Research On Papr Suppression For OFDM Systems Based On Deep Learning

With the rapid development of society,China's mobile communication technology has made a qualitative leap.5G,or the fifth generation of mobile communication technology,emerged as the new generation of mobile communication system.Orthogonal frequency division multiplexing(OFDM)plays an important role in 5G due to its characteristics of high spectrum utilization,anti-multipath fading and anti-intersymbol interference.However,the OFDM system adopts parallel transmission technology and uses multiple carriers for modulation,which will lead to a higher peak-to-average power ratio(PAPR)of the system,making signal distortion and distortion seriously affect the communication quality.Therefore,how to reduce PAPR value simply and effectively is one of the key problems in OFDM technology.Aiming at this topic,this paper carries out technical research on reducing PAPR value of OFDM system based on deep learning.The specific work is as follows:(1)this paper introduces the principle of OFDM technology,the definition and distribution of PAPR,and analyzes three traditional technologies for reducing the PAPR value of OFDM systems.The three traditional technologies have their own advantages and disadvantages,but there is still no algorithm with excellent performance in all aspects.Therefore,it is considered to reduce the PAPR value of OFDM system by combining deep learning.(2)the TFNN scheme and PRNet scheme based on deep learning to reduce the PAPR value of OFDM system are introduced,and the simulation results are analyzed.TFNN scheme is an improvement on ACE scheme.It uses deep learning training to obtain ACE signal,which can reduce IFFT operation times,simplify operation and reduce operation complexity.The PRNet scheme can be regarded as an improvement of the coding technology.Its principle is to find the mapping with the lowest PAPR value ofOFDM signal through the training of noise reduction automatic encoder,and ensure that the signal can be solved and reconstructed from the original signal at the receiving end.(3)as PRNet scheme is significantly superior to TFNN scheme in performance,this paper proposes PRNet-RP scheme through improvement and innovation of PRNet scheme.PRNet-RP scheme mainly improves and innovates the following three parts on the basis of the original PRNet model: firstly,the input data set is preprocessed to make the input of the self-encoder conform to the input data format requirements of the deep learning model,and the virtual part information of the original data set is retained.Then,the loss function is improved to further clarify that the optimization target only considers PAPR performance,so as to reduce PAPR value of OFDM system more effectively and reduce the number of iterations.Finally,the model structure of the PRNet scheme was improved.The improved model changed the activation function of the last sub-block of the encoder and decoder to Tanh activation function,which could avoid PAPR performance degradation after the encoded OFDM signal remove the direct current and ensure that the self-encoder could reconstruct the input sequence.The above improvements and optimizations enable PRNet-RP scheme to be better applied in actual production.