Signal processing

Signal processing operates on signals—functions that convey information about physical phenomena. Analog signals are continuous, while digital signals are discrete in time and amplitude, obtained via sampling and quantization.³ The fundamental operations include filtering, transformation, and estimation. The Fourier transform, which decomposes a signal into its frequency components, is central; the fast Fourier transform (FFT) makes this computationally efficient.⁴ Other key tools include convolution, correlation, and the z-transform for discrete systems. Statistical methods, such as Wiener filtering and Kalman filtering, handle noise and uncertainty.⁵

Signal processing is ubiquitous in modern technology. In telecommunications, it enables data compression, error correction, and modulation for mobile networks and the internet.⁶ In medical imaging, techniques like computed tomography (CT) and magnetic resonance imaging (MRI) rely on reconstruction algorithms. Audio processing uses it for noise cancellation, equalization, and speech recognition. Radar and sonar systems employ it for target detection and ranging. Image processing, a subfield, handles tasks from enhancement to object recognition.⁷

The field's roots lie in classical mathematics, such as Fourier's work on heat conduction (1822). Early 20th-century developments in electrical engineering, like the vacuum tube amplifier, enabled analog signal processing. The 1948 paper by Claude Shannon formalized information theory, and the 1965 FFT algorithm by Cooley and Tukey made digital processing practical.⁴ The advent of microprocessors and DSP chips in the 1970s–80s drove widespread adoption. Today, machine learning and deep learning are reshaping signal processing, enabling data-driven approaches to feature extraction and pattern recognition.⁸

Beyond mainstream uses, signal processing has niche applications. In seismology, it helps detect earthquakes and nuclear explosions. In astronomy, it enhances images from telescopes, such as the Event Horizon Telescope's black hole image. In finance, it analyzes market data for algorithmic trading. A notable edge case is the Nyquist–Shannon sampling theorem, which dictates that a signal must be sampled at twice its highest frequency to avoid aliasing—a principle often violated in practice, leading to artifacts.³ Another obscure fact: the term "signal processing" was coined in the 1940s at MIT's Radiation Laboratory, which developed radar during World War II.²

Aliasing

Distortion that occurs when a signal is sampled below the Nyquist rate, causing high-frequency components to appear as lower frequencies.

Convolution

A mathematical operation that combines two signals to produce a third, representing the output of a linear time-invariant system.

Digital signal processing (DSP)

The use of digital computers or specialized processors to analyze and modify signals.

Fast Fourier transform (FFT)

An efficient algorithm to compute the discrete Fourier transform and its inverse, reducing complexity from O(n^2) to O(n log n).

Filter

A system that selectively passes or suppresses certain frequency components of a signal.

Nyquist rate

The minimum sampling rate (twice the highest frequency) required to avoid aliasing.

Sampling

The process of converting a continuous-time signal into a discrete-time signal by measuring its amplitude at regular intervals.