Fourier Transform

In simple terms

Any repeating signal — a sound wave, a heartbeat, a daily temperature pattern — can be represented as a sum of sine waves of different frequencies. The Fourier transform is the tool that finds those frequencies: it tells you “this audio clip contains 440 Hz (concert A), 880 Hz (an octave up), and 1760 Hz (two octaves up)” and how much of each. This insight — that everything is a sum of sines — is one of the most powerful ideas in mathematics, underlying MP3 compression, MRI scanners, JPEG images, signal filtering, and machine learning on sequential data.

More detail

The continuous Fourier transform: F(ω) = ∫_{-∞}^{∞} f(t) e^{-iωt} dt

Given a function f(t) in the time domain, F(ω) gives its representation in the frequency domain. e^{-iωt} = cos(ωt) - i·sin(ωt) (Euler’s formula) — the integral computes the “overlap” of f with each frequency ω.

The inverse Fourier transform recovers f(t) from F(ω): f(t) = (1/2π) ∫_{-∞}^{∞} F(ω) e^{iωt} dω

Key properties:

• Linearity: F(af + bg) = aF(f) + bF(g).
• Convolution theorem: f * g in time domain ↔ F · G in frequency domain (pointwise multiplication). Convolution becomes multiplication — a massive computational saving.
• Parseval’s theorem: total energy is preserved between domains: ∫|f(t)|² dt = (1/2π)∫|F(ω)|² dω.
• Shift theorem: a time shift ↔ a phase shift in frequency.

Discrete Fourier Transform (DFT): for sampled signals (computer data), the DFT operates on N samples: X[k] = Σ_{n=0}^{N-1} x[n] · e^{-2πink/N}

Computing all N outputs naively requires O(N²) operations.

Fast Fourier Transform (FFT): the Cooley-Tukey FFT algorithm (1965) reduces DFT to O(N log N) by exploiting the recursive structure of the DFT when N is a power of 2. A 1 million-point DFT: naively 10¹² operations; with FFT, ~2×10⁷ operations — 50,000× faster. The FFT is one of the most important algorithms in computing.

Applications:

Audio (MP3, AAC): the MDCT (Modified Discrete Cosine Transform — a variant of DFT) transforms audio frames to frequency domain; frequencies where the ear is less sensitive are quantised more aggressively (perceptual coding). MP3 achieves 10:1 compression with near-transparent audio quality.

Image (JPEG): DCT (Discrete Cosine Transform) applied to 8×8 pixel blocks; high-frequency coefficients are discarded or heavily quantised. JPEG’s frequency-domain insight: the human eye is less sensitive to high-frequency detail.

Signal filtering: apply FFT, multiply by a filter function (remove frequencies above 20 kHz from audio), apply inverse FFT. Equivalent to convolution in the time domain — but O(N log N) vs. O(N²).

MRI: MRI scanners directly acquire data in k-space (frequency domain); an inverse FFT reconstructs the image.

Convolution in ML: convolutional neural networks compute cross-correlation; for large kernels, FFT-based convolution is more efficient.

Spectral analysis: FFT of a vibration signal reveals dominant frequencies — used for fault detection in rotating machinery (a bearing fault at frequency f_bearing × RPM).

Nyquist-Shannon sampling theorem: to accurately represent a signal with maximum frequency f_max, you must sample at ≥ 2f_max (the Nyquist rate). Sampling below the Nyquist rate causes aliasing — high-frequency components fold back as false low-frequency artefacts. CD audio samples at 44.1 kHz (> 2 × 20 kHz human hearing limit).

Windowing: the DFT assumes the signal is periodic. A sharp truncation (rectangular window) introduces spectral leakage — energy from one frequency bleeds into adjacent bins. Window functions (Hann, Hamming, Blackman) taper the signal at the edges, reducing leakage at the cost of frequency resolution.

Why it matters

The FFT is one of the ten most important algorithms in science and engineering. It is the computational engine behind MP3, JPEG, video codecs, radar, MRI, digital communications (OFDM — the modulation scheme in WiFi and 4G/5G), earthquake seismology, and gravitational wave detection (LIGO). The convolution theorem powers efficient polynomial multiplication, large integer multiplication (Karatsuba, Schönhage-Strassen), and neural network inference for large kernels. Understanding the Fourier transform is foundational for signal processing, audio/image/video engineering, and scientific computing.

Real-world examples

• MP3/AAC: frequency-domain perceptual coding; basis of all lossy audio compression.
• JPEG: DCT-based image compression; basis of most photo storage and web images.
• WiFi 802.11/4G/5G LTE: OFDM modulation — parallel frequency-domain sub-carriers; the entire wireless network is built on FFT.
• LIGO (gravitational wave detector): matched filtering using FFT to detect the characteristic signal of black hole mergers in noisy data.

Common misconceptions

• “The Fourier transform only works for periodic signals.” The Fourier transform applies to any square-integrable function. Non-periodic functions have a continuous frequency spectrum rather than discrete spectral lines.
• “FFT and DFT are different algorithms.” FFT is an efficient algorithm for computing the DFT — same result, much faster. “FFT” is often used loosely to mean “DFT computed efficiently.”

Learn next

The Fourier transform is grounded in calculus and linear algebra (it’s a linear operator on function spaces). Numerical methods covers the practical implementation of FFT and discrete signal processing. Information theory explains why frequency-domain representations enable compression.