Machine learning based null allocation design for adaptive beamforming in unmanned aerial vehicle communications

In order to communicate with target unmanned aerial vehicles (UAVs), ground control stations (GCSs) typically adopt adaptive beamforming with high antenna gain and co-channel interference rejection. Multiple interfering signals arriving from different directions arise from other UAVs and other GCSs, and the beamformer installed in the home GCS will usually attempt to null unwanted signals from all these directions of arrival (DoAs) without analyzing the distribution of the angles of arrival. Consequently, the beamformer will fail to allocate nulls in some directions, and the signal-to-interference-plus-noise (SINR) performance of the home GCS is impaired. In this paper, a new approach to null allocation is proposed, based on machine learning using k-means clustering. The design first involves the collection of information about the DoAs and the corresponding received signal strengths of all the interfering signals into a two-dimensional dataset. Secondly, this dataset is broken down into clusters by using k-means clustering, and the cluster centroids are calculated. In each cluster, the interfering signal that has the shortest Euclidean distance to the centroid is identified as the approximated centroid. Only the approximated centroids are selected as input to the beamformer, with the aim that each complete cluster of interference sources can be nulled by allocating one null per cluster. To optimize the number of clusters k used in the null allocation process, the design adopts the particle swarm optimization technique to adaptively update the value of k to maximize the SINR at the home GCS. Simulation results show that our design yields a maximum SINR improvement of about 12 dB when compared to cases where no null allocation is considered. Moreover, our design also outperforms null steering in the UAV scenarios. Advantageously, this enhanced performance is obtained without the need for additional power amplification or hardware modification to the beamformer.