Correlated Shadowing

Shadowing correlation describes the statistical dependence between the slow-varying signal power measured at two or more sensing nodes. Unlike fast multipath fading, shadowing is caused by macroscopic obstructions—buildings, terrain, foliage—that create log-normal signal variations over tens to hundreds of wavelengths. When nodes are separated by less than the decorrelation distance, they experience highly similar shadowing realizations.

• Log-normal model: Signal power in dB follows a Gaussian distribution with a standard deviation typically 4–12 dB
• Decorrelation distance: Ranges from 10–100 meters in urban microcells to 500+ meters in rural macro-cells
• Exponential correlation model: The most common mathematical representation, where correlation decays as ρ(d) = exp(-d/d_corr) with distance d

The fundamental premise of cooperative spectrum sensing is that spatially distributed nodes experience independent fading, so the probability of all nodes simultaneously experiencing a deep fade is low. Correlated shadowing erodes this spatial diversity by introducing a common fading component.

• Degraded detection probability: When shadowing is positively correlated, multiple nodes may simultaneously miss a primary user signal, defeating the purpose of cooperation
• False alarm clustering: Nodes under a common shadow may collectively misclassify noise as a signal
• Diminishing returns: Adding more correlated nodes yields marginal improvement beyond a certain density threshold
• Mitigation strategy: Node selection algorithms that choose a subset of sensors with minimum inter-node correlation

Correlated shadowing arises from common propagation paths and shared physical obstructions in the environment. Understanding these mechanisms is essential for accurate channel modeling and sensor placement optimization.

• Common obstacle shadowing: Two nodes behind the same building or hill experience nearly identical attenuation
• Angle-of-arrival similarity: Signals arriving from similar azimuth angles traverse correlated scattering environments
• Site-specific factors: Urban canyons, indoor wall layouts, and vegetation density create deterministic correlation patterns
• Distance-dependent decay: Correlation decreases monotonically with node separation, but the rate depends heavily on the propagation environment type

Accurate modeling of shadowing correlation is critical for realistic simulation of cooperative sensing networks. Several mathematical frameworks capture this phenomenon with varying complexity.

• Gudmundson model: A 2D autocorrelation model where shadowing is a stationary Gaussian process with an exponential autocorrelation function
• Joint log-normal distribution: Models the shadowing at N nodes as a multivariate Gaussian with a covariance matrix parameterized by inter-node distances
• Measurement-based models: Site-specific ray-tracing or empirical campaigns that build correlation maps for deployment planning
• 3GPP spatial channel models: Standardized models incorporating both intra-site and inter-site shadowing correlation for cellular network evaluation

When shadowing correlation cannot be avoided, intelligent fusion strategies can partially recover lost diversity by accounting for the correlation structure in the decision process.

• Correlation-aware node selection: Greedy or optimization-based algorithms that select a sensing subset with a covariance matrix below a condition number threshold
• Mahalanobis distance weighting: Assigning fusion weights inversely proportional to the statistical dependence between nodes
• Cluster-based architectures: Grouping highly correlated nodes into clusters, fusing within clusters first, then treating clusters as independent sensors
• Adaptive threshold adjustment: Raising or lowering detection thresholds based on estimated instantaneous correlation to maintain a target false alarm rate

It is critical to distinguish correlated shadowing from small-scale multipath fading, as they operate on fundamentally different spatial and temporal scales and require different countermeasures.

• Shadowing: Large-scale, log-normal, correlated over 10–500 m, caused by macroscopic obstructions
• Multipath fading: Small-scale, Rayleigh/Rician, decorrelated over λ/2 (centimeters at GHz frequencies), caused by constructive/destructive interference
• Combined effect: Real channels exhibit superimposed shadowing and multipath fading, with shadowing determining the local mean power around which fast fading fluctuates
• Sensing implication: Cooperative sensing primarily combats multipath fading through micro-diversity; correlated shadowing requires macro-diversity with much larger node separations

The core benefit of cooperative sensing that correlated shadowing directly degrades. It relies on geographically separated nodes experiencing independent fading realizations. When shadowing is correlated, the signal-to-noise ratios (SNRs) at multiple nodes fade simultaneously, eliminating the diversity branch that a fusion center relies on to make a robust global decision. The diversity gain is maximized only when the inter-node distance exceeds the decorrelation distance.

The physical separation required for the shadow fading experienced by two nodes to become statistically uncorrelated. This parameter is highly dependent on the propagation environment:

• Urban microcells: Typically 10–50 meters due to dense clutter.
• Suburban macro cells: Often 50–100 meters.
• Indoor environments: Can be as low as a few meters due to walls. Network planners must ensure sensor separation exceeds this distance to avoid correlated measurements.

The standard statistical model where large-scale path loss variation follows a log-normal distribution (dB values are Gaussian). The correlation is often modeled using an exponential decay function: ρ(d) = exp(-a * d) where ρ is the correlation coefficient, d is the distance between nodes, and a is an environment-specific decay constant. This model is critical for simulating realistic cooperative sensing performance in tools like MATLAB or ns-3.

The mechanism by which correlated shadowing causes global decision errors at the fusion center. In a K-out-of-N rule, correlation causes multiple nodes to simultaneously miss a primary user (a missed detection) or falsely trigger (a false alarm). This breaks the assumption of independent observations, invalidating the theoretical performance bounds of the Likelihood Ratio Test (LRT) and making the network more vulnerable to the hidden node problem.

A practical mitigation strategy to combat correlated shadowing. Instead of using all available nodes, the fusion center or a cluster head selects a subset of sensors with low mutual correlation. Techniques include:

• Geographic separation: Selecting nodes spaced beyond the decorrelation distance.
• Cluster-based CSS: Grouping correlated nodes and treating each cluster as a single uncorrelated reporting entity. This optimizes the trade-off between spatial diversity and network overhead.

A blind sensing approach that can be adapted to account for correlation. Instead of treating correlation as a nuisance, the spatial covariance matrix of the received signals is estimated. The ratio of the maximum to minimum eigenvalues of this matrix is used for detection. This method is robust to noise uncertainty and, when the correlation structure is known or can be learned, can actually exploit the correlation to distinguish between a correlated primary user signal and uncorrelated noise.