Research On LED Nonlinear Equalization Technique For Visible Light Communication Based On Improved KNN Algorithm

Visible Light Communication(VLC),as a new mode of combining lighting and communication,is one of the current research hotspots at home and abroad because of its ability to achieve high capacity and ultra-high transmission rates.However,the nonlinearity of light emitting diodes(LEDs)greatly affects the system performance of VLC.To enhance the system capacity,the use of orthogonal frequency division multiplexing(OFDM)has been proposed.However,optical OFDM inherits the high peak-to-average power ratio(PAPR)characteristic of traditional OFDM,which makes visible light communication OFDM systems more sensitive to LED nonlinearity,severely impacting system performance.Therefore,equalizing the nonlinearity of LEDs in VLC is crucial for improving system performance.As a component of artificial intelligence,machine learning(ML)can fit various transfer functions and has the potential to alleviate nonlinearity.ML can significantly advance VLC research by improving signal processing through learning nonlinear mappings.The article begins by exploring the impact of LED nonlinearity distortion on the optical orthogonal frequency division multiplexing(O-OFDM)system for visible light communication.Additionally,it discusses the high complexity problem of the K-nearest neighbor(KNN)algorithm.To address these issues,the article proposes a cascade equalizer based on the improvement of KNN using rough set(RS)theory,called RS-KNN.Further research is then conducted on the limitations of the rough set theory,and a weighted variable precision rough set theory improved KNN(WVPRS-KNN)cascade equalizer is proposed to overcome these limitations.The specific research contents are as follows:1.To address the issue of LED nonlinearity distortion severely impacting the performance of the VLC system,the article proposes a cascade equalizer consisting of the rough set theory improvement of the KNN algorithm(RS-KNN)and the least mean square(LMS)algorithm.Firstly,based on the distribution law of the constellation points under different signal-to-noise ratios,the RS-KNN algorithm is proposed.This algorithm improves the KNN algorithm by dividing the training set data space into different regions and using a different classification strategy for each,thus reducing the complexity of the traditional KNN algorithm.Next,the LMS and RS-KNN cascade equalizer is proposed.In the first stage,the LMS equalizer reduces the dispersion of the sample points,while the RS-KNN equalizer in the second stage further improves the overall system performance.Finally,the Monte Carlo bit error rate simulation method is used to analyze the complexity of the algorithm and the suppression effect of LED nonlinearity in VLC.The results demonstrate that the RS-KNN algorithm achieves a complexity that is only 1/9 of the traditional KNN algorithm without sacrificing algorithm accuracy.Moreover,the RS-KNN cascade equalizer is the most effective in suppressing LED nonlinearity,outperforming the deep neural network(DNN)+LMS cascade equalizer,the traditional KNN single-stage equalizer,and the LMS single-stage equalizer.2.To address the problem that the traditional KNN algorithm suffers from the accuracy of classification results due to the uneven distribution of training samples,and the problem that the classical rough set model is very sensitive to noisy data,an equalizer of WVPRS-KNN cascaded with the LMS algorithm is proposed.Firstly,the weighted KNN algorithm(W-KNN)is proposed to weight the test points in the classification process based on the number of training set data and the Euclidean distance from the test points to the training set data points to alleviate the problem of poor classification accuracy caused by uneven distribution of training set data.Then,a weighted KNN algorithm improved by variable precision rough set theory is proposed to address the limitations of rough set theory for classification of controllable uncertainty classification points or error points.By introducing precision values to change the lower approximation region of the set to an arbitrary precision level,the complexity of the system at different precision of the variable precision rough set theory improved KNN algorithm(VPRSKNN)is discussed,and the best precision value is selected to make the complexity and classification accuracy of WVPRS-KNN optimal.Then,the LMS and WVPRS-KNN cascade equalization is proposed.Finally,Monte Carlo BER simulation is used to analyse the complexity of the algorithm and the suppression effect on LED non-linearity in VLC.The results show that the complexity of the WVPRS-KNN algorithm is further reduced compared to that of the RS-KNN algorithm,and the WVPRS-KNN cascade equaliser is more effective at high signal-to-noise ratios.