|
665 | 665 |
|
666 | 666 | \paragraph{(ii) Maximum Entropy (MEM).} |
667 | 667 | Choose $D$ to maximize Shannon entropy subject to matching measured harmonics (constraints): |
668 | | - \begin{align} |
669 | | - \max_{D\ge0}\quad & \mathcal{H}[D] = -\int_{0}^{2\pi} D\ln D\, d\theta,\\ |
670 | | - \text{s.t.}\quad & \int D\,d\theta=1,\quad |
671 | | - \int D \cos n\theta\, d\theta = a_n,\quad |
672 | | - \int D \sin n\theta\, d\theta = b_n,\; n=1,\ldots,N_h. |
673 | | - \end{align} |
| 668 | + \begin{equation} |
| 669 | + \begin{aligned} |
| 670 | + \max_{D\ge 0}\quad |
| 671 | + & \mathcal{H}[D] |
| 672 | + = -\int_{0}^{2\pi} D \ln D \, d\theta, |
| 673 | + \\ |
| 674 | + \text{s.t.}\quad |
| 675 | + & \int D\, d\theta = 1, |
| 676 | + \\ |
| 677 | + & \int D \cos(n\theta)\, d\theta = a_n, |
| 678 | + \\ |
| 679 | + & \int D \sin(n\theta)\, d\theta = b_n, |
| 680 | + \qquad n=1,\ldots,N_h. |
| 681 | + \end{aligned} |
| 682 | + \end{equation} |
674 | 683 | The solution has the exponential family form |
675 | 684 | \begin{equation} |
676 | 685 | D(\theta)\propto |
|
681 | 690 | \paragraph{(iii) Maximum Likelihood/Beamforming (MLM/EMLM).} |
682 | 691 | For an array with steering $\mathbf{a}(\theta;f)$ and CSD $\mathbf{C}(f)$, the classic Capon/MLM estimator is |
683 | 692 | \begin{equation} |
684 | | - \widehat{D}_{\text{MLM}}(f,\theta) |
685 | | - = \frac{1}{\mathbf{a}(\theta;f)^{\mathrm{H}}\, |
686 | | - \mathbf{C}(f)^{-1}\,\mathbf{a}(\theta;f)}\;\bigg/ \; |
687 | | - \int_{0}^{2\pi}\frac{d\varphi}{\mathbf{a}(\varphi;f)^{\mathrm{H}}\, |
688 | | - \mathbf{C}(f)^{-1}\,\mathbf{a}(\varphi;f)}. |
| 693 | + \begin{aligned} |
| 694 | + \widehat{D}_{\text{MLM}}(f,\theta) |
| 695 | + &= |
| 696 | + \frac{ |
| 697 | + \displaystyle |
| 698 | + \frac{1}{ |
| 699 | + \mathbf{a}(\theta;f)^{\mathrm{H}} |
| 700 | + \mathbf{C}(f)^{-1} |
| 701 | + \mathbf{a}(\theta;f) |
| 702 | + } |
| 703 | + }{ |
| 704 | + \displaystyle |
| 705 | + \int_{0}^{2\pi} |
| 706 | + \frac{d\varphi}{ |
| 707 | + \mathbf{a}(\varphi;f)^{\mathrm{H}} |
| 708 | + \mathbf{C}(f)^{-1} |
| 709 | + \mathbf{a}(\varphi;f) |
| 710 | + } |
| 711 | + }. |
| 712 | + \end{aligned} |
689 | 713 | \end{equation} |
690 | 714 | optionally corrected for bias/noise in EMLM. This provides high directional resolution when aperture and SNR are adequate. |
691 | 715 |
|
|
713 | 737 | \paragraph{Outputs and Integral Measures.} |
714 | 738 | From $\widehat{S}(f,\theta)$ compute directional moments and spreads: |
715 | 739 | \begin{align} |
716 | | - \bar{\theta}(f) &= \arg\!\Big(\int e^{\mathrm{i}\theta} D(f,\theta)\, d\theta\Big), |
717 | | - & |
718 | | - \sigma_\theta^2(f) &= 2\Big(1-\big|\int e^{\mathrm{i}\theta} D(f,\theta)\, d\theta\big|\Big),\\ |
719 | | - m_n &= \int f^n \bigg(\int S(f,\theta)\,d\theta\bigg)\, df, |
720 | | - & |
721 | | - H_s &= 4\sqrt{m_0},\quad T_z=\sqrt{m_0/m_2},\quad T_{-10}=m_{-1}/m_0. |
| 740 | + \bar{\theta}(f) |
| 741 | + &= \arg\!\left( |
| 742 | + \int e^{\mathrm{i}\theta} D(f,\theta)\, d\theta |
| 743 | + \right), |
| 744 | + \\ |
| 745 | + \sigma_\theta^2(f) |
| 746 | + &= 2\left( |
| 747 | + 1-\left| |
| 748 | + \int e^{\mathrm{i}\theta} D(f,\theta)\, d\theta |
| 749 | + \right| |
| 750 | + \right), |
| 751 | + \\ |
| 752 | + m_n |
| 753 | + &= \int f^n |
| 754 | + \left( |
| 755 | + \int S(f,\theta)\, d\theta |
| 756 | + \right)\, df, |
| 757 | + \\ |
| 758 | + H_s &= 4\sqrt{m_0}, |
| 759 | + \qquad |
| 760 | + T_z = \sqrt{m_0/m_2}, |
| 761 | + \qquad |
| 762 | + T_{-10} = m_{-1}/m_0. |
722 | 763 | \end{align} |
723 | 764 | Directional partitions (wind sea vs.\ swells) can then be obtained by algorithms in \Cref{sec:wave-partitioning}. |
724 | 765 |
|
|
856 | 897 |
|
857 | 898 | \end{thebibliography} |
858 | 899 |
|
859 | | -\end{document} |
| 900 | +\end{document} |
0 commit comments