Top Banner

of 118

compressive sensing theory notes

Jun 03, 2018

Download

Documents

Madhav Gumma
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
  • 8/12/2019 compressive sensing theory notes

    1/118

    Introduction to Compressive Sensing

    Collection Editor:

    Marco F. Duarte

  • 8/12/2019 compressive sensing theory notes

    2/118

  • 8/12/2019 compressive sensing theory notes

    3/118

    Introduction to Compressive Sensing

    Collection Editor:

    Marco F. Duarte

    Authors:

    Mark A. Davenport

    Ronald DeVore

    Marco F. Duarte

    Chinmay Hegde

    Jason Laska

    Michael A Lexa

    Shriram Sarvotham

    Mona Sheikh

    Wotao Yin

    Online:< http://cnx.org/content/col11355/1.2/ >

    C O N N E X I O N S

    Rice University, Houston, Texas

  • 8/12/2019 compressive sensing theory notes

    4/118

  • 8/12/2019 compressive sensing theory notes

    5/118

  • 8/12/2019 compressive sensing theory notes

    6/118

  • 8/12/2019 compressive sensing theory notes

    7/118

    BA

    BA:={fL2(R) :

    f () = 0,|| A}.

    f

    f () := 1

    2 Rf(t) eit dt.

    f L1 f L2

    f(t) := 1

    2

    R

    f () eit d.

    fBA f fnA

    f(t) =nZ

    f n

    A

    sinc ( (At n)) ,

    sinc (t) = sint

    t

    A= 1 fBA=1

    f(t) = 1

    2

    f () eit d.

  • 8/12/2019 compressive sensing theory notes

    8/118

    F() 2

    f

    F() :=nZ

    f (

    2n) .

    F()

    F() =nZ

    cnein,

    cn

    cn = 12

    F() e

    in d

    = 12

    f () ein d.

    cn= 1

    2f(n) .

    F() = 1

    2

    nZ

    f(n) ein.

    f() = F() [,]= 1

    2

    nZ

    f(n) ein[,],

    F[,] = 12 sinc () F(g (t n)) =einF(g (t)) ,

    f(t) =nZ

    f(n) sinc ( (t n)) .

    {sinc ( (t

    n))

    }n

    Z

    L2

    f2L2= 2nZ

    |f(n) |2

    f

    nZ

    |sinc ( (t n)) | nZ

    1

    |t n| + 1

  • 8/12/2019 compressive sensing theory notes

    9/118

    f () = 0 || , >0,

    [,]

    g ()

    g

    g

    g

    g

    g () = 1 || g= 0 ||> g m1 g

    |g (t) | C(|t| + 1)

    m

    C >0

    g

    f () = F()

    g () = 12

    nZ

    f(n) ein

    g () .

    f(t) =nZ

    f(n) g (t n) ,

    [ , ]

  • 8/12/2019 compressive sensing theory notes

    10/118

    [, ] [ , ] [,]

    f(t) =

    nZ f(n) sinc ( (t n)) ,f(t) =

    nZ fn

    g (t n) .

    f BA f A

    f f

    f [T, T] T >0

    f BA=1 [T, T] L [T, T]

    BA:={fL2(R) :|

    f() |= 0, || A}.

    fBA f=n

    f n

    A

    (At n) .

    g()

    1A 2A , f (x) = (x)x

    > 1

    f=

    f n

    A

    g(At n) .

  • 8/12/2019 compressive sensing theory notes

    11/118

    1A 2A

    g() g()

    |g(t) | c,k(1 + |t|)

    k, k= 1, 2,...

    g()

    ()

    [T, T] [cT,cT]

    t [T, T] [cT,cT]

    c > 1 [T, T]

    [T, T] T > 0 BA K

    X

    K

    K X (E, D)E K D

    X

  • 8/12/2019 compressive sensing theory notes

    12/118

    d

    d (K, E , D, X) := fKf D (Ef) X .

    n (K, E) = fK#Ef #Ef Ef n fK

    n(K, X) := (E,D){n (K, E) :d (K, E , D, X)} K

    N d (K, E , D, X) E, D n (K, E)

    N

  • 8/12/2019 compressive sensing theory notes

    13/118

    K

    >0 K K

    KUNi=1b (fi, ) . N := {N : } N(K) K

    X K N= N(K, X)

    H(K, X) K X

    H(K, X) = logN(K, X) .

    KX n(K, X) =H(K, X)

    f K N

    K, X

    logN

    K, X

  • 8/12/2019 compressive sensing theory notes

    14/118

    BA(M) ={fBA:|f(t) | M, t R} M

    L [T, T] T > 0 BA(M) L BA(M)L [T, T]

    nA

    [T(1 + ) , T(1 + )]

    BA

    f () = 0 || A |f| M f = (T) > 1 > 0 k 2k1 < 2k f f f nA

    nA [T(1 + ) , T(1 + )] (T)> 0 f

    [T, T] f nA k + k0(T) f f

    nA

    f nA

    k+ k0(T) f

    nA

    f nA

    f nA

    f n

    A

    2kk0 M.

    g

    f(t) = nNT

    f nA g(At n) ,

    NT :={n: T(1 + ) nA

    T(1 + )}.

    |f(t) f(t) | nNT f nA f nA |g(At n) |+| nA|>T(1+)f nA |g(At n) |

  • 8/12/2019 compressive sensing theory notes

    15/118

    f nA

    f nA M 2kk0

    f

    nA

    M

    |f(t) f(t) | nNT M 2kk0 |g(At n) |+| nA|>T(1+) M |g(At n) |=: S1+ S2

    S1

    S1 =

    nNT M 2kk0 |g(At n) | M 2kk0 n |g(At n) | M C0() 2kk0 g ()

    k0 S1M C0() 2kk0 2 S2 /2 g

    BA T T 0 T 1 T 2T A k+ k0 (k+ k0) 2AT(1 + ) k

    limT

    n(BA(M) , LT, T)2T

    n

    [

    T, T]

    [T, T] k 2k

    A= 106

    T = 24 105 k = 10 A k 2T = 106 10 105 = 1012

    2k

    1km

  • 8/12/2019 compressive sensing theory notes

    16/118

  • 8/12/2019 compressive sensing theory notes

    17/118

    R3

    N R

    N R

    N p

    p[1, ]

    xp={N

    i=1 |xi|p 1p

    , p[1, ) ;max

    i=1,2,...,N|xi|, p=.

    RN

    < x, z >=zTx=Ni=1

    xizi.

    2 x2 =

    < x, x > p p

  • 8/12/2019 compressive sensing theory notes

    18/118

    x0 :=|supp(x) | supp(x) ={i: xi= 0} x

    |supp(x) | supp(x) 0

    limp0x0xp =|supp (x) |,

    p p {x:xp= 1}, R

    2 p < 1

    R2

    p p= 1, 2, p p= 1

    2

    1 2

    p

    x R2

    A

    p

    x A x xp p A x p p x A

    xA x p p p

  • 8/12/2019 compressive sensing theory notes

    19/118

    R2

    p

    p= 1, 2,

    p p= 1

    2

    1

    2 p

    ={i}iI V V

    V= RN I={1,...,N} x RN

    x=iIa

    ii,

    ai =< x, i > {i}iI x

    ={i}iI

    < i, j >= 0 i= j = T

    {i}ni=1 Rd d < n Rdn x Rd

    Ax2

    2 T

    x2

    2Bx2

    2

    0 < A B 0 A B A B A= B A A = B = 1

    > 0 i2 = i= 1,...,N = 1 d N A B T

  • 8/12/2019 compressive sensing theory notes

    20/118

    x

    x =

    T = T =I

    =

    T1

    A > 0 T

    d= T

    T1

    x.

    2 d2 2

    x=

    x K K x0K

    K={x:x0K} K

    x K x x= 0K

  • 8/12/2019 compressive sensing theory notes

    21/118

    Xk =N1

    n=0 xncosN

    n + 12

    k

    k = 0, , N 1 xn

  • 8/12/2019 compressive sensing theory notes

    22/118

    K p K K

    K

  • 8/12/2019 compressive sensing theory notes

    23/118

    K

    x

    x= ,

    x x s

    |s| C1sq

    , s= 1, 2, ....

    q

  • 8/12/2019 compressive sensing theory notes

    24/118

    0 1 2 3 4 5 6

    x 104

    0

    0.5

    1

    1.5

    Sorted indices

    Waveletcoefficientmagnitude

    K N K K K K K(x)

    K(x) = arg minK

    x 2.

  • 8/12/2019 compressive sensing theory notes

    25/118

    K(x)C2K1/2s.

    K(x)2

    Kr

    i

    ir+1/2 K

    p

    p x (n) p p p

    xp=

    i

    |xi|p 1

    p

  • 8/12/2019 compressive sensing theory notes

    26/118

    p

    p

    p p p

    x [n] p q > 1/p

    p p1

  • 8/12/2019 compressive sensing theory notes

    27/118

    x RN M

    y= x,

    M N y RM R

    N N RM M N

    x

    x

    x x

    y

    MN

  • 8/12/2019 compressive sensing theory notes

    28/118

    N() ={z : z= 0}. x x

    x, x K={x:x0K} x= x

    x

    x

    y

    x = x

    x x = 0 x x 2K xK N() 2K

    y RM xK y = x spark ()> 2K

    y RM xK y = x spark () 2K 2K h N() h 2K h 2K h = x x x, x K h N()

    x x

    = 0 x = x x

    K y = x

    spark ()> 2K spark ()> 2K y x, x K

    y = x = x

    x x = 0 h = x x h = 0 spark ()> 2K 2K h= 0 x= x

    spark()[2, M+ 1] M2K

    N()

    {1, 2, , N} c ={1, 2, , N} \

    x N x c

  • 8/12/2019 compressive sensing theory notes

    29/118

    M N c

    K

    C > 0

    h2Chc1

    K

    h N() || K

    h K hc1 = 0 h= 0 K N() h= 0

    x : RM

    R

    N

    (x) x2CK(x)1

    K

    x

    K(x)p= min

    xKx xp.

    K K x

    x

    x

    p p

    K(x)1/

    K K(x)2

    2K

    2K

    : RN RM : RM RN (, ) 2K

    ||

    M||

  • 8/12/2019 compressive sensing theory notes

    30/118

    h N() 2K h 0 1 |0| =|1| = K x = h1+ hc x =h0 h= x

    x

    x

    K x

    = x h N()

    h=

    x x = 0 x = x x = (x)

    h2 h2 =x x 2 =x (x) 2CK(x)1

    K=

    2C

    hc12K

    ,

    (x) x2CK(x)1

    K,

    K K(0, 1)

    (1 K) x22 x22(1 + K) x22, xK={x:x0K}

    2K K

    x22 x22x22 0<

    0

    K=c1log (p)

    p = /4

    M c0log (p)2

    =16c0K

    c12 .

    K K/2 Klog (N/K)

    2K 2K | u1

    u2sgn(u) 2 sgn(u) K 1 uK sgn(u) =

    K

    K u u

    h h N()

    h N()

    1

    2K h RN h= 0 0 {1, 2,...,N} |0| K 1 K hc

    0 = 0 1

    h2hc

    01

    K+

    ||h2

    ,

    =

    22K

    1 2K, = 1

    1 2K.

    h

    h N()

    h N()

    h2Chc1

    K

    2K h

    0

    K

    h

    h= 0

    h2hc

    01

    K.

  • 8/12/2019 compressive sensing theory notes

    36/118

    hc01

    =h11+ hc1

    Kh12+ hc1

    h2

    h12+hc1

    K

    .

    h12 h2

    (1 ) h2hc1

    K.

    2K 0 M = 2K 2K 2K 1 2K 2K>0 2K

    NK

    K RN

    N K 2K 2K

    E

    2ij

    = 1

    M,

    1/M

    c > 0

    E

    eijtec2t2/2

  • 8/12/2019 compressive sensing theory notes

    37/118

    t R

    1/

    M

    c2 = E

    2ij

    =1M

    M N ij ij c2 = 1/M Y = x x RN > 0 xRN

    E Y22=x22

    PY22 x22x222exp

    M

    2

    = 2/ (1 log (2))6.52

    (0, 1) M N ij ij c2 = 1/M

    M1Klog

    NK

    ,

    K 1 2e2M

    1 2 = 2/2 log (42e/) /1

    x

    M

  • 8/12/2019 compressive sensing theory notes

    38/118

    NK

    ()

    i j

    () = max1i|i2j2

    .

    ()

    NMM(N1) , 1

    N M ()1/M

    N N

    M

    mij

    1i, jN

    N

    di = di(ci, ri) 1iN ci= mii ri=

    j=i|mij | G= T

    spark ()1 + 1 ()

    .

    spark () {1,...,N} ||= p G= T

    gii = 1 1ip |gij | () 1i, jp i=j

    j=i|gij |

  • 8/12/2019 compressive sensing theory notes

    39/118

    K 0 a=

    X = [X1, X2,...,XN] Xi Xi Sub

    c2

    RN < X, > Sub c2 22 XiSSub

    2

    RN < X, >SSub 2 22

    Xi

    E

    exp

    tN

    i=1 iXi

    = EN

    i=1 exp (tiXi)

    =N

    i=1 E (exp (tiXi))

    Ni=1 exp

    c2(it)2/2

    =expN

    i=1 2

    i

    c2

    t2

    /2

    .

    Xi c2 =2

    E

    < X, >2

    = 2 22

    X

    Xi Xi

    Sub(c) X2

    X t > 0

    P (Xt) E (X)t

    .

    f(x) X

    E (X) =

    0

    xf(x)dxt

    xf(x) dxt

    tf(x) dx = tP (Xt) .

    XSub c2 E

    exp

    X2/2c2 1

    1 ,

  • 8/12/2019 compressive sensing theory notes

    42/118

    [0, 1)

    = 0 (0, 1) f(x) X X

    exp (tx) f(x) dxexp c2t2/2

    t R exp c2t2/2

    exp

    tx c2t2/2 f(x) dxexp c2t2 ( 1) /2 . t

    exp

    tx c2t2/2

    dt

    f(x) dx

    exp

    c2t2 ( 1) /2

    dt,

    1

    c

    2

    exp

    x2/2c2

    f(x)dx 1c

    2

    1 .

    X= [X1, X2,...,XM] Xi XiSub

    c2

    E

    X2i

    = 2

    EX22= M 2.

    (0, 1) c2/2, max 4 max 2/c2

    PX22M 2exp M(1 )2/

    PX22M2exp M( 1)2/ .

    Xi

    EX22= M

    i=1

    E

    X2i

    =

    Mi=1

    2 =M 2

    PX22M2 = P exp X22exp M2

    E(exp(X2

    2))exp(M2)

    =QMi=1

    E(exp(X2i ))exp(M2) .

  • 8/12/2019 compressive sensing theory notes

    43/118

    XiSub

    c2

    E

    exp

    X2i

    = E

    exp

    2c2X2i/2c

    2

    1

    1

    2c2

    .

    Mi=1

    E

    exp

    X2i 1

    1 2c2M/2

    PX22M2

    exp22

    1 2c2

    M/2.

    =

    2

    c2

    2c22 (1 + ) .

    PX22M2

    2

    c2exp

    1

    2

    c2

    M/2.

    PX22M 2

    2

    c2exp

    1

    2

    c2

    M/2.

    = max

    4, 2

    max2/c 12(max2/c 1) log (max2/c)

    0, max2/c log ()( 1) 2( 1)

    2

    ,

    exp

    ( 1) 2( 1)2

    .

    = 2/c2 = 2/c2

    M2 M c2/2 1 c2 = 2

  • 8/12/2019 compressive sensing theory notes

    44/118

  • 8/12/2019 compressive sensing theory notes

    45/118

    MN

    M

    K(0, 1)

    (1

    K)

    x

    22

    x

    22

    (1 + K)

    x

    22

    xK K x K 2K > 0 M = 2K

    2K 2K 1 2K 2K>0 2K

    NK

    K RN

    N K 2K 2K 2K

    MN ij ij SSub (1/M) Y = x x RN > 0 x RN

    EY22=x22

    PY22 x22x222expM2

    = 2/ (1 log (2))6.52

    E

    2ij

    = 1

    M,

    1/M

    K

    K

    (0, 1) Q q2 = 1 q Q|Q| (3/)K x RK x2 = 1 qQ x q2

  • 8/12/2019 compressive sensing theory notes

    46/118

    Q

    q1 RK q12 = 1

    Q

    i

    qi RK qi2 = 1

    qi

    qj

    2 > j < i

    x RK x2 = 1 q Q x q2 |Q| /2 Q 1 + /2 BK(r) r R

    K

    |Q| BK (/2) BK (1 + /2)

    |Q| (BK(1+/2))

    (BK(/2))

    = (1+/2)K

    (/2)K

    (3/)

    K.

    (0, 1) M N ij ij SSub (1/M)

    M1Klog

    N

    K

    ,

    K 1 2e2M

    1 > 1 2 = 2/2 1/1 log (42e/)

    x2 = 1 T {1, 2,...,N} |T|= K XT K T QT QT XT q2 = 1 qQT xXT x2 = 1

    minqQT

    x q2/14.

    QT |QT| (42/)K

    T

    QT

    Q=

    T:|T|=KQT.

    NK

    T

    N

    K

    =

    N(N 1) (N 2) (N K+ 1)K!

    NK

    K!

    eN

    K

    K,

    K! (K/e)K |Q| (42eN/K)K

  • 8/12/2019 compressive sensing theory notes

    47/118

    = /

    2

    1

    2(42eN/K)

    KeM2/2 ,

    1 /

    2q22q22

    1 + /

    2q22, for all qQ.

    M

    log

    42eN

    K

    KK

    log

    N

    K

    + log

    42e

    M

    1+ Mlog

    42e

    1 2e2M A

    x

    2

    1 + A, for all x

    K,

    x

    2 = 1.

    A xK x2 = 1

    qQ x q2/14 x qK xXT qQT XT x q2/14

    x2 q2+ (x q)2

    1 + /

    2 +

    1 + A /14.

    A

    1 + A

    1 + /

    2 +

    1 + A /14

    1 + A

    1 + /

    2

    1 /14

    1 + ,

    x2 q2 (x q)2

    1 /

    2 1 + /14 1 ,

    = I K K

  • 8/12/2019 compressive sensing theory notes

    48/118

  • 8/12/2019 compressive sensing theory notes

    49/118

    x

    y = x x

    x

    x=argminz

    z0 subject to z B (y) ,

    B (y) x y z0 =|supp(z) | z

    B (y) ={z : z = y}

    B (y) ={z :z y2 } x y

    x x=

    =argminz

    z0 subject to z B (y)

    B (y) ={z : z = y} B (y) ={z :z y2 } =

    = I

    xx2 =

    cc2=

    2

    cc2 xx

    0

  • 8/12/2019 compressive sensing theory notes

    50/118

    0 1

    x=argminz z1 subject to z B (y) .

    B (y) B (y) ={z : z = y}

    R2

    1

    p p= 1

    2 1 p

    1

    1 p p < 1 1

    1

    1

    1

    1

    1

  • 8/12/2019 compressive sensing theory notes

    51/118

    1

    1

    1

    x=argminz

    z1 subject to z B (y) . B (y)

    1

    2K

    2K 0

    J H J(x) = x1 1 x H(x, y) = 12 xy22 2

  • 8/12/2019 compressive sensing theory notes

    62/118

    x H x1 J()

    J(x) = T V (x) + x1

    1024 1024

    1 J(x) =x1 O

    N3

    N

    M

    1

    minx

    x

    1+ H(x) ,

    H

    H

  • 8/12/2019 compressive sensing theory notes

    63/118

    shrink (t, ) ={

    t ift > ,

    0 if t, andt + ift < .

    i= 1,...,N ith x (k+ 1)th

    xk+1i = shrink

    xk [U+25BD]Hxki,

    >0 k xk+1 xk H() [U+25BD]H xk

    J

    H(x)

    H(x) =y x22[U+25BD]H(x) = 2 (y x) .

    xk+1i = shrink

    xk [U+25BD]Hy xki,

    y n

    x

    x 0 = 0 r= y k= 0

    kk + 1 x x T r

    xshrink (x, k ) ry x

  • 8/12/2019 compressive sensing theory notes

    64/118

    x x

    yk+1 =yk + y xkxk+1 =argmin J(x) + 2 x yk+1

    2.

    J(x) J(x) =x1 > 0

    1

    1

    x

    x y

    minI

    {|I|: y =iI

    ixi},

    I i = 1,...,N i ith

    I

    y

  • 8/12/2019 compressive sensing theory notes

    65/118

  • 8/12/2019 compressive sensing theory notes

    66/118

    K

    O (MN K) .

    M=CKlogN K M=CKlogN

    K

    y

    x

    0 = 0 r= y = i= 0

    ii + 1 b T r supp(T (b, 1))

    x

    i| x x i| C0 ry x i

    x x i

    K

    1

    r0 = y Trk1 kth

    S= 10

    O (KNlogN)

  • 8/12/2019 compressive sensing theory notes

    67/118

    1 K K

    i

    xi1

    eTr 2K e T supp xi1 b b|T Ty b|TC 0 b K xi ry xi

    y K

    K

    x x

    x 0 = 0 r= y i= 0

    e T r supp (T (e, 2K)) T supp

    x i 1

    b| T T y b| T C

    x

    i T (b, K) ry x i

    x x i

    O (MN)

    K

  • 8/12/2019 compressive sensing theory notes

    68/118

    x0

    xi+1 = T

    xi+ T

    y xi

    , K

    .

    x

    x

    1

    N

    K

    x

    xi= 0

    K

    xi = 0

    x ij j

    th

    ith

    x

    xi

  • 8/12/2019 compressive sensing theory notes

    69/118

    i x

    N= 232 x

    y = x M N M N y

    y

    xi xi i y ith y xi y = x x x x

    H h :{1,...,N} {1,...,m} H mN h H (h) m N j j = h (i) d h1,...,hd H M = md M N d

    x y= x y {h1,...,hd} ith

    y

    h

    yi

    yi=

    j:h(j)=i xj .

    j yi xj i h

    y jth

    xj =minl

    yi : hl(j) = i.

    xj xj d = ClogN

    m= 4/K

    x

    x x/K x x1, x K x 1

  • 8/12/2019 compressive sensing theory notes

    70/118

    xj

    xj =medianl

    yi : hl(j) = i.

    N x N x N

    K

    x

    K

    x

    x

    y = x x

  • 8/12/2019 compressive sensing theory notes

    71/118

    x y x (i) y (i)

    x

    x x y

    x (i) y (i)

  • 8/12/2019 compressive sensing theory notes

    72/118

  • 8/12/2019 compressive sensing theory notes

    73/118

    N M MN M 1 y

    y x yx

    M N N

    N

    x

    x

    x M N {1, ....M} N > M x N y =

    Mm=1 ixi = f

    y

    y= x + e e

    e

    x

    e

    (N M) N

  • 8/12/2019 compressive sensing theory notes

    74/118

    = 0

    y = y= x + e= e

    e

    e

    e

    y= y e x= y= y e

    m K M < NCKlogN/K C

    N

    K

    x xi= 0 K xi= 0

    x ij j

    th

    ith x

    xi

    i

    x

    N= 232 x y= x M N MN y y

    xi xi i y ith y xi y = x x x x

  • 8/12/2019 compressive sensing theory notes

    75/118

    x (t) x (t) x (t) x (t)

  • 8/12/2019 compressive sensing theory notes

    76/118

    x (t) {j(t)}Mj=1 t [0, T] M

    y [j] =

    T0

    x (t) j(t) dt.

    j(t)

    M

    x (t)

    j(t)

    M

    y [j]

    j(t)

    j(t) M

    j(t) j(t) =(t tj) {tj}Mj=1 M x (t)

    [0, T]

    M

    x (t) 1 pc(t) Na Na x (t) 1/Ma Ma Na

    y [j] =

    j/Ma(j1)/Ma

    pc(t) x (t)dt.

  • 8/12/2019 compressive sensing theory notes

    77/118

  • 8/12/2019 compressive sensing theory notes

    78/118

    T = 1

    =

    1 +1 +1

    1 +1 1+1 +1 1

    +1 1 1

    .

    M N/M O (N) O (MN)

    NN

    M=O Klog2 (N/K) ,

    pc[n] x (t) x (t) 1 K

    MCKlog (N/K+ 1) C1.7

    k/Na k

    K

    B

    KB

    y [j] =< x,j > N x N {j}Mj=1

  • 8/12/2019 compressive sensing theory notes

    79/118

  • 8/12/2019 compressive sensing theory notes

    80/118

    N/2

    {j} [n j]

    M

    N

    x

    x N= 256 256 M=N/50

    N= 256 256 M=N/10

    1

  • 8/12/2019 compressive sensing theory notes

    81/118

    256256

    M= 1300 50

    256 256

    M= 6500

    j j {0, 1} N

    N Wlog2N N N

    W0 = 1 Wj

    Wj = 1

    2

    Wj1 Wj1

    Wj1 Wj1

    .

    1/N O (NlogN)

    W1 = 1

    2

    1 1

    1 1

  • 8/12/2019 compressive sensing theory notes

    82/118

    W2 = 12

    1 1 1 1

    1 1 1 11 1 1 11 1 1 1

    .

    N = 2B WB I

    M N I M IWB

    WB WB D N N

    =

    1

    2

    N IWB+

    1

    2

    D.

    12

    N IWB +

    12 IWB

    {0, 1

    }

    D WB

    x y

    f(x, y, )

    f(x, y) ={f(x, y, )}

  • 8/12/2019 compressive sensing theory notes

    83/118

  • 8/12/2019 compressive sensing theory notes

    84/118

    x y

  • 8/12/2019 compressive sensing theory notes

    85/118

  • 8/12/2019 compressive sensing theory notes

    86/118

  • 8/12/2019 compressive sensing theory notes

    87/118

    = x,y=

    [1, 1] x,y [1, 2] x,y [2, 1] x,y [2, 2] x,y

    .

    K K

    K

    N

    RN

  • 8/12/2019 compressive sensing theory notes

    88/118

    F K

  • 8/12/2019 compressive sensing theory notes

    89/118

    K = 2 (1, 2) 1 2 RN

    K M

    RN

    M = CKlog (N) K < M < < N M

    RN

    RM

  • 8/12/2019 compressive sensing theory notes

    90/118

    RN

    N

  • 8/12/2019 compressive sensing theory notes

    91/118

    K

    K

    K

    N

  • 8/12/2019 compressive sensing theory notes

    92/118

    s s

    SIR = 6

    3 4

  • 8/12/2019 compressive sensing theory notes

    93/118

    2

    K M=O (KlogN)

    M

    M

    M

    x RN

    f(x)

    y = x M N p

    f x f M

  • 8/12/2019 compressive sensing theory notes

    94/118

    x

    K

    K N (KlogN)

    M

    M

    kM n2

    k

    M

  • 8/12/2019 compressive sensing theory notes

    95/118

    M

    L

    NL

    M2/(2+1) M2

  • 8/12/2019 compressive sensing theory notes

    96/118

    A

    T

    G

    C

    A

    T

    G

    C

  • 8/12/2019 compressive sensing theory notes

    97/118

    M

    N

    M N 1iM 1j N i j i,j j

    xj i yi=N

    j=1 i,jxj =ix j x y ={yi}i = 1,...,M y= x

    x

  • 8/12/2019 compressive sensing theory notes

    98/118

  • 8/12/2019 compressive sensing theory notes

    99/118

    [U+FFFD]

    [U+FFFD]

  • 8/12/2019 compressive sensing theory notes

    100/118

    [U+FFFD]

    [U+FFFD] [U+FFFD]

    [U+FFFD]

    [U+FFFD] [U+FFFD]

    [U+FFFD]

    [U+FFFD] [U+FFFD]

    [U+FFFD]

    [U+FFFD]

    [U+FFFD]

    [U+FFFD]

    [U+FFFD]

    [U+FFFD]

    [U+FFFD]

    [U+FFFD]

  • 8/12/2019 compressive sensing theory notes

    101/118

  • 8/12/2019 compressive sensing theory notes

    102/118

  • 8/12/2019 compressive sensing theory notes

    103/118

    [U+FFFD]

  • 8/12/2019 compressive sensing theory notes

    104/118

  • 8/12/2019 compressive sensing theory notes

    105/118

  • 8/12/2019 compressive sensing theory notes

    106/118

  • 8/12/2019 compressive sensing theory notes

    107/118

  • 8/12/2019 compressive sensing theory notes

    108/118

  • 8/12/2019 compressive sensing theory notes

    109/118

  • 8/12/2019 compressive sensing theory notes

    110/118

  • 8/12/2019 compressive sensing theory notes

    111/118

  • 8/12/2019 compressive sensing theory notes

    112/118

  • 8/12/2019 compressive sensing theory notes

    113/118

  • 8/12/2019 compressive sensing theory notes

    114/118

  • 8/12/2019 compressive sensing theory notes

    115/118

  • 8/12/2019 compressive sensing theory notes

    116/118

  • 8/12/2019 compressive sensing theory notes

    117/118

  • 8/12/2019 compressive sensing theory notes

    118/118