基于最大相关熵的多凸组合滤波器

Multi-convex combined filter based on maximum correntropy criterion

相关熵被广泛地应用于非高斯信号处理中,但是也不可避免步长与失调量之间的矛盾。为了解决这个矛盾,基于最大相关熵的凸组合(Convex combination of Maximum Correntropy Criterion,CMCC)滤波器被提出,以同时获得大步长滤波器的快收敛速度和小步长滤波器的低失调量。但是,当两个步长不同的滤波器进行凸组合时,由于步长选取的差异,会导致组合滤波器的收敛速度和追踪最优值的能力下降。因此,提出了基于最大相关熵的多凸组合滤波器(Multi-convex Combination of Maximum Correntropy Criterion, MCMCC)。MCMCC算法能够提供数量更多不同步长的自适应滤波器,使得组合滤波器可以灵活地调节多个步长的权重占比,从而具有更好的追踪能力。仿真结果表明,在针对混合高斯噪声和突变的系统辨识中,MCMCC算法相比CMCC算法具有更快的收敛速度、再收敛能力和追踪能力。

Correntropy based algorithms are widely used in non-Gaussian signal processing,but they also suffer from the conflict between the step size and the misadjustment.In order to solve this problem,a convex combination filter based on maximum correntropy criterion(CMCC)was proposed to obtain the fast convergence speed of the filter with large step size as well as the low misadjustment of the filter with small step size.However,the convex combination of two filters with different step sizes will result in the penalties in terms of the combined filter′s convergence speed and the ability to track the optimal value.In this paper,a multi-convex combination filter based on maximum correntropy criterion(MCMCC)is proposed to provide more adaptive filters with different step sizes,so that the weight ratio can be flexibly adjusted for more step sizes,and thus having better tracking ability.Simulation results show that compared with the CMCC algorithm,the proposed MCMCC algorithm has faster convergence speed,stronger re-convergence performance and better tracking ability in the system identification for in the presence of mixed Gaussian noise and abrupt change.