Consensus-Based Sensing

Each node initializes its state with a local measurement or log-likelihood ratio and iteratively updates it by computing a weighted average of its own state and the states received from immediate neighbors. Over successive iterations, all node states converge to the global average of the initial values, enabling a fully decentralized computation of the optimal test statistic without any node having global knowledge of the network topology.

Nodes employ asynchronous gossip algorithms where, in each time slot, a randomly selected node wakes up, contacts a randomly chosen neighbor, and they exchange and average their current state values. This pairwise communication model is robust to dynamic topologies and node failures, as the global averaging process does not depend on a fixed routing structure or synchronized rounds, making it ideal for mobile ad-hoc cognitive radio networks.

The convergence properties of the consensus algorithm are governed by the weight matrix used in the averaging step. The Metropolis-Hastings method constructs a doubly-stochastic weight matrix using only local degree information, where the weight on edge (i,j) is set to 1/(1+max(d_i, d_j)). This ensures fast, guaranteed convergence even on irregular network graphs without requiring any node to know the global topology or perform centralized matrix factorization.

In adversarial environments where malicious nodes inject false data to corrupt the consensus, robust consensus variants replace the weighted averaging step with trimmed or median-based aggregation rules. By discarding extreme values at each iteration, the network can converge to a value within the convex hull of the honest nodes' initial states, providing resilience against Spectrum Sensing Data Falsification attacks without requiring a centralized reputation manager.

Nodes must autonomously determine when the consensus process has sufficiently converged to make a local decision. Common stopping criteria include monitoring the maximum deviation between a node's current state and its neighbors' states over a sliding window, or using the Wasserstein distance between local belief distributions. Once the deviation drops below a pre-defined threshold ε, each node independently applies the same detection threshold to its converged test statistic, yielding a unanimous global decision without explicit coordination.

To operate within the severe bandwidth constraints of cognitive radio control channels, nodes transmit quantized versions of their state values rather than continuous real numbers. Probabilistic quantization schemes, where a node transmits a rounded integer representation of its state while retaining the quantization error for the next iteration, preserve the convergence properties of the unquantized algorithm while reducing per-message payload to just a few bits, enabling practical deployment in low-data-rate reporting channels.

The foundational peer-to-peer communication protocol enabling consensus-based sensing. Nodes iteratively select a random neighbor and exchange state variables (e.g., local test statistics). Through repeated pairwise averaging, the network converges to a global average without any central coordination. Variants include broadcast gossip and push-sum protocols, each offering different tradeoffs between convergence speed and communication overhead on resource-constrained cognitive radios.

A specific consensus objective where all nodes in the network iteratively compute the global arithmetic mean of their initial local measurements. In spectrum sensing, nodes average their received signal energy levels. The final converged value serves as a distributed test statistic, which each node independently compares against a local threshold to make a global decision. This eliminates the single point of failure inherent in fusion center architectures.

The mathematical framework underpinning consensus-based sensing, extending classical Neyman-Pearson and Bayesian detection to multi-node systems with constrained inter-node communication. Unlike parallel fusion architectures, distributed detection with iterative information exchange requires analyzing the asymptotic convergence of local likelihood ratios. Key challenges include maintaining constant false alarm rates as the network topology changes.

The study of how quickly a consensus algorithm reaches a stable global decision, governed by the algebraic connectivity (second-smallest eigenvalue of the graph Laplacian) of the network topology. Faster convergence reduces sensing latency but typically requires more frequent message exchanges. Metropolis-Hastings weights are often used to optimize the convergence rate in networks where nodes have heterogeneous degrees without requiring global topology knowledge.

The resilience of the consensus process against Spectrum Sensing Data Falsification (SSDF) attacks where malicious nodes inject false state updates. Robust consensus algorithms use outlier rejection techniques, such as median-based consensus or trimmed averaging, to isolate Byzantine neighbors. Unlike reputation-based fusion center approaches, these methods provide inherent security by preventing a single compromised node from dominating the global decision.

Mechanisms for maintaining consensus performance when nodes join, leave, or move within the network. Push-sum consensus protocols handle dynamic topologies by tracking a network mass variable that rebalances as the node count changes. This is critical for mobile ad-hoc cognitive radio networks where the communication graph is constantly evolving due to node mobility and changing channel conditions.