A frequency‑domain function that describes how the power or variance of a time‑series signal is distributed across frequency; for a wide‑sense stationary stochastic process the (auto)power spectral density S_x(f) is the Fourier transform of the autocorrelation function and has units of power per hertz (e.g., W·Hz⁻¹).

Principle

Integrating the PSD over a frequency band yields the mean power or variance contributed by that band; for electrical signals instantaneous dissipated power relates to the PSD through appropriate transfer functions and impedance.

Demonstration

Situation: An engineer assesses vibration on a machine bearing. Recognition: The vibration signal is sampled and assumed stationary over the measurement interval. Action: Compute the PSD estimate and integrate S_a(f) between 50 and 200 Hz to obtain the band‑limited mean square acceleration; convert to RMS acceleration by square root. Consequence: The engineer identifies a resonant band contributing most variance and designs a filter or damping targeted at that frequency range.

Misapplication

Interpreting a PSD estimate as a deterministic amplitude spectrum (mistaking units and meaning), or failing to account for one‑sided versus two‑sided PSD conventions and the required conversion; the semantic error is conflating power density with amplitude spectrum.

Consequence

Proper use yields correct band power, RMS estimates and informed filter/structural design; misuse produces erroneous RMS values, incorrect filter bandwidths and potential under‑ or over‑specification of mitigation measures.

Reversal

PSD theory assumes stationarity and ergodicity for simple interpretation; for non‑stationary signals the PSD is not a complete descriptor and time‑frequency methods (spectrograms, wavelets) or evolutionary spectra are required. Discrete finite‑length data require estimation methods (periodogram, Welch) with associated bias‑variance tradeoffs.

Boundary

Clearly within: the PSD of a wide‑sense stationary random vibration process estimated with Welch's method over a long stationary interval. Boundary case: a finite, quasi‑stationary signal where the PSD estimate's resolution and variance depend on windowing choices. Clearly outside: a single transient pulse described by its Fourier transform amplitude spectrum but not meaningfully by a stationary PSD.

Semantic Tension

Frequency Resolution ↔ Estimation Variance: increasing spectral resolution (narrower frequency bins) reduces averaging and increases variance of the PSD estimate; averaging to reduce variance lowers resolution.

Synthesis

The PSD apportions a signal's variance across frequency and is the correct tool for band‑power and RMS calculations under stationarity assumptions; practical interpretation requires attention to estimation methods, sidedness conventions, and the signal's time‑variation.