svd (singular value decomposition) · 2016-08-25 · chosun university 170 nano system control lab....

65
Nano System Control Lab. Chosun University 170 SVD (Singular Value Decomposition) 1 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 nm K 1 2 ;n m real matrix where 0 K H H S.V.D A UΣV U AV Where H -1 H -1 matrix V U: n n unitary matrix U =U V: m m unitary =V max SV min SV

Upload: others

Post on 10-Mar-2020

2 views

Category:

Documents


0 download

TRANSCRIPT

Nano System Control Lab.Chosun University 170

SVD (Singular Value Decomposition)

1

2

0 0 0 00 0 0 00 0 0 00 0 0 0

0 0 0 0 0

n m K

1 2

; n m real matrixwhere 0K

H

H

S.V.D A UΣVU AV

Where

H -1

H -1matrix V

U : n n unitary matrix U =U

V : m m unitary =V

max SV

min SV

Nano System Control Lab.Chosun University 171

Where ; left singular vector of is right eigenvector of associated with

H 2: AA i i i

H1 2 m i jV v :v : v ; v v ij

H 2: A A i i i

HAA

i Hiλ (A A)

H1 2 i jU u , u u ; u un ij : Kronecker delta

Hλ (AA )i

Aiu

Where right singular vector of is right eigenvector of associated with

iv ; AHA A

i

Column vectors of U(V) are orthonormal

Nano System Control Lab.Chosun University 172

Geometrical Interpretation

(1) is a real matrix, exists n n -1AAn

2

n2

y Ax, x R x x x

y R y y y

T

T

Spectral norm of A

2

22 2X 0 X 1

2

AXA max max AX

X

Singular value relation

2

2 2

max 2 2X 1

min 2 1X 12

(A) max AX A

1(A) min AXA

: spectral norm

Nano System Control Lab.Chosun University 173

Visualization (2D)A

Unit Circle

0

1 1

2 2

Vector OA=v Vector OA =uVector OB=v Vector OB =u

maxH

min

1 2 1 2

0SVD : A U V ,

0

U u u , V v v

AB

B A

1y

2x

1x

2y

B

00

Nano System Control Lab.Chosun University 174

max 1

min 2

Length of vector OA σ σLength of vector OB σ σ

.

.

max SV, min SV, Range of gain

Nano System Control Lab.Chosun University 175

1) Singular values “size” of A

HA A 1, 2, Ki i i

2) SVD “Direction” of A,B

H

H

A U VU AV

max min, for

G(s)u(t) y(t)

MIMO Freq. Response

u(t) uy(t) yy G(jω)u

j t

j t

ee

Singular value

2

2

max max2 2u 1

min min2 2u 1

G(jω) max G jω u y jω

G(jω) min G jω u y jω

Nano System Control Lab.Chosun University 176

max maxy (jω)

min miny (jω) Singular value of y exists in this region.

S.V. Facts

(1) -1 -1max min

min max

1 1(A ) , (A )(A) (A)

(2) max max max(A) 1 (I+A) 1 (A)

(3) max max max(A+B) (A) (B)

(4) max max max(AB) (A) (B)

Nano System Control Lab.Chosun University 177

Feedback Performance Specs. In Frequency Domain

Use of SV to establish MIMO Performance specs.

Proof. Attributes

Review

r(s)

d(s)

u(s) y (s )

n(s)

e(s)K(s) G(s)

-Command Following-Disturbance Rejection-Insensitivity to Sensor noise-Robustness

Sinusoidal

Nano System Control Lab.Chosun University 178

1

1

t

Loop TFM : T(s) G(s)K(s)

Sens TFM : S(s) 1 T(s)

Closed-Loop TFM : C(s) 1 T(s) T(s)

e (s) S(s) r(s) d(s) C(s)n(s) e(s) r(s) y(s)

conflict S(s) C(s) I

1) Command Following : d(s) n(s) 0

tr(t) r e (t) ej t j te e : error vectore

Relation

2 2 2

max 2

S(jω)S(jω) r

S(jω) r

e re

True error

Nano System Control Lab.Chosun University 179

Define;

range of frequency that r has energyr

We have good command following by making max S(jω) 1, r

Interpretation

max2 2,maxr 1 S(jω)e

The worst error at is attained when points along the right singular vector associated with ,

rmax 1V

min2,minS(jω)e

The best error r VK

In General

min max2S(jω) S(jω)e

min

Nano System Control Lab.Chosun University 180

Good Command Following in terms of T(s)=G(s)K(s)

max S(jω) 1 ; r

1

maxmin

1(I+T) 1I+T

min min min1 T I+T 1, T(jω) 1

Overall Loop Gain

To visualize good C.F

max (S)

min (S)

dBLoop TFM

max (T)

min (T)ω

1

r

1

dB

11

Nano System Control Lab.Chosun University 181

2) Disturbance Rejection : r(s)=n(s) 0

Sinusoidal Disturbance

td(t) d e ej t j te e

Relation

max2 2

e S(jω) de S(jω) d

We have good disturbance rejection by

max

min

making S(jω) 1

also, T(jω) 1

; ω d

Define :d Range of frequency that d has energy

Interpretation ; 2d 1

Same on the C.F.

Nano System Control Lab.Chosun University 182

To visualize good C.F. & good D.R. P r d

dBdB

max (T)

min (T)ω

max (S)

min (S)

ω

P

1

:margin of design.

Quantity Relation

Let : 0 1

If max S(jω) , ω P

Then, min11 T(jω) ; ω P

and min max1 C(jω) C(jω) 1

11

C(jω) I “Plant Inversion”

Nano System Control Lab.Chosun University 183

Visualization

dB

P max S( jω)

Sensitivity TFM

dB

P

min T(jω)1

Loop TFM

dB max C(jω)

min C(jω)

1

1 Closed-Loop TFM

Nano System Control Lab.Chosun University 184

Example of the FP (FP : Feedback Performance)

(Temp. regulation in a leaky pipe line)

(1) Problem ; monitor & control temperature.

(2) Model of the system

• Dynamics of medium Flow in the pipeConstitutive relation of the medium

• Dynamics of leak• Dynamic of a heater (time-lag)

ideal gas

Assumptions what? Why? Validity, MIMO, TITO

(3) Solution-Further assumption, linearization

Application of tools & concepts

(4) Evaluation of your solution & problem

heater leak (pressure drop)

FlowPV=nRT

Nano System Control Lab.Chosun University 185

Quantitative Relations

max

min

min max

: 0 1; S(jω) 1 , ω

;1(1) 1 T(jω) ;

(2) 1 C(jω) C(jω) 1

p

p

Let

If

Then

Proof of (1)

1max max

min min

max

min

min min

1 1S(jω) I+T(jω)(I+T(jω)) 1 T(jω)

Given , S(jω)1

1 T(jω)1 1 11 T(jω) , or T(jω) 1 1

Nano System Control Lab.Chosun University 186

Proof of (2)

max

1

-1 -1 -1

-1 -1 -1max max max

min

min

(1 ) C(jω)

C(jω) I+T(jω) T(jω)

C (jω) T (jω) I+T(jω) I+T (jω)1 C (jω) I+T (jω) 1 T (jω)C(jω)

11T(jω)

Since min1T(jω) from

(1)

min

min min

min

1T(jω) 1

1 1 11 1C(jω) T(jω) 1 1

C(jω) 1

Nano System Control Lab.Chosun University 187

max

max max max

max max

max

C(jω) 1C(jω) I S(jω)

C(jω I S(jω) 1 S(jω)

1 S(jω) , since S(jω)

C(jω) 1

Summary for good C.F & D.R

• Large Loop Gain

• Small Sensitivity

• Flat CL response

min T(jω) 1

max S(jω) 1

min maxC(jω) C(jω) I

for

Constraint

Nano System Control Lab.Chosun University 188

3) Insensitivity to S.N. (Sinusoidal Noise) ; r(s) d(s) 0

n( ) n ( ) ej t j ttt e e t e

Relation

max2 2

e C(jω)nC(jω)e n

Define range of frequency of noise with singnificant energyn

We can desensitize the system from S.N. by making for all n

Large separatebetween the two

( , )p n max C(jω) 1 for all n

suppose max C(jω) ; 0 1

① Then, min T(jω) 1 ;(low loopgain)1

② max1 1 S(jω) ; high sensitivity

Nano System Control Lab.Chosun University 189

Proof: ① -1 -1min min

max max

1 1C I+T 1(C) (T)

maxmax

maxmax

1 1Since (C) ,(C)

1 11 (T)(T) 1

Proof ②1

max max minmin min max

maxmax

1 1 1(S) (I+T) 1 (T)(I+T) 1 (T) (S)

11 1 (S)1 (S)

Nano System Control Lab.Chosun University 190

Design Implication

We need wide frequency separation between sets.

andp r d n

P

n

ω

dB

min S(jω)

1

max S(jω)

Nano System Control Lab.Chosun University 191

dB

1

1 P

P max C(jω)

min C(jω)

n

ω

dB

1

P

n

1 min T(jω)

max T(jω)

ω

Nano System Control Lab.Chosun University 192

Direction information in SV plot

SVDA

m m

m m (A)i

-1AComplex matrix , exists

Real diagonal matrix of

1 maxH

min

0 00 0 (A) (A A)0 0

i i

m

H HA U V : Σ=U AV

• right singular vector , orthonormal• left singular vector , orthonormal

iv ;iu ;

miv C

miu C

Nano System Control Lab.Chosun University 193

For

AX Y Y=AX

SVD HY U V X

Suppose ;H

i iX=v y=U V v

Since

H Hi

0 00

0V v , V v1

0 0

0 0

i i

rowthi

Nano System Control Lab.Chosun University 194

1 i

0

y u0

0

ii m iu u u

Directional Information

Max : 1 max , 1 max max 2

1 max max 2

v V , V 1

u U , U 1

If maxX V

max max max 2

max2

Y= U ; U 1

Y (A) max. .Amp

Min : min ,m m min min 2

m min min 2

v V , V 1

u U , U 1

Nano System Control Lab.Chosun University 195

If minX V

min min min2Y σ U Y σ (A) min. .Amp

Input

minV

Unit sphere

min (A)

maxU

minU

maxy

miny

max (A)

“ can be less than 1”max miny , y

.

maxV

output

Nano System Control Lab.Chosun University 196

Analytical Expression

Real Amp.Complex sinusoid

x(t) A j te

Re x(t) A cosωt

Im x(t) Asinωt

Complex Amp.complex sinusoid

x(t) B j te B=B j

je

Re x(t) B cos (ωt )

Im x(t) B sin (ωt )

2 2

1 ( )

B

tan x( ) B j tt e

Complex vectors 1

x( ) x , xj ti

n

x

t e x

x

Nano System Control Lab.Chosun University 197

Summary of SVD

System : Y(s)=G(s)X(s)

Pick ω , calculate G(jω) SVD

• Max - direction

Find max max max, V , U

Set max max

( )max

x ( ) V a , U b

y b

j ji i ii i

j ti i

t e e

e

• Min-minimum direction

Find min min min, V , U0

0

& : 90 difference*

& : 90 difference

min min

( )min

x ( ) V C , U d

y ( ) d

j ii i ii i

j ti i

t e e

t e

Nano System Control Lab.Chosun University 198

F-404 turbofan engine control (GE)

1 11

2 22

3 3

x xu

x * x *u

x x

1

2

3

: LP roto speed: HP roto speed

: Turbine temper

xx

x

1

2

u : Fuel Flowu : Nozzle Area

11

32

1

xyy

xy

G(s) C(sI A) B at ω 0.1 rad/sec0.863 0.11 4.609 0.47

G(j0.1)0.9347 0.25 0.88 0.32

j jj j

G(0) :steadystate

max

min

4.846 (13.7 )0.757 ( 2.4 )

DBDB

G(s)K(s)

Nano System Control Lab.Chosun University 199

Open loop TR function

1 rad/sec

2.4

13.70

max

min

( )G j

From Matrix X

0

0

max 0.22

180

min 0.23

0.22 0 0.22V

0.975 0.009 0.98

0.98V0.22

j

j

j

j

j ej e

ee

*

6.15

max 19.38

0

min 166.6

0.94U

0.22

0.23U

0.97

j

j

j

j

ee

ee

( )i max iy b j te

0.1

,

,

,( )

i min iy d ; y = Gxj te

Nano System Control Lab.Chosun University 200

For max ;

01

max 02

u 0.22sin (0.1 0) 4.55sin(0.1 6.15 )x y

u 0.98sin (0.1 0.22) 1.115sin(0.1 19.38 )t t

t t

Input to study Max. amplitude

Phases are differentFor min

min min min 0

0.174sin(0.1 0)y U

0.7341sin(0.1 166.6 )t

t

Min amplitude

0

min 0

0.98sin(0.1 180 )x u V

0.22sin(0.1 0.23 )tt

max max max maxx V , y U

Nano System Control Lab.Chosun University 201

SVD problem

x

F

v

0x

Car suspension

0M F K( ) b( v)x x x x

0If 0x

0 1 0 0 FK b 1 b v- -M M M M

x xx x

1

2 2

1 0 0 0 Fy

v 0 1 0 1 v

1 b- -1 M MG(s) C(sI A) B D b K s Ks s -s -M M M M

x xx x

M

b k

Nano System Control Lab.Chosun University 202

For n0.5 , ω 10, M 1

At ω 10.01 0.001 0.1 0.01

G(j1)0.001 0.01 0.99 0.1

kg

j jj j

max min

max min max min

U. .V SVD G(j1)1.002, 0.01

U= U U V= V V

If input u (not U) is chosen as

(1) max max max maxu V , then y Uj t j te e

(2)min min min minu V , then y Uj t j te e

1 1 1 1max min

2 2 2 2

a sin(ωt ) b sin (ωt )U U

a sin(ωt ) b sin (ωt )

,

,

Nano System Control Lab.Chosun University 203

u1u

2u

y

Nano System Control Lab.Chosun University 204

Stability of MIMO Feedback system Nominal stability

• Time domain - locations of poles• Frequency domain- Nyquist criterion

For

1T(s) C(sI A) B T(s)r(s) e(s) y(s)

1CL: C(s) I+T(s) T(s)

For u(t) r(t) y(t) r(t) Cx(t)

x (A BC)x Bry=Cx

“Close” Loop TF1C(s) C(sI A BC) B

Stability Re (A BC) 0 for alli i

Nano System Control Lab.Chosun University 205

Key relation

0 (s) det(sI A) ; T(s)L

(s) det(sI A+BC) ; C(s)CL

0* (s) ( ) det( I T(s))CL L m m m ms

Proof ;

1m

1m

1m

m

(s) det (sI A BC) det(sI A) det(I C(sI A) B)

det (sI A) (I C(sI A) B)

det (sI A) I (sI A) BC

(s) I +T(s)

CL

OL

Nano System Control Lab.Chosun University 206

Complex variable theory facts

• = Complex scalar variable ; “Lives” on s-plane• = Scalar-valued analytic function : maps

s-plane to another complex plane

(s)fs

(s)f Analytic in R

(1) Derivative exist at each point of R

(2) Unique in R

Nano System Control Lab.Chosun University 207

Ex)If

s+1(s)= , s 1 j2s

f

then2+j2 6 2(s)= j1+j2 5 5

f

..(s)f

Im

Re

The principle of the Argument

A theorem in complex variable theory

Suppose

C = closed contour in the s-plane.

(s)f = complex-valued scalar function

(1) Analytic in C(2) z zeros inside C(3) p poles inside C

Nano System Control Lab.Chosun University 208

Then, The image of C under mapping (s)f

(A) Generates a closed contour in the C.P.(B) encircle the origin 0, z-p times in a clockwise direction.

Im

Re

o

o

j

Notation

N( , (s), ) #a f C of clockwise encirclement of point by the image of under mapping (s)f

a C

Nano System Control Lab.Chosun University 209

The principle of Argument

N(0, (s),C) z pf

z p 3

Suppose , 1 2(s) (s) (s)f f f

1 1,z p 2 2,z p

Then, 1 2N(0, (s), C) N(0, (s), C) N(0, (s),C)f f f

Nano System Control Lab.Chosun University 210

S-plane

Scalar mapping plane to plane det I T(s)

Im

Re

MIMO Nyquist contour,

R

Semi circle with radius

Define RD ;

R0

RD

Nano System Control Lab.Chosun University 211

The MIMO CL system is stable iff

R u R uN 0, det I T(s) , D P , or N 1, 1 det I T(s) , D P

uP # of unstable poles of T(s) (open-loop)

Proof)

Recall (s) det I+T(s)CL OL

and R R RN 0, (s), D N 0, (s), D N 0, det I T(s) ,DCL OL

MIMO Nyquist Criterion

T(s)=G(s)K(s)e(s)r(s) y(s)

T(s)

0 0 0(z p ) ?

Nano System Control Lab.Chosun University 212

Let g(s) det I+T(s)

2 2

kT(s)s(s 2 s )n n

A B

C

1

j

s-plane

Nano System Control Lab.Chosun University 213

( )g j g(j ) je

( )

. A ; s j 0g(jω) g(jω) j

sege

. B ; s Re , RT(s) 0

g(s) det I+T(s) 1

jseg

Re

Im

1

B .

A

0

C

.

. C ; s j 0seg

Nano System Control Lab.Chosun University 214

* ( ) G(s) K(s)t s

( j )K(jω)

G(jω)K(jω) ( j ) G(jω)

t

t

For SISO case

u R

R R

T(s) ( ) det I+T(s) 1 ( )

-P N(0, det I T(s) , D

N 0,1 (s), D N 1, (s), D (open- loop circlement)

t s t s

t t

G(jω) = Unstable plant

T(s): desired

1mPositive phase

margin

1

1 1G( jω)

Positive gain margin

Nano System Control Lab.Chosun University 215

MIMO Case

0

det I+G(jω)

det I+T(jω) : desired T jω G(jω)K(jω)

1

Nano System Control Lab.Chosun University 216

MIMO Nyquist Criterion

• SISO 1C(s) 1 ( ) ( )t s t s

1R R RN 0, ( ), D N 0, 1 ( ) , D N 0, ( ), DC s t s t s

C C C R 0 0

C R 0 0

Z P N 0, 1 ( ), D Z P

N 1, ( ), D Z P

t s

t s

We want CP 0

C 0 0 C R

0 C R

Z Z +P =N 1, ( ), D

P =N 1, ( ), D

t s

t s

: Nyquist criterion for SISO (ccw)

cZ Pc 0 0Z Pccw

Nano System Control Lab.Chosun University 217

• MIMO

1

1 21 2 1 0

11 1 0

C(s) I+T(s) T(s)

(s)det I+T(s)

( ) det sI A BC s s s s

only eros poles of CL(s) det(sI A) s s s

eros poles of OL

CL OL

n n nCL n n

n nOL n

s

z

z

R R R

SC UC

SO UO

N 0, (s), D N 0, (s), D N 0, det I+T(s) , D

#of Z( ) Z Z#of Z( ) Z Z

CL OL

CL

OL

nn

Stable Unstable open loop

Nano System Control Lab.Chosun University 218

UC UO R

UC UO R

Z =Z N 0, det I T(s) , D

P =P N 0, det I T(s) , D

Counter clockwise

0

det I T(s)

Stable

: roots of characteristic eq.

: poles of closed loopC(s) T(s)

UC

R UO

C R UO

We want "P 0"

N 0, det I+T(s) , D P

N 0, det I+T(s) , D P

Nano System Control Lab.Chosun University 219

Stability Robustness : SISO

Central theme- to provide guarantee that compensator designed on the basis of nominal plant model, will not yield unstable feedback system

Assume

Question: then,

Ng (s) ,NK (s)

Ng (s)Ne (s) Ny (s)NK (s) Nu (s)

stable

Is this stable?

Ag (s)e(s) y(s)NK (s)

u(s)

Nano System Control Lab.Chosun University 220

A Ng (s) g(s) g (s)

Source of g(s)

• Low order approximation of high-order system.

Directional Uncertainty

At (jω) >1g( jω) ?

1

Nt (jω)At (jω)

g( jω) has 180 uncertaintyo

• actuator of sensor dynamics.

• “Fast” mechanical dynamics : bending, torsional.

• “Small” time delay.

Nano System Control Lab.Chosun University 221

SISO Model Error (cascade Representation)

; Error reflected at the plant output

A 0 N N 1, g (s) (s) g (s) g (s) (s)l l

0 (s)lu(s) y(s)

Ng (s)

1(s)lu(s) y(s)

Ng (s)

For SISO

11ω

2ωAt (jω)

Nt (jω)

; modelling error0l

0

Nano System Control Lab.Chosun University 222

Test 1 (For Robustness)

N N

N N

N N

Vector sum-1 d (ω) t (jω)

d (ω) 1 t (jω)

d (ω) 1 t jω

" make this resonable big"

Test 2 NN

N

t (s)C (s)1+t (s)

Assume NC (s), (s)l are stable.

Then, NN

1 1If ( jω) 1 or C (jω)C ( jω) ( jω) 1

ll

The actual Loop is still stable Sufficient condition

Nt (jω)

1

Nd (ω)0

A N; t (s) (s)t (s)l

Nano System Control Lab.Chosun University 223

Proof)

A N

A N N

A N

N

t (s) (s)t (s)t (s) t (s) (s) 1 t (s)

t (s) t (s)(s) 1t (s)

ll

l

A N Nt ( jω) t ( jω) 1 t ( jω)

“The inverse is true”

1

N1 t (jω)

Nt (jω)

A Nt (jω) t (jω)

At (jω)

0

N

N N

1 t ( jω) 1( jω) 1t (jω) C ( jω)

l

divided by at the both sidesNt ( jω)

Nano System Control Lab.Chosun University 224

Alternate model

Percent deviation

m

m

m

(s) 1 e (s) (s)(s) e (s) (s)

(s) 1 e (s)

A N

N

g gg g

l

u(s) y(s)Ng (s)

me (s)

12

unstable region N

1C (jω)

me (jω)

Nano System Control Lab.Chosun University 225

N

1C (jω)

me (jω)stable

NC (jω)

Robustnessboundary

m

1e (jω)

Nano System Control Lab.Chosun University 226

Design Implication

Need ; (1) Ng (s)

(2) Some upper bound of m maxe (jω) e (ω)

Conservatism.“sufficient condition”

maxe

max

1e

NC (jω)minimize the bandwidth NC (jω)

Nano System Control Lab.Chosun University 227

Robustness Condition for SISO case

mN

Nm

1e (jω)C (jω)

1C (jω)e (jω)

; “MIMO Nyquist criterionNec. Sufficient condition”

NC (jω)

1me (jω)

Bode plot

m maxe (jω) e (ω)

Nano System Control Lab.Chosun University 228

MIMO Stability Robustness

Assume : Nominal Feedback Loop is stable

Need : Actual Feedback Loop

Nominal plant ; NG (s)Actual plant ; AG (s)

• Plant Output Error

• Plant input Error

oL (s)u(s) y(s)

NG (s)

A 0 NG (s)=L (s)G (s)

NG (s)y(s)

IL (s)

A N IG (s)=G (s)L (s)

u(s)

0 IL (s) L (s) for MIMO

Nano System Control Lab.Chosun University 229

Robustness Test (MIMO Case)

For a stable nominal loop,

L(s)NK (s) NG (s)

The actual loop is stable if

(1) L(s) is stable

and

-1max min N

N N N

-1max min N

1

L(jω) I I T (jω) , ω

T (jω) G (jω)K (jω)or

L(jω) I C (jω)

C I+T T

Nano System Control Lab.Chosun University 230

NT (s)

mE (s)

mL(s)= I+E (s)

C

Alternate Test

1max N max m

max m

1C (jω) E (jω)E (jω)

max NC (jω)

1max mE (jω)

Nano System Control Lab.Chosun University 231

Robustness Test proof;

Step ;

(1) Assume is stable

(2) Find “Smallest” model error in “worst” direction that will cause actual Feedback System to be on the verge of instability

(3) Express (2) on Nyquist plot.

(4) Use a SV fact to obtain a sufficient condition for stability

NC (s)

Nano System Control Lab.Chosun University 232

SV factmax minIf ( B ) ( A )

Then , det(A+B) 0n n n n

0. 1

N

N u

det I+T (jω)

N 0, det I T (jω) ,D PR

Adet I+T ( jω)

i

A

at ω ω*det I T ( jω) 0

For stability ;

Adet I T (jω) 0, ω

Set

-1m N

-1 -1 -1 -1m N m N N N N N

-1m N N N

B E ( ) , A=I+T ( )

0 det E +I+T det E T T +T T +T

det E T +T +I det T

s s

-1Nsince det(T ) 0,

Nano System Control Lab.Chosun University 233

A m NT I+E T

m N N m Ndet E T +T +I det I+(I+E )T 0

AT (s)

NT

mE

* Sufficient condition for stability Robustness

-1max m min N

-1min N

max N

E (jω) I+T (jω)

C (jω)

1C (jω)

Nano System Control Lab.Chosun University 234

max Nmax m

1C (jω)E (jω)

max NC (jω)

1max mE (jω)

max NT (jω)