论文部分内容阅读
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
In this paper, the neural network method for coherent demodulation of continuous phase modulated signals is studied, and the demodulation scheme of the radix function network based on decision feedback preprocessing and transform domain feature extraction is proposed. Through the decision feedback preprocessing can effectively reduce the number of network input signal samples, making it better separability. By using the correlation of CPM signals at high sampling rate, the method of introducing transform domain can greatly reduce the dimension of network input signal and sample space. The use of a radix function network as a decision classifier can not only approximate the maximum likelihood demodulation performance, but also does not require noise training. Simulation results show that the scheme proposed in this paper has near-optimal error performance while reducing training and implementation complexity