应用深度学习的信号解调

摘要

针对噪声干扰信道下的信号解调问题,提出了应用深度学习的信号识别方法,通过识别信号完成信号解调.深层置信网络使用受限波尔兹曼机为基本单元,设计针对通信信号识别的多层深层置信网络.通信信号首先变换为特定表征序列,并以此构建完备的训练集合对深度置信网络进行逐层的无监督学习和全局有监督的微调反馈学习,在深层置信网络的权重参数优化过程中实现对通信信号的特征提取与识别.仿真实验表明,与传统调制信号解调方法相比,应用深度学习的信号解调方法的检测性能有约0.4 dB的提升.%This paper proposes a deep learning based demodulation method by identifying the modulated signal in radio channel.The proposed deep belief network is composed of multilayer restricted Boltzmann machines.The communication signal is transformed into a new form,which is used as the input of the deep belief network and system training.The deep belief network extracts the characteristics of communication signals by top-down depth learning and bottom-up feedback fine tuning.The algorithm's practicability is verified by simulation.The simulation also indicates that the bit error rate(BER) of the proposed method is almost 0.4 dB better than that of traditional one.