Research On Underwater Acoustic Communication Technology Based On Comprehensive Performance Optimization

In underwater acoustic sensor networks,node energy consumption,delay,network throughput and communication bit error rate and communication task execution status are a set of mutually affecting performance parameters.Therefore,in the harsh underwater acoustic channel environment,improving the effectiveness and robustness of data communication and data collection based on underwater acoustic sensor networks has become a hot research topic in underwater communication.To solve this problem,the research proposes to establish a comprehensive optimization function of network performance based on the above performance parameters,transform the problem into an optimal solution problem,and use reinforcement learning to solve the problem.The main innovations of the paper are as follows:1.Starting from point-to-point underwater communication technology,a robust point-to-point underwater acoustic communication technology(Reinforcement learning based adaptive modulation and coding technology,RLMC)is proposed based on reinforcement learning.In response to the problem of large delay in underwater acoustic communication,the technology first divides the underwater acoustic channel into two independent frequency bands,a data channel and a feedback channel,to improve the efficiency of data transmission.Next,for the fast time-varying underwater acoustic channel environment,the channel impulse response and bit error rate are used as feedback information,and reinforcement learning is used to predict channel changes to solve the comprehensive performance optimization problem of point-to-point communication.The performance of the proposed algorithm was verified by the pool experiment and sea experiment data.Compared with the reference algorithm,RLMC improves throughput,reduces bit error rate and transmission time,and saves energy consumption.What's more,compared to pure Q-learning,its convergence speed is faster.For example,in the pool experiment,the BER of RLMC was reduced by about 44%,the transmission time was reduced by about 53%,the energy consumption was reduced by about 25%,and the reward was increased by about 63%.2.For multi-point communication,the underwater wireless sensor network is studied,and a task-based adaptive deep routing protocol(MC-DBR)is proposed.The protocol divides the data packets generated by source nodes of different task types into three different levels,and different levels of data packets have different retransmission strategies in the case of packet loss.Then,based on multi-agent reinforcement learning,an efficient multi-agent reinforcement learning based adaptive modulation and coding technology(MARL-MC)is designed,which regards each node in MC-DBR as an independent agent.The agents perform cooperative learning by exchanging V-value function information,and then select the best modulation and coding method and relay node for the data packet.The goal of MARL-MC is to find a path with the highest reward from the source node to the target node.The performance of the proposed algorithm is verified using marine experimental data.Compared with the reference algorithm,MARL-MC reduces the single-hop bit error rate and transmission time.Compared with RLMC which uses distributed independent Q-learning,the cooperative learning method MARL-MC improves the overall performance of the communication network.3.Based on OPNET,a simulation platform for underwater acoustic sensor network communication and data collection has been established.The platform modified OPNET's corresponding pipeline stage model to enable it to truly simulate the true environment of the underwater wireless communication network,effectively improving the credibility of the simulation performance;Then implement the data model and node model defined in MC-DBR,and the corresponding algorithm.Finally,based on different network models,the performance of DBR protocol is compared,and the superiority of MC-DBR is verified again.