基于室内可见光通信系统的一种高斯核辅助的深度神经网络信道均衡方案

In this paper, we propose a GKI-DNN (Gaussian Kerne Improved-Deep Neural Network) scheme for channel equalization in indoor visible light ACO-OFDM system. The scheme use the method of deep learning implicitly estimate the channel state information (CSI), use the simulation data generated by channel statistics to offline train network model, the model is directly used to recover the online transmit data, in order to accelerate the speed of the network training, improve the system ber performance, put a gaussian kernel function layer between the network\'s input layer and hidden layer. The simulation results show it makes the network number of iterations shrank by 44.44%. After parameters debugging, GKI-DNN network improved the channel phase deflection and amplitude distortion, constellation diagram of receiving signals are concentrated. When each frame using the 64 pilot, training network model\'s error performance contrast to the traditional LS equalizer increased 10 db, Compared with MMSE equalization, the bit error performance is improved 5dB, The bit error performance is obviously improved.