FBMC Modulation Signal Based On Gated Recurrent Unit Research On Detection Technology

With the rapid development of mobile communication technology,5G has gradually grown into a large-scale basic network that maintains the normal operation of all walks of life in society.The substantial increase in transmission rate and further expansion of the service range pose many challenges to the physical layer technology.Especially Filter Bank Multi-Carrier(FBMC),which is one of the key technologies of the physical layer.In the past ten years,the outstanding performance of neural network in image processing and speech recognition has attracted extensive attention of scholars.Its flexibility and strong pertinence have a great application space in FBMC system.The above characteristics can break the limitations of traditional FBMC system modules and get performance gains.Therefore,this paper studies how to optimize the signal detection module of the FBMC system using the Gated Recurrent Unit(GRU)in deep neural network.The main contents include:(1)The application of FC-GRU and CNN-GRU on FBMC system is studied.Aiming at the problem that the neural network lacks the library functions for dealing with complex numbers,the paper preprocesses the complex number data filtered by analyzing filter bank.Although GRU has advantages in processing time-series signal data,a single GRU network is not very capable of fitting nonlinear data.Therefore,a FC-GRU-based FBMC signal detection system is proposed.The signal detection model was improved by FC-GRU network model.However,the fully connected layers(FC)have insufficient ability to extract data features after filtering,so the FC layer was replaced with Convolutional Neural Networks(CNN),and a FBMC signal detection system based on CNN-GRU was proposed.Experimental results show that the FC-GRU and CNN-GRU models outperform the systems respectively using FC and CNN alone,respectively,due to the addition of the GRU network.Compared with the LS channel estimation algorithm,the FBMC signal detection systems based on FC-GRU and CNN-GRU both show better Bit Error Ratio(BER)performance,the latter not only has better BER performance but also has less dependence on the number of pilots.(2)The application of GRU self-encoding network in FBMC system is studied.inspired by the design of the single-carrier system based on the self-encoding network,it is found that the characteristics that the output is consistent with the input of the autoencoder network are very suitable for communication systems.However,when the autoencoder network performs data reconstruction,the accuracy rate will decrease with the increase of data sequence length.And the long-term memory ability of GRU can solve the above problem,so the GAE-SC system is proposed.After verifying its effectiveness through simulation,the idea is extended to the FBMC multi-carrier system,and the GAE-FBMC system is proposed.The experimental results show that the GAE-FBMC system performs better in BER performance compared with the FBMC signal detection system based on FC-GRU and CNN-GRU,and its BER performance is almost the same as that of the MMSE channel estimation algorithm,and The dependence on the number of pilots of the system is very low.After simulation analysis,the FBMC signal detection system based on FC-GRU and CNN-GRU and the GAE-FBMC system proposed in this paper have significantly improved the BER performance.