lecture7 - improved spread spectrum

Upload: shervinsh

Post on 03-Apr-2018

215 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/28/2019 Lecture7 - Improved Spread Spectrum

    1/18

    Improved Spread Spectrum:

    A New Modulation Techniquefor Robust Watermarking

    IEEE Trans. On Signal Processing, April 2003

    Multimedia Security

  • 7/28/2019 Lecture7 - Improved Spread Spectrum

    2/18

    Spread-spectrum based watermarking,

    where b is the bit to be embedded

  • 7/28/2019 Lecture7 - Improved Spread Spectrum

    3/18

    3

    u: the chip sequence (reference pattern), with zeromean and whose elements are equal to +uor -u (1-bit message coding)

    inner product:

    norm: ||x||

    Embedding: s =x +bu

    Distortion in the embedded signal:

    D = ||s - x|| = ||bu|| = ||u|| =u2

    Channel noise: y =s +n

    =

    1

    0

    1 N

    iiuixN

  • 7/28/2019 Lecture7 - Improved Spread Spectrum

    4/184

    Detection is performed by first computing

    the (normalized) sufficient statistic r:normalized correlation

    u 2

    ||u|| ||u||and estimating the embedded bit by

    b = sign( r )

    r

    b + + b + +x n~ ~

    ^

  • 7/28/2019 Lecture7 - Improved Spread Spectrum

    5/185

    We usually assume simple statistical

    models for the original signal x and theattack noise n:

    both to be samples from uncorrelated white

    Gaussian random process

    xi N(0,x2)

    ni N(0,n2)

  • 7/28/2019 Lecture7 - Improved Spread Spectrum

    6/186

    Then, it is easy to show that the sufficientstatistic ris also Gaussian, i.e.,

    r N(mr,r2)

    where

    mr= E( r ) = b b {0,1}

    x2 +n

    2

    Nu2

    r2

  • 7/28/2019 Lecture7 - Improved Spread Spectrum

    7/187

    Lets consider the casewhen b = 1.

    Then, an error occurs

    when r< 0, and therefore,the error probability p isgiven by

    for b = 1, mr= E( r) =b = 1

    ( )

    ( )

    function.errorarycomplement

    theiserfcwhere

    12

    1

    2

    1

    22

    1

    22

    1

    )1|0Pr(

    2

    2

    2

    2

    22

    2

    +

    =

    +=

    =

    =

  • 7/28/2019 Lecture7 - Improved Spread Spectrum

    8/18

    8

    The same error probability is obtained

    under the assumption that b = -1 but r> 0^

    ( mr/r )

    If we want an errorprob. Better than 10-3,

    then we need

    mr/

    r> 3

    N

    u

    2 > 9(x

    2+n

    2)

    -3

  • 7/28/2019 Lecture7 - Improved Spread Spectrum

    9/18

    9

    In general, to achieve an error probability p,

    we need

    Nu2 > 2 (erfc-1(p))2(x

    2+n2)

    One can trade the length of the chip

    sequence N with the energy of thesequenceu

    2 !!

  • 7/28/2019 Lecture7 - Improved Spread Spectrum

    10/18

    10

    Main idea:by using the encoder knowledge about thesignal x (or more precisely, the projection

    ofx on the watermark), one can enhanceperformance by modulating the energy of

    the inserted watermark to compensate forthe signal interference.

    New Approach via Improved-SS

  • 7/28/2019 Lecture7 - Improved Spread Spectrum

    11/18

    11

    We vary the amplitude of the inserted chip

    sequence by a function(x,b):

    s =x +(x,b)uwhere, as before

    x = / ||u|| : signal interference

    SS is a special case of the ISS in whichthe function is made independent ofx.

    ~

    ~

    ~

  • 7/28/2019 Lecture7 - Improved Spread Spectrum

    12/18

    12

    Linear Approximation:

    is a linear function of x

    s

    =x

    + (b-x

    )u

    The parameters and control thedistortion level and the removal of the

    carrier distortion on the detection statistics.Traditional SS is obtained by setting= 1

    and= 0.

    ~

  • 7/28/2019 Lecture7 - Improved Spread Spectrum

    13/18

    13

    With the same channel noise model as

    before, the receiver sufficient statistic is||u||

    The closer we make to 1, the more theinfluence ofx is removed from r.

    The detector is the same as in SS, i.e., thedetected bit is sign(r).

    r b + (1 -) x +n~ ~

    ~

  • 7/28/2019 Lecture7 - Improved Spread Spectrum

    14/18

    14

    The expected distortion of the new system is given by

    To make the average distortion of the new system to

    equal that of traditional SS, we force E[D]=

    u

    2

    , andtherefore

    [ ]

    [ ]2

    2

    222

    22

    1

    ~

    u

    u

    x

    u

    N

    xbE

    xsEDE

    44 344 21

    =

    +=

    =

    =

    2uN

    222

    xuN

    =

  • 7/28/2019 Lecture7 - Improved Spread Spectrum

    15/18

    15

    To compute the error probability, all we need isthe mean and variance of the sufficient statistic r.

    They are given by

    Therefore, the error probability p is

    2

    2222 )1(

    u

    xn

    r

    r

    N

    bm

    +=

    =

    { }

    +

    =

    =

    =

  • 7/28/2019 Lecture7 - Improved Spread Spectrum

    16/18

    16

    We can also write p as a function of the relativepower of the SS sequence Nu

    2/x2and

    the SNR x2/n2

    By proper selection of the parameter, the errorprobability in the proposed method can be made

    several orders of magnitude better than usingtraditional SS.

    +

    =

    +

    =2

    2

    2

    2

    2

    2

    2

    2

    )1(1

    1

    21

    21

    )1(21

    21

    SNR

    sspowererfc

    N

    erfcp

    x

    n

    x

    u

  • 7/28/2019 Lecture7 - Improved Spread Spectrum

    17/18

    17

    The three lines correspond to values to 5, 10,and 20dB SNR (with higher values having

    smaller error probability).

    Solid linesrepresent a 10-

    dB SIR and dashlines represent a7-d SIR

    SIR: Signal-to-interference ratio

  • 7/28/2019 Lecture7 - Improved Spread Spectrum

    18/18

    18

    As can be inferred from the above figure, the errorprobability varies with, with the optimum value

    usually close to 1.The expression for the optimum value for canbe computed from the error probability p byand is given by

    Note: for N large enough,opt 1 as SNR

    0=

    p

    ++

    ++=

    2

    22

    2

    2

    2

    2

    2

    2

    2

    2

    4112

    1

    x

    u

    x

    u

    x

    n

    x

    u

    x

    n

    opt

    NNN