1
$\begingroup$

I have a table with transform operations, e.g. scaling:

\begin{equation} \tag{0} \label{0} x(at) \iff \frac{1}{a} X(\frac{\omega}{a}) \end{equation}

or frequency shifting:

\begin{equation}\tag{1} \label{1} x(t)e^{jw_ot} \iff X(\omega - \omega_o) \end{equation}

I also have a table with transform pairs, for example:

\begin{equation}\tag{2} \label{2} e^{-a|t|} \iff \frac{2a}{a^2+\omega^2} \end{equation}

Now, I would like to use these tables but with hertz $f$ instead. Note that there are many more operations and transform pairs in the tables than I have mentioned. I hope those can be translated similary as the examples I have given. Is there an easy way to do this?

For instance, I have tried to find the inverse transformation of $X(f - f_o)$ using equation $\eqref{0}$ and $\eqref{1}$

\begin{equation} X(f - f_o) =^? X(\frac{\omega - \omega_o}{2\pi}) \iff //eq. \eqref{0}// \iff 2\pi \cdot x(2\pi t)\cdot e^{j2\pi w_ot} = \end{equation}

\begin{equation}\tag{3} \label{3} = 2\pi \cdot x(2\pi t)\cdot e^{j4\pi^2 f_ot} \end{equation}

but equation $\eqref{3}$ gives \begin{equation}\tag{4} \label{4} \frac{1}{2}(X(f - f_o) + X(f + f_o)) \iff 2\pi \cdot x(2\pi t)\cdot \frac{e^{j4\pi^2 f_ot} + e^{-j4\pi^2 f_ot}}{2} \end{equation}

which simplifies to

\begin{equation}\tag{5} \label{5} \frac{1}{2}(X(f - f_o) + X(f + f_o)) \iff 2\pi \cdot x(2\pi t)\cdot \cos(j4\pi^2 f_ot) \end{equation}

which doesn't equal the correct expression

\begin{equation}\tag{6} \label{6} \frac{1}{2}(X(f - f_o) + X(f + f_o)) \iff x(t)\cdot \cos(j2\pi f_ot) \end{equation}.

Even if equation $\eqref{5}$ is correct, there is non-trivial calculation involved. I would like a faster way similar to "replace $\omega$ with $2\pi f$"

Thanks

$\endgroup$
3
$\begingroup$

The properties that you cite are for actually changing the frequency components, by stretching/compressing or shifting, not for just changing the units on the axis. When you do the conversion, $\omega=2\pi f$, you are doing a mapping from radians per second to Hertz but the frequencies in the signal are not changing.

enter image description here

When you do the frequency stretching/compressing or shifting, you are truly changing the frequency components of the signal. The best comparison I can think of is to consider loading up barbell with 110 pounds, but then a friend comes over who insists on using kilograms as a measure instead so you take off the 100 pounds and put on 55 kilograms (55 kilograms = 110 pounds). You haven't changed the weights, only the units that it is measured in (this is what is happening when changing from $\omega \rightarrow f$). However, imagine if you actually had loaded with 55 pounds, now you've truly changed something! You've scaled the amount of weight (this is what is happening when changing from $\omega \rightarrow \frac{\omega}{a}$).

enter image description here

To more directly answer your question: If you call the Fourier transform of $x(t)$ using radians as $X_{\omega}(\omega)=\int x(t) e^{-j\omega t} dt$ and the Fourier transform using $\text{Hz}$ as $X_f(f)=\int x(t) e^{-j 2\pi ft} dt$, then your question boils down to showing that $X_{\omega}(\omega)=X_f(f)$ for each property. First replace $f$ with $\frac{\omega}{2\pi}$, then simplify.

$$ \begin{align} X_{\omega}(\omega-\omega_0) &= X_f(f-f_0) \\ &= X_f(\frac{\omega-\omega_0}{2\pi}) \\ &= \int x(t) e^{-j2\pi \big(\frac{\omega-\omega_0}{2\pi}\big) t} dt \\ &= \int x(t) e^{-j (\omega-\omega_0) t} dt \\ &= X_{\omega}(\omega-\omega_0) \end{align} $$

And you can do a similar thing for the scaling property.

$$ \begin{align} \frac{1}{a}X_{\omega}\bigg(\frac{\omega}{a}\bigg) &= \frac{1}{a}X_f\bigg(\frac{f}{a}\bigg) \\ &= \frac{1}{a}X_f\bigg(\frac{\omega}{2\pi a}\bigg) \\ &= \frac{1}{a}\int x(t) e^{-j2\pi \big(\frac{\omega}{2\pi a}\big) t} dt \\ &= \frac{1}{a}\int x(t) e^{-j \big(\frac{\omega}{a}\big) t} dt \\ &= \frac{1}{a}X_{\omega}\bigg(\frac{\omega}{a}\bigg) \end{align} $$

| improve this answer | |
$\endgroup$
3
$\begingroup$

Your confusion comes from the fact that you use $X(\cdot)$ for denoting both functions, the function of $\omega$ and the function of $f$, but they are really two different functions, because

$$X(\omega)=\hat{X}(f)=\int_{-\infty}^{\infty}x(t)e^{-j2\pi ft}dt,\quad\omega=2\pi f\tag{1}$$

That's why in all the correspondences you mentioned you can just replace $\omega$ by $f$ inside the function argument (and note that the two frequency domain functions are actually different).

Note, however, that outside the function you have to replace $\omega$ by $2\pi f$, as for instance in the following correspondence:

$$\frac{d^nx(t)}{dt^n}\Longleftrightarrow (j\omega)^nX(\omega)=(j2\pi f)^n\hat{X}(f)\tag{2}$$

There is one more complication that arises due to the transform as defined by $(1)$ being unitary if $\hat{X}(f)$ is used, but being non-unitary when using $X(\omega)$. This can be seen by the factor $1/(2\pi)$ for the inverse transform when using $X(\omega)$. This implies, for instance, that you need a factor $1/(2\pi)$ in the following correspondence when using $X(\omega)$, but not when you use $\hat{X}(f)$:

$$x(t)y(t)\Longleftarrow \frac{1}{2\pi}(X\star Y)(\omega)=(\hat{X}\star\hat{Y})(f)\tag{3}$$

Also be careful when a Dirac delta impulse is involved, because clearly $\delta(\omega)\neq\delta(f)$. The correct correspondence is

$$\delta(f)=2\pi\delta(\omega)\tag{2}$$

Consequently,

$$\mathcal{F}\{1\}=2\pi\delta(\omega)=\delta(f)\tag{3}$$

and

$$\mathcal{F}\{e^{j\omega_0t}\}=\mathcal{F}\{e^{j2\pi f_0t}\}=2\pi \delta(\omega-\omega_0)=\delta(f-f_0)\tag{4}$$

from which the Fourier transforms of sine and cosine follow easily.

| improve this answer | |
$\endgroup$

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.