Multipath Time-Delay Estimation With Impulsive Noise via Bayesian Compressive Sensing

Xingyu Ji, Lei Cheng, Hangfang ZhaoIEEE Signal Processing Letters 2023 · OpenReview

TL;DR

This paper introduces a novel Bayesian Compressive Sensing (BCS) method, called LL-BCS, designed to accurately estimate time delays in signals corrupted by impulsive noise, a common problem in radar and sonar. By employing a heavy-tailed Laplacian distribution for noise modeling and a Laplacian prior for signal sparsity, LL-BCS significantly outperforms existing methods in challenging, noisy environments.

What Problem Does This Paper Address?

Accurate time-delay estimation is crucial for many applications, including radar and sonar, where it helps determine the location and characteristics of objects. A significant challenge in these real-world scenarios is the presence of "impulsive noise" – sudden, sharp spikes in the signal that can severely degrade the performance of standard estimation techniques. Traditional methods, and even existing advanced techniques like Bayesian Compressive Sensing (BCS), often assume Gaussian noise, making them vulnerable to these outliers. This paper aims to bridge this gap by developing a robust estimation framework specifically designed to handle the detrimental effects of impulsive noise.

Methods

The core innovation of this paper lies in adapting the Bayesian Compressive Sensing (BCS) framework to explicitly model and mitigate impulsive noise. The authors propose a new method, termed LL-BCS, which leverages two key probabilistic modeling choices:

  • Laplacian Likelihood: Instead of the standard Gaussian assumption for noise, LL-BCS uses a heavy-tailed Laplacian distribution. This distribution is more adept at modeling the characteristics of impulsive noise, which are prone to extreme values.
  • Laplacian Prior: To encourage sparsity in the multipath signal representation (meaning only a few time delays are active), a Laplacian prior is employed. This is a common practice in compressive sensing to promote sparse solutions.

The method then formulates a Bayesian learning problem that involves estimating both the signal parameters (time delays and amplitudes) and the model hyperparameters. This is achieved using the BCS inference framework, which iteratively updates these parameters. A key aspect of their approach is the use of a Gaussian Mixture Model (GMM) representation for the Laplacian distribution, which facilitates the derivation of closed-form Bayesian inference steps. The hyperparameters are updated using an Expectation-Maximization (EM) algorithm, and the posterior distribution of the time delay is derived.

Datasets

The paper does not use pre-existing datasets for training or evaluation. Instead, it relies on simulated data generated to mimic realistic scenarios. The transmitting signal used is a linear frequency modulation (LFM) signal. The impulsive noise is synthetically generated using a Gaussian Mixture Model (GMM) with specific parameters for variances and mixing probabilities, as described in reference [13]. The simulations are conducted across various Signal-to-Impulse-Noise-Ratio (SINR) levels to rigorously test the performance of the proposed method against benchmarks.

Key Insights and Ideas

  • Laplacian is Key for Impulsive Noise: The most significant insight is that a heavy-tailed distribution like the Laplacian is far more effective than a Gaussian distribution for modeling impulsive noise in time-delay estimation. This allows the algorithm to "down-weight" the influence of these outliers.
  • BCS is Adaptable: The paper demonstrates the flexibility of the Bayesian Compressive Sensing framework. By simply changing the prior and likelihood distributions, BCS can be tailored to specific noise characteristics and signal properties, leading to improved performance.
  • Sparsity is Still Crucial: Even in the presence of impulsive noise, the underlying assumption of sparse multipath signals remains valid and is effectively exploited by the Laplacian prior.
  • LL-BCS Outperforms: The LL-BCS method shows a clear advantage in scenarios with significant impulsive noise, suggesting its practical utility in real-world radar and sonar systems.

Key Results

The paper presents simulation results that highlight the superior performance of LL-BCS, particularly in low Signal-to-Impulse-Noise-Ratio (SINR) conditions.

  • Improved Accuracy: In scenarios with intense impulsive noise, LL-BCS accurately estimates multipath parameters (both time delays and amplitudes). Visualizations (Figure 2) show LL-BCS closely matching the ground truth, while other methods like CVX and L-BCS struggle.
  • Lower RMSE: Quantitative results (Figure 3) demonstrate that LL-BCS achieves a significantly lower Root Mean Squared Error (RMSE) for time-delay estimation compared to standard CS, BCS, and Laplacian-prior BCS (L-BCS) when SINR is low (e.g., -10 dB and below).
  • Robustness: As SINR increases, the performance gap between LL-BCS and other methods decreases, indicating that while LL-BCS excels in noisy conditions, other methods become more competitive in cleaner environments.

Limitations and Future Work

  • Computational Complexity: While LL-BCS offers improved accuracy, the iterative nature of the Bayesian inference and hyperparameter updates can lead to higher computational complexity compared to simpler methods. The paper mentions that convergence is guaranteed towards a local optimum, which is a common characteristic of EM-based algorithms.
  • Hyperparameter Tuning: The performance of LL-BCS relies on the appropriate selection of hyperparameters. While the EM algorithm helps in their estimation, further research might explore more robust or adaptive hyperparameter tuning strategies.
  • Generalization to Other Noise Types: The current work focuses on impulsive noise modeled by a Laplacian distribution. Future work could explore extending this framework to other types of non-Gaussian noise or complex noise mixtures.
  • Real-World Data Validation: The current evaluation is based on simulated data. Validating the performance of LL-BCS on actual radar and sonar datasets would be a crucial next step to confirm its practical applicability.

Summary updated

Abstract

Multipath time-delay estimation is commonly encountered in radar and sonar signal processing. In some real-life environments, impulse noise is ubiquitous and significantly degrades estimation performance. Here, we propose a Bayesian approach to tailor the Bayesian Compressive Sensing (BCS) to mitigate impulsive noises. In particular, a heavy-tail Laplacian distribution is used as a statistical model for impulse noise, while Laplacian prior is used for sparse multipath modeling. The Bayesian learning problem contains hyperparameters learning and parameter estimation, solved under the BCS inference framework. The performance of our proposed method is compared with benchmark methods, including compressive sensing (CS), BCS, and Laplacian-prior BCS (L-BCS). The simulation results show that our proposed method can estimate the multipath parameters more accurately and have a lower root mean squared estimation error (RMSE) in intensely impulsive noise.
BibTeX
@article{ji2023multipath,
  author = {Xingyu Ji and Lei Cheng and Hangfang Zhao},
  title = {{Multipath Time-Delay Estimation With Impulsive Noise via Bayesian Compressive Sensing}},
  journal = {IEEE Signal Processing Letters},
  year = {2023},
  url = {https://openreview.net/forum?id=8jZZ5zlOZc},
}
Figure