基于Sanger神经网络的TDDM-BOC信号组合码序列盲估计

针对低信噪比（SNR）下时分数据调制二进制偏移载波调制信号（TDDM-BOC）的组合码序列盲估计问题，提出一种基于Sanger神经网络（Sanger NN）的新方法。首先将已分段的信号作为输入信号并利用Sanger NN提取各主分量的权值向量；然后通过其多次输入反复训练权值向量，直至权值向量达到收敛；最终利用各个权值向量的符号函数重建信号的组合码序列，实现TDDM-BOC组合码序列的盲估计。此外，采用最优变步长的方法来提高收敛速度。理论分析和仿真实验表明，Sanger NN可以实现-20.9~0 dB信噪比下TDDM-BOC信号组合码序列的盲估计，且其复杂度明显低于传统奇异值分解（SVD）法和自适应特征提取的在线无监督学习神经网络（LEAP）；尽管Sanger NN收敛所需数据组数大于LEAP，但收敛时间明显少于LEAP算法。

Blind estimation of combination code sequence for TDDM-BOC based on Sanger neural network

Concerning the blind estimation of the combination code sequence of Time Division Data Modulation-Binary Offset Carrier (TDDM-BOC) modulation signal under low Signal-to-Noise Ratio (SNR), a new method based on Sanger Neural Network (Sanger NN), a kind of multi-principal component neural network, was proposed. Firstly, the segmented TDDM-BOC signal was used as input signal, and the weight vectors of multi-feature components of the segmented TDDM-BOC signal were adaptively extracted by Sanger NN algorithm. Secondly, the weight vectors were trained repeatedly until convergence by continuously inputing segmented TDDM-BOC signal. Finally, the combination signal code sequence was rebuilt by the symbolic function of each weight vector, thus realizing the blind estimation of the TDDM-BOC signal. Furthermore, an optimal variable step method was used in Sanger NN algorithm to greatly improve the convergence speed. Theoretical analysis and simulation results demonstrate that the Sanger NN algorithm can achieve blind estimation of the TDDM-BOC combined code sequence with low SNR of -20.9~0 dB, and its complexity is significantly lower than that of Singular Value Decomposition (SVD) and on-line unsupervised learning neural network for adaptive feature extraction via principal component analysis (LEAP). Although the number of data group required for the convergence of Sanger NN algorithm is larger than that of LEAP algorithm, but the convergence time of Sanger NN is lower than that of LEAP algorithm.