论文部分内容阅读
针对大多数现有盲分离算法都假定信源数已知,分析了信源数未知且动态变化时的超定盲分离(接收天线数多于信源数)算法,通过奇异值分解在线估计信源个数,并修正自适应神经网络的维数。文章在经典Cichocki-Unbenauen算法中加入动量项,设置可变步长,使分离性能得到较大的改进,同时避免训练陷入局部最小。与文献[1]新提出的ANA算法相比,仿真结果证明本文算法具有更快的收敛速度和更小的稳态误差。
Most existing algorithms for blind separation assume that the number of sources is known, and the algorithm of overdetermined blind separation (the number of receiving antennas is more than the number of sources) when the number of sources is unknown and dynamically changes is analyzed. By using the singular value decomposition The number of sources, and correct the dimension of adaptive neural network. The article adds the momentum term to the classical Cichocki-Unbenauen algorithm, sets the variable step size, greatly improves the separation performance, and avoids the training getting into the local minimum. Compared with the new ANA algorithm proposed in [1], the simulation results show that the proposed algorithm has faster convergence rate and smaller steady-state error.