Fourier Transforms Intro: A Complete Student Guide

The Fourier transform extends Fourier series from periodic functions to arbitrary signals: F(ω) = ∫−∞∞ f(t)e−iωtdt decomposes f(t) into a continuous spectrum of frequencies, telling you exactly which frequencies are present and how strongly. The inverse transform rebuilds the signal from its spectrum.

This is the single most-used tool in modern technology. Every JPEG photo, MP3 song, MRI scan, and Wi-Fi signal passes through Fourier transforms (computed blindingly fast by the FFT algorithm). Engineers design filters, solve PDEs on infinite domains, and analyse vibrations entirely in the frequency domain.

Key Concepts

• Time domain vs frequency domain: f(t) shows when things happen; F(ω) shows what oscillates — two views of the same information.
• Inverse transform: f(t) = (1/2π)∫−∞∞ F(ω)eiωtdω; the pair is symmetric up to the 1/2π factor.
• Differentiation becomes multiplication: the transform of f′(t) is iωF(ω) — calculus turns into algebra, just like Laplace.
• Convolution theorem: convolution in time equals pointwise multiplication in frequency — the basis of all filtering.
• Parseval/Plancherel: total energy is preserved: ∫|f|2 = (1/2π)∫|F|2 — no energy is created or lost by transforming.
• Uncertainty principle: Δt·Δω ≥ 1/2 — a signal cannot be simultaneously short in time and narrow in frequency.

Formulas

• F(ω) = ∫−∞∞ f(t)e−iωt dt; f(t) = (1/2π)∫−∞∞ F(ω)eiωt dω.
• Linearity, time shift: f(t−a) → e−iωaF(ω); scaling and modulation have similar clean rules.
• Derivative rule: f′(t) → iωF(ω); hence differential equations become algebraic in ω.
• LTI systems: output spectrum Y(ω) = H(ω)·X(ω), where H is the frequency response.
• Assumption: f absolutely integrable (or finite energy). Limit: conventions for the 2π factor vary by field — always check.

Worked Example

Find the Fourier transform of the decaying pulse f(t) = e−at for t ≥ 0 (and 0 for t < 0), with a > 0. F(ω) = ∫0∞ e−ate−iωtdt = ∫0∞ e−(a+iω)tdt = [−e−(a+iω)t/(a+iω)]0∞. At the upper limit the exponential vanishes because a > 0; at 0 it equals 1. So F(ω) = 1/(a+iω). Its magnitude |F(ω)| = 1/√(a2+ω2): at ω = 0 this is 1/a (for a = 2, exactly 1/2), and it falls off at high frequencies — a smooth pulse is dominated by low frequencies. The phase is −arctan(ω/a).

Common Misconceptions

• Fourier transforms are only for periodic signals. That is Fourier series; the transform is precisely the tool for non-periodic signals.
• Negative frequencies are meaningless. Real signals need symmetric positive and negative frequency content; dropping them breaks the reconstruction.
• The FFT equals the continuous transform. The FFT computes a discrete approximation — sampling rate and record length impose Nyquist and resolution limits.

Real-World Applications

• Audio and image compression (MP3, JPEG) discard inaudible/invisible frequency components.
• MRI and CT scanners reconstruct body images from Fourier-sampled measurements.
• Solving PDEs on infinite domains: the heat equation becomes an ODE in ω.
• Filter and antenna design, vibration analysis, and quantum mechanics (position and momentum are Fourier pairs).

The Fourier transform F(ω) = ∫f(t)e−iωtdt reveals any signal's frequency content, with differentiation turning into multiplication by iω and convolution into multiplication — which is why filtering and PDE-solving become algebra. The worked pulse e−at transforms to 1/(a+iω), a low-pass spectrum. Master this pair of domains and most of modern signal technology opens up.