CPGD: Cadzow Plug-and-Play Gradient Descent for Generalised FRI

被引:24
作者
Simeoni, Matthieu [2 ,3 ]
Besson, Adrien [1 ]
Hurley, Paul [4 ]
Vetterli, Martin [2 ]
机构
[1] E Scop, F-13760 St Cannat, France
[2] Ecole Polytech Fed Lausanne EPFL, Lab Commun Audiovisuelles LCAV, CH-1015 Lausanne, Switzerland
[3] Ibm Res Lab Zurich, Fdn Cognit Solut Grp, Zurich, Switzerland
[4] Western Sydney Univ WSU, Ctr Res Data Sci & Math, Parramatta, NSW 2150, Australia
基金
瑞士国家科学基金会;
关键词
Technological innovation; Noise reduction; Optimization; Signal processing algorithms; Mathematical model; Convolution; Convergence; Finite rate of innovation; non bandlimited sampling; Dirac streams; non convex optimisation; Cadzow denoising; proximal gradient descent; alternating projections; APPROXIMATION; CONVERGENCE; ALGORITHM; CONVEX;
D O I
10.1109/TSP.2020.3041089
中图分类号
TM [电工技术]; TN [电子技术、通信技术];
学科分类号
0808 ; 0809 ;
摘要
Finite rate of innovation (FRI) is a powerful reconstruction framework enabling the recovery of sparse Dirac streams from uniform low-pass filtered samples. An extension of this framework, called generalised FRI (genFRI), has been recently proposed for handling cases with arbitrary linear measurement models. In this context, signal reconstruction amounts to solving a joint constrained optimisation problem, yielding estimates of both the Fourier series coefficients of the Dirac stream and its so-called annihilating filter, involved in the regularisation term. This optimisation problem is however highly non convex and non linear in the data. Moreover, the proposed numerical solver is computationally intensive and without convergence guarantee. In this work, we propose an implicit formulation of the genFRI problem. To this end, we leverage a novel regularisation term which does not depend explicitly on the unknown annihilating filter yet enforces sufficient structure in the solution for stable recovery. The resulting optimisation problem is still non convex, but simpler since linear in the data and with less unknowns. We solve it by means of a provably convergent proximal gradient descent (PGD) method. Since the proximal step does not admit a simple closed-form expression, we propose an inexact PGD method, coined Cadzow plug-and-play gradient descent (CPGD). The latter approximates the proximal steps by means of Cadzow denoising, a well-known denoising algorithm in FRI. We provide local fixed-point convergence guarantees for CPGD. Through extensive numerical simulations, we demonstrate the superiority of CPGD against the state-of-the-art in the case of non uniform time samples.
引用
收藏
页码:42 / 57
页数:16
相关论文
共 53 条
[1]  
Andersson F., 2011, PROC PROJECT REV, V1, P37
[2]  
[Anonymous], 2014, Foundations of signal processing
[3]  
[Anonymous], 2012, MATRIX COMPUTATIONS
[4]  
[Anonymous], 2017, ARXIV171102151
[5]   A Fast Iterative Shrinkage-Thresholding Algorithm for Linear Inverse Problems [J].
Beck, Amir ;
Teboulle, Marc .
SIAM JOURNAL ON IMAGING SCIENCES, 2009, 2 (01) :183-202
[6]  
Bezzam E., 2018, PROC IEEE INT ULTRAS, P1
[7]  
Blu T, 2002, 2002 IEEE INTERNATIONAL SYMPOSIUM ON BIOMEDICAL IMAGING, PROCEEDINGS, P777
[8]   Sparse sampling of signal innovations [J].
Blu, Thierry ;
Dragotti, Pier-Luigi ;
Vetterli, Martin ;
Marziliano, Pina ;
Coulot, Lionel .
IEEE SIGNAL PROCESSING MAGAZINE, 2008, 25 (02) :31-40
[9]   SIGNAL ENHANCEMENT - A COMPOSITE PROPERTY MAPPING ALGORITHM [J].
CADZOW, JA .
IEEE TRANSACTIONS ON ACOUSTICS SPEECH AND SIGNAL PROCESSING, 1988, 36 (01) :49-62
[10]   Convergence of the alternating minimization algorithm for blind deconvolution [J].
Chan, TF ;
Wong, CK .
LINEAR ALGEBRA AND ITS APPLICATIONS, 2000, 316 (1-3) :259-285