Transfer Learning for Tilt-Dependent Radio Map Prediction

Machine learning will play a major role in handling the complexity of future mobile wireless networks by improving network management and orchestration capabilities. Due to the large number of parameters that can be monitored and configured in the network, collecting and processing high volumes of data is often unfeasible or too expensive at network runtime, which calls for taking resource management and service orchestration decisions with only a partial view of the network status. Motivated by this fact, this paper proposes a transfer learning framework for reconstructing the radio map corresponding to a target antenna tilt configuration by transferring the knowledge acquired from another tilt configuration of the same antenna, when no or very limited measurements are available from the target. The performance of the framework is validated against standard machine learning techniques on a data set collected from a 4G commercial base stations. In most of the tested scenarios, the proposed framework achieves notable prediction accuracy with respect to classical machine learning approaches, with a mean absolute percentage error below 8%.