基扩展模型下基于深度学习的双选信道估计方法

Channel Estimation over Doubly Selective Channel Based on Deep Learning under Basis Expansion Model

摘要:

针对OFDM系统在高速移动环境下信道的双选衰落和非平稳特性导致下行链路信道估计性能受限的问题，提出一种基于基扩展模型(basis expansion model，BEM)并结合深度学习(deep learning，DL)的信道估计方法。针对高速移动环境信道的双选衰落特性，使用BEM对信道进行建模，把估计大量的信道冲激响应转变为估计少量的基系数，减少了待估参数，有效降低了估计复杂度；针对高速移动环境信道非平稳特性，建立了深度神经网络，通过离线训练使其学习到双选信道的变化特征，提高了信道估计的准确度。仿真结果表明，在高速移动环境下，与传统的方法相比，所提信道估计方法，性能提升明显。

关键词:基扩展模型;双选信道;深度学习;压缩感知

Abstract:

Aiming at the problem that the doubly-selective fading and non-stationary characteristics of the Orthogonal Frequency Division Multiplexing (OFDM) system in the high-speed mobile environment lead to the limited performance of the downlink channel estimation, a basis extension model (BEM) combined with deep learning (DL) channel estimation method. Aiming at the doubly-selective fading characteristics of the channel in the high-speed mobile environment, BEM is used to model the channel, which converts the estimation of a large number of channel impulse responses into a small number of basis coefficients, which reduces the parameters to be estimated and effectively reduces the complexity of channel estimation. Aiming at the non-stationary characteristics of the channel in the highly mobile environment, a deep neural network is established, and the change characteristics of the doubly-selective channel are learned through offline training, which improves the accuracy of channel estimation. The simulation results show that in the highspeed mobile environment, com-pared with the traditional methods, the proposed channel estimation method has significant perfor-mance improvement.

Key words:Basic Expansion Model; Doubly Selective Channel; Deep Learning; Compressive Sensing