The spatial frequency domain
MIT 2.71/2.710 Optics 10/31/05 wk9-a-1 Recall: plane wave propagation
x path delay increases linearly with x π λ ⎛ λ x θ E0 exp⎜i2 sin + ⎝ π z ⎞ i2 λcosθ ⎟ ⎠ θ z=0 z
plane of observation
MIT 2.71/2.710 Optics 10/31/05 wk9-a-2 Spatial frequency ⇔ angle of propagation?
x
λ electric field
θ z=0 z
plane of observation
MIT 2.71/2.710 Optics 10/31/05 wk9-a-3 Spatial frequency ⇔ angle of propagation?
x
λ
θ z=0 z
plane of observation
MIT 2.71/2.710 Optics 10/31/05 wk9-a-4 Spatial frequency ⇔ angle of propagation?
The cross-section of the optical field with the optical axis is a sinusoid of the form π θ ⎛ sin ⎞ E0 exp⎜i2 λ x +φ0 ⎟ ⎝ ⎠ i.e. it looks like π sinθ E exp()i2 u x +φ where u ≡ 0 0 λ
is called the spatial frequency
MIT 2.71/2.710 Optics 10/31/05 wk9-a-5 2D sinusoids
corresp. phasor E0 cos[]2π ()ux + vy E0 exp[i2π (ux + vy)] y
x
MIT 2.71/2.710 Optics 10/31/05 wk9-a-6 2D sinusoids
corresp. phasor E0 cos[]2π ()ux + vy E0 exp[i2π (ux + vy)] y
x
min max
MIT 2.71/2.710 Optics 10/31/05 wk9-a-7 2D sinusoids
corresp. phasor E0 cos[]2π ()ux + vy E0 exp[i2π (ux + vy)] y
1/u
1/v x
min max
MIT 2.71/2.710 Optics 10/31/05 wk9-a-8 2D sinusoids
corresp. phasor E0 cos[]2π ()ux + vy E0 exp[i2π (ux + vy)] u tanψ = y v 1 Λ = 2 2 u + v Λ x ψ min max
MIT 2.71/2.710 Optics 10/31/05 wk9-a-9 Spatial (2D) Fourier Transforms
MIT 2.71/2.710 Optics 10/31/05 wk9-a-10 The 2D2D Fourier integral
(aka inverseinverse FourierFourier transformtransform) g()x, y = ∫ G(u,v) e+i2π (ux+vy) dudv
superposition sinusoids complex weight, expresses relative amplitude (magnitude & phase) of superposed sinusoids MIT 2.71/2.710 Optics 10/31/05 wk9-a-11 The 2D2D Fourier transform
The complex weight coefficients G(u,v), aka FourierFourier transformtransform of g(x,y) are calculated from the integral G()u,v = ∫ g(x, y) e−i2π (ux+vy) dxdy
(1D so we can draw it easily ... ) g(x) Re[e-i2πux] Re[G(u)]=∫ x dx
MIT 2.71/2.710 Optics 10/31/05 wk9-a-12 2D2D Fourier transform pairspairs
(from Goodman, Introduction to Fourier Optics, page 14)
MIT 2.71/2.710 Optics 10/31/05 wk9-a-13 Space and spatial frequency representations
SPACE DOMAIN g(x,y)
G()u,v = ∫ g x, y ( e−i2π )(ux+vy) dxdy g(x, y)= ∫ G(u,v) e+i2π (ux+vy) dudv
2D2D FourierFourier transformtransform 2D2D FourierFourier integralintegral aka inverseinverse 2D2D FourierFourier transformtransform SPATIAL FREQUENCY DOMAIN G(u,v)
MIT 2.71/2.710 Optics 10/31/05 wk9-a-14 Periodic Grating /1: vertical
y v
x u
Frequency Space (Fourier) domain y v domain
x u
MIT 2.71/2.710 Optics 10/31/05 wk9-a-15 Periodic Grating /2: tilted
y v
x u
Frequency Space (Fourier) domain y v domain
x u
MIT 2.71/2.710 Optics 10/31/05 wk9-a-16 Superposition: multiple gratings ++
Frequency Space (Fourier) domain domain
MIT 2.71/2.710 Optics 10/31/05 wk9-a-17 Superpositions: spatial frequency representation
Frequency Space (Fourier) domain domain
MIT 2.71/2.710 Optics 10/31/05 wk9-a-18 Superpositions: spatial frequency representation
space domain spatial frequency domain g()x, y G(u,v)= ℑ{g(x, y)} MIT 2.71/2.710 Optics 10/31/05 wk9-a-19 Fourier transform properties /1
• Fourier transforms and the delta function ℑ{}δ ()x, y =1
ℑ{}exp[]i2π (u0 x + v0 y) = δ (u − u0 )δ (v − v0 ) • Linearity of Fourier transforms
if ℑ{}g1()x, y = G1(u,v)() and ℑ{g2 x, y }= G2 (u,v) {}()()() () then ℑ a1g1 x, y + a2 g2 x, y = a1G1 u,v + a2G2 u,v
for any pair of complex numbers a1, a2.
MIT 2.71/2.710 Optics 10/31/05 wk9-a-20 Fourier transform properties /2 Let ℑ{}g()x, y = G(u,v) • Shift theorem (space → frequency)
ℑ{}g()x − x0 , y − y0 = G(u,v)exp[− i2π (ux0 + vy0 )]
• Shift theorem (frequency → space)
ℑ{}g()x, y exp[i2π (ux0 + vy0 )] = G(u − u0 ,v − v0 )
• Scaling theorem 1 ⎛ u v ⎞ ℑ{}g()ax,by = G⎜ , ⎟ ab ⎝ a b ⎠
MIT 2.71/2.710 Optics 10/31/05 wk9-a-21 Modulation in the frequency domain: the shift theorem
Frequency Space (Fourier) domain domain
MIT 2.71/2.710 Optics 10/31/05 wk9-a-22 Size of object vs frequency content: the scaling theorem
Frequency Space (Fourier) domain domain
MIT 2.71/2.710 Optics 10/31/05 wk9-a-23 Fourier transform properties /3 Let ℑ{}f ()x, y = F(u,v) and{} ℑ h(x, y) = H (u,v) Let g()x, y = ∫ f (x′, y′)⋅h(x − x′, y − y′) dx′dy′ • Convolution theorem (space → frequency) ℑ{}g()x, y = F(u,v)⋅ H (u,v)
Let Q()u,v = ∫ F(u′,v′)⋅ H (u − u′,v − v′) du′dv′ • Convolution theorem (frequency → space) Q()u,v = ℑ{f (x, y)⋅h(x, y)}
MIT 2.71/2.710 Optics 10/31/05 wk9-a-24 Fourier transform properties /4 Let ℑ{}f ()x, y = F(u,v) and{} ℑ h(x, y) = H (u,v) Let g()x, y = ∫ f (x′, y′)⋅h(x′ − x, y′ − y) dx′dy′ • Correlation theorem (space → frequency) ℑ{}g()x, y = F(u,v)⋅ H * (u,v) Let Q()u,v = ∫ F(u′,v′)⋅ H (u′ − u,v′ − v) du′dv′
• Correlation theorem (frequency → space) Q()u,v = ℑ{f (x, y)⋅h* (x, y)}
MIT 2.71/2.710 Optics 10/31/05 wk9-a-25 Spatial frequency representation
space domain Fourier domain 3 sinusoids (aka spatial frequency domain) MIT 2.71/2.710 Optics 10/31/05 wk9-a-26 Spatial filtering
space domain Fourier domain 2 sinusoids (1 removed) (aka spatial frequency domain) MIT 2.71/2.710 Optics 10/31/05 wk9-a-27 Spatial frequency representation
space domain Fourier domain g()x, y (aka spatial frequency domain) MIT 2.71/2.710 Optics 10/31/05 wk9-a-28 G(u,v)= ℑ{g(x, y)} Spatial filtering (low–pass)
removed high-frequency content
space domain Fourier domain (aka spatial frequency domain) MIT 2.71/2.710 Optics 10/31/05 wk9-a-29 Spatial filtering (band–pass)
removed high- and low-frequency content
space domain Fourier domain (aka spatial frequency domain) MIT 2.71/2.710 Optics 10/31/05 wk9-a-30 Example: optical lithography
original pattern mild (“nested L’s”) low-pass filtering Notice: (i) blurring at the edges (ii) ringing MIT 2.71/2.710 Optics 10/31/05 wk9-a-31 Example: optical lithography
original pattern severe (“nested L’s”) low-pass filtering Notice: (i) blurring at the edges (ii) ringing MIT 2.71/2.710 Optics 10/31/05 wk9-a-32 2D2D linear shift invariant systems
input output convolution with impulse response gi(x,y) go(x’, y’) inverse Fourier g ()x′, y′ = g (x, y) h(x′ − x, y′ − y) dxdy inverse Fourier o ∫ i transform transform
LSI Fourier Fourier transform transform multiplication with transfer function
Gi(u,v) Go(u,v)
Go (u,v)= Gi (u,v)( H u,v )
MIT 2.71/2.710 Optics 10/31/05 wk9-a-33 2D2D linear shift invariant systems SPACE DOMAIN input output convolution with impulse response gi(x,y) go(x’, y’) inverse Fourier g ()x′, y′ = g (x, y) h(x′ − x, y′ − y) dxdy inverse Fourier o ∫ i transform transform
LSI Fourier Fourier transform transform multiplication with transfer function
Gi(u,v) Go(u,v)
Go (u,v)= Gi (u,v)( H u,v )
SPATIAL FREQUENCY DOMAIN MIT 2.71/2.710 Optics 10/31/05 wk9-a-34 2D2D linear shift invariant systems
input output convolution with impulse response gi(x,y) go(x’, y’) inverse Fourier g ()x′, y′ = g (x, y) h(x′ − x, y′ − y) dxdy inverse Fourier o ∫ i transform transform
LSI h(x,y), H(u,v) Fourier Fourier transform transform are 2D F.T. pair multiplication with transfer function
Gi(u,v) Go(u,v)
Go (u,v)= Gi (u,v)( H u,v )
MIT 2.71/2.710 Optics 10/31/05 wk9-a-35 Sampling space and frequency
δx :pixel size δu :frequency resolution
umax
space ∆x :field size spatial domain frequency Nyquist 1δ 1 umax = δu = domain relationships: 2 x 2∆x MIT 2.71/2.710 Optics 10/31/05 wk9-a-36 The Space–Bandwidth Product
Nyquist relationships:
1 from space → spatial frequency domain: u = max 2δx
∆x 1 from spatial frequency → space domain: = 2 2δu
∆x 2u δ = max ≡ N : 1D Space–Bandwidth Product (SBP) x δu aka number of pixels in the space domain
2D SBP ~ N 2
MIT 2.71/2.710 Optics 10/31/05 wk9-a-37 SBP: example
space domain Fourier domain δx =1 ∆x =1024 (aka spatial frequency domain) MIT 2.71/2.710 Optics −4 10/31/05 wk9-a-38 umax = 0.5 δu =1/1024 = 9.765625×10