ECE 552 Numerical Circuit Analysis Chapter Six NONLINEAR DC ANALYSIS OR: Solution of Nonlinear Algebraic Equations Copyright © I. Hajj 2012 All rights reserved
ECE 552Numerical Circuit Analysis
Chapter Six
NONLINEAR DC ANALYSIS
OR: Solution of Nonlinear Algebraic Equations
Copyright © I. Hajj 2012 All rights reserved
Nonlinear Algebraic Equations
• A system of linear equations Ax = b has a unique solution, unless A is singular.
• However, a system of nonlinear equations f(x) =y may have one solution, multiple finite solutions, no solution, or infinite number of solutions.
Example
Example
Example
y = x2 has two solutions for y > 0 one solution for y = 0 no solution for y < 0
Example
y = x3 has a unique solution for every y.
n-dimensional case
• f(x) = y
or,
f1(x1,x2, ...,xn) = y1
f2(x1,x2, ...,xn) = y2
:
fn(x1,x2, ...,xn) = yn
Problem
• Given y ε Rn, find x* ε Rn, if it exists, such that f(x*) = y.
• Some Theorems on the Existence and Uniqueness of Solutions of Nonlinear Resistive Networks.
Definition
• Given a mapping f(.): Rn → Rn.
• f ε C1 means f is continuously differentiable; and
f is C1 diffeomorphism means that the inverse
function f-1 exists, and is also of class C1.
Palais's Theorem
• The necessary and sufficient conditions that the mapping
f(.): Rn → Rn to be a C1 diffeomorphism of Rn onto itself are:
(i) f is of class C1
(iii) lim ||f(x)|| → ∞ as ||x|| → ∞
• For existence and uniqueness of solution, can allow det [J] = 0 at
isolated points as long as it does not change signs and
lim||f(x)||→ ∞ when ||x||→ ∞
Circuit-Theoretic Theorems
• Theorem 2 (Duffin): In a network consisting of independent voltage and current sources, and voltage-controlled two-terminal resistors (i = g(v)), there exists at least one solution provided that each resistor's v-i characteristic function g(v) is continuous in v and satisfies:
g(v) → + ∞ (or - ∞) as v → + ∞ (or - ∞)
Circuit-Theoretic Theorems (cont.)
• Theorem 3 (Duffin): In a network consisting of independent voltage and current sources and voltage-controlled resistors (i = g(v)), there exists at most one solution provided that each resistor's v-i characteristic function g is strictly monotone increasing:
For existence and uniqueness, both Theorems 2 and 3 should be satisfied.
Circuit-Theoretic Theorems (cont.)
• Theorem 3 (Desoer and Katzenelson): A sufficient condition for the existence of a unique solution for a network consisting of time-varying voltage-controlled and current-controlled resistors characterized by continuous (not necessarily strictly) monotone increasing functions, and independent voltage and current sources, is that the resistor network formed by short-circuiting all voltage sources and open-circuiting all current sources has a tree (or forest) such that all tree branches correspond to current-controlled elements and all links correspond to voltage-controlled elements.
Now back to the numerical solution of:
f(x) = y
Given y, find x (assuming it exists)
Fixed-Point Iteration
x = g(x)
•A given problem can be recast into fixed-point problem, where x = g(x) is a suitably chosen function whose solutions are the solution of f(x) = y.
•For example, f(x) = y can be written as x = f (x) - y + x = g(x).
•Givenx = g(x)•Fixed-Point Iteration: xk+1 = g(xk)
Repeat until ||xk+1 - xk|| < ε
Examples
Examples (cont.)
Multiple solutions
Contraction mapping theorem
• Suppose g: D ε Rn → Rn maps a closed set D0 ε D into
itself and ||g(x) - g(y)|| ≤ α ||x-y||, x, y ε D0 for some
α < 1
Then for any x0 ε D0, the sequence xk+1 = g(xk), k = 0, 1,2, ..., converges to a unique x* of g in D0.
• Proof:
||x*-xk|| = ||g(x*) – g(xk -1)||
≤ α ||x*-xk-1||
≤ αk ||x*-x0||
Since α < 1, α k → 0 and ||x* - xk||→ 0 or xk → x*
Parallel Chord Method
Parallel Chord Method
xk+1= xk + A-l(y - f(xk)
Or A(xk+1- xk) = (y - f(xk)
Parallel Chord Method
A remains constant; it is usually chosen to be = where
“Jacobian matrix”
Instead of computing A-1, the following equation is solved:
A(xk+l - xk) = y-f(xk) or A ∆xk = ∆yk
At every iteration, ∆yk changes, while A (and its LU factors) remain unchanged.
Parallel Chord Algorithm
Parallel Chord Example (nonconvergence)
Newton or Newton-Raphson Method
Newton's (or Newton-Raphson) Method
Instead of solving:
It is sometimes more convenient to solve directly for x:
Newton's (or Newton-Raphson) Method
The slope changes with every iteration.
Convergence Properties of Newton's Method
scalar case
xk+1 =
quadratic convergence.
n-dimensional case
• Applying Taylor Series expansion at xk:
• y = f(x*) = f(xk) + Jk(x* - xk) + R(x* - xk)
where x* is the solution
• If the derivative of Jk (i.e., second derivative of f) is bounded, then:
||R(xk - x*)|| ≤ α ||xk - x*||2
• Newton's Method:
xk+1 = xk + [Jk]-1 (y - f(xk))
or, [Jk](xk+1 - xk) = y - f(xk)
n-dimensional case (cont.)
From Taylor Series:
• y - f(xk) = Jk(x* - xk) + R(x* - xk)
• Jk(xk+l - xk) = Jk(x* - xk) + R(x* - xk)
• xk+l - xk = x* - xk + [Jk]-1R(x* - xk)
• xk+l - x* = [Jk]-lR(x* - xk)
• ||xk+1 - x*|| ≤ c ||x* - xk||2
Provided J(x*) is nonsingular
Rate of Convergence
• Define ek = x* - xk
• A method is said to converge with rate r if:
||ek+1|| = c ||ek||r
for some nonzero constant c
• If r = 1, and c<1, the convergence is linear.
• If r > 1, c>0, the convergence rate is superlinear.
• If r = 2, c>0, the convergence rate is quadratic.
Newton's Method
Has quadratic convergence if xk is
"close enough" to the solution
and J(x*) is nonsingular.
Convergence Problems of Newton’s Method
Convergence Problems of Newton’s Method
(6) Homotopy: λf(x) + (1 - λ)(x - x°) = λ y When λ = 0, x = x° (chosen)
Put λ = 0,...,1 and solve starting from the last solution. f(x) = y when λ = 1.
Application to Electronic Circuits: DC Analysis
• Capacitors are open and inductors are short-circuited. Why?
Tableau Equations:
To formulate in Newton’s method, linearize element
characteristic equations at iteration point v1k, i2
k and obtain linearized circuit tableau equations, then use MNA., or any other formulation method.
Linearization
Stamp
Stamp
Stamp
Example
Example (cont.)
MNA Stamp
Other Methods of Linearization
Secant method
i = av + b
Other Methods of Linearization (cont.)
Line through origin
i = av
a = , b = 0
Multiport or Multiterminal y = f(x)
Taylor Series
Two-Port or Three-Terminal Elements
Multiport or Multiterminal: y = f(x) (cont.)
Transistor Models:NPN Bipolar Junction Transistor Model:Ebers-Moll Model:
Reference: VLACH & SINGHAL, Chap. 11
Multiport or Multiterminal y = f(x) (cont.)
In addition, αfIes = αrIcs = Is
Characteristics are of the form:
ic = f1 (vbc,vbe)
ie = f2 (vbc,vbe)
Multiport or Multiterminal y = f(x) (cont.)
PNP Bipolar Junction Transistor Model:
Multiport or Multiterminal y = f(x) (cont.)
V-I Characteristics of MOSFETs - NMOS
saturationlinear
Multiport or Multiterminal y = f(x) (cont.)
Linear Region: VGS ≥ VTn, VDS ≤ (VGS - VTn)
IDS = Kn [(VGS-VTn)VDS-(1/2)V2DS]
Kn = μn Cox , Cox = , μ n electron mobility
(Threshold Voltage)
Saturation Region:
VGS ≥ Vtn , VDS ≥ (VGS-VTn)
Note that in DC, the current in the gate IG = 0.
Cut-off:
PMOS
Modified Newton's Method for DC Analysis of Electronic Circuit with
Exponential Nonlinearities
e.g., diodes, bipolar junction transistors
• Possibility of overflow if v becomes "large."
Voltage-Current Iteration
- Prevents overflow and improves convergence
DC Input-Output Characteristics
Driving-Point Characteristics: Iin vs. Vin Transfer Characteristics: Vout vs. Vin