• Support PF! Buy your school textbooks, materials and every day products via PF Here!

Sawtooth function Fourier transform

1,263
11
1. The problem statement, all variables and given/known data
For a periodic sawtooth function ##f_p (t) = t## of period ##T## defined over the interval ##[0, T]##, calculate the Fourier transform of a function made up of only a single period of ##f_p (t),## i.e.

$$f(t)=\left\{\begin{matrix}f_p (t) \ \ 0<t<T\\0 \ \ elsewhere \end{matrix}\right.$$

Use the result that

$$FT \Big[ f(t) = t \Big] = \frac{j \delta' (\nu)}{2 \pi}$$

2. Relevant equations


3. The attempt at a solution

I am not sure how to approach this problem. I think somehow we need to evaluate ##f(\nu) = \frac{j \delta' (\nu)}{2 \pi}## over the interval ##[0, T]##, but it's not possible since we are in Fourier space. If this is the case, do we need to write the interval in terms of the corresponding frequency (##[\nu = 0, \ 1/T ]##)?

Any help would be greatly appreciated.
 

blue_leaf77

Science Advisor
Homework Helper
2,629
784
Note that ##f(t)## can be obtained by multiplying a function ##y_1(t)=t## and a top hat function
$$
y_2(t)=\left\{\begin{matrix}1 \ \ 0<t<T\\0 \ \ elsewhere \end{matrix}\right.
$$
i.e. ##f(t)=y_1(t)y_2(t)##. The Fourier transform of ##f(t)## is then the convolution between ##\tilde{y}_1(\nu)## and ##\tilde{y}_2(\nu)##. Now, you have been given ##\tilde{y}_1(\nu)##, which is equal to ##\frac{j \delta' (\nu)}{2 \pi}##, and your task now is to find ##\tilde{y}_2(\nu)##.
 
Last edited:
1,263
11
Thank you very much for this hint. So if I write the answer as:

$$F(\nu)=\left\{\begin{matrix}\frac{j \delta'(\nu)}{2 \pi} \ \ 0<t<T\\0 \ \ elsewhere \end{matrix}\right.$$

is that alright?
 

blue_leaf77

Science Advisor
Homework Helper
2,629
784
As I said, the Fourier transform of ##f(t)## will be a convolution between the transform of the individual functions in the time domain. The one you proposed above doesn't seem to be one. Have you calculated the transform of ##y_1## and ##y_2##?
 
1,263
11
I see. So for the case ##0<t<T## the convolution would be:

$$\tilde{y_1} * \tilde{y_2} = \int^T_0 \tilde{y_1} (\tau) \tilde{y_2} (t- \tau) \ d\tau = \frac{j \delta' (\nu)}{2 \pi} \int^T_0 1 \ d \tau = \frac{j \delta' (\nu)}{2 \pi} T$$

So, we have

$$F(\nu)=\left\{\begin{matrix}\frac{j \delta'(\nu)}{2 \pi} T \ \ 0<t<T\\0 \ \ elsewhere \end{matrix}\right.$$

Is this correct now?
 

vela

Staff Emeritus
Science Advisor
Homework Helper
Education Advisor
14,342
1,049
No, it's not correct. The Fourier transforms are functions of frequency, not time.
 

blue_leaf77

Science Advisor
Homework Helper
2,629
784
I see. So for the case ##0<t<T## the convolution would be:

$$\tilde{y_1} * \tilde{y_2} = \int^T_0 \tilde{y_1} (\tau) \tilde{y_2} (t- \tau) \ d\tau = \frac{j \delta' (\nu)}{2 \pi} \int^T_0 1 \ d \tau = \frac{j \delta' (\nu)}{2 \pi} T$$

So, we have

$$F(\nu)=\left\{\begin{matrix}\frac{j \delta'(\nu)}{2 \pi} T \ \ 0<t<T\\0 \ \ elsewhere \end{matrix}\right.$$

Is this correct now?
No, that's wrong. What the convolution theorem states is that when you have two functions ##y_1(t)## and ##y_2(t)## multiplied in the time domain ##f(t)=y_1(t)y_2(t)##, then Fourier transform of the product function in frequency domain will be a convolution of the form
$$
\tilde{f}(\nu) = FT[f(t)] = \int_{-\infty}^\infty \tilde{y}_1(\nu')\tilde{y}_2(\nu-\nu') d\nu'
$$
 
1,263
11
Since ##\tilde{y}_2## is equal to ##1## on (0, T) the convolution would become

$$\int_{-\infty}^\infty \tilde{y}_1(\nu')\tilde{y}_2(\nu-\nu') d\nu' = \frac{j}{2 \pi} \int_{-\infty}^\infty \delta' (\nu') . 1 \ d \nu'=- \frac{j}{2 \pi} \Big[ \delta(\nu') \Big]^{\infty}_{-\infty}$$

But the last expression is equal to zero since the Dirac delta is zero everywhere other than on ##0## (or alternatively it equals to ##1## if 0 is contained within the integration limits).

So ##\tilde{f} (\nu) = 0## in this case? Is that right?
 

blue_leaf77

Science Advisor
Homework Helper
2,629
784
Since ~y2y~2\tilde{y}_2 is equal to 111 on (0, T) the convolution would become
No. It's ##y_2(t)## which is unity in the interval [0,T], not ##\tilde{y}_2(\nu)## which is the Fourier transform of ##y_2(t)##.
 
1,263
11
No. It's ##y_2(t)## which is unity in the interval [0,T], not ##\tilde{y}_2(\nu)## which is the Fourier transform of ##y_2(t)##.
Thanks, so I know that the Fourier transform of the uniform function is equal to ##\delta(\nu)##. So the convolution integral becomes:

$$\frac{j}{2 \pi} \int^\infty_{-\infty} \delta ' (\nu ') \ \delta (\nu - \nu') d \nu '$$

I know that there is a property that says: ##\int^\infty_{-\infty} \delta' (\nu) \phi (\nu ') d \nu' = - \phi' (0)##. Is it possible to use this? Or do I need to proceed with integration by parts?
 

blue_leaf77

Science Advisor
Homework Helper
2,629
784
so I know that the Fourier transform of the uniform function is equal to δ(ν)
##y_2(t)## is not a constant function, it has top hat form and hence its Fourier transform is not equal to a delta function.
 
1,263
11
##y_2(t)## is not a constant function, it has top hat form and hence its Fourier transform is not equal to a delta function.
What do you mean? Because you said earlier that it is equal to unity in the interval [0, T], so the Fourier transform of 1 is just a delta function.
 

blue_leaf77

Science Advisor
Homework Helper
2,629
784
Look at the form of ##y_2(t)## in post#2. Does that look like a constant function over the entire time axis?
 
1,263
11
In post #2 you have ##y_2(t) =1## which is a constant over the interval under consideration (##0<t<T##). So I looked the FT only in this region.

But if we consider the time axis outside this region, the function looks like a rectangular pulse function of duration ##T## (in our course it is denoted by ##\Pi##). In this case it would be ##\Pi(\frac{t}{T})## whose FT is ##T sinc(T \nu)##. Is this what you meant?
 

blue_leaf77

Science Advisor
Homework Helper
2,629
784
Is this what you meant?
Yes. However, there is a small difference. The rectangular pulse we have in this problem is not centered with respect to the origin. If you know the shifting theorem of Fourier transform, this can help you find the corresponding transform.
 
1,263
11
Yes. However, there is a small difference. The rectangular pulse we have in this problem is not centered with respect to the origin. If you know the shifting theorem of Fourier transform, this can help you find the corresponding transform.
Okay so if we rewrite the rectangular function as centered at T/2:

$$\Pi \left( \frac{t- (\frac{T}{2})}{T} \right)$$

Using the shifting property (##f(t-t_0) \iff e^{-j 2 \pi \nu t_0} F(\nu)##) the FT becomes: ##T \ sinc (T \nu) e^{-j 2 \pi (T/2)}.## So

$$\tilde{f} = \frac{jT}{2 \pi} \int^{\infty}_{- \infty} \delta ' (\nu ') \ sinc (T \nu) e^{-j 2 \pi \nu (T/2)} \ d \nu'$$

I am not sure how to proceed from here. The delta function still becomes zero over this integration limit. Are there any special properties we could use?
 

blue_leaf77

Science Advisor
Homework Helper
2,629
784
It should be
$$
\tilde{f} = \frac{jT}{2 \pi} \int^{\infty}_{- \infty} \delta ' (\nu ') \ sinc (T (\nu-\nu')) e^{-j 2 \pi (\nu-\nu') (T/2)} \ d \nu'
$$
writing ##\delta ' (\nu ')d \nu' = d\delta(\nu ')##, that integral becomes
$$
\tilde{f} = \frac{jT}{2 \pi} \int^{\infty}_{- \infty} \ sinc (T (\nu-\nu')) e^{-j 2 \pi (\nu-\nu') (T/2)} \ d\delta(\nu ')
$$
At this point, consider using integration by parts.
 
1,263
11
Thank you so much for the help.
 

Want to reply to this thread?

"Sawtooth function Fourier transform" You must log in or register to reply here.

Physics Forums Values

We Value Quality
• Topics based on mainstream science
• Proper English grammar and spelling
We Value Civility
• Positive and compassionate attitudes
• Patience while debating
We Value Productivity
• Disciplined to remain on-topic
• Recognition of own weaknesses
• Solo and co-op problem solving
Top