The integration of sensing and communication has become essential to next-generation vehicular networks. In this paper, we investigate a vehicle-to-infrastructure (V2I) network with multiple roadside units (RSUs) based on the dual-functional radar-communication (DFRC) technique. Since there are multiple RSUs in the system, we first propose a signal-switching model between vehicles and different RSUs. These RSUs estimate and predict vehicles’ motion parameters based on the DFRC signal echoes and the state evolution model. Accordingly, we utilise a neural network to extract angle information from signal echoes instead of traditional methods, thus improving the angle estimation accuracy. To further improve the estimation performance, we formulate an optimisation problem to minimise the Cramer-Rao bound (CRB) on angle estimation by properly allocating power to each RSU. Finally, we propose a novel weighting method to further improve the cooperative localisation accuracy of the multi-RSU system. Simulation results show that the performance of angle estimation can be improved by utilising the proposed neural network method and the novel power allocation scheme. In addition, the novel weighting method can considerably improve the localisation accuracy.

传感和通信的集成对于下一代车辆网络至关重要。在本文中，我们研究了基于双功能雷达通信（DFRC）技术的具有多个路边单元（RSU）的车辆到基础设施（V2I）网络。由于系统中有多个 RSU，我们首先提出车辆和不同 RSU 之间的信号切换模型。这些 RSU 基于 DFRC 信号回波和状态演化模型来估计和预测车辆的运动参数。因此，我们利用神经网络代替传统方法从信号回波中提取角度信息，从而提高角度估计精度。为了进一步提高估计性能，我们制定了一个优化问题，通过向每个 RSU 正确分配功率来最小化角度估计的 Cramer-Rao 界 (CRB)。最后，我们提出了一种新的加权方法来进一步提高多 RSU 系统的协作定位精度。仿真结果表明，利用所提出的神经网络方法和新颖的功率分配方案可以提高角度估计的性能。此外，新颖的加权方法可以显着提高定位精度。