基于深度学习的OFDM系统信号检测设计

Signal detection design of OFDM system based on deep learning

摘要:

传统信号检测技术在通信系统、雷达信号处理、生物医学信号处理等方面有着广泛的应用。正交频分复用系统中传统信号检测技术存在信道失真、载波间干扰、码间干扰等问题。为解决此问题，提出一种基于深度学习的OFDM系统信号检测方法，实验中根据信道统计仿真的结果数据离线训练深度学习模型，利用该模型恢复在线传输的数据。通过与传统算法之间的对比，实验展示了在OFDM系统中使用深度学习方法进行信道估计和符号检测的成果。仿真实验结果表明，在训练导频较少、循环前缀省略和非线性削波噪声的条件下，基于深度学习方法的信号检测比传统方法的鲁棒性更强。该方法可以应用在大部分信道失真和干扰的无线通信系统中，具有较强的实用价值。

Abstract:

Traditional signal detection techniques have a wide range of applications in communication systems, radar signal processing, biomedical signal processing, and so on. The traditional signal detection techniques in orthogonal frequency division multiplexing (OFDM) systems have problems such as channel distortion, inter-carrier interference, and inter-code interference. In order to solve this problem, a deep learning based signal detection method for OFDM system is proposed, and in the experiment, a deep learning model is trained offline based on the resultant data from the channel statistical simulation, and the model is utilized to recover the data transmitted online. Through the comparison between the traditional algorithms, the experiment demonstrates the results of using deep learning methods for channel estimation and symbol detection in OFDM systems. The results of simulation experiments show that the signal detection based on the deep learning method is more robust than the traditional method under the conditions of fewer training guides, cyclic prefix omission and nonlinear clipping noise. The method can be applied in most wireless communication systems with channel distortion and interference, and has strong practical value.