Research On Wireless Optical Communication And Positioning Based On Machine Learning

In recent years,with the exponential growth of the number of global mobile devices,more and more spectrum bandwidth has been occupied,resulting in the depletion of traditional radio spectrum resources.Wireless optical communication technology is rich in spectrum resources and does not require authorization.It is one of the key technologies for indoor and outdoor high-speed wireless communication in the future.Among them,Underwater Wireless Optical Communication(UWOC)and Visible Light Communication(VLC)are typical wireless optical communication application scenarios.However,when the wireless optical signal is transmitted in an underwater channel,it will be interfered by non-linear factors such as absorption and scattering,resulting in high bit error rate in the traditional signal demodulation method.In addition,the integrated mechanism of Visible Light Positioning and Communication(VLPC)for mobile users has not been established,and the resource allocation problem of dynamic communication systems is relatively complex,and traditional optimization methods are difficult to solve.In response to the above problems,this thesis proposes a method for underwater wireless optical signal demodulation based on machine learning,and a power distribution method for mobile users in the VLPC integrated system,which solves the problem of nonlinear demodulation,and resource allocation of dynamic visible light communication and positioning integrated system.The main research work and innovations of this thesis are summarized as follows:1.Firstly,a universal end-to-end UWOC hardware experiment platform is built,and the received signal vectors of nine single-carrier modulation schemes and DCO-OFDM modulation schemes are collected to establish the UWOC actual measurement database.The database URL is as follows: https: //pan.baidu.com/s/1Swj P78KZXyd6 a F7Tu Umu QQ(the extraction code is soz9).2.Furthermore,aiming at the high bit error rate of UWOC system signal demodulation,a demodulator based on DBN and a demodulator based on Ada Boost are designed.Based on the measured database,the performance of these two machine learning demodulators is studied.Experimental results show that the demodulation performance of the Ada Boost demodulator is the best.When the received optical power is fixed,the higher the demodulation order,the lower the demodulation accuracy.3.Finally,aiming at the problem that the VLPC dynamic system integration mechanism has not yet been established,the RSS-based visible light high-precision positioning technology is adopted to establish an integrated indoor visible light communication and positioning mechanism.In order to solve the power allocation problem of the VLPC dynamic system,the discrete deep reinforcement learning(DDRL)resource allocation scheme and the continuous DRL(Continuous DRL,CDRL)resource allocation scheme are further proposed,which can optimize the allocation of positioning power and communication power to maximize the downlink transmission rate of the system.Numerical simulation results show that the performance of the proposed CDRL scheme and DDRL scheme is better than that of the random power allocation method.The CDRL solution can provide a positioning accuracy of about 1.23 cm and a communication rate of about 8.33 bits/s/Hz.The DDRL resource allocation solution can provide a positioning accuracy of about 1.43 cm and a communication rate of about 7.08 bits/s/Hz.There are 41 figures,15 tables and 112 references in this thesis.