Research on Spectrum Sensing Algorithm Based on Long Short-Term Memory Network and Goodness of Fit of Distribution of Signal Energy

The performance of traditional spectrum sensing is not ideal at low signal to noise ratio(SNR) and good at high SNR. As the SNR decreases, the performance of algorithms gradually deteriorates. This paper proposes a spectrum sensing algorithm based on long short-term memory(LSTM) network and the goodness of fit of the distribution of signal energy. This method calculates the distance value form the energy distribution of the received signal when primary user signals exist. Then the feature vector consists of the distance value is input into a LSTM network training to obtain a model. Finally, test data is input into the model for prediction to achieve spectrum sensing. Results show that the new method proposed in this paper has a detection probability of 96.21% when the SNR is -13dB and the number of sampling points is 28, which is significantly better than the traditional energy detection algorithm and the traditional goodness-of-fit algorithm.