I just need to generate the equivalent train in the frequency domain by hand.

I know that the Fourier Transform of a pulse train is a pulse train, with the intervals of the pulses changed by (1/T).

However I'm confused on the implementation when you are also working with sampling rate of the original signal.

Can anyone quick point me in the correct direction.


  • $\begingroup$ Can you please clarify what your question is? The text seems unrelated to the title. $\endgroup$ – MBaz Dec 14 '16 at 18:47

The train of pulses in the time domain: $$x(t)=|||(t)=\sum_{i=1}^{\infty} \delta(t-i)$$

Equals to the following train of pulses in the frequency domain: $$\mathbb{F}x=|||(f)=\sum_{i=1}^{\infty} \delta(f-i)$$

Where $|||$ is the train of unitary pulses, comb standard signal.

Hence: $$|||(t) \rightarrow |||(f)$$

The train of unitary pulses in time domain separated by $\tau$ is (do you know why we are having $1/\tau$?): $$1/\tau|||(t/\tau) \rightarrow |||(\tau f)$$

And the train of unitary pulses in frequency domain separated by $1/\tau$ is: $$|||(t/\tau) \rightarrow \tau|||(\tau f)$$

NOW when having a sampling time of $dt$, each delta pulse in the instant $i$ can be represented in discrete time as: $$\delta(t-i)=\frac{1}{dt}\delta_{ti}$$ Where $\delta_{ti}=1$ for $t=i$ and 0 elsewhere, as in the discrete kronecker delta sense, approximating each pulse as the smallest possible rectangle (other approximations are possible though...).

In discrete time/frequency, we just have to replace the proper $delta$ implementation, as already defined.

Finally, the $\infty$ in the time sum is bounded to $T$, the total measuring time, and to $F=0.5/dt$, the sampling frequency band (nyquist?). So the period $\tau$ must be divisor of $T$, $T=\phi\tau$, and $1/\tau$ must be a divisor of $F=0.5/dt$, $T\phi=0.5\tau N$ for the numbers in the FFT algorithm to be squared in order to create a perfect periodic sequence in both the time and frequency streams.

HENCE, the standard FFT algorithm, for example with $\tau=6$, $\phi=3$ and $dt=1$ (for simplicity):


Implements the non unitary train of pulses in time domain:

x=1/3*[1 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0]

with the unitary train of pulses in frequency domain:

Fx=[1 0 0 1 0 0 1 0 0 1 0 0 1 0 0 1 0 0]

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.