Neyman-Pearson Criterion

The Neyman-Pearson criterion solves a constrained optimization problem: maximize the probability of detection (P_d) subject to an upper bound constraint on the probability of false alarm (P_f). Unlike Bayesian approaches that minimize average cost, this framework treats missed detections and false alarms asymmetrically. In cognitive radio, this asymmetry is critical—failing to detect a primary user causes harmful interference, while a false alarm merely results in a missed transmission opportunity. The constraint P_f ≤ α, where α is typically set to 0.1 or 0.01, ensures primary users receive guaranteed protection.

The Neyman-Pearson lemma proves that the most powerful test at a given significance level α is the Likelihood Ratio Test (LRT). The decision rule compares the ratio of probability density functions under hypotheses H₁ (signal present) and H₀ (signal absent) against a threshold λ:

• L(y) = p(y|H₁) / p(y|H₀) > λ → Decide H₁
• L(y) < λ → Decide H₀

The threshold λ is chosen to satisfy the false alarm constraint P_f = α. This structure guarantees that no other test can achieve a higher P_d for the same P_f, making it the gold standard for spectrum sensing fusion rules.

The detection threshold λ is not arbitrary—it is uniquely determined by the false alarm constraint. The process involves:

• Computing the distribution of the test statistic under H₀ (noise-only hypothesis)
• Setting the threshold such that the tail probability beyond λ equals α
• For energy detection with Gaussian noise, this yields a threshold based on the chi-squared distribution

This direct coupling between λ and P_f means that as the noise environment changes, the threshold must adapt to maintain a Constant False Alarm Rate (CFAR). In cooperative sensing, the fusion center applies this principle to aggregate local test statistics.

In cooperative spectrum sensing, the Neyman-Pearson criterion extends naturally to fusion rule design. For soft decision fusion, the optimal combining rule is a weighted sum of local likelihood ratios. For hard decision fusion, the K-out-of-N rule approximates the Neyman-Pearson solution when local decisions are treated as binary observations. The fusion center:

• Collects local test statistics or decisions from N sensing nodes
• Forms a global likelihood ratio by exploiting spatial diversity
• Applies a global threshold satisfying the network-wide P_f constraint

This framework optimally mitigates the hidden node problem while guaranteeing primary user protection across the entire cooperative network.

The Receiver Operating Characteristic (ROC) curve is the primary visualization tool for evaluating Neyman-Pearson detectors. It plots P_d against P_f as the threshold varies, revealing the fundamental tradeoff:

• Ideal performance: ROC curve hugs the upper-left corner (P_d → 1, P_f → 0)
• Random guessing: ROC follows the diagonal line P_d = P_f
• Area Under Curve (AUC) quantifies overall detection capability

In spectrum sensing, ROC curves enable direct comparison of different detection algorithms—energy detection, cyclostationary feature detection, and matched filter detection—under identical signal-to-noise ratio conditions. The Neyman-Pearson criterion selects the operating point on the ROC curve corresponding to the specified P_f constraint.

While theoretically optimal, implementing the Neyman-Pearson criterion faces practical hurdles:

• Channel state information: The full likelihood ratio requires knowledge of the primary user's signal parameters and channel gains, which are often unavailable
• Noise uncertainty: Imprecise noise power estimation shifts the effective P_f, potentially violating the constraint and creating an SNR wall below which detection becomes impossible
• Reporting channel errors: In cooperative sensing, imperfect reporting channels between nodes and the fusion center corrupt the likelihood ratio, requiring robust modifications

These challenges motivate suboptimal but robust alternatives like blind sensing and double threshold detection, which trade some optimality for practical resilience.

The Receiver Operating Characteristic is the graphical plot that visualizes the tradeoff dictated by the Neyman-Pearson Criterion. It maps the Probability of Detection (Pd) on the y-axis against the Probability of False Alarm (Pfa) on the x-axis. The Neyman-Pearson optimal detector achieves the highest possible Pd for any given constrained Pfa, pushing the ROC curve as close to the top-left corner as possible. The area under the ROC curve (AUC) is a standard aggregate metric for comparing sensing algorithm performance independent of a specific threshold.

The Likelihood Ratio Test is the mathematical mechanism that implements the Neyman-Pearson Criterion. It constructs a test statistic by computing the ratio of the probability density functions under two hypotheses: H1 (signal present) and H0 (signal absent). If this ratio exceeds a threshold λ, the detector declares H1. The threshold λ is uniquely determined by the maximum tolerable probability of false alarm constraint. While optimal, the full LRT often requires perfect knowledge of channel state information and signal parameters, making it a theoretical gold standard rather than a practical blind sensing solution.

The Probability of False Alarm is the constrained variable in the Neyman-Pearson framework. It quantifies the likelihood that a spectrum sensor incorrectly declares a frequency band occupied when it is actually vacant. This represents a missed transmission opportunity for the secondary user. In cooperative sensing, a fusion center's global Pfa is a function of the local sensor Pfa values and the fusion rule. The Neyman-Pearson Criterion explicitly sets an upper bound on this value (e.g., Pfa ≤ 0.1) to ensure a minimum level of spectral efficiency.

The Probability of Detection is the metric maximized by the Neyman-Pearson Criterion. It measures the statistical likelihood that a sensor correctly identifies a primary user's transmission when the signal is actually present. A high Pd is critical for primary user protection to prevent harmful interference. In a cooperative sensing network with a Fusion Center, the global Pd is typically improved through spatial diversity, as the Neyman-Pearson optimal fusion rule combines observations to achieve a higher Pd for the same global Pfa constraint.

Constant False Alarm Rate processing is the practical engineering implementation of the Neyman-Pearson Criterion in dynamic noise environments. Since the optimal threshold λ depends on noise power, CFAR algorithms dynamically estimate the ambient noise floor and continuously adjust the detection threshold to maintain a fixed, pre-defined Pfa. This prevents noise uncertainty from causing excessive false alarms. Common CFAR methods include cell-averaging CFAR, which estimates local noise from adjacent range bins or frequency cells, ensuring the detector operates at the designed point on the ROC curve.

Hard Decision Fusion is a bandwidth-efficient cooperative strategy where each sensing node independently applies the Neyman-Pearson Criterion locally. Instead of transmitting raw test statistics, nodes send a single binary decision (1 or 0) to the Fusion Center. The center then applies a global voting rule, such as the K-out-of-N rule. The global probabilities of detection and false alarm are derived from the binomial distribution of local decisions. While suboptimal compared to soft combining, it minimizes reporting channel overhead and is robust to individual node failures.