A Spectrum Sensing Algorithm Based on LSTM Neural Network

The spectrum sensing plays very important role in Cognitive Radio system. The classical energy detection (ED) algorithm has low detection performance at low Signal-to-Noise Ratio (SNR) and difficulty in the estimation of the threshold required by ED. However, the algorithm based on machine learning is limited by the construction of detecting statistic which will cause the loss of the original structural information of the received signal. Aiming at these problems, this paper proposes a LSTM neural network(NN) based spectrum sensing method. Firstly, the sampling sequence of the receiver signal is used as the input of the NN, and then the LSTM neural network is employed to train a classifier. Finally, the classifier for spectrum sensing is followed. This method has no need of either the detection thresholds or the construction of the feature vector, and the simulation results show that the proposed algorithm is superior to the comparison algorithm in the case of fewer sampling points and second users.