EE 511: Introduction to Fourier Optics and Image Understanding Volume 1 I. History and Background II. Fourier Transforms and Linear Systems Dwight L. Jaggard University of Pennsylvania 308 Moore <jaggard@seas.upenn.edu> 215.898.4411 ©2000, D. L. Jaggard EE 511 1 Course Goal Understand the fundamentals of physical and ray optics and their application to current science and technology ©2000, D. L. Jaggard EE 511 2 What good is optics? ©2000, D. L. Jaggard EE 511 3 Course Outline (I) I. History and Background A. History B. Types of Optics C. Applications of Physical Optics II. Fourier Transforms and Linear Systems III. Scalar Diffraction Theory (Physical Optics) IV. Fresnel and Fraunhofer Approximations V. Vector Diffraction Theory VI. Geometrical and Ray Optics ©2000, D. L. Jaggard EE 511 4 Course Outline (II) VII. Properties of Lenses VIII. Coherent and Incoherent Imaging IX. Partial Coherence Theory X. Special Topics (topics selected according to time and interest of class) A. Image classification and understanding B. Non-destructive evaluation and testing C. Fractal antennas and arrays D. “Electromagnetic bullets” or focused beams E. Inverse problems ©2000, D. L. Jaggard EE 511 5 Administrative Information l Contacts l Dwight Jaggard , instructor l Lois Clearfield, secretary (appointments and handouts) <jaggard@seas. upenn.edu> or 215.898.4411 <lois@ee.upenn.edu> or 215.898.8241 l l Office Hours: 4:30 - 5:30 W Review Sessions: as needed ©2000, D. L. Jaggard EE 511 6 Course Information l Grades l l l l l l l Homework Midterm Final Mini-Project ~15 - 20% ~35 - 40% ~35 - 40% ~10% Late homework not accepted Midterm Exam: Wednesday, March 7 Guidelines on collaboration ©2000, D. L. Jaggard EE 511 7 Mini-Project l Idea: Take topic related to course and present topic to class Work in small groups Turn in presentation (Power Point) plus paper on topic l Talks given last two weeks of class l Paper due April 30 ©2000, D. L. Jaggard EE 511 8 Potential Mini -Project Topics - I l l l l l Image and pattern classification Signal reconstruction (from limited data) Non-uniform or 2-D sampling Rough surface scattering Electromagnetic scattering (physical optics with polarization or exact methods) l l l Knife edge diffraction for both polarizations Low frequency diffraction by apertures Optical computing/neural nets ©2000, D. L. Jaggard EE 511 9 Potential Mini -Project Topics - II l l l l l l l l l l Diffraction by fractals Zernike polynomials and optics Diffuse beam propagation/imaging Higher-order Gaussian beam propagation Focused beams/“diffractionless” propagation Applications of partial coherence theory Ray optics and lens design Multilayers: application & design Spectroscopy Antenna radiation ©2000, D. L. Jaggard EE 511 10 Potential Mini -Project Topics - III l l l l l l l l l Phase retrieval problem Holography Inverse problems Tomography & radon transform Ultrasound imaging X-ray diffraction/crystallography MRI Adaptive optics Wavelets ©2000, D. L. Jaggard EE 511 11 Links l Course website: http://www.seas. upenn.edu/~ee511 (homework posted here) l Photonics information & news l Some interesting sites for physical optics www.optics.org www.opticalimaging.org/fourieroptics.html http://dukemil.egr.duke.edu/Ultrasound/k space/bme265.htm http://wyant.optics.arizona.edu/fresnelZones/fresnelZone s.htm http://hyperphysics.phy astr.gsu.edu/hbase/phyopt/diffracon.html#c1 ©2000, D. L. Jaggard EE 511 12 Related Journals l l l l l l l l l l l l l Journal of the Optical Society of America – A Optics Letters Optics Communications Applied Optics Optical Engineering Journal of Lightwave Technology Journal of Optics A: Pure and Applied Optics Optik Journal of Modern Optics Optica Acta Applied Physics Letters Applied Physics B: Lasers and Optics Optics Express (online journal) ©2000, D. L. Jaggard EE 511 13 I. History and Background A. History B. Types of Optics C. Applications of Physical Optics ©2000, D. L. Jaggard EE 511 14 History (I) l Classical Times & Greek Philosophers l l l Empedocles (circa 490-430 B.C.) Euclid (circa 300 B.C.) The Golden 1600’s l Descartes (1596-1650) l l Considered the nature of light Light was pressure transmitted through the aether l Galileo (1564-1642) l Snell (1621) l l Experimental methods Refraction of light at interface ©2000, D. L. Jaggard EE 511 15 History (II) l More 1600’s l Fermat (1601-1665) l Father Grimaldi (1618-1663) l l l l l “Principle of Least Time” Refraction laws verified First noticed “diffraction” Note: diffraction is the bending of light not caused by refraction Newton (1642-1727) l l l Discovered basic qualities of color White light could be split up into colors Experiments with prisms and light and “refrangibility ” or bending of light at an interface ©2000, D. L. Jaggard EE 511 16 History (III) l Still More 1600’s l Huygens (1629-1695) l l l l Wave propagation of light Polarization of light Laws of reflection and refraction Progress in the 1700’s l Young (1773-1829) l Fresnel (1788-1827) l l l l l l l Wave theory Interference (colors of thin films) Confirmed wave theory of propagation and diffraction Influence of earth’s motion of light propagation Interference of polarized rays of light (light no longitudinal) Reflection and polarization Cause of dispersion ©2000, D. L. Jaggard EE 511 17 History (IV) l The Maxwell Era l Faraday (1791-1867) l Maxwell (1831-1879) l l l l l l Experiments in electricity and magnetism Work independent of optics experiments Theoretically unified electricity and magnetism Showed possibility of electromagnetic waves propagating with velocity that could be calculated Electrostatics, magnetostatics , induction, EM waves and optics unified under single theory Lord Rayleigh (scientific work 1899-1920) l l l Investigated waves propagation and scattering Examined scattering from small particles Studied wave interactions with periodic structures ©2000, D. L. Jaggard EE 511 18 History (V) l Atomic Nature of Light - The Beginning l Fraunhofer (1787-1826) l Kirchhoff (1824-1887) l l l Plank, Bohr and Einstein (early 1900’s) l l l l l Discovered absorption lines in the solar spectrum Experimentally measured absorption lines of solar spectrum Quantum theory makes inroads Applications of quantum mechanics to atomic structure and line spectra (materials have quantized atomic systems) Photons postulated Certain effects (e.g., photo -electric effect) explained only by photons Dirac (1927) l l Field quantization (electromagnetic fields are quantized) Quantum optics ©2000, D. L. Jaggard EE 511 19 What are the “brands” of optics? ©2000, D. L. Jaggard EE 511 20 Types of Optics l Ray or geometrical optics l Wave/physical/Fourier optics Scalar theory (no polarization) Vector or EM theory (polarization) l Beam optics l l l Statistical optics Optics of atomic systems/materials l Quantum optics l ©2000, D. L. Jaggard EE 511 21 Hierarchy of Optics Quantum Optics Electromagnetic (Vector) Optics Scalar Wave Optics Geometrical (Ray) Optics ©2000, D. L. Jaggard EE 511 22 Applications of Physical Optics l Remote sensing & inverse scattering l Imaging & image systems l Image processing, pattern discrimination and classification l Holography and non -destructive evaluation and testing (NDE & NDT) Rough surface scattering l Antenna and array design l Spectroscopy l l l Inteferometry Optical computing ©2000, D. L. Jaggard EE 511 23 Course Outline I. II. History and Background Fourier Transforms and Linear Systems III. Scalar Diffraction Theory (Physical Optics) IV. Fresnel and Fraunhofer Approximations V. VI. Vector Diffraction Theory Geometrical and Ray Optics VII. Properties of Lenses VIII. Coherent and Incoherent Imaging IX. X. Partial Coherence Theory Special Topics ©2000, D. L. Jaggard EE 511 24 II. Fourier Transforms and Linear Systems A. B. C. D. E. F. G. H. I. Requirements “Inventing” the Fourier Transform Dirac Delta Function Use of the F.T. in Optics F.T. Properties Some Useful Transforms Two-Dimensional F.T. More Useful Transforms Sampling ©2000, D. L. Jaggard EE 511 25 A. Requirements l To use Fourier Transforms (F.T.) there are requirements on: l System l Signal ©2000, D. L. Jaggard EE 511 26 System Requirements l To use F.T. the system must be: l Linear l Nonlinear systems often use specialized methods unique to each system l No general theory exists l Time invariant l Memoryless ©2000, D. L. Jaggard EE 511 27 Signal Requirements l To use F.T. the signal g(t) must: l Satisfy ∞ ∫ g(t) dt exists −∞ l Must have finite number of discontinuities e.g., cannot be the function +1 for t rational g( t) = −1 for t irrational l Have a finite number of max and min e.g., cannot be the function g(t) = sin(t−1) ©2000, D. L. Jaggard EE 511 28 More on Signal Requirements l Fourier transforms for signals not satisfying three conditions can often be found: For signals whose absolute value has infinite area, one can use a damping function and take the limit l For signals with discontinuities impose a Lipschitz condition l Signals from real systems are most often well-behaved l ©2000, D. L. Jaggard EE 511 29 How can we discover the Fourier Transform? ©2000, D. L. Jaggard EE 511 30 One Method Joseph ©2000, D. L. Jaggard Fourier (1768-1839) EE 511 31 B. “Inventing” the F.T. l The spectrum of g(t) is the amount of each frequency f contained in g(t) l Mathematically the F.T. is the graph of the spectrum of g(t) l Need a way to find out how much of each frequency f is in g(t) ©2000, D. L. Jaggard EE 511 32 Inner Product and F.T. l The inner product is a measure of how much a signal is like another signal l Define the inner product of g(t) and ∞ h(t) as ∆ * < g(t)h(t) > = ∫ g(t )h (t)dt −∞ Clearly this is max when g(t) = h(t) l If inner product is zero, g(t) and h(t) are orthogonal (just like vector dot product) l ©2000, D. L. Jaggard EE 511 33 Phasors and Signal of “Pure Frequency” l l l l An example of a signal of “pure frequency” is exp(j2πft) This is a rotating “phasor” of unit amplitude and angle 2πft and so rotates counterclockwise for f > 0 Real and imaginary parts of this phasor are cos(2πft) and sin(2πft) A phasor exp(-j2πft) simply rotates clockwise for f > 0 [gives meaning to “negative frequency’] ©2000, D. L. Jaggard EE 511 34 Signal of “Pure Frequency” and g(t) l Let h(t) = exp(j2πft) to find how much of frequency “f” is in signal “g(t)” through the inner product l Define the F.T. as: F {g(t )} =< g(t)e j 2πft ∞ >= ∫ g(t)e − j 2πft dt −∞ ©2000, D. L. Jaggard EE 511 35 More on F.T. of g(t) l Fourier transform of g(t) gives frequency-domain function G(f) which provides the graph of the spectrum of g(t) ∞ F {g(t )} = G( f ) =< g(t)e j 2πft >= ∫ g(t)e − j2 πft dt −∞ l l Note: the “spectrum” may also refer to |G(f)| 2 - so beware! Sometimes all we have is |G(f)| 2 which does not give unique g(t) [“phase retrievel problem”] ©2000, D. L. Jaggard EE 511 36 F.T. and Its Inverse l Fourier transform is not unique l Only combined pair of the Fourier transform F{g(t)} = G(f) and its associated inverse transform F–1 {G(f)} = g(t) are unique where F l −1 F {g(t)} = g(t) Many minor variations in notation (e.g., factors of 2 π, changes in sign of exponent) ©2000, D. L. Jaggard EE 511 37 Celebrated F.T. Pair l The Fourier transform pair used here is: F {g(t )} = G( f ) = ∞ ∫ g(t)e − j 2πft dt −∞ ∞ F −1 {G( f ) } = g(t) = ∫ G( f )e + j 2 πftdf −∞ l Note: we have only defined the first equation and postulated the second l Here f and t are conjugate variables ©2000, D. L. Jaggard EE 511 38 Notes on F.T. Pair l We use the shorthand g(t ) ⇔ G( f ) to indicate a function g(t) and its transform G(f) l There are many other F.T. conventions that change signs in the exponent and/or change the position of the factor (2 π) l If g(t) is discontinuous, F.T. construction will not give discontinuity but will converge to average value at discontinuity ©2000, D. L. Jaggard EE 511 39 F.T. Inversion Formula ∞ g(t ) = ∫ G( f )e + j2 πft df Assume inversion formula is correct, now check −∞ = ∞ ∞ ∫ ∫ [g(t' )e − j 2 πft' + j 2πft dt']e df −∞ −∞ = ∞ ∞ ∫ ∫ [g(t' )e + j2 πf (t − t ) ' dt' df −∞ −∞ ∞ e + j2 πf (t −t ) − e − j2π f (t − t ) = lim ∫ g(t' ) dt' j2π (t − t' ) f →∞ −∞ = ' ∞ ' sin[ 2πf (t − t' )] dt' π (t − t' ) ∫ g(t' ) lim −∞ f →∞ ∞ = g(t' ) t' = t sin[ 2πf (t − t' )] dt' π (t − t' ) ∫ lim −∞ f→∞ [sin(t)/t] function is sharply peaked and “pulls out” or “sifts” g(t’) at t=t’ Integral has a value of unity = g(t ) ©2000, D. L. Jaggard EE 511 40 Sine and Cosine Integrals t Si(t) = ∫ sin(t' ) dt' t' Si(t) = π 2 0 lim t→∞ ∞ ∞ sin(t' ) sin(2 πft' ) dt' = ∫ dt' = π t' t' −∞ −∞ ∫ t Ci(t) = γ + ln(t ) + ∫ 0 cos(t ' ) − 1 dt' t' where γ ≈ 0.5772156649 ... is Euler's constant lim t →∞ Ci(t) = 0 ©2000, D. L. Jaggard EE 511 41 C. Dirac Delta Function l Definition l Candidates l Some useful relations ©2000, D. L. Jaggard EE 511 42 Delta Function Definition ∞ This is not your usual function ∫ δ (t)dt = 1 −∞ and δ (t) = 0 l for t≠0 Note: l l l Dirac delta function δ(t) is highly peaked where its argument is zero Its “weight” or “strength” is defined by its integral (unity) This is a “generalized function” or “distribution” ©2000, D. L. Jaggard EE 511 43 Candidates for δ(t) l Functions that are sharply peaked and have an area of unity l Example: define the rectangle function or “rect function” of width T as rect(t/T): 1 for rect(t / T ) = 0 for ©2000, D. L. Jaggard 1 2 1 t /T > 2 t /T < EE 511 44 Delta Function and T–1 rect(t/T) l Normalize rect function to give unit area and take limit: δ(t) = lim T → 0 1 1 T for t / T < 1 2 rect(t / T ) = 1 T 0 for t / T > 2 T–1rect(t/T) T 1/T Take limit T–> 0 t ©2000, D. L. Jaggard EE 511 45 Candidates for δ(t) δ (t) = δ (t) = δ (t) = δ (t) = δ (t) = lim T→ 0 1 rect(t / T ) T lim α →∞ lim α →∞ lim α →∞ lim α →0 l sin(αt) πt α 2 exp(−αt ) π α exp( −α t ) 2 l α 2 2 π(α + t ) ©2000, D. L. Jaggard Candidates can be continuous or discontinuous; monotonic or oscillatory; have finite support or infinite support All are peaked at the origin and are normalized to have unity area EE 511 46 Properties of Dirac Delta ∞ l Sifting Property ∫ f(t)δ(t − a)dt = f (a) −∞ l Scaling Property l Derivative δ (at) = 1 δ (t) a ∞ ∫ f ( t)δ' (t − a)dt = − f ' (a ) −∞ l δ(t) = δ( −t) Parity δ' (t) = −δ' (− t) ©2000, D. L. Jaggard EE 511 47 Additional Properties l Shift & δ (at + b) = 1a δ (t + ba ) Scale l Powers l Roots (a ≠ 0) t δ (t) = (−1) n!δ(t) n ( n) n δ [ f ( t)] = ∑ n δ ( t − t0 ) f ' (t n ) where f (t) = 0 has and f' (tn ) is and exists ©2000, D. L. Jaggard roots tn non - zero and EE 511 48 Some Useful Relations ∞ ∫ exp(+ j2πft)df = δ (t) −∞ ∞ ∫ exp(− j2πft)dt = δ ( f ) −∞ l l Prove by using Euler identities and property of the Sine integral This implies F.T. and F.T. –1 of unity yields a delta function F {δ(t)} = 1 F {1} = δ( f ) ©2000, D. L. Jaggard EE 511 49 ∞ Example: Show ∫ exp( + j2πft)df = δ(t) −∞ ∞ ∞ ∫ exp( + j2πft) df = ∫ [cos( 2πft ) + j sin(2πft )]df −∞ −∞ ∞ = 2∫ [cos( 2πft)]df sine function is odd and its integral vanishes 0 = lim f → ∞ 2 sin(2πft ) 2πt = δ(t) ©2000, D. L. Jaggard EE 511 50 II. Fourier Transforms and Linear Systems A. B. C. D. E. F. G. H. I. Requirements “Inventing” the Fourier Transform Dirac Delta Function Use of F.T. in Optics F.T. Properties Some Useful Transforms Two-Dimensional F.T. More Useful Transforms Sampling ©2000, D. L. Jaggard EE 511 51 D. Use of the Fourier Transforms in Optics l Spatial frequency l Analogy between electrical and optical quantities l Essence of Fourier Optics ©2000, D. L. Jaggard EE 511 52 Spatial Frequency l Conjugate variables for g(t) l In space we have conjugate variables for g(x) t (seconds) x(meters) f (seconds –1) f x (meters –1) l fx is denoted spatial frequency along coordinate x ©2000, D. L. Jaggard EE 511 53 More Spatial Frequency l For two-dimensional surfaces f x is the spatial frequency along the x-axis while f y is the spatial frequency along the y-axis l Note: l fx l l is the number of cycles per meter of variation of a spatial signal fx = k x /2π (normalized wavenumber) fy = k y /2π (normalized wavenumber) ©2000, D. L. Jaggard EE 511 54 Examples: Surface Height h(x,y) h(x,y) = surface height h(x,y) = surface height y x y x Which fractal surface has more higher spatial frequencies? ©2000, D. L. Jaggard EE 511 55 Summary l Signals in Time l Signals in Space l g(t) signal l g(x) signal l l x fx variable spatial l ω=2π f variable cyclic frequency radian frequency l l t f l kx=2π fx frequency wavenumber Note: Spatial frequency (=1/λ) is the normalized wavenumber k (=2π/λ) ©2000, D. L. Jaggard EE 511 56 Electrical/Optical Analog l l Electrical l l l l l l l Fourier transform Quadratic phase filter Linear FM generator Filtering Pulse shaping Autocorrelation Narrow band filter Optical l l l l l l l ©2000, D. L. Jaggard Fraunhofer diffraction Fresnel diffraction Lens Contrast improvement Apodization Coherence Fabry-Perot cavity Interferometer EE 511 57 Essence of Fourier Optics l Far-zone field U(f x,f y) is the twodimensional Fourier transform of the aperture field U 0(x,y) x’ fx ~x U0 (x’,y’) U(fx,fy) z y’ fy ~y Diffracted field is the F.T. of the aperture field ©2000, D. L. Jaggard EE 511 58 E. Fourier Transform Properties 1. 2. Linearity Oddness & Evenness 3. 4. Scaling Shifting 5. 6. 7. Modulation Convolution Correlation 8. 9. Rayleigh , Parseval and Power Theorems Differentiation 10. 11. Moments Periodic Functions ©2000, D. L. Jaggard EE 511 59 Fourier Transform Properties - Linearity Ag(t ) + Bh(t) ⇔ AG( f ) + BH ( f ) Proof: Follows from definition Fourier transforms add and scale in amplitude as do the original functions ©2000, D. L. Jaggard EE 511 60 Example: rect function 1/ 2 F {rect(t)} = ∫e − j 2 πft dt −1 / 2 e − jπ f − e+ jπf − j2πf sin(πf ) = πf = ≡ sinc( f ) or rect( t) ⇔ sinc( f ) ©2000, D. L. Jaggard EE 511 61 Fourier Transform Properties - Oddness and Evenness g(t) G(f) Real & ev en Real & ev en Real & odd Imaginary & odd Imaginary & even Imaginary & even Comp lex & even Complex & even Comp lex & odd Complex & odd Real & asymmetrical Complex & hermitian Symmetry of function gives rise to symmetry of its transform Imaginary and asymmet. Complex and antiherm. Real even plus imaginary odd Real odd plus imaginary even Even Real Odd Odd Imaginary Even ©2000, D. L. Jaggard Note: g(t) = g*(–t) indicates Hermitian function EE 511 62 ©2000, D. L. Jaggard EE 511 63 Fourier Transform Properties - Scaling g( at) ⇔ 1 G( f / a) a Proof: F {g( at)} = ∞ ∫ g( at)e −j 2 πft dt let y = at −∞ 1 = a ∞ ∫ g( y)e − j2 πfy / a dy for a>0 −∞ 1 G( f / a) a similar proof = for a < 0 There is an inverse scaling relation between functions and their transforms ©2000, D. L. Jaggard EE 511 64 Example: rect function F {rect(t / T)} = T/ 2 ∫ [1]e − j2 πft dt −T/ 2 = − j2 πft T/ 2 e − j2πf =T −T /2 sin(πTf ) (πTf ) = T sinc(Tf ) If rect function is stretched by factor “T” then its transform is compressed by same factor “T” and its amplitude is also scaled (to conserve power) ©2000, D. L. Jaggard EE 511 65 Fourier Transform Properties - Shifting g( t − a) ⇔ G( f )exp(− j2πfa) Proof: F {g (t − a )} = ∞ ∫ g (t − a )e − j 2 π ft dt let y =t − a −∞ = ∞ − j 2 π ( y +a ) f dy ∫ g (y )e −∞ ∞ = e− j 2 π fa ∫ g ( y )e − j 2 π f y dy −∞ = e− j 2 π fa G ( f ) Shifting in the time -domain leads to phase delay in the frequency -domain (no shift in frequency -domain) so F.T. amplitude is unaltered ©2000, D. L. Jaggard EE 511 66 Two Time Shifts Makes Difference Example: F { g (t − a) +g (t + a )} = ∞ ∫ [g (t − a) + g (t + a )]e − j 2 π ft dt −∞ =L = G ( f )[e− j 2π fa + e+ j 2π fa ] = 2G( f )cos(2π fa) What does this mean for two -slit diffraction? In Fourier optics this represents the interference for two-slit diffraction ©2000, D. L. Jaggard EE 511 67 Fourier Transform Properties - Modulation g(t) exp(+ j2πf0t) ⇔ G( f − f0 ) Proof : F {g(t )e + j2πf 0 t }= ∞ ∫ g(t )e ∞ = + j 2πf 0t − j2πft e dt −∞ ∫ g(t )e − j 2π ( f − f 0 )t dt −∞ = G( f − f0 ) Modulation in the time -domain leads to frequency shifting in the frequency -domain Useful for modulated pulses ©2000, D. L. Jaggard EE 511 68 Example: Modulated Pulse l Consider F. T. of rect(t/T) cos(2πf0t) F {rect(t / T)cos(2πf0t } = T /2 ∫ −T / 2 T /2 ∫ = −T / 2 Half the transform is shifted to positive frequency and half to negative frequency = e+ j 2πf 0t + e − j 2πf 0t − j 2πft e dt 2 e+ j 2π ( f 0 − f ) t + e− j2π ( 2 T/ 2 f0 + f )t dt T /2 e e + j4π( f0 − f ) − j4π( f 0 + f ) − T /2 − T /2 + j 2π ( f 0 − f )t − j2π ( f0 + f )t =L = T T sinc[T( f − f0 )] + sinc[T( f + f 0)] 2 2 How many cycles need to be in the pulse so that the relative bandwidth of the transform is ~10%? ©2000, D. L. Jaggard EE 511 69 Fourier Transform Properties - Convolution g(t) ⊗ h(t) ≡ ∞ ∫ g(τ )h(t −τ )dτ −∞ g( t) ⊗ h(t ) ⇔ G( f )H ( f ) Proof: F {g(t )⊗ h(t )} = ∞ ∞ ∫ ∫ g(τ )h(t − τ)e − j2 πft dτdt −∞−∞ This is an inner product of a function and a shifted & reversed function integrate over t = ∞ ∫ g(τ )H(f )e − j2 πfτ dτ −∞ = G( f )H( f ) Convolution in the time -domain leads to multiplication in the frequency -domain ©2000, D. L. Jaggard EE 511 70 Sufficient Conditions for Convolution l For g(t) = f(t) V h(t) to exist (assuming f and h are reasonably well-behaved and single valued): l l l Both f(t) and h(t) are absolutely integrable on (–8 ,0); or Both f(t) and h(t) are absolutely integrable on (0, 8 ); or Either f(t) or h(t) are absolutely integrable on (–8 , 8 ) ©2000, D. L. Jaggard EE 511 71 Properties of Convolution l Delta Function l Commutative Property l Distributive Property l Shift Invariance l l l l l g(t) V d(n)(t) = g(n)(t) = nth derivative of g(t) f(t) V h(t) = h(t) V f(t) [av(t) + bw(t)] V h(t) = a[v(t) V h(t)] + b[w(t) V h(t)] If f(t) V h(t) = g(t) then f(t –t 0) V h(t) = g(t –t0 ) Associative Property l [v(t) V w(t)] V h(t) = v(t) V [w(t) V h(t)] ©2000, D. L. Jaggard EE 511 72 Physical Interpretation of Convolution i(t) h(t) o(t) I(f) H(f) O(f) Input Linear System Output l O(f) = H(f) I(f) [system transfer function] l If we want O(f) = I(f) –> H(f) = 1 l Or h(t) = δ(t) since F–1{1} = δ(t) l o(t) = i(t) V h(t) so h(t) = δ(t) = F–1{H(f)} l Therefore, h(t) is called the “impulse response” ©2000, D. L. Jaggard EE 511 73 More on Convolution l What happens if the linear system distorts input? l Convolution is the natural operation to find the time -domain response o(t) to the system for arbitrary input i(t) ©2000, D. L. Jaggard EE 511 74 Convolution Example “Flip and Shift” Let e − t E(t )= 0 aE(ατ ) bE(βt − βτ) τ t> 0 t <0 ∞ aE(αt )⊗ bE(βt) = ab ∫ E(ατ)E(βt − βτ)dτ −∞ t = abE(βt )∫ E (ατ − βτ)dτ aE(αt) ⊗ bE( βt) 0 t Convolution calculation ©2000, D. L. Jaggard = abE(βt ) = ab E (αt − βt)− 1 β −α E(αt) − E(βt) β −α EE 511 75 Result of Multiple Convolutions l sinc(f) rect(t) t f t f t f t f l l Convolution usually smoothes function (running average) Width of convolution is sum of widths of individual functions Central Limit Theorem yields Gaussian for many convolutions Multiple convolutions of rect(t) and its F.T. ©2000, D. L. Jaggard EE 511 76 Fourier Transform Properties - Correlation Rgh (t) ≡ g( t)Ηh( é t) ≡ ∞ ∫ g(τ )h (τ − t)dτ ∗ −∞ ∗ éh(t) ⇔ G( f )H ( f) g(t)Η This is an inner product of a function and a shifted & conjugate function • Proof is similar to convolution relation proof • Definitions vary with authors • Autocorrelation is Rgg = g(t) é g(t) Correlation in the time -domain leads to multiplication in the frequency -domain ©2000, D. L. Jaggard EE 511 77 Properties of Correlation l Cross-correlation does not commute l Autocorrelation is Hermitian l Maximum modulus of autocorrelation occurs at the origin l Autocorrelation decay provides “correlation time” or “correlation length” or characteristic scale of signals and surfaces l g(t) é h(t) ? h(t) é g(t) l Rgg (t) = l Rgg*(–t) |Rgg(t)| < R gg (0) ©2000, D. L. Jaggard EE 511 78 Fourier Transform Properties - Rayleigh/Parseval Relation ∞ ∫ g(t) ∞ 2 dt = −∞ ∫ G( f ) 2 df −∞ P roof: ∞ ∞ ∫ g(t) g (t)dt = ∫ g(t)g (t)e ∗ ∗ −∞ −j 2 πf ' t dt f' = 0 for −∞ = G( f ' ) ⊗ G∗ (− f ' ) = ∞ ∫ G( f )G ( f − f ' )df ∗ for f' = 0 for f' = 0 −∞ = ∞ ∫ G( f )G∗ ( f )df −∞ Area under the absolute value squared of a function is equal to area under the absolute value squared of its transform ©2000, D. L. Jaggard EE 511 79 Rayleigh Theorem Example g(t) G(f) t |g(t)|2 f |G(f)|2 t l f Equal areas under |g(t)| 2 and |G(f)|2 imply conservation of power for optical (and other wave) signals ©2000, D. L. Jaggard EE 511 80 Fourier Transform Properties - Differentiation d n g(t) = g ( n )(t) ⇔ ( j2πf )n G( f ) n dt l l l l Prove by usual means Differentiation in the time -domain leads to multiplication by frequency in the frequency -domain Useful for solving D.E.s Useful for finding F.T. of piecewise continuous functions: l l Take second derivative Find inverse transform as sum of delta functions ©2000, D. L. Jaggard EE 511 81 Fourier Transform Properties - Moments ∞ m n = ∫ t n g(t)dt = m th moment of g(t) −∞ mn = ( n) G (0) (− j2π )n The moment of a function is related to the behavior of its transform near the origin l Therefore low frequency regime of transform may be valuable in classification & identification l ©2000, D. L. Jaggard EE 511 82 Properties of Moments m0 = G(0 ) zeroth moment or area (1) m1 = G (0) − j2π G (2 ) (0) m2 = −4π 2 m2 = m0 first moment or centroid second moment G ( 2) (0) 2 −4π G(0) radius or moment of inertia of gyration Moments can be used for signal or image classification ©2000, D. L. Jaggard EE 511 83 Second Moment Examples G(f) g(t) l t f t f l g(t) Moments of g(t) affect G(f) at the origin G(f) t f ©2000, D. L. Jaggard Infinite moment of g(t) gives cusp at origin for G(f) EE 511 84 Fourier Transform Properties Periodic Functions If g(t) is periodic g(t ) = with period ∞ ∑G −∞ exp( j2nπf0t ) n where Gn = ∞ ∑ G F {exp( j2 nπf t)} n 0 −∞ = ∞ ∑ G δ ( f − nf ) n T/2 ∫ g(t) exp( − j2 nπf t )dt 0 − T /2 ∞ F {g(t )}= F ∑ Gn exp( j 2nπf0t ) −∞ = T = 1 / f0 0 The spectrum of a periodic function is a “line spectrum” This is the complex Fourier series −∞ To find the F.T. of a periodic function, find the F.T. of its Fourier series ©2000, D. L. Jaggard EE 511 85 Some Notes l F{F{g(t)}} = g(–t) l l l If we know a F.T. pair, we can invert this pair by inverting the coordinate If one lens in optics gives a F.T., then a second lens can be used to provide an inverted image If g(t) and its first (n–1) derivatives are continuous, its F.T. decays as least as rapidly as |f| –(n+1) ©2000, D. L. Jaggard EE 511 86 II. Fourier Transforms and Linear Systems A. B. C. D. E. F. G. H. I. Requirements “Inventing” the Fourier Transform Dirac Delta Function Use of the F.T. in Optics F.T. Properties Some Useful Transforms Two-Dimensional F.T. More Useful Transforms Sampling ©2000, D. L. Jaggard EE 511 87 F. Some Useful Transforms l Unity and Delta Function Rect l Triangle l Gaussian l Signum and Step l Comb l Sine and Cosine l ©2000, D. L. Jaggard EE 511 88 Some Useful Functions t / T <1 / 2 t / T >1 / 2 1 rect(t / T ) = 0 1 − t / T t / T < 1 Λ(t / T ) = t /T >1 0 Gaus(t / T ) = exp[−π (t / T )2 ] ∞ comb(t / T ) = ∑ δ (t / T − n ) = T −∞ 1 sgn( t) = −1 1 u(t) = 0 ∞ ∑δ (t − n / T ) −∞ t >1 t <1 t >1 t <1 ©2000, D. L. Jaggard = 1 1 + sgn(t) 2 2 EE 511 89 Some Useful F.T. Pairs 1⇔ δ( f ) δ (t) ⇔ 1 rect(t / T) ⇔ T sinc (Tf ) Λ(t / T) ⇔ T sinc 2 (Tf ) comb(t / T) ⇔ T comb(Tf ) Gaus(t / T) ⇔ T Gaus(Tf ) 2 exp(−t ) ⇔ 4π 2 f 2 +1 t −1 ⇔ − jπ sgn(−πf 2 ) sgn( t) ⇔ ( jπf )−1 u(t) ⇔ 2 −1 δ (t) + ( j2πf )−1 cos(2πf0t) ⇔ 2 −1 [δ( f − f0 ) + δ( f + f0 )] sin(2πf0t) ⇔ (2 j)−1 [δ ( f − f0 ) − δ( f + f 0 )] ©2000, D. L. Jaggard EE 511 90 G. Two-Dimensional F.T. l Separable Function l Circular Symmetry l l Hankel transforms l Fourier-Bessel transforms Some Two -Dimensional F.T. Pairs ©2000, D. L. Jaggard EE 511 91 Separable Functions -- Easy Fourier Transforms l If the two -dimensional function is separable, its F.T. is the product of two one-dimensional F.T.s: l Let g(x,y) = j(x) h(y) l Then F{g(x,y) = F{j(x)} F{ h(y)} = J(f) H(f) ©2000, D. L. Jaggard EE 511 92 Polar Separable Case l Let g(r,θ) be separable in polar coordinates: l Let g(r,θ) = g R(r)ejm θ l Then I.C.B.S.T. F {g(r,θ)} = (–j)m ejm θ Hm {g R(r)} where ∞ Hm {g R (r)} = 2π ∫ r g R(r) Jm (2πρr)dr 0 is the Hankel transform of order m and Jm is the Bessel function of order m ©2000, D. L. Jaggard EE 511 93 Case of Circular Symmetry l Let g(r,θ) = gR(r) = g(r) [circular symmetry]: l Then I.C.B.S.T. ∞ F {g(r)} = 2 π ∫ r g R(r) J0(2πρr)dr 0 is the Fourier-Bessel transform of g (r) where J 0 is the Bessel function of order 0 l This gives the Fourier-Bessel pair ∞ G( ρ) = 2π ∫ rg(r)J0 (2πrρ)dr 0 ∞ g(r) = 2π ∫ ρG( ρ)J 0 (2πrρ)dρ 0 g( r) ⇔ G (ρ) ©2000, D. L. Jaggard Also known as Hankel transform of order zero EE 511 94 Fourier-Bessel Proof ∞ G( f x , f y ) = ∞ ∫ ∫ g(x,y)e − j2π ( f x x + fy y) dxdy −∞ −∞ r= x +y 2 ρ= 2 θ = tan −1 (y / x) fx 2 + f y 2 φ = tan − 1 (f y / f x ) fx cosφ f = ρ y sinφ y cosθ = r y sinθ F {g(r)} = G(ρ) 2π ∞ = ∫ ∫ g(r)e − j2π rρ (cosθ cos φ +sin θ sin φ ) rdrdθ 0 0 = 2π ∞ ∫ ∫ g(r)e − j2π rρ cos(θ −φ ) rdrdθ 0 0 2π ∫e Note: − ja cos(θ −φ ) dθ = 2πJ0 (a) 0 ∞ ©2000, L. Jaggard F {g(r)} = G(ρ) = 2πD.∫rg(r )J 0 (2πrρ)dr 0 EE 511 95 Some Useful Relations x ∫ yJ 0 (y)dy = x J1 (x) 0 ∞ ∫x ν +1 exp( −αx )J ν (βx)dx = 2 0 Jν (x) → x→0 (1 / 2 )x ν Γ(ν +1) βν 2 exp( −β / 4α) 2α ( ν +1) (v ≠ −1,−2, −3,...) 2 cos[ x − (1 / 2 )νπ − (1 / 4)π Jν (x) → x→∞ πx where Γ (ν + 1) = ν! for ν = integer Γ(ν + 1) =ν Γ (ν ) ©2000, D. L. Jaggard EE 511 96 Bessel Functions J 0 & J 1 l l l All Bessel functions except J 0 are zero at the origin J0 is unity at the origin Bessel functions have significant value in the regime where their order equals their argument ©2000, D. L. Jaggard EE 511 97 Two-Dimensional Functions r / a <1 1 circ(r / a) ≡ r / a >1 0 J (2π ρ) jinc(ρ) ≡ 2 1 2πρ Gaus(ax, by) ≡ exp[−π (a 2 x2 + b 2 y2 )] Gaus(ar) ≡ exp[−π a2 r 2 ] = exp[−πa 2 (x2 + y2 )] δ ( x, y) ≡ δ (x)δ( y) = comb( ax)comb(by) ≡ ∞ δ (r) πr ∞ ∑ ∑δ (ax − n, by − m) n= −∞ m= −∞ ©2000, D. L. Jaggard EE 511 98 “sinc” and “ jinc” Functions sinc(x) jinc(r/2) sinc2(x) jinc2(r/2) What are the differences between sinc and jinc functions? ©2000, D. L. Jaggard EE 511 99 Two-Dimensional F.T. Pairs rect (x / a)rect (y / b) ⇔ absinc(afx )sinc(bfy ) circ(r / a) ⇔ π a jinc(aρ) 2 δ (r) ⇔1 πr 1 ab 1 1 ⇔ r ρ δ (ax,by ) ⇔ 1 ⇔δ ( f x , f y ) cos(πr2 ) ⇔ sin(πρ2 ) exp( ± jπr2 ) ⇔ ± j exp( mjπρ 2 ) Gaus(ax,by ) ⇔ Gaus(ar) ⇔ 1 Gaus( f x / a, f y / b ) ab 1 Gaus( f ρ / a) 2 a ©2000, D. L. Jaggard comb(x / a)comb(y / b) ⇔ abcomb( afx )comb(bfy ) EE 511 100 How Often Does a Signal Need to be Sampled in Order to be Exactly Replicated? ©2000, D. L. Jaggard EE 511 101 Sampling of Signals l Whittaker-Shannon Sampling Theorem A signal which is bandlimited (i.e., all frequencies less than f max) can be exactly reconstructed by accurate samples at times T < ( 2 f max ) −1 l l The frequency 2 fmax (number of samples per second) is known as the Nyquist frequency Requirement on sampling frequency is fs > 2 fmax ©2000, D. L. Jaggard EE 511 102 Miracle of Sampling sinc function is “interpolating function” for exact reconstruction l Exact reconstruction if overlap avoided by T < 1 2 fm ©2000, D. L. Jaggard EE 511 103 Imperfect Sampling l l l l What if samples of finite width are used so that each sample is an average of a part of the signal? What if the signal is only sampled for a finite length of time? What if unequally spaced samples are used? What if the samples are descretized? ©2000, D. L. Jaggard EE 511 104 Two-Dimensional Sampling l l Various geometries can be used Space-bandwidth product: l If g(x,y) has significant value in the region |x|<LX and |y|<LY, and if g(x,y) is sampled on a rectangular lattice (Nyquist rate) with spacing (2B X)–1 and (2BY) –1 in the x and y directions, then the total number of (possibly complex) samples needed to represent g(x,y) is l M = 16 LX LY BX BY ©2000, D. L. Jaggard EE 511 105