| With the rapid development of artificial intelligence and wireless communication,deep learning is widely used in the upper field of wireless communication systems.In order to improve the communication quality of the Filter Bank Multi-Carrier(FBMC)system,as well as to address the problems of channel estimation and signal recovery in the FBMC system,a FBMC signal detection system based on deep learning algorithms is proposed.This type of system replaces Traditional modules such as channel estimation in the FBMC system are used.Without knowing the prior channel statistical characteristics,the detection algorithm proposed in this paper can approach the actual channel environment through iterations during the training process,thereby improving the performance of the system in a complex channel environment.The thesis mainly studies how to use deep learning algorithms to improve the traditional multi-carrier detection system in FBMC system.First,the research background and development status of multi-carrier modulation principle and deep learning technology are introduced,and the FBMC system model is given.Then the channel estimation and channel equalization algorithms of the FBMC system are analyzed.After that,a brief description of common deep learning models and principles is given.Then,the thesis analyzes the characteristics of the signal form received by the FBMC system and the data processed by the deep learning,and performs preprocessing operations on the FBMC demodulated data according to these characteristics.Aiming at the problem of single network processing FBMC time-frequency signal difference,a FBMC signal detection algorithm based on Convolutional Neural Networks(CNN)and neural network is proposed.The CNN+NN network model is used to replace the channel estimation,equalization,and demapping modules in the traditional FBMC system.The simulation results show that,compared with the traditional LS channel estimation algorithm and the use of a single deep neural network,the FBMC signal detection system based on CNN+NN has better Bit Error Ratio(BER)performance and guides the system,and the signal detection system is less dependent on pilot frequency.Finally,for the time-frequency domain offset and fading distortion caused by the multipath channel in the FBMC system,the CNN extracting signal features is inaccurate,and a FBMC signal detection algorithm based on Recurrent Convolutional Neural Networks(RCNN)is proposed.In view of the long signal sequence,RCNN cannot connect the signal features at a long distance,which leads to the problem of gradient disappearance in the RCNN network during the training process.A FBMC signal detection algorithm combining CNN and Long Short-Term Memory(LSTM)is proposed.In view of the fact that the traditional dropout method cannot be directly applied to the loop layer of LSTM,the gaussian dropout method is introduced and applied to the FBMC signal detection algorithm of CNN+LSTM.The simulation results show that the FBMC system based on gaussian dropout CNN+LSTM not only has better BER performance than FBMC signal detection systems based on other deep learning algorithms,but also has no dependence on the pilot frequency in the system.Under high signal-noise ratio,its BER performance is also slightly better than the MMSE channel estimation algorithm. |
