A statistical framework for deciding between competing hypotheses about observed signals in the presence of noise and interference, typically by defining decision rules that manage tradeoffs between error types (false alarms and misses) under specified probabilistic models or cost constraints.

Principle

For simple hypotheses under common regularity conditions, likelihood-ratio–based tests provide decision rules that optimize specified criteria (e.g., maximize detection probability for a fixed false-alarm rate); more generally, choice of decision rule reflects the tradeoff between error types and the assumed priors or costs.

Demonstration

Illustrative scenario → A radar receiver must decide whether a target is present (H1) or absent (H0) from noisy returns. The receiver computes a test statistic (e.g., matched-filter output) and compares it to a threshold set to achieve a target false-alarm rate; raising the threshold reduces false alarms but increases missed detections, illustrating the operational tradeoff.

Misapplication

Interpreting a detection decision as precise localization or as proof that the estimated signal parameters are accurate. The semantic error is conflating hypothesis decision (presence/absence) with parameter estimation or with an absolute statement of truth independent of model and threshold choices.

Consequence

Detection theory provides the basis for setting decision thresholds, computing receiver operating characteristic (ROC) curves, allocating sensing resources, and specifying performance guarantees under the chosen model and error-cost framing.

Reversal

When noise statistics, independence assumptions, or cost/prior information are unknown or incorrect, classical optimality results (e.g., Neyman–Pearson) may not apply; robust, minimax, or Bayesian alternatives and generalized detectors are then appropriate.

Boundary

Within: hypothesis testing for presence/absence or between a small number of specified signal models under an explicit statistical model. Boundary case: detection among many closely spaced hypotheses where problem blends into classification. Outside: detailed parameter estimation or inference about continuous-valued parameters where decision outcomes are insufficient summaries.

Semantic Tension

Sensitivity (Detection Probability) ↔ Specificity (False-Alarm Rate) — improving one typically worsens the other, requiring contextual selection of acceptable tradeoffs.

Synthesis

Detection Theory formalizes decision-making about signal presence as a tradeoff problem: it prescribes how to convert noisy observations into thresholded decisions given models and performance criteria, while distinguishing decision correctness from parameter accuracy or semantic certainty.