Modeling Local Demand for Mobile Spectrum using Large Crowdsourced Datasets

With the deployment and expansion of 5G networks underway in many countries, the demand for mobile spectrum continues to grow, particularly in frequency bands below 6 GHz. The emergence of 6G networks will also further amplify current challenges associated with spectrum scarcity. This paper discusses the need to model mobile spectrum demand - defined in terms of the demand for mobile services - at the local level. More accurate spectrum demand modeling can help regulatory bodies to better plan spectrum allocations and to make more informed spectrum policy decisions to support future technological developments. While existing research typically estimates demand using simplistic models based on factors such as network capacity, spectral efficiency, and population-based proxies, this work proposes a data-driven approach to estimate mobile spectrum demand using machine learning. We also derive a more accurate proxy for demand using a large dataset of over 2.5 billion crowdsourced commercial mobile measurements. The data-driven nature of this proxy eliminates the need for various theoretical assumptions associated with current demand proxies. Finally, we employ the SHapley Additive exPlanations (SHAP) method for global model interpretation to demonstrate that, contrary to intuition, population is not the sole contributing factor of demand. Instead, a diverse set of real-world factors can influence demand patterns and, therefore, should be used to create more accurate demand models.