Constant False Alarm Rate (CFAR)

The foundational adaptive threshold technique that estimates the local noise power by calculating the arithmetic mean of the amplitudes within a set of reference cells surrounding the Cell Under Test (CUT).

• Optimal Performance: Achieves the highest probability of detection in homogeneous, exponentially distributed background clutter.
• Guard Cells: A ring of empty cells immediately adjacent to the CUT prevents signal energy from leaking into the reference window, which would bias the noise estimate upward.
• Key Limitation: Suffers severe masking effects in multi-target environments, where a closely spaced interfering target in the reference window inflates the threshold and suppresses detection of the primary target.

Designed specifically to control false alarms at clutter edges—the transition zone between two regions of vastly different noise power. GO-CFAR splits the reference window into a leading and lagging half, calculates the mean of each, and selects the greater of the two as the noise estimate.

• Clutter Edge Robustness: Prevents the excessive false alarms that plague CA-CFAR when the CUT is in the low-power region but the reference window includes high-power clutter.
• Trade-off: Exhibits degraded detection performance in homogeneous environments compared to CA-CFAR, as the maximum operator introduces a constant bias into the threshold multiplier.

The complementary strategy to GO-CFAR, optimized to resolve closely spaced targets. SO-CFAR computes the mean of the leading and lagging reference windows independently and selects the smaller of the two as the noise estimate.

• Multi-Target Resolution: If an interfering target falls entirely within one half of the reference window, the uncontaminated half provides a lower, uncorrupted noise estimate, allowing the primary target to be detected.
• Critical Weakness: Cannot control false alarms at clutter edges, as the smaller operator will select the low-power region's mean when the CUT is in the high-power region, causing a catastrophic spike in false detections.

A rank-based approach that abandons the arithmetic mean in favor of selecting the k-th largest sample from the sorted reference window as the noise estimate. This non-linear technique is inherently immune to multiple interfering targets, provided the number of interferers is less than the rank order parameter.

• Rank Parameter (k): Typically set to 3N/4, where N is the total number of reference cells. This value provides a robust compromise between CA-CFAR-like performance in homogeneous noise and tolerance to outliers.
• Computational Cost: Requires real-time sorting of the reference window, which historically limited its use to systems with dedicated digital signal processing hardware, though modern FPGAs have largely eliminated this constraint.

An iterative approach that actively identifies and excises interfering targets from the reference window before computing the noise estimate. C-CFAR applies an initial threshold to the reference cells, removes those exceeding it, and recalculates the mean using only the remaining, presumed noise-only, samples.

• Adaptive Censoring: The censoring depth dynamically adjusts to the number of detected interferers, making it highly effective in dense target environments where OS-CFAR's fixed rank parameter might fail.
• Convergence: Requires multiple iterations to stabilize the set of censored cells, introducing latency that must be accounted for in real-time system design.

An adaptive, composite architecture that dynamically selects between CA-CFAR, GO-CFAR, and SO-CFAR based on real-time statistical tests performed on the reference window. VI-CFAR computes a variability index and a mean ratio to classify the environment as homogeneous, clutter edge, or multi-target.

• Hypothesis Testing: The variability index tests whether the reference cells are identically distributed, while the mean ratio compares the leading and lagging window averages.
• Decision Logic: A deterministic logic table maps the outcomes of these two tests to the optimal CFAR variant, providing near-optimal performance across all canonical interference scenarios without prior environmental knowledge.

A graphical plot illustrating the diagnostic ability of a binary classifier as its discrimination threshold varies. In spectrum sensing, the ROC curve plots Probability of Detection (Pd) against False Alarm Probability (Pfa). CFAR algorithms are explicitly designed to operate at a single, fixed point on this curve, maintaining a constant Pfa regardless of noise fluctuations. The area under the ROC curve (AUC) serves as a single-figure metric for comparing detector performance.

The inherent fluctuation in ambient noise power that fundamentally limits the performance of energy detectors. In practical receivers, thermal noise variance is not perfectly known and drifts due to temperature changes, component aging, and calibration errors. This uncertainty creates an SNR Wall below which reliable detection becomes impossible, regardless of sensing duration. CFAR processors mitigate this by continuously estimating the local noise floor from reference cells.

A blind spectrum sensing technique that measures the energy of received samples and compares it against a threshold. It requires no prior knowledge of the primary user's signal structure, making it computationally simple. However, its performance is highly sensitive to noise power estimation errors. CFAR is the standard threshold-setting mechanism for energy detectors, enabling them to maintain a stable false alarm rate in dynamic noise environments.

The theoretical minimum SNR threshold below which a non-coherent detector cannot reliably distinguish a signal from noise, even with an infinite observation time. This fundamental limit arises directly from noise uncertainty. CFAR algorithms do not eliminate the SNR wall but ensure the detector operates at the optimal threshold for the estimated noise floor, preventing catastrophic failure due to threshold misplacement.

The conditional probability that a sensing algorithm incorrectly declares a frequency band as occupied when it is actually vacant. A false alarm represents a missed transmission opportunity for a secondary user. CFAR algorithms hold this value constant—typically at 0.01 or 0.001—by adaptively scaling the detection threshold based on the estimated noise power in surrounding reference cells.

The conditional probability that a sensing algorithm correctly identifies an active primary user signal. Maximizing Pd is critical to avoid harmful interference. In CFAR systems, Pd is not fixed but varies with the instantaneous Signal-to-Interference-plus-Noise Ratio (SINR). The design goal is to maximize Pd for a given, constant Pfa constraint—a criterion known as the Neyman-Pearson optimal detector.