Sampling Theorem
📂Fourier Analysis Sampling Theorem Buildup[^1] Consider a physical signal f f f being measured over time t 1 < t 2 < t 3 < ⋯ t_{1} < t_{2} < t_{3} < \cdots t 1 < t 2 < t 3 < ⋯ . Even if we know f ( t 1 ) , f ( t 2 ) , … f(t_{1}), f(t_{2}), \dots f ( t 1 ) , f ( t 2 ) , … , we generally cannot know the values for arbitrary f ( t ) f(t) f ( t ) . However, let’s assume that signal f f f contains only frequencies within a certain range. That is, we consider a signal f f f that contains only frequencies smaller than some constant Ω \Omega Ω , which is referred to as a band-limited signal .
In the language of Fourier analysis, this is the same as saying that the function values are all 0 0 0 in the region where f ^ ( ω ) \hat{f}(\omega) f ^ ( ω ) is ∣ ω ∣ ≥ Ω |\omega| \ge \Omega ∣ ω ∣ ≥ Ω . Therefore, the condition that signal f f f is band-limited is equivalent to the condition f ^ ( ω ) = 0 for ∣ ω ∣ ≥ Ω \hat{f} (\omega) = 0\ \text{for } | \omega | \ge \Omega f ^ ( ω ) = 0 for ∣ ω ∣ ≥ Ω , meaning f ^ ∈ L 1 \hat{f} \in L^{1} f ^ ∈ L 1 . Under such conditions, the following powerful theorem holds.
Theorem Assuming that f ∈ L 2 f\in L^{2} f ∈ L 2 and f ^ ( ω ) = 0 for ∣ ω ∣ ≥ Ω \hat{f} (\omega) = 0\ \text{for } | \omega | \ge \Omega f ^ ( ω ) = 0 for ∣ ω ∣ ≥ Ω , then f ( t ) f(t) f ( t ) is determined by the values in n π / Ω ( n = 0 , ± 1 , ± 2 , … ) n\pi / \Omega (n=0, \pm 1, \pm 2, \dots) nπ /Ω ( n = 0 , ± 1 , ± 2 , … ) . That is, the following applies.
f ( t ) = ∑ n = − ∞ ∞ f ( n π Ω ) sin ( Ω t − n π ) Ω t − n π
f(t) = \sum \limits_{n = -\infty}^{\infty} f \left( \dfrac{n\pi}{\Omega}\right) \dfrac{\sin (\Omega t -n\pi )}{\Omega t - n\pi}
f ( t ) = n = − ∞ ∑ ∞ f ( Ω nπ ) Ω t − nπ sin ( Ω t − nπ )
Explanation This is referred to as the sampling theorem . The sampling theorem provides the condition under which all f ( t ) f(t) f ( t ) for all t t t can be determined using a countable number of function values f ( n π Ω ) f \left( \dfrac{n\pi}{\Omega}\right) f ( Ω nπ ) .
From the perspective of signal analysis, the condition f ^ ( ω ) = 0 for ∣ ω ∣ ≥ Ω \hat{f} (\omega) = 0\ \text{for } | \omega | \ge \Omega f ^ ( ω ) = 0 for ∣ ω ∣ ≥ Ω means that the frequency of signal f f f is limited.
Also, a similar formula applies to the frequency function f ^ ∈ L 2 \hat{f} \in L^{2} f ^ ∈ L 2 and time-limited signal f ( t ) f(t) f ( t ) .
Frequency Sampling Theorem Assume that f ^ ∈ L 2 \hat{f} \in L^{2} f ^ ∈ L 2 and f ( t ) = 0 for ∣ t ∣ ≥ L f (t) = 0\ \text{for } | t | \ge L f ( t ) = 0 for ∣ t ∣ ≥ L . Then f ^ ( ω ) \hat{f}(\omega) f ^ ( ω ) is determined by the values in n π / L ( n = 0 , ± 1 , ± 2 , … ) n\pi / L(n=0, \pm 1, \pm 2, \dots) nπ / L ( n = 0 , ± 1 , ± 2 , … ) . That is, the following applies.
f ^ ( ω ) = ∑ n = − ∞ ∞ f ^ ( n π L ) sin ( L ω − n π ) L ω − n π
\hat{f}(\omega) = \sum \limits_{n = -\infty}^{\infty} \hat{f} \left( \dfrac{n\pi}{L}\right) \dfrac{\sin (L \omega -n\pi )}{L \omega - n\pi}
f ^ ( ω ) = n = − ∞ ∑ ∞ f ^ ( L nπ ) L ω − nπ sin ( L ω − nπ )
The proof is similar to that of the time sampling theorem.
Proof Since f ^ ∈ L 1 \hat{f} \in L^{1} f ^ ∈ L 1 , f ^ \hat{f} f ^ can be expressed as a Fourier series. f ^ ( ω ) \hat{f} (\omega) f ^ ( ω ) is defined at [ − Ω , Ω ] [-\Omega, \Omega] [ − Ω , Ω ] , so the complex Fourier series is as follows.
f ^ ( ω ) = ∑ − ∞ ∞ c − n e − i n π ω / Ω ( ∣ ω ∣ ≤ Ω )
\begin{equation}
\hat{f}(\omega)=\sum_{-\infty}^{\infty} c_{-n} e^{-i n \pi \omega / \Omega} \quad(|\omega| \leq \Omega)
\label{eq1}
\end{equation}
f ^ ( ω ) = − ∞ ∑ ∞ c − n e − inπω /Ω ( ∣ ω ∣ ≤ Ω )
For convenience in the proof process, n n n is denoted by − n -n − n . The coefficient c − n c_{-n} c − n is as follows.
c − n = 1 2 Ω ∫ − Ω Ω f ^ ( ω ) e i n π ω / Ω d ω = 1 2 Ω ∫ − ∞ ∞ f ^ ( ω ) e i n π ω / Ω d ω = π Ω 1 2 π ∫ − ∞ ∞ f ^ ( ω ) e i ω ( n π / Ω ) d ω
\begin{align*}
c_{-n} &= \dfrac{1}{2 \Omega} \int_{-\Omega}^{\Omega} \hat{f}(\omega) e^{i n \pi \omega / \Omega} d \omega
\\ &= \dfrac{1}{2 \Omega} \int_{-\infty}^{\infty} \hat{f}(\omega) e^{i n \pi \omega / \Omega} d \omega
\\ &= \dfrac{\pi}{\Omega} \dfrac{1}{2\pi} \int_{-\infty}^{\infty} \hat{f}(\omega) e^{i \omega (n \pi / \Omega)} d\omega
\end{align*}
c − n = 2Ω 1 ∫ − Ω Ω f ^ ( ω ) e inπω /Ω d ω = 2Ω 1 ∫ − ∞ ∞ f ^ ( ω ) e inπω /Ω d ω = Ω π 2 π 1 ∫ − ∞ ∞ f ^ ( ω ) e iω ( nπ /Ω ) d ω
Since f ^ ( ω ) = 0 f o r ∣ ω ∣ ≥ Ω \hat{f} (\omega) = 0\ for\ | \omega | \ge \Omega f ^ ( ω ) = 0 f or ∣ ω ∣ ≥ Ω , then f ^ ( ω ) ∈ L 2 \hat{f} (\omega) \in L^{2} f ^ ( ω ) ∈ L 2 , and by the Fourier inverse transform theorem, the following applies.
f ( t ) = 1 2 π ∫ − ∞ ∞ f ^ ( ω ) e i ω t d ω
\begin{equation}
f(t) =\dfrac{1}{2\pi} \int _{-\infty} ^{\infty}\hat{f}(\omega) e^{i\omega t}d\omega
\label{eq2}
\end{equation}
f ( t ) = 2 π 1 ∫ − ∞ ∞ f ^ ( ω ) e iω t d ω
Therefore, one obtains the following equation.
c − n = π Ω 1 2 π ∫ − ∞ ∞ f ^ ( ω ) e i ω ( n π / Ω ) d ω = π Ω f ( n π Ω )
\begin{align*}
c_{-n} &= \dfrac{\pi}{\Omega} \dfrac{1}{2\pi} \int_{-\infty}^{\infty} \hat{f}(\omega) e^{i \omega (n \pi / \Omega)} d\omega
\\ &= \dfrac{\pi}{\Omega} f \left( \dfrac{n\pi}{\Omega}\right)
\end{align*}
c − n = Ω π 2 π 1 ∫ − ∞ ∞ f ^ ( ω ) e iω ( nπ /Ω ) d ω = Ω π f ( Ω nπ )
Inserting this into ( eq1 ) \eqref{eq1} ( eq1 ) yields the following.
f ^ ( ω ) = ∑ − ∞ ∞ c − n e − i n π ω / Ω ( ∣ ω ∣ ≤ Ω ) = ∑ − ∞ ∞ π Ω f ( n π Ω ) e − i n π ω / Ω ( ∣ ω ∣ ≤ Ω )
\hat{f}(\omega) = \sum_{-\infty}^{\infty} c_{-n} e^{-i n \pi \omega / \Omega} \quad(|\omega| \leq \Omega) = \sum_{-\infty}^{\infty} \dfrac{\pi}{\Omega} f \left( \dfrac{n\pi}{\Omega}\right) e^{-i n \pi \omega / \Omega} \quad(|\omega| \leq \Omega)
f ^ ( ω ) = − ∞ ∑ ∞ c − n e − inπω /Ω ( ∣ ω ∣ ≤ Ω ) = − ∞ ∑ ∞ Ω π f ( Ω nπ ) e − inπω /Ω ( ∣ ω ∣ ≤ Ω )
Again inserting this into ( eq2 ) \eqref{eq2} ( eq2 ) yields the equation below.
f ( t ) = 1 2 π ∫ − ∞ ∞ f ^ ( ω ) e i ω t d ω = 1 2 π ∫ − Ω Ω f ^ ( ω ) e i ω t d ω = 1 2 Ω ∫ − Ω Ω ∑ − ∞ ∞ f ( n π Ω ) e − i n π ω / Ω e i ω t d ω
\begin{align*}
f(t) &= \frac{1}{2 \pi} \int_{-\infty}^{\infty} \hat{f}(\omega) e^{i \omega t} d \omega = \frac{1}{2 \pi} \int_{-\Omega}^{\Omega} \hat{f}(\omega) e^{i \omega t} d \omega
\\ &=\frac{1}{2 \Omega} \int_{-\Omega}^{\Omega} \sum_{-\infty}^{\infty} f\left(\frac{n \pi}{\Omega}\right) e^{-i n \pi \omega / \Omega} e^{i \omega t} d \omega
\end{align*}
f ( t ) = 2 π 1 ∫ − ∞ ∞ f ^ ( ω ) e iω t d ω = 2 π 1 ∫ − Ω Ω f ^ ( ω ) e iω t d ω = 2Ω 1 ∫ − Ω Ω − ∞ ∑ ∞ f ( Ω nπ ) e − inπω /Ω e iω t d ω
In the function space , the inner product is defined as a definite integral, and since the inner product is continuous, the limit can be taken out, so the equation is as follows.
f ( t ) = 1 2 Ω ∫ − Ω Ω ∑ − ∞ ∞ f ( n π Ω ) e − i n π ω / Ω e i ω t d ω = 1 2 Ω ⟨ ∑ − ∞ ∞ f ( n π Ω ) e − i n π ω / Ω , e − i ω t ⟩ = 1 2 Ω ∑ − ∞ ∞ ⟨ f ( n π Ω ) e − i n π ω / Ω , e − i ω t ⟩ = 1 2 Ω ∑ − ∞ ∞ ∫ − Ω Ω f ( n π Ω ) e − i n π ω / Ω e i ω t d ω = 1 2 Ω ∑ − ∞ ∞ f ( n π Ω ) ∫ − Ω Ω e − i n π ω / Ω e i ω t d ω
\begin{align*}
f(t) &= \frac{1}{2 \Omega} \int_{-\Omega}^{\Omega} \sum_{-\infty}^{\infty} f\left(\frac{n \pi}{\Omega}\right) e^{-i n \pi \omega / \Omega} e^{i \omega t} d \omega
\\ &= \frac{1}{2 \Omega} \left\langle \sum_{-\infty}^{\infty} f\left(\frac{n \pi}{\Omega}\right) e^{-i n \pi \omega / \Omega}, e^{-i \omega t} \right\rangle
\\ &= \frac{1}{2 \Omega} \sum_{-\infty}^{\infty} \left\langle f\left(\frac{n \pi}{\Omega}\right) e^{-i n \pi \omega / \Omega}, e^{-i \omega t} \right\rangle
\\ &= \frac{1}{2 \Omega} \sum_{-\infty}^{\infty} \int_{-\Omega}^{\Omega} f\left(\frac{n \pi}{\Omega}\right) e^{-i n \pi \omega / \Omega} e^{i \omega t} d \omega
\\ &= \frac{1}{2 \Omega} \sum_{-\infty}^{\infty} f\left(\frac{n \pi}{\Omega}\right) \int_{-\Omega}^{\Omega} e^{-i n \pi \omega / \Omega} e^{i \omega t} d \omega
\end{align*}
f ( t ) = 2Ω 1 ∫ − Ω Ω − ∞ ∑ ∞ f ( Ω nπ ) e − inπω /Ω e iω t d ω = 2Ω 1 ⟨ − ∞ ∑ ∞ f ( Ω nπ ) e − inπω /Ω , e − iω t ⟩ = 2Ω 1 − ∞ ∑ ∞ ⟨ f ( Ω nπ ) e − inπω /Ω , e − iω t ⟩ = 2Ω 1 − ∞ ∑ ∞ ∫ − Ω Ω f ( Ω nπ ) e − inπω /Ω e iω t d ω = 2Ω 1 − ∞ ∑ ∞ f ( Ω nπ ) ∫ − Ω Ω e − inπω /Ω e iω t d ω
Calculating the integral above yields the following.
∫ − Ω Ω e − i n π ω / Ω e i ω t d ω = ∫ − Ω Ω e i ( Ω t − n π ) ω / Ω d ω = e i ( Ω t − n π ) ω / Ω i ( Ω t − n π ) / Ω ∣ − Ω Ω = 1 i ( Ω t − n π ) / Ω [ e i ( Ω t − n π ) − e − i ( Ω t − n π ) ] = 2 ( Ω t − n π ) / Ω e i ( Ω t − n π ) − e − i ( Ω t − n π ) 2 i = 2 sin ( Ω t − n π ) ( Ω t − n π ) / Ω
\begin{align*}
\int_{-\Omega}^{\Omega} e^{-i n \pi \omega / \Omega} e^{i \omega t} d \omega &= \int_{-\Omega}^{\Omega} e^{i(\Omega t- n \pi )\omega/\Omega} d \omega
\\ &= \left. \frac{e^{i(\Omega t-n \pi) \omega / \Omega}}{i(\Omega t-n \pi) / \Omega}\right|_{-\Omega} ^{\Omega}
\\ &= \frac{1}{i(\Omega t-n \pi) / \Omega}\left[ e^{i(\Omega t-n \pi)} - e^{-i(\Omega t-n \pi)} \right]
\\ &= \frac{2}{(\Omega t-n \pi) / \Omega}\dfrac{ e^{i(\Omega t-n \pi)} - e^{-i(\Omega t-n \pi)}}{2i}
\\ &= \frac{2 \sin (\Omega t-n \pi)}{(\Omega t-n \pi) / \Omega}
\end{align*}
∫ − Ω Ω e − inπω /Ω e iω t d ω = ∫ − Ω Ω e i ( Ω t − nπ ) ω /Ω d ω = i ( Ω t − nπ ) /Ω e i ( Ω t − nπ ) ω /Ω − Ω Ω = i ( Ω t − nπ ) /Ω 1 [ e i ( Ω t − nπ ) − e − i ( Ω t − nπ ) ] = ( Ω t − nπ ) /Ω 2 2 i e i ( Ω t − nπ ) − e − i ( Ω t − nπ ) = ( Ω t − nπ ) /Ω 2 sin ( Ω t − nπ )
Substituting and rearranging yields the following.
f ( t ) = 1 2 Ω ∑ − ∞ ∞ f ( n π Ω ) ∫ − Ω Ω e − i n π ω / Ω e i ω t d ω = 1 2 Ω ∑ − ∞ ∞ f ( n π Ω ) 2 sin ( Ω t − n π ) ( Ω t − n π ) / Ω = ∑ − ∞ ∞ f ( n π Ω ) 2 sin ( Ω t − n π ) Ω t − n π
\begin{align*}
f(t) &= \frac{1}{2 \Omega} \sum_{-\infty}^{\infty} f\left(\frac{n \pi}{\Omega}\right) \int_{-\Omega}^{\Omega} e^{-i n \pi \omega / \Omega} e^{i \omega t} d \omega
\\ &= \frac{1}{2 \Omega} \sum_{-\infty}^{\infty} f\left(\frac{n \pi}{\Omega}\right) \frac{2 \sin (\Omega t-n \pi)}{(\Omega t-n \pi) / \Omega}
\\ &= \sum_{-\infty}^{\infty} f\left(\frac{n \pi}{\Omega}\right) \frac{2 \sin (\Omega t-n \pi)}{\Omega t-n \pi}
\end{align*}
f ( t ) = 2Ω 1 − ∞ ∑ ∞ f ( Ω nπ ) ∫ − Ω Ω e − inπω /Ω e iω t d ω = 2Ω 1 − ∞ ∑ ∞ f ( Ω nπ ) ( Ω t − nπ ) /Ω 2 sin ( Ω t − nπ ) = − ∞ ∑ ∞ f ( Ω nπ ) Ω t − nπ 2 sin ( Ω t − nπ )
■