| Since the 1990s,with the rapid development of wireless communication technology and the improvement of material and cultural level,people have gradually had higher requirements for communication quality and communication efficiency.At the same time,the number of devices connected to mobile networks has increased.Higher requirements are also placed on the load capacity of the wireless communication network.With the full coverage of my country’s"four vertical and four horizontal"high-speed rail network,high-speed rail has become the primary way of travel for people,and mobile communication in high-speed scenarios has become a research hotspot in the academic world.In high-speed scenarios,the high Doppler frequency shift caused by high-speed movement makes it impossible to guarantee the orthogonality of subcarriers.In high-speed scenarios,the performance of traditional OFDM is poor,and Orthogonal Time Frequency Space Modulation is expected to solve the communication problem in high-speed scenarios.On the other hand,in recent years,the application of machine learning in the field of communication has become more and more extensive,and its outstanding performance in channel estimation and signal detection has attracted the attention of scholars.Based on the above background,this paper studies the Orthogonal Time Frequency Space Modulation system and its related signal detection algorithms in high-speed mobile scenarios.Firstly,this paper expounds the development history of wireless communication,points out that it is of great significance to study wireless communication technology in high-speed scenarios,introduces the research status of OTFS technology at home and abroad,and then briefly introduces the application of machine learning in the field of wireless communication.Secondly,this paper studies and analyzes the principle and system model of OTFS technology modulation and demodulation,and then this paper compares and analyzes the message passing algorithm（Message Passing,MP）and the approximate message passing algorithm（Generalized Approximate Message Passing,GAMP）.The conclusion shows that GAMP algorithm solves the problem of MP algorithm complexity NP,which can reduce the computational complexity while ensuring a certain detection performance.Thirdly,because the improper selection of the damping factor of the GAMP algorithm will bring about the decline of the detection performance,this paper proposes the GA-GAMP algorithm based on the genetic algorithm to optimize the damping factor.The algorithm regards the damping factor of each iterative layer as an independent parameter,the bit error rate of a specific signal-to-noise ratio is minimized through multi-objective optimization,and the corresponding simulation and analysis are carried out.The results show that the detection performance of the GA-GAMP algorithm has a certain improvement compared with the GAMP algorithm with a fixed damping factor.Finally,since the damping factor of each iteration layer affects the convergence performance of the current iteration layer,it also affects the initial value of subsequent iterations,thereby affecting the overall detection performance of the system.Therefore,the training network that pays attention to both the iteration process and the final bit error rate is more important.to be reasonable.In this paper,the deep reinforcement learning method is used to optimize the signal detection of OTFS,and the A2C-GAMP detection algorithm is proposed.The algorithm expands the iterative process of GAMP into different states of reinforcement learning.Completing all iterations of GAMP is equivalent to the end of an episode.The action of the agent is to select the training parameters of each iteration.To avoid sparse rewards,the Reward of each episode is composed of RewardT of different iteration layers and RewardBER of the bit error rate reward,and the Actor’s policy network and the Critic’s value network are fitted through a neural network.The simulation results show that compared with the traditional GAMP detection algorithm,the A2C-GAMP algorithm has obvious performance improvement. |
