| Coherent optical orthogonal frequency division multiplexing（CO-OFDM）systems are formed based on multi-carrier modulation and have the advantages of high reception sensitivity and easy compatibility with wavelength division multiplexing systems.There is also an unavoidable problem in CO-OFDM systems,which is the high peak-to-average power ratio（PAPR）.The presence of high PAPR can lead to problems such as spectral interference and BER performance degradation,which influence the system performance.To achieve PAPR suppression in CO-OFDM systems,the work of this thesis is shown below:1.A DCT cascade iterative limiting filtering（ICF）algorithm is proposed.The PAPR reduction can also be achieved by changing sequence correlation through DCT,and there is a low computational complexity of DCT.The ICF algorithm can mitigate the nonlinear distortion caused by clipping operation through multiple iterations of filtering,the system performance is improved.But the PAPR suppression ability of the system decreased after iteration.Therefore,the DCT-ICF cascade algorithm is proposed to increase the PAPR suppression capability of the ICF algorithm at the sacrifice of a little computational complexity.The results of simulation show that the PAPR value of the DCT-ICF（CR=4,iter=4）algorithm is 6.795d B,when the CCDF is10-4.2.A DCT cascaded low complexity selective mapping（LCSLM）algorithm is proposed.The SLM algorithm achieves the PAPR suppression in the system by interfering with the signal phase.During the operation of the algorithm,some sideband information is attached,which increases the computational complexity.A LCSLM algorithm is proposed,which splits the complex signal,increases the number of groups without increasing the number of IFFT/FTT,reduces the computational complexity.The simulation results show that when the CCDF is 10-4,the PAPR value of the DCT-LCSLM（D=4）algorithm is 6.435d B.3.A neural network（NN）model is proposed to achieve the suppression of PAPR in CO-OFDM systems based on the cascade algorithm.The NN model is trained according to the labeled data generated by the proposed algorithm to achieve the purpose of data optimization.The improved algorithm uses the ICF algorithm and the improvedμcompanding algorithm,the two sets of data are trained by the NN model respectively.A neural network model is built to train the label data of ICF algorithm and improvedμcompanding algorithm to achieves the optimization of BER performance in the system.The ISLM-NN-ICF algorithm and the ISLM-NN-improvedμcompanding algorithm are proposed respectively,both of which can achieve further reduction of the system PAPR.The results of the data indicate that when the CCDF is 10-4,the value of PAPR for the ISLM-NN-ICF（D=4,iter=4）algorithm is 5.910 d B,and the PAPR value of the proposed ISLM-NN-improvedμ（D=4,μ=2,m=2）companding algorithm is 5.675d B. |
