An Integrated New Deep Learning Framework for Reliable CSI Acquisition in Connected and Autonomous Vehicles

According to society of automotive engineers (SAE)'s standard J3016, there are six levels of driving automation, ranging from no driving automation (level 0) to full driving automation (level 5). In order to make autonomous vehicles (AVs) fully automated when driving, they must be capable of comprehending the driving environment, which requires the assistance of 5G networks and multi-access edge computing (MEC). Through the use of advanced precoding techniques, 5G networks are able to provide low latency and high transmission rates. The real-time precoding is correlated with the quality of estimated instantaneous channel state information (CSI). However, two obstacles prevent acquiring accurate CSI. One is CSI feedback overhead, and the other is channel aging. The aforementioned challenge can be successfully solved through deep learning, but there are two aspects that have been overlooked. The conventional way to solve the challenge separately with two deep learning models lead to a sub-optimal results. Besides, deep learning models are static and unable to adapt to new data. In this article, we propose an integrated framework to address the performance degradation caused by acquiring CSI via two deep learning models. Further, we introduce continual learning to train the deep learning model to enhance its adaptability to new data. The simulation results demonstrate the integrated framework can address the challenge of CSI acquisition with a satisfactory performance. In addition, continual learning can help the deep learning model to adapt to changing data in a dynamic environment.