The vigorous developments of the Internet of Things make it possible to extend its computing and storage capabilities to computing tasks in the aerial system with the collaboration of cloud and edge, especially for artificial intelligence (AI) tasks based on deep learning (DL). Collecting a large amount of image/video data, unmanned aerial vehicles (UAVs) can only hand over intelligent analysis tasks to the back-end mobile edge computing (MEC) server due to their limited storage and computing capabilities. How to efficiently transmit the most correlated information for the AI model is a challenging topic. Inspired by task-oriented communication in recent years, we propose a new aerial image transmission paradigm for the scene classification task. A lightweight model is developed on the front-end UAV for semantic block transmission with the perception of images and channel states. To achieve the tradeoff between transmission latency and classification accuracy, deep reinforcement learning (DRL) is applied to explore the semantic blocks which have the greatest contribution to the back-end classifier under various channel states. Experimental results show that the proposed method can significantly improve classification accuracy by more than 4% under the same conditions, compared to other semantic saliency learning methods.

A 5G network is the key driving factor in the development of vehicle-to-vehicle (V2V) communication technology, and V2V communication in 5G has recently attracted great interest. In the V2V communication network, users can choose different transmission modes and power levels for communication, to guarantee their quality-of-service (QoS), high capacity of vehicle-to-infrastructure (V2I) links and ultra-reliability of V2V links. Aiming at V2V communication mode selection and power adaptation in 5G communication networks, a reinforcement learning (RL) framework based on slow fading parameters and statistical information is proposed. In this paper, our objective is to maximize the total capacity of V2I links while guaranteeing the strict transmission delay and reliability constraints of V2V links. Considering the fast channel variations and the continuous-valued state in a high mobility vehicular environment, we use a multi-agent double deep Q-learning (DDQN) algorithm. Each V2V link is considered as an agent, learning the optimal policy with the updated Q-network by interacting with the environment. Experiments verify the convergence of our algorithm. The simulation results show that the proposed scheme can significantly optimize the total capacity of the V2I links and ensure the latency and reliability requirements of the V2V links.

Reasonable design and effective detection of the random access preamble has become a challenging task due to the unique characteristics of mobile satellite communications. To tackle this challenge, we first design a universal long sequence structure by concatenating multiple short Zadoff-Chu sequences that are insensitive to carrier frequency offset (CFO), and then propose the new principles of parameter selection for short sequences to ensure the minimum utilization of root sequence and the independence of the cyclic shift offset on the beam radius. To further reduce the detection complexity and improve the multi-user access performance, a fast timing detection approach is also presented by leveraging the piecewise cumulative detection and the multi-peaks joint estimation to obtain an accurate timing advance for each access user. Simulation results and complexity analysis validate the effectiveness of the new preamble in a typical satellite communication environment, and reveal that the proposed timing detection can achieve the robustness to CFO and offer outstanding performance improvements especially in multi-user scenarios while having a notably reduced computational complexity.

Vehicle location prediction or vehicle tracking is a significant topic within connected vehicles. This task, however, is difficult if merely a single modal data is available, probably causing biases and impeding the accuracy. With the development of sensor networks in connected vehicles, multimodal data are becoming accessible. Therefore, we propose a framework for vehicle tracking with multimodal data fusion. Specifically, we fuse the results of two modalities, images and velocities, in our vehicle-tracking task. Images, being processed in the module of vehicle detection, provide visual information about the features of vehicles, whereas velocity estimation can further evaluate the possible locations of the target vehicles, which reduces the number of candidates being compared, decreasing the time consumption and computational cost. Our vehicle detection model is designed with a color-faster R-CNN, whose inputs are both the texture and color of the vehicles. Meanwhile, velocity estimation is achieved by the Kalman filter, which is a classical method for tracking. Finally, a multimodal data fusion method is applied to integrate these outcomes so that vehicle-tracking tasks can be achieved. Experimental results suggest the efficiency of our methods, which can track vehicles using a series of surveillance cameras in urban areas.