CAJ | 학술논문

负熵是一种重要的非高斯性度量方法，最大化负熵使随机变量的非高斯性达到最大，从而使输出的各分量之间相互独立。负熵最大化算法以负熵作为目标函数，牛顿迭代法作为优化算法，针对牛顿迭代法中对初始值选择敏感的问题，用牛顿下山法代替牛顿迭代法，通过改变下山因子，使目标函数呈下降趋势，降低算法对初始值的依赖性。实验结果表明，改进后的算法在不同初始值下均能较好地分离语音音乐混合信号，改善了初值敏感问题。
부적시일충중요적비고사성도량방법，최대화부적사수궤변량적비고사성체도최대，종이사수출적각분량지간상호독립。부적최대화산법이부적작위목표함수，우돈질대법작위우화산법，침대우돈질대법중대초시치선택민감적문제，용우돈하산법대체우돈질대법，통과개변하산인자，사목표함수정하강추세，강저산법대초시치적의뢰성。실험결과표명，개진후적산법재불동초시치하균능교호지분리어음음악혼합신호，개선료초치민감문제。
Negative entropy is an important method of measuring non-gaussian. Each output component is independent of each other by maximizing the negentropy that makes the non-Gaussian maximum. Negentropy maximization takes negentropy as the objective function and Newton iteration method as the optimization algorithm. In order to solve the sensitivity problem of the initial value of Newton iteration, Newton downhill method is proposed instead of the original method. The Newton downhill reduces the dependence of the initial value by changing the downhill factor that makes the objective function on a declining trend. The simulation experiment results show that the proposed method can separate mixed signal of speech and music better under different initial values. Thus the Newton downhill method solves the initial value sensitivity problem effectively.