基于神经网络的OFDM信道补偿与信号检测

针对非线性失真和多径效应混合的复杂信道条件，提出一种基于神经网络的正交频分复用（OFDM）信道补偿与信号检测的方法。首先接收端信号利用最小二乘（LS）算法和迫零（ZF）算法做预处理，然后再输入到一层全链接层的神经网络进行进一步的信道补偿与信号检测，并恢复数据流。仿真结果表明，在没有进行输入信号功率回退（IBO）时，所提方法的误比特率（BER）性能比LS算法提升2个数量级，比线性最小均方误差（LMMSE）、最小均方误差（MMSE）提升一个数量级；在进行IBO后，所提方法能避免LS信道估计下至少4 dB的功率损失，能避免LMMSE、MMSE信道估计下至少2 dB的功率损失。所提方法在一定程度上验证了机器学习结合通信的先验知识的这种新的网络结构更能提升系统数据传输的准确率。 

Channel compensation and signal detection of OFDM based on neural network

A method for Orthogonal Frequency Division Multiplexing (OFDM) channel compensation and signal detection based on neural network is proposed for the complex channel conditions of nonlinear distortion and multi-path effects. First, the receiver uses the Least Squares (LS) and Zero Forcing (ZF) algorithm to preprocess the data, and then the processed data are input to neural network with only one fully connected layer for further channel compensation and signal detection, and finally the data flow is recovered. The simulation results show that, without Input Back-Off (IBO), the Bit Error Rate (BER) performance of the proposed method is two orders of magnitude higher than that of LS algorithm, and one order of magnitude higher than that of Linear Minimum Mean Square Error (LMMSE) and Minimum Mean Square Error (MMSE); with IBO, the proposed method can avoid at least 4 dB power loss under LS channel estimation and at least 2 dB power loss under LMMSE and MMSE channel estimation. To some extent, this paper verifies that the new network structure of machine learning combined with prior knowledge of communication can improve the accuracy of data transmission.