相移键控信号的深度神经网络解调器  

Realization of SSAE demodulator of MPPSK communication system

作　　者：杨耀栋 吴乐南[1] 

机构地区：[1]东南大学信息科学与工程学院

出　　处：《电子测量与仪器学报》2018年第4期144-150,共7页Journal of Electronic Measurement and Instrumentation

基　　金：国家自然科学基金(61271204)资助项目

摘　　要：多元位置相移键控(MPPSK)信号具有高带宽利用率的特性,但是,带限信道下传统解调方案的解调性能大幅度下降。针对这个问题,提出了基于栈式稀疏自编码(SSAE)网络的码元判决方案。从接收信号提取特征信息和码元之间的码间干扰,并使用深度学习的思想对SSAE网络进行训练,使SSAE网络在信道环境恶劣的条件下对码元进行正确分类。此外,提出'多码元联合判决'方案并应用到SSAE网络的训练中,有效提高网络的解调性能。仿真结果表明,SSAE网络比传统方案的解调性能提高1~2个数量级,并且对信道环境的适应性更强。The multi-position phase shift keying( MPPSK) can transform much information occupying limited frequency spectrum. The traditional scheme can ’t demodulate the received signal effectively under limited channel condition. This paper proposes a new demodulation scheme based on stacked sparse auto-encoder neural network( SSAE). The SSAE network can extractfeatures and intersymbol interference fromreceived signal,thus it can judge the received signal correctly after training completely. Additionally,this paper applies a scheme named"multi-symbol joint judgement"to the SSAE network to improve the demodulation performance. The simulation result shows that SSAE network has understanding demodulation performance and stronger adaptability than traditional scheme.

关 键 词：多元位置相移键控 栈式稀疏自编码网络 深度学习 带限信道 联合判决 

分 类 号：TN763[电子电信—电路与系统]