CAJ | 학술논문

在加性高斯白噪声的影响下，对于三阶多项式相位信号(CPS)，经典的字典学习算法，如K-means Singular Value Decomposition(K-SVD)，递归最小二乘字典学习算法(RLS-DLA)和K-means Singular Value Decomposition Denoising (K-SVDD)得到的学习字典，通过稀疏分解，不能有效去除信号的噪声。为此，该文提出了针对CPS去噪的字典学习算法。该算法首先利用 RLS-DLA 对的字典进行学习；其次采用非线性最小二乘(NLLS)法修改了该算法对字典更新的部分；最后对训练后的字典通过对信号的稀疏表示得到重构信号。对比其它的字典学习算法，该算法的信噪比(SNR)值明显高于其它算法，而均方误差(MSE)显著低于其它算法，具有明显的降噪效果。实验结果表明，采用该算法得到的字典通过稀疏分解，信号的平均信噪比比K-SVD, RLS-DLS 和K-SVDD高出9.55 dB ,13.94 dB和9.76 dB。
재가성고사백조성적영향하，대우삼계다항식상위신호(CPS)，경전적자전학습산법，여K-means Singular Value Decomposition(K-SVD)，체귀최소이승자전학습산법(RLS-DLA)화K-means Singular Value Decomposition Denoising (K-SVDD)득도적학습자전，통과희소분해，불능유효거제신호적조성。위차，해문제출료침대CPS거조적자전학습산법。해산법수선이용 RLS-DLA 대적자전진행학습；기차채용비선성최소이승(NLLS)법수개료해산법대자전경신적부분；최후대훈련후적자전통과대신호적희소표시득도중구신호。대비기타적자전학습산법，해산법적신조비(SNR)치명현고우기타산법，이균방오차(MSE)현저저우기타산법，구유명현적강조효과。실험결과표명，채용해산법득도적자전통과희소분해，신호적평균신조비비K-SVD, RLS-DLS 화K-SVDD고출9.55 dB ,13.94 dB화9.76 dB。
Under the influence of additive white Gaussian noise, the classical dectionary learning algorithms, such as K-means Singular Value Decomposition (K-SVD), Recursive Least Squares Dictionary Learning Algorithm (RLS-DLA) and K-means Singular Value Decomposition Denoising (K-SVDD), can not effectively remove the noise of Cubic Phase Signal (CPS). A novel dictionary learning algorithm for denoising CPS is proposed. Firstly,the dictionary is learned by using the RLS-DLA algorithm. Secondly,the update stage of the RLS-DLA algorithm is modified by using Non-Linear Least Squares (NLLS) in the algorithm. Finally, the signal is reconstructed via sparse representations over learned dictionary.Signal to Noise Ratio (SNR) obtained by using the novel dictionary learning algorithm is obviously higher than other algorithms,and the Mean Squares Error (MSE) obtained by using the novel dictionary learning algorithm is obviously lower than other algorithms. Therefore there is obviously denoising performance for using the dictionary learned by the algorithm to sparsely represent CPS. The experimental results show that the average SNR obtained by using the algorithm is 9.55 dB , 13.94 dB and 9.76 dB higher than K-SVD, RLS-DLS and K-SVDD.