# 一种基于压缩感知的无需超参数的测向算法A Hyperparameter-Free Direction-Finding Algorithm in the Compressive Sensing Framework

• 全文下载: PDF(621KB)    PP.1-7   DOI: 10.12677/HJWC.2019.91001
• 下载量: 220  浏览量: 319

The fully automatic sparsity-parameter estimation algorithms do not require the user to make any hard decision (possibly via trial-and-error) on the values of the hyperparameters, making them more pragmatic in practice. This paper provides a unified interpretation of the existing approaches including covariance matrix fitting (CMF), sparse iterative covariance based estimation (SPICE) and likelihood-based estimation of sparse parameters (LIKES). The point of view taken is that they are all covariance-fitting-based algorithms under different statistical distances. Following this, we present a new covariance-fitting scheme trying to minimize one of the two asymmetrical Itakura-Saito distances. Simulations show that the proposed method appears to be preferable as it outperforms the aforementioned algorithms in general.

1. 引言

2. 数学模型

$x\left(t\right)={\sum }_{m=1}^{M}{a}_{m}{s}_{m}\left(t\right)+n\left(t\right)$ (1)

${\left\{{\theta }_{k}\right\}}_{k=1}^{K}$ 为可以覆盖所有可能入射区域的栅格，并覆盖了(至少是近似覆盖了)上述M个信号的入射角度，相应的导向矢量为，其中， $K\gg N$ 。于是，式(1)可以重新表述为：

$x\left(t\right)={\sum }_{k=1}^{K}{d}_{k}{\stackrel{¯}{s}}_{k}\left(t\right)+n\left(t\right)$ (2)

$R=E\left\{x\left(t\right){x}^{\text{H}}\left(t\right)\right\}={\sum }_{k=1}^{K}{\sigma }_{k}{d}_{k}{d}_{k}^{\text{H}}+ϵI$ (3)

$R={\sum }_{k=1}^{K+N}{\sigma }_{k}{d}_{k}{d}_{k}^{\text{H}}$ (4)

$\stackrel{^}{R}=\frac{1}{L}{\sum }_{l=1}^{L}x\left[l\right]{x}^{\text{H}}\left[l\right]$ (5)

2.1. 协方差矩阵拟合算法(CMF)

$\begin{array}{l}\underset{{\left\{{\sigma }_{k}\right\}}_{k=1}^{K},ϵ}{\mathrm{min}}\left\{{d}_{\text{SE}}\left(\stackrel{^}{R},R\right)={‖\stackrel{^}{R}-\underset{k=1}{\overset{K}{\sum }}{\sigma }_{k}{d}_{k}{d}_{k}^{\text{H}}-ϵI‖}^{2}\right\}\\ \text{s}.\text{t}.\text{\hspace{0.17em}}\text{\hspace{0.17em}}{\left\{{\sigma }_{k}\right\}}_{k=1}^{K}\ge 0,\text{\hspace{0.17em}}\text{\hspace{0.17em}}{\sum }_{k=1}^{K}{\sigma }_{k}\le \lambda ,\text{\hspace{0.17em}}\text{\hspace{0.17em}}\lambda =-N{\gamma }_{N}+{\sum }_{n=1}^{N}{\gamma }_{n},\text{\hspace{0.17em}}\text{\hspace{0.17em}}ϵ\ge 0\end{array}$ (6)

$\begin{array}{l}\underset{{\left\{{\sigma }_{k}\right\}}_{k=1}^{K+N}}{\mathrm{min}}\left\{{d}_{\text{SE}}\left(\stackrel{^}{R},R\right)={‖\stackrel{^}{R}-\underset{k=1}{\overset{K+N}{\sum }}{\sigma }_{k}{d}_{k}{d}_{k}^{\text{H}}‖}^{2}\right\}\\ \text{s}.\text{t}.\text{\hspace{0.17em}}\text{\hspace{0.17em}}{\left\{{\sigma }_{k}\right\}}_{k=1}^{K+N}\ge 0\end{array}$ (7)

2.2. 基于稀疏迭代协方差的估计(SPICE)

SPICE算法 [6] 实际上是将两种高斯分布 之间的Jeffreys距离(亦称为对称Kullback-Leibler距离) [13] 最小化：

$\begin{array}{l}\underset{{\left\{{\sigma }_{k}\right\}}_{k=1}^{K+N}}{\mathrm{min}}\left\{{‖{R}^{-1/2}\left(\stackrel{^}{R}-R\right){\stackrel{^}{R}}^{-1/2}‖}^{2}=\text{tr}\left({R}^{-1}\stackrel{^}{R}\right)+\text{tr}\left({\stackrel{^}{R}}^{-1}R\right)-2N=2{d}_{KL}\left(\stackrel{^}{R},R\right)\right\}\\ \text{s}.\text{t}.\text{\hspace{0.17em}}\text{\hspace{0.17em}}\text{\hspace{0.17em}}R={\sum }_{k=1}^{K+N}{\sigma }_{k}{d}_{k}{d}_{k}^{\text{H}},\text{\hspace{0.17em}}\text{\hspace{0.17em}}\text{\hspace{0.17em}}{\left\{{\sigma }_{k}\right\}}_{k=1}^{K+N}\ge 0\end{array}$ (8)

2.3. 基于似然的稀疏参数估计(LIKES)

LIKES算法 [7] 使用的是高斯最大似然(GML)标准：

$\begin{array}{l}\underset{{\left\{{\sigma }_{k}\right\}}_{k=1}^{K+N}}{\mathrm{min}}\left\{{X}^{\text{H}}{R}^{-1}X+\text{ln}|R|\right\}\\ \text{s}.\text{t}\text{.}\text{\hspace{0.17em}}\text{\hspace{0.17em}}\text{\hspace{0.17em}}R={\sum }_{k=1}^{K+N}{\sigma }_{k}{d}_{k}{d}_{k}^{\text{H}},\text{\hspace{0.17em}}\text{\hspace{0.17em}}\text{\hspace{0.17em}}{\left\{{\sigma }_{k}\right\}}_{k=1}^{K+N}\ge 0\end{array}$ (9)

$\begin{array}{l}\underset{{\left\{{\sigma }_{k}\right\}}_{k=1}^{K+N}}{\mathrm{min}}\left\{{d}_{\text{IS}}\left(\stackrel{^}{R},R\right)=\text{tr}\left(\stackrel{^}{R}{R}^{-1}-I\right)-\text{ln}|\stackrel{^}{R}{R}^{-1}|\right\}\\ \text{s}.\text{t}\text{.}\text{\hspace{0.17em}}\text{\hspace{0.17em}}\text{\hspace{0.17em}}R={\sum }_{k=1}^{K+N}{\sigma }_{k}{d}_{k}{d}_{k}^{\text{H}},\text{\hspace{0.17em}}\text{\hspace{0.17em}}\text{\hspace{0.17em}}{\left\{{\sigma }_{k}\right\}}_{k=1}^{K+N}\ge 0\end{array}$ (10)

3. 本文提出的协方差矩阵拟合方案

$\begin{array}{l}\underset{{\left\{{\sigma }_{k}\right\}}_{k=1}^{K+N}}{\mathrm{min}}\left\{{d}_{\text{IS}}\left(R,\stackrel{^}{R}\right)=\text{tr}\left(R{\stackrel{^}{R}}^{-1}-I\right)-\text{ln}|R{\stackrel{^}{R}}^{-1}|\right\}\\ \text{s}.\text{t}\text{.}\text{\hspace{0.17em}}\text{\hspace{0.17em}}\text{\hspace{0.17em}}R={\sum }_{k=1}^{K+N}{\sigma }_{k}{d}_{k}{d}_{k}^{\text{H}},\text{\hspace{0.17em}}\text{\hspace{0.17em}}\text{\hspace{0.17em}}{\left\{{\sigma }_{k}\right\}}_{k=1}^{K+N}\ge 0\end{array}$ (11)

${d}_{\text{IS}}\left(R,\stackrel{^}{R}\right)=\text{tr}\left(R{\stackrel{^}{R}}^{-1}\right)-\text{ln}|R|+\text{const}={\sum }_{k=1}^{K+N}{\sigma }_{k}{d}_{k}^{\text{H}}{\stackrel{^}{R}}^{-1}{d}_{k}-\text{ln}|{\sum }_{k=1}^{K+N}{\sigma }_{k}{d}_{k}{d}_{k}^{\text{H}}|+\text{const}$ (12)

$\stackrel{¯}{R}=\stackrel{^}{\alpha }I+\stackrel{^}{\beta }\stackrel{^}{R}$ (13)

$\stackrel{^}{\rho }=\frac{1}{{L}^{2}}\underset{l=1}{\overset{L}{\sum }}{‖x\left[l\right]‖}^{4}-\frac{1}{L}{‖\stackrel{^}{R}‖}^{2}$

$\stackrel{^}{\nu }=\frac{\text{tr}\left(\stackrel{^}{R}\right)}{N}$

$\stackrel{^}{\beta }=1-\frac{\stackrel{^}{\alpha }}{\stackrel{^}{\nu }}$ (14)

$\begin{array}{l}\underset{{\left\{{\sigma }_{k}\right\}}_{k=1}^{K+N}}{\mathrm{min}}\left\{\underset{k=1}{\overset{K+N}{\sum }}{\sigma }_{k}{d}_{k}^{\text{H}}{\stackrel{^}{R}}^{-1}{d}_{k}-\text{ln}|\underset{k=1}{\overset{K+N}{\sum }}{\sigma }_{k}{d}_{k}{d}_{k}^{\text{H}}|\right\}\\ \text{s}.\text{t}.\text{\hspace{0.17em}}\text{\hspace{0.17em}}\text{\hspace{0.17em}}{\left\{{\sigma }_{k}\right\}}_{k=1}^{K+N}\ge 0\end{array}$ (15)

$c={\left[{d}_{1}^{\text{H}}{\stackrel{^}{R}}^{-1}{d}_{1},{d}_{2}^{\text{H}}{\stackrel{^}{R}}^{-1}{d}_{2},\cdots ,{d}_{K+N}^{\text{H}}{\stackrel{^}{R}}^{-1}{d}_{K+N}\right]}^{\text{T}}$ (16)

$p={\left[{\sigma }_{1},{\sigma }_{2},\cdots ,{\sigma }_{K+N}\right]}^{\text{T}}$ (17)

$D=\left[{d}_{1},{d}_{2},\cdots ,{d}_{K+N}\right]$ (18)

cvx_begin quiet

cvx_precision best

cvx_expert true

variable p(K+N,1)

minimize (c'*p-log_det(D*diag(p)*D'))

subject to

p >= zeros(K+N,1);

cvx_end

4. 计算机仿真

$\text{RMSE}=\sqrt{\frac{{\sum }_{n=1}^{300}\left[{\left({\stackrel{^}{\theta }}_{1,n}-{\theta }_{1}\right)}^{2}+{\left({\stackrel{^}{\theta }}_{2,n}-{\theta }_{2}\right)}^{2}+{\left({\stackrel{^}{\theta }}_{3,n}-{\theta }_{3}\right)}^{2}\right]}{900}}$ (19)

Figure 1. RMSE versus varied SNR

Figure 2. RMSE versus varied numbers of samples

5. 结论

 [1] ITU (2011) Spectrum Monitoring Handbook. ITU, Geneva. [2] Capon, J. (1969) High-Resolution Frequency-Number Spectrum Analysis. Proceedings of the IEEE, 57, 1408-1418. https://doi.org/10.1109/PROC.1969.7278 [3] Schmidt, R.O. (1986) Multiple Emitter Location and Signal Parameter Estimation. IEEE Transactions on Antennas and Propagation, 34, 276-280. https://doi.org/10.1109/TAP.1986.1143830 [4] Malioutov, D., ?etin, M. and Willsky, A. (2005) A Sparse Signal Reconstruction Perspective for Source Localization with Sensor Arrays. IEEE Transactions on Signal Processing, 53, 3010-3022. https://doi.org/10.1109/TSP.2005.850882 [5] Yardibi, T., Li, J., Stoica, P., et al. (2008) Sparsity Constrained Deconvolution Approaches for Acoustic Source Mapping. Journal of Acoustical Society of America, 123, 2631-2642. https://doi.org/10.1121/1.2896754 [6] Stoica, P., Babu, P. and Li, J. (2011) SPICE: A Sparse Covariance-Based Estimation Method for Array Signal Processing. IEEE Transactions on Signal Processing, 59, 629-638. https://doi.org/10.1109/TSP.2010.2090525 [7] Stoica, P. and Babu, P. (2012) SPICE and LIKES: Two Hyperparameter-Free Methods for Sparse-Parameter Estimation. Signal Processing, 92, 1580-1590. https://doi.org/10.1016/j.sigpro.2011.11.010 [8] Tibshirani, R. (1996) Regression Shrinkage and Selection via the Lasso. Journal of Royal Statistical Society: Series B (Statistical Methodology), 58, 267-288. https://doi.org/10.1111/j.2517-6161.1996.tb02080.x [9] Gorodnitsky, I.F. and Rao, B.D. (1997) Sparse Signal Reconstruction from Limited Data Using FOCUSS: A Re-Weighted Minimum Norm Algorithm. IEEE Transactions on Signal Processing, 45, 600-616. https://doi.org/10.1109/78.558475 [10] Cotter, S.F., Rao, B.D., Engan, K., et al. (2005) Sparse Solutions to Linear Inverse Problems with Multiple Measurement Vectors. IEEE Transactions on Signal Processing, 53, 2477-2488. https://doi.org/10.1109/TSP.2005.849172 [11] Xu, D.Y., Hu, N., Ye, Z.F., et al. (2012) The Estimate for DOAs of Signals Using Sparse Recovery Method. Proceedings of the 37th IEEE International Conference on Acoustics, Speech, and Signal Processing, Kyoto, 2573-2576. https://doi.org/10.1109/ICASSP.2012.6288442 [12] Ottersten, B., Stoica, P. and Roy, R. (1998) Covariance Matching Estima-tion Techniques for Array Signal Processing Applications. Digital Signal Processing, 8, 185-210. https://doi.org/10.1006/dspr.1998.0316 [13] Kullback, S. (1997) Information Theory and Statistics. Dover Edition. Dover, New York. [14] Bensaid, S. and Slock, D. (2012) Blind Audio Source Separation Exploiting Periodicity and Spectral Envelops. Pro-ceedings of International Workshop on Acoustic Signal Enhancement, Aachen. [15] Vandenberghe, L., Boyd, S. and Wu, S.P. (1998) Determinant Maximization with Linear Matrix Inequality Constraints. SIAM Journal on Matrix Analysis and Applications, 19, 499-533. https://doi.org/10.1137/S0895479896303430 [16] Landi, L., De Maio, A., De Nicola, S., et al. (2008) Knowledge-Aided Covariance Matrix Estimation: A MAXDET Approach. IET Radar, Sonar & Navigation, 3, 341-356. https://doi.org/10.1109/RADAR.2008.4720823 [17] Li, J., Du, L. and Stoica, P. (2008) Fully Automatic Computation of Di-agonal Loading Levels for Robust Adaptive Beamforming. Proceedings of the 33rd IEEE International Conference on Acoustics, Speech, and Signal Processing, Las Vegas, 2325-2328. [18] Wu, S.P., Vandenberghe, L. and Boyd, S. (1996) MAXDET: Software for Determinant Maximization Problems. Information Systems Laboratory, Stanford University, Stanford. [19] Grant, M. and Boyd, S. (2012) CVX: MATLAB Software for Disciplined Convex Programming, Version 2.0 Beta. http://cvxr.com/cvx [20] Bhattacharyya, A. (1943) On a Measure of Divergence between Two Statistical Populations Defined by Their Probability Distributions. Bulletin of the Calcutta Mathematical Society, 35, 99-109. [21] Zolotarev, V.M. (1984) Probability Metrics. Theory of Probability and Its Applications, 28, 278-302. https://doi.org/10.1137/1128025