Elsevier

Signal Processing

Volume 134, May 2017, Pages 249-260
Signal Processing

p-Based complex approximate message passing with application to sparse stepped frequency radar

https://doi.org/10.1016/j.sigpro.2016.12.007Get rights and content

Highlights

  • A new compressed sensing algorithm called adaptive lp-CAMP is proposed.

  • The proposed algorithm can be applied to the sparse stepped frequency radar signal processing.

  • The performance of the proposed algorithm is verified by simulations and real data.

Abstract

Compressed sensing exploits the sparsity of the signal to reduce the sampling rate while keeping the resolution fixed, and has been widely used. In this paper we propose a new algorithm called adaptive ℓp-CAMP and show its application in the sparse stepped frequency radar signal processing. Our algorithm is inspired by the complex approximate message passing algorithm (CAMP) that solves complex-valued LASSO. The following properties of the proposed algorithm make it superior to existing algorithms: (1) All the parameters of the algorithm are tuned dynamically and optimally. The algorithm does not require any information about the signal and is still capable of tuning the parameters as well as an oracle that has all the signal information. (2) Adaptive ℓp-CAMP is designed to solve the complex-valued ℓp-regularized least squares for 0p1. Hence, it can outperform CAMP. The performance of the proposed algorithm is verified by simulations and the data collected by a real radar system.

Introduction

Compressed sensing (CS) theory suggests that a high-dimensional unknown sparse signal can be recovered exactly from a small number of measurements with high probability by solving a sparsity-regularized optimization problem [17], [31], [33], [34]. It can significantly reduce the sampling cost and is of great interest in a wide range of applications, such as magnetic resonance imaging (MRI) [11], [29], wireless communications [26] and radar signal processing [37], [35], [25], [13].

Compressed sensing for complex-valued signal is required in many applications such as radar and MRI [19], [21]. The last decade has witnessed a surge of reconstruction algorithms for complex signal [27], [17], [21]. One algorithm that has attracted significant attention is the complex approximate message passing (CAMP) algorithm [21]. The following properties of CAMP have made it appealing [1]: (1) the algorithm converges very fast, and hence is suitable for real-time processing. (2) Its mean-squared-error (MSE) can be accurately predicted at every iteration, which enables a dynamic tuning of its parameters on the fly. (3) It is straightforward to control the false alarm rate, which is important for applications such as radar.

Despite the appealing features of CAMP, some evidence inspires us to research better reconstruction algorithms. It is shown in CS that the ℓp-norm constraint with 0p<1 can lead to more accurate solutions than the ℓ1-norm constraint [8], [23], which encourages researchers to use the other algorithms such as SL0 and OMP despite all the nice properties of CAMP. The main objective of this paper is to develop an algorithm, called adaptive ℓp-CAMP. As an application, we apply our ℓp-CAMP algorithm to reconstruct the range profiles in sparse stepped frequency waveform (SFW) radar, which has the following properties: (1) It is guaranteed to outperform the original CAMP on any recovery problem. (2) All the parameters are tuned dynamically and efficiently (to achieve minimum MSE) in the algorithm, and hence it does not require any parameter tuning from the user. (3) Its MSE performance can be analyzed theoretically. In applications such as radar, there is a trade-off between the SNR of reconstruction and the data rate, so the third property helps in designing the parameter in sensing to improve the overall system performance.

The remainder of the paper is organized as follows. We start with the problem formulation and the signal model of the sparse SFW radar in Section 2. The adaptive ℓp-CAMP is developed in Section 3. The performance of ℓp-CAMP is theoretically analyzed by characterizing the statistical properties of the recovered signal in Section 4. In Section 5, the performance of the proposed algorithm is verified by using the real data collected by an experimental radar system. Finally we conclude the paper in Section 6.

Section snippets

Motivating example: sparse stepped frequency radar

To motivate this work by an application, we introduce the signal model of sparse SFW radar and the problem formulation. Conventional full-band SFW radars transmit a series of narrowband pulses, which are stepped in frequency from pulse to pulse and form a burst1 that covers a wide bandwidth. For high resolution radar, the target is usually composed of several scatterers which take up a small fraction of the

Solving c-LPLS with CAMP

In this section, we develop an algorithm for solving (5), (6) under the framework of AMP, along with an analysis framework reviewed in Section 4, which enables the evaluation and automatic optimization of the parameters.

State evolution

In this section, we first explain the state evolution (SE) framework that characterizes the performance of the p-CAMP and adaptive p-CAMP. Then the performance of the adaptive p-CAMP is compared with that of the adaptive 1-CAMP. In Theorem 1 only, we write the vectors and matrices as x(N), A(N), w(N), xt(N), zt(N) to emphasize the dimensions of x. In [36], the SE for p-AMP has been given. The result can be extended to the complex-valued signal case and we obtain the following theorem.

Theorem 1

Let {x

Data description

In this section, we verify the performance of the proposed algorithm in the sparse SFW radar application. The adaptive ℓp-CAMP algorithm is run on the data collected by an experimental full-band SFW radar system and its performance is compared with that of the other existing methods. The experimental radar is an S-band synthetic wideband pulse doppler radar reported in [14]. The waveform parameter of the radar in this experiment is listed in Table 1. The radar transmits 256 pulses in each burst

Conclusions

In this paper, we have proposed the adaptive ℓp-CAMP algorithm and showed its application to the sparse SFW radar signal processing. The experiments on real data show that the proposed algorithm provides better MSE performance compared to ℓ1-CAMP and some other existing compressed sensing reconstruction algorithms. It is also shown that the performance of the proposed algorithm can be accurately predicted by a theoretical framework, known as state evolution (SE). SE allows the system designer

Acknowledgements

The work of Quanhua Liu is supported by the National Natural Science Foundation of China (Grant No. 61301189), and 111 Project of China (Grant No. B14010).

References (37)

  • C.-A. Deledalle, G. Peyré, J. Fadili, Stein consistent risk estimator (score) for hard thresholding. 2013, arXiv...
  • D.L. Donoho et al.

    Message-passing algorithms for compressed sensing

    Proc. Natl. Acad. Sci.

    (2009)
  • S. Foucart et al.

    A Mathematical Introduction to Compressive Sensing

    (2013)
  • K. Gerlach et al.

    Detection of a spatially distributed target in white noise

    IEEE Signal Process. Lett.

    (1997)
  • J.P. Haldar et al.

    Compressed-sensing MRI with random encoding

    IEEE Trans. Med. Imag.

    (2011)
  • S. Haykin

    Cognitive dynamic systems: Radar, control, and radio [point of view]

    Proc. IEEE

    (2012)
  • M.A. Herman et al.

    High-resolution radar via compressed sensing

    IEEE Trans. Signal Process.

    (2009)
  • F. Hua-yu, R. Li-xiang, Z. Kang, M. Er-ke, A fully coherent motion compensation method of ISAR imaging based on...
  • Cited by (17)

    • Sparse representation of complex-valued fMRI data based on spatiotemporal concatenation of real and imaginary parts

      2021, Journal of Neuroscience Methods
      Citation Excerpt :

      However, finding sparse representation of complex-valued fMRI data is challenging due to the highly noisy nature of phase data. Furthermore, existing complex-valued sparse representation algorithms are not suitable for analyzing fMRI data, having been developed for inverse synthetic aperture radar imaging (Zheng et al., 2016; Shi et al., 2018; Yang et al., 2019), magnetic resonance imaging (Chaabene et al., 2018; Da Silva et al., 2018; Cao et al., 2019), or direction-of-arrival estimation (Panahi and Viberg, 2012; Wu and Fang, 2017; Liu et al., 2018). Most recently, Iqbal et al. (2020) proposed an adaptive complex-valued dictionary learning (cDL) algorithm for analyzing complex-valued fMRI data.

    • Signal denoising using the minimum-probability-of-error criterion

      2020, APSIPA Transactions on Signal and Information Processing
    • Waveform optimization for SFA radar based on evolutionary particle swarm optimization

      2022, Xi Tong Gong Cheng Yu Dian Zi Ji Shu/Systems Engineering and Electronics
    View all citing articles on Scopus
    View full text