(10 intermediate revisions by 3 users not shown)
Line 19: Line 19:
 
</math>  
 
</math>  
  
 +
 +
<span style="color:red"> This is the long way. Do not do this if you can help it!!! </span>
 
The period of the input is N, so we will calculate the N-point DFT:
 
The period of the input is N, so we will calculate the N-point DFT:
  
Line 29: Line 31:
 
</math>
 
</math>
  
 +
<span style="color:red"> This is the short way: write your signal as a sum of complex exponentials, and then compare with IDFT formula to extract the DFT coefficients.</span>
 +
 +
 +
<math>
 +
\begin{align}
 +
x[n] &=s_N[n] \text{ (Remember, that function we defined when looking at downsampling in the frequency domain?) } \\
 +
&=\frac{1}{N} \sum_{k=0}^{N-1} e^{jk \frac{2\pi}{N} n} \text{ (Writing }s_N[n] \text{ as its Fourier series.)} \\
 +
\end{align}
 +
</math>
 +
 +
By comparison with the inverse-DFT expression for x[n], namely
 +
 +
<math> x[n]=\frac{1}{N}\sum_{k=0}^{N-1} X[k] e^{j \frac{2\pi}{N} kn }</math>
 +
 +
we find that <math> X[k]=1</math>, for k=1,…,N-1. Using the periodicity of X[k] (period N), we conclude that X[k]=1, for all k.
  
 
b) <math>x[n]= e^{j \frac{2}{5} \pi n}</math>
 
b) <math>x[n]= e^{j \frac{2}{5} \pi n}</math>
Line 37: Line 54:
 
\begin{align}
 
\begin{align}
 
x[n]&=\frac{1}{5} \sum_{k=0}^{4} X_5[k] e^{j2\pi kn/5} \\
 
x[n]&=\frac{1}{5} \sum_{k=0}^{4} X_5[k] e^{j2\pi kn/5} \\
&= \frac{1}{5} \left ( X_5[0]e^{j2\pi k0/5}  + X_5[1]e^{j2\pi k1/5} + X_5[2]e^{j2\pi k2/5} + X_5[3]e^{j2\pi k3/5} + X_5[4]e^{j2\pi k4/5}  \right ) \\
+
&= \frac{1}{5} \left ( X_5[0]e^{j2\pi n0/5}  + X_5[1]e^{j2\pi n1/5} + X_5[2]e^{j2\pi n2/5} + X_5[3]e^{j2\pi n3/5} + X_5[4]e^{j2\pi n4/5}  \right ) \\
 
&= e^{j2\pi n/5}  
 
&= e^{j2\pi n/5}  
 
\end{align}
 
\end{align}
Line 59: Line 76:
 
Notice that the period is 5, so we will calculate the 5-point DFT. Beginning with the inverse-DFT:
 
Notice that the period is 5, so we will calculate the 5-point DFT. Beginning with the inverse-DFT:
  
<math>x[n]= e^{-j \frac{2}{5} \pi n} = e^{-j \frac{2}{5} \pi n} e^{2\pi n} = e^{j2\pi n \frac{4}{5}} </math>
+
<math>x[n]= e^{-j \frac{2}{5} \pi n} = e^{-j \frac{2}{5} \pi n} e^{j 2\pi n} = e^{j2\pi n \frac{4}{5}} </math>
  
 
<math>
 
<math>
 
\begin{align}
 
\begin{align}
x[n]&=\frac{1}{5} \sum_{k=0}^{4} X_5[k] e^{j2\pi kn} \\
+
x[n]&=\frac{1}{5} \sum_{k=0}^{4} X_5[k] e^{j2\pi kn/5} \\
&= \frac{1}{5} \left ( X_5[0]e^{j2\pi n0 \frac{4}{5}}  + X_5[1]e^{j2\pi n1 \frac{4}{5}} + X_5[2]e^{j2\pi n2 \frac{4}{5}} + X_5[3]e^{j2\pi n3 \frac{4}{5}} + X_5[4]e^{-j2\pi k4 \frac{4}{5}}  \right ) \\
+
&= \frac{1}{5} \left ( X_5[0]e^{j2\pi n0/5}  + X_5[1]e^{j2\pi n1/5} + X_5[2]e^{j2\pi n2/5 } + X_5[3]e^{j2\pi n3/5} + X_5[4]e^{j2\pi n4/5}  \right ) \\
 
&= e^{j2\pi n \frac{4}{5}}
 
&= e^{j2\pi n \frac{4}{5}}
 
\end{align}
 
\end{align}
Line 78: Line 95:
  
 
<math>
 
<math>
X_5[k]=\begin{cases} 5\mbox{, }k=1\\ 0\mbox{, } k=0, 2, 3, 4 \end{cases} \mbox{ , periodic with } = 5
+
X_5[k]=\begin{cases} 5\mbox{, }k=4\\ 0\mbox{, } k=0, 1, 2, 3 \end{cases} \mbox{ , periodic with } = 5
 
</math>
 
</math>
  
Line 87: Line 104:
  
  
e) <math>x[n]= \cos\left( \frac{2}{1000} \pi n\right) ;</math>
+
e) <math class="inline">x[n]= e^{j \frac{\pi}{3} n } \cos ( \frac{\pi}{6} n )</math>
 
+
First, use Euler's formula
+
 
+
<math> x[n]=\frac{1}{2} e^{j2\pi n/1000} + \frac{1}{2}e^{-j2\pi n/1000} </math>
+
 
+
As in d), change the exponents so they are both positive:
+
 
+
<math>\begin{align}
+
x[n]&=\frac{1}{2} e^{j2\pi n/1000} + \frac{1}{2}e^{-j2\pi n/1000}e^{2\pi n} \\
+
&=\frac{1}{2} e^{j2\pi n/1000} + \frac{1}{2}e^{j2\pi 999n/1000}
+
\end{align}</math>
+
 
+
Then looking at the 1000-point inverse DFT:
+
 
+
<math>\begin{align}
+
x[n] &= \frac{1}{1000} \sum_{k=0}^{999} X_{1000}[k]e^{j2\pi kn/1000} \\
+
&= \frac{1}{2} e^{j2\pi n/1000} + \frac{1}{2}e^{j2\pi 999n/1000}
+
\end{align}</math>
+
 
+
Again, by matching terms we can see that
+
 
+
<math>
+
X_{1000}[k] = \begin{cases} 500&\mbox{, if } k=1 \mbox{ or }k=999 \\ 0 &\mbox{, if } k=0 \mbox { or } k=2,\ldots,998 \end{cases} \mbox{, periodic with period } 1000
+
</math>
+
 
+
 
+
f) <math class="inline">x[n]= e^{j \frac{\pi}{3} n } \cos ( \frac{\pi}{6} n )</math>
+
  
 
Using Euler's formula
 
Using Euler's formula
Line 136: Line 126:
  
  
g) <math>x[n]= (-j)^n .</math>
+
f) <math>x[n]= (-j)^n .</math>
  
 
First, rewrite the signal as
 
First, rewrite the signal as
Line 165: Line 155:
 
<math>X_4[k]=\begin{cases} 4 &\mbox{, if }k=3 \\ 0 &\mbox{, if } k=0,1,2 \end{cases} \mbox{, periodic with period } 4</math>
 
<math>X_4[k]=\begin{cases} 4 &\mbox{, if }k=3 \\ 0 &\mbox{, if } k=0,1,2 \end{cases} \mbox{, periodic with period } 4</math>
  
h) <math class="inline">x[n] =(\frac{1}{\sqrt{2}}+j \frac{1}{\sqrt{2}})^n </math>
+
 
 +
g) <math class="inline">x[n] =(\frac{1}{\sqrt{2}}+j \frac{1}{\sqrt{2}})^n </math>
  
 
First, rewrite the signal
 
First, rewrite the signal
Line 189: Line 180:
 
==Question 2 ==
 
==Question 2 ==
 
Compute the inverse DFT of  <math class="inline">X[k]= e^{j \pi k }+e^{-j \frac{\pi}{2} k} </math>.
 
Compute the inverse DFT of  <math class="inline">X[k]= e^{j \pi k }+e^{-j \frac{\pi}{2} k} </math>.
 +
 +
''' Solution '''
  
 
Rewrite so that the exponents are negative:
 
Rewrite so that the exponents are negative:
Line 251: Line 244:
 
The change in the limits follows because <math>x[n'] e^{j2\pi k n'/N} </math> has a period of N. If we're summing over one full period, it doesn't matter where we start the summation.
 
The change in the limits follows because <math>x[n'] e^{j2\pi k n'/N} </math> has a period of N. If we're summing over one full period, it doesn't matter where we start the summation.
  
 +
 +
----
 +
 +
==Question 4==
 +
 +
Under which circumstances can one recover the DTFT of a finite duration signal from the DFT of its periodic repetition? Justify your answer mathematically.
 +
 +
''' Solution '''
 +
 +
Suppose the length of finite duration signal <math>x[n]</math> is <math>M</math>. The number of points of its DFT is <math>N</math>.
 +
 +
Using IDFT we have
 +
 +
<math>x'[n]=\frac{1}{N}\sum_{k=0}^{N-1}X(k)e^{\frac{j2\pi nk}{N}}</math>
 +
 +
Noticing that <math>x'[n]=x'[n+N]</math>, so the <math>x'[n]</math> obtained by IDFT is periodic with <math>N</math>.
 +
 +
We can reconstruct DTFT by substituting <math>x[n]</math> by <math>x'[n]</math> as long as <math>x[n]=x'[n]</math> for <math>n=0,1,...,M-1</math>. i.e.
 +
 +
<math>\begin{align}
 +
X(e^{j\omega}) &= \sum_{n=0}^{M-1}x[n]e^{-j\omega n} \\
 +
&= \sum_{n=0}^{M-1}x'[n]e^{-j\omega n} \\
 +
&= \sum_{n=0}^{M-1}\frac{1}{N}\sum_{k=0}^{N-1}X(k)e^{\frac{j2\pi nk}{N}}e^{-j\omega n}
 +
\end{align}</math>
 +
 +
Since <math>x'[n]</math> is periodic with <math>N</math>, we must guarantee that <math>N\ge M</math> in order to fully reconstruct DTFT using DFT.
  
 
----
 
----

Latest revision as of 14:36, 20 October 2015


Homework 6 Solution, ECE438, Fall 2015, Prof. Boutin

Questions 1

Compute the DFT of the following signals x[n] (if possible). How does your answer relate to the Fourier series coefficients of x[n]?

a) $ x[n] = \left\{ \begin{array}{ll} 1, & n \text{ multiple of } N\\ 0, & \text{ else}. \end{array} \right. $


This is the long way. Do not do this if you can help it!!! The period of the input is N, so we will calculate the N-point DFT:

$ \begin{align} X_n[k]&=\sum_{n=0}^{N-1} x[n] e^{-j2\pi kn /N} \\ &= 1e^{-j2\pi k 0 /N} + 0e^{-j2\pi k1 /N} + \ldots + 0e^{-j2\pi k(N-1) /N} \\ &= 1 \text{ for all } k \end{align} $

This is the short way: write your signal as a sum of complex exponentials, and then compare with IDFT formula to extract the DFT coefficients.


$ \begin{align} x[n] &=s_N[n] \text{ (Remember, that function we defined when looking at downsampling in the frequency domain?) } \\ &=\frac{1}{N} \sum_{k=0}^{N-1} e^{jk \frac{2\pi}{N} n} \text{ (Writing }s_N[n] \text{ as its Fourier series.)} \\ \end{align} $

By comparison with the inverse-DFT expression for x[n], namely

$ x[n]=\frac{1}{N}\sum_{k=0}^{N-1} X[k] e^{j \frac{2\pi}{N} kn } $

we find that $ X[k]=1 $, for k=1,…,N-1. Using the periodicity of X[k] (period N), we conclude that X[k]=1, for all k.

b) $ x[n]= e^{j \frac{2}{5} \pi n} $

Notice that the period is 5, so we will calculate the 5-point DFT. Beginning with the inverse-DFT:

$ \begin{align} x[n]&=\frac{1}{5} \sum_{k=0}^{4} X_5[k] e^{j2\pi kn/5} \\ &= \frac{1}{5} \left ( X_5[0]e^{j2\pi n0/5} + X_5[1]e^{j2\pi n1/5} + X_5[2]e^{j2\pi n2/5} + X_5[3]e^{j2\pi n3/5} + X_5[4]e^{j2\pi n4/5} \right ) \\ &= e^{j2\pi n/5} \end{align} $

From this we can see that

$ X_5[1]=5 \mbox{, and } X_5[0]=X_5[2]=X_5[3]=X_5[4]=0 $

or

$ X_5[k]=\begin{cases} 5\mbox{, }k=1\\ 0\mbox{, } k=0, 2, 3, 4 \end{cases} \mbox{ , periodic with } = 5 $


c) $ x[n]= e^{-j \frac{2}{5} \pi n} $

Notice that the period is 5, so we will calculate the 5-point DFT. Beginning with the inverse-DFT:

$ x[n]= e^{-j \frac{2}{5} \pi n} = e^{-j \frac{2}{5} \pi n} e^{j 2\pi n} = e^{j2\pi n \frac{4}{5}} $

$ \begin{align} x[n]&=\frac{1}{5} \sum_{k=0}^{4} X_5[k] e^{j2\pi kn/5} \\ &= \frac{1}{5} \left ( X_5[0]e^{j2\pi n0/5} + X_5[1]e^{j2\pi n1/5} + X_5[2]e^{j2\pi n2/5 } + X_5[3]e^{j2\pi n3/5} + X_5[4]e^{j2\pi n4/5} \right ) \\ &= e^{j2\pi n \frac{4}{5}} \end{align} $

From this we can see that

$ X_5[4]=5 \mbox{, and } X_5[0]=X_5[1]=X_5[2]=X_5[3]=0 $

or

$ X_5[k]=\begin{cases} 5\mbox{, }k=4\\ 0\mbox{, } k=0, 1, 2, 3 \end{cases} \mbox{ , periodic with } = 5 $


d) $ x[n]= e^{j \frac{2}{\sqrt{3}} \pi n} $

The period of the input is $ \sqrt{3} $. We cannot take a $ \sqrt{3} $-point DFT (only integer values).


e) $ x[n]= e^{j \frac{\pi}{3} n } \cos ( \frac{\pi}{6} n ) $

Using Euler's formula

$ \begin{align} x[n] &= e^{j\frac{\pi}{3}n} \left ( \frac{1}{2} e^{j\frac{\pi}{6}} + \frac{1}{2}e^{-j\frac{\pi}{6}}\right ) \\ &= \frac{1}{2}e^{\frac{\pi}{2}n} + \frac{1}{2}e^{j\frac{\pi}{6}n} \\ &= \frac{1}{2}e^{\frac{2\pi}{12}3n} + \frac{1}{2}e^{j\frac{2\pi}{12}n} \mbox{, this will make comparing with the IDFT easier} \end{align} $

The period for the signal is 12. Looking at the 12-point IDFT:

$ \begin{align} x[n]&=\frac{1}{12}\sum_{k=0}^{11} X_{12}[k] e^{j2\pi kn /12} \\ &= \frac{1}{2}e^{j2\pi 3n/12} + \frac{1}{2}e^{j2\pi n/12} \end{align} $

We can see that

$ X_{12}[k]=\begin{cases} 6 &\mbox{, if } k=1 \mbox{ or } k=3 \\ 0 &\mbox{, if} k=0 \mbox{, } k=2 \mbox{, or} k=4,\ldots,11 \end{cases}\mbox{, periodic with period } 12 $


f) $ x[n]= (-j)^n . $

First, rewrite the signal as

$ \begin{align} x[n] &= (-j)^n \\ &= e^{j3\pi n/2} \\ &= e^{j2\pi 3n /4} \mbox{, again, this makes comparison with the IDFT easier} \end{align} $

The period of the signal is 4. You can see this by observing that the sequence is {-j, -1, j, 1, -j, ...}. Or you can find it using the general form $ e^{j\omega_0n} $. Then you can solve for the period M by solving $ \omega_0M=2\pi m $ for some integer m. This signal, for example, would be

$ \frac{3\pi}{2} M = 2\pi n \Rightarrow M=\frac{4}{3}m \mbox{, where M and m are both integers} $

When m=3, M is the integer 4.

So, looking at the 4-point IDFT

$ \begin{align} x[n]&= \sum_{k=0}^{3} X_4[k] e^{j2\pi kn/4} \\ &= e^{j2\pi 3n/4} \end{align} $

From this, we can see that

$ X_4[k]=\begin{cases} 4 &\mbox{, if }k=3 \\ 0 &\mbox{, if } k=0,1,2 \end{cases} \mbox{, periodic with period } 4 $


g) $ x[n] =(\frac{1}{\sqrt{2}}+j \frac{1}{\sqrt{2}})^n $

First, rewrite the signal

$ \begin{align} x[n] &=(\frac{1}{\sqrt{2}}+j \frac{1}{\sqrt{2}})^n \\ &=(e^{j\pi/4})^n \\ &=e^{j2\pi n/8} \end{align} $

Now, looking at the 8-point IDFT

$ \begin{align} x[n] &= \sum_{k=0}^{7} X_8[k] e^{j2\pi kn /8} \\ &=e^{j2\pi n/8} \end{align} $

We can see that

$ X_8[k] = \begin{cases} 8 &\mbox{, if } k=1 \\ 0 &\mbox{, if} k=0 \mbox{ or } k=2,\ldots,7 \end{cases}\mbox{, periodic with period } 8 $


Question 2

Compute the inverse DFT of $ X[k]= e^{j \pi k }+e^{-j \frac{\pi}{2} k} $.

Solution

Rewrite so that the exponents are negative:

$ \begin{align} x[n] &= e^{j\pi k}e^{-j2\pi k} + e^{-j\pi k/2} \\ &= e^{-j\pi k} + e^{-j\pi k/2} \end{align} $

The period of the signal is 4. We can again rewrite the signal so that the period is in the denominator of the exponent (this makes the next steps easier):

$ x[n] = e^{-j2\pi 2 k/4} + e^{-j2\pi k/4} $

Looking at the 4-point DFT

$ \begin{align} X_4[k] &= \sum_{k=0}^{3} x[n] e^{-j2\pi kn /4} \\ &= e^{-j2\pi 2 k/4} + e^{-j2\pi k/4} \end{align} $

As in question 1, we can see that

$ \begin{align} x[n]&=\begin{cases} 1 &\mbox{, if }n=1 \mbox{ or } n=2 \\ 0 &\mbox{, if } n=0 \mbox{ or } n=3 \end{cases} \mbox{, periodic with period } 4 \\ &=\delta(n-1) + \delta(n-2) \mbox{, periodic with 4} \end{align} $


Question 3

Prove the time shifting property of the DFT.

Method 1

Given that

$ X_n[k] = \text{DFT} \left \{ x[n] \right \} $

Then the shifted form can be written as

$ x[n-n_0] = x[n]\ast \delta[n-n_0] $

Then it follows that

$ \begin{align} \text{DFT}\left \{ x[n-n_0] \right \} &= \text{DFT} \left \{ x[n] \right \} \text{DFT} \left \{\delta[n-n_0] \right \} \\ &= X_n[k] e^{-j2\pi kn_0/N} \end{align} $

Method 2

Or, you can use a change of variables. Let $ X_N[k] = \text{DFT} \left \{x[n] \right \} $, then

$ \begin{align} \text{DFT}\left \{ x[n-n_0] \right \} &= \sum_{k=0}^{N-1} x[n-n_0] e^{-j2\pi kn /N} \\ &= \sum_{n'=-n_0}^{N-n_0-1} x[n']e^{-j2\pi k(n' + n_0)/N} \mbox{ using the variable substitution } n'=n-n_0 \\ &=e^{-j2\pi k n_0/N} \sum_{n'=-n_0}^{N-n_0-1}x[n'] e^{-j2\pi k n'/N} \\ &=e^{-j2\pi k n_0/N} \sum_{n'=-0}^{N-1}x[n'] e^{-j2\pi k n'/N} \mbox{ (details below)}\\ &=e^{-j2\pi k n_0/N} X_N[k] \end{align} $

The change in the limits follows because $ x[n'] e^{j2\pi k n'/N} $ has a period of N. If we're summing over one full period, it doesn't matter where we start the summation.



Question 4

Under which circumstances can one recover the DTFT of a finite duration signal from the DFT of its periodic repetition? Justify your answer mathematically.

Solution

Suppose the length of finite duration signal $ x[n] $ is $ M $. The number of points of its DFT is $ N $.

Using IDFT we have

$ x'[n]=\frac{1}{N}\sum_{k=0}^{N-1}X(k)e^{\frac{j2\pi nk}{N}} $

Noticing that $ x'[n]=x'[n+N] $, so the $ x'[n] $ obtained by IDFT is periodic with $ N $.

We can reconstruct DTFT by substituting $ x[n] $ by $ x'[n] $ as long as $ x[n]=x'[n] $ for $ n=0,1,...,M-1 $. i.e.

$ \begin{align} X(e^{j\omega}) &= \sum_{n=0}^{M-1}x[n]e^{-j\omega n} \\ &= \sum_{n=0}^{M-1}x'[n]e^{-j\omega n} \\ &= \sum_{n=0}^{M-1}\frac{1}{N}\sum_{k=0}^{N-1}X(k)e^{\frac{j2\pi nk}{N}}e^{-j\omega n} \end{align} $

Since $ x'[n] $ is periodic with $ N $, we must guarantee that $ N\ge M $ in order to fully reconstruct DTFT using DFT.


Discussion

You may discuss the homework below.

  • write comment/question here
    • answer will go here

Back to Homework6

Back to ECE438, Fall 2015, Prof. Boutin

Alumni Liaison

Ph.D. 2007, working on developing cool imaging technologies for digital cameras, camera phones, and video surveillance cameras.

Buyue Zhang