DEEP LEARNING BASED SPECTRUM SENSING TECHNIQUE FOR SMARTER COGNITIVE RADIO NETWORKS

Abstract

 Accurate spectrum sensing is crucial for avoiding interference and maximizing spectrum band utilization or spectral efficiency in a cognitive radio network (CRN).However, conventional spectrum sensing techniques such as energy detection (ED) and cyclostationary feature detection (CFD) suffer limitations such as unreliable sensing and the problem of finding optimal detection threshold. In this paper, deep learning based detection (DLbD) technique is proposed to overcome these limitations for non-cooperative CRN under low SNR scenario .The proposed technique uses the Long Short Term Memory (LSTM) model of deep learning to learn the features of a modulated received signal in order to accurately distinguish between such signal and noise within a spectrum band. An LSTM classifier was trained using some generated signals of different modulation schemes and noise as features. The proposed DLbD technique was compared with the ED and CFD techniques using probability of detection and probability of missing as performance metrics. The simulation results reveal that the proposed DLbD technique outperforms both the ED and CFD techniques. The proposed DLbD technique is essential in fifth-generation (5G) mobile telecommunications ultra-dense networks like smart cities.

KEYWORDS: Spectrum sensing (SS), cognitive radio network (CRN),deep learning (DL), long short term memory (LSTM), spectral efficiency, smart network.