Spectrum Cartography

A geostatistical method that predicts unknown RF signal values at unmeasured locations by computing a weighted average of known neighboring measurements. Unlike simple distance-based interpolation, Kriging weights are derived from a modeled variogram that quantifies the spatial autocorrelation of the signal. This provides the Best Linear Unbiased Predictor (BLUP) for spatial data, making it the gold standard for constructing Radio Environment Maps from sparse sensor networks.

A non-parametric Bayesian machine learning method that models the spectrum as a distribution over functions. For every spatial coordinate, GPR outputs both a predicted mean power spectral density and a quantified uncertainty variance. This inherent uncertainty quantification is critical for risk-aware dynamic spectrum access, allowing a cognitive radio to assess the confidence of a spectrum opportunity before transmitting. GPR is defined by its kernel function, which encodes assumptions about signal smoothness.

A computational technique that estimates missing spectrum data by leveraging correlations across two dimensions simultaneously: spatial (correlation between nearby sensors at the same time) and temporal (correlation of historical measurements at a single sensor). This is essential for tracking moving interferers or adapting to dynamic traffic patterns. Methods often combine Kalman filters for temporal dynamics with Kriging or Graph Neural Networks for spatial smoothing.

A signal processing technique that enables the reconstruction of a wideband spectrum map from sub-Nyquist rate samples. It exploits the inherent sparsity of spectrum occupancy—most frequencies are idle at any given moment. By solving an L1-minimization optimization problem, compressed sensing recovers the full spectral map from far fewer measurements than traditional Shannon-Nyquist sampling requires, dramatically reducing sensor hardware costs and data throughput.

A deep learning architecture that models the spatial relationships between distributed sensors as a graph, where nodes represent sensors and edges represent communication links or spatial proximity. GNNs perform interpolation by iteratively passing messages between connected nodes, learning complex non-linear propagation patterns directly from data. This approach outperforms classical Kriging in highly cluttered urban environments where propagation defies simple statistical models.

A core geostatistical function that quantifies the spatial autocorrelation of RF measurements as a function of distance. The empirical variogram plots the average squared difference between measurement pairs against their separation distance. A theoretical model (e.g., spherical, exponential, or Matérn) is then fitted to this cloud. This variogram model is the foundational input for Kriging, dictating how quickly signal correlation decays over space.

The primary geostatistical estimator for spatial spectrum prediction. Kriging computes a weighted average of known sensor measurements, where weights are derived from a variogram model that quantifies spatial correlation decay with distance.

• Produces a Best Linear Unbiased Estimator (BLUE)
• Minimizes estimation variance at every grid point
• Requires modeling of the empirical variogram before interpolation
• Ordinary Kriging assumes an unknown but constant local mean

A non-parametric Bayesian alternative to Kriging that defines a prior distribution over functions and updates it with sensor observations. Critically, GPR provides both a predicted mean and a variance estimate at every spatial coordinate.

• The variance map serves as a confidence interval for REM data
• Kernel function selection (e.g., Matérn, RBF) encodes smoothness assumptions
• Computationally intensive: scales as O(N³) with sensor count
• Enables active sensing by identifying high-uncertainty regions

Mathematical prediction of path loss between transmitters and receivers, accounting for distance, frequency, terrain, and clutter. Propagation models translate sparse sensor measurements into physically consistent spatial estimates.

• Empirical models (Longley-Rice, Hata) use measurement-driven formulas
• Deterministic models (ray tracing) simulate individual multipath components
• Semi-empirical models combine statistical fits with terrain diffraction calculations
• Essential for extrapolating beyond sensor coverage areas

The regulatory-mandated three-tier spectrum coordination system for the 3.5 GHz CBRS band. A SAS maintains a geolocation database of incumbent users and uses REM data to dynamically authorize secondary transmissions.

• Tier 1: Incumbent federal radar and satellite ground stations
• Tier 2: Priority Access License holders with interference protection
• Tier 3: General Authorized Access users on an opportunistic basis
• Environmental Sensing Capability (ESC) sensors detect incumbent radar activity

A continuously synchronized virtual replica of the physical electromagnetic environment. Unlike static REMs, a digital twin ingests real-time sensor feeds and updates propagation predictions to enable what-if scenario simulation.

• Integrates 3D city models with material properties for ray tracing
• Supports spectrum policy testing before physical deployment
• Enables predictive interference analysis for network planning
• Used in defense for electromagnetic battle management

A signal processing breakthrough enabling wideband spectrum reconstruction from sub-Nyquist samples. Exploits the inherent sparsity of spectrum occupancy—most frequencies are idle at any moment.

• Reduces the sampling rate required for wideband cartography
• Uses ℓ₁-norm minimization to recover the sparse frequency vector
• Enables cheaper, lower-power spectrum sensors
• Pairs with random demodulator or multi-coset sampling hardware