Scaling DCN Models for Indoor White Space Exploration

With the fast growth of the wireless spectrum demand, people have been focusing on utilizing indoor white spaces. In the past few years, several indoor white space exploration methods have been proposed. These methods focus on the utilization of spatial and spectral correlations of white spaces. However, these correlations change over time. In this paper, we perform indoor white space synchronous measurement to demonstrate the volatility of white spaces. Then, we propose a DCN (Deep Convolutional Network)-based method to capture the statistical dependencies among the features and combinatorial features extracted from white spaces, which are not limited to spatial or spectral features, and construct the white space availability map. After demonstrating the instability of spectral and spatial correlations, we scale our DCN models to a time-agnostic model. We conduct real-world experiments to evaluate our system. The evaluation results show that our time-specific DCN model and time-agnostic model outperforms the state-of-the-art method.