Spectrum sensing based on graph mixture statistic

Abstract

Graph transformation-based spectrum sensing algorithms have demonstrated effectiveness in enhancing detection performance with limited samples. However, existing methods primarily rely on topological features derived from a single graph constructed using either the power spectrum (PS) or the autocorrelation function (ACF). This approach overlooks the potential benefits of integrating information from multiple graphs. Moreover, these methods often fail to effectively combine topological features with the statistical properties of the original signals, resulting in reduced detection performance in low signal-to-noise ratio (SNR) environments and fading channels. To address these limitations, we propose a novel spectrum sensing algorithm based on graph mixture graph statistic. Our method employs quantization-based graph transformation to generate two distinct subgraphs from the block sums of the PS (BSPS) and the ACF. We define the vertex probability vector of the BSPS and the mean and standard deviation sum of the ACF quantization subsets as the respective graph signals. By calculating the Laplacian quadratic form for the BSPS-based graph and the one-hop graph filter statistic for the ACF-based graph, we derive a graph mixture statistic that serves as a new detection criterion. Simulations demonstrate that our proposed approach outperforms existing methods with only a modest increase in computational complexity.