为了有效地去除信号中的宽带噪声,提出了一种基于自适应稀疏表示的宽带噪声去除算法。根据噪声成分与信号特征成分之间的不相关或弱相关特点,自适应地确定稀疏分解的终止条件,实现信号的稀疏表示。降噪过程中使用染噪信号构造学习样本,由信号的自适应稀疏表示和原子库的更新迭代实现原子库的训练。染噪信号在训练后的原子库上进行自适应稀疏表示,实现信号的噪声去除。仿真信号和齿轮振动信号的降噪试验表明:该方法具有比小波阈值降噪、匹配追踪降噪方法更好的降噪性能,能够有效地去除信号中的宽带噪声。
In order to remove wideband noise from signal, a denoising method based on adaptive sparse representation was proposed. According to the independence or weak correlation between noise and signal feature components, the termination condition of sparse decomposition is adaptively determined, and then the sparse representation of the sig- nal is achieved. The method trains the initialized dictionary based on learning samples constructed from noised sig- nal. The training process is completed by an iteration algorithm, which alternates between adaptive sparse represen- tation and dictionary update. Based on the trained dictionary, noise reduction is conducted via adaptive sparse repre- sentation of the noised signal. Experiment results of simulated data and vibration signals of a gear show that the pro- posed method is better than wavelet denoising and matching pursuit denoising. It could effectively remove wideband noise from signal.