A generic interpolation program

Module interp1 implements a generic 1-D interpolation program. Analogous modules also exist for the 2-D and 3-D cases. The module includes the initialization (constructor) subroutine interp1_init, which takes an array coord to define the data coordinates and the SEPlib-style values o1, d1, and n1 to define the model grid. Additionally, it accepts an external function interp to compute the interpolation filter W (x,n). The size of the filter is defined by the integer parameter nfilt. An example of a function with the interface of interp is lagrange , which implements the local Lagrange interpolation. The initialization program allocates and computes three private arrays: the integer array x that defines the mapping from the data coordinates to the model grid, the logical array m that masks the data values outside the grid, and the real-value array w that stores coefficients of the interpolation filter. Computing these arrays outside the actual interpolation program interp1_op not only complies with the object-oriented design of linear operators Fomel and Claerbout (1996), but also significantly improves the efficiency when the interpolation operator is applied more than once (e.g., in iterative least-square optimization.) The arrays are deallocated by the ``destructor'' program interp1_close.

 

function lagrange (x, w) result (stat) 
  integer             :: stat
  real, intent (in)   :: x
  real, dimension (:) :: w

integer             :: i, j, nf
  real                :: f, xi

nf = size (w)
  do i = 1, nf
     f = 1.
     xi = x + 1. - i
     do j = 1, nf
        if (i /= j) f = f * (1. + xi / (i - j))
     end do
     w (i) = f
  end do
  stat = 0
end function lagrange

 

module interp1	
  use adj_mod
  integer,                               private :: nd, nf
  integer, dimension (:),   allocatable, private :: x
  logical, dimension (:),   allocatable, private :: m
  real,    dimension (:,:), allocatable, private :: w

contains
  subroutine interp1_init (coord, o1, d1, n1, interp, nfilt)
    real, dimension (:), intent (in) :: coord
    real,                intent (in) :: o1, d1
    integer,             intent (in) :: n1, nfilt
    interface
       integer function interp (x, w)
         real, intent (in)   :: x
         real, dimension (:) :: w
       end function interp
    end interface
    integer                          :: id, ix, stat
    real                             :: rx

nd = size (coord) ; nf = nfilt
    if (.not. allocated (x)) allocate (x (nd), m (nd), w (nf,nd))
    do id = 1, nd ; rx = (coord (id) - o1)/d1 ; ix = rx 
                    rx = rx - ix ; x (id) = ix + 1 - 0.5*nf
       m (id) = .true. ; w (:, id) = 0. 
       if ((x (id) + 1 >= 1) .and. (x (id) + nf <= n1)) then
          m (id) = .false. ; stat = interp (rx, w (:,id))
       end if
    end do
  end subroutine interp1_init

function interp1_op (adj, add, mod, ord) result (stat)
    integer              :: stat
    logical, intent (in) :: adj, add
    real, dimension (:)  :: mod, ord
    integer              :: id, i1, i2

call adjnull (adj, add, mod, ord)
    do id = 1, nd ; if (m (id)) cycle 
       i1 = x (id) + 1 ; i2 = x (id) + nf
       if (adj) then
          mod (i1:i2) =   mod (i1:i2) + w (:,id)  * ord (id)
       else
          ord (id) = sum (mod (i1:i2) * w (:,id)) + ord (id)
       end if
    end do
    stat = 0
  end function interp1_op

subroutine interp1_close ()
    deallocate (x, m, w)
  end subroutine interp1_close
end module interp1

To illustrate the forward interpolation operator, I chose a regularly sampled chirp function $\exp{-t^2/\sigma^2} \cos{\omega t}$ as the input model (Figure 2). Figures 3, 4, and 5 show the result of forward interpolation with different interpolators.

 

alias Figure 2 The input is a regularly sampled chirp function.

 

bin
Figure 3 Left is the result of forward nearest-neighbor interpolation; right, linear interpolation.

 

lgg
Figure 4 Result of forward Lagrange interpolation. Left is the 4-point interpolator; right, 10-point. Increasing the number of coefficients may lead to instabilities.

 

sinc
Figure 5 Result of forward Muir interpolation. Left is the 4-point interpolator; right, 10-point.