[Paper Review] Properties of spatial coupling in compressed sensing

Francesco Caltagirone, Lenka Zdeborová|arXiv (Cornell University)|Jan 24, 2014

Sparse and Compressive Sensing Techniques17 references4 citations

TL;DR

This paper investigates the design of spatially coupled measurement matrices in compressed sensing for finite systems, showing that interaction range, shape, and seed optimization critically affect reconstruction performance. It demonstrates that asymmetric coupling with backward bias improves wave propagation, reduces required seed size, and approaches information-theoretic limits even under finite coupling conditions.

ABSTRACT

In this paper we address a series of open questions about the construction of spatially coupled measurement matrices in compressed sensing. For hardware implementations one is forced to depart from the limiting regime of parameters in which the proofs of the so-called threshold saturation work. We investigate quantitatively the behavior under finite coupling range, the dependence on the shape of the coupling interaction, and optimization of the so-called seed to minimize distance from optimality. Our analysis explains some of the properties observed empirically in previous works and provides new insight on spatially coupled compressed sensing.

Motivation & Objective

• To address open questions in spatially coupled compressed sensing for practical, finite-size implementations.
• To quantify the impact of coupling range, interaction shape, and seed parameters on reconstruction thresholds.
• To optimize the seed size and strength to minimize termination cost while maximizing wave propagation.
• To explain empirical observations from prior work using finite-size analysis.
• To provide design guidelines for hardware implementations of spatially coupled compressed sensing systems.

Proposed method

• The study uses a block-structured spatially coupled measurement matrix with L blocks, where each block interacts with neighboring blocks via a coupling function g(x).
• The interaction range w defines the number of neighboring blocks influencing each block, with g(x) modeling the coupling strength and directionality.
• The authors employ state evolution analysis to compute the algorithmic threshold α_w, which marks the transition between successful and failed reconstruction.
• They analyze the propagation speed of the reconstruction wave as a function of α_b, w, and g(x), particularly focusing on asymmetric shapes with angular coefficient A.
• The seed is modeled as a region of higher measurement density (α_s) and size (w_s) to initiate successful reconstruction wave propagation.
• Phase diagrams and effective measurement ratios α_eff are computed to evaluate performance across parameter space.

Experimental results

Research questions

• RQ1 How does the finite coupling range w affect the reconstruction threshold α_w in spatially coupled compressed sensing?
• RQ2 What is the optimal shape of the coupling function g(x), particularly its asymmetry, for maximizing reconstruction wave speed and minimizing seed requirements?
• RQ3 How should the seed size w_s and strength α_s be chosen to minimize termination cost while ensuring robust wave propagation?
• RQ4 To what extent does the interaction range w influence the effective measurement ratio α_eff along the transition line?
• RQ5 Why does asymmetric coupling (A < 0) outperform symmetric or forward-biased coupling in wave propagation and threshold saturation?

Key findings

• For small interaction range w=1, the effective measurement ratio α_eff is minimized, indicating optimal performance under finite coupling.
• The transition threshold α_w moves closer to the information-theoretic limit α_c when the coupling function g(x) is asymmetric with negative angular coefficient A.
• A negative A (stronger backward coupling) increases the propagation speed of the reconstruction wave and reduces the required seed size and strength.
• When w=1 is insufficient for wave propagation (e.g., ρ=0.2, α_b=0.26), increasing w to 2 becomes optimal, showing that interaction range must be tuned to signal sparsity.
• The phase diagram shows that for ρ=0.4, Δ=10⁻¹², and w=1, the system achieves α_w=0.4619, which is above the α_c threshold, confirming improved performance over standard compressed sensing.
• The study confirms that strong asymmetry in coupling, as used in prior works, is theoretically optimal for minimizing termination cost and maximizing robustness.