CPM信号相干解调的神经网络方法研究

本文研究连续相位调制信号相干解调的神经网络方法，提出了基于判决反馈预处理和变换域特征提取的射基函数网络解调方案。通过判决反馈预处理可以有效地减少网络输入信号样本个数，使其具有更好的可分性。利用高采样率下ＣＰＭ信号的相关性，引入变换域的处理方法可以大幅度地降低网络输入信号和样本空间维数。采用射基函数网络作为判决分类器，不仅可以逼近最大似然解调性能，而且无需加噪训练。模拟结果表明：本文提出的方案在降低训练及实现复杂度的同时具有接近最优的误比特性能

A Study of Neural Network Based Coherent Demodulation for CPM Signals

This paper investigates neural network coherent demodulation for continuousphase modulation A radial basis function network demodulation scheme is proposed with decision feedback pre processing and feature extraction in frequency domains Through decision feedback preprocessing,the size of the signal sample set can be reduced and the signal patterns have better separableness Considering the excellent relativity of CPM signals with high sampling rates,KLT and DCT are employed to reduce the signal dimension Byusing RBF network as classifier,maximum likelihood demodulation performance can be approximated,and network training with noise isn't required Despite of the low complexity in training and implementation,Computer simulation results show that the scheme can achieve nearly optimal demodulation performance.