelSim R become ef fecti v e solutions [7]. In this section, se v eral interf acing instances for analog and digital co-simulation of circuit simulators are discussed. 3.2.1. Saber R and ModelSim R According to [7], the Saber/M o de lSim co-simulation interf ace enables Saber , an analog and mix ed-signal simulator , to support VHDL modeling. More specifically , a designer can model the analog part by using Saber MAST IJPEDS V ol. 6, No. 4, December 2015: 665 682 Evaluation Warning : The document was created with Spire.PDF for Python.