Research On PAPR Reduction Of FBMC-OQAM Signals Based On Autoencoder

Filter bank multicarrier with offset quadrature amplitude modulation(FBMC-OQAM)scheme is a popular multicarrier modulation scheme for next-generation mobile communications.Compared to the orthogonal frequency division multiplexing(OFDM)scheme used in 5G,it has the advantages of low out-of-band(OOB)energy leakage,good time-frequency localization,high spectrum utilization,but also has the same disadvantage of high peak to average power ratio(PAPR).In this paper,we propose an improved method using the autoencoder model in deep learning to reduce the PAPR of FBMC-OQAM signal.Based on the ability to reconstruct FBMC-OQAM signals without distortion,a parallelly distributed autoencoder scheme of fully-connected layer neural network is the key to this proposed method,and we call it DMCMNet(Distributed Multicarrier Modulation Net,DMCMNet).The bit error rate(BER)and PAPR are used as joint optimized objectives in DMCMNet to suppress the PAPR without increasing BER in communication.The proposed DMCMNet scheme draws on the classical idea of partial transmission sequence(PTS),each autoencoder can optimize the phase of a set of sub-carrier data,and then combine the output of all autoencoder into a FBMC-OQAM signal with low PAPR to transmit.Because the neural networks in autoencoder have the characteristic of freezing parameters after training,it does not need to produce random sequence like PTS method,which saves the transmission of side band information.Compared with the original FBMCOQAM system,the simulation results can prove that the proposed method has superior BER and PAPR performance in the case of low signal to noise ratio(SNR),and can be applied to future communication scenarios with low SNR as the main feature.