Research On PAPR Suppression Algorithm In OFDM System Based On Deep Learning

Orthogonal frequency division multiplexing(OFDM)technology plays an important role in the fifth generation of mobile communication technology due to its excellent performance in frequency selective fading channels.However,as a multi-carrier transmission scheme,the inherent high PAPR problem of OFDM symbols can cause severe nonlinear distortion.Traditional PAPR suppression algorithms have limited performance and will cause certain negative effects.However,due to strong mapping capabilities and flexible learning capabilities,deep learning can consider PAPR performance,bit error rate performance,complexity,and spectrum utilization at the same time.Therefore,this article uses deep learning methods to improve the existing PAPR suppression algorithm in the OFDM system to further improve its suppression capability.The main work of this paper is as follows:(1)The basic principles of the OFDM system,the definition and distribution of PAPR are introduced.(2)Programmed to realize the common PAPR suppression algorithm,and their implementation principles,advantages and disadvantages,and their performance are analyzed.An improved SLM-Clipping joint algorithm is proposed.The algorithm is implemented by two algorithms used in series,to achieve a progressive inhibition of the PAPR.The proposed algorithm uses the SLM algorithm to perform the first PAPR suppression,and on this basis,the Clipping algorithm is used for secondary PAPR suppression.Compared with the direct use of the Clipping algorithm,the introduction of limiting noise is greatly reduced,and the PAPR suppression performance is also significantly improved.(3)The basic principles of deep learning,the simplified limiting filtering algorithm based on neural network and the PRNet scheme based on autoencoder are introduced and the realization principle,system structure,advantages and disadvantages and performance of these scheme are also introduced.(4)We propose a novel tone reservation network(TRNet)based on deep learning to further enhance the PAPR performance of the conventional TR technique.In the proposed scheme,we utilize the feedforward neural network(FFNN)to adaptively generate the peak-canceling signal according to the characteristics of the input signal.Compared to the existing PAPR reduction schemes that apply neural networks both at the transmitter and the receiver,our proposed scheme only requires the FFNN at the transmitter,which reduces the cost of the system.The simulation results demonstrate that the proposed scheme can significantly reduce the PAPR of the OFDM system without causing any additional signal distortion.Besides,in our simulation process,only a few number of data sets are required for the learning and the model can converge quickly,which provides a feasible solution for solving real-time problems.