Digital-Twin-Enabled City-Model-Aware Deep Learning for Dynamic Channel Estimation in Urban Vehicular Environments

Abstract

Most of the existing works on vehicle-to-everything (V2X) communications assume some deterministic or stochastic channel models, which is unrealistic for highly-dynamic vehicular channels in urban environments under the influence of high-speed vehicle motion, intermittent connectivity, and signal attenuation in urban canyon. Enabled by the concept of digital twin, the digital replica of a real-world physical system, this paper proposes a city-model-aware deep learning algorithm for dynamic channel estimation in urban vehicular environments. Specifically, the digital twin simulation allows us to accurately model radio ray reflection and attenuation in urban canyon, and the data can be supplemented and validated with empirical measurements. Our results indicates that the city-model-aware deep neural network (CMA DNN) estimator performs much better than conventional methods and has more than 32% improvement in BER when compared with generic DNN approaches that do not take the 3D city model into account. Since some geometry-based models like ray-tracing techniques used in the digital twin simulation for dynamic channel modeling could be computational expensive, we also propose a basis expansion model (BEM) approach to simplify the computation load of the overall methodology to gain a good balance between accuracy and timeliness.