Fusion Center

In hard combining, each cooperating node makes a local binary decision (1 for occupied, 0 for vacant) and transmits only this single bit to the fusion center. The fusion center then applies a logical rule to reach a global decision.

• AND Rule: Declares a channel occupied only if all nodes report a detection. Minimizes false alarms but increases missed detection probability.
• OR Rule: Declares a channel occupied if any node reports a detection. Maximizes sensitivity but increases false alarm rate.
• K-out-of-N Rule: Declares a channel occupied if at least K out of N nodes report a detection, providing a tunable balance between the AND and OR extremes.

Hard combining minimizes control channel bandwidth requirements but discards valuable signal quality information from individual sensors.

Soft combining transmits the raw or quantized sensing statistics from each node to the fusion center, which constructs a global test statistic before making a decision. This preserves more information than hard combining.

• Equal Gain Combining (EGC): The fusion center sums the received signal energies from all nodes with equal weighting. Simple but suboptimal when nodes experience different channel conditions.
• Maximal Ratio Combining (MRC): The fusion center weights each node's contribution by its instantaneous signal-to-noise ratio (SNR), giving more influence to nodes with better reception. This is the optimal linear combining scheme.
• Likelihood Ratio Test (LRT): The fusion center computes the ratio of probability densities under the signal-present and signal-absent hypotheses, providing the theoretically optimal detection performance at the cost of requiring full channel knowledge.

Soft combining achieves superior detection performance, especially in fading environments, at the expense of higher control channel bandwidth.

The physical link between cooperating nodes and the fusion center is called the reporting channel, and its imperfections directly degrade cooperative sensing performance.

• Fading and Shadowing: Deep fades on the reporting channel can cause individual node decisions or soft statistics to be lost entirely, effectively removing that node from the cooperative pool.
• Bandwidth Constraints: The reporting channel has finite capacity, forcing a trade-off between the number of cooperating nodes and the quantization resolution of their reported observations.
• Latency and Synchronization: Time misalignment between sensing observations arriving at the fusion center can cause decisions to be made on stale or temporally mismatched data, reducing correlation benefits.
• Censoring Strategies: To conserve bandwidth, nodes may be programmed to transmit only when their local test statistic exceeds a threshold, a technique known as censoring that reduces reporting channel load while maintaining most of the cooperative gain.

The centralized architecture of a fusion center creates a single point of failure and an attractive target for adversarial attacks that can cripple the entire cooperative sensing network.

• Spectrum Sensing Data Falsification (SSDF): Malicious nodes deliberately report false sensing data to the fusion center. An attacker sending a constant 'occupied' signal can trigger a denial-of-service, while a constant 'vacant' signal can cause harmful interference to primary users.
• Fusion Center Compromise: If the fusion center itself is taken over, the attacker gains control over all global spectrum access decisions for the entire cooperative network.
• Reputation-Based Mitigation: Defense mechanisms assign trust scores to each cooperating node based on the historical consistency of their reports with the global decision. Nodes with low reputation are progressively de-weighted or excluded.
• Sequential Probability Ratio Test (SPRT): A robust fusion technique that continuously accumulates evidence over time and makes a decision only when sufficient confidence is reached, making it more resistant to intermittent falsification attacks than fixed-sample-size tests.

In large-scale networks, a flat architecture with all nodes reporting to a single fusion center becomes impractical due to reporting channel congestion and scalability limits. Hierarchical fusion addresses this by introducing intermediate aggregation layers.

• Cluster Head Selection: Nodes are organized into geographic or logical clusters. Each cluster elects a cluster head that performs local fusion of its members' observations before forwarding a condensed report to the global fusion center.
• Two-Stage Decision: The first stage occurs at the cluster head using hard or soft combining. The second stage occurs at the global fusion center, which combines the cluster-level decisions. This dramatically reduces backbone traffic.
• Energy Efficiency: By limiting long-range transmissions to cluster heads only, hierarchical architectures significantly extend the operational lifetime of battery-powered sensing nodes in wireless sensor networks.
• Scalability: The hierarchical model supports networks with hundreds or thousands of cooperating nodes without overwhelming the fusion center or the common control channel.

The design of the fusion rule directly determines the receiver operating characteristic (ROC) of the cooperative sensing system. The optimal rule depends on the available information and the performance criteria.

• Chair-Varshney Rule: The optimal likelihood ratio-based fusion rule for hard decisions that accounts for each node's individual probability of detection and false alarm. It assigns higher weight to more reliable nodes.
• Neyman-Pearson Criterion: The fusion center maximizes the global probability of detection subject to a constraint on the global probability of false alarm, a common requirement in regulatory frameworks where primary user protection is paramount.
• Bayesian Criterion: The fusion center minimizes the average cost of decision errors by incorporating prior probabilities of primary user activity and assigning different costs to false alarms versus missed detections.
• Deep Learning Fusion: Modern approaches replace analytically derived fusion rules with neural networks trained end-to-end to map raw node observations directly to global decisions, learning optimal combining strategies from data without explicit channel models.

The collaborative detection framework where multiple secondary users share local sensing observations with the fusion center to overcome the hidden node problem and multipath fading. By leveraging spatial diversity, cooperative sensing dramatically improves detection reliability compared to single-node sensing.

• Mitigates shadowing and fading effects
• Increases aggregate signal-to-noise ratio (SNR)
• Reduces individual sensing time requirements

A class of fusion logic where each cooperating node transmits a binary local decision (1 for occupied, 0 for vacant) to the fusion center. The center then applies a logical rule to generate the global decision.

• AND Rule: Declares channel occupied only if all nodes report occupied; minimizes false alarms
• OR Rule: Declares occupied if any single node detects a signal; minimizes missed detections
• K-out-of-N Rule: Requires at least K nodes to report occupied; balances the tradeoff between AND and OR logic

Fusion strategies where nodes transmit their raw energy measurements or full test statistics rather than binary decisions. The fusion center aggregates these analog values to compute a more nuanced global test statistic.

• Equal Gain Combining (EGC): Sums all energy values with equal weight
• Maximal Ratio Combining (MRC): Weights each node's observation by its instantaneous SNR
• Likelihood Ratio Test (LRT): The statistically optimal fusion rule requiring channel state knowledge

A dedicated signaling channel used by cognitive radio nodes to transmit sensing reports to the fusion center and receive global decisions. The CCC must be always available and protected from interference.

• Must operate independently of the sensed spectrum bands
• Requires robust modulation to ensure reliable report delivery
• Vulnerable to saturation attacks if not properly managed

A critical sensing vulnerability where a secondary user is physically obstructed from detecting a primary transmitter due to terrain, buildings, or distance. The node reports the channel as vacant, but its transmission would cause harmful interference to a primary receiver in its vicinity.

• Primary cause of missed detection in single-node sensing
• The fundamental motivation for cooperative sensing architectures
• Fusion centers mitigate this by aggregating spatially diverse observations

The two error metrics that define fusion center performance. The false alarm rate (P_fa) measures wasted transmission opportunities when a vacant channel is declared occupied. The missed detection probability (P_md) measures harmful interference events when an occupied channel is declared vacant.

• Receiver Operating Characteristic (ROC) curves visualize this tradeoff
• Regulatory bodies mandate maximum P_md thresholds (often < 0.1)
• Fusion rules are selected to optimize one metric under a constraint on the other