Constant False Alarm Rate (CFAR)

The core mechanism of CFAR is the dynamic adjustment of the detection threshold. Unlike a fixed threshold that would be overwhelmed by noise spikes, CFAR continuously estimates the local noise power from surrounding reference cells and scales the threshold accordingly.

• Cell Under Test (CUT): The specific range/Doppler bin being evaluated for a target.
• Reference Window: A sliding window of adjacent cells used to estimate the instantaneous noise power.
• Guard Cells: Buffer cells immediately adjacent to the CUT, excluded from the noise estimate to prevent target signal energy from leaking in and biasing the threshold upward.

OS-CFAR replaces the arithmetic mean with the k-th ordered statistic of the reference window. By sorting the reference cell values and selecting a specific percentile, OS-CFAR gains inherent robustness against interfering targets.

• Multi-Target Resilience: If a few reference cells contain interfering targets, their high values are sorted to the top and ignored when selecting a lower-order statistic like the median.
• Clutter Edge Performance: Exhibits much lower false alarm rate spiking at noise-to-clutter transitions compared to CA-CFAR.
• Trade-off: Suffers from slightly higher CFAR loss in homogeneous noise compared to CA-CFAR due to the higher variance of order statistics.

Variants designed to handle specific non-homogeneities by splitting the reference window into leading and lagging halves and applying a logical operation.

• GO-CFAR (Greatest-Of): Selects the maximum of the two half-window noise estimates. Excels at controlling false alarms at clutter edges, where one half-window enters a high-power region before the other.
• SO-CFAR (Smallest-Of): Selects the minimum of the two half-window estimates. Optimized for detecting a target closely followed by a stronger interfering target, where the interfering target contaminates only one half-window.

CFAR detection incurs a performance penalty compared to a theoretical optimal detector with perfect noise knowledge. This CFAR loss is the additional signal-to-noise ratio required to achieve the same detection probability.

• Finite Reference Window: The estimation error from a limited number of reference cells is the primary source of CFAR loss. More cells reduce variance but limit spatial resolution.
• Noise Uncertainty: In practical systems, the true noise power is never known perfectly. Below a certain SNR wall, no detector can reliably distinguish signal from noise regardless of observation time, rendering detection impossible.

In spectrum sensing, CFAR principles are applied to energy detection to maintain a constant probability of false alarm (PFA) against an uncertain noise floor. This directly controls the likelihood of declaring a vacant channel occupied, which would waste transmission opportunities.

• Neyman-Pearson Criterion: The theoretical foundation—maximize probability of detection subject to a fixed PFA constraint.
• Noise Estimation: The reference window estimates the ambient noise plus interference floor, allowing the threshold to adapt to changing electromagnetic environments.
• Cooperative Sensing: In distributed networks, CFAR thresholds can be applied locally at each node before hard decisions are sent to a fusion center, or globally at the fusion center on combined soft statistics.