| 6G will expand to higher terahertz(THz)bands to obtain larger transmission bandwidth to meet the rapidly growing demand for traffic and connections.By seamlessly integrating THz wireless communication and fiber optic communication technologies,the advantages of low loss fiber optic communication in communication bandwidth and transmission distance are effectively utilized,as well as the advantages of THz wireless communication in mobility and seamless coverage.However,the signal-to-noise ratio(SNR)limitation caused by severe atmospheric attenuation in the THz band significantly hinders the improvement of THz wireless transmission rate and distance.At the same time,in the photonics-aided THz communication systems,the large optical power from the fiber optic,the electro-optical modulator and photodiode in the optical link,the electrical amplifier in the wireless link,and the mixer at the receiver end for down-conversion and so on,will introduce nonlinear impairments and reduce system transmission performance.This dissertation focuses on the key technologies that can solve these problems,and conducts in-depth research in the application scenarios of broadband THz wireless communication systems.In the THz communication system,we propose an advanced machine learning training method,which can establish the correlation between the real and imaginary components of the quadrature amplitude modulation(QAM)signal and effectively preserve the phase information of the signal,to effectively solve the nonlinear damage problem of the THz communication system and realize the improvement of receiver sensitivity;We propose an advanced Delta-Sigma modulation(DSM)technology,in order to ensure a lower signal bandwidth and to reduce various noises.Meanwhile,the DSM technology can also effectively alleviate nonlinearity and realize long-distance wireless transmission of high-order QAM signals;We propose the combination of advanced machine learning training methods and DSM and so on,which can not only effectively solve the problem of nonlinear damage,but also effectively improve spectrum utilization.It can be applied in the future high-speed THz transmission with an electronic bandwidth bottleneck.The main innovative achievements are as follows:1.A fully connected complex-valued deep neural network nonlinear equalization algorithm,a gated recurrent unit nonlinear equalization algorithm based on in-phase/quadrature(I/Q)joint complex input,and a complex-valued deep neural network nonlinear classification algorithm are proposed to establish the correlation between the real and imaginary components of the QAM signal and to save the phase information of the signal,so as to effectively solve the nonlinear damage problem of the THz communication system.And in the scenarios of different baud rates and different modulation formats,the suppression effect of the complex neural network nonlinear equalization algorithm on the nonlinear damage in the long-distance large-capacity photonics-aided THz system is verified.In the three corresponding experimental demonstrations,4Gbaud PS-16QAM signal(sub-THz band,wireless distance of 4.6km),4Gbaud PS-64QAM signal(sub-THz band,wireless distance of 4.6km),10Gbaud PS-64QAM signal(THz band,wireless distance of 200m)can reach the net bit rate of 11.04Gbit/s,17.6Gbit/s and 44Gbit/s,respectively.The time complexity of the algorithm is reduced by more than 27.7%.2.The combination technology of DSM modulation and OFDM multi-carrier modulation as well as the combination technology of DSM modulation and polarization division multiplexing(PDM)are proposed to convert high-order QAM signals into low-order signals,effectively alleviating nonlinearity and realizing long-distance wireless transmission of high-order QAM signals.In the corresponding experimental demonstrations,we achieve the successful transmission of 1.25Gbaud OFDM-256QAM-1024QAM signals(THz band)on 20km single-mode fiber and 400m wireless links,as well as 1 Gbaud PDM1024QAM/2048QAM signals(sub-THz band,wireless distance of 4.6 km)with net bit rates of 16Gbit/s and 17.6Gbit/s,respectively.3.A technology combining DSM modulation and complex-valued deep neural network nonlinear equalization is proposed,and the application prospect of this technology in future high-speed THz transmission with electronic bandwidth bottleneck is verified.Based on this technology,in a 0.15THz free space wireless link system,a 1024QAM signal with a transmission rate of 15 Gbit/s can be successfully transmitted over a 2m free space link.When the soft-decision error threshold is 2×10-2,the sensitivity of the receiver using complex-valued deep neural network nonlinear equalization is increased by 1dB.The above experiments show that the introduction of machine learning algorithms into the nonlinear equalization technology can effectively solve the nonlinear damage problem at the THz receiver. |
