Abstract:
Significant progress has been made in developing optimal frameworks for spatial prediction in radio propagation environments, as they enhance various aspects of wireless networking. These predictions rely on a limited set of measurements from an active transmitter whose radio envi- ronment is to be predicted. This work draws inspiration from the paper titled "ProSpire: Proactive Spatial Prediction of Radio Environment Using Deep Learning," which introduces a supervised deep learning-based framework to enable spectrum sharing through proactive spatial prediction. Since deep learning models are data-intensive and require large datasets, this project explores novel data augmentation techniques beyond those proposed in the framework, aiming to reduce the overall mean absolute error of the trained model. To achieve these results, various methods were experimented with, including:
a. Usage of K-means for optimal data segmentation,
b. Developed a new Dynamic Probability Allocation Heuristic approach
c. Transformed Gaussian Data to uniform Distribution
