Parallel Multi-Relay Selection Method Based on DRL in Terminal Cooperative Communication

In terminal cooperative communication, the existing relay selection schemes with relay buffer do not consider the source data size and the transmit sequence of source packets, and are not suitable for parallel multi relay selection scenarios. Therefore, this paper proposes a parallel multi relay selection method based on deep reinforcement learning (DRL), which comprehensively considers the throughput of source-relay link and relay-destination link, and the buffer status of all optional relays, uses LSTM to predict the source data size of the next moment and divides the source data into appropriately sized packets, then uses DRL to perform parallel multi relay selection. In addition, in the process of relay selection, the sequence of source data packets, the complete process of packet forwarding, and the delay of packet forwarding are also considered. Therefore, the proposed parallel multi relay selection method based on DRL can achieve dynamic selection of multiple relays based on the number of divided data packets. Simulation results show that, compared with Max-Ratio method, proposed method can achieve higher throughput under the constraints of forwarding times and packet transmitting delay.