13

1388 -
:

)(fundamental subspaces) R ( A), N ( A), R( A T ), N ( A T

) (pseudo inverse 1

Amn x n1 = b m1
x Ax b . Amn rank ( A) = n

- - QR
Amn rank ( A) = k AT A ill condition
)(SVD
2

)( Pseudo - Inverse
A# AA# A = A -1
A# AA# = A# -2

( AA# )T = AA# -3
( A# A)T = A# A -4
- ) (Moore - Penrose Conditions .
#
-1 Am n An m .

# #
( A ) = A -2 . ( AT ) # = ( A# )T

. A# = ( AT A) # AT = AT ( AAT ) # -3
#
#
#
#
-4 A A, AA , I A A, I AA .

 )( Pseudo - Inverse
-

A 1 = (UV T ) 1 = V 1U T

if Ann ,

full rank

) 1 = diag (1 1 , 1 2 , K , 1 n
-

An#m = V #U T

1 2 L k > 0

Amn

) n#m = diag(1 1 , 1 2 , K , 1 k ,0, K ,0

#
- x = A b .

Amn A# = ( AT A) 1 AT . Ann A# = A1 .4

1
#
A AA
.

1 0 1 2
A = 1 2 1
0
0 1 1
1

A
0
0 0
0.5774 0.7071 0.4082 3

A = UV = 0.5774 0.7071 0.4082 0 2.4495 0 0

0.5774
0
0.8165 0
0
0 0
T

0.5774
0.1066 0.8095
0
0.5774 0.5774
0.3515
0.4581

0.5774
0
0.8095 0.1066

0.5774 0.5774 0.4581 0.3515

rank( A) = 2 # A#
5

0.1667
0
0.1667
0.0556
0.2778 0.1111
#
#
T

= A = V U
0.1111 0.1111 0.1111

0.2778 0.0556 0.1111

0
0
0

0
0.3333
0
0.4082
= #
0
0

0
0

#
AA

0.1667
0
0.1667
1 0 1 2
0.0556
0.2778 0.1111

0
AA = 1 2 1
0.1111 0.1111 0.1111
0 1 1
1

0.2778 0.0556 0.1111

0.8333 0.1667 -0.3333
= 0.1667 0.8333 0.3333
-0.3333 0.3333 0.3333
6

) pinv(A MATLAB .;]A = [1 0 - 1 - 2;1 2 1 0;0 1 1 1

)pinv(A
= ans
- 0.0000

0.1667

0.1667

0.1111

0.2778

0.0556

0.1111

0.1111

- 0.1111

0.1111

- 0.0556

- 0.2778

.
7

2

1 0 1
1

Ax = b
1 2 1 x = 1
0 1 1
1

.

. rank ( A) = 2 rank ( A | b) = 3
.
A x = ( AT A) 1 AT b
.
8

A A = UV T
0
0 0.2852 0.7651
0.5774
0.4082 2.7651
0.0849 0.9089

0.2650 0.4082 0
1.5344 0 0.8052 0.1355 0.5774
A = 0.8736
0.4792 0.3220 0.8165 0
0
0 0.5199 0.6295 0.5774

T

A # = V #U T

0.4444 0.2222 0.1111

A = 0.0556 0.2778 0.1111
0.3889 0.0556 0.2222

#

0
0
1 2.7651

= 0
1 5344 0
0
0
0
#

0.5556
x = A b
x = 0.4444

-0.1111

#

A AT A ill condition .
9

 : A # = V #U T x = A#b
Ax = b
Ax = b . b R( A)
) b R( A b b b .
Ax = b x
. b ) R( A.
A x = b b
43

w2 ,b
w3,b
w1 , b
w1 +
w2 +
w3
= projbR(A) = b
w1
w2
w3

w 1 , w 2 , w 3 ) R( A . A

0 v 1T

0 v T2 = u1 1 v 1T + u 2 2 v T2 + u 3 3 v T3
0 v T3

1 0
u 4 ] 0 2
0 0

0
0

u2

u3

A = UV T = [u1

u1 , u 2 , u 3 ) R( A b
b = u1 , b u1 + u 2 , b u 2 + u 3 , b u 3 = (u1T b)u1 + (u T2 b)u 2 + (u T3 b)u 3
A
1

Av i = u i

A = u i i v Ti

b
) (u T b
)(u T b
)(u T b
)(u T b
)(u T b
)(u T b
b = 1 Av 1 + 2 Av 2 + 3 Av 3 = A 1 v 1 + 2 v 2 + 3 v 3 = Ax
1
2
3
2
3
1

= u T3 b

u T2 b + v 3

u1T b + v 2

1
0 0 u1T
1

0 u T2 b = A # b
v 3 ] 0 1
2

1 T
0
0
3 u 3

v2

v 3 = v1

)(u T3 b

v2 +

)(u T2 b

v1 +

)(u1T b

= x

1
1 T
1 T
= v 1 u1T + v 2
u2 + v3
u 3 b = [v 1
2
3
1

0 u1T

0 u T2
1
1 T
1 T
= v 1 u1T + v 2
u2 + v3
u3
T
1

1
2
3
3 u3
T
0 u 4

0
1

0
0

1
1
0
] v 3
0

v2

A # = V T #U = [v 1

x = A#b .
10

Low rank matrix approximation

- A

rank ( A) = r

0 V1T
,
V2T

0
1

O
] U 2
0
r

A = [U1

B k < r A B . )(Data Compression11

A rank ( A) = r

V1Ta
T
V1b
V2T

k +1

O
0
0

A = UV T

0
1

0
k

] A = [U 1a U 1b U 2

A = u1 1 v1T + L + u k k v Tk + u k +1 k +1 v Tk +1 + L + u r r v Tr

A . k +1
u k +1 k +1 vTk +1 + L + u r r vTr
A .
12

 - B

V1Ta
T
V1b
V2T

0
1

0
k

0
0

rank ( B) = k < r

A B = k +1

] B = [U 1a U 1b U 2

B = u1 1 v1T + L + u k k v Tk

A B = U AV T U BV T = U ( A B )V T
13

3
A .

14

.15

x 2 m +1 L
x 2 m + 2 L
M
M

x3m L

x m +1
xm+2
M
x2m

x1
x
= 2
M

xm

x3 L x r ] Amn

x2

X = [x1

200 405 .
A A .
.
5 = 0.7289

4 = 0.7768 ,

3 = 0.9626 ,

2 = 1.0639 ,

1 = 16.0649 ,
16

 Clean Signal
4

10

10

10

-2

Noisy Signal
4

-2
5

Enhanced Signal

-2
5

17

5 = 0

4 = 0,

3 = 0,

2 = 0,

1 = 16.0234 ,

5 = 0.7289

4 = 0.7768 ,

3 = 0.9626 ,

2 = 1.0639 ,

1 = 16.0649 ,

18

 A
1000 4025 .
25
5 = 2.8992
10 = 2.2070
15 = 1.7182
20 = 1.1944
25 = 0.5875

1 = 36.1256 , 2 = 3.3546 , 3 = 3.1330 , 4 = 2.9426 ,

6 = 2.7629 , 7 = 2.5845 , 8 = 2.4875 , 9 = 2.3294 ,
11 = 2.1174 , 12 = 1.9658 , 13 = 1.9043, 14 = 1.7803,
16 = 1.5574 , 17 = 1.4762 , 18 = 1.2959 , 19 = 1.2511,
21 = 1.0308 , 22 = 0.9332 , 23 = 0.8637 , 24 = 0.7150 ,
19

Clean Signal
4

-2
10

10

10

Noisy Signal
4

-2

Enhanced Signal
4

-2

20

 Clean Signal
4

-2
10

10

10

Noisy Signal
4

-2

Enhanced Signal
4

-2

21

.
) (weighting
2 0,5
.

22



5 80
5 = 0.3941

4 = 0.4875 ,

3 = 0.5069 ,

2 = 0.5418 ,

1 = 89.1145 ,

output of the noisy system

5

4.5

3.5

60

50

60

50

40

30

20

10

output of the system after noise reduction with SVD method

5

4.5

3.5

40

30

20

10

23

.
y = 3.2410 + 0.0732t 0.0012t 2 + 7 10 6 t 3
output of the system after noise reduction with SVD method
5

4.5

3.5

10

20

30

40

50

60

50

60

3rd order least square data fit

5

4.5

3.5

24

10

20

30

40

 )(Data Compression

) ;
4
a
( 6
: e
g
1 a
m
a i

MATLAB MATLAB =

0.5

0.6

0.7

0.8

0.9

1.1

1.2

1.3

1.4

60

50

30

40

10

20

1.5

25

.
0.5
1
1.5
2
2.5
3
3.5
4

5.5

4.5

3.5

2.5

1.5

4.5
0.5

A
26

)
; A
)
5
*
,
4
0
(
0
d
1
( n
a e
r
g
a
m
A i

72.6078 51.1643
18.6431 17.6336
34.5255 67.9415

70.2714 93.0307

23.3649 36.3295 26.9719

93.4402 89.2667 16.7493
= 100A
49.7758 4.8464 5.4033

81.9026 53.5329 13.6772

. rank ( A) = 4 27

A = UV T
0.4063 0.3543 0.7843 0.3070
0.4981 0.8393 0.0371 0.2146

= U
0.3649 0.3505
0.5927 0.6267

0.1795
0.6833
0.6736 0.2171
0
0
0
0
224.7721

0
85
.
0690
0
0
0

0
0
44.7287
0
0

0
0
0
7.3295 0

0.5755 - 0.4105 0.5009 - 0.3346 0.3706

0.4318 - 0.5728 - 0.4321 0.4413 - 0.3225

V = 0.1356 0.0042 - 0.3604 - 0.8071 - 0.4475

0.4392 0.4401 - 0.5492 0.0118 0.5583

0.5206 0.5565 0.3617 0.2043 - 0.4967

A 28

rank ( A1 ) = 1,
52.5564
64.4383
A1 =
47.1990

87.1385
rank ( A2 ) = 2 ,

1 = 224.7721 ,

39.4285 12.3849 40.1043 47.5420

48.3424 15.1849 49.1710 58.2902
35.4093 11.1225 36.0162 42.6958

65.3724 20.5342 66.4929 78.8247

1 = 224.7721,

40.1854 22.1638
93.7458 89.24.31
A2 =
34.9594 18.3280

79.5569 54.7917
rank ( A3 ) = 3 ,

1 = 224 .7721 ,

22.6119
92.9140
A3 =
48.2388

83.5784
29

2 = 3 = 4 = 0

2 = 85.0690 ,

3 = 4 = 0

12.5130 53.3675 64.3145

14.8816 17.7497 18.5555
11.2491 49.1386 59.2901

20.6127 74.6213 89.1037

2 = 85 .0690 ,

3 = 44 .7287 ,

37.3225 25.1555 72.6343

89.9606 15.4799 18.6616
48.2388 1.6958 34.5797
83.5784 17.7196 70.2123

51.6239
17.9548
68.8797

92.0078

4 = 0

 - A2 A1 A3

5.5

5.5

4.5

3.5

2.5

1.5

0.5

0.5

1.5

1.5

2.5

2.5

3.5

3.5

4.5
0.5

5.5

A1

4.5

3.5

2.5

1.5

A2

4.5

3.5

2.5

1.5

4.5
0.5

0.5

0.5

1.5

1.5

2.5

2.5

3.5

3.5

4.5
0.5

5.5

4.5

3.5

2.5

1.5

4.5
0.5

A3

A3 A .
30

 15 20 2

10

12

14

20

16

18

14

12

10

1 = 912.0093 ,
6 = 148.8492 ,
11 = 72.6961,

2 = 208.4180 , 3 = 193.4322 , 4 = 172.7841, 5 = 159.7591

7 = 130.7079 , 8 = 120.6242 , 9 = 112.0716 , 10 = 97.8217
12 = 64.4814 , 13 = 50.5712 , 14 = 43.4440 , 15 = 40.0670

31

10

10

10

12

12

12

14

20

18

16

14

12

10

14

20

1 ,K, 5

20

18

16

14

12

10

14

18

16

14

12

10

20

18

1 ,K, 3

1 ,K, 11

16

14

12

10

10

10

10

12

12

12

14

14

14

20

18

16

14

12

10

1 ,K, 9

20

18

16

14

12

10

1 ,K, 7
32

20

18

16

14

12

10

10

10

12

12

14

14

20

18

16

14

1 ,K, 11

12

10

33

- .

362 500 362 1 = 150.2370 362 = 0.1005 .

34

1 = 150.2370,K, 10 = 11.2150

1 = 150.2370,K, 5 = 20.8949

1 = 150.2370,K, 50 = 3.2094

1 = 150.2370,K, 20 = 7.1647

1 = 150.2370,K, 150 = 1.0507

1 = 150.2370,K, 100 = 1.6042

150 .
35