Qucs

A Tutorial
Component, compact device and circuit modelling using symbolic equations

Mike Brinson
Copyright c 2007 Mike Brinson <[email protected]> Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.1 or any later version published by the Free Software Foundation. A copy of the license is included in the section entitled GNU Free Documentation License.

Introduction
Qucs releases 0.0.11 and 0.0.12 mark a turning point in the development of the Qucs component and circuit modelling facilities. Release 0.0.11 introduced component values dened by equations and for the rst time allowed subcircuits with parameters. Release 0.0.12 extends these features to add model development using symbolic equations that are similar to compact device code written in the Verilog-A modelling language. In designing the latest Qucs modelling features the Qucs team has made a central focus of their work the need to provide the package with an interactive and easy to use modelling system which allows fast model prototype construction. Much of these new aspects have up to now been undocumented and are likely to be very new to most Qucs users. The aim of this tutorial note is to outline the background to these important package extensions and to provide real help to Qucs users who are interested in writing and experimenting with their own models. The text includes a number of illustrative examples for readers to try and experiment with.

Qucs electronic device and circuit modelling

Circuit simulation packages are complex software systems which often take years to mature to a stage where they are capable of analysing the current generation of integrated and discrete electronic circuits. Most circuit simulators have a number of common basic attributes; rstly circuits are represented by a textual netlist or a schematic diagram which contains all the information required by a simulator to analyse the performance of a circuit, and secondly a simulation engine which undertakes the calculation of circuit performance in one or more dierent circuit domains such as DC, AC or transient, and thirdly a post simulation processing system which structures and displays the simulation data in both tabular and graphical forms. All circuit simulators have one other important attribute, namely that they represent individual electronic components by a model, or abstraction, in a way that can be understood and analysed by the simulation engine when undertaking a simulation task. Without component models the science of circuit simulation would not have developed to the stage it has today. From a users point of view component models are the key to simulator productivity; the greater the number of dierent models the easier it becomes to analyse mixed analogue and digital electronic systems. Shown in Fig. 1 is a block diagram of the analogue component modelling and simulation facilities currently provided by the Qucs package. The diagram is structured as a ow chart which emphasises the dierent device modelling routes. When Qucs was rst released only two of these were available for users to develop new device models. The rst of these has been used extensively by the package developers to construct the built-in models that are distributed with each Qucs release. This fundamental route involves hand coding the C++ code for a new model1 , its compilation and linking with the core Qucs C++
1

The technical details of the built-in models are described in: Qucs Technical Papers, Stefan Jahn,

code. Obviously, this does require a specialised knowledge of the Qucs model programming interface2 , the necessary C++ skills, including a good working knowledge of the Trolltech Qt toolkit3 . At the time of writing these notes the latest device to be added to Qucs using this approach is the exponential pulse source4 . Models based on hand written C++ code are normally restricted to basic devices that form the fundamental component core of a simulator - particularly where simulation computational eciency is important. One disadvantage of this approach, is the obvious one, in that the time to implement a new model increases disproportionately with increasing model complexity. For most Qucs users this route would not be the most natural to use when developing new models. However, for the specialist who spends a signicant amount of time researching new device models this has always in the past, been the route of choice. Unfortunately, modern semiconductor device models are becoming so complex that the model development time can stretch into months or even years and requires typically thousands of lines of C or C++ code to characterise a model5 . With the more complex models the problem of nding bugs in the model code also acts as a limit to fast model development. For the average Qucs user their rst introduction to the software is probably through constructing circuit schematics made entirely from the standard component models built into the package and the testing of their performance by launching the simulator from one of the Qucs simulation icons.6 The next natural stage in the Qucs modelling and simulation learning curve is the use of subcircuits where groups of built-in components are collected together to form a higher level circuit block. These blocks are often arranged with a common theme, forming a Qucs library. The process of modelling new devices/circuits is normally done by connecting existing component models and user dened subcircuits. With this type of modelling higher level functional models can only be constructed from existing fundamental components or previously constructed subcircuits. Engineers often call this approach to modelling, macromodelling. Qucs releases up to 0.0.10 relied on macromodelling for functional model development via the Qucs schematic interface. This route remains popular amongst most Qucs users because it is easy to understand, is fully interactive and allows straight forward testing of new models. One feature that is common to all components included in Qucs releases up to 0.0.10 may not be immediately obvious to readers, namely that, with the exception of sweep variables, component values could only be numbers, for example R1 = 1k, and were not allowed to be represented by algebraic expressions like R1 = Value1, where Value1 = 100.0+50 X. Its also worth pointing out at
Michael Margraf, Vincent Habchi and Raimund Jacob, http://qucs.sourceforge.net/technical.html. 2 Writing the documentation for the Qucs model programming interface is on the to do list and will be completed, when time allows, sometime in the future. 3 Qt is a registered trademark of Trolltech, Norway; http://www.trolltech.com/copyright. 4 Added by Gunther Kraut on 15 April 2007. This device has been added for compatibility with SPICE. 5 A good introduction to writing compact device models is given in How to (and how not to) write a compact model in Verilog-A, Georey J. Coram, 2004, Proc. 2004 IEEE International Behavioral Modeling and Simulation Conference (BMAS 2004), pp 97- 106. 6 The Getting Started with Qucs tutorial by Stefan Jahn outlines a number of basic simulation techniques; http://qucs.sourceforge.net/docs.html.

Verilog-A Compact device code

Hand coded device model C++ code

Nonlinear equation defined devices

Component data processing using Qucs equations

SPICE netlist

SPICE preprocessor

SPICE parameterised netlist

ADMS compiler Schematic capture symbols QUCS GUI Circuit entered using schematic capture

User defined subcircuit schematic capture symbol Qucs Library components

C++ code

User defined subcircuits Symbolic equations

Qucs components

Simulate

Generate Qucs netlist code from GUI schematic, including conversion of SPICE code to Qucs format

Qucs Tools: Line Calculator Attenuator Design Matching Circuits Filter Design

QUCSATOR C++ component code compiled and linked to Qucsator core C++ code via API

Post simulation data processing using Qucs equations

Simulation output data

Qucs plots and tables

Figure 1: Qucs analogue component modelling and simulation block diagram (not including optimisation)

this point that during simulation, again performed by Qucs releases up to 0.0.10, component values were required to remain constant and could not be a function of the circuit variables such as voltage, current or charge. One way to remove the component value restrictions imposed by early Qucs releases is to model devices and circuits using preprocessor extended forms of the SPICE netlist language. Circuit design equations can then be embedded in SPICE netlists and the calculation of component values completed by the SPICE preprocessor. Both the SPICE to Qucs and OP AMP tutorials7 outline in detail the steps required to merge circuit design and simulation in this way. This modelling route is a very important and powerful model development tool. So much so that ongoing tests to identify how compatible Qucs is with the industrial standard SPICE 2g6 and 3f5 syntax are currently being undertaken as part of the Qucs development schedule8 . Although perfectly viable as a model development tool the use of an extended SPICE netlist language has a number of serious disadvantages, namely that not all the Qucs built-in component models have equivalent SPICE models and secondly text netlists are the only entry medium for describing models. The previous paragraphs give a brief statement of the dierent component modelling routes that were available up to release 0.0.11. Qucs 0.0.11 is very much a modelling water shed in that symbolic equations were introduced for the calculation of component values, previously equations were only allowed when structuring simulation output data for post simulation listing or plotting. Release 0.0.11 allows the following types of variable; 1. sweep variables, 2. equations left hand side, 3. component parameters left hand side (e.g. R1.R), 4. subcircuit parameters and 5. simulation output data. With each Qucs release the number of analysis functions, and other data processing features, included in the Qucs equation set continues to expand9 . From release 0.0.11 parameters are also allowed with subcircuits so that data can be passed to a model. This allows generalised subcircuit/macromodels to be developed for popular devices such as operational ampliers. Through the use of embedded design equations within subcircuits and parameter passing it became possible to construct powerful models that mix both circuit
7 Qucs simulation of SPICE netlists and Modelling Operational Ampliers, Mike Brinson, http://qucs. sourceforge.net/docs.html. 8 Qucs: Report Book; SPICE to Qucs test reports, Mike Brinson, http://qucs.sourceforge.net/ docs.html. 9 See Measurement Expressions Reference Manual, Gunther Kraut and Stefan Jahn, http://qucs. sourceforge.net/docs.html.

design procedures and the calculation of individual component values. Qucs 0.0.11 still imposed the restriction that equations could not be functions of voltage, current or charge. With the release of Qucs 0.0.12 the voltage, current and charge restrictions imposed on equations will nally be relaxed. The introduction of a new device modelling component called the equation dened device (EDD) allows rstly device current to be formulated as a function of voltage, and secondly device charge to be calculated as a function of voltage and current. The syntax adopted for the new model borrows heavily on the compact device modelling approach taken by the Verilog-A modelling language. Some readers will probably have noted that so far these notes make no reference to the ADMS model development route illustrated in Fig. 1. ADMS stands for Automated device model synthesizer10 and includes a Verilog-A to C/C++ compiler. It allows compact device models to be described in the Verilog-A language then compiled to C/C++ and the resulting code linked with the Qucs core simulation code11 . Model development using ADMS is similar to the fundamental hand coded C++ model development route except that model development is greatly simplied by the power of the high level Verilog-A language. A strong relationship exists between the ADMS and EDD modelling procedures in that EDD can be considered a fast interactive model prototyping method whose equations can easily be expressed in Verilog-A and compiled into C/C++ code for permanent inclusion in the Qucs simulator12 . The opening paragraphs attempt to outline the available device modelling techniques that are central to the functioning of the Qucs package. The remaining sections of this tutorial note are devoted to illustrating the power of Qucs modelling through the introduction of a number of illustrative examples. Initially these start from a simple, and hopefully familiar, point and then proceed to more complex examples which present many of the concepts lightly touched upon in the opening text.

Extending circuit simulation capabilities with equations

Just adding component value calculations, via equations, to a circuit simulator immediately increases the underlying design and simulation capabilities way beyond that found in earlier generation simulators. Consider the simple RC circuit shown in Fig. 2. Capacitor Cap is stepped from 0.1F to 1.1F and the small signal AC response of the network calculated. In this example the values for both R1 and Cap are given as numeric values. The simulation test shows the eect of stepping the value of one component through a series of values and
L.Lemaitre, C.C. McAndrew, and S. Hamm, ADMS - Automated Device Model Synthesizer, Proc. IEEE CICC, 2002. 11 For more details see, Qucs Description: Verilog-AMS interface, Stefan Jahn and Hl`ne Parruitte, ee http://qucs.sourceforge.net/docs.html. 12 Appendix A gives an operator and function comparison table for Qucs and Verilog-A.
10

recording the eect of component changes on circuit performance. In other words this is a classical circuit analysis use of a circuit simulator. In a real design situation dierent data is often required. Most designers would prefer to nd the value of Cap that gives a specic RC cut-o frequency (fc ) for a specied value of R1. This is the type of investigative problem where adding equations into the simulation process generates more informative results. Shown in Fig. 3 is a similar RC network to that illustrated in Fig. 2. Capacitor voltage V Cap is given by: VCap = V1
2 2 1 + 2 C1 R1

(1)

where the cut-o frequency in the voltage transfer function is fc = 1 2 R1 C1 (2)

Hence, by expressing Cap as a function of fc and stepping fc through a range of frequencies, the eect of capacitance changes on the voltage transfer function can be found. More importantly a nomogram of Cap values against fc can be plotted giving the circuit designer a visual aid for determing the value of Cap required for given values of R1 and fc . Although the circuits shown in Figs. 2 and 3 are very basic they do demonstrate how much more powerful a circuit simulator becomes when component values are calculated using equations.

Low pass active lter design with embedded design equations

In this section a more advanced circuit design example is introduced to illustrate the power of embedded design equations in a Qucs simulation schematic. A second order Sallen-Key low pass lter is employed for this task because it is so well known and most readers are likely to have met its design in the past. A second order low pass lter is represented by the voltage transfer function: A(S) = A0 Vout = Vin (1 + a2 S + b2 S 2 ) (3)

where A0 is the passband DC gain and coecients a2 , b2 are for Bessel, Butterworth, Tschebysche or similar polynomials. The following list13 gives the second order coecients for the Bessel 1.3617, 0.618; Butterworth 1.4142, 1.000; and 3dB ripple Tschebysche 1.065, 1.9305, polynomials. The second order Sallen-Key low pass lter circuit is shown in Fig. 4. This circuit has a voltage gain transfer function given by:
See OP Amps for everyone, Chapter 16: Active lter design technology, Texas Instruments, August 2002, SL0D006B, PP 16.1,16.63.
13

VCap

Parameter sweep
SW1 Sim=AC1 Type=lin Param=Cap Start=0.1u Stop=1.1u Points=11

V1 U=1 V

R1 R=1k

C1 C=Cap

ac simulation
AC1 Type=log Start=1Hz Stop=1 MHz Points=61

VCap.v

0.5

0 1 10 100 1e3 1e4 acfrequency 1e5 1e6

1e-6

Cap

5e-7

0 2 4 6 number 8 10

Figure 2: A simple RC circuit simulation using numerical component values

VCap

ac simulation
AC1 Type=log Start=1Hz Stop=1 MHz Points=61

Parameter sweep
SW1 Sim=AC1 Type=log Param=fc Start=10 Stop=1000 Points=21

V1 U=1 V

R1 R=Rvalue

C1 C=Cap

VCap.v

0.5

Equation Eqn1 Rvalue=1000 Cap=1/(2*pi*Rvalue*fc)

0 1 10 100 1e3 1e4 acfrequency 1e5 1e6

1e-5

Cap

1e-6

1e-7 10 100 fc 1e3

Figure 3: A simple RC circuit simulation employing equation determined component values

A(S) = where

A0 2 1 + c [C1 (R1 + R2 ) + (1 A0) R1 C2 ] S + c R1 R2 C1 C2 S 2 A0 = 1 +

(4)

R3 (5) R4 This can be simplied by letting R1 = R2 = R and C1 = C2 = C; the transfer function then becomes: A(S) = By comparison a2 = c R C (3 A0) and b2 = (c R C)2 Fixing C and solving for R and A0, yields a2 b2 , and A0 = 3 . R= c C b2 Also once A0 is known the value for R4 can be calculated using equation A0 = 1 + R3 . R4 (10) (8) (7) A0 . 1 + [c R C (3 A0)] S + (c R C)2 S 2 (6)

(9)

Hence by providing values for C and R3 the values for R and A0, and of course R4, can be determined for a specied cut o frequency f c. Figure 5 shows the nal design schematic and the simulation results for this example. A number of important observations can be made from Fig. 5: 1. One or more equation blocks hold both design and post simulation data processing equations plus assignments for named items: C, f c and R3 are given numerical values, the a and b polynomial coecients are set to the values introduced in the text, and nally the design equations for R, A0 and R4 calculations are listed. 2. The order of entries in equation blocks is not important because Qucs automatically sorts out the data it requires when calculating equations. 3. The lefthand quantities in the assignment entries in the equation blocks are linked to the component values in the schematic, see for example C and R. 4. The OP27 operational amplier model is from the modied Qucs 0.0.11 OPAMP library. This model was generated using the SPICE to Qucs modelling route. 9

C2

SUB1

+
V1 U=1 V R1 R2 C1 OPA27(TI)

VEE Vout

V2 U=15 V

VCC V3 U=15 V

R3

R4

Figure 4: The Sallen-Key lowpass active lter circuit 5. To design and simulate a Sallen-Key low pass lter with a dierent cut o frequency14 simply change the value of f c and rerun the Qucs simulator. 6. On completion of a simulation, pressing key F5 (Show last messages) causes the simulation log to be displayed. This includes the calculated values of the components and the netlist for the circuit, see Fig. 6. 7. One nal point of signicance that some readers may have noticed - all numerical values in equation blocks must be specied in scientic notation; electronic notation like 1k or 3nF is not allowed15 .

14 If the design calculations result in impractical values for the lter components then the value of C should be changed and the simulation repeated. 15 In long term it is expected that electronic notation will be allowed. The changes for this are on the to do list but at the moment the work has a low priority.

10

C2 C=C SUB1

+
V1 U=1 V R2 R=R R1 R=R C1 C=C OPA27(TI)

VEE Vout

V2 U=15 V

dc simulation
DC1

VCC V3 U=15 V

ac simulation
Equation AC1 Type=log Start=1 Hz Stop=100 kHz Points=101 R3 R=R3_calc 4 R4 R=R4_calc

Eqn1 C=22e-9 a2=1.065 b2=1.9305 fc=3000 R=sqrt(b2)/(2*pi*fc*C) A0=3-a2/(sqrt(b2)) R3_calc=4700 R4_calc=(A0-1)*R3_calc gain_dB=dB(Vout.v) gain_phase=rad2deg(unwrap(angle(Vout.v)))

Vout.v

0 1 10 100 1e3 acfrequency 1e4 1e5

0 0 gain_phase -50 -200 1 10 100 1e3 acfrequency 1e4 1e5 1 10 100 1e3 acfrequency 1e4 1e5 gain_dB

-100

Figure 5: The Sallen-Key lowpass active lter schematic with embedded design equations

11

Output : n e t l i s t content 13 R i n s t a n c e s 5 C instances 2 VCCS i n s t a n c e s 5 CCCS i n s t a n c e s 2 VCVS i n s t a n c e s 1 CCVS i n s t a n c e s 8 Vdc i n s t a n c e s 1 Idc i n s t a n c e s 1 Vac i n s t a n c e s 4 Diode i n s t a n c e s 2 BJT i n s t a n c e s 1 DC i n s t a n c e s 1 AC i n s t a n c e s creating netlist . . . c h e c k e r n o t i c e , v a r i a b l e Vout . v in e q u a t i o n g a i n dB not y e t d e f i n e d c h e c k e r n o t i c e , v a r i a b l e Vout . v in e q u a t i o n g a i n phase not y e t d e f i n e d kB = 1 . 38065 e 23 e = 2 . 71828 p i = 3 . 14159 C = 2 . 2 e 08 a2 = 1 . 065 b2 = 1 . 9305 f c = 3000 R = 3350 . 51 A0 = 2 . 2335 R3 c a l c = 4700 R4 c a l c = 5797 . 43 kB = 1 . 38065 e 23 e = 2 . 71828 p i = 3 . 14159 kB = 1 . 38065 e 23 e = 2 . 71828 p i = 3 . 14159 kB = 1 . 38065 e 23 e = 2 . 71828 p i = 3 . 14159

Figure 6: Message output log for the simulation of the Sallen-key low pass circuit: for brevity only the component value section is given

12

Introduction to Qucs subcircuit parameters

Subcircuits are a concept that has been part of the simulation scene for a long time. All circuit simulators based on SPICE have subcircuits as part of their basic device compliment. This is not surprising because they form a natural way of breaking an electronic system down into a number of smaller self contained functional blocks. What is surprising however, is the fact that a signicant number of simulators, including SPICE 2g6 and 3f516 , do not allow parameters to be passed to a subcircuit. Parameter passing appears to have been rst introduced when a number of the popular commercial circuit simulators were being developed17 . Qucs releases up to version 0.0.10 are similar to SPICE in that they also did not allow parameters with subcircuits. This very important limitation has been removed with release 0.0.11, which allows parameters to be attached to component symbols and used in subcircuit equation calculations. Shown in Fig. 7 are the circuit schematic and user generated symbol for a simple harmonic generator with a fundamental and three harmonic sinusoidal components. Parameters f1 to f4 determine these frequency components. Notice that an equation block, at the circuit schematic level, is used to calculate the harmonic frequencies. Parameters ph1 to ph4 set the phase of the individual sinusoidal oscillators. The process of attaching parameters, and their default values, to a subcircuit symbol is straightforward; simply right click on the symbol subcircuit name, SUB1 in Fig. 7, and an Edit Subcircuit Properties dialog box appears allowing parameter names and their default values to be entered18 . Subcircuit parameters and their values are normally displayed as a list underneath the subcircuit name. Changing parameter values is done in a similar fashion to changing the values of the standard built-in components. The diagram and simulation results illustrated in Fig. 8 show a waveform formed from a fundamental and two harmonics. An equation block is employed to calculate and plot the amplitude and power spectral densities of the harmonic waveform. By changing the fundamental frequency, signal amplitudes and phases dierent wave shapes can be generated by resimulating the circuit. In this example transient analysis is used to generate the harmonic waveform with the run time set to 10ms and the number of points equal to 50019 . This gives a sampling time of 20s and a sampling frequency of 50kHz. Equation block Eqn1 demonstrates how the Qucs functions20 can be used to postprocess simulation generated data - in this example they are used to compute the DFT of the harmonic generator waveform, convert the resulting
One of the reasons SPICE preprocessors were developed was to allow parameter passing to subroutines, for more details see Qucs Tutorial: Qucs simulation of SPICE netlists, Mike Brinson, http://qucs. sourceforge.net/. 17 See, for example, the extended netlist format originally designed by the MicroSim Corporation for the PSpice circuit simulator. 18 See Appendix B for a more detailed description of the procedure. 19 Qucs function length() determines the correct data length in equation block Eqn1 calculations. 20 If you have used a program like Octave, or indeed Matlab, many of these functions should be familiar to you. These functions provide Qucs with powerful numerical resource which signicantly extends the range of problems that Qucs can analyse.
16

13

V1 U=f1_amp P_sig f=f1 Phase=ph1 Equation V2 U=f2_amp f=f2 Phase=ph2 V3 U=f3_amp f=f3 Phase=ph3 V4 U=f4_amp f=f4 Phase=ph4 Eqn1 f2=2*f1 f3=3*f1 f4=4*f1

HG1
SUB1 f1=1000 f1_amp=1.0 f2_amp=0.0 ph1=0.0 ph2=0.0 f3_amp=0.0 f4_amp=0.0 ph3=0.0 ph4=0.0

Figure 7: Harmonic generator subcircuit schematic and symbol spectra from double sided to single sided form, compute and plot the amplitude and power spectral densities.

14

Equation hg_sig Eqn1 ts=(max(time)-min(time))/length(time) fs=1/ts Adft=dft(hg_sig.Vt) LAdft=length(hg_sig.Vt) Amp2=2*Adft[1:(LAdfto2)-1] LAdfto2=LAdft/2 Amp_squared=Adft[:LAdfto2]*conj(Adft[:LAdfto2]) Amp=sqrt(Amp_squared) f_bin=linspace(1, LAdfto2, LAdfto2) f=(f_bin-1)*fs/LAdft PLAmp=PlotVs(2*Amp/LAdft,f) PLPower=PlotVs(4*Amp*Amp/(LAdft*LAdft),f)

transient simulation
TR1 Type=lin Start=0 Stop=10 ms Points=500

HG1
R1 R=50 Ohm SUB1 f1=1000 f1_amp=5.0 f2_amp=2.0 ph1=0 ph2=0 f3_amp=2 f4_amp=0 ph3=0 ph4=90

5 hg_sig.Vt

-5

1e-3

0.002

0.003

0.004

0.005 time

0.006

0.007

0.008

0.009

0.01

6 Power Spectral density (V^2) PLPower Amplitude Spectral density (V) PLAmp

30

20

10

0 0 5e3 1e4 1.5e4 Frequency Hz 2e4 2.5e4

0 0 5e3 1e4 1.5e4 Frequency Hz 2e4 2.5e4

Figure 8: Harmonic generator subcircuit test circuit and simulation waveforms

15

Building universal macromodels using subcircuits and parameters

Passing parameters to subcircuits allows universal macromodels to be built. One obvious application of this technique is the modelling of operational ampliers (OP AMP) and other integrated circuits. The approach adopted is similar to that outlined in the last section. However, because of the complexity of the models it is advisable to break a model into a series of smaller blocks. These are then combined to form a complete subcircuit macromodel. Two techniques are possible when partitioning models, these are demonstrated next. Shown in Fig. 9 is a simple AC OP AMP model21 consisting of an input stage, an intermediate gain stage and an output stage22 . An equation block, if needed, is associated with each stage. These blocks contain the equations for calculating the component values in a given stage. A single schematic symbol represents the model. This has a list of parameters attached. The ow of information into a macromodel starts with parameters passed into a subcircuit, via a schematic symbol, then onto the equation blocks, where it is nally used to calculate the component values. Hence, by simply changing the subcircuit parameters dierent OP AMPs can be simulated using a single generalised macromodel. However, please note that dierent OP AMP circuit structures, or indeed technologies, naturally result in a series of generalised subcircuit macromodels to cover all possible types in a given device family. The second technique involves breaking a model down into smaller blocks and associating subcircuit symbols with each block. This approach is illustrated in Fig. 10. Again parameters are passed from the top level symbol (called AC in the schematic) to the inner subcircuits. These pass their own parameters down a subcircuit level where the component calculations are completed. The second technique results in two levels of subcircuit, accounting for the change in parameter name when passing a parameter from top to lower hierarchy. A second more detailed example showing how to construct nested subcircuits is presented later in these notes. In reality the macromodel for a typical OP AMP that models DC, AC and transient domains is much more complex than the model given in Fig. 9. The schematic for a typical multi-domain OP AMP modular macromodel is shown in Fig. 11, where each section of the macromodel is represented, if needed, by its own equation block. The test schematics shown in Figures. 12 and 13 show two OP AMPs with dierent subcircuit parameters. In Fig. 12 the small signal characteristics of unity gain closed loop ampliers clearly show the dierence in performance of the OP AMPs. Figure 13 is particularly interesting in that it illustrates how Qucs can be used to determine the eect
The term AC here refers to the fact that the OP AMP model chosen for demonstration purposes is a simplied version of a multi-domain OP AMP model. It only models small signal AC parameters and device input stage bias and oset properties. 22 The schematic shown in Fig. 9 forms part of a modular OP AMP macromodel. A detailed description of the function of individual networks and the derivation of the component equations is given in Qucs tutorial Modelling Operational Ampliers, Mike Brinson, http://qucs.sourceforge.net/docs.html.
21

16

Voff1 U=voff1 Ib1 I=ib Ioff1 I=ioff1 Ib2 I=ib R1 R=r1 Cin1 C=cd R2 R=r2

P_INN1

GMSRT1 G=0.01 S

Intermediate gain stage

RSRT1 R=1 RADO1 R=aoldc CP1 C=cp1

Voff2 U=voff2

GMP1 G=1 S

Equation Eqn2 cp1=1/(2*pi*gbp)

P_INP1

Equation Eqn1 voff1=voff/2 voff2=voff/2 ioff1=ioff/2 r1=rd/2 r2=rd/2

AC

+
SUB1 voff=0.7e-3 ioff=80e-3 rd=2e6 cd=1.4e-12 aoldc=200e3 gbp=1e6 ro=75 EOS1 G=1

ROS1 R=ro

P_OUT1

Output stage

Input Stage

Figure 9: Expanded AC OP AMP model showing circuitry and equation blocks of amplier oset voltage on integrator DC saturation by stepping resister rp through a series of values. The low oset voltage of the OP27 makes this device much more suitable for integrator circuits when compared to the popular UA741. These results can be conrmed by a simple calculation: the oset voltage for the UA741 is set at 0.7 mV and the amplier open loop DC gain at roughly 200, 000. The UA741 goes into saturation when rp is approximately 20 M. In saturation the OP AMP gain becomes open loop giving a DC output voltage of roughly 0.7e-3 2e5 or 14 V, which agrees with the Qucs simulation results.

17

IN AC P_INN1

ON

IN Inter stage IP SUB3 gbp=g_bp aoldc=a_oldc O

output IN O Stage SUB4 ro=r_o P_OUT1

+
SUB1 v_off=0.7e-3 i_off=20e-9 r_d=2e6 c_d=1.4e-12 i_b=80e-9 g_bp=1e6 a_oldc=200e3 r_o=75 P_INP1

Input Stage IP OP

SUB2 voff=v_off ioff=i_off rd=r_d ib=i_b cd=c_d

Voff1 U=voff1 Ib1 I=ib Ioff1 I=ioff1 Ib2 I=ib R1 R=r1

P_INN3

IN

ON

P_INN2

Input Stage IP OP Voff2 U=voff2

Cin1 C=cd R2 R=r2 P_INP3

SUB7 voff=v_off ioff=i_off rd=r_d ib=i_b cd=c_d

P_INP2

Equation Eqn1 voff1=voff/2 voff2=voff/2 ioff1=ioff/2 r1=rd/2 r2=rd/2

P_INP4 IN Inter stage IP SUB8 gbp=g_bp aoldc=a_oldc P_INN4 O

GMSRT1 G=0.01 S RSRT1 R=1 RADO1 R=aoldc CP1 C=cp1 P_OUT2 GMP1 G=1 S

Equation Eqn2 cp1=1/(2*pi*gbp)

output IN O Stage SUB9 ro=r_o PO1 ROS1 R=ro EOS1 G=1 P_OUT3

Figure 10: Modular AC OP AMP model showing subcircuits

18

Equation Eqn1 voff1=voff/2 voff2=voff/2 ioff1=ioff/2 r1=rd/2 r2=rd/2

Equation Eqn2 ecm1=1e6/cmrrdc ccm1=1/(2*pi*1e6*fcmz)

Equation Eqn3 p1=(100*pslewr)/(2*pi*gbp)-0.7 p2=(100*nslewr)/(2*pi*gbp)-0.7 psum=p1+p2

Voff1 U=voff1

P_INN

Ib1 I=ib Ioff1 I=ioff1 Ib2 I=ib

R1 R=r1 Cin1 C=cd R2 R=r2

RSUM1 R=1

SRC1 G=1 S

Voff2 U=voff2

P_INP RDCMZ R=650M RCM1 R=1M RCM2 R=1 SRC2 G=1 S Equation ECM1 G=ecm1 CCM1 C=ccm1 GMSRT1 G=0.01 S RSCALE1 R=100 VSR1 U=p1 RSRT1 R=1 RADO R=aoldc CP1 C=cp1 Eqn4 cp1=1/(2*pi*gbp)

SRC3 G=1 S

D1 Is=1e-12 A Bv=psum Ibv=20 mA

GMP1 G=1 S Equation Eqn5 cp2=1/(2*pi*fp2)

RP2 R=1

CP2 C=cp2

ROS1 R=ro P_VCC EOS1 G=1

GMP2 G=1 S

VLIM1 U=vlim1 Equation DVL1 Is=8e-16 A Eqn7 vlim1=vcc-vccm+1 vlim2=-vee+veem+1 P_OUT DVLM2 Is=8e-16 A

D2 Is=1e-15 A Cj0=0.0

D3 Is=1e-15 A Cj0=0.0 HCL1 G=hcl1

RDCCL1 R=100M

ECL G=1

Equation Eqn6 hcl1=0.9/idcoutm VLIM2 U=vlim2

P_VEE

Figure 11: Modular OP AMP subcircuit schematic with embedded component calculation equations 19

R1 R=4.7k

vout_op27

V3 U=1 V

R2 R=4.7k MOD

V1 U=15 V VCC

VEE

V2 U=15 V

SUB1 voff=30e-6 ib=15e-9 ioff=12e-9 rd=4e6 cd=1.4e-12 cmrrdc=1778279.4 fcmz=2009.0 aoldc=1778279.4 gbp=8e6 fp2=17e6 pslewr=2.8e6 nslewr=2.8e6 vcc=15 vee=-15 vccm=14 veem=-14 ro=75 idcoutm=32e-3

vout_ua741

R3 R=4.7k

R4 R=4.7k MOD

VCC

VEE

Equation

dc simulation
DC1

Eqn1 gain_ua741=dB(vout_ua741.v) phase_ua741=phase(vout_ua741.v) phase_op27=phase(vout_op27.v) gain_op27=dB(vout_op27.v)

number 1

vout_op27.V -3.87e-05

vout_ua741.V 0.001

ac simulation
AC1 Type=log Start=1 Hz Stop=100MHz Points=161

SUB2 voff=0.7e-3 ib=80e-9 ioff=10e-9 rd=2e6 cd=1.4e-12 cmrrdc=31622.77 fcmz=200.0 aoldc=199526.3 gbp=1e6 fp2=3e6 pslewr=0.5e6 nslewr=0.5e6 vcc=15 vee=-15 vccm=14 veem=-14 ro=75 idcoutm=34e-3

vout_ua741.v vout_op27.v

0.5

0 1 10 100 1e3 1e4 Frequency Hz 1e5 1e6 1e7 1e8

gain_ua741 gain_op27

-20

-40

-60 1 10 100 1e3 1e4 Frequency Hz 1e5 1e6 1e7 1e8

200

phase_ua741 phase_op27

100

-100 1 10 100 1e3 1e4 Frequency Hz 1e5 1e6 1e7 1e8

Figure 12: Unity gain OP AMP test circuit and waveforms 20

R3 R=rp

vout_ua741

R4 R=rp

vout_op27

C1 C=1 uF V1 U=15 V R2 R=1k

C2 C=1 uF V3 U=15 V

R1 R=1k MOD

VCC

MOD

VCC

V2 U=15 V VEE

V4 U=15 V VEE

SUB1 voff=0.7e-3 ib=80e-9 ioff=10e-9 rd=2e6 cd=1.4e-12 cmrrdc=31622.77 fcmz=200.0 aoldc=199526.3 gbp=1e6 fp2=3e6 pslewr=0.5e6 nslewr=0.5e6 vcc=15 vee=-15 vccm=14 veem=-14 ro=75 idcoutm=34e-3

Parameter sweep
SW1 Sim=DC1 Type=log Param=rp Start=1e3 Stop=1e9 Points=31

dc simulation
DC1

SUB2 voff=30e-6 ib=15e-9 ioff=12e-9 rd=4e6 cd=1.4e-12 cmrrdc=1778279.4 fcmz=2009.0 aoldc=1778279.4 gbp=8e6 fp2=17e6 pslewr=2.8e6 nslewr=2.8e6 vcc=15 vee=-15 vccm=14 veem=-14 ro=75 idcoutm=32e-3

15

vout_op27.V vout_ua741.V

10

0 1e3 1e4 1e5 1e6 rp 1e7 1e8 1e9

Figure 13: Integrator test circuits for determining DC saturation

21

More complex nested subcircuit models

In the previous two sections the example circuits only included subcircuits nested to one or two levels. Qucs does however, allow subcircuits to be nested to an arbitrary level and parameters can be passed down the nested chain to any depth required. Some care is needed when setting up the parameter passing sequence. Shown in Fig. 14 is a top level subcircuit with temperature swept between 10 and 110 centigrade. A simple resistor voltage divider network is at the bottom of a series of linked subcircuits, three levels down. R2 in the divider is a function of temperature. A schematic representation of the coupled subcircuits parameter passing sequence is also given in the right hand side of Fig. 14. Each level passes the value of temperature to its next lower member in the hierarchy. The Qucs generated netlist given in Fig. 15 clearly shows the parameter passing mechanism employed by Qucs. The ability to nest subcircuits and pass parameters down a hierarchy is an important feature in Qucs because it allows both circuit design and device data to be passed to dierent sections of the circuit/system being simulated. These parameters can, of course, be at dierent levels in a problem hierarchy providing a very exible and powerful design/analysis tool.
dc simulation
DC1 V1 U=1 V SUB 3 IN OUT SUB1 sp1=tsweep vp01 SUB3 IN OUT

Sp1 = tsweep

Parameter sweep
SW1 Sim=DC1 Type=lin Param=tsweep Start=10 Stop=110 Points=100

1 SUB2 vp01.V 0.8 IN OUT

0.6

Sp2=Sp1

0.4 20 40 60 tsweep 80 100 SUB1 IN OUT

tscan=Sp2

P1

R2 R=10k Temp=26.85

R1 R=10k P2 Temp=tscan Tc1=0.01 Tc2=0.015

Figure 14: A nested subcircuit showing parameter passing sequence

22

# Qucs 0 . 0 . 12 / media / hda2 /Qucs e q u a t i o n m o d e l l i n g p r j / r d i v t e s t tsweep 3 l . s c h . Def : r d i v sub1 temp n e t 1 n e t 0 t s c a n=27 R: R2 gnd n e t 0 R =10 k Temp= t s c a n Tc1=0 . 01 Tc2=0 . 015 Tnom=26 . 85 R: R1 n e t 1 n e t 0 R =10 k Temp=26 . 85 Tc1=0 . 0 Tc2=0 . 0 Tnom=26 . 85 . Def : End . Def : r d i v t e s t 6 temp n e t 1 n e t 0 sp2=27 Sub : SUB1 n e t 1 n e t 0 Type= r d i v sub1 temp t s c a n=sp2 . Def : End . Def : r d i v sub3 temp n e t 0 n e t 1 sp1=27 Sub : SUB1 n e t 0 n e t 1 Type= r d i v t e s t 6 temp sp2=sp1 . Def : End Vdc : V1 n e t 0 gnd U =1 V .DC:DC1 Temp=26 . 85 r e l t o l=0 . 001 a b s t o l=1 pA v n t o l=1 uV saveOPs=no MaxIter=150 s a v e A l l=no c o n v H e l p e r=none S o l v e r=CroutLU .SW:SW1 Sim=DC1 Type= l i n Param=tsweep Start=10 Stop=110 P o i n t s=100 Sub : SUB1 n e t 0 vp01 Type= r d i v sub3 temp sp1=tsweep

Figure 15: Qucs netlist for nested subcircuit showing parameter passing sequence

Introduction to equation dened devices (EDD)

Although adding symbolic equations to a simulator merges circuit design and analysis, it is by making these equations functions of circuit variables that the real power of modern circuit simulator is fully exploited. Equations that are functions of voltage, current and charge have to be continuously evaluated as a simulation progresses. This is in contrast to the type of equations previously introduced, which are only evaluated at the start of a simulation sequence. When component properties are functions of circuit variables considerable complexity is added to a simulation engine and as a result most simulators restrict such properties to a small number of component types, the most common being controlled current and voltage generators23 . Qucs version 0.0.12 introduces an equation dened device (EDD) which allows its terminal currents to be functions of voltage, and its stored charge to be functions of voltage and current. The EDD is similar, but more advanced, to the B type controlled source implemented in SPICE 3f5. It is capable of realising the same models as the SPICE B type device plus an extensive range of more complex compact device models. At this stage in Qucs development only the explicit
Probably the most well known non-linear controlled generators are the SPICE 2g6 and 3f5 forms, see A. Vladimirescu, Kaihe Zhang, A.R. Newton, D.O. Pederson and A. Sangiovanni-Vincentelli, SPICE Version 2G Users Guide, 1981, Department of Electrical Engineering and Computer Sciences, University of California, Berkeley, Ca. 94720, section 11, Appendix B: Nonlinear dependent sources., and B. Johnson, T. Quarles, A.R. Newton, D.O. Pederson and A. Sangiovanni-Vincentelli, SPICE3 Version f Users Manual, 1992, Department of Electrical Engineering and Computer Sciences, University of California, Berkeley, Ca. 94720, section 3.2.2.4, Non-linear dependent sources.
23

23

form of EDD is implemented24 . EDD is an advanced component that allows Qucs users to construct their own device models from a set of equations derived from the physical properties that characterise a device. The explicit form of EDD can only be used to develop models for devices where their dening equations can be transformed into the explicit analysis form required by Qucs25 . A range of functions similar to those dened in the Verilog-A compact device modelling language are provided by Qucs, making the equation modelling language easy to use and powerful. The ternary ? : form of the C language if statement has also been implemented to allow selection of model equations that change with diering device voltage, current and charge conditions. Before introducing the EDD symbol and its properties consider the following circuit simulation modelling problem: a model for a device is required where the output voltage is a function of two input voltages V IN1 and V IN2 , such that Vout (V IN1 , V IN2 ) = V IN1 V IN2 , where V IN1 and V IN2 can be arbitrary varying voltages. This type of model is dicult to simulate at functional level26 using the pre-version 0.0.12 built-in devices. A linear voltage controlled voltage source can be used to multiply a voltage by a constant. Multiplying by a second voltage is not possible with the linear controlled sources. Qucs AM modulated and PM modulated sources are the nearest that Qucs has to the source dened above. These sources however, only allow sinusoidal carrier signals. Illustrated in Fig. 16 is a four quadrant multiplier EDD which allows multiplication of two varying signals27 . The EDD device generates current I1 = V 2 V 3. This in turn is transformed to the output voltage by a unity gain current controlled voltage source SRC1. An EDD device can consist of up to 8 branches. The branches have currents, I1 to I8, voltages V1 to V8 and internal charges Q1 to Q8 respectively. Overall the total device current depends how these branches are connected. A similar comment applies to the total device charge. In Fig. 16 currents I2 and I3 are set to zero, charges Q2 and Q3 are also zero, and voltages V 2 = V IN1 and V 3 = V IN2 . Hence current I1 becomes the multiplication of V IN1 and V IN2 . The fact that currents I2 and I3 are set to zero implies that the terminals connected to the external input voltages have high impedance and act as voltage probes. The test circuit in Fig. 16 is shown with signal inputs generated by sinusoidal oscillators; V1 acts as a modulating signal and V2 as a carrier signal. The bottom right hand corner of Fig. 16 includes a second graph which illustrates the eect of changing signal V2 to a square wave source with 0.05ms period. (11)

24 See Qucs Technical Papers, Section 10.7: Equation dened models, Stefan Jahn, Michael Margraf, Vincent Habchi and Raimund Jacob, http://qucs.sourceforge.net/technical.html. 25 The Y parameters of the device being modelled must also exist for the explicit form of the EDD to be valid. 26 It is, of course, possible to model the multiplier operation at discrete component level e.g. using a Gilbert cell mixer circuit. 27 This model is based on an idea suggested by Stefan Jahn, during the EDD development phase.

24

transient simulation
TR1 Type=lin Start=0 Stop=1 ms Points=401

5 V1 U=1 V f=1 kHz Out.Vt Out V2 U=5 V f=10 kHz R1 R=50 Ohm VMULT1 0

-5 0 2e-4 4e-4 time 6e-4 8e-4 1e-3

D1 I1=V2*V3 Q1=0 I2=0 Q2=0 I3=0 Q3=0 Out1 Num=1

1

SRC1 G=1
2

1 vmul_2_tb:Out.Vt

In1 Num=2
3

In2 Num=3

-1 0 2e-4 4e-4 time 6e-4 8e-4 1e-3

Figure 16: Qucs EDD four quadrent multiplier model and test circuit

The Qucs EDD component

A two terminal model for a universal non-linear component with resistive, capacitive and inductive parallel branches is shown in Fig. 17. All three branches have elements that can be functions of either voltage or current or charge28 . The Qucs EDD component can be used to model this nonlinear device. One EDD element is needed to model the resistive and capacitive branches. A second EDD device, plus a gyrator, models the inductive branch. The total terminal current is the sum of the individual branch currents. Equations for the three branch currents are given by the following equations: I = I1 + IC + IL,
28

(12)

Each branch can be a function of one or more of these circuit variables but not necessarily all three at the same time.

25

where I1 = f (V ), IC = C(V, I) Also V 1 = i2, V 2 = IL, i2 = L(I) Giving IL = and VL=V2=V1= Hence I = f (V ) + C(V, I) 1 L(I)

dV 1 dQ1 = dt dt

(13)

dV 2 dIL , V 1 = L(I) dt dt

(14)

V 2 dt d dt V 1 dt

(15)

(16)

dV 1 1 + dt L(I)

(17)

The EDD is characterised by eight parallel branches each comprising a current component In and a charge component Qn, where n ranges from 1 to 8. The currents may be constants or dened by equations that are functions of the EDD branch voltages (these are designated V 1 to V 8). This form of the EDD component is known as the explicit EDD model. Please note, EDD currents cannot be functions of current. However, with release 0.0.12 implementation of the explicit EDD the device charge can be a function of either voltage or current29 . The current in the resistive branch being a function of EDD voltage allows a range of two terminal30 devices to be modelled, allowing, for example, nonlinear resistors and diode models to be easily developed. Similarly, the fact that the EDD charge can be a function of voltage or current extends the range of allowed Qucs capacitor types opening new areas of application. The same comments apply to the nonlinear inductors where components that have inductance values which are functions of current allow modelling of nonlinear transformer and coupled inductor eects. This was not possible with earlier Qucs releases. The EDD current and charge values may be dened by symbolic equations that include the operators and functions listed in the Short description of mathematical functions entry in the Qucs help index31 .

29 This allows modelling of semiconductor capacitive eects where the amount of stored charge is either a function of voltage (depletion layer capacitance), or a function of current (diusion capacitance). 30 The number of device terminals can be increased to model transistors and other devices. 31 The Qucs operators and functions are a superset of those dened in the Verilog-A language manual. However, in some cases the name of the operator or function diers slightly. For example Verilog-A uses pow(x, y) for the power function whilst Qucs uses to denote xy . An example of diering function names are the inverse trigonometric functions. A list of the available functions is given in Appendix A.

26

I IC Q C1 C=f(V,I)

I1 R1 R=f(V)

IL L1 L=f(I)

V1

I I1+IC IL i2 D1 I1=I1 Q1=C(V,I)*V1

1

V1

D2 I1=0 Q1=L(I)*V2
1

V2

X1 R=1 Equation defined device (EDD) Equation defined device (EDD)

Gyrator

Figure 17: A non-linear two terminal branch with parallel resistive, capacitive and inductive components

27

Modelling nonlinear resistors

In many measurement applications a transducer is employed to transform changing values of a physical quantity to, say, changes in resistance. Often the resistive characterstics of these devices are nonlinear. To demonstrate how the EDD can be used to model a nonlinear resistance the example shown in Fig. 18 is introduced. In this schematic an EDD represents a resistance that is a function of the applied voltage across its terminals. This example deliberately shows an extreme case where the resistance changes in a resistive pulse like fashion as the terminal voltage increases. The example also introduces for the rst time the ternary ? : operator and illustrates how it can be nested to give an if then else structure to dene the component properties. A point of note with these very nonlinear devices centres around the fact that it is possible to dene components that have discontinuities in their I-V characteristics32 . The EDD current equation denes how the resistance of this device changes with changing terminal voltage. This equation is given by I1=V1/((V1<1.0) ? ? ? ? 1000 : (V1<2.0) 1000+4000*(V1-1) : (V1<5.0) 5000 : ((V1 >=5.0) && (V1<6.0)) 5000-4500*(V1-5.0) : 500)

Which in terms of an if then else type statement is equivalent to: I1 = V1/( if (V1 < 1.0) then 1000 else if (V1 < 2.0) then 1000 + 4000*(V1-1) else if (V1 < 5.0) then 5000 else if ((V1 >= 5.0) && (V1 < 6.0)) then 5000 - 4500*(V1-5.0) else 500 )

One eect of such a discontinuity is the introduction of rapidly changing circuit conditions which can cause the simulator diculties in converging to a correct solution. Sometimes, if this happens, simulation run times may be dramatically increased or simulation fails altogether.

32

28

Parameter sweep
Equation Eqn1 R=Vs/Pr1.I

dc simulation

DC1

Vs

SW1 Sim=DC1 Type=lin Param=Vs Start=0 Stop=7 Points=100

V1 U=Vs

Pr1

D1 I1=V1/((V1<1.0) ? 1000 : (V1<2.0) ? 1000+4000*(V1-1) : (V1<5.0) ? 5000 : ((V1 >=5.0) && (V1<6.0)) ? 5000-4500*(V1-5.0) : 500)
1

Pr1.I

Figure 18: Qucs nonlinear resistor model

1e-4

R 2e3 1e3 2 Vs 3 4 5 6 7 0 0 1 2 3 Vs 4 5 6 7

29
5e3 4e3 3e3

0.1

0.01

1e-3

1e-5

1e-6

Modelling nonlinear capacitors and inductors

Nonlinear capacitors, whos C value is a function of terminal voltage, and nonlinear inductors, whos L value is a function of terminal current, commonly act as control elements in electronic systems. SPICE 2g6 includes a nonlinear symbolic polynomial form of C and L33 . The schematic shown in Fig. 19 illustrates how a nonlinear capacitor can be modelled by an EDD. This model is based on a SPICE like polynomial function with four coecients; C0, C1, C2 and C334 . The test circuit is a simple RC network with nominally identical R and C component values to those shown in Fig. 2. Increasing the value of DC source V1 also increases C which in turn decreases the RC low pass lter -3dB frequency. This eect is very visible in Fig. 19. The nonlinear changes in C are also clearly illustrated in the output voltage and phase curves. The schematic symbol for the nonlinear capacitor is shown in Fig. 19 with a red ring drawn around the normal capacitor symbol. This denotes an EDD based component. An alternative convention is to use red lettering within a symbol. The test circuit and simulation results for a nonlinear inductance are shown in Fig. 20. The EDD model is similar to the SPICE 2g6 nonlinear inductance model with four coecients. This number can be increased, if required, by extending the EDD polynomial expression. A gyrator is employed with the EDD to model the nonlinear inductance. The eect of nonlinear inductance on the inductance current is shown by the dierence between probe currents Pr1 and Pr2.

33 The details of these polynomial functions are presented in Test Reports 4 and 5 of the SPICE to Qucs testing Series, Mike Brinson, http://qucs.sourceforge.net/docs.html. 34 SPICE 2g6 allows up to twenty coecients. Simply add more higher order terms to the Qucs polynomial if required.

30

V2 U=1 V V1 U=Vb

Vout

R1 R=1k

dc simulation
DC1

SUB1 C0=1u C1=0.5u C2=0.2u C3=0.1u

Vb 1 2 3 4 5 6 7 8 9 10

Vout.V 1 2 3 4 5 6 7 8 9 10

ac simulation
AC1 Type=log Start=1 Hz Stop=10kHz Points=201

Parameter sweep
SW1 Sim=AC1 Type=lin Param=Vb Start=1 Stop=10 Points=10

Equation Eqn1 Ph_Vout=phase(Vout.v) Vout_dB=dB(Vout.v)

D1 I1=0 Q1=C0*V1+(C1/2)*V1^2+(C2/3)*V1^3+(C3/4)*V1^4 POUT1

1

PIN1

Vout.v

0.5

0 1 10 100 acfrequency 1e3 1e4

0 -20 Vout_dB -40 -60 -80 1 10 100 acfrequency 1e3 1e4

200

Ph_Vout

150

100 1 10 100 acfrequency 1e3 1e4

Figure 19: Qucs nonlinear capacitor model 31

P_inp1
1

D1 I1=0 Q1=L*V1+(L2/2)*V1^2+(L3/3)*V1^3+(L4/4)*V1^4 X1 R=1 in

P_inn1

V1 U=vin f=1 MHz

IND=L+L2*I(L)+L3*I(L)^2+L4*I(L)^3

Pr1

transient simulation
TR1 Type=lin Start=0 Stop=4 us

SUB1 L=1e-6 L2=5e-7 L3=1e-7 L4=5e-8

Parameter sweep
SW1 Sim=TR1 Type=lin Param=vin Start=0 Stop=100 Points=3

dc simulation
DC1

Pr2

L1 L=1e-6

100

in.Vt

-100 0 5e-7 1e-6 1.5e-6 2e-6 time 2.5e-6 3e-6 3.5e-6 4e-6

Pr1.It

0 0 40 5e-7 1e-6 1.5e-6 2e-6 time 2.5e-6 3e-6 3.5e-6 4e-6

Pr2.It

20

0 0 5e-7 1e-6 1.5e-6 2e-6 time 2.5e-6 3e-6 3.5e-6 4e-6

Figure 20: Qucs nonlinear inductor model 32

Compact device modelling using EDD

Semiconductor device models are a corner stone of all circuit simulators. Often they are characterised by the same parameters as those found in the SPICE 2g6 and 3f5 diode, BJT, FET and MOS models.35 . Since the original SPICE semiconductor device models where rst developed many new extensions to these models have been proposed. Unfortunately, adding such models to a circuit simulator is a complex process, being both time consuming and requiring specialised knowledge. For the average Qucs user the hand coded C++ model generation route is one that they would not contemplate attempting because of the depth of knowledge and specialised skills required. The Qucs EDD was devised to promote fast, and straight forward, prototyping of semiconductor compact models, allowing a wider Qucs population the opportunity to try their hand at device model construction. To demonstrate the stages needed to generate an EDD model of a semiconductor device a compact model of a diode is introduced in this section36 . The DC diode current Id is given by the following functions of diode voltage Vd 37 . Id = Is (exp (Vd /(n V t) 1) + Vd GM IN, Id = Is + Vd GM IN, (5 n V t Vd ) (18) (19) (20) (Vd < BV ). (21)

(BV < Vd ) and (Vd < 5 n V t Vd ) (Vd = BV )

Id = IBV,

Id = Is (exp ((BV + Vd )/V t) 1 + BV /V t) , In these equations: Is = the saturation current. n = the emission coecient.

GM IN = a small conductance in parallel with the diode38 V t = kB T /q, where T is the diode temperature in Kelvin, kB is Boltzmanns constant and q the charge on the electron.
The SPICE 2g6 and 3f5 device parameters are a subset of those commonly provided with current generation of circuit simulators, including Qucs. 36 A second three terminal MESFET transistor example is available for downloading from the Qucs Web site. 37 These equations are for the SPICE 2g6 diode model, see Giuseppe Massobrio, Chapter 1, Pn-junction diode and Schottky diode, Semiconductor device modeling with SPICE, Edited by Paolo Antognetti, Giuseppe Massobrio, 1988, McGraw-Hill,Inc, ISBN 0-07-002107-4. 38 GMIN is added to help Qucs DC convergence. The SPICE default value is 1e-12S.
35

33

 BV = reverse breakdown voltage (positive number) IBV = reverse breakdown current (positive number). Figure 21 gives the EDD model for the experimental semiconductor diode. The ternary operator ?: is used to select the correct equation for each diode operating region. The diode current Id : content.tex, v1.22007/06/0316 : 58 : 59elaExp is the sum of EDD branch currents I1 to I4, where I1 represents the diode forward bias region, I2 the reverse bias region and I3 plus I4 the diode reverse bias breakdown region. When calculating diode current a special form of the exponential function exp(), called limexp(), is employed to assist Qucs to converge to a solution during DC and transient large signal analysis. The function limexp() linearises the exponential function at large argument values minimising the possibility of oating point overow and generation of software exceptions. The Id Vd characteristic curves shown in Fig. 21 are for the forward bias region with series resistance rs set to 0.01. For completeness the simulation data for the Qucs built-in diode are also given. Clearly the two sets of results are very similar. The DC simulation results for the diode reverse breakdown region of operation are shown in Fig. 22. Again for comparison an Id Vd plot for the Qucs built-in diode is also provided. In this region of operation some slight dierences are apparent: although for both devices the reverse breakdown is very close to 100V the slope of the Id V d curve at negative voltages beyond -BV is dierent, emphasising that the SPICE diode model does not model breakdown or zener eects well39 . The next stage in the development of the diode model is to add capacitance eects: depletion layer capacitance for the reverse bias region and diusion capacitance for the forward bias region. Diode capacitance is given by: Depletion layer capacitance Cdep Diusion capacitance Cdif f = dQdif f dId = tt dVd dVd (23) Vd dQdep = = Area Cj0 1 dVd Vj
m

(22)

Where the total stored charge Qd = Qdep + Qdif f . Using the same notation as the SPICE diode model: Qdif f = tt Id
Vd

(24) (Vd <= F C Vj ) (25)

Qdep = Area Cj0

0
39

Vd Vj

dV,

See Steven M. Sandler, SPICE subcircuit accurately models zener characteristics, Personal Engineering, November 1998, pp 45-48 for more information on this subject.

34

PANODE1

RS1 R=rs

Equation Eqn2 GMIN=1e-12 Vt=vt(300)

D2 I1=(V1>-5.0*n*Vt) ? Is*(limexp(V1/(n*Vt))-1.0)+V1*GMIN : 0 Q1=0 I2=(-BV<V1) ? (V1<-5.0*n*Vt) ? -Is+V1*GMIN : 0 : 0 Q2=0 I3=(V1==-BV) ? -IBV : 0 Q3=0 I4=(V1<-BV) ? -Is*(limexp(-(BV+V1)/Vt)-1.0+BV/Vt) : 0 Q4=0

Vs U=Vd

dc simulation
DC1

PCATHODE1

Pr2 Pr1 SUB1 n=1.0 rs=0.01 Is=1e-14 BV=100.0 IBV=1e-3 Vj=1.0 D1 Is=1e-14 A N=1 Vj=1.0 Rs=0.01 Bv=100.00 Ibv=1e-3

Parameter sweep
SW1 Sim=DC1 Type=lin Param=Vd Start=0 Stop=1 Points=190

Equation Eqn1 Id=Pr1.I Id_Q=Pr2.I lnId=ln(Pr1.I) lnId_Q=ln(Pr2.I)

10

Id (A)

ln(Id) 0 0.2 0.4 0.6 Vd (V) 0.8 1

-20

0 -40 0 0.2 0.4 0.6 Vd (V) 0.8 1

10

ID_Q (A)

ln(ID_Q) 0 0.2 0.4 0.6 Vd (V) 0.8 1

-20

0 -40 0 0.2 0.4 0.6 Vd (V) 0.8 1

Figure 21: Compact diode model DC test circuit and simulation results: SUB1 is the EDD diode model and D1 the Qucs diode model with the same parameters as SUB1. 35

-5

ID_Q (A) -100.5 Vd (V) -100 -99.5

Id (A)

-50

-10 -101 -101 -100.5 Vd (V) -100 -99.5

Figure 22: Compact diode model DC simulation results for the reverse breakdown region of operation Using integration formula (ax + b)n dx = Qdep = 1 (ax + b)1+n and simplifying yields: a 1+n
1m

Vd Area Cj0 Vj 1 1 1m Vj

(26)

Also, in the forward bias region Area Cj0 = Area Cj0 F 1 + F2

Vd

Qdep

F3 +
F C Vj

m Vd Vj

dV,

(Vd >= F C Vj ) (27)

On integrating 1 F2 m 2 Vj

Qdep = Area Cj0 F 1 + Where

F 3 (Vd F C Vj ) +

Vd2 + (F C Vj )2 (28)

F1 =

Vj 1 (1 F C)1m , F 2 = (1 F C)1+m , F 3 = 1 F C (1 + m) 1m

(29)

In these equations: F C = Coecient for forward-bias depletion capacitance. m = Grading coecient. tt = Transit time. Area = Device area. Cj0 = Zero-bias junction capacitance.

36

Figure 23 shows the extended diode model. The Cdep and Cdif f components of the device capacitance have been included in the EDD model as stored charge Q1 and Q2. Again the ternary operator ?: is employed to select the correct equation for each section of the diode DC operating range. An equation block is used to simplify the charge equations through the use of factors F1, F2 and F3.40 . An area factor has also been added to the EDD model in Fig. 23. This is introduced to allow simulation of two or more equivalent parallel devices. The diode variables scaled by area are: Is (A) = Is Area, Cj0(A) = Cj0 Area, and rs (A) = rs/Area. (30)

The test circuit shown in Fig. 23 illustrates how device capacitance and resistance can be determined as a function of diode bias voltage. Firstly, the diode S parameters are determined at a given bias voltage, secondly these are converted to Y parameters and the diode capacitance (Cap) and resistance (RD) extracted from Y [1, 1], and nally the variation of Cap and RD with diode voltage Vd plotted using the Qucs plotting function PlotVs. Notice that the value of Cap at Vd = 0V agrees with the value of Cj0. To complete the demonstration EDD diode model all that remains to do is to add temperature dependence to the current and capacitance equations. Circuit simulators normally use two temperatures to determine device temperature dependence; the rst called Tnom represents the temperature that the device parameters were measured, and the second called Temp represents the current device temperature. A high percentage of the diode parameters are temperature dependent. However, to simplify the demonstration diode model only the temperature dependence of parameters Is , V j and Cj0 will be included in the model. Adding extra temperature dependence to the diode model is left to readers as an exercise41 . One of the great advantages of the EDD style of modelling is that it is interactive allowing easy experimentation with models to any given level. The following equations list the temperature dependence of Is , V j and Cj0. Let T1 = Tnom and T2 = Temp, then Is (T 2) = Is (T 1) T2 T1
XT I n

exp T2 T1

q Eg(300) kB T 2
1.5

T2 T1

(31)

V j(T 2) =

T2 2 kB T 2 V j(T 1) ln T1 q

T2 Eg(T 1) Eg(T 2) T1 V j(T 2) V j(T 1) V j(T 1)

(32) (33)

Cj0(T 2) = Cj0(T 1) 1 + m 400 106 (T 2 T 1) In these equations:

In complex current and charge expressions precalculating subexpressions in equation blocks ensures that they are only calculated once at the beginning of a simulation, ensuring minimum run times for an EDD model. 41 For example, parameters m and BV are both temperature dependent.

40

37

PANODE1 RS1 R=rs

Equation

PCATHODE1

Eqn2 GMIN=1e-12 F1=(Vj/(1-m))*(1-(1-FC)^(1-m)) F2=(1-FC)^(1+m) F3=1-FC*(1+m) Vt=vt(300) X1 Vs1 U=Vs P1 Num=1 Z=50 Ohm

D1 I1=(V1>-5.0*n*Vt) ? Is*(limexp(V1/(n*Vt))-1.0)+V1*GMIN : 0 Q1=(V1 < FC*Vj) ? tt*I1+Area*(Cj0*Vj/(1-m))*(1-(1-V1/Vj)^(1-m)) : 0 I2=(-BV<V1) ? (V1<-5.0*n*Vt) ? -Is+V1*GMIN : 0 : 0 Q2=(V1 >= FC*Vj) ? tt*I1+Area*Cj0*(F1+(1/F2)*(F3*(V1-FC*Vj)+(m/(2*Vj))*(V1*V1-FC*FC*Vj*Vj))) : 0 I3=(V1==-BV) ? -IBV : 0 Q3=0 I4=(V1<-BV) ? -Is*(limexp(-(BV+V1)/Vt)-1.0+BV/Vt) : 0 Q4=0

Cap (F)

1e12 Rd ( ) 5e11 0 -4 -3 -2

LN_RD

dc simulation
DC1

S parameter simulation
SP1 Type=const Values=[100 kHz] Equation

Eqn1 Y=stoy(S) LN_RD=ln(RD) RD=PlotVs(1/(real(Y[1,1])),Vs) Cap=PlotVs(imag(Y[1,1])/Omega,Vs) Omega=2*pi*frequency

Figure 23: Compact diode model capacitance and resistance simulation

Vd (V)

Vd

Pr1 SUB1 n=1.0 rs=0.01 Is=1e-14 BV=100.0 IBV=1e-3 Vj=1.0 Cj0=1e-12 FC=0.5 tt=1e-12 Area=1 m=0.5

Parameter sweep
SW1 Sim=SP1 Type=lin Param=Vs Start=-4 Stop=0.8 Points=200

1e-11

5e-12

0 -4 -3 -2 Vd (V) -1 0 1

20

0 -1 0 1 -4 -3 -2 -1 VD (V) 0 1

38

 XT I = Saturation current temperature exponent. Eg(T ) = EG(0) 7.02e 4 T 2 , the energy gap. 1108 + T

Figure 24 shows the extended EDD for the experimental diode model. Again the limexp() function is used in preference to the standard exp() function in the temperature calculations listed in equations block Eqn2. The test circuit in Fig. 24 sweeps the device temperature from 20 to 80 degrees Centigrade. The graph inlay illustrates the experimental diode current Id plotted as a function of temperature. The temperature of the built-in Qucs diode is held constant, at room temperature, and its current Id Q plotted as an overlay. The two curves cross at room temperature, indicating identical currents at this temperature.

39

Pr2

dc simulation
DC1

Vs U=0.6

Pr1 SUB1 n=1.0 rs=0.01 Is=1e-14 BV=100.0 IBV=1e-3 Vj=1.0 Cj0=1e-12 m=0.5 Area=1 FC=0.5 tt=1e-12 XTI=3.0 Tnom=26.85 Temp=Temp_sw Eg=1.16

Equation Eqn1 Id=Pr1.I Id_Q=Pr2.I lnId=ln(Pr1.I) lnId_Q=ln(Pr2.I)

D1 Is=1e-14 A N=1 Cj0=1e-12 Vj=1.0 Rs=0.01 Bv=100.00 Ibv=1e-3 Temp=26.85 Xti=3.0 Eg=1.11 Tnom=26.85 Area=1

Parameter sweep
SW2 Sim=DC1 Type=lin Param=Temp_sw Start=-20 Stop=80 Points=100

0.1 0.01 1e-3 Id_Q Id 1e-4 1e-5 1e-6 1e-7

PANODE1 RS1 R=rs_AREA

1e-8 -20

20 40 Temp (Centigrade)

60

80

D2 I1=(V1>-5.0*n*Vt) ? Area*Is_T2*(limexp(V1/(n*Vt))-1.0)+V1*GMIN : 0 Q1=(V1 < FC*Vj) ? tt*I1+Area*(Cj0_T2*Vj_T2/(1-m))*(1-(1-V1/Vj_T2)^(1-m)) : 0 I2=(-BV<V1) ? (V1<-5.0*n*Vt) ? -Area*Is_T2+V1*GMIN : 0 : 0 Q2=(V1 >= FC*Vj) ? tt*I1+Area*Cj0_T2*(F1+(1/F2)*(F3*(V1-FC*Vj_T2)+(m/(2*Vj_T2))*(V1*V1-FC*FC*Vj_T2*Vj_T2))) : 0 I3=(V1==-BV) ? -IBV : 0 Q3=0 I4=(V1<-BV) ? -Area*Is_T2*(limexp(-(BV+V1)/Vt)-1.0+BV/Vt) : 0 Q4=0

Eqn3 F1=(Vj/(1-m))*(1-(1-FC)^(1-m)) F2=(1-FC)^(1+m) F3=1-FC*(1+m) Vt=vt(300)

PCATHODE1 Equation Equation Eqn2 Cj0_T2=Cj0*(1+m*(400e-6*(T2-T1)-(Vj_T2-Vj)/Vj)) rs_AREA=rs/AREA GMIN=1e-12 A=7.02e-4 B=1108 T1=Tnom+273.15 Vj_T2=(T2/T1)*Vj-(2*kB*T2/q)*ln((T2/T1)^1.5)-((T2/T1)*Eg_T1-Eg_T2) Is_T2=Is*(T2/T1)^(XTI/n)*limexp((-(q*Eg)/(kB*T2))*(1-T2/T1)) Eg_T1=Eg-A*T1*T1/(B+T1) Eg_T2=Eg-A*T2*T2/(B+T2) T2=Temp+273.15

Figure 24: Compact diode model with temperature dependence

40

Constructing EDD compact device models and circuit macromodels

Component equations, subcircuits with parameters and EDD models are major developments for the Qucs circuit simulator. They provide advanced modelling capabilities with enough power and exibility to allow a much greater range of models to be developed than the ones currently provided with each Qucs release. In the future it is proposed to add new models to the Qucs Web site. The Qucs team is very keen to encourage all Qucs users to support the modelling eort. If you have constructed a new model and would like to share it with other Qucs users please post your model on the qucs-devel or qucs-help mailing lists. Both the model schematic le and a brief outline of its operation and specication are requested. An example model specication for the Curtice MESFET device can be found on the Qucs Web site. Please use the same format when writing model descriptions.

End Note
This tutorial note introduces a large number of new modelling concepts and shows how equations, subcircuits with parameters and the new equation dened device perform a central role in constructing Qucs models. The EDD approach to modelling makes possible, for the rst time, the construction of equation dened compact device models and circuit macromodels using the Qucs schematic capture facilities as an interactive modelling medium. This is a major step forward for Qucs. Once again these notes are very much a record of work in progress: much still remains to be done in the future to improve the modelling capabilities provided by Qucs. A major short term task will be the development of additional models covering as wide a range of applications as possible. If Qucs is to fulll its mission to become a truly universal circuit simulator then it must be supported by models. Some readers will have noticed that these notes include very little information about the ADMS-Verlog-A and hand coded C++ model development routes. This was a deliberate decision on my part. Sometime in the future I intend to return to these subjects and update the tutorial. A very special thank you must go to Stefan Jahn for all his hard work, skill, and dedication during the period he has worked on programming the amazing modelling capabilities now embedded in Qucs.

41

Appendix A: Qucs constants, operators and functions

This appendix lists the constants, operators and a number of functions that are available for constructing Qucs equations. Items in [...] indicate the equivalent object in the Verilog-A language. The functions listed are common to Qucs and Verilog-A. A number of other functions have been implemented in Qucs. The full list can be found in the Qucs help system; Short Description of mathematical Functions or in the Qucs Measurement Expression Reference Manual by Gunther Kraut and Stefan Jahn, http: //qucs.sourceforge.net/docs.html. Constants 1. pi = 3.141593... 2. e = 2.718282... 3. kB = 1.380651e-23 J/K 4. -q = -1.602177e-19 C Operators 1. +x unary plus 2. -x unary minus 3. x+y addition 4. x-y subtraction 5. x*y multiplication 6. x/y division 7. x%y modulo (remainder) 8. x^y power [pow(x,y)] 9. ?: ternary (condition) ? (expression if true) : (expression if false) 10. || logical or 11. && logical and 12. == equal 13. < less than 14. <= less than or equal to 15. > greater than 16. >= greater than or equal to 17. != not equal to 18. ( ) brackets 42

 Functions 1. ln(x) natural logarithm 2. log10(x) decimal logarithm 4. sqrt(x) square root 5. min(x,y) minimum 6. max(x,y) maximum 7. abs(x) absolute value 8. sin(x) sine 9. cos(x) cosine 10. tan(x) tangent 11. arcsin(x) inverse sine 12. arccos(x) inverse cosine 14. sinh(x) hyperbolic sine 15. cosh(x) hyperbolic cosine 16. tanh(x) hyperbolic tangent 17. arsinh(x) inverse hyperbolic sine 18. arcosh(x) inverse hyperbolic cosine 19. artanh(x0 inverse hyperbolic tangent 21. hypot(x,y) Euclidean distance function [asinh(x)] [acosh(x)] [atanh(x)] [asin(x)] [acos(x)] [atan2(x,y)] [log(x)] 3. exp(x) exponential function base e

13. arctan(x[,y]) inverse tangent

20. limexp(x) argument limited exponential function

43

Appendix B: Constructing subcircuits with parameters

In this appendix a series of screen dumps illustrate the sequence needed to construct a subcircuit with parameters. A simple series resonance circuit has been chosen for the demonstration.

Enter the series resonance circuit and add input and output pins

Figure 25: Stage 1: screen dump showing LCR circuit

44

Change the component names to Ls, Cs and Rs

Figure 26: Stage 2: screen dump showing LRC circuit

Figure 27: Stage 2: screen dump after name changes

45

Construct symbol for new subcircuit

Right click on the Qucs drawing area and select Edit Circuit symbol or press key F9. Edit the drawing symbol to give the design shown in Fig. 28.

Figure 28: Stage 3: the subcircuit symbol

46

Add the names of the subcircuit parameters to the LCR symbol

Right click on the SUB / File=name caption and enter names of subcircuit parameters with their default values.

Figure 29: Stage 4: entering subcircuit parameter names and default values

Figure 30: Stage 4: resulting subcircuit and parameter list with default values

47

Test the LCR subcircuit

Figure 31 gives a simple AC transfer function test circuit and resulting waveforms. Parameter R SW is swept over the range 1 to 10 and the AC transfer function recorded and plotted.
Vin I V1 U=1 V SUB1 Rs=R_SW Cs=1u Ls=1m

LCR
O

Vout

dc simulation
DC1

ac simulation
AC1 Type=log Start=100 Hz Stop=100kHz Points=100

Parameter sweep
SW1 Sim=AC1 Type=lin Param=R_SW Start=1 Stop=10 Points=10

Equation Eqn1 gain=dB(Vout.v) phase=phase(Vout.v)

Vout.v

0.5

0 100 1e3 acfrequency 0 1e4 1e5

100

phase -50 100 1e3 1e4 acfrequency 1e5

gain

-100 100 1e3 1e4 acfrequency 1e5

Figure 31: Stage 5: Subcircuit test circuit and output waveforms

48