Multi-Band Spectrum Prediction Method via Singular Spectrum Analysis and A-BiLSTM

In wireless communication, the demand for communication services is increasing while the spectrum resources remain limited. This leads to a shortage of spectrum resources, which in turn affects the quality of the electromagnetic environment. To make better use of the spectrum resources, it is necessary to analyze and forecast the spectrum. Spectral prediction is a technique for inferring unknown spectral data from a sample of known spectral data. However, spectral prediction has always been a challenging problem due to the complexity and uncertainty of spectral data. In order to improve prediction performance, this paper studies the bidirectional long and short time memory (BiLSTM) network prediction method based on singular spectrum analysis (SSA) and attention mechanism. The method uses an SSA locus matrix to construct the original time series and extract signals representing different components of the time series. Finally, the BiLSTM model with attentional mechanism (A-BiLSTM) is used to predict the spectral data and obtain the prediction results. Experimental results show that this method can further improve prediction accuracy and model performance.