Hierarchical Cooperative LSTM-Based Spectrum Sensing

In this letter, we design a hierarchical cooperative long short-term memory (LSTM) network-based cooperative spectrum sensing (CSS) method which utilizes convolutional neural network (CNN) and LSTM network. The CNN extracts spatial features from the input covariance matrices (CMs) which are generated by sensing data of each secondary user (SU) and the sequence of spatial features corresponding to multiple sensing periods are fed into secondary user LSTM (SU-LSTM) so that the PU activity pattern at SU level can be learned. The cooperative LSTM learns the group-level PU activity pattern from all SU-level temporal feature representations. The aim of learning the PU activity pattern at SU-level and group-level is to improve the detection performance further. To demonstrate the robustness of the proposed model, the scenario of an imperfect reporting channel is taken into account. With a sufficient amount of simulations, the effectiveness of the proposed method is proven and simulation results demonstrate that the proposed method outperforms the state-of-the-art in terms of detection probability and classification accuracy.

Automated summary

In this letter, a hierarchical cooperative LSTM network-based CSS method is proposed, which utilizes CNN and LSTM network. The CNN extracts spatial features from input CMs generated by sensing data of each SU, and the sequence of spatial features is fed into SU-LSTM to learn the PU activity pattern at SU level. The cooperative LSTM learns the group-level PU activity pattern from all SU-level temporal feature representations. The effectiveness of the proposed method is proven through sufficient simulations, outperforming the state-of-the-art in terms of detection probability and classification accuracy.