Predicting Channel Quality Indicators for 5G Downlink Scheduling in a Deep Learning Approach (2008.01000v1)

Abstract: 5G networks provide more bandwidth and more complex control to enhance user's experiences, while also requiring a more accurate estimation of the communication channels compared with previous mobile networks. In this paper, we propose a channel quality indicator (CQI) prediction method in a deep learning approach in that a Long Short-Term Memory (LSTM) algorithm. An online training module is introduced for the downlink scheduling in the 5G New Radio (NR) system, to reduce the negative impact of outdated CQI for communication degradation, especially in high-speed mobility scenarios. First, we analyze the impact of outdated CQI in the downlink scheduling of the 5G NR system. Then, we design a data generation and online training module to evaluate our prediction method in ns-3. The simulation results show that the proposed LSTM method outperforms the Feedforward Neural Networks (FNN) method on improving the system performance of the downlink transmission. Our study may provide insights into designing new deep learning algorithms to enhance the network performance of the 5G NR system.