Research On Signal Processing Technologies Of Microwave Photonics Link Based On Artificial Neural Networks

With new services and applications of the commercialization of 5G and the development of next-generation mobile communication systems such as B5G and 6G,the huge number of nodes and connections will change the transmission environment of simple single point to point into the transmission environment of complicated multi-point.Common public radio interface(CPRI)is the main interface of current network architecture,with the step of quantization and sampling,the data rate of interface will rapidly increase.Furthermore,with the use of massive MIMO for capacity expansion,the data rate of CPRI will largely increase due to the increasement of used antennae,it will put a lot of pressure on transmission bandwidth of interface.Meanwhile,microwave photonics link employs analog waveform signal rather than digital I/Q signal,the quantization and sampling of digital signal which is based on CPRI are avoided.Therefore,the microwave photonic link which combines narrowband microwave technology and wideband optical technology is regarded as a vital role in post-5G mobile communication.The radio over fiber technology(RoF)which is based on microwave photonic link could meet variety demands like low loss,wide bandwidth,multi-band frequency and flexibility.Therefore,it is considered an effective solution which could realize new type broadband wireless access.This paper mainly focuses on two problems.Firstly,in wideband and multi-carrier microwave photonic link,there are various nonlinear distortions like harmonics distortions(HD),inter-modulation distortions(IMD),crossmodulation distortion(XMD).These distortions would jointly degrade dynamic range(SFDR)of link.Secondly,in multi-user and multi-frequency spectrum broadband wireless access network,the integrated nonlinear interference generated by a single user and multiple users will act simultaneously.In order to solve these problems,we adopt various advanced machine learning algorithms like artificial neural network(ANN)to realize multi-user and compatible linearized microwave photonics link.The proposed method is expected to be applied to multi-user and multi-frequency spectrum broadband wireless access.The main research works are as follows:1.For the purpose of mitigating cross-modulation nonlinear distortion(XMD)which appears in wideband and multi-carrier microwave photonic link.A novel linearization approach which based on full connected artificial neural network(ANN)is proposed.In simulation,how to design the specific structure of neural network is discussed.In experiment,we adopt intensity modulated and direct detection system(IM/DD),two dual-tone signals separated by 15 GHz are used to simulate a broadband multi-carrier link.The experimental results show that XMD suppression ratio is improved by 22.18 dB.The inter-modulation distortion(IMD)suppression ratio is also improved by 29.76 dB after proposed ANN linearized algorithm.The proposed method has simple structure and good linearization performance,it could realize effective mitigation of XMD in wideband and multi-carrier microwave photonics link.2.For wideband microwave photonics link which contains multi-users,in order to realize compensation of various nonlinear distortion simultaneously.A high compatibility and low-cost linearization scheme based on improved ANN with transfer learning(TL)is proposed.The use of transferred parameters of different ANN could reduce the overall learning cost of proposed method and realize various nonlinear distortion mitigation of different number of users.This approach could realize low-cost suppression of XMD and IMD simultaneously.In experiment,the signals are compatible linearized by using networks for two users separated by 15 GHz and three users separated by 5 GHz and 10 GHz,respectively.By using the network for three users,the experimental results show that XMD is suppressed by 22.38 dB and IMD3 is suppressed by 16.73 dB,the SFDR of link is improved by 14.3 dB.This scheme has low training cost and high compatibility.With simultaneous mitigation of various nonlinear distortions,linearized broadband microwave photonics link which contains multi-users could be achieved.3.In the cases of broadband wireless access for multi-frequency spectrum and multi-user application scenario,in order to mitigate the integrated nonlinear interference generated by a single user and multiple users,an effective and compatible scheme based on waveform regression ANN with TL is demonstrated.This proposed method uses machine learning algorithms to reduce learning costs and improve compatibility in different application scenarios.In experiment,an analog RoF access system with 2 km multi-core fiber transmission and 0.5 m wireless transmission for Sub-6 GHz is demonstrated.We adopt OFDM signal to carry out the analog RoF transmission experiment with multi-core fiber.The experimental results show that the proposed method could effectively decrease bit error ratio(BER)of signal to 3.8×10-3 of the forward error correction threshold(FEC)in the case of RoF link which contains four users.Moreover,when realizing the compatible nonlinear equalization between different number of users,this proposed scheme possesses good compatibility for different scenarios which contain various numbers of users.The improved optical receiver sensitivity is larger than 1.5 dB when simple system which contains two users adopts same nonlinear equalization network of system which contains four users.The proposed method could realize compatible wideband wireless access for multi-user and multi-frequency spectrum signals.