The growing demand for wireless connectivity, combined with limited spectrum resources, calls for more efficient spectrum management. Spectrum sharing is a promising approach; however, regulators need accurate methods to characterize demand dynamics and guide allocation decisions. This paper builds and validates a spectrum demand proxy from public deployment records and uses a graph attention network in a hierarchical, multi-resolution setup (HR-GAT) to estimate spectrum demand at fine spatial scales. The model captures both neighborhood effects and cross-scale patterns, reducing spatial autocorrelation and improving generalization. Evaluated across five Canadian cities and against eight competitive baselines, HR-GAT reduces median RMSE by roughly 21% relative to the best alternative and lowers residual spatial bias. The resulting demand maps are regulator-accessible and support spectrum sharing and spectrum allocation in wireless networks.
Figures & tables
Figure 1: Overview of the proposed methodology.
Figure 2: Geospatial processing for proxy validation: (a) LTE cell locations and estimated coverage areas; (b) aggregated busy-hour throughput per grid tile.
Figure 3: Deployed bandwidth heatmaps at zoom level 15 for Calgary, Montreal, GTA, Vancouver, and Ottawa.
Figure 4: Feature processing pipeline for geospatial demand modeling.
Feature Category
Description (per grid tile)
Population & Demographics
Residents by age, income, and worker type
Daytime Population
Workers/visitors present during working hours (non-residents included)
Households
Number of dwelling units
Mobility (trip distance)
Counts/shares traveling in distance bands (e.g., 0–3, 3–7, 7–10, 10–15 km)
Economic Activity
Business counts by sector (e.g., NAICS) and estimated employees
Built Environment (Buildings)
Building count, total built-up area, and building density
Table 1: Summary of features processed for demand modeling
Figure 5: Visualization of HR-GAT. (a) Hierarchical graph construction across multiple zoom levels. (b) Model architecture for spectrum demand estimation.
Figure 6: Moran’s I analysis of spatial autocorrelation across different cities.
Proxy Variable
R²
F-Statistic
p-Value
Deployed Bandwidth ( BWg )
0.727
4477
<0.001
Table 2: ols regression results for proxy validation
Model
Median MAE ↓
Median RMSE ↓
R² ↑
Gradient Boosting
23.44
44.27
0.86
XGBoost
23.53
44.72
0.86
LightGBM
30.78
55.70
0.80
Decision Tree
41.60
79.92
0.53
Random Forest
45.89
77.48
0.55
Linear Model
55.50
89.31
0.43
Table 3: performance comparison of hr-gat vs. alternative ml models
Figure 7: Scatter plot of actual vs. predicted spectrum demand values for HR-GAT.
Figure 8: Empirical cumulative distribution function (eCDF) of RMSE values across all models.
Figure 9: Log-scaled residual error distribution across cities.
Model
Median MAE (Ottawa) ↓
Gradient Boosting
37.34
XGBoost
39.49
LightGBM
47.20
Decision Tree
52.47
Random Forest
50.66
Linear Model
60.93
Table 4: median mae comparison for ottawa as the test set
Figure 10: SHAP summary plot illustrating the relative importance of the top 10 features in predicting mobile spectrum demand. Higher absolute SHAP values indicate stronger influence on the model estimates.