Research On Visible Light Transmission Technology Based On Neural Networks

In recent years,data services and wireless applications have grown rapidly,and spectrum resources have become scarcer.Visible light communication is an important supplement to radio frequency communication,and has a broad application prospect.Indoor visible light communication can use ubiquitous LED to real-ize lighting and communication at the same time.It has several advantages such as no conflict with radio frequency bandwidth,no authorization required,energy saving,environmental protection,and good con-fidentiality.In recent years,it has become a research hotspot.Although the potential bandwidth of visible light communication may be large,the poor characteristics of LED devices,such as memory nonlinearity of LEDs and multicolor crosstalk of nonlinear multicolor LEDs,severely limit the performance of visible light communication systems.This article mainly introduces the neural network structure based on the traditional scheme to mitigate the impact of these device defects on the performance of the VLC system,and provides new solutions for the design of VLC transceivers.Firstly,from the two directions of Volterra model and neural network model,the pre-distortion compen-sation technology for memory nonlinearity of LED channel is studied.The memory nonlinearity of the LED channel and the characteristic of the limited dynamic range of the LED is explained.The specific mathemat-ical model is used to characterize the nonlinearity of the LED memory.Predistortion schemes are used to suppress the memory nonlinearity of the LED channel from two directions:(1)Based on the Volterra model,first use least squares to estimate the kernel coefficients,and then explore the feasible schemes to reduce the complexity of the predistortion model system,Based on truncation and compressed sensing theory respec-tively,a better trade-off is achieved between system performance and system complexity.(2)The structure of the predistortion model based on neural network is proposed,and the parameter training process using sliding windows is explained.The predistortion model using a neural network structure reduces the exponential level of system complexity to the square level.For a nonlinear system with a memory depth of Q,the complexity of the predistortion module is O(Q²).Simulation experiments have been carried out for different models.The simulation experiment is carried out by first using the data collected using the actual visible light com-munication platform to model the LED,and using the trained model as the ideal model for offline simulation experiments.Through experiments,the effectiveness of the pre-distortion scheme proposed in this paper on the non-linearity of LED memory is verified.Next,a type of feed-forward neural network called auto-encoder is used to build an end-to-end multicolor visible light communication transceiver.The basic principles of the multicolor VLC system and self-encoder are explained,and then the causes and mathematical expression of the multicolor crosstalk matrix are dis-cussed by mathematically modeling the multicolor VLC channel.In addition to meeting communication requirements,multicolor VLC systems also need to meet lighting constraints and dynamic range constraints.Aiming at two cases of fixed channel of visible light and variable channel of incident angle,two multicolor VLC systems based on self-encoder were designed.The encoder and decoder of the autoencoder are used to realize the mapping between transmission symbols and color space constellation points.The LED output meets the optical power constraints and dynamic range constraints through the optical power control module.Simulation experiments for these two schemes respectively verify that under the same noise power condition,the multicolor VLC transceiver based on the autoencoder has better symbol error rate performance than the traditional scheme.Finally,the DCO-OFDM system based on self-encoder was studied,from the perspective of changing the OFDM symbol waveform to reduce the linearity requirements of the LED,and to reduce the impact of nonlinear distortion on the OFDM system due to clipping.DCO-OFDM frequency-domain subcarrier symbol constellation points are mapped and demapped using encoder and decoder respectively.The combination of the mean square error of the input and output symbols and the original cubic metric of the time domain OFDM symbol is used as the loss function.Under the premise that the output symbol distortion is within an acceptable range,the cubic metric of DCO-OFDM time-domain symbols is reduced.The simulated channel environment uses the measured frequency selected LED channel.Through experimental verification,when the complementary accumulation function of RCM is 10^-3,RCM is reduced by about 11.2d B,which can decrease the linearity requirements of LED to a certain extent.