Spectrum sensing for cognitive radio based on convolution neural network

TLDR: Experimental results show that a reasonable CNN model is built and the proposed algorithm has higher detection probability than cyclostationary feature detection (CFD) about 0.5 in −20dB.

Abstract: The problem in the process of spectrum sensing that the detection rate of the the primary user (PU) signal is low in the environment of low signal-to-noise (SNR) is present, a novel spectrum sensing algorithm based on convolution neural network (CNN) is proposed. The CNN is widely used in image recognition and speech recognition, and has good classification performance. Therefore, the CNN is employed to solve spectrum sensing which can be viewed as a binary hypothesis-testing problem. Firstly, the feature of the presence of the PU signal and the presence of only the noise signal are extracted, including cyclostationary feature and energy feature. And then, the extracted features should be pre-processed, which are used as the training input of the CNN model. Finally, the test data is fed into the trained CNN model, which is aiming to detect the presence of the PU. Experiment results show that a reasonable CNN model is built and the proposed algorithm has higher detection probability than cyclostationary feature detection (CFD) about 0.5 in −20dB.