基于联合稀疏表示和同时稀疏近似的并行坐标下降去噪算法

PARALLEL COORDINATE DESCENT DENOISING ALGORITHM BASED ON JOINT SPARASE REPRESENTATION AND SIMULTANEOUS SPARSE APPROXIMATION

针对并行坐标下降(Parallel Coordinate Descent,PCD)在音频信号去噪过程中的运行时间成本问题,构建一种新的时域处理框架,并在此基础上提出基于联合稀疏表示和同时稀疏近似的Joint-PCD算法。新的框架是将每个分割的音频帧作为一个列向量生成信号矩阵,利用超完备字典,Joint-PCD算法每执行一次是对一个音频信号(矩阵)而不仅仅是对一个音频帧(向量)实施去噪。仿真结果表明,Joint-PCD不仅具有与PCD相同的去噪性能,而且加快了算法的收敛。

Aiming at the running time cost of parallel coordinate descent(PCD)in audio signal denoising process,a joint-PCD algorithm based on simultaneous sparse approximation and joint sparse representation is proposed by constructing a new time domain processing framework.The new framework used each segmented audio frame as a column vector to generate a signal matrix.Using an over-complete dictionary,the joint-PCD denoised an audio signal(matrix)rather than just an audio frame(vector).Simulation results show that joint-PCD not only has the same denoising performance as PCD,but also accelerates the convergence of the algorithm.