EI331 Signals and Systems Lecture 14 Bo Jiang John Hopcroft Center for Computer Science Shanghai Jiao Tong University April 11, 2019
Contents 1. CT Fourier Transform 2. Fourier Transform of L 1 Signals 3. Fourier Transform of More General Functions 4. Fourier Transform of Periodic Signals 1/30
CT Fourier Transform Fourier transform (analysis equation) � ∞ x ( t ) e − j ω t dt X ( j ω ) = F { x } ( j ω ) = −∞ X ( j ω ) called spectrum of x ( t ) Inverse Fourier transform (synthesis equation) � ∞ x ( t ) = F − 1 { X } ( t ) = 1 X ( j ω ) e j ω t d ω 2 π −∞ Superposition of complex exponentials at continuum of 1 frequencies; frequency ω has density 2 π X ( j ω ) 2/30
CT Fourier Transform Two equivalent representations of same signal F • time domain vs. frequency domain: x ( t ) ← − − → X ( j ω ) Also widely used elsewhere. In probability theory, • characteristic function of random variable X with density p ( x ) � ∞ ϕ X ( t ) = E [ e jtX ] = p ( x ) e jxt dx −∞ In quantum mechanics • position representation: wave function ψ ( x ) ◮ | ψ ( x ) | 2 probability density of finding particle at position x • momentum representation: Ψ( p ) � ∞ 1 ψ ( x ) e − jpx / � dx √ Ψ( p ) = 2 π � −∞ ◮ | Ψ( p ) | 2 probability density of finding particle with momentum p 3/30
Contents 1. CT Fourier Transform 2. Fourier Transform of L 1 Signals 3. Fourier Transform of More General Functions 4. Fourier Transform of Periodic Signals 4/30
Fourier Transform of L 1 Signals Recall from calculus, for real-valued x , improper integral � T 2 � x ( t ) dt � lim x ( t ) dt T 1 , T 2 →∞ R − T 1 � is well-defined if x ∈ L 1 ( R ) , i.e. � x � 1 = R | x ( t ) | dt < ∞ Same is true for complex-valued x = u + jv , since | u | , | v | ≤ | x | and � � � x ( t ) dt � u ( t ) dt + j v ( t ) dt R R R For signal x ∈ L 1 ( R ) • x ( t ) e − j ω t ∈ L 1 ( R ) , since | x ( t ) e − j ω t | = | x ( t ) | � ∞ • Fourier transform X ( j ω ) = −∞ x ( t ) e − j ω t dt is well-defined improper integral for each ω 5/30
Fourier Transform of L 1 Signals Theorem. If x ∈ L 1 , then X is bounded, in fact � X � ∞ ≤ � x � 1 Proof. Consequence of following lemma. Lemma. For complex-valued function f of real variable t , � � � � � � f ( t ) dt � ≤ | f ( t ) | dt � � � � � Proof. If | f ( t ) | dt = ∞ , trivial. Assume | f ( t ) | dt < ∞ . Let � � � f ( t ) dt | e j φ . Then phase of f ( t ) dt be φ , i.e. f ( t ) dt = | � � � � � � = e − j φ e − j φ f ( t ) dt � � f ( t ) dt f ( t ) dt = � � � Taking real part � � � � � � e − j φ f ( t ) dt = Re [ e − j φ f ( t )] dt ≤ | e − j φ f ( t ) | dt � � f ( t ) dt � = Re � � � 6/30
Fourier Transform of L 1 Signals Theorem. If x ∈ L 1 , then X is uniformly continuous Proof. For any T > 0 , � | x ( t ) e − j ω 1 t − x ( t ) e − j ω 2 t | dt | X ( j ω 1 ) − X ( j ω 2 ) | ≤ R � � � � 2 sin ∆ ω t � � = | x ( t ) | · (∆ ω = ω 1 − ω 2 ) � dt � � 2 R � � ≤ 2 | x ( t ) | dt + | ∆ ω | T | x ( t ) | dt | t | > T | t |≤ T � | x ( t ) | dt + | ∆ ω | T � x � 1 � I 1 + I 2 ≤ 2 | t | > T Given ǫ > 0 , since x ∈ L 1 ( R ) , exists T > 0 s.t. I 1 < ǫ/ 2 . Fix such T . For | ∆ ω | ≤ ǫ/ ( 2 T � x � 1 ) , I 2 < ǫ/ 2 . Thus | X ( j ω 1 ) − X ( j ω 2 ) | < ǫ . 7/30
Example: Right-sided Decaying Exponential x ( t ) | X ( j ω ) | x ( t ) = e − at u ( t ) , a > 0 1 1 / a √ 1 2 e 2 a � ∞ t 0 1 / a e − ( a + j ω ) t dt X ( j ω ) = 0 ∞ 1 1 a + j ω e − ( a + j ω ) t � = − 0 = � ω a − a 0 a + j ω � arg X ( j ω ) NB. Above formula for X also works for complex a with Re a > 0 . π 2 π 1 4 For a > 0 , √ a | X ( j ω ) | = a 2 + ω 2 , − a ω − π 4 arg X ( j ω ) = − arctan ω − π a 2 8/30
Example: Two-sided Decaying Exponential x ( t ) x ( t ) = e − a | t | , a > 0 1 1 / e � ∞ e − a | t | e − j ω t dt X ( j ω ) = −∞ � 0 � ∞ t − 1 / a 0 1 / a e ( a − j ω ) t dt + e − ( a + j ω ) t dt = −∞ 0 1 1 X ( j ω ) = a − j ω + a + j ω 2 a 2 / a = a 2 + ω 2 1 / a NB. Above formula for X also works for complex a with Re a > 0 . ω a − a 0 9/30
Example: Gaussian For a > 0 , � π ae − ω 2 F x ( t ) = e − at 2 ← − − → X ( j ω ) = 4 a x ( t ) X ( j ω ) √ a = 1 / 2 2 π a = 1 a = 2 1 ω t √ √ 2 ω 2 = F In particular, x ( t ) = e − 1 2 t 2 2 π e − 1 ← − − → X ( j ω ) = 2 π x ( ω ) , √ i.e. F { x } = 2 π x 10/30
Example: Gaussian For a > 0 , � π ae − ω 2 x ( t ) = e − at 2 F ← − − → X ( j ω ) = 4 a Proof. � d � ∞ � ∞ dte − at 2 � d e − at 2 ( − jt ) e − j ω t dt = j e − j ω t dt d ω X ( j ω ) = 2 a −∞ −∞ e − at 2 � d � ∞ � = − j dte − j ω t (integration by parts) dt 2 a −∞ � ∞ = − ω e − at 2 e − j ω t dt = − ω 2 aX ( j ω ) 2 a −∞ � π d ω 2 ⇒ X ( j ω ) = X ( j 0 ) e − ω 2 ae − ω 2 � � 4 a = X ( j ω ) e = 0 = 4 a 4 a d ω 11/30
Fourier Inversion for L 1 Signals Given Fourier transform, � ∞ x ( t ) e − j ω t dt X ( j ω ) = F { x } ( j ω ) = −∞ Is inverse Fourier transform well-defined? Is it equal to x ? � ∞ x ( t ) ? = F − 1 { X } ( t ) = 1 X ( j ω ) e j ω t d ω 2 π −∞ Theorem. If x ∈ L 1 ( R ) is continuous and X = F { x } ∈ L 1 ( R ) , then x = F − 1 { X } . • e.g. Two-sided decaying exponential, Gaussian But, for x ( t ) ∈ L 1 ( R ) , X ( j ω ) is not necessarily in L 1 ( R ) • e.g. one-sided decaying exponential, rectangular pulse • if X ∈ L 1 ( R ) , x must be continuous 12/30
Fourier Inversion for L 1 Signals Inverse FT typically interpreted as principal value, i.e. � ∞ � W 1 1 X ( j ω ) e j ω t d ω = lim X ( j ω ) e j ω t d ω 2 π 2 π W →∞ −∞ − W may converge without being absolutely convergent Theorem. If x ∈ L 1 ( R ) satisfies Dirichlet conditions on all finite intervals, then � W X ( j ω ) e j ω t d ω = x ( t + ) + x ( t − ) 1 lim pointwise 2 π 2 W →∞ − W NB. Gibbs phenomenon at discontinuity Often also need to interpret FT as principal value � ∞ � T x ( t ) e − j ω t dt = lim x ( t ) e − j ω t dt T →∞ −∞ − T 13/30
Example: Rectangular Pulse x ( t ) x ( t ) = u ( t + T ) − u ( t − T ) 1 � T e − j ω t dt = 2 sin( ω T ) X ( j ω ) = ω − T − T t T Inverse FT � ∞ X ( j ω ) sin( ω T ) d ω = x ( 0 ) = 1 2 T πω −∞ π As T → ∞ , T • frequency domain ω 0 sin( ω T ) lim = δ ( ω ) πω T →∞ sinc( θ ) � sin( πθ ) • time domain: x ( t ) → 1 , DC � πθ 14/30
Example: Ideal Lowpass Filter Frequency response H ( j ω ) 1 H ( j ω ) = u ( ω + ω c ) − u ( ω − ω c ) Impulse response ω ω c − ω c � ω c e j ω t d ω = sin( ω c t ) h ( t ) = 1 h ( t ) 2 π π t − ω c π ω c ω c π NB. h ( t ) / ∈ L 1 ( R ) but H ( j ω ) ∈ L 1 ( R ) t 0 As ω c → ∞ , • time domain ◮ h ( t ) → δ ( t ) , becomes identity system • frequency domain ◮ H ( j ω ) → 1 , passes all frequencies � 15/30
Duality X 1 ( j ω ) 2 T x 1 ( t ) F π 1 T ω − T t 0 T 0 x 2 ( t ) X 2 ( j ω ) F π 1 W π W ω t − W W 0 0 16/30
Contents 1. CT Fourier Transform 2. Fourier Transform of L 1 Signals 3. Fourier Transform of More General Functions 4. Fourier Transform of Periodic Signals 17/30
Fourier Transform of More General Functions If x n → x , define Fourier transform of x by X ( j ω ) = F { x } � lim n F { x n } i.e. � � x ( t ) e − j ω t dt � lim x n ( t ) e − j ω t dt X ( j ω ) = n R R If x ∈ L 1 ( R ) , above definition is consistent with old one In general, convergence interpreted in distributional sense, i.e. for nice test function φ � � �� � x n ( t ) e − j ω t dt X ( j ω ) φ ( ω ) d ω � lim φ ( ω ) d ω n R R R Interchanging order of integration leads to alternative definition �� � � � � φ ( ω ) e − j ω t d ω X ( j ω ) φ ( ω ) d ω = lim x n ( t ) dt = x ( t )Φ( jt ) dt n R R R R 18/30
Schwarz Space Space of test functions is so-called Schwarz space on R , denoted S = S ( R ) Function φ ∈ S if it is • infinitely differentiable: φ ( k ) exists for all k ∈ N • rapidly decreasing: | t ℓ φ ( k ) ( t ) | < ∞ , � φ � ℓ, k � sup ∀ ℓ, k ∈ N t ∈ R Example. Gaussian g ( t ) = e − at 2 ∈ S Note φ ∈ S = ⇒ φ ∈ L 1 , Fourier transform Φ well-defined. Theorem. If φ ∈ S , then Φ ∈ S Example. Gaussian G ( j ω ) = � π a e − ω 2 4 a ∈ S 19/30
Example: Unit Impulse δ Method 1. Recall for ideal lowpass filter, h W ( t ) = sin( Wt ) F ← − − → H W ( j ω ) = u ( ω + W ) − u ( ω − W ) π t Since h W → δ as W → ∞ F { δ } = lim W →∞ H W ( j ω ) = 1 Can also use Gaussian instead of sinc . Recall 1 √ π ae − t 2 → G a ( j ω ) = e − a ω 2 F g a ( t ) = ← − − a 4 Since g a → δ as a → 0 F { δ } = lim a → 0 G a ( j ω ) = 1 20/30
Recommend
More recommend