Comparison

Power Spectral Density vs Energy Spectral Density

Connecting a spectrum analyzer to an FM transmitter and to a single radar pulse gives you two completely different spectral measurements — PSD and ESD. The FM transmitter runs continuously; its total energy is infinite, so you characterize it by power per hertz. The radar pulse is finite in duration; its total energy is finite and measurable in joules. Choosing the wrong spectral density definition for your signal type leads to infinite integrals and nonsense results — a mistake that costs marks in every random signal theory examination.

ECE, EI

Side-by-side comparison

ParameterPower Spectral DensityEnergy Spectral Density
Applicable Signal TypePower signals — periodic or stationary random (finite average power, infinite energy)Energy signals — finite-duration or finite-energy signals
Mathematical DefinitionS_x(f) = lim(T→∞) E[|X_T(f)|²/T] W/HzE_x(f) = |X(f)|² J/Hz
Total Area Under CurveEquals average power (watts)Equals total energy (joules)
Fourier Transform RequirementDTFT or PSD via autocorrelation (Wiener-Khinchin)Standard Fourier transform exists (signal is square-integrable)
Example SignalsAWGN noise, sinusoid, random binary data streamSingle rectangular pulse, Gaussian pulse, finite-length ECG segment
UnitW/Hz or V²/HzJ/Hz or V²·s/Hz
Wiener-Khinchin TheoremPSD = Fourier transform of autocorrelation R_x(τ)ESD = |X(f)|² directly
Existence ConditionSignal must be wide-sense stationary (WSS)∫|x(t)|² dt < ∞ (square-integrable)
Practical ToolSpectrum analyzer displaying dBm/Hz (e.g., Rigol DSA815)FFT magnitude squared of a captured waveform segment
Cross-spectral VariantCross-PSD: S_xy(f) = FT of cross-correlationCross-ESD: X*(f)·Y(f)

Key differences

The dividing line is energy: a sinusoid running forever has infinite total energy but finite average power (P = A²/2), so PSD applies. A 10 µs radar pulse has finite total energy, so ESD applies — compute |X(f)|² directly from the Fourier transform. The Wiener-Khinchin theorem links PSD to the autocorrelation function via R_x(τ) ↔ S_x(f), a tool that works only for WSS random processes — it breaks down for transient or deterministic finite-duration signals. The units give you the check: if your spectral measure integrates to watts, it's PSD; if it integrates to joules, it's ESD.

When to use Power Spectral Density

Use PSD for any signal that runs continuously or indefinitely — AWGN channel noise in a 5G NR receiver, thermal noise across a 50 Ω resistor at 290 K (N₀/2 = kT = 4×10⁻²¹ W/Hz), or a periodic clock signal.

When to use Energy Spectral Density

Use ESD for signals with a definite start and end — a single BPSK symbol burst, a captured 5 ms ultrasound echo, or a finite-length ECG heartbeat segment extracted for arrhythmia analysis.

Recommendation

Choose PSD for noise and continuous signals; choose ESD for pulses and transients. The Wiener-Khinchin theorem is your computational shortcut for PSD — never try to Fourier-transform an infinite-duration noise signal directly. Most GATE problems on this topic hinge on correctly identifying whether the signal has finite energy or finite power first.

Exam tip: GATE repeatedly tests Parseval's theorem in both forms: ∫|x(t)|²dt = ∫E_x(f)df for energy signals and (1/T)∫|x(t)|²dt = ∫S_x(f)df for power signals — write both forms and state which applies to which signal type.

Interview tip: Interviewers at communication system companies ask you to compute noise power in a bandwidth B given N₀ — the answer is N₀×B, which comes directly from integrating PSD over the band — show that you know PSD integrates to power, not energy.

More Digital Signal Processing comparisons