Deep Learning-Based Spectrum Sensing in Cognitive Radio: A CNN-LSTM Approach

Abstract

For most existing spectrum sensing detectors, the design of their test statistics relies on certain signal-noise model assumptions and hence, their detection performance heavily depends on the accuracy of the assumed models. Therefore, recently, much attention in the research of spectrum sensing is focused on deep learning which is free from model assumptions. Note that, in deep learning, the convolutional neural networks (CNNs) and the long-short term memory (LSTM) networks have the powerful capabilities in extracting spatial and temporal features of the input, respectively. In this letter, we propose a CNN-LSTM detector which first uses the CNN to extract the energy-correlation features from the covariance matrices generated by the sensing data, then the series of energy-correlation features corresponding to multiple sensing periods are input into the LSTM so that the PU activity pattern can be learned. The purpose of learning PU activity pattern is to further promote the detection probability. With sufficient simulations, the superiority of the CNN-LSTM detector is proven in scenarios with and without noise uncertainty.