基于因子图的同频数字混合信号单通道盲分离

摘要

To solve the single channel blind source separation (BSS) problem of co-frequency digital modulated mixtures with complex codes, the unified frame of the separation algorithm is given under factor graph. By combining sum-product (SP) algorithm with ECM algorithm organically, a novel code-aided single channel blind source separation algorithm based on factor graph is proposed. The expressions to estimate all the synchronization parameters are derived under maximum likelihood law. The proper factor graph model is established for the mixture with inter-symbol interferences, using the generalized distribution law, and the posteriori probability estimation of the symbols is given. The simulation results show that, the new algorithm can not only gain almost equivalent performance towards the mixture using convolutional codes, but effectively estimate the parameters and sequence of the mixtures using complex codes such as LDPC.%本文针对复杂编码下同频数字调制混合信号单通道盲分离问题,构建了利用因子图设计盲分离算法的统一框架.通过将和积消息传递算法与ECM参数估计算法有机地结合,提出一种基于因子图的编码辅助同频混合信号单通道盲分离算法.确定了在最大似然准则下混合信号同步参数的估计式.利用广义分配率准则,为存在码间干扰的混合信号建立了合理的因子图模型,得到了符号序列的后验概率估计式.仿真结果表明,本文提出的算法不仅能够对卷积码混合信号取得较好的分离性能,还能有效实现对LDPC码等复杂编码混合信号的参数和序列的联合估计.