Compressive Domain Interference Cancellation

Mark A. Davenport, Petros T. Boufounos, Richard G. Baraniuk2009

TL;DR

This paper introduces a novel method for removing unwanted interference from signals acquired using compressive sensing (CS). By operating directly on the compressed measurements, their technique efficiently cancels interference while preserving the signal's essential properties, leading to improved recovery accuracy and speed compared to traditional methods.

What Problem Does This Paper Address?

Compressive sensing (CS) allows us to acquire signals using far fewer measurements than traditional methods, provided the signal is sparse (can be represented with few non-zero coefficients in some basis). However, CS systems can be more susceptible to interference than classical methods. When a signal of interest is acquired alongside unwanted interference, traditional approaches might involve recovering the entire signal first and then trying to separate the interference. This can be computationally expensive and may degrade the quality of the signal of interest. This paper addresses the challenge of efficiently removing interference before or during the signal recovery process, specifically within the compressed domain itself.

Methods

The core of the paper's contribution is a compressive domain filtering technique. The authors propose an interference cancellation method that leverages the mathematical properties of compressive sensing, particularly the Restricted Isometry Property (RIP). The key idea is to construct a projection operator, denoted by $P$, that operates directly on the compressed measurements $y$. This operator is designed to nullify the component of the measurements corresponding to the interference ($x_I$) while preserving the information about the signal of interest ($x_S$).

The method relies on the assumption that the interference signal ($x_I$) lives in a known subspace that is orthogonal to the signal of interest ($x_S$). The projection operator $P$ is constructed such that its nullspace is precisely this interference subspace. This ensures that when $P$ is applied to the combined measurements $y = \Phi(x_S + x_I)$, the interference term vanishes: $P\Phi x_I = 0$. Crucially, the authors prove that this projection operator maintains a relaxed version of the RIP for the remaining signal of interest, meaning the geometric properties of the signal space are preserved, which is vital for subsequent CS recovery algorithms. The construction of $P$ is designed to be computationally efficient, especially when the sensing matrix $\Phi$ has a fast transform-based implementation.

Datasets

Not discussed in the paper. The experiments use synthetic data generated with controlled parameters for signal sparsity, interference sparsity, and noise levels.

Key Insights and Ideas

  • Interference cancellation in the compressed domain is feasible and beneficial: Instead of recovering the full signal and then filtering, operating directly on compressed measurements offers significant advantages.
  • Preserving the RIP is crucial: The proposed method's ability to maintain the RIP for the signal of interest is key to ensuring that subsequent recovery algorithms can still perform effectively.
  • Orthogonality is a powerful simplifying assumption: The assumption that the interference subspace is orthogonal to the signal of interest subspace greatly simplifies the design and analysis of the cancellation operator.
  • Computational efficiency gains: By avoiding the full signal recovery before cancellation, the proposed method can be significantly faster, especially when interference is complex or abundant.

Key Results

The paper demonstrates through experiments that their "cancel-then-recover" approach significantly outperforms "modified recovery" and "recover-then-cancel" methods, especially when the interference signal's complexity ($K_I$) is much larger than the signal of interest's complexity ($K_S$).

  • Improved Recovery Accuracy: The "cancel-then-recover" method achieved significantly better Signal-to-Noise Ratio (SNR) for the recovered signal of interest, performing almost as well as an "oracle" (which knows the exact signal and interference) across a wide range of $K_I/K_S$ ratios.
  • Faster Recovery Time: The proposed method was also considerably faster than the other approaches, particularly as $K_I$ increased relative to $K_S$. This speedup is attributed to avoiding the full signal recovery step and potentially more efficient computation of the projection operator.

Limitations and Future Work

  • Orthogonality Assumption: The current method relies on the interference subspace being orthogonal to the signal of interest subspace. While this simplifies the problem, real-world scenarios might not always satisfy this condition.
  • Known Interference Support: The method assumes knowledge of the interference's sparsity basis and, implicitly, its support. If the interference is not sparse in a known basis or its support is unknown, this approach would need modification.

Future work suggested includes generalizing the filters beyond simple projection schemes to more arbitrary compressive domain filters and exploring even more efficient and structured approaches for practical CS systems (e.g., random samplers). The authors also suggest studying the impact of such filtering approaches on a wider variety of compressive domain processing algorithms.

Summary updated

Abstract

In this paper we consider the scenario where a compressive sensing system acquires a signal of interest corrupted by an interfering signal. Under mild sparsity and orthogonality conditions on the signal and interference, we demonstrate that it is possible to efficiently filter out the interference from the compressive measurements in a manner that preserves our ability to recover the signal of interest. Specifically, we develop a filtering method that nulls out the interference while maintaining the restricted isometry property (RIP) on the set of potential signals of interest. The construction operates completely in the compressive domain and has computational complexity that is polynomial in the number of measurements.
BibTeX
@misc{davenport2009compressive,
  author = {Mark A. Davenport and Petros T. Boufounos and Richard G. Baraniuk},
  title = {{Compressive Domain Interference Cancellation}},
  year = {2009},
}
Figure