BAYESIAN OPTIMISATION-BASED DYNAMIC CHANNEL PREDICTION FOR THE URLLC-ORIENTED UAV COMMUNICATION SYSTEM

Qiuyun Zhang, Hong Jiang, Fanrong Shi, Liping Deng, Qiumei Guo, Ying Luo, and Tingting Yan

Keywords

Unmanned aerial vehicles (UAVs), channel prediction, ultra-reliable and low-latency communications (URLLC), Bayesian optimisation, long short-term memory (LSTM)

Abstract

The acquisition of real-time and accurate channel state information (CSI) is the foundation for achieving ultra-reliable and low-latency communication (URLLC). However, in high-speed mobile application scenarios, the wireless channel between the transmitter and receiver exhibits rapid changes and non-stationary characteristics. This makes obtaining CSI increasingly difficult, and conventional data- driven prediction models are difficult to adapt. Therefore, this paper proposes a new model-driven prediction framework to tackle the challenge. Specifically, we propose a new prediction framework based on Bayesian optimisation and long short-term memory (LSTM). This method continuously acquires and updates CSI data in real time, evaluates prediction accuracy, and optimises the model based on these evaluations. This iterative process allows the model to adapt to changes, leading to more accurate CSI predictions. To demonstrate the effectiveness of our forecasting framework, we designed a UAV operational context and conducted experiments on the UAV control channels. In comparing various data-driven models, including LSTM, gated recurrent unit (GRU), TCN, and recurrent neural network (RNN), the simulation outcomes revealed that our approach performed better on dynamic non-stationary UAV control channels. Our method, evaluated by mean squared error (MSE), shows superior performance over other prediction methods, reducing errors by 3.86% on average.

Important Links:



Go Back