The Wireless Channel Estimation Research Base On Deep Learning

As a common multi-carrier transmission technology,orthogonal frequency division multiplexing(OFDM)is widely used in wireless communication system because of its good spectrum selection and anti fading performance.Channel estimation is to estimate the channel model parameters according to the received data information,which is the key of OFDM communication system and largely determines the performance of wireless communication system.Traditional channel estimation algorithms usually need to predict the prior statistical characteristics of the channel,and fail to make full use of the channel characteristics information,so it is difficult to achieve better estimation performance;moreover,in practical application,due to the influence of environmental factors such as noise,poor robustness,so it is difficult to meet the actual demand.Deep learning technology has developed rapidly in video analysis,image processing and speech recognition due to its powerful learning ability,researchers in the field of communication system also apply deep learning technology to all aspects of wireless communication system.In this paper,by studying the current excellent deep-learning channel estimation methods,we first propose a channel estimation algorithm based on deconvolution network;secondly,in order to improve the performance of channel estimation,we further propose a channel estimation algorithm based on dilated convolution.The main contents of this paper are as follows:(1)For the channel interpolation problem,this paper innovatively introduces the image processing method,and models the channel estimation as the problem of image super-resolution reconstruction.Specifically,we regard the pilot channel state information calculated by the least square channel estimation as a low resolution image,and the complete channel information as a high resolution image to be reconstructed.A channel estimation method based on deconvolution network is proposed.Deconvolution network can make use of deconvolution operation to achieve better channel estimation with less network layers while achieving channel data size interpolation and amplification.Meanwhile,the channel estimation method based on deconvolution network can enhance the estimation accuracy of channel state information and reduce the computational complexity.The experiment results prove that the proposed the deconvolution network is able to fully utilize the channel characteristic information and improve the performance of channel estimation.(2)Aiming at the problem of the channel de-noising,we propose a channel estimation method based on dilated convolution network by combining the characteristics of wireless channel and the advantages of dilated convolution network,we further construct an dilated convolution network to suppress channel noise.The residual structure of the dilated convolution network can learn the mapping of the input channel image to the noise to suppress the noise,so as to improve the accuracy of channel estimation.At the same time,the channel estimation method based on the dilated convolution network can not only learn the nonlinear mapping of the channel transfer function,but also has good adaptability for different channel models.It can converge quickly according to different channel states information,and improve the adaptability and robustness of the algorithm.The experiment results show that the channel estimation method based on deconvolution and dilated convolution network is better for the reconstruction of complete channel state information.And the channel de-noising ability of dilated convolution network is better than other de-noising algorithms.