Research on a radio-frequency spectrum prediction algorithm for unmanned aerial vehicle communication
Xin Dai and Mian Huang
With the widespread use of unmanned aerial vehicles (UAVs), spectrum resources are becoming less and less available. Better utilization of the spectrum can be achieved by spectrum prediction. In this paper, four algorithms, hidden Markov model (HMM), support vector machine (SVM), multi-layer perceptron (MLP), and long short-term memory neural network (LSTM-NN) algorithms, were introduced for the prediction of channel states in the radio-frequency spectrum. Experiments are conducted on a dataset. It was found from results that when the number of output sequence was fixed as 1 and the number of input channel state information (CSI) sequence was 1-20, the more input sequences were, the lower the error rate of the algorithm was; when the number of input CSI sequences was 20, the error rate of the LSTM algorithm was the lowest among several algorithms, which was 4.82%; when the number of input CSI sequenced was fixed as 1 and the future 1-15 CSI sequences were predicted, the more sequences need to be predicted was, the higher the error rate of the algorithm was; when 15 CSI sequences were predicted, the error rate of the LSTM algorithm was the lowest, 42.33%. The experimental results demonstrate that the LSTM algorithm is reliable in radio-frequency spectrum prediction and can be applied in practical UAV communication.
Keywords: unmanned aerial vehicle communication, radio-frequency spectrum, spectrum prediction, channel state