Field Guide · algorithm

Also known as: energy detection, radiometry, radiometer detector

Energy detection decides whether a signal occupies a band by measuring the total energy (or power) in that band over an observation window and comparing it to a threshold: above the threshold means “signal present”, below means “noise only”.1 It is the classic radiometer, and its defining virtue is that it needs no knowledge of the signal’s modulation, timing, or structure — it is a blind detector that works on anything with more power than the background.

bandpassfilter (·)² ∫ overwindow E ≷ λ ?decide x(t) H₁/H₀ threshold λ set from the noise floor and the target false-alarm rate
Energy detection filters to the band of interest, squares and integrates to form an energy statistic, then thresholds it to choose between "signal present" and "noise only".

How it works

The detector forms a test statistic E = Σ |x[n]|² — the accumulated squared magnitude of the (band-limited) samples over an observation window of N points. Under the noise-only hypothesis H₀ this statistic is a scaled chi-square variable centred on the noise power; under H₁ it is shifted upward by the signal energy. Pick a threshold λ: the noise floor and N fix the false-alarm probability P_fa (how often E crosses λ on noise alone), and the SNR and N then fix the detection probability P_d. Longer integration (larger N) trades latency for sensitivity — you can dig a weak signal out by averaging longer, up to the limits set by noise-power uncertainty.

In practice

Energy detection’s weakness is the flip side of its blindness: it cannot tell signal from noise-like interference, and it is only as good as its knowledge of the noise power. If the true noise floor is uncertain by even a fraction of a dB, there is an SNR wall below which no amount of integration makes the decision reliable, because the signal energy is indistinguishable from a slightly-higher noise estimate. Making the threshold adaptive — estimating the local noise from neighbouring bins, exactly the CFAR idea — is the standard fix. Where the signal is known, a matched filter or cyclostationary detector beats energy detection by exploiting that structure, at the cost of needing it.

Relevance to SDR

Energy detection is the backbone of spectrum sensing in cognitive radio (is this channel free to use?), spectrum-occupancy monitoring, and simple presence/squelch decisions, and it is a natural companion to a Welch-averaged power spectrum. GopherTrunk relies on power/quality thresholding to decide whether a control or voice channel carries a usable signal — an energy-detection-style decision in spirit — while its actual symbol recovery uses structured, matched-filter-based demodulation rather than a bare radiometer. Energy detection is the simplest member of the detection-theory family that also includes CFAR and matched filtering.

Sources

  1. Detection theory — Wikipedia, on hypothesis testing and the energy/radiometer detector for signal presence. 

See also