Machine Learning for Predictive Deployment of UAVs With Multiple Access

Abstract

This paper presents a machine learning-based framework for the predictive deployment of unmanned aerial vehicles (UAVs) as flying base stations (BSs) to offload heavy traffic from ground BSs. To account for time-varying traffic distribution, a long short-term memory (LSTM)-based prediction algorithm is introduced to predict future cellular traffic. A joint K-means and expectation maximization (EM) algorithm based on Gaussian mixture models (GMM) is proposed to determine the service area of each UAV based on the predicted user service distribution. Based on the predicted traffic, the optimal positions of UAVs are derived, and four multiple access techniques, namely, rate splitting multiple access (RSMA), frequency domain multiple access (FDMA), time domain multiple access (TDMA), and non-orthogonal multiple access (NOMA), are compared to minimize the total transmit power. Simulation results show that the proposed method can reduce up to 24% of the total power consumption compared to the conventional method without traffic prediction. Furthermore, RSMA is found to require the lowest transmit power among the four multiple access techniques. Therefore, this paper focuses on the comparison of multiple access techniques for UAV deployment, which is essential for the efficient and effective use of UAVs as flying BSs.