Deep Learning-Based Channel Prediction in Realistic Vehicular Communications

Abstract

Access to reliable estimates of the wireless channel, such as the channel state information (CSI) and the received signal strength would open opportunities for timely adaptation of transmission parameters and consequently increased throughput and transmission efficiency in vehicular communications. To design the adaptive transmission schemes, it is important to understand the realistic channel properties, especially in vehicular environments where the mobility of communication devices causes rapid channel variation. However, getting CSI estimates is challenging due to the lack of support for obtaining CSI from the chipset. In this paper, we present our efforts towards enabling reliable, up-to-date channel estimates in vehicular communications. We begin by designing and conducting a measurement campaign where we collect IQ (in-phase and quadrature) samples of the IEEE 802.11p transmission and implement CSI extraction algorithms to obtain and analyze wireless channel estimates from various real-world environments. We then propose a deep learning-based channel prediction for predicting future CSI and received signal levels. The trace-based evaluation demonstrates that our prediction approach improves the future power level estimate by 15% to 25% in terms of the root-mean-square-error compared to the latest known channel properties, thus, providing a sound basis for future efforts in anticipatory vehicular communication transmission adaptation.

Keywords

• Channel state information
• channel prediction
• vehicular communications
• neural networks
• LSTM