CAJ | 학술논문

针对基于传统梯度下降算法的前馈神经网络盲均衡在输入线性相关条件下收敛速度慢的问题,提出了一种修正解相关前馈神经网络盲均衡改进算法。对接收观测数据进行解相关处理,使梯度方向保持正交,同时,设定判断阈值,如果前馈神经网络输入相关系数大于阈值,说明输入向量强相关,保持梯度更新大小和方向不变,以克服强相关输入条件下解相关算法收敛停滞的问题。计算机仿真结果表明,文中提出的算法与基于直接梯度下降算法和传统解相关前馈神经网络盲均衡算法相比具有更快的收敛速度,有效提高了均衡性能。
침대기우전통제도하강산법적전궤신경망락맹균형재수입선성상관조건하수렴속도만적문제,제출료일충수정해상관전궤신경망락맹균형개진산법。대접수관측수거진행해상관처리,사제도방향보지정교,동시,설정판단역치,여과전궤신경망락수입상관계수대우역치,설명수입향량강상관,보지제도경신대소화방향불변,이극복강상관수입조건하해상관산법수렴정체적문제。계산궤방진결과표명,문중제출적산법여기우직접제도하강산법화전통해상관전궤신경망락맹균형산법상비구유경쾌적수렴속도,유효제고료균형성능。
To solve the problem of slow convergence rate in blind equalization by feedforward neural network(FNN),a modified decorrelation algorithm is proposed and combined with FNN to implement blind equalization.In the algorithm the gradient vector can keep orthogonally by decorrelating to the input signals.Meanwhile,we set a threshold value to the correlate coefficient.If the correlate coefficient of input signals of FNN is bigger than threshold value,then the updating gradient value is kept for overcoming the stagnate of decorrelation algorithm for input signals.Computer simulation shows the effectiveness of the method proposed in this paper.