Properties of Fourier Transform

Introduction

The properties of the Fourier Transform are the analytical tools that make frequency-domain signal processing practical. Rather than computing the transform integral for every new signal, these properties allow you to derive complex transform pairs from simple ones, predict how time-domain operations affect the spectrum, and convert between time-domain and frequency-domain operations. Mastering these properties is essential — they appear in virtually every signal processing derivation, from filter design to modulation theory.

When you encounter a new signal in practice, the strategy is almost never to evaluate the Fourier integral directly. Instead, you decompose the signal into simpler components whose transforms you already know, then apply properties to assemble the final result. This approach is both computationally efficient and conceptually illuminating — it reveals how each time-domain manipulation has a precise frequency-domain counterpart.

Linearity

$$\mathcal{F}\{a_1 x_1(t) + a_2 x_2(t)\} = a_1 X_1(j\omega) + a_2 X_2(j\omega)$$

The transform of a linear combination equals the linear combination of individual transforms. This follows directly from the linearity of integration.

Why it matters: Linearity allows you to break complicated signals into simpler pieces, transform each piece independently, and combine the results. For instance, if a signal consists of a rectangular pulse plus a decaying exponential, you can look up each transform separately and add them with appropriate scaling constants.

Example: Find $\mathcal{F}\{3e^{-2t}u(t) + 5e^{-4t}u(t)\}$.

Using known pairs and linearity: $X(j\omega) = \frac{3}{2+j\omega} + \frac{5}{4+j\omega}$

No integration was needed — just table lookup and addition.

Time Shifting

$$\mathcal{F}\{x(t - t_0)\} = e^{-j\omega t_0} X(j\omega)$$

A time delay adds a linear phase factor. The magnitude spectrum is unchanged ($|e^{-j\omega t_0}| = 1$); only the phase changes by $-\omega t_0$ (linear phase).

Physical meaning: Delaying a signal does not change its frequency content — only the timing (phase) of each component shifts. This is why you can recognize a song regardless of when it starts playing — the frequencies are the same, only the phase differs.

Worked Example: Given $x(t) = e^{-3t}u(t)$ with $X(j\omega) = \frac{1}{3+j\omega}$, find the transform of $y(t) = e^{-3(t-2)}u(t-2)$.

Recognizing $y(t) = x(t-2)$, we apply the time-shift property: $$Y(j\omega) = e^{-j2\omega} \cdot \frac{1}{3+j\omega}$$

The magnitude $|Y(j\omega)| = \frac{1}{\sqrt{9+\omega^2}}$ is identical to $|X(j\omega)|$. Only the phase has changed by an additional $-2\omega$ radians.

Frequency Shifting (Modulation)

$$\mathcal{F}\{x(t)e^{j\omega_0 t}\} = X(j(\omega - \omega_0))$$

Multiplying by a complex exponential shifts the spectrum to the right by $\omega_0$. For real modulation: $$\mathcal{F}\{x(t)\cos(\omega_0 t)\} = \frac{1}{2}[X(j(\omega-\omega_0)) + X(j(\omega+\omega_0))]$$

This is the mathematical basis of amplitude modulation — the baseband spectrum is shifted to $\pm\omega_0$.

Application in AM Radio: A voice signal occupying $0$ to $4$ kHz is multiplied by $\cos(2\pi f_c t)$ where $f_c$ might be $1$ MHz. The result occupies two bands centered at $\pm f_c$, each $8$ kHz wide. The receiver reverses this by multiplying again by the carrier and low-pass filtering — a process called coherent demodulation.

Time Scaling

$$\mathcal{F}\{x(at)\} = \frac{1}{|a|}X\left(\frac{j\omega}{a}\right)$$

Compressing a signal in time ($|a|>1$) expands its spectrum and reduces its amplitude. This embodies the time-bandwidth tradeoff: $\Delta t \cdot \Delta\omega \geq$ constant.

Intuition: A short pulse must contain high frequencies (its spectrum is wide). A long, slow pulse has a narrow spectrum concentrated near DC. You cannot simultaneously have a short signal and a narrow bandwidth — this is the uncertainty principle of signal processing.

Example: If $x(t)$ has bandwidth $W$, then $x(2t)$ (compressed by factor 2) has bandwidth $2W$ but half the spectral amplitude. The total energy is preserved (the wider spectrum compensates for the reduced amplitude).

Convolution Theorem

$$\mathcal{F}\{x(t) * h(t)\} = X(j\omega) \cdot H(j\omega)$$

Convolution in time becomes multiplication in frequency. This is arguably the most powerful property — it transforms the computationally expensive convolution integral into simple pointwise multiplication.

The dual statement: multiplication in time becomes convolution in frequency (scaled by $1/2\pi$): $$\mathcal{F}\{x(t) \cdot h(t)\} = \frac{1}{2\pi} X(j\omega) * H(j\omega)$$

Why this matters for filtering: When a signal passes through an LTI system with impulse response $h(t)$, the output is $y(t) = x(t) * h(t)$. In the frequency domain, this becomes $Y(j\omega) = X(j\omega) \cdot H(j\omega)$. Therefore, designing a filter means specifying $H(j\omega)$ — which frequencies to pass, which to reject — and the convolution theorem guarantees the system implements the corresponding time-domain convolution.

Worked Example: Find the output when $x(t) = e^{-t}u(t)$ passes through a system with $h(t) = e^{-2t}u(t)$.

$X(j\omega) = \frac{1}{1+j\omega}$, $H(j\omega) = \frac{1}{2+j\omega}$

$Y(j\omega) = \frac{1}{(1+j\omega)(2+j\omega)}$

By partial fractions and inverse transform: $y(t) = (e^{-t} - e^{-2t})u(t)$

Differentiation

$$\mathcal{F}\left\{\frac{d^n x}{dt^n}\right\} = (j\omega)^n X(j\omega)$$

Differentiation in time corresponds to multiplication by $j\omega$ in frequency. Each derivative boosts high frequencies by a factor of $\omega$, which is why differentiation amplifies noise (noise is typically broadband, including high frequencies).

Practical consequence: In control systems and signal processing, taking derivatives of noisy measurements is problematic because noise power at high frequencies gets multiplied by $\omega^2$ (for a second derivative). This is why smooth estimators (Kalman filters, low-pass pre-filtering) are applied before differentiation.

Integration

$$\mathcal{F}\left\{\int_{-\infty}^t x(\tau)d\tau\right\} = \frac{X(j\omega)}{j\omega} + \pi X(0)\delta(\omega)$$

Integration divides by $j\omega$ in frequency (attenuates high frequencies), which is why integration smooths signals. The delta term accounts for the DC value.

Interpretation: Integration is a low-pass operation. It suppresses high-frequency oscillations and accumulates low-frequency content, producing smoother output signals. This is the frequency-domain explanation of why running averages and integrators smooth noisy data.

Parseval's Theorem (Energy Conservation)

$$\int_{-\infty}^{\infty}|x(t)|^2 dt = \frac{1}{2\pi}\int_{-\infty}^{\infty}|X(j\omega)|^2 d\omega$$

Signal energy computed in time equals energy computed in frequency. This guarantees that the Fourier Transform preserves energy — no energy is lost or created in the transformation.

Application: Parseval's theorem lets you compute signal energy in whichever domain is more convenient. For a signal with $X(j\omega) = \frac{1}{1+j\omega}$, computing $\frac{1}{2\pi}\int\frac{1}{1+\omega^2}d\omega = \frac{1}{2}$ is often easier than integrating $|x(t)|^2$ directly.

Duality

If $x(t) \xleftrightarrow{\mathcal{F}} X(j\omega)$, then $X(jt) \xleftrightarrow{\mathcal{F}} 2\pi x(-\omega)$.

Duality means that if you know a transform pair, you automatically get another pair by swapping the roles of time and frequency.

Example: Since $\text{rect}(t) \leftrightarrow T\text{sinc}(\omega T/2\pi)$, by duality $\text{sinc}(t) \leftrightarrow \text{rect}(\omega/2\pi)$ (appropriately scaled). This duality is particularly useful for deriving the transform of a sinc function without integration.

Conjugation and Symmetry

For real signals $x(t) = x^*(t)$: $$X(-j\omega) = X^*(j\omega)$$

This means the magnitude spectrum is even: $|X(j\omega)| = |X(-j\omega)|$, and the phase spectrum is odd: $\angle X(j\omega) = -\angle X(-j\omega)$.

Consequence: For real signals, you only need to plot the spectrum for $\omega \geq 0$ — the negative frequency side is completely determined by symmetry. This is why spectrum analyzers and FFT plots typically show only the positive frequency axis.

Summary Table

Key Takeaways

• Linearity and time-shifting are used most frequently in practice
• Convolution theorem ($x*h \leftrightarrow X \cdot H$) is the foundation of frequency-domain system analysis
• Time compression expands the spectrum (time-bandwidth tradeoff)
• Modulation (frequency shifting) is multiplication by $e^{j\omega_0 t}$
• Differentiation boosts high frequencies; integration suppresses them
• Parseval's theorem guarantees energy conservation between domains
• For real signals: magnitude spectrum is even, phase spectrum is odd
• The strategy for solving problems is: decompose, apply properties, recombine — avoid direct integration whenever possible