Deep Reinforcement Learning based reliable spectrum sensing under SSDF attacks in Cognitive Radio networks

Abstract

In Spectrum Sensing Data Falsification (SSDF) attacks, malicious Secondary Users (SUs) actively send erroneous local sensing results to the Fusion Centre (FC) that influence the actual outcomes of Cooperative Spectrum Sensing (CSS). Existing trust value-based algorithms are partially successful as SUs can quickly change their characteristics from honest to malicious and vice versa. The present work explores the Deep Reinforcement Learning algorithm (DRL) for adapting the behavioural changes of SUs during CSS to overcome the issue mentioned earlier. The agent in the DRL algorithm wisely avoids the malicious data from the received sensed-energy values at FC and reduces the sensing error effectively. Simulation results show that the proposed work outperforms the conventional and Support Vector Machine-based approaches in CSS reliabilities under SSDF attacks.