结合卷积神经网络的DFT-Like信道估计算法

DFT-Like Channel Estimation Algorithm Combining with Convolution Neural Network

针对快速移动环境中存在频率选择性衰落和导频开销急剧增加造成性能受限的问题,提出了一种基于复数神经网络的信道估计算法。在正交频分复用(Orthogonal Frequency Division Multiplexing,OFDM)系统中,提出残差网络直接处理复数信号,估计出信道的频域响应,并且设计了类似离散傅里叶变换(Discrete Fourier Transform,DFT)的卷积神经网络,通过训练学习转换波形,实现信道估计与均衡。仿真结果表明,所提方法的性能优于理想线性最小均方误差方法,加入循环前缀后性能接近理想信道估计方法,能够缓解码间干扰。

Aiming at the problems of limited performance caused by frequency domain selective fading and sharp increase in conduction overhead in the fast-moving environment,a channel estimation and equalization algorithm based on complex-valued neural network is proposed.The complex signal is processed directly through residual networks and the frequency domain response of the channel is recovered for OFDM(Orthogonal Frequency Division Multiplexing)systems.Moreover,DFT/IDFT-Like’s complexvalued convolution neural network is designed,and the channel estimation and equalization are realized by training and learning the waveform converting.Simulation results indicate that in the performance of the proposed method is better than that of the ideal linear minimum mean square error algorithm.In addition,the performance of the method with cyclic prefix almost achieves the ideal channel estimation method.The proposed channel estimation method can mitigate inter-symbol interference.