Research On Cubic Metric Suppression Of OFDM Signal Based On Deep Neural Networks

Orthogonal frequency division multiplexing(OFDM),as a multi carrier modulation technology,has attracted much attention because of its high frequency spectral efficiency,robustness to multipath effect and realizability based on fast Fourier transform(FFT).At present,it has been widely used in modern communication systems.However,in OFDM system,large signal envelope fluctuation is its main disadvantage.Due to the limited linear range of the power amplifier(PA)at the transmitter,if the signal peak is too high,PA will produce serious nonlinear distortion in the process of signal amplification.It will not only increase the bit error ratio(BER)of the system,but also cause serious out of band leakage.Cubic metric(CM)is an index used to measure the envelope fluctuation,and CM suppression is a key technology in OFDM system.Traditional CM suppression techniques for OFDM systems have many problems,such as many iterations,high computational complexity,and the performance of envelope suppression can not be taken into account with other performance.In order to overcome the above problems,this paper introduces the deep neural network(DNN)technology and proposes a series of algorithms based on DNN to reduce the CM of OFDM system.The main contents of this paper are as follows:Firstly,an algorithm based on ensemble learning to suppress the CM of OFDM signal is proposed.In this paper,DNN technology is used to simulate the process of simplified clipping and filtering(SCF)algorithm,which avoids the problem of multiple iterations of traditional algorithm.Aiming at the problem that the process is too complex,a new network structure is designed,which has the characteristics of low complexity.In order to further improve the performance,the ensemble learning method is used to optimize the combination of network models.The algorithm can approximately achieve the suppression performance and error performance of SCF algorithm,and its complexity is greatly reduced.Secondly,an algorithm based on multi task learning to suppress the CM of OFDM signal is proposed.Based on the idea of DNN and active constellation extension(ACE),a neural network structure based on multi task learning is proposed.The model can comprehensively consider the performance of suppression and error rate.By simplifying the structure,the complexity of the model is reduced,the model has better robustness,greatly reduces the requirements of training set,and has good suppression effect on CM and low BER.