Fourier Transform

The Fourier transform is a generalization of the complex Fourier series in the limit as L->infty. Replace the discrete A_n with the continuous F(k)dk while letting n/L->k. Then change the sum to an integral, and the equations become
f(x)=int_(-infty)^inftyF(k)e^(2piikx)dk
(1)
F(k)=int_(-infty)^inftyf(x)e^(-2piikx)dx.
(2)

Here,

F(k)=F_x[f(x)](k)
(3)
=int_(-infty)^inftyf(x)e^(-2piikx)dx
(4)

is called the forward (-i) Fourier transform, and

f(x)=F_k^(-1)[F(k)](x)
(5)
=int_(-infty)^inftyF(k)e^(2piikx)dk
(6)

is called the inverse (+i) Fourier transform. The notation F_x[f(x)](k) is introduced in Trott (2004, p. xxxiv), and f^^(k) and f^_(x) are sometimes also used to denote the Fourier transform and inverse Fourier transform, respectively (Krantz 1999, p. 202).

Note that some authors (especially physicists) prefer to write the transform in terms of angular frequency omega=2pinu instead of the oscillation frequency nu. However, this destroys the symmetry, resulting in the transform pair

H(omega)=F[h(t)]
(7)
=int_(-infty)^inftyh(t)e^(-iomegat)dt
(8)
h(t)=F^(-1)[H(omega)]
(9)
=1/(2pi)int_(-infty)^inftyH(omega)e^(iomegat)domega.
(10)

To restore the symmetry of the transforms, the convention

g(y)=F[f(t)]
(11)
=1/(sqrt(2pi))int_(-infty)^inftyf(t)e^(-iyt)dt
(12)
f(t)=F^(-1)[g(y)]
(13)
=1/(sqrt(2pi))int_(-infty)^inftyg(y)e^(iyt)dy
(14)

is sometimes used (Mathews and Walker 1970, p. 102).

In general, the Fourier transform pair may be defined using two arbitrary constants a and b as

F(omega)=sqrt((|b|)/((2pi)^(1-a)))int_(-infty)^inftyf(t)e^(ibomegat)dt
(15)
f(t)=sqrt((|b|)/((2pi)^(1+a)))int_(-infty)^inftyF(omega)e^(-ibomegat)domega.
(16)

The Fourier transform F(k) of a function f(x) is implemented as FourierTransform[f, x, k], and different choices of a and b can be used by passing the optional FourierParameters-> {​a, b} option. By default, Mathematica takes FourierParameters as (0,1). Unfortunately, a number of other conventions are in widespread use. For example, (0,1) is used in modern physics, (1,-1) is used in pure mathematics and systems engineering, (1,1) is used in probability theory for the computation of the characteristic function, (-1,1) is used in classical physics, and (0,-2pi) is used in signal processing. In this work, following Bracewell (1999, pp. 6-7), it is always assumed that a=0 and b=-2pi unless otherwise stated. This choice often results in greatly simplified transforms of common functions such as 1, cos(2pik_0x), etc.

Since any function can be split up into even and odd portions E(x) and O(x),

f(x)=1/2[f(x)+f(-x)]+1/2[f(x)-f(-x)]
(17)
=E(x)+O(x),
(18)

a Fourier transform can always be expressed in terms of the Fourier cosine transform and Fourier sine transform as

 F_x[f(x)](k)=int_(-infty)^inftyE(x)cos(2pikx)dx-iint_(-infty)^inftyO(x)sin(2pikx)dx.
(19)

A function f(x) has a forward and inverse Fourier transform such that

 f(x)={int_(-infty)^inftye^(2piikx)[int_(-infty)^inftyf(x)e^(-2piikx)dx]dk   for f(x) continuous at x; 1/2[f(x_+)+f(x_-)]   for f(x) discontinuous at x,
(20)

provided that

1. int_(-infty)^infty|f(x)|dx exists.

2. There are a finite number of discontinuities.

3. The function has bounded variation. A sufficient weaker condition is fulfillment of the Lipschitz condition

(Ramirez 1985, p. 29). The smoother a function (i.e., the larger the number of continuous derivatives), the more compact its Fourier transform.

The Fourier transform is linear, since if f(x) and g(x) have Fourier transforms F(k) and G(k), then

int[af(x)+bg(x)]e^(-2piikx)dx=aint_(-infty)^inftyf(x)e^(-2piikx)dx+bint_(-infty)^inftyg(x)e^(-2piikx)dx
(21)
=aF(k)+bG(k).
(22)

Therefore,

F[af(x)+bg(x)]=aF[f(x)]+bF[g(x)]
(23)
=aF(k)+bG(k).
(24)

The Fourier transform is also symmetric since F(k)=F_x[f(x)](k) implies F(-k)=F_x[f(-x)](k).

Let f*g denote the convolution, then the transforms of convolutions of functions have particularly nice transforms,

F[f*g]=F[f]F[g]
(25)
F[fg]=F[f]*F[g]
(26)
F^(-1)[F(f)F(g)]=f*g
(27)
F^(-1)[F(f)*F(g)]=fg.
(28)

The first of these is derived as follows:

F[f*g]=int_(-infty)^inftyint_(-infty)^inftye^(-2piikx)f(x^')g(x-x^')dx^'dx
(29)
=int_(-infty)^inftyint_(-infty)^infty[e^(-2piikx^')f(x^')dx^'][e^(-2piik(x-x^'))g(x-x^')dx]
(30)
=[int_(-infty)^inftye^(-2piikx^')f(x^')dx^'][int_(-infty)^inftye^(-2piikx^(''))g(x^(''))dx^('')]
(31)
=F[f]F[g],
(32)

where x^('')=x-x^'.

There is also a somewhat surprising and extremely important relationship between the autocorrelation and the Fourier transform known as the Wiener-Khinchin theorem. Let F_x[f(x)](k)=F(k), and f^_ denote the complex conjugate of f, then the Fourier transform of the absolute square of F(k) is given by

 F_k[|F(k)|^2](x)=int_(-infty)^inftyf^_(tau)f(tau+x)dtau.
(33)

The Fourier transform of a derivative f^'(x) of a function f(x) is simply related to the transform of the function f(x) itself. Consider

 F_x[f^'(x)](k)=int_(-infty)^inftyf^'(x)e^(-2piikx)dx.
(34)

Now use integration by parts

 intvdu=[uv]-intudv
(35)

with

du=f^'(x)dx
(36)
v=e^(-2piikx)
(37)

and

u=f(x)
(38)
dv=-2piike^(-2piikx)dx,
(39)

then

 F_x[f^'(x)](k)=[f(x)e^(-2piikx)]_(-infty)^infty-int_(-infty)^inftyf(x)(-2piike^(-2piikx)dx).
(40)

The first term consists of an oscillating function times f(x). But if the function is bounded so that

 lim_(x->+/-infty)f(x)=0
(41)

(as any physically significant signal must be), then the term vanishes, leaving

F_x[f^'(x)](k)=2piikint_(-infty)^inftyf(x)e^(-2piikx)dx
(42)
=2piikF_x[f(x)](k).
(43)

This process can be iterated for the nth derivative to yield

 F_x[f^((n))(x)](k)=(2piik)^nF_x[f(x)](k).
(44)

The important modulation theorem of Fourier transforms allows F_x[cos(2pik_0x)f(x)](k) to be expressed in terms of F_x[f(x)](k)=F(k) as follows,

F_x[cos(2pik_0x)f(x)](k)=int_(-infty)^inftyf(x)cos(2pik_0x)e^(-2piikx)dx
(45)
=1/2int_(-infty)^inftyf(x)e^(2piik_0x)e^(-2piikx)dx+1/2int_(-infty)^inftyf(x)e^(-2piik_0x)e^(-2piikx)dx
(46)
=1/2int_(-infty)^inftyf(x)e^(-2pii(k-k_0)x)dx+1/2int_(-infty)^inftyf(x)e^(-2pii(k+k_0)x)dx
(47)
=1/2[F(k-k_0)+F(k+k_0)].
(48)

Since the derivative of the Fourier transform is given by

 F^'(k)=d/(dk)F_x[f(x)](k)=int_(-infty)^infty(-2piix)f(x)e^(-2piikx)dx,
(49)

it follows that

 F^'(0)=-2piiint_(-infty)^inftyxf(x)dx.
(50)

Iterating gives the general formula

mu_n=int_(-infty)^inftyx^nf(x)dx
(51)
=(F^((n))(0))/((-2pii)^n).
(52)

The variance of a Fourier transform is

 sigma_f^2=<(xf-<xf>)^2>,
(53)

and it is true that

 sigma_(f+g)=sigma_f+sigma_g.
(54)

If f(x) has the Fourier transform F_x[f(x)](k)=F(k), then the Fourier transform has the shift property

int_(-infty)^inftyf(x-x_0)e^(-2piikx)dx=int_(-infty)^inftyf(x-x_0)e^(-2pii(x-x_0)k)e^(-2pii(kx_0))d(x-x_0)
(55)
=e^(-2piikx_0)F(k),
(56)

so f(x-x_0) has the Fourier transform

 F_x[f(x-x_0)](k)=e^(-2piikx_0)F(k).
(57)

If f(x) has a Fourier transform F_x[f(x)](k)=F(k), then the Fourier transform obeys a similarity theorem.

 int_(-infty)^inftyf(ax)e^(-2piikx)dx=1/(|a|)int_(-infty)^inftyf(ax)e^(-2pii(ax)(k/a))d(ax)=1/(|a|)F(k/a),
(58)

so f(ax) has the Fourier transform

 F_x[f(ax)](k)=|a|^(-1)F(k/a).
(59)

The "equivalent width" of a Fourier transform is

w_e=(int_(-infty)^inftyf(x)dx)/(f(0))
(60)
=(F(0))/(int_(-infty)^inftyF(k)dk).
(61)

The "autocorrelation width" is

w_a=(int_(-infty)^inftyf*f^_dx)/([f*f^_]_0)
(62)
=(int_(-infty)^inftyfdxint_(-infty)^inftyf^_dx)/(int_(-infty)^inftyff^_dx),
(63)

where f*g denotes the cross-correlation of f and g and f^_ is the complex conjugate.

Any operation on f(x) which leaves its area unchanged leaves F(0) unchanged, since

 int_(-infty)^inftyf(x)dx=F_x[f(x)](0)=F(0).
(64)

The following table summarized some common Fourier transform pairs.

functionf(x)F(k)=F_x[f(x)](k)
Fourier transform--11delta(k)
Fourier transform--cosinecos(2pik_0x)1/2[delta(k-k_0)+delta(k+k_0)]
Fourier transform--delta functiondelta(x-x_0)e^(-2piikx_0)
Fourier transform--exponential functione^(-2pik_0|x|)1/pi(k_0)/(k^2+k_0^2)
Fourier transform--Gaussiane^(-ax^2)sqrt(pi/a)e^(-pi^2k^2/a)
Fourier transform--Heaviside step functionH(x)1/2[delta(k)-i/(pik)]
Fourier transform--inverse function-PV1/(pix)i[1-2H(-k)]
Fourier transform--Lorentzian function1/pi(1/2Gamma)/((x-x_0)^2+(1/2Gamma)^2)e^(-2piikx_0-Gammapi|k|)
Fourier transform--ramp functionR(x)piidelta^'(2pik)-1/(4pi^2k^2)
Fourier transform--sinesin(2pik_0x)1/2i[delta(k+k_0)-delta(k-k_0)]

In two dimensions, the Fourier transform becomes

F(x,y)=int_(-infty)^inftyint_(-infty)^inftyf(k_x,k_y)e^(-2pii(k_xx+k_yy))dk_xdk_y
(65)
f(k_x,k_y)=int_(-infty)^inftyint_(-infty)^inftyF(x,y)e^(2pii(k_xx+k_yy))dxdy.
(66)

Similarly, the n-dimensional Fourier transform can be defined for k, x in R^n by

F(x)=int_(-infty)^infty...int_(-infty)^infty_()_(n)f(k)e^(-2piik·x)d^nk
(67)
f(k)=int_(-infty)^infty...int_(-infty)^infty_()_(n)F(x)e^(2piik·x)d^nx.
(68)
  • 0
    点赞
  • 0
    收藏
    觉得还不错? 一键收藏
  • 0
    评论

“相关推荐”对你有帮助么?

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值