1
$\begingroup$

I found the answers to this question and this question to be extremely helpful in understanding the derivation of the downsampling or decimation property of the DTFT. Thank you! I am now struggling to come up with a good example that shows clearly the role of the frequency-shifted term(s).

To summarize, let $y[n]$ be a discrete-time signal formed by downsampling $x[n]$ by an integer factor $M$, $$ y[n] = x[Mn]. $$ Then the discrete-time Fourier transforms of $y[n]$ and $x[n]$ are related through $$ Y \left( e^{j \omega} \right) = \frac{1}{M} \sum_{k=0}^{M-1} X \left( e^{j (\omega - 2 \pi k)/M} \right). $$

The simplest case is for $M=2$, $$ Y \left( e^{j \omega} \right) = \frac{1}{2} \, X \left( e^{j \omega/2} \right) + \frac{1}{2} \, X \left( e^{j (\omega/2 - \pi)} \right) . $$

I want to develop an example that clearly shows the role of the second term. I have already derived the result for a causal exponential signal, but for that example the role of the second term is not at all obvious from a plot of the spectrum. Therefore, let $x[n]$ be a sinc signal, $$ x[n] = \frac{\sin(\omega_0 n)}{\pi n} \qquad \Rightarrow \qquad X \left( e^{j \omega} \right) = \begin{cases} 1, & |\omega| < \omega_0 \\ 0, & \omega_0 < |\omega| < \pi. \end{cases} $$ I am suppressing the periodic repetition of the rectangular spectrum here for simplicity.

Assume for the moment that $\omega_0 < \pi/4$. The computation of the DTFT of $y[n]$ by the downsampling property gives $$ \begin{array}{rclcl} Y_1 \left( e^{j \omega} \right) & = & \dfrac{1}{2} \, X \left( e^{j \omega/2} \right) & = & \begin{cases} \frac{1}{2}, & |\omega| < 2 \omega_0 \\ 0, & 2 \omega_0 < |\omega| < \pi. \end{cases} \\ \\ Y_2 \left( e^{j \omega} \right) & = & \dfrac{1}{2} \, X \left( e^{j (\omega - 2 \pi)/2} \right) & = & \begin{cases} \frac{1}{2}, & \pi - 2 \omega_0 < |\omega| < \pi\\ 0, & |\omega| < \pi - 2 \omega_0. \end{cases}\\ \\ Y \left( e^{j \omega} \right) & = & Y_1 \left( e^{j \omega} \right) + Y_2 \left( e^{j \omega} \right) & = & \begin{cases} \frac{1}{2}, & |\omega| < 2 \omega_0 \\ \frac{1}{2}, & \pi - 2 \omega_0 < |\omega| < \pi\\ 0, & 2 \omega_0 < |\omega| < \pi - 2 \omega_0. \end{cases}\\ \end{array} $$

This expression tells us that the DTFT of $y[n]$ consists of two rectangular spectra, one centered around $\omega = 0$, and the other around $\omega = \pm \pi$. This would seem to be exactly what I am looking for. However, if we return to the original expression for $y[n]$, $$ y[n] = x[2n] = \frac{\sin(2 \omega_0 n)}{2 \pi n} . $$

Therefore, the DTFT of $y[n]$ should be $$ Y \left( e^{j \omega} \right) = \begin{cases} \frac{1}{2}, & |\omega| < 2 \omega_0 \\ 0, & 2 \omega_0 < |\omega| < \pi, \end{cases} $$

which contains only the low-frequency rectangular component. How can I resolve this apparent contradiction?

$\endgroup$
1
$\begingroup$

You got the term $X \left( e^{j (\omega - 2 \pi)/2} \right)$ wrong. It is centered at $2\pi$ and it is non-zero in the interval $(2\pi-2\omega_0,2\pi+2\omega_0)$. So the results obtained in the frequency domain and in the time domain, respectively, are identical.

Note that the term $X \left( e^{j \omega/2} \right)$ is $4\pi$-periodic, so the term $X \left( e^{j (\omega - 2 \pi)/2} \right)$ makes sure that the spectrum of the downsampled signal is $2\pi$-periodic. Clearly, there is no aliasing as long as $\omega_0<\pi/2$ is satsified.

| improve this answer | |
$\endgroup$
  • $\begingroup$ Oh dear. You're right of course, thank you for the help. Then in the general case $X \left( e^{j \omega/M} \right)$ is $2M \pi$-periodic, and the $M-1$ frequency-shifted terms are required to make $Y \left( e^{j \omega} \right)$ $2 \pi$-periodic. This in turn means that any example showing the effect of the shifted spectral copies would be the result of aliasing, i.e. that $X \left( e^{j \omega} \right)$ has some nonzero-value for $\omega > \pi/M$. That makes sense of course. $\endgroup$ – Steve J. Jul 3 at 17:50
  • $\begingroup$ @SteveJ.: That's right! $\endgroup$ – Matt L. Jul 3 at 17:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.