What Is Signal Processing?

Signal processing is the discipline that studies, modifies, and extracts information from signals. A signal is any quantity that varies over time, space, or another independent variable and represents or encodes information: audio, images, sensor readings, and biological measurements all qualify. Signal processing provides the mathematical tools and algorithms to clean, compress, transform, and interpret these quantities, forming a foundational layer beneath communications, control, biomedical engineering, and machine learning.

The field spans both continuous-time and discrete-time formulations. Modern practice is dominated by digital signal processing, in which signals are represented as sequences of numbers and manipulated by software or specialized hardware. Research from the IEEE Signal Processing Society spans theory, algorithms, and applications across all major sub-disciplines.

Digital Signal Processing and the Discrete Fourier Transform

Digital signal processing (DSP) converts continuous physical quantities into discrete numerical sequences through sampling and quantization, then applies computational algorithms to those sequences. The sampling theorem, established by Nyquist and Shannon, requires that the sampling rate exceed twice the highest frequency present in the signal to avoid aliasing. Once digitized, signals can be processed with arbitrary precision, stored without degradation, and processed reproducibly.

The discrete Fourier transform (DFT) is the central analytical tool of DSP. It converts a finite sequence of samples into a set of complex coefficients representing amplitude and phase at uniformly spaced frequencies. The fast Fourier transform algorithm computes the DFT in O(N log N) operations, enabling spectral analysis in real time. NIST's Digital Library of Mathematical Functions provides rigorous definitions for the transforms and special functions that DSP relies upon.

Filtering and Windowing

Filters selectively pass or reject frequency components of a signal. Finite impulse response (FIR) filters are inherently stable and can achieve linear phase, meaning all frequencies are delayed by the same amount, which preserves waveform shape. Infinite impulse response (IIR) filters achieve a given roll-off with fewer coefficients but require care in design to maintain stability and manage phase nonlinearity.

Windowing addresses the spectral leakage that occurs when a finite-length signal segment is treated as periodic. Applying a smooth window function (Hann, Hamming, Blackman) tapers the signal to zero at its edges, concentrating spectral energy in the main lobe at the cost of slightly reduced frequency resolution. Window selection involves a tradeoff between main-lobe width and side-lobe level that depends on the application's requirements.

Sampling and Convolution

Sampling is the process of measuring a continuous signal at discrete instants. In addition to the Nyquist condition, practical sampling requires anti-aliasing filters to remove frequency content above half the sampling rate before digitization. Oversampling, combined with noise shaping in sigma-delta converters, trades sample rate for amplitude resolution and is the basis for high-quality audio ADCs.

Convolution is the operation that describes how a linear time-invariant system transforms an input. In the discrete case, the output is the sum of shifted, scaled copies of the impulse response. The convolution theorem connects convolution in the time domain to multiplication in the frequency domain, allowing filters to be applied efficiently via the FFT. These relationships are treated rigorously in IEEE Transactions on Signal Processing, which is the field's principal journal.

Array Signal Processing

Array signal processing uses multiple spatially distributed sensors to exploit both temporal and spatial structure of signals. Beamforming steers the array's sensitivity toward a desired direction by applying weights and delays to sensor outputs before summing. Adaptive beamformers such as MVDR (minimum variance distortionless response) suppress interference from specific angles while maintaining gain toward the target. Applications include phased-array radar, medical ultrasound imaging, and wireless base-station antennas. Recent research on massive MIMO arrays addresses beamforming with hundreds of antenna elements.

Applications

Signal processing techniques are applied across virtually every technical domain:

• Audio and speech: noise suppression, echo cancellation, and codec compression rely on digital filtering and spectral analysis.
• Wireless communications: channel equalization, OFDM modulation, and MIMO detection are DSP algorithms running in every modern handset.
• Medical imaging: MRI reconstruction, ultrasound beamforming, and ECG analysis use convolution, Fourier transforms, and adaptive filtering.
• Radar and sonar: matched filtering, pulse compression, and Doppler processing extract target information from reflected signals.
• Seismology: spectral and time-frequency analysis of seismic records characterizes earthquake sources and subsurface geology.