Beamforming
How it works
A wave hitting an array of N sensors reaches successive elements with delay differences that depend on the angle of arrival and the array geometry. (1) For each element a phase shift (narrowband systems) or time delay (wideband systems) is computed to align in-phase the components coming from the desired direction — the steering vector. (2) Element signals are multiplied by complex weights and summed; in the delay-and-sum variant the weights share equal magnitude and differ only in phase, giving constructive interference toward the beam direction. (3) In adaptive beamforming the weights are derived from the signal covariance matrix: MVDR/Capon minimises output power subject to unit gain toward the target, while LCMV/GSC and subspace methods (MUSIC) place nulls precisely toward interferers. The beam can be swept electronically by changing the weights, and wideband signals are processed per frequency bin.
Problem solved
A single omnidirectional sensor cannot separate the desired signal from noise and interference arriving from other directions, degrading the signal-to-noise ratio and hampering tasks such as speech recognition in noise or source localisation. Beamforming solves this spatially — without physically aiming the sensor — improving SNR, suppressing interference, and enabling direction-of-arrival (DOA) estimation.
Components
A set of spatially distributed elements (microphones, antennas, hydrophones) with known geometry, so that delay differences encode source direction.
The set of phase shifts/delays describing how a wave from a given direction maps onto the array elements; defines where the beam is aimed.
Multiplies each element signal by a complex weight and sums the results, shaping the directional response (main lobe and nulls).
Official
Derives weights from the signal covariance matrix (e.g. MVDR) or a subspace decomposition (MUSIC), enabling interference nulling and DOA estimation.
Evolution
Jack Capon introduces the minimum-variance distortionless-response (MVDR) adaptive beamformer, the basis of high-resolution directional analysis.
Ralph Schmidt proposes MUSIC (MUltiple SIgnal Classification), a subspace method for high-resolution direction-of-arrival (DOA) estimation.
Commercial 5G NR deployments make beamforming (massive MIMO, mmWave) a standard element of directional radio transmission.
Computational complexity
Time complexity: O(N) per próbkę (konwencjonalny), O(N^3) dla adaptacyjnego (inwersja macierzy kowariancji).
Execution paradigm
Output is conditional on direction (angle) and scene statistics: the same sensors yield different responses depending on the chosen weights.
Parallelism
Per-element and per-frequency-bin weighting and summation are inherently parallel; radar typically requires hardware acceleration.
Hardware requirements
Real-time processing of radar and high-throughput arrays typically requires dedicated, reconfigurable hardware (FPGA).
Acoustic microphone arrays (sonar/speech) allow real-time software processing on CPUs.