Fusion Center

A bandwidth-efficient strategy where individual cognitive radios transmit only a binary local decision (0 or 1) to the fusion center. The center then applies a logical combining rule to form a global inference.

• Logical Rules: AND (conservative, minimizes interference), OR (aggressive, maximizes secondary access), and K-out-of-N (flexible majority logic).
• Trade-off: Minimizes reporting channel overhead at the cost of information loss, degrading detection performance compared to soft fusion.
• Vulnerability: Highly susceptible to Byzantine attacks where a single malicious node flips its bit to corrupt the global decision.

A high-fidelity strategy where sensors transmit raw or quantized sensing statistics (e.g., energy levels, likelihood ratios) to the fusion center, preserving granular information for optimal global detection.

• Common Schemes: Equal Gain Combining (EGC), Maximal Ratio Combining (MRC), and Likelihood Ratio Test (LRT) fusion.
• Performance: Achieves superior detection probability and resilience against the SNR wall compared to hard fusion, at the expense of higher reporting channel bandwidth.
• Quantization: Practical implementations often use multi-bit quantization to balance the trade-off between precision and backhaul load.

A security framework addressing the threat of Spectrum Sensing Data Falsification (SSDF) attacks, where compromised nodes send manipulated reports to the fusion center to cause harmful interference or selfishly monopolize spectrum.

• Reputation-Based Fusion: The fusion center maintains a dynamic trust score for each sensor, weighting their reports based on historical consistency with the global decision.
• Outlier Detection: Statistical methods like Dixon's Q-test or clustering algorithms identify and isolate anomalous reports before fusion.
• Sequential Probability Ratio Test (SPRT): A robust sequential detection method that adaptively evaluates sensor trustworthiness over time to mitigate persistent attackers.

A practical compromise between hard and soft fusion where sensors transmit multi-bit quantized versions of their local test statistics to the fusion center, optimizing the bandwidth-performance trade-off.

• Optimal Quantization: Lloyd-Max algorithms design non-uniform quantizers that minimize mean-squared error for the specific distribution of the sensing statistic.
• Censoring: Sensors only transmit when their local observation exceeds a threshold, dramatically reducing reporting channel traffic in sparse signal environments.
• Channel-Aware Fusion: The fusion center incorporates the instantaneous quality of the reporting channel itself into the global likelihood ratio, preventing corrupted reports from degrading the final decision.

A fundamental degradation in cooperative sensing caused when a cognitive radio is shadowed or in deep fade relative to a transmitting primary user, leading to a missed local detection.

• Mechanism: Physical obstructions like buildings or terrain attenuate the primary signal at the sensor's location, causing its energy detector to report a false 'vacant' state.
• Mitigation via Fusion: Cooperative architectures with a fusion center inherently solve this by aggregating spatially diverse observations—a node with a clear line-of-sight compensates for the shadowed node.
• Spatial Correlation Modeling: Advanced fusion algorithms model the correlated shadowing between nearby sensors to avoid overconfident global decisions when multiple nodes are simultaneously obstructed.

The statistically optimal fusion rule under the Neyman-Pearson criterion, where the fusion center computes the ratio of the joint probability density of all sensor reports under the signal-present hypothesis to that under the signal-absent hypothesis.

• Optimality: Minimizes the missed detection probability for a given constraint on the false alarm probability, serving as the theoretical performance benchmark.
• Implementation Complexity: Requires precise knowledge of each sensor's local detection and false alarm probabilities, as well as the statistical distribution of their observations.
• Approximations: Suboptimal but practical variants like the Chair-Varshney rule use locally optimal quantized data to approach LRT performance with significantly reduced computational burden.