Lecture IX: Fourier transform Maxim Raginsky BME 171: Signals and Systems Duke University October 8, 2008 Maxim Raginsky Lecture IX: Fourier transform This lecture Plan for the lecture: 1 Recap: Fourier series representation of periodic signals 2 Frequency content of aperiodic signals: the Fourier transform 3 The inverse Fourier transform 4 Properties of the Fourier transform 5 Generalized Fourier transform 6 Bandlimited and timelimited signals 7 Frequency response of LTI systems Maxim Raginsky Lecture IX: Fourier transform Recap: Fourier series Recall from the last lecture that any sufficiently regular T -periodic continuous-time signal x(t) can be expanded, e.g., in a complex exponential Fourier series: x(t) = ∞ X ck ejkω0 t , k=−∞ where ω0 = 2π/T is the fundamental frequency, and the Fourier coefficients {ck } are given by ck = 1 T Z T /2 x(t)e−jkω0 t dt, k = . . . , −2, −1, 0, 1, 2, . . . −T /2 The Fourier coefficients {ck } tell us about the frequency content (or spectral content) of x(t). Maxim Raginsky Lecture IX: Fourier transform Spectral content of aperiodic signals: the Fourier transform What about aperiodic signals? Any continuous-time signal x(t) that has finite “energy”, i.e., Z ∞ x2 (t)dt < +∞, −∞ can be represented in the frequency domain via the Fourier transform: Z ∞ x(t)e−jωt dt X(ω) = −∞ We will also write X(ω) = F [x(t)] to denote the fact that X(ω) is the Fourier transform of x(t). Maxim Raginsky Lecture IX: Fourier transform Example: rectangular pulse Consider the rectangular pulse pτ (t) = F [pτ (t)] = = 1, |t| ≤ τ /2 0, |t| > τ /2 Z ∞ pτ (t)e−jωt dt −∞ Z τ /2 e−jωt dt −τ /2 = = = = Maxim Raginsky − 1 −jωt e jω τ /2 −τ /2 ejωτ /2 − e−jωτ /2 jω 2 sin(ωτ /2) ω τω τ sinc . 2π Lecture IX: Fourier transform Inverse Fourier transform The signal x(t) can be recovered from its Fourier transform X(ω) = F [x(t)] using the inverse Fourier transform formula Z ∞ 1 x(t) = F −1 [X(ω)] = X(ω)ejωt dω 2π −∞ Note: There is a factor of 1/2π in front of the integral. The integration is with respect to ω, for a fixed value of t. We will also write x(t) ↔ X(ω) and say that x(t) [time domain] and X(ω) [freq. domain] are a Fourier transform pair. Maxim Raginsky Lecture IX: Fourier transform Proof: 1 2π Z ∞ Z ∞ ′ 1 x(t′ )e−jωt dt′ ejωt dω X(ω)ejωt dω = 2π −∞ −∞ −∞ Z ∞ Z ∞ ′ 1 = x(t′ ) ej(t−t )ω dω dt′ . 2π −∞ −∞ Z ∞ 1 2π Z Z πΩ ′ 1 lim ej(t−t )ω dω Ω→∞ 2π −∞ −πΩ iπΩ h 1 j(t−t′ )ω e = lim ′ Ω→∞ 2πj(t − t )Ω −πΩ ′ sin(πΩ(t − t )) = lim Ω→∞ πΩ(t − t′ ) = lim sinc(Ω(t − t′ )) ∞ ′ ej(t−t )ω dω = Ω→∞ = δ(t − t′ ) Hence, 1 2π Q.E.D. Z ∞ X(ω)ejωt dω = −∞ Z ∞ x(t′ )δ(t − t′ )dt′ = x(t) −∞ Maxim Raginsky Lecture IX: Fourier transform Properties of the Fourier transform The Fourier transform has many useful properties that make calculations easier and also help thinking about the structure of signals and the action of systems on signals. The properties are listed in any textbook on signals and systems. We will look at and prove a few of them. Maxim Raginsky Lecture IX: Fourier transform Linearity The Fourier transform is linear: if x1 (t) ↔ X1 (ω) and x2 (t) ↔ X2 (ω), then c1 x1 (t) + c2 x2 (t) ↔ c1 X1 (ω) + c2 X2 (ω) for any two numbers c1 and c2 . Proof: obvious – F [c1 x1 (t) + c2 x2 (t)] = Z ∞ [c1 x1 (t) + c2 x2 (t)] e−jωt dt −∞ = c1 Z ∞ x1 (t)e −jωt dt + c2 −∞ Z ∞ x2 (t)e−jωt dt −∞ = c1 X1 (ω) + c2 X2 (ω) Q.E.D. Maxim Raginsky Lecture IX: Fourier transform Time shift If x(t) ↔ X(ω), then x(t − c) ↔ X(ω)e−jωc for any constant c. Proof: F [x(t − c)] = Z ∞ x(t − c)e−jωt dt −∞ Z ∞ x(t)e−jω(t+c) dt Z ∞ −jωc = e x(t)e−jωt dt = −∞ −∞ = X(ω)e−jωc . Q.E.D. Maxim Raginsky Lecture IX: Fourier transform Multiplication by a complex exponential If x(t) ↔ X(ω), then x(t)ejω0 t ↔ X(ω − ω0 ) for any real ω0 . Proof: F x(t)e jω0 t = = Z ∞ x(t)ejω0 t e−jωt dt −∞ Z ∞ x(t)e−j(ω−ω0 )t dt −∞ = X(ω − ω0 ). Q.E.D. Maxim Raginsky Lecture IX: Fourier transform Multiplication by a cosine If x(t) ↔ X(ω), then x(t) cos(ω0 t) ↔ 12 [X(ω + ω0 ) + X(ω − ω0 )]. Proof: use linearity and the last property to get 1 jω0 t −jω0 t x(t) e +e F [x(t) cos(ω0 t)] = F 2 1 1 = F x(t)ejω0 t + F x(t)e−jω0 t 2 2 1 = [X(ω − ω0 ) + X(ω + ω0 )] . 2 Q.E.D. Maxim Raginsky Lecture IX: Fourier transform Convolution in time domain If x(t) ↔ X(ω) and v(t) ↔ V (ω), then x(t) ⋆ v(t) ↔ X(ω)V (ω) Proof: F [x(t) ⋆ v(t)] Z ∞ Z ∞ [x(t) ⋆ v(t)]e−jωt dt Z ∞ = x(λ)v(t − λ)dλ e−jωt dt −∞ −∞ Z ∞ Z ∞ = x(λ) v(t − λ)e−jωt dt dλ −∞ −∞ | {z } = −∞ Z ∞ F [v(t−λ)] x(λ)V (ω)e−jωλ dλ Z ∞ x(λ)e−jωλ dλ = V (ω) = −∞ −∞ = X(ω)V (ω). Q.E.D. Maxim Raginsky Lecture IX: Fourier transform Parseval’s theorem Let x(t) and v(t) be real-valued signals. Then Z ∞ Z ∞ 1 X(ω)V (ω)dω x(t)v(t)dt = 2π −∞ −∞ Proof: Z ∞ x(t)v(t)dt = −∞ Z ∞ −∞ = = = V (ω)ejωt dω) dt −∞ Z ∞ V (ω) x(t)ejωt dt dω x(t) 1 2π Z ∞ Z ∞ 1 2π −∞ −∞ Z ∞ 1 V (ω)X(−ω)dω 2π −∞ Z ∞ 1 X(ω)V (ω)dω, 2π −∞ where we used the fact that, since x(t) is real, Z ∞ X(ω) = x(t)ejωt dt = X(−ω). −∞ Q.E.D. Maxim Raginsky Lecture IX: Fourier transform Parseval’s theorem: cont’d An important consequence of Parseval’s theorem is that Z ∞ Z ∞ 1 2 x (t)dt = |X(ω)|2 dω. 2π −∞ −∞ In other words, signal energy can be computed both in time domain and in frequency domain (up to a factor of 1/2π). Maxim Raginsky Lecture IX: Fourier transform Duality If x(t) ↔ X(ω), then X(t) ↔ 2πx(−ω). Proof: F [X(t)] = = = Z ∞ X(t)e−jωt dt −∞ Z ∞ 1 X(t)e−jωt dt 2π · 2π −∞ Z ∞ ′ 1 2π · X(ω ′ )e−jωω dω ′ 2π −∞ | {z } =F −1 [X(ω)](−ω) = 2π · 1 x(−ω) 2π Q.E.D. Maxim Raginsky Lecture IX: Fourier transform Duality: an example Let x(t) = τ sinc τt 2π . Then by duality we have X(ω) = 2πpτ (ω). In more detail: pτ (t) ↔ τ sinc Thus, by duality, τ sinc τt 2π Maxim Raginsky τω 2π ↔ 2πτ (ω). Lecture IX: Fourier transform Generalized Fourier transform The Fourier transform is defined only for signals with finite energy. However, we can extend its scope by allowing singularity functions. We begin by computing the Fourier transform of the unit impulse δ(t). Z ∞ F [δ(t)] = δ(t)e−jωt dt −∞ Z ∞ = δ(t)dt −∞ = 1, where we used the sifting property of the unit impulse. By duality, we have 1 ↔ 2πδ(−ω) = 2πδ(ω). Maxim Raginsky Lecture IX: Fourier transform Fourier transform of the cosine The cosine signal x(t) = cos(ω0 t) does not have the Fourier transform in the ordinary sense. It does, however, have a generalized Fourier transform: 1 jω0 t −jω0 t F [cos(ω0 t)] = F (e +e ) 2 1 1 F 1 · ejω0 t + F 1 · e−jω0 t = 2 2 1 = [2πδ(ω − ω0 ) + 2πδ(ω + ω0 )] 2 = πδ(ω − ω0 ) + πδ(ω + ω0 ). Q.E.D. Maxim Raginsky Lecture IX: Fourier transform Fourier transform of a periodic signal Using the generalized Fourier transform, we can analyze periodic signals that do not have a Fourier transform in the ordinary sense. Thus, if x(t) is a T -periodic signal, we can expand it in a complex exponential Fourier series as ∞ X ck ejkω0 t . x(t) = k=−∞ X(ω) = = = F " ∞ X k=−∞ ∞ X k=−∞ ∞ X ck e jkω0 t # ck F ejkω0 t 2πck δ(ω − kω0 ). k=−∞ Thus, the (generalized) Fourier transform of a periodic signal is a train of impulses located at integer multiples of the fundamental frequency ω0 . Maxim Raginsky Lecture IX: Fourier transform Bandlimited and timelimited signals A signal x(t) is called: bandlimited if there exists a number B > 0 (called the bandwidth), such that X(ω) = 0, for all |ω| ≥ B. timelimited if there exists a number T > 0, such that x(t) = 0, for all |t| ≥ T. It can be proved that a bandlimited signal cannot be timelimited, and vice versa. We’ve seen an example of this with the transform pairs τω τt and τ sinc ↔ 2πpτ (ω) pτ (t) ↔ τ sinc 2π 2π However, a signal can be approximately timelimited and bandlimited — that is, there exist numbers B > 0 and T > 0, such that |x(t)| is small for |t| ≥ T and |X(ω)| is small for |ω| ≥ B. Maxim Raginsky Lecture IX: Fourier transform Frequency response of LTI systems Consider an LTI system with the impulse response h(t). Then the output of the system due to input x(t) is given by the convolution integral, Z ∞ x(λ)h(t − λ)dλ. y(t) = x(t) ⋆ h(t) = −∞ In frequency domain, the action of the system can be described as follows: Y (ω) = H(ω)X(ω). This is a consequence of the fact that convolution in time domain corresponds to multiplication in frequency domain. The Fourier transform H(ω) of the impulse response h(t) is called the frequency response of the system. Maxim Raginsky Lecture IX: Fourier transform