Fourier Analysis: Decomposing Signals into Frequencies
Introduction
Fourier analysis is the mathematical art of breaking complex signals into simple building blocks — sine and cosine waves of different frequencies. Just as a prism splits white light into its constituent colors, the Fourier transform decomposes any signal into its frequency components. This perspective reveals structure invisible in the original time or space representation.
The impact of Fourier analysis extends across science and engineering. Audio compression removes frequencies the ear cannot hear. Medical MRI scanners reconstruct images from frequency-domain measurements. Radar systems identify targets by analyzing frequency shifts. The Fourier transform is so fundamental that it has been called the most important numerical algorithm of our time.
Fourier Series
A Fourier series represents a periodic function as an infinite sum of sines and cosines. Any reasonable periodic function with period T can be written as f(t) = a₀/2 + Σ_{n=1}∞ [a_n cos(2πnt/T) + b_n sin(2πnt/T)], where the coefficients a_n and b_n are computed by integrating the function against the corresponding basis functions.
Computing Fourier Coefficients
The coefficient a₀ represents the average value of the function over one period. The coefficient a_n measures how much the function resembles cos(2πnt/T), and b_n measures resemblance to sin(2πnt/T). These coefficients are computed by orthogonality: the integral over one period of the product of different basis functions is zero, while the integral of a basis function with itself equals T/2.
The amplitude spectrum √(a_n² + b_n²) shows the strength of each frequency component. The phase spectrum arctan(−b_n/a_n) shows the phase shift. Together, amplitude and phase completely describe the periodic function. A square wave, for example, contains only odd harmonics with amplitudes decaying as 1/n.
Convergence and the Gibbs Phenomenon
Fourier series converge to the function at points of continuity. At jump discontinuities, the series overshoots by about 9% of the jump — the Gibbs phenomenon. This overshoot does not disappear as more terms are added; its width shrinks but its height remains constant. The Gibbs phenomenon is important in signal processing, where sharp transitions create ringing artifacts.
Complex Fourier Series
The complex form of the Fourier series uses complex exponentials e^{iωt} instead of sines and cosines: f(t) = Σ_{n=−∞}∞ c_n e^{2πint/T}. The coefficients c_n are complex and combine amplitude and phase information. The relationship between real and complex coefficients is c_n = (a_n − ib_n)/2 for n > 0 and c_{-n} = (a_n + ib_n)/2.
The complex form simplifies many calculations. The derivative of the series becomes multiplication by iω. The convolution theorem takes its natural form. The complex representation also connects Fourier analysis to complex analysis through the Laplace transform, which extends the Fourier transform to handle growing or decaying signals.
The Fourier Transform
The Fourier transform extends the ideas of Fourier series to nonperiodic functions on the infinite line. The transform F(ω) = ∫{−∞}∞ f(t) e^{−iωt} dt converts a time-domain signal into a frequency-domain representation. The inverse transform f(t) = (1/2π) ∫{−∞}∞ F(ω) e^{iωt} dω reconstructs the original signal.
Properties of the Fourier Transform
The Fourier transform has elegant properties that make it invaluable. The transform of a derivative is iω times the transform of the function, converting differential equations into algebraic equations. Convolution in the time domain becomes multiplication in the frequency domain: F{f*g} = F(ω)G(ω). This convolution theorem underlies efficient filtering and correlation operations.
Time shifts become phase shifts in the frequency domain: F{f(t − a)} = e^{−iωa}F(ω). Scaling in time produces reciprocal scaling in frequency: F{f(at)} = (1/|a|)F(ω/a). The energy of the signal is preserved: ∫|f(t)|² dt = (1/2π)∫|F(ω)|² dω (Parseval’s theorem).
The Uncertainty Principle
The Fourier transform pairs time localization with frequency localization. A signal that is perfectly localized in time (an impulse) has infinite bandwidth. A signal perfectly localized in frequency (a pure tone) extends infinitely in time. The uncertainty principle Δt·Δω ≥ 1/2 quantifies this tradeoff.
This principle has profound implications. You cannot simultaneously know both the exact time and exact frequency of a signal. Short-time Fourier transforms and wavelets address this by analyzing signals at multiple time-frequency resolutions, providing a tradeoff appropriate for different applications.
The Discrete Fourier Transform
The discrete Fourier transform (DFT) applies Fourier analysis to sampled data. Given N samples f_k, the DFT produces N frequency components F_n = Σ_{k=0}^{N−1} f_k e^{−2πink/N}. The fast Fourier transform (FFT) computes the DFT in O(N log N) operations instead of O(N²), making Fourier analysis practical for large datasets.
The Fast Fourier Transform
The FFT recursively divides the DFT into smaller DFTs, exploiting the symmetry of the complex exponential. Cooley and Tukey’s 1965 algorithm revolutionized signal processing by making Fourier analysis computationally feasible. Modern FFT libraries compute transforms of million-point signals in seconds.
The FFT is used everywhere: JPEG and MP3 compression use block-based FFTs. Convolution via FFT accelerates polynomial multiplication, digital filtering, and cross-correlation in particle image velocimetry. Synthetic aperture radar uses FFTs to form high-resolution images from raw radar returns.
The FFT is used everywhere: JPEG and MP3 compression use block-based FFTs. Spectral analysis of financial time series uses FFTs to detect periodic patterns. Numerical solution of partial differential equations uses FFT-based spectral methods.
Sampling and Aliasing
Sampling a continuous signal at rate f_s allows reconstruction of frequencies up to f_s/2 (the Nyquist frequency). Frequencies above Nyquist are aliased to lower frequencies, appearing as spurious components. The sampling theorem (Nyquist-Shannon) states that a bandlimited signal can be perfectly reconstructed from samples taken at more than twice its maximum frequency.
Aliasing is a critical practical concern. Audio recordings must filter out frequencies above the Nyquist limit before sampling. Images must antialias before downsampling. Radar systems must avoid ambiguous range measurements due to aliasing in the pulse repetition frequency.
The Laplace Transform
The Laplace transform extends the Fourier transform to handle signals that grow exponentially. Defined as F(s) = ∫₀∞ f(t) e^{−st} dt for complex s = σ + iω, it converges for functions that grow no faster than e^{σ₀t}. The Laplace transform is the standard tool for analyzing linear time-invariant systems, converting differential equations into algebraic equations in the s-domain.
The transfer function H(s) of a system relates the Laplace transform of the output to that of the input. Poles of H(s) determine stability: poles in the left half-plane indicate stable, decaying responses; poles on the imaginary axis indicate sustained oscillations; poles in the right half-plane indicate unstable, growing responses. Control system design places poles to achieve desired performance specifications.
Applications of Fourier Analysis
Fourier analysis is fundamental to modern technology. Audio compression (MP3, AAC) transforms audio into the frequency domain and discards components below the threshold of human hearing. Image compression (JPEG) performs block-based discrete cosine transforms, keeping low-frequency components while discarding high-frequency details.
Medical imaging relies heavily on Fourier analysis. MRI scanners acquire data in the spatial frequency domain (k-space) and reconstruct images via inverse Fourier transform. Computed tomography uses the Radon transform, closely related to the Fourier transform, to reconstruct cross-sectional images from projections.
Spectral analysis identifies periodic components in noisy data. Economists use it to detect business cycles. Astronomers use it to find exoplanets from radial velocity variations. Engineers use it to identify vibration frequencies in rotating machinery.
Wavelets and Time-Frequency Analysis
Wavelet analysis extends Fourier analysis to signals whose frequency content changes over time. Unlike the Fourier transform, which provides only frequency information with no time localization, wavelets use scaled and translated basis functions that capture both time and frequency information.
The continuous wavelet transform convolves the signal with scaled versions of a mother wavelet. The discrete wavelet transform uses filter banks to decompose signals into approximation and detail coefficients at multiple resolution levels. Wavelets excel at detecting transient events, compressing images (JPEG 2000), and denoising signals with nonstationary characteristics.
What is the difference between a Fourier series and a Fourier transform? A Fourier series decomposes a periodic function into discrete frequency components. A Fourier transform extends this to nonperiodic functions, producing a continuous spectrum. A Fourier series represents a periodic function as a discrete sum of harmonics. A Fourier transform represents a nonperiodic function as a continuous integral over all frequencies.
What does the convolution theorem say? The Fourier transform of the convolution of two functions equals the product of their individual Fourier transforms. This makes convolution efficient to compute via FFT.
What is aliasing and how do you prevent it? Aliasing occurs when a signal contains frequencies above half the sampling rate, causing them to appear as lower frequencies. Prevent it by low-pass filtering the signal before sampling.
Why is the FFT so important? The FFT reduces the computational complexity of the discrete Fourier transform from O(N²) to O(N log N), making frequency-domain analysis practical for large datasets in real-time applications.
Partial Differential Equations — Numerical Analysis — Computational Mathematics