NS-3-based Real-Time Spatiotemporal Spectrum Prediction Dataset

The real-time spatiotemporal spectrum prediction dataset is collected by setting up a 5G downlink scenario with 4 static base stations and 10 mobile users uniformly distributed throughout a 100m \u00d7 100m region of interest along with randomly spatially distributed sensors. The sensor measurements are organized into a 2D PSD matrix of size N_nodes \u00d7 N_t. The N_t samples are partitioned into non-overlapping sensing windows of length N_w = N_s + N_p, with each window forming an input\u2013output pair of size N_s and N_p, respectively, for each sensing node. Total data becomes N_data = N_t\/N_w samples where input and output matrix have sizes of N_nodes \u00d7 N_s and N_nodes \u00d7 N_p, respectively.  This dataset is split into training and testing subsets with ratio \u03b1 and 1 \u2212 \u03b1, respectively, where \u03b1 = 0.8 in our evaluation, resulting in \u03b1N_data training samples and (1 \u2212 \u03b1)N_data testing samples.