| Visible Light Communication(VLC),as one of the most promising optical wireless communications in commercial applications,has great potential for unauthorized transmission bandwidth.In order to make efficient use of spectrum resources,orthogonal frequency division multiplexing(OFDM)is widely used in VLC systems.Because OFDM signals have high peak-to-average ratio(PAPR).When the OFDM signal with high PAPR passes through the light-emitting diode(LED),it will exceed its linear operating range and produce nonlinear effects.Discrete Fourier Transform-Spread-OFDM(DFT-S-OFDM)is an active method to achieve low PAPR and better bit error rate performance.However,when the nonlinearity of VLC system is very high,the performance of DFT-S-OFDM will also deteriorate.At the same time,the complex nonlinearity makes the mathematical derivation of the system more difficult,and the nonlinear interference can be reduced by machine learning.Starting from the main problems of signal decision demodulation and channel estimation in VLC system,this thesis uses the idea of machine learning to carry out innovative solutions to the above two problems:(1)A signal decision demodulation method for DFT-S-OFDM communication system based on Gaussian Mixture Model(GMM)clustering is proposed.Starting from DFT-S-OFDM system,this thesis analyzes the nonlinear effects in VLC system.At the same time,because the OFDM signal is composed of multiple subcarriers that are transmitted synchronously in the time domain,these subcarriers are superimposed at the same time,which will produce a large PAPR.When the signal with high PAPR passes through the LED,it will exceed its linear operating range,resulting in a decline in signal quality and an increase in bit error rate.GMM estimates the Gaussian distribution parameters of all constellation points directly according to the received signal,then calculates the probability of the Gaussian distribution of the received signal to each constellation point,and selects the constellation point corresponding to the maximum probability as the decision result of the received signal for demodulation.Compared with the traditional method based on hard decision demodulation,in the LED nonlinear channel,the decision demodulation method using GMM clustering can obtain about 0.6～2.7d B and about 0.2～1.7d B signal-to-noise ratio gain under16-order quadrature amplitude modulation(QAM)and 32 QAM,respectively.(2)A channel estimation method based on Long Short Term Memory(LSTM)for DFT-S-OFDM communication systems is proposed to mitigate nonlinear effects.For DFT-S-OFDM communication systems,using LSTM networks to compensate for signal nonlinearity,the obtained bit error rate performance is superior to traditional channel estimation algorithms.At the signal receiving end,offline training and online deployment methods are used to replace traditional channel estimation algorithms.In the simulation,we obtained better performance than traditional channel estimation algorithms without using pilots,which means less spectrum resources are consumed.This further indicates that the proposed method can achieve better performance without requiring channel assistance information,and to some extent enhance the stability of the system.Compared with the Least Squares(LS)LS channel estimation algorithm,the channel estimation method based on LSTM can achieve a gain of about 1 d B at 32 QAM;Under 64 QAM and 128 QAM,when the LS channel estimation algorithm cannot reach the 7% forward error correction code tolerance,the LSTM channel estimation algorithm can achieve this.It is further explained that the LSTM channel estimation algorithm is more effective in mitigating the nonlinear effects of the channel. |
