非合作高阶调制卫星同频混合信号单通道盲分离技术

Non-cooperative high-order modulated single-channel co-frequency mixed signal blind separation research work includes three major aspects, i.e. blind separation performance, parameter estimation algorithm and blind separation algorithm. Firstly, aimed at the single-channel co-frequency mixed signals blind separation, based on the maximum likelihood criterion of dual-signal joint sequence detection, this paper obtain the lower bound of the blind separation by removing the correlation between error symbols in Forney performance boundary algorithm. A joint blind parameters estimation algorithm based on minimum channel stability is taken into account, which is for non-cooperative receiving high order modulated single-channel co-frequency mixed signals. Based on the derivation of the channel stability function, the hierarchical searching method is used to estimate the parameters. Aiming at the high-order modulated of the same frequency mixed signal, a low complexity blind separation is proposed by using the set segmentation algorithm. A blind separation algorithm is taken into account for single-channel co-frequency mixed signals with different symbol rates based on Double Grid Per-Survivor Processing(DG-PSP). The channel states and input signals components are treated as two dynamic grids, receiving mixed-signal reconstructed by respectively iterative updating the two groups of grid status, thus achieving blind signal separation. On this basis, the neural network structure of single channel mixed frequency signal is proposed to study the mixed signals neural network learning separation algorithm with low complexity.

非合作高阶调制单通道同频混合信号盲分离研究工作包括三大方面，即盲分离性能界、参数估计算法和盲分离算法。首先从双信号联合序列检测的最大似然准则出发，针对单通道同频混合信号，拟去除Forney性能界算法错误符号间相关性，得到单通道同频混合信号分离紧致性能下界；针对非合作接收高阶调制单通道同频混合信号，拟利用基于最小信道稳定度的参数联合盲估计算法，在推导信道稳定度函数的基础上，采用分级搜索方法估计信道参数；针对高阶调制同频混合信号，拟利用集分割算法实现低复杂度盲分离；针对非合作接收不同符号速率单通道同频混合信号，拟采用基于双网格逐幸存路径处理的分离算法，将两路信号分量的信道状态与输入分别视为两组动态网格，通过分别迭代更新两组网格状态实现混合信号重构，进而实现单通道同频混合信号盲分离；在此基础上，拟搭建分离单通道同频混合信号的神经网络结构，研究混合信号分离低复杂度神经网络学习算法。

鉴于成对载波多址（Paired Carrier Multiple Access，PCMA）通信体制在信道容量以及安全性上的优势，其相关研究已成为热点。课题针对非合作单通道PCMA信号低复杂度分离课题展开系统性研究，主要包括分离性能界推导、高性能参数估计算法、不同符号速率非合作PCMA信号分离算法以及相同符号速率非合作PCMA信号低复杂度分离算法等方面。主要成果及创新点如下：.1. 针对PCMA混合信号，从发送信号模型出发构造传输比特与混合信号符号的空间映射，利用最大似然准则推导具有更广泛适用性的分离性能下界表达式，数值计算结果与参数已知情况下维特比算法仿真结果吻合，验证了所推导性能界的合理性。.2. 针对非合作接收PCMA信号，提出一种基于最小信道稳定度的参数联合估计算法。在推导信道稳定度函数的基础上，采用分级搜索方法估计时延、幅度以及频偏，并给出存在有限集模糊度的相偏估计，最后借助参数分级迭代处理结构提升算法性能，并通过改进算法降低了运算复杂度。.3. 针对非合作接收不同符号速率PCMA混合信号，提出一种基于双网格逐幸存路径处理（DG-PSP, double grid per-survivor processing）的分离算法。所提算法复杂度与传统PSP算法相当，同时仿真结果表明，算法性能逼近联合界。.4. 针对当前非合作接收PCMA混合信号分离的高复杂度束缚，提出一种联合格基规约（Lattice Reduction, LR）与连续干扰抵消(Serial Interference Cancellation, SIC)的迭代分离算法。仿真实验表明，算法具有拟合分离联合界的性能，QPSK调制时LR-SIC算法复杂度比PSP算法复杂度降低了90%，证明算法合理性与有效性。.5. 针对非合作接收PCMA混合信号分离问题，提出一种基于前馈神经网络的分离算法,通过搭建神经网络分离平台，实现PCMA混合信号低复杂度高性能盲分离。仿真实验表明，神经网络能够极大挖掘信号内在信息，针对QPSK调制PCMA混合信号，达到信噪比13dB时10-2数量级误比特率分离性能，并伴随着较相同误比特率性能的PSP算法指数级下降的复杂度，实验条件下约为PSP算法复杂度的1/10。