Shadow Fading Map

Shadow fading is modeled as a zero-mean Gaussian random variable when expressed in decibels. The standard deviation (σ) typically ranges from 4 to 12 dB depending on terrain clutter density—urban environments exhibit higher variance than rural areas. This log-normal behavior arises from the multiplicative effect of multiple random obstructions along the propagation path, which becomes additive in the logarithmic domain via the central limit theorem.

Unlike fast fading, shadow fading exhibits strong spatial correlation over tens to hundreds of meters. The autocorrelation is typically modeled using an exponential decay function with a decorrelation distance (d_corr) parameter:

• Urban microcells: 10-50 meters
• Suburban macrocells: 50-100 meters
• Rural macrocells: 100-500 meters This correlation means two receivers in close proximity experience similar shadowing, a critical property for Kriging interpolation in REM construction.

A complete large-scale propagation model decomposes received power into three additive dB components:

1. Distance-dependent path loss: Deterministic function of log-distance
2. Shadow fading: Zero-mean Gaussian random variable N(0, σ²)
3. Building penetration loss: Additional attenuation from indoor environments The Shadow Fading Map isolates component #2, enabling network planners to compute cell edge reliability—the probability that a user at the cell boundary exceeds the minimum sensitivity threshold.

The classic Gudmundson correlation model extends Jakes' framework to shadow fading, describing the spatial autocorrelation as:

• R(Δx) = σ² · exp(-|Δx| / d_corr) Where Δx is the spatial separation between two points. This exponential model is widely adopted in 3GPP channel models (TR 38.901) and enables the generation of realistic 2D shadow fading maps using Cholesky decomposition of the covariance matrix for Monte Carlo simulations.

Shadow fading maps are not purely statistical—they can be deterministically enhanced by incorporating terrain and clutter data:

• Diffraction loss from terrain ridges computed via knife-edge models
• Clutter classes (dense urban, urban, suburban, rural) mapped to σ values
• Building footprint databases for ray-tracing validation This hybrid approach combines the computational efficiency of statistical models with the accuracy of site-specific propagation prediction, forming a critical layer in a Radio Environment Map (REM).

The primary operational use of a shadow fading map is computing cell coverage probability:

• P_coverage = Q((P_min - P_rx(d)) / σ) Where Q() is the Gaussian Q-function, P_min is the receiver sensitivity, and P_rx(d) is the median received power at distance d. For a 90% cell edge reliability with σ = 8 dB, a fade margin of approximately 10.3 dB must be added to the link budget. The map visualizes this margin spatially across the service area.

Path loss is the deterministic, distance-dependent decay of signal power, typically modeled by a log-distance equation. Shadow fading is the stochastic, large-scale variation superimposed on this trend, caused by macroscopic obstructions like buildings or hills. While path loss predicts the mean signal at a distance, shadow fading captures the log-normal deviation around that mean, typically with a standard deviation of 6–12 dB in urban environments. Separating these two components is critical for accurate coverage prediction in a REM.

Shadow fading is universally modeled as a log-normal random process because signal power in dB is the sum of many independent multiplicative attenuation factors, invoking the central limit theorem. Key properties:

• Mean: 0 dB (unbiased around path loss estimate)
• Standard deviation (σ): 4–13 dB depending on terrain and frequency
• Decorrelation distance: 10–100 meters in urban microcells, defining the spatial granularity of the map This distribution enables probabilistic coverage metrics, such as the likelihood that a receiver exceeds a sensitivity threshold at a given location.

Shadow fading is not spatially white; nearby locations experience correlated shadowing due to shared obstructions. This is modeled using an exponential autocorrelation function: R(Δd) = σ² · exp(-Δd/d_corr), where d_corr is the decorrelation distance. In REM construction, this correlation structure is essential for Kriging interpolation—it determines how much weight a nearby sensor measurement receives when estimating shadowing at an unobserved grid point. Ignoring correlation leads to overconfident, speckled maps.

Constructing a shadow fading map involves:

1. Measurement collection: Drive tests or distributed sensor RSSI readings
2. Path loss removal: Subtract the known distance-dependent loss to isolate the residual shadowing component
3. Spatial interpolation: Apply Kriging or Gaussian Process Regression using the estimated variogram to predict shadowing at all grid points
4. Validation: Compare predicted vs. held-out measurements to tune the variogram model The resulting layer is a smooth, continuous surface representing large-scale obstruction effects across the coverage area.

In a complete REM, shadow fading is combined with other loss components to produce a total channel gain map:

• Path loss: Distance and frequency dependent (deterministic)
• Shadow fading: Large-scale log-normal variation (stochastic, spatially correlated)
• Small-scale fading: Rayleigh or Rician multipath effects (fast variation, often averaged out in REM layers) This composite model enables network simulators to generate realistic coverage probability maps and identify outage hotspots before deployment.

The 3rd Generation Partnership Project (3GPP) specifies shadow fading parameters for standardized network simulations in TR 38.901. Key values:

• Urban Macro (UMa): σ = 4 dB (LOS), 6 dB (NLOS)
• Urban Micro (UMi): σ = 3 dB (LOS), 7.9 dB (NLOS)
• Indoor Hotspot (InH): σ = 3 dB (LOS), 8.03 dB (NLOS) These standardized models ensure that REMs built for cellular planning align with industry-accepted propagation assumptions, enabling consistent coverage benchmarking across vendors.