Core Theory of Fourier Analysis

Introduction to Fourier Analysis What is Fourier Analysis? Fourier analysis is a mathematical technique for decomposing functions into simpler oscillatory components. At its core, the idea is elegantly simple: instead of viewing a function as a time-varying signal, we can represent it as a sum of sinusoids (sines and cosines) or complex exponentials at different frequencies. Fourier analysis is the process of breaking down a function into these frequency components. Fourier synthesis is the reverse process—rebuilding the original function from its frequency components. The mathematical operation that performs the decomposition is called a Fourier transformation, and its output is the Fourier transform. Consider this intuitive example: a musical recording contains many overlapping notes played simultaneously. Fourier analysis can decompose that audio signal into individual frequencies, telling you which notes are present and how loud each one is. This is exactly what an audio equalizer does—it shows you the frequency content of sound. Understanding the Frequency Domain A Fourier transform converts a time-domain signal (what you see when you plot a function against time) into a frequency-domain representation (what you see when you plot the signal's frequency content). The image above shows a time-domain signal—oscillations that vary over time. A Fourier transform converts this into frequency information. This frequency-domain view shows amplitude (strength) at different frequencies. The key insight is that magnitude (how tall each peak is) tells you the amplitude of that frequency component, and the phase angle (orientation of the complex number) tells you the initial phase shift of that component. The inverse Fourier transform reconstructs the original time-domain signal from this frequency-domain information, allowing you to convert back and forth between representations. Fundamental Properties: Why Fourier Analysis Works So Well Linearity Fourier transforms are linear operators, meaning that the transform of a sum equals the sum of the transforms: $$\mathcal{F}\{af(t) + bg(t)\} = a\mathcal{F}\{f(t)\} + b\mathcal{F}\{g(t)\}$$ This property is remarkably useful because it allows you to analyze complex signals by breaking them into simple parts, transforming each part, and combining the results. The Convolution Theorem One of the most powerful properties is the convolution theorem: the Fourier transform of a convolution of two functions equals the pointwise product of their individual transforms. $$\mathcal{F}\{f(t) g(t)\} = \mathcal{F}\{f(t)\} \cdot \mathcal{F}\{g(t)\}$$ This matters because convolution is computationally expensive to compute directly. By transforming to the frequency domain, multiplying (which is fast), and transforming back, you can perform convolution-based operations like signal filtering, polynomial multiplication, and large-integer multiplication much more efficiently. Eigenfunctions and Differentiation Complex exponential functions $e^{i\omega t}$ are eigenfunctions of the differentiation operator. This means when you differentiate an exponential, you simply multiply it by a constant. Consequently, differentiating a function in the time domain corresponds to multiplying its Fourier transform by $i\omega$ in the frequency domain. This turns difficult differential equations into simple algebraic equations in the frequency domain. Time-Invariant Systems Because exponential functions are eigenfunctions of linear time-invariant systems (like filters and circuits), the behavior of such systems can be analyzed completely and separately at each frequency. This is why we use frequency-domain analysis for designing filters and analyzing system stability. Four Variants of Fourier Analysis Fourier analysis takes different mathematical forms depending on whether your signal is continuous or discrete, and whether it's periodic or non-periodic. Understanding which variant applies to your situation is critical. The image above shows how these four variants relate to each other. Let's examine each one. Continuous-Time Fourier Transform (CTFT) The continuous-time Fourier transform applies to functions $x(t)$ defined on the entire real line. It's defined as: $$X(\omega) = \int{-\infty}^{\infty} x(t)\,e^{-i\omega t}\,dt$$ The inverse continuous-time Fourier transform reconstructs the original function: $$x(t) = \frac{1}{2\pi}\int{-\infty}^{\infty} X(\omega)\,e^{i\omega t}\,d\omega$$ Here, $\omega$ is angular frequency (in radians per second), and $X(\omega)$ is a continuous function giving the frequency content at all frequencies. When to use it: When analyzing non-periodic, continuous-time signals like an analog electrical signal or a physical phenomenon. Continuous-Time Fourier Series (CTFS) For periodic functions with period $P$, we use the continuous-time Fourier series instead: $$x(t) = \sum{k=-\infty}^{\infty} ck e^{i2\pi kt/P}$$ The series coefficients are: $$ck = \frac{1}{P}\int{0}^{P} x(t)\,e^{-i2\pi kt/P}\,dt$$ Notice the key difference: instead of a continuous spectrum $X(\omega)$, we get discrete coefficients $ck$ at integer multiples of the fundamental frequency $1/P$. These coefficients represent the amplitude and phase of each harmonic. The Fourier series converges to the original periodic function under the Dirichlet conditions, which are satisfied by most practical signals. These conditions essentially require the signal to be reasonably well-behaved (finite number of discontinuities and extrema in each period). When to use it: When analyzing periodic continuous-time signals, like repeating waveforms in circuits or mechanical vibrations. Discrete-Time Fourier Transform (DTFT) The discrete-time Fourier transform applies to sequences $x[n]$ (values at integer indices like $n = ..., -2, -1, 0, 1, 2, ...$). It's defined as: $$X(\omega) = \sum{n=-\infty}^{\infty} x[n]\,e^{-i\omega n}$$ The inverse discrete-time Fourier transform recovers the sequence: $$x[n] = \frac{1}{2\pi}\int{-\pi}^{\pi} X(\omega)\,e^{i\omega n}\,d\omega$$ A crucial point: $X(\omega)$ is a periodic function with period $2\pi$. This periodicity occurs because we can't distinguish frequencies separated by more than $2\pi$ when working with discrete samples—a phenomenon called aliasing. When to use it: When analyzing sequences obtained by sampling a continuous signal, or when working directly with discrete-time signals like digital audio or sensor data. Discrete-Time Fourier Series (DTFS) and Discrete Fourier Transform (DFT) For a periodic discrete sequence with period $N$, the discrete-time Fourier series represents it as: $$x[n] = \sum{k=0}^{N-1} ck e^{i2\pi kn/N}$$ with coefficients: $$ck = \frac{1}{N}\sum{n=0}^{N-1} x[n]\,e^{-i2\pi kn/N}$$ The discrete Fourier transform (DFT) is slightly different—it takes a finite sequence $x[0], x[1], ..., x[N-1]$ and produces $N$ complex coefficients: $$X[k] = \sum{n=0}^{N-1} x[n]\,e^{-i2\pi kn/N}$$ The inverse DFT reconstructs the sequence: $$x[n] = \frac{1}{N}\sum{k=0}^{N-1} X[k]\,e^{i2\pi kn/N}$$ When to use them: The DFT is the most practical variant for real-world digital signal processing because it works with finite-length data. The fast Fourier transform (FFT) is an efficient algorithm for computing the DFT that reduces computational complexity from $O(N^2)$ to $O(N \log N)$. Important note about zero-padding: If you add zeros to the end of your sequence before computing the DFT, you increase the frequency resolution of your spectrum—the frequency samples become more closely spaced. However, zero-padding does not actually add information about the signal; it only interpolates the spectrum you would have obtained without padding. <extrainfo> Decimation and Aliasing: If you reduce the number of samples (decimation) without properly accounting for the signal's frequency content, frequency components can overlap in the frequency domain—the phenomenon called aliasing. This is why the Nyquist sampling theorem is important: you must sample at least twice as fast as the highest frequency in your signal to avoid aliasing. </extrainfo> Summary: Choosing the Right Transform The decision tree is straightforward: Is your signal periodic? If yes, use a Fourier series (CTFS for continuous, DTFS for discrete). If no, use a Fourier transform. Is your signal continuous or discrete? Continuous → CTFT or CTFS. Discrete → DTFT or DTFS/DFT. Do you have finite data? Use the DFT with possible zero-padding for frequency resolution. Each variant is a mathematical expression of the same fundamental idea: representing a function as a sum of sinusoids at different frequencies. The differences reflect practical concerns about whether your signal is periodic, continuous, or sampled.