I know that for a given signal $s(t)$ and a given delay $ \tau $, by the shift theorem:
$$ \mathcal{F}\{s(t-\tau)(f) \} = e^{-j 2 \pi f \tau} \mathcal{F}\{s(t)(f) \} \tag{1} $$ However, when I try to derive the Fourier Transform of a delayed LFM chirp, written as:

$$ s_{LFM}(t-\tau) = A e^{j (\omega_o (t-\tau) + \alpha (t-\tau)^2)} \Pi_{\frac{T}{2}}(t-\tau) $$ it gives:

$$ s_{LFM}(t-\tau) = A e^{j( \omega_o (t-\tau) + \alpha (t-\tau)^2)} \Pi_{\frac{T}{2}}(t-\tau) \\ = A e^{j \omega_o t} \ e^{-j \omega_o \tau} e^{j \alpha t^2} e^{j \alpha \tau^2} e^{j \alpha 2t\tau} \Pi_{\frac{T}{2}}(t-\tau)$$ so: $$ \mathcal{F}\{s_{LFM}(t-\tau)(f) \} = A e^{-j \omega_o \tau} e^{j \alpha {\tau}^2} \ \mathcal{F}\{ e^{j \omega_o t} e^{j \alpha t^2} e^{j \alpha 2t{\tau}} \Pi_{\frac{T}{2}}(t-\tau)(f) \} \tag{2} $$ The term $ e^{j \alpha 2t{\tau} } $ shifts the spectrum of $s_{LFM}$.

Why do not we get the same result between $(1)$ and $(2)$?

  • $\begingroup$ Are you sure your equation for an LFM chirp is correct? shouldnt $j$ apply to both terms in the exponential? $$Ae^{j \left(\omega_o(t-\tau)+\alpha(t-\tau)^2\right)}$$ $\endgroup$
    – Jdip
    Commented Aug 17, 2023 at 9:35
  • $\begingroup$ Sorry for the typo, you are right $\endgroup$
    – user68882
    Commented Aug 17, 2023 at 10:37
  • $\begingroup$ I think it does indeed shift the spectrum, but until you include the effects of the $\Pi$ function, aren't you just shifting the spectrum of a rectangle function with infinite support? That is, is there really a difference between $\mathrm{rect}(f / \infty)$ and $\mathrm{rect}((f-2\alpha\tau) / \infty)$? $\endgroup$ Commented Aug 21, 2023 at 23:55
  • $\begingroup$ Hay you might be interested in this answer or this one. $\endgroup$ Commented Aug 23, 2023 at 5:37

2 Answers 2


Let me get rid of $\Pi_{\frac{T}{2}}(t)$ in the derivation:

$$s(t) = Ae^{j\left(\omega_o t + \alpha t^2\right)}$$

Let's show that $$\mathcal{F}\left\{s(t-\tau)\right\} = e^{-j\omega \tau}\mathcal{F}\left\{s(t)\right\}$$

  1. Start with the Fourier Transform of the chirp (I'm omitting the integral's limits, they are $-\infty$ and $\infty$): \begin{align} \mathcal{F}\left\{s(t)\right\} &= \int s(t)e^{-j\omega t}dt\\ &= \int Ae^{j\left(\omega_o t + \alpha t^2\right)}e^{-j\omega t}dt \end{align}
  2. Now let's compute the Fourier Transform of the delayed chirp (notice the substitution $\tilde{t} = t-\tau$): \begin{align} \mathcal{F}\left\{s(t-\tau)\right\} &=\int s(t-\tau)e^{-j\omega t}dt\\ &=\int Ae^{j\left(\omega_o(t-\tau) + \alpha(t-\tau)^2\right)}e^{-j\omega t}dt\\ &= \int Ae^{j\left(\omega_o\tilde{t} + \alpha\tilde{t}^2\right)}e^{-j\omega (\tilde{t}+\tau)}d\tilde{t}\\ &=e^{-j\omega \tau} \int Ae^{j\left(\omega_o\tilde{t} + \alpha\tilde{t}^2\right)}e^{-j\omega \tilde{t}}d\tilde{t}\\ &= e^{-j\omega \tau}\mathcal{F}\left\{s(t)\right\} \end{align}
  • $\begingroup$ Here the Shift theorem is used, but I was mentioning a direct derivation from the delayed chirp, without the use of that theorem.. $\endgroup$
    – user68882
    Commented Aug 18, 2023 at 9:19
  • $\begingroup$ Where do you see the shift theorem being used? I use the delayed chirp, compute its Fourier transform (step 2) and show that it indeed equals the Fourier transform of the chirp (step 1) multiplied by $e^{j\omega \tau}$, effectively deriving the shift theorem from these two steps. $\endgroup$
    – Jdip
    Commented Aug 18, 2023 at 9:35
  • 1
    $\begingroup$ @user68882 You should accept this answer to close it out. Jdip addressed your question in it's entirety. He proved the time-shift property, he didn't use it directly to show that it works. $\endgroup$
    – Envidia
    Commented Aug 19, 2023 at 22:19
  • $\begingroup$ I think the OPs confusion arises from an infinite duration assumption on the LFM. Once the finite nature of the signal is accounted for by the inclusion of the $\Pi$ term, I think the shift will take care of itself. At least I think so. $\endgroup$ Commented Aug 21, 2023 at 23:58
  • $\begingroup$ I don’t think that’s where the confusion comes from, but I agree that the inclusion of $\Pi$ doesn’t change the derivation. $\endgroup$
    – Jdip
    Commented Aug 22, 2023 at 5:40

Ignoring constant terms, the FT of an infinite duration LFM is given by

$$ e^{j\pi\alpha t^2} \leftrightarrow e^{-j\pi f^2/ \alpha}. \tag{1} $$

Using the delay property of FTs, we know that

$$ e^{j\pi\alpha (t-\tau)^2} = e^{j\pi\alpha t^2} \ast \delta (t-\tau)\leftrightarrow e^{-j\pi f^2/ \alpha} e^{-j2\pi f \tau}. \tag{2} $$

If we expand the phase term of the delayed signal prior to the FT, we get

$$ e^{j\pi\alpha (t-\tau)^2} = e^{j\pi\alpha \tau^2} e^{j\pi\alpha t^2} e^{-j2\pi\alpha t \tau}, \tag{3} $$

which is the original LFM multiplied by a constant and modulated by the linear phase signal $\exp(-j2\pi \alpha t \tau)$, which, as you correctly point out, must shift the spectrum of original LFM by $-\alpha \tau$. This seems to be the cause of some confusion because a delay in time shouldn't lead to a shift in frequency, but for an LFM it does. This is due to the one-to-one relationship between time and frequency for LFMs.

Let us look at what happens when we shift the spectrum of the undelayed LFM as given in $(1)$ by $-\alpha \tau$:

$$ e^{-j\pi f^2/ \alpha} \rightarrow e^{-j\pi (f + \alpha \tau )^2 / \alpha} = e^{-j\pi / \alpha (f^2 + 2 f \alpha \tau + \alpha^2 \tau^2 ) } = e^{-j\pi f^2 / \alpha } e^{-j2\pi f \tau} e^{-j\pi \alpha \tau^2}, \tag{4} $$

which, aside from the constant, is the same as $(2)$. If you went through a careful derivation that accounted for the constant phase terms, you would see that the two are indeed equal.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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