基于高阶累积量的自适应IIR线谱增强器

将四阶累积量的递推算法推广到求正弦信号的四阶混合累积量，并提出了一种基于四阶混合累积量切片的自适应IIR结构的线谱增强器算法。该线谱增强器通过自适应前向线性预测法求得滤波器的系数，从而使带噪信号的线谱成份得到增强。仿真结果表明，在平稳或非平稳输入信号情况下，该算法从色噪声中提取正弦信号的能力优于基于二阶自相关函数的自适应IIR线谱增强器。

Designing an Adaptive Line Enhancer for Processing Sinusoidal Signals Embedded in Colored Noise

We propose an adaptive IIR(Infinite Impulsive Response) notch filter based on a fourth-order mixed cumulant slice to suppress additive colored noise, thus enhancing the desired spectral peaks due to the sinusoids. The IIR filter coefficients are estimated by applying the adaptive forward linear prediction method to a fourth-order mixed cumulant slice of the input signals that contain highly colored noise. In the full paper, we explain in some detail how to design a Fourth-Order Mixed Cumulant-Based Linear Prediction Adaptive Line Enhancer(FOMCBLPALE). We performed simulations and compared the results obtained with FOMCBLPALE algorithm designed by us with those obtained with an existing algorithm which we call Autocorrelation-Based Linear Prediction ALE(ACBLPALE) algorithm. In the full paper, our comparison included three kinds of highly colored noise. Here we give only the comparison for Gaussian colored noise. In our simulations, we used three different types of input signals: (1) signal fixed at 300 Hz; (2) signal including two parts: one part fixed at 200 Hz and the other part fixed at 300 Hz; (3) signal varying from 240 Hz to 300 Hz. For Gaussian colored noise and signal fixed at 300 Hz, FOMCBLPALE algorithm gives estimated frequency of 300.822 3 Hz, much better than the 329.0163 Hz estimated by ACBLPALE algorithm. For Gaussian colored noise and the other two types of signals, FOMCBLPALE algorithm also gives estimated frequencies much more accurate than those estimated by ACBLPALE algorithm.