LAUSR

The unmanned aerial vehicles (UAVs) have been widely considered as one of the key applications for future wireless communication systems, where UAVs can be used as aerial base stations (BSs) for coverage extension, transmission improvement, emergency communication, and etc. Against this background, each UAV BS is expected to select the optimal codeword to form directional analog beams, and it is capable of achieving concurrent transmissions from multiple other UAV BSs simultaneously. However, in such a kind of UAV networks, due to the vast number of connected mobile users (MUEs), UAV BSs cannot timely and preciously select the codeword from the pre-defined codebook. Fortunately, machine learning (ML) is suitable for decreasing complexity in codeword selection, because ML could extract features from the data samples acquired in real environments. In this paper, we propose an ML approach to achieve an efficient and low complexity codeword selection for UAV networks. Specifically, we first derive the probabilities that multiple UAV BSs serve one MUE to obtain the average sum rate (ASR) in UAV networks. On that basis, we develop an ML approach to maximize the ASR, where we design a classifier based on support vector machine (SVM), where our ML approach is used for selecting the optimal codeword and maximizing the ASR in UAV networks. Third, we proposed an iterative sequential minimal optimization (SMO) training algorithm to train the data of all links between UAV BSs and MUEs, where the algorithm convergence is also discussed. Finally, we show the comparison between our proposed algorithm and the traditional methods by the simulation results. The simulation at last demonstrate our method is a more efficient solution for obtain a higher performance, where a much lower computational complexity can achieved than the traditional algorithm based on channel estimation.