Receiver Operating Characteristic (ROC)

The ROC curve is a graphical plot with the Probability of False Alarm (P_fa) on the x-axis and the Probability of Detection (P_d) on the y-axis. Each point on the curve represents a (P_fa, P_d) pair for a specific detection threshold. A perfect classifier would have a point at (0,1). The diagonal line from (0,0) to (1,1) represents the performance of a random guess. The curve visually demonstrates that increasing sensitivity to detect signals inevitably increases the chance of mistaking noise for a signal.

The Area Under the ROC Curve (AUC) is a single scalar metric summarizing overall classifier performance, independent of any specific threshold. It represents the probability that the classifier will rank a randomly chosen positive instance higher than a randomly chosen negative one. An AUC of 1.0 signifies a perfect classifier, while an AUC of 0.5 indicates performance no better than random chance. In spectrum sensing, a higher AUC means the sensor is fundamentally better at distinguishing a primary user's signal from noise across all possible threshold settings.

A spectrum sensor's operating point on the ROC curve is selected by tuning its detection threshold. A lower threshold increases the Probability of Detection (P_d) , providing better protection for primary users, but also increases the Probability of False Alarm (P_fa) , wasting secondary transmission opportunities. Conversely, a higher threshold reduces false alarms but risks missing active primary users. The choice of operating point is a critical design decision governed by regulatory constraints, such as the IEEE 802.22 standard's requirement for a P_d of 0.9.

The Neyman-Pearson (NP) criterion provides the theoretical foundation for setting an optimal detection threshold. The NP framework formulates the problem as: maximize the Probability of Detection (P_d) subject to an upper-bound constraint on the Probability of False Alarm (P_fa) . This is the most common design objective in spectrum sensing, where a regulator might mandate a maximum P_fa of 0.1. The resulting NP detector is a Likelihood Ratio Test (LRT), which is the most powerful test for a given P_fa constraint.

The Signal-to-Noise Ratio (SNR) is the dominant factor shaping the ROC curve. At high SNR, the curve bows sharply toward the top-left corner, indicating near-perfect classification. As SNR decreases, the curve sags toward the diagonal, reflecting degraded discriminability. This creates the SNR Wall phenomenon for certain detectors like the energy detector: below a critical SNR level, the constraints on P_d and P_fa cannot be simultaneously met, no matter how long the sensing duration, making reliable detection impossible.

Cooperative Spectrum Sensing (CSS) directly improves the ROC curve by exploiting spatial diversity. When multiple geographically separated nodes experience independent fading, their combined decision at a fusion center yields a higher P_d for the same P_fa compared to a single node. The global ROC curve shifts closer to the ideal top-left corner. The choice of fusion rule—such as the K-out-of-N rule or soft decision combining—determines the exact shape of this improved global ROC curve.

The true positive rate—the likelihood that a spectrum sensing algorithm correctly identifies a primary user signal when it is actually transmitting. On the ROC curve, Pd is plotted on the y-axis. A high Pd (e.g., > 0.9) is mandated by regulators to ensure primary user protection. In cooperative sensing, Pd is the key metric for evaluating global decision accuracy at the fusion center.

The false positive rate—the likelihood that a sensing algorithm incorrectly declares a frequency band occupied when it is actually vacant. Plotted on the x-axis of the ROC curve, Pfa represents a missed spectrum opportunity for secondary users. The Neyman-Pearson criterion formulates the detection problem as maximizing Pd subject to a fixed Pfa constraint, typically Pfa ≤ 0.1.

A single scalar metric summarizing the overall diagnostic ability of a classifier across all possible thresholds. An AUC of 1.0 represents a perfect classifier; 0.5 indicates random guessing. In spectrum sensing, AUC provides a threshold-agnostic comparison between different detection algorithms (e.g., energy detection vs. cyclostationary feature detection) without committing to a specific operating point.

The fundamental design conflict where longer sensing durations improve Pd (moving the operating point up the ROC curve) but reduce the time available for data transmission. The optimal sensing time is found by solving a constrained optimization that maximizes secondary user throughput while satisfying a minimum Pd requirement. This tradeoff is the practical engineering consequence of the ROC curve's shape.

A fundamental limit below which reliable detection is impossible, regardless of sensing duration, due to irreducible noise uncertainty. The SNR wall manifests on the ROC curve as a collapse of Pd and Pfa toward the diagonal (random guessing). This phenomenon motivates the use of cyclostationary feature detection and cooperative sensing architectures that exploit spatial diversity to overcome individual node limitations.