Abstract:We consider the problem of recovering two unknown vectors, $\boldsymbol{w}$ and $\boldsymbol{x}$, of length $L$ from their circular convolution. We make the structural assumption that the two vectors are members of known subspaces, one with dimension $N$ and the other with dimension $K$. Although the observed convolution is nonlinear in both $\boldsymbol{w}$ and $\boldsymbol{x}$, it is linear in the rank-1 matrix formed by their outer product $\boldsymbol{w}\boldsymbol{x}^*$. This observation allows us to recast the deconvolution problem as low-rank matrix recovery problem from linear measurements, whose natural convex relaxation is a nuclear norm minimization program. We prove the effectiveness of this relaxation by showing that for "generic" signals, the program can deconvolve $\boldsymbol{w}$ and $\boldsymbol{x}$ exactly when the maximum of $N$ and $K$ is almost on the order of $L$. That is, we show that if $\boldsymbol{x}$ is drawn from a random subspace of dimension $N$, and $\boldsymbol{w}$ is a vector in a subspace of dimension $K$ whose basis vectors are "spread out" in the frequency domain, then nuclear norm minimization recovers $\boldsymbol{w}\boldsymbol{x}^*$ without error. We discuss this result in the context of blind channel estimation in communications. If we have a message of length $N$ which we code using a random $L\times N$ coding matrix, and the encoded message travels through an unknown linear time-invariant channel of maximum length $K$, then the receiver can recover both the channel response and the message when $L\gtrsim N+K$, to within constant and log factors.

A geometrical investigation on the generalized lp/lq norm for blind deconvolution.

Lp (1≤p≤∞) BLIND ADAPTIVE DECONVOLUTION AND ITS APPLICATIONS

Multi-frame Blind Deconvolution Using Sparse Priors

L∞ Blind Deconvolution for the Generalized AR Model.

Blind Deconvolution Using an Improved L0 Sparse Representation

Identifiability of Multichannel Blind Deconvolution and Nonconvex Regularization Algorithm

Nonlinear Blind Deconvolution Based on Generalized Normalized l p / l q Norm for Early Fault Detection

Non-blind image deconvolution using natural image gradient prior

Restricted $Q$-Isometry Properties Adapted to Frames for Nonconvex $L_q$-Analysis

On the Schatten P-Quasi-norm Minimization for Low-Rank Matrix Recovery

Generalized variational framework with minimax optimization for parametric blind deconvolution

Blind Image Deblurring Using Spectral Properties of Convolution Operators

Blind Deconvolution Technique Based on Improved Correlated Generalized Lp / Lq Norm for Extracting Repetitive Transient Feature

A L-0 Sparse Analysis Prior for Blind Poissonian Image Deconvolution

Blind Image Deblurring by Spectral Properties of Convolution Operators

Sparse Blind Deconvolution with Nonconvex Optimization for Ultrasonic NDT Application

Rapid, Robust, and Reliable Blind Deconvolution Via Nonconvex Optimization

Blind Deconvolution using Convex Programming

Blind Deconvolution on Graphs: Exact and Stable Recovery

Fundamental Limits of Blind Deconvolution Part II: Sparsity-Ambiguity Trade-offs

Blind Image Deconvolution Assisted By Robust Non-Gaussian Priors