## TOEPLITZ-STRUCTURED COMPRESSED SENSING MATRICES

Citations: | 58 - 5 self |

### BibTeX

@MISC{Bajwa_toeplitz-structuredcompressed,

author = {Waheed U. Bajwa and Jarvis D. Haupt and Gil M. Raz and Stephen J. Wright and Robert D. Nowak},

title = {TOEPLITZ-STRUCTURED COMPRESSED SENSING MATRICES},

year = {}

}

### OpenURL

### Abstract

The problem of recovering a sparse signal x ∈ R n from a relatively small number of its observations of the form y = Ax ∈ R k, where A is a known matrix and k ≪ n, has recently received a lot of attention under the rubric of compressed sensing (CS) and has applications in many areas of signal processing such as data compression, image processing, dimensionality reduction, etc. Recent work has established that if A is a random matrix with entries drawn independently from certain probability distributions then exact recovery of x from these observations can be guaranteed with high probability. In this paper, we show that Toeplitz-structured matrices with entries drawn independently from the same distributions are also sufficient to recover x from y with high probability, and we compare the performance of such matrices with that of fully independent and identically distributed ones. The use of Toeplitz matrices in CS applications has several potential advantages: (i) they require the generation of only O(n) independent random variables; (ii) multiplication with Toeplitz matrices can be efficiently implemented using fast Fourier transform, resulting in faster acquisition and reconstruction algorithms; and (iii) Toeplitz-structured matrices arise naturally in certain application areas such as system identification. Index Terms — Compressed sensing, restricted isometry property, system identification, Toeplitz matrices, underdetermined systems of linear equations 1.1. Background 1.

### Citations

1715 | Compressed sensing - Donoho - 2006 |

1297 | Robust Uncertainty Principles: Exact Signal Reconstruction from Highly Incomplete Frequency Information
- Candes, Romberg, et al.
(Show Context)
Citation Context ...by solving the convex program x[ℓ] = arg�min z∈R n �z�1 subject to y = Az�, (21) − 1] y[n] where, in this case, y =����y[n . and . y[n + k − 2]����, − 1] a[n − 2] . . . a[1] a[0] a[n] a[n − 1] . . . a=-=[2]-=- a[1] A =����a[n . . . . . . .. . .. . a[n + k − 2] a[n + k − 3] . . . . . . . . a[k − 1]����. 3.4. Beyond Sparse Signals We have proven above that Toeplitz (and circulant) matrices, having entries dr... |

1230 |
Adaptive Filter Theory
- Haykin
- 1996
(Show Context)
Citation Context ...dge of its input and output signals, commonly termed as system identification, is of considerable importance in signal processing because of its applicability to a wide range of problems – see, e.g., =-=[11, 12]-=-. In the case of a finite impulse response (FIR) LTI system, this typically involves probing the system with a (known) white noise sequence of duration orders of magnitude greater than that of the imp... |

1190 |
System Identification: Theory for the User
- Ljung
- 1999
(Show Context)
Citation Context ...dge of its input and output signals, commonly termed as system identification, is of considerable importance in signal processing because of its applicability to a wide range of problems – see, e.g., =-=[11, 12]-=-. In the case of a finite impulse response (FIR) LTI system, this typically involves probing the system with a (known) white noise sequence of duration orders of magnitude greater than that of the imp... |

831 | Practical Signal Recovery from Random
- Candes, Romberg, et al.
(Show Context)
Citation Context ...mensionality reduction etc. and has recently received a lot of attention under the rubric of compressed sensing (CS) – starting in particular with some of the earlier works of Candes, Romberg and Tao =-=[1, 2, 3]-=- and Donoho [4]. One of the fundamental problems in CS is to identify the observation matrices that are sufficient to ensure exact recovery of x from This research was supported in part by the Nationa... |

816 |
Introduction to Graph Theory
- West
- 2001
(Show Context)
Citation Context ... hence, satisfies (2) with probability at least columns in expectation). Then, from the definition of A j T that each Ãj T 1 − e −f(|C j|,m,δ3m) ≥ 1 − e −f(⌊k/q⌋,m,δ3m) . (11) 2We refer the reader to =-=[7]-=- for a review of basic terminology in graph theory. 3Recall that coloring of a graph means that each Cj ⊂ V , ∪jCj = V , Cj ∩ Cj ′ = φ and no two vertices in a color class share an edge.s|C j | Also, ... |

654 | Decoding by linear programming
- Candès, Tao
- 2005
(Show Context)
Citation Context ...mensionality reduction etc. and has recently received a lot of attention under the rubric of compressed sensing (CS) – starting in particular with some of the earlier works of Candes, Romberg and Tao =-=[1, 2, 3]-=- and Donoho [4]. One of the fundamental problems in CS is to identify the observation matrices that are sufficient to ensure exact recovery of x from This research was supported in part by the Nationa... |

302 | A simple proof of the restricted isometry property for random matrices
- Baraniuk, Davenport, et al.
- 2008
(Show Context)
Citation Context ...andom matrices with entries drawn independently from certain probability distributions satisfy RIP of order 3m with high probability for every δ3m ∈ (0,1/3) provided k ≥ const · mln(n/m) – see, e.g., =-=[1, 3, 4, 6]-=-; we refer to such matrices as independent and identically distributed (IID) CS matrices. 1.2. Main Result In this paper, we show that if a probability distribution P(a) yields an IID CS matrix (havin... |

286 | Gradient projection for sparse reconstruction: Application to compressed sensing and other inverse problems
- Figueiredo, Nowak, et al.
- 2007
(Show Context)
Citation Context ... arg�min z∈R n �z�1 subject to y = Az�, (3) which is attractive because it can be solved in a computationally tractable manner using linear programming and convex optimization techniques – see, e.g., =-=[1, 4, 5]-=-. Note that the RIP of order 3m is equivalent to saying that the singular values of all k × 3m submatrices of A lie in the interval�Ô2/3,Ô4/3�. And while the definition of RIP does not guarantee the e... |

63 |
Random filters for compressive sampling and reconstruction
- Tropp, Wakin, et al.
- 2006
(Show Context)
Citation Context ...eplitz matrices of equal dimensions, in fact, support this intuition. It is also interesting to note that somewhat similar numerical results (without any performance guarantees) have been reported in =-=[10]-=- in the context of random filters. 3. EXTENSIONS In this section, we discuss natural extensions of the result of Section 2 to circulant and left-shifted Toeplitz-structured matrices. Further, we also ... |

54 |
Proof of a conjecture of Erd'os. Combinatorial theory and its applications
- Hajnal, Szemerédi
- 1969
(Show Context)
Citation Context ...gree ∆ of G (defined as the maximum number of edges originating from any vertex) is given by ∆ ≤ (q − 1). 2 Consequently, using the well-known Hajnal-Szemerédi theorem on equitable coloring of graphs =-=[8]-=-, we can always partition G using q (or more) colors such that ⌊k/q⌋ ≤ min |Cj| ≤ max |Cj| ≤ ⌈k/q⌉, (9) j∈{1,2,...,q} j∈{1,2,...,q} where {Cj} q j=1 correspond to the different color classes.3 Next, l... |

19 |
Equitable coloring extends Chernoff-Hoeffding bounds
- Pemmaraju
- 2001
(Show Context)
Citation Context ...r of independent random variables is not new and has been previously used by researchers to derive deviation bounds for sums of dependent random variables that exhibit limited dependence – see, e.g., =-=[9]-=-. Loosely speaking (and for k sufficiently large), Lemma 1 says that for any subset T of cardinality 3m, if an IID submatrix satisfies (2) with probability at least 1 − e −O(k) , then a Toeplitz subma... |

2 |
FIR system modeling and identification in the presence of noise with band-limited inputs
- Rabiner, Crochiere, et al.
- 1978
(Show Context)
Citation Context ...of a finite impulse response (FIR) LTI system, this typically involves probing the system with a (known) white noise sequence of duration orders of magnitude greater than that of the impulse response =-=[13]-=-, which may be prohibitive because of the delay incurred in solving for the impulse response and the difficulty of generating a truly white noise sequence. For the purposes of deconvolving an LTI syst... |

2 |
Sparsity and incoherence in compressive sampling,” Preprint - available at http ://www.dsp.ece.rice.edu/cs
- Candès, Romberg
(Show Context)
Citation Context ...hly O(m 2 ) IID submatrices having dimensions approximately equal to O(k/m 2 ) × |T |. It is interesting to note that certain special types of graphs can be equitably colored 4 We refer the reader to =-=[14]-=- for a different take on solving the problem of computational and storage complexity in CS applications. using far fewer colors – see, e.g., [9]. This implies interesting extensions to the work presen... |