Research On Nonlinear Compensation Techniques Based On Optimized BP Neural Network In Visible Light Communication Systems Deep Learning

With the advent of 5G mobile communication era and the rapid development of artificial intelligence technology,the spectrum resources of wireless RF communication are in a shortage.Visible light communication(VLC)has attracted the attention of scholars due to its abundant spectrum resources,ultra-high propagation speed and low cost.At the same time,the potential reuse factor of existing light-emitting diode(LED)lighting infrastructure has added a great attraction to the system.However,in visible light communication systems,the signal transmission is affected by the nonlinear characteristics of LEDs,and the high peak-to-average power ratio of the OFDM modulated signal can also exacerbate signal distortion and seriously affect system performance.Therefore,it is particularly urgent to study the signal nonlinear distortion compensation technology.Based on the above background,this thesis firstly briefly introduces the commonly used VLC system model,the nonlinear model of LED and two common neural network models,and simulates and analyzes the signal distortion phenomenon caused by LED nonlinearity.Secondly,focusing on the key problem of signal distortion caused by LED nonlinearity,the thesis focuses on the distortion compensation of the signal at the receiving end using optimized BP techniques,and the main innovation points are as follows:(1)A post-distortion compensation algorithm based on a hybrid algorithm of Genetic Algorithm(GA)and Particle Swarm Optimization(PSO)combined with a multilayer Back Propagation(BP)neural network is studied.The main idea is to feed the received signal of the system into the network and take the original transmit signal as the desired output,pre-optimize the initial weights of the BP network by the hybrid algorithm of genetic algorithm and particle swarm,and then estimate the transmit signal with the help of the trained network to achieve distortion suppression.On the one hand,the powerful self-learning ability of the neural network can characterize the nonlinear relationship between the input and output,and then achieve the compensation at the receiver side;on the other hand,the excellent optimization-seeking ability of the hybrid algorithm can obtain the optimal initial weights of the BP network while avoiding falling into the local optimum.(2)To address the specificity problem of the two swarm intelligence algorithms,two improvement schemes are proposed.One,the particle swarm algorithm adopts nonlinear inertia weights characterized by the cosine function,and the genetic algorithm successively introduces adaptive crossover and variation probabilities;second,on the basis of the improved algorithm I,the particle asynchronous update method is adopted so as to achieve maximum exploration.The simulation results show that the improved algorithm can achieve better BER performance,which confirms the effectiveness of the algorithm.(3)Another hybrid algorithm based on Grey Wolf Optimization(GWO)and particle swarm is investigated to optimize the distortion compensation algorithm of BP networks.Compared with the genetic algorithm,the Grey Wolf algorithm relies on fewer parameters,has a simple model,and has a lower algorithm complexity.Despite its many advantages,it is difficult to avoid the common problems of most intelligent optimization algorithms,such as,easy early convergence,slow convergence speed and low accuracy.Therefore,logistic chaotic mapping is introduced to ensure that the initialization positions are dispersed as evenly as possible to improve the accuracy;meanwhile,catfish effect is introduced to ensure the population dynamics and prevent the algorithm from maturing prematurely.The results show that the improved optimization algorithm has some improvement in the system performance.