Soft Fusion based Cooperative Spectrum Prediction using LSTM
Niranjana Radhakrishnan, Sithamparanathan Kandeepan, Xinghuo Yu, Gianmarco Baldini · 2021
Spectrum prediction is an important solution proposed to efficiently manage the scarce spectrum resource in various Dynamic Spectrum Access (DSA) applications. Deep Learning based models such as Long Short Term Memory (LSTM) have been increasingly applied to perform temporal and multi-dimensional prediction of future spectrum characteristics. These models have shown excellent capabilities to learn the correlations in the historical spectrum observations and make next-step predictions. Moreover, to achieve better accuracy, cooperative spectrum prediction using multiple local predictors is known to be a promising technique compared to a single local predictor. Cooperative prediction can also potentially lead to increased spectrum utilization efficiency and energy efficiency. Therefore, in this work, we study different soft fusion and hard fusion methods to perform cooperative spectrum prediction of spectrum occupancy in a cognitive radio environment with trained LSTM-based local predictors. The proposed methods indicate a reduction in the prediction error compared to local prediction and most hard fusion methods.