Performance Prediction in UAV-Terrestrial Networks With Hardware Noise

Abstract

To enhance the service quality of the unmanned aerial vehicle (UAV), the UAV-aided Internet of Things (IoT) systems could deploy a Deep Neural Network (DNN) for performance prediction for the users. Non-orthogonal multiple access (NOMA) is applied to such networks in order to improve spectrum efficiency, and results in improved quality of service at the ground users under the mobility of UAV. The outage and ergodic capacity requirements of the IoT users may not be satisfied due to some imperfect system parameters such as hardware noise. A DNN-based algorithm for performance prediction and the design of multiple antennas at the UAV under hardware noise is proposed. In this DNN-based UAV-NOMA, the central controller (server) collects system parameters periodically based on observing the state of IoT system and performs adjustments to the dynamic environment. The closed-form expressions for the outage probability and the ergodic capacity are derived to evaluate the performance of a group of IoT users. Our numerical results demonstrate that: i) In contrast to the traditional UAV-NOMA system, the UAV cannot know the performance at each IoT user in order to adjust the parameters (i.e. power allocation factors) before transmitting the signals to the devices; while the proposed DNN-based IoT system is capable of predicting the performance; ii) The performance of the IoT users can be significantly improved by integrating more antennas at the UAV and limiting levels of hardware noise; iii) By designing NOMA, the UAV-NOMA-based IoT system can increase the throughput to the tune of 40% compared with the benchmark (the orthogonal multiple access (OMA)-based IoT).

Keywords

• Deep neural networks
• non-orthogonal multiple access (NOMA)
• line-of-sight (LoS)
• unmanned aerial vehicles (UAVs)
• the Internet of Things (IoT)
• hardware noise