Primary user detection in cognitive radio using spectral-correlation features and stacked denoising autoencoders based on signal classification

摘要

For the solution to settle problems of either spectrum scarcity as well as spectrum allocation, which would be severely aggravated due to the hunger of bandwidth and richness of feature of wire applications, the cognitive radio (CR) system has been proposed to fit into those circumstances. This paper provides a novel primary user (PU) presence detection approach in CR system on account of applying the algorithm of stacked denoising autoencoders (SDAE) network when analyzing and processing the cyclic spectral correlation features. With regard to the non-known bandwidths nor carrier frequency, conventional cyclic spectral analysis can be the answer to application of detection or classifying signals, however, serious time of observation make a depravation in approving performance. At this point, the stacked denoising autoencoders network, which is freshly employed over the signal classification scenario, could help pick up the feature of input analyzed spectral correlation automatically after appropriate preprocessing and give out the PU detection results, which contain existence or non-existence as two classified categories. In addition to that, in contrast to the conventional methods for CR system, the presented results demonstrate our approach's superiority for only a short allowed detection executing time.