next up previous
Next: Equispaced knots Up: NDFT Previous: NDFT

Direct NDFT

The first problem to be adressed can be regarded as a matrix vector multiplication. For a finite number of given Fourier coefficients $ \hat f_{\mbox{\boldmath\scriptsize {${k}$}}} \in \mathbb{C}\;(\mbox{\boldmath {${k}$}}\in I_{\mbox{\boldmath\scriptsize {${N}$}}})$ one wants to evaluate the trigonometric polynomial

$\displaystyle f\left(\mbox{\boldmath {${x}$}}\right) := \sum_{\mbox{\boldmath\s...
...e {i}}} \mbox{\boldmath\scriptsize {${k}$}}\mbox{\boldmath\scriptsize {${x}$}}}$ (2.1)

at given non equispaced knots $ {x}$$ _j \in \mathbb{T}^d$. Thus, our concern is the evaluation of

$\displaystyle f_j = f\left(\mbox{\boldmath {${x}$}}_j\right) := \sum_{\mbox{\bo...
...iptsize {${k}$}}\mbox{\boldmath\scriptsize {${x}$}}_j} \qquad (j=0,\hdots,M-1).$    

In matrix vector notation this reads as

$\displaystyle \mbox{\boldmath {${f}$}}$$\displaystyle =$$\displaystyle \mbox{\boldmath {${A}$}}$   $\displaystyle \mbox{\boldmath {${\hat f}$}}$ (2.2)

where

$\displaystyle \mbox{\boldmath {${f}$}}$$\displaystyle :=\left(f_j\right)_{j=0,\hdots,M-1},$   $\displaystyle \mbox{\boldmath {${A}$}}$$\displaystyle :=\left({\rm e}^{-2\pi{\mbox{\scriptsize {i}}} \mbox{\boldmath\sc...
...ht)_{\mbox{\boldmath\scriptsize {${k}$}}\in I_{\mbox{\boldmath\tiny {${N}$}}}}.$    

As already mentioned, all 'vec'-operation are omitted, and the matrix vector product is defined by $ {A}$   $ {\hat f}$$ :={\rm Vec}\left(\mbox{\boldmath {${A}$}}\right) {\rm vec}\left(\mbox{\boldmath {${\hat f}$}}\right)$. The straight forward algorithm of this matrix vector product, which is called ndft, takes $ {\cal O}\left(M N_{\text{\tiny $\Pi$}}\right)$ arithmetical operations.

Related matrix vector products are the adjoint ndft

$\displaystyle \mbox{\boldmath {${\hat f}$}}$$\displaystyle =$$\displaystyle \mbox{\boldmath {${A}$}}$$\displaystyle ^{\rm H}$   $\displaystyle \mbox{\boldmath {${f}$}}$$\displaystyle , \qquad \hat f_{\mbox{\boldmath\scriptsize {${k}$}}}=\sum_{j=0}^...
...i}}} \mbox{\boldmath\scriptsize {${k}$}}\mbox{\boldmath\scriptsize {${x}$}}_j},$    

the conjugated ndft

$\displaystyle \mbox{\boldmath {${f}$}}$$\displaystyle =\overline {\mbox{\boldmath {${A}$}}} \mbox{\boldmath {${\hat f}$...
...i}}} \mbox{\boldmath\scriptsize {${k}$}}\mbox{\boldmath\scriptsize {${x}$}}_j},$    

and the transposed ndft

$\displaystyle \mbox{\boldmath {${\hat f}$}}$$\displaystyle =$$\displaystyle \mbox{\boldmath {${A}$}}$$\displaystyle ^{\rm T}$   $\displaystyle \mbox{\boldmath {${f}$}}$$\displaystyle , \qquad \hat f_{\mbox{\boldmath\scriptsize {${k}$}}}=\sum_{j=0}^...
...i}}} \mbox{\boldmath\scriptsize {${k}$}}\mbox{\boldmath\scriptsize {${x}$}}_j},$    

where $ {A}$$ ^{\rm H}=\overline {\mbox{\boldmath {${A}$}}}^{\rm T}$.


next up previous
Next: Equispaced knots Up: NDFT Previous: NDFT
Stefan Kunis 2004-09-03