28
Optimal Formation of Sensors for Statistical Target Tracking 1/27 April 24, 2015 Sung-Ho Kim (金聲浩) Korea Advanced Institute of Science and Technology (KAIST) South Korea 2015 Workshop on Combinatorics and Applications at SJTU April 21-27 Shanghai Jiao Tong University, China.

Optimal Formation of Sensors for Statistical Target Tracking€¦ · Linear State-Space Models 4/27 1 (state eq.). (measurement eq.) where is an -dimensional state vector, dimensional

  • Upload
    others

  • View
    0

  • Download
    0

Embed Size (px)

Citation preview

  • Optimal Formation of Sensors for Statistical Target Tracking

    1/27

    April 24, 2015

    Sung-Ho Kim (金聲浩) Korea Advanced Institute of Science and Technology

    (KAIST) South Korea

    2015 Workshop on Combinatorics and Applications at SJTU

    April 21-27 Shanghai Jiao Tong University, China.

  • Contents

    Back ground and problem

    Probability model of range difference

    Fisher information

    Optimal ring formation of sensors

    Numerical result

    Conclusion

    2/27

  • Multiple Missile Tracking

    3/27

    Launcher Moving Target

    Loitering Missiles (with LADAR seeker)

    cooperative sensing and

    precise target tracking

    Attack Missiles (with semi-active seeker)

    cooperative target attack

  • Linear State-Space Models

    4/27

    1 (state eq.). (measurement eq.)

    whereis an -dimensional state vector,

    dimensional oberservation vector,and Gaussian white noises.

    t t t t t

    t t t t t

    t

    t

    t t

    s a F sy b G s

    s my k

    ηε

    η ε

    + = + + = + +

    2ts − 1ts − ts3tη − 2tη − 1tη −

    2ty − ty1ty −

    2tε − tε1tε −

  • Range difference measurements by TDOA method TDOA=Time Difference Of Arrival

    5/27

  • Range difference ( ) geometry

    Range difference between

    sensors 0 and i:

    An approximation to :

    6/27

    ,0 ,i t t ir d d= −

    ,

    2,0 ,02

    2,0 2

    ,0

    1 1 2 cos( )

    cos( ) sin ( )

    i t t i

    i it i t

    t t

    ii i t i t

    t

    r d d

    d dd

    d d

    dd

    d

    θ θ

    θ θ θ θ

    = −

    = − + − −

    ≈ − − −

    ir

    ir

  • .

    7/27

    Probability model for range difference

    Let , 0,1, 2, , , where is the time of observation

    and the light speed.jz c j n

    cκ κ = × =

    2j,t( + d ,j jz N α σ )

    0 , 1, 2, , .j jy z z j n= − =

    1 2( , , , ) '.ny y yy =

    ir

  • 8/27

    Probability model for range difference ir

  • 9/27

    Probability model for range difference ir

  • Fisher information

    10/27

    [ ]

    1

    1

    1212

    1

    Let , , be random variables from ( ; ).Let ( , , ) satisfy that

    E ( ) .Then, under some conditions on ( ; ),

    E ( ) E log ( ; ) ( ) .

    (

    n

    n

    X X f xW W X X

    Wf x

    W n f X I

    I

    θ

    θ θ

    θ

    θθ

    θ θ θθ

    =

    =

    ∂ − ≥ = ∂

    1) is called the Fisher information for in , , .nX Xθ θ

  • Fisher information for target location on a plane (1/2)

    11/27

  • 12/27

    Fisher information for target location on a plane (2/2)

  • 13/27

    Fisher information for target location on a plane (2/2)

  • Target location problem in 3-D

    14/27

    ( , , )t t tt d φ θ=

    ( , , )i i ii d φ θ =

    • Target location

    • location of sensor i

    • location of sensor 0 = origin

  • Range difference between sensors 0 and i:

    15/27

    2

    , 21 1 2 sin sin cos( ) 2 cos cosi i i

    i t t i t t i i t t it t t

    d d dr d d dd d d

    φ φ θ θ φ φ

    = − = − − − − +

    Target location problem in 3-D

    i 0

  • 16/27

    Target location problem in 3-D

  • 17/27

    Target location problem in 3-D

  • 18/27

    Target location problem in 3-D

  • 19/27

    Geometric Interpretation of Iφφ

  • Optimal ring formation

    20/27

  • 21/27

    Optimal ring formation

  • 22/27

    Optimal ring formation

  • 23/27

    Optimal ring formation

  • 24/27

    Optimal ring formation

    ˆ ˆ ˆ( ) ( )( ) '.c t c c c cVar t E t t t t= − −

  • 25/27

    Optimal ring formation

  • MSE from (24, K) ring formations

    26/27

    Optimal ring formation

  • Conclusions The ring formation renders the estimators stochastically independent. Optimal sensor formations are half-and-half arrangement

    between the center and the outer-most ring.

    (n,4)-ring performs better to worse than (n,3)-ring as approaches from either direction.

    27/27

    , ,t t td φ θ

    tθ / 4π

  • Thanks (謝謝)

    Optimal Formation of Sensors for Statistical Target TrackingContentsMultiple Missile TrackingLinear State-Space ModelsRange difference measurements by TDOA methodRange difference ( ) geometry Slide Number 9Fisher information Fisher information for target location on a plane (1/2)Fisher information for target location on a plane (2/2)Fisher information for target location on a plane (2/2)Target location problem in 3-DTarget location problem in 3-DSlide Number 16Target location problem in 3-DTarget location problem in 3-DSlide Number 19Optimal ring formationOptimal ring formationOptimal ring formationOptimal ring formationOptimal ring formationOptimal ring formationOptimal ring formationConclusionsSlide Number 28