Probability of Detection

The single most dominant factor. Pd increases monotonically with the received SNR. In low-SNR environments, such as deep fade or shadowed conditions, even optimal detectors struggle. The SNR Wall phenomenon dictates that below a certain threshold, dictated by noise uncertainty, no amount of sensing time can guarantee reliable detection. System designers must account for worst-case path loss and fading margins.

A longer sensing duration allows the detector to integrate more signal samples, effectively averaging out random noise and improving the Pd for a fixed false alarm rate. However, this creates the Sensing-Throughput Tradeoff: every millisecond spent sensing is a millisecond not spent transmitting data. Sequential detection techniques optimize this by stopping the sensing process as soon as a reliable decision can be made.

The choice of algorithm creates a performance hierarchy. Energy detection is simple but vulnerable to noise uncertainty. Cyclostationary feature detection exploits the periodicity of modulated signals, offering robustness at low SNR but at high computational cost. Eigenvalue-based detection and matched filter detection provide superior performance when signal or noise statistics are known, representing the upper bound of achievable Pd.

A single sensor is vulnerable to the hidden node problem, where multipath fading or shadowing causes a missed detection. Cooperative spectrum sensing mitigates this by fusing observations from spatially diverse nodes. The fusion strategy matters: soft decision fusion (combining raw energy levels) significantly outperforms hard decision fusion (combining binary votes) by preserving statistical information, albeit at the cost of higher backhaul bandwidth.

Wireless channels are not static. Rayleigh fading and shadowing cause deep fluctuations in instantaneous received power. A detector that performs perfectly in an AWGN channel may fail during a deep fade. Robust systems employ diversity techniques, such as multiple antennas or cooperative sensing, to create a composite channel with reduced variance and a higher probability of a non-faded signal path.

The Pd is highly dependent on the signal's structure. A constant-envelope FM signal is easier for an energy detector to find than a low-duty-cycle bursty packet transmission. Known patterns like pilot tones or cyclic prefixes (in OFDM) provide structured features that enable high-Pd semi-blind detection without requiring a full signal template, bridging the gap between blind and matched filter approaches.

The ROC curve is the canonical visualization of the trade-off between Probability of Detection (Pd) and Probability of False Alarm (Pfa). It plots Pd against Pfa as the detector's discrimination threshold varies. A perfect detector achieves Pd=1 for any Pfa>0, while a random guess yields a diagonal line. The Area Under the Curve (AUC) quantifies overall sensing efficacy, with values approaching 1.0 indicating robust performance even at low signal-to-noise ratios.

Defined as Pmd = 1 - Pd, this is the conditional probability that a sensing algorithm fails to declare a band occupied when a primary user is actively transmitting. A missed detection is the most catastrophic sensing error, as it causes the secondary user to transmit simultaneously with the licensed incumbent, generating harmful co-channel interference. Regulatory bodies often mandate a maximum Pmd (e.g., < 0.01) to ensure incumbent protection.

The conditional probability that a detector incorrectly declares a vacant band as occupied. While not harmful to primary users, a high Pfa directly degrades spectral efficiency by causing the cognitive radio to miss viable transmission opportunities. The Constant False Alarm Rate (CFAR) algorithm dynamically adjusts the detection threshold to maintain a fixed Pfa despite fluctuating noise power, ensuring predictable spectrum access behavior.

A distributed architecture where multiple spatially separated cognitive radios share local sensing observations to combat the hidden node problem. By fusing data at a Fusion Center, the network mitigates the effects of multipath fading and shadowing that can cause a single sensor to miss a primary transmission. Hard decision fusion (combining binary local votes via AND/OR/K-out-of-N rules) and soft decision fusion (combining raw energy statistics) both dramatically improve the global Probability of Detection compared to standalone sensing.

A fundamental theoretical limit for non-coherent detectors like the energy detector. Due to noise uncertainty—the inevitable imprecision in estimating ambient noise power—there exists a minimum SNR below which reliable detection is impossible, regardless of how long the sensor observes the channel. Below the SNR Wall, Pd cannot be driven to 1 while maintaining a low Pfa. This motivates the use of cyclostationary feature detection or matched filter detection, which exploit known signal structure to bypass this limit.

The fundamental frame-structure tension in cognitive radio. A longer sensing duration increases the number of samples, improving Pd and reducing Pfa for a given SNR. However, this directly reduces the time remaining for data transmission within the fixed frame, lowering the secondary user's achievable throughput. The optimal sensing time maximizes the average throughput subject to a regulatory constraint on the maximum acceptable Missed Detection Probability (Pmd), balancing incumbent protection against spectral efficiency.