CSE 245: Computer Aided Circuit Simulation and Verification Fall 2004, Sep 28 Lecture 2: Dynamic Linear System
CSE 245: Computer Aided Circuit Simulation and Verification
Fall 2004, Sep 28
Lecture 2:
Dynamic Linear System
Lecture2.2
OutlineTime Domain Analysis
State Equations RLC Network Analysis by Taylor ExpansionImpulse Response in time domain
Frequency Domain AnalysisFrom time domain to Frequency domainCorrespondence between time domain and frequency domainSerial expansion of (sI-A)-1
Lecture2.3
Outline (Cont’)Model Order Reduction
Moments Passivity, Stability and Realizability
Symbolic Analysis Y-Delta TransformationBDD Analysis
Lecture2.4
State of a systemThe state of a system is a set of data, the value of which at any time t, together with the input to the system at time t, determine uniquely the value of any network variable at time t. We can express the state in vector form
x =
Where xi(t) is the state variables of the system
)(...
)()(
2
1
tx
txtx
k
Lecture2.5
State VariableHow to Choose State Variable?
The knowledge of the instantaneous values of all branch currents and voltages determines this instantaneous stateBut NOT ALL these values are required in order to determine the instantaneous state, some can be derived from others.choose capacitor voltages and inductor currents as the state variables! But not all of them are chosen
Lecture2.6
Degenerate NetworkA network that has a cut-set composed only of inductors and/or current sources or a loop that contains only of capacitors and/or voltage sources is called a degenerate networkExample: The following network is a degenerate network since C1, C2 and C5 form a degenerate capacitor loop
Lecture2.7
Degenerate NetworkIn a degenerated network, not all the capacitors and inductors can be chosen as state variables since there are some redundancy
On the other hand, we choose all the capacitor voltages and inductors currents as state variable in a nondegenerate network
We will give an example of how to choose state variable in the following section
Lecture2.8
Order of Circuitn = bLC – nC - nL
n the order of circuit, total number of independent state variablesbLC total number of capacitors and inductors in the networknC number of degenerate loops (C-E loops)nL number of degenerate cut-sets (L-J cut-sets)n = 4 – 1 = 3
In a nondegenerate network, n equals to the total number of energy storage elements
Lecture2.9
State Equations
StateInput
Output
)()()(
)()(
tDutQxty
tButAxdtdx
+=
+=
Linear system of ordinary differential equations
Lecture2.10
State Equation for RLC Circuits
The state equation is of the form
Or
vt: voltage in the trunk, capacitor voltageil: current in the loop, inductor current.Y and R are the admittance matrix and impedance matrix of cut-set and meshE covers the co-tree branches in the cut-set–ET covers the tree trunks in the mesh analysis
L
C0
0
l
t
iv&&
− RE
EYT
l
t
iv
= - + Pu
= Gx(t) + Pu(t))(tx&M
Lecture2.11
State Equations
If we shift the matrix M to the right hand side, we have
Let A = M-1G and B = M-1P, we have the state equation
Together with the output equation
are called the State Equations of the linear system
= Gx(t) + Pu(t))(tx&M
)(tx& = M-1Ax(t) + M-1Bu(t)
= Ax(t) + Bu(t))(tx&
)(ty = Qx(t) + Du(t)
Lecture2.12
RLC Network AnalysisA given RLC network
Degenerate Network, Choose only voltages of C1 and C5, current of L6 as our state variable
Vs
g3
g4C1
C2
C5
L6
1 2
0
Lecture2.13
Tree StructureTake into tree as many capacitors as possible and,as less inductors as possibleResistors can be chosen as either tree branches or co-tree branches
Vs
g3
g4C1
C2
C5
L6
1 2
0
g3C1 C5 g4
1 2C2/L6
0
Vs
Lecture2.14
Linear State EquationBy a mixed cut-set and mesh analysis, consider capacitor cut-sets and inductor loops only. we can write the linear state equation as follows
M = Gx(t) + Pu(t))(tx&
Cut-set KCL
Loop KVLCut-set KCL
+−
−+
6
522
221
0000
LCCC
CCC
6
2
1
ivv
&
&
&
−−01110
10
4
3
gg
6
2
1
ivv
003g
=- + Vs
Lecture2.15
OutlineTime Domain Analysis
State Equations RLC Network Analysis by Taylor ExpansionImpulse Response in time domain
Frequency Domain AnalysisFrom time domain to Frequency domainCorrespondence between time domain and frequency domainSerial expansion of (sI-A)-1
Lecture2.16
Solving RCL Equation by Taylor Expansion(1)General Circuit Equation
Consider homogeneous form first
BUAXX +=•
AXX =•
0XeX At=
...!
...!2!1
22
+++++=ktAtAAtIe
kkAt
Q: How to Compute Ak ?
and
Lecture2.17
Assume A has non-degenerate eigenvalues and corresponding linearly
independent eigenvectors , then A can be decomposed as
where and
Solving RCL Equation by Taylor Expansion (2)
1−ΧΛΧ=A
kΧΧΧ ,...,, 21
kλλλ ,...,, 21
=Λ
kλ
λλ
LL
MOLM
ML
L
0
000
2
1
[ ]kΧΧΧ=Χ ,...,, 21
Lecture2.18
What’s the implication then?
To compute the eigenvalues:
1−ΧΛΧ=A
01
1 ...)det( ccAI nn
n +++=− −− λλλ
0)...(...))(( 212
0 =++−= ppp λλλ
realeigenvalue
ConjugativeComplexeigenvalue
122 −ΧΧΛ=A
1−Λ ΧΧ= tAt ee where
=Λ
ke
ee
e t
λ
λ
λ
LL
MOLM
ML
L
0
000
2
1
Solving RCL Equation by Taylor Expansion (3)
Lecture2.19
In the previous example
==
−−
)0()0(
1
2//1/10
01
2
iv
eXeiv t
lrlc
At
11
00
1110 −
−
+− Χ
=ΛΧ=
−−
=λ
λXXA
231
231
j
j
−−=
+−=
−
+
λ
λwhere
−
−−−
=Χ1
231
12
31
3 j
jj
−−+−=Χ−
231
231
111 jj
hence 1
00 −Χ
Χ=
−
+
λ
λ
ee
eAt
Let c=r=l=1, we have
Solving RCL Equation by Taylor Expansion (4)
Lecture2.20
What if matrix A has degenerated eigenvalues? Jordan decomposition !
1−ΧΧ= JAJ is in the Jordan Canonical form
And still 1−ΧΧ= JtAt ee
Solving RCL Equation by Taylor Expansion (5)
Lecture2.21
Jordan Decomposition
=
λλ0
1J
=+
+
= t
ttJt
etee
teλ
λλ
λλ
001
1001
L
=
λλ
λ
001001
J
=+
+
=
t
tt
ttt
Jt
etee
ettee
teλ
λλ
λλλ
λλ
λ
000
!2
001001
100010001
2
L
similarly
Lecture2.22
OutlineTime Domain Analysis
State Equations RLC Network Analysis by Taylor ExpansionResponse in time domain
Frequency Domain AnalysisFrom time domain to Frequency domainCorrespondence between time domain and frequency domainSerial expansion of (sI-A)-1
Lecture2.23
Response in time domainWe can solve the state equation and get the closed form expression
The output equation can be expressed as
Note: * denotes convolution
Lecture2.24
Impulse ResponseThe Impulse Response of a system is defined as the Zero State Response resulting from an impulse excitation
Thus, in the output equation, replace u(t) by the impulse function δ(t), and let x(t0)=0 we have
h(t) = y(t) = QeAt B
Lecture2.25
OutlineTime Domain Analysis
State Equations RLC Network Analysis by Taylor ExpansionImpulse Response in time domain
Frequency Domain AnalysisFrom time domain to Frequency domainCorrespondence between time domain and frequency domainSerial expansion of (sI-A)-1
Lecture2.26
Solutions in S domainBy solving the state equation in s domain, we have
Suppose the network has zero state and the output vector depends only on the state vector x, that is, x(t0) = 0 and D = 0, we can derive the transfer function of the networkH(s) = = Q(sI-A)-1B
x(s) = (sI-A)-1 x(t0)+ (sI-A)-1 Bu(s)y(s) = Qx(s) +Du(s) = Q(sI-A)-1(x(t0) + Bu(s)) +Du(s)
)()(
ss
uy
Lecture2.27
Key Key Transform Transform Property:Property:
Bilateral Bilateral LaplaceLaplaceTransform:Transform:
)()()()()(
sQxsysBusAxssx
=+=
)()(
)()()(
tQxty
tButAxdt
tdx
=
+=
dtdx )(ssx
dtetxsx st∫∞
∞−
−= )()()(tx
Frequency Domain Representation
Lecture2.28
Express y(s) as a Express y(s) as a function of u(s)function of u(s)
)()()()()(
sQxsysBusAxssx
=+=
Transfer Function:Transfer Function: )(sH
)()()( 1 sBuAsIQsy −−−=
System Transfer Function
Lecture2.29
Transfer FunctionTime Domain Impulse ResponseFrequency domain representationFrequency domain representation
H(s)u(s) y(s) = H(s) u(s)
Linear systemLinear system
h(t)u(t) ∫ −=t
duthty0
)()()( τττ
Linear systemLinear system
Time domain representationTime domain representation
The transfer function H(s) is the The transfer function H(s) is the LaplaceLaplace Transform Transform of the impulse response h(t)of the impulse response h(t)
Lecture2.30
OutlineTime Domain Analysis
State Equations RLC Network Analysis by Taylor ExpansionImpulse Response in time domain
Frequency Domain AnalysisFrom time domain to Frequency domainCorrespondence between time domain and frequency domainSerial expansion of (sI-A)-1
Lecture2.31
Correspondence between time domain and frequency domain
We can derive the time domain solutions of the network from the s domain solutions by inverse Laplace Transformation of the s domain solutions.
State Equations in S domain
State Equations in time Domain
Inverse LaplaceTransform
sx(s) – x(t0)= Ax(s) +Bu(s)
y(s) = Qx(s) +Du(s)
x(t) = L-1[(sI-A)-1x(t0) + (sI-A)-1 Bu(s)]
= L-1[(sI-A)-1]x(t0) + L-1[(sI-A)-1]B*u(t)
y(t) = L-1[Q(sI-A)-1(x(t0) + Bu(s)) +Du(s)]
= Q L-1[(sI-A)-1] x(t0) + {QL-1 [(sI-A)-1]B +Dδ(t)}* u(s)
Lecture2.32
Correspondence between time domain and frequency domain
(sI-A)-1 eAt
multiplication of u(s) in s domain corresponds to the convolution in time domain
Solution from time domain analysis
Solution by inverse Laplacetransform
x(t) = L-1[(sI-A)-1x(t0) + (sI-A)-1 Bu(s)]
= L-1[(sI-A)-1]x(t0) + L-1[(sI-A)-1]B*u(t)
y(t) = L-1[Q(sI-A)-1(x(t0) + Bu(s)) +Du(s)]
= Q L-1[(sI-A)-1] x(t0) + {QL-1 [(sI-A)-1]B +Dδ(t)}* u(s)
Lecture2.33
OutlineTime Domain Analysis
State Equations RLC Network Analysis by Taylor ExpansionImpulse Response in time domain
Frequency Domain AnalysisFrom time domain to Frequency domainCorrespondence between time domain and frequency domainSerial expansion of (sI-A)-1
Lecture2.34
Serial expansion of (sI-A)-1
When s 0 we can write (sI-A)-1 as
Thus, the transfer function can be written as
When s ∞ we can write (sI-A)-1 as
The transfer function can be written as
(sI-A)-1 = -A-1(I – sA-1) = -A-1(I + sA-1 + s2A-2 + … + skA-k + …)
H(s) = Q(sI-A)-1B = -QA-1(I + sA-1 + s2A-2 + … + skA-k + …)B
(sI-A)-1 = s-1(I – s-1A)-1 = s-1(I + s-1A + s-2A2 + … + s-kAk + …)
H(s) = Q(sI-A)-1B = s-1(I + s-1A + s-2A2 + … + s-kAk + …)B
Lecture2.35
Assume A has non-degenerate eigenvalues and corresponding linearly
independent eigenvectors , then A can be decomposed as
where and
Matrix Decomposition
1−ΧΛΧ=A
kΧΧΧ ,...,, 21
kλλλ ,...,, 21
=Λ
kλ
λλ
LL
MOLM
ML
L
0
000
2
1
[ ]kΧΧΧ=Χ ,...,, 21
Lecture2.36
Matrix DecompositionThen we can write (sI-A)-1 in the following form
(sI-A)-1 in s domain corresponds to the exponential function eAt in time domain, we can write eAt as
(sI-A)-1 = (SI – XΛX-1)-1 = X-1(sI – Λ)-1X = X-1
−
−
−
ns
s
s
λ
λ
λ
1.
.2
1
1
1
X
eAt = X-1
t
t
t
ne
ee
λ
λ
λ
..
2
1
X