-
随着对多输入多输出(Multiple Input Multiple Output, MIMO)雷达和频率分集阵列(Frequency Diverse Array, FDA)技术研究的深入,研究人员将两种技术相结合,提出了FDA-MIMO雷达[1-7]。FDA-MIMO雷达采用天线阵列发射和接收信号,在各天线发射信号的载波间引入一个微小的频率差。同时要求各天线发射的基带信号相互正交,在接收端利用发射导向向量的距离、角度耦合性,提高雷达目标检测、参数估计以及对主瓣干扰的抑制能力。文献[1-3]针对工作于均匀高斯干扰背景下的FDA-MIMO雷达,分别设计了应用于单目标、分布式目标以及多目标的自适应检测器。文献[4]针对采用长基线发射阵列的FDA-MIMO雷达,提出了一种实现目标距离和角度无模糊估计的方法。文献[5-6]针对工作于多径背景下的双基地FDA-MIMO雷达,提出了用于目标距离、波达角和波离角的估计算法。文献[7]基于FDA-MIMO雷达在距离维的自由度,研究其对主瓣干扰的抑制方法。
大型阵列技术已成为雷达、通信等领域的重要研究方向[8-9],采用大的天线阵列能增强雷达对目标的检测概率、提高角度估计精度和分辨率,然而,大型阵列高的接收数据维数会造成计算复杂和工程实现困难。波束空间处理是解决上述问题的常用方法[10-12],它将各天线单元接收的数据与一个降维矩阵相乘,将高维的阵元空间数据变换成低维的波束空间数据,在波束空间设计相应的处理算法实现对目标的检测和定位等功能。文献[10]提出了用于MIMO雷达角度估计的波束空间实多项式求根算法。文献[11]针对低角跟踪背景下相干目标的角度估计问题,提出了一种波束空间角度估计算法。文献[12]提出一种用于目标角度估计的波束空间稀疏贝叶斯算法。
由于传统的波束空间变换矩阵破坏了阵元数据矩阵的旋转不变性和范德蒙结构,并在波束空间的带外引入新的零点,因此需要对旋转不变技术 (Estimating Signal Parameter Via Rotational Invariance Techniques, ESPRIT)、求根MUSIC (Multiple Signal Classification)等超分辨算法做一定的调整,才能实现对目标角度的估计。卷积波束空间(Convolutional Beamspace, CBS)处理不改变均匀线阵接收数据的旋转不变性和范德蒙结构,能有效解决上述问题[13-16]。文献[13]提出用于均匀线阵目标角度估计的卷积波束空间处理方法,该方法是将阵列接收数据与一个满足奈奎斯特性质的FIR滤波器的单位冲激响应进行卷积处理,再对卷积处理结果进行抽取运算以减小数据维数。
本文将卷积波束空间处理方法应用到FDA-MIMO雷达中,研究了FDA-MIMO雷达卷积波束空间目标距离和角度估计问题。考虑文献[13]所提的波束空间ESPRIT方法(CBS ESPRIT)需要对抽取数据的协方差矩阵进行特征分解[17],为进一步减小运算量,本文给出基于传播算子(Propagator Method, PM)[18-20]的卷积波束空间角度、距离估计算法,并根据卷积滤波器的通带特性和FDA阵列角度、距离间的耦合关系,实现目标角度和距离参数的解模糊处理。
-
本文研究的FDA-MIMO雷达的发射天线和接收天线分别为由
${N_{\mathrm{t}}}$ 个和${M_{\mathrm{r}}}$ 个阵元构成的均匀线阵。第$n$ 个发射阵元的工作频率和发射信号分别为:$$ {f_n} = {f_0} + n\Delta f\quad n = 0,1, \cdots ,{N_t} - 1 $$ (1) $$ {s_n}( t ) = {g_n}( t ){{\mathrm{e}}^{{\mathrm{j}}2 {\text π} {f_n}t}}\quad 0 \leqslant t \leqslant {T_p} $$ (2) 式中,
${f_0}$ 表示载波频率;$ \Delta f $ 表示相邻天线的频率增量(对应的雷达最大无模糊距离等于${c \mathord{\left/ {\vphantom {c {2\Delta f}}} \right. } {2\Delta f}}$ ;$c$ 表示电磁波速度,且$ \Delta f \ll {f_0} $ );$ {T_p} $ 表示脉冲宽度;$ {g_n}( t ) $ 表示发射的基带波形,满足式(3)所给的正交性质[21]:$$ \int_0^{{T_p}} {{g_{{n_1}}}( t )} g_{{n_2}}^ * ( {t - \tau } ){{\mathrm{e}}^{{\mathrm{j}}2 {\text π} \Delta f( {{n_1} - {n_2}} )t}}{\mathrm{d}}t = \left\{ {\begin{array}{*{20}{c}} 1&{{n_1} = {n_2}} \\ 0&{{n_1} \ne {n_2}} \end{array}} \right. $$ (3) 式中,
$ \tau $ 表示时延;$ * $ 表示共轭。假设有$P$ 个静止目标位于雷达的远场区域,第$p$ 个目标相对于雷达的径向距离为$ {r_p} $ ,相对于天线法线的角度为$ {\theta _p} $ ,且${r_p} \in ( {0,{c \mathord{\left/ {\vphantom {c {2\Delta f}}} \right. } {2\Delta f}}} )$ ,${\theta _p} \in ( { - 0.5 {\text π} ,0.5 {\text π} } )$ ,发射和接收天线单元间距均等于$0.5\lambda $ ($\lambda $ 对应载波${f_0}$ 的波长)。雷达工作时,每个接收天线首先对接收信号进行下变频处理,然后将下变频信号输入到与${N_{\mathrm{t}}}$ 个发射波形相对应的匹配滤波器组,基于式(3)所给的正交性,可分离出目标对${N_{\mathrm{t}}}$ 个发射波形的反射回波,因此接收阵列在第$l$ 个脉冲收到第$p$ 个目标对发射信号$ {s_n}( t ) $ 的反射信号为:$$ {{\boldsymbol{y}}_{n,p}}( l ) = {\alpha _p}( l ){{\mathrm{e}}^{{\mathrm{j}}( {n - 1} ) {\text π} \sin ( {{\theta _p}} )}}{{\mathrm{e}}^{\tfrac{{ - {\mathrm{j}}4 {\text π} ( {n - 1} )\Delta f{r_p}}}{c}}}{{\boldsymbol{a}}_r}\big( {{\theta _p}} \big) $$ (4) $$ {{\boldsymbol{a}}_r}( {{\theta _p}} ) = {\left[ {1,{{\mathrm{e}}^{{\mathrm{j}} {\text π} \sin ( {{\theta _p}} )}}, \cdots ,{{\mathrm{e}}^{{\mathrm{j}} {\text π} ( {{M_r} - 1} )\sin ( {{\theta _p}} )}}} \right]^{\rm{T}}} $$ (5) 式中,
$ {{\boldsymbol{a}}_r}\big( {{\theta _p}} \big) $ 表示接收线阵导向向量;$ {\alpha _p}( l ) $ 表示目标回波强度;${\mathrm{T}}$ 表示转置。由式(4)可得接收阵列在第$l$ 个脉冲收到$P$ 个目标的回波信号为:$$ {\boldsymbol{y}}( l ) = \sum\limits_{p = 1}^P {{\alpha _p}( l ){\boldsymbol{a}}\big( {{\theta _p},{r_p}} \big)} + {\boldsymbol{w}}( l )= {\boldsymbol{A\alpha }}( l ) + {\boldsymbol{w}}( l ) $$ (6) 式中,
$ {\boldsymbol{A}} = \left[ {{\boldsymbol{a}}( {{\theta _1},{r_1}} ), {\boldsymbol{a}}( {{\theta _2},{r_2}} ), \cdots ,{\boldsymbol{a}}( {{\theta _P},{r_P}} )} \right] $ ;$ {\boldsymbol{\alpha }} ( l ) = \left[ {\alpha _1} ( l ),\right. \left.{\alpha _2}( l ), \cdots ,{\alpha _P}( l ) \right]^{\rm{T}} $ ;$ {\boldsymbol{w}}( l ) $ 为高斯白噪声。其中:$$ {\boldsymbol{a}}\big( {{\theta _p},{r_p}} \big) = {\boldsymbol{b}}\big( {{\theta _p},{r_p}} \big) \otimes {{\boldsymbol{a}}_r}\big( {{\theta _p}} \big) $$ (7) $$ {\boldsymbol{b}}\big( {{\theta _p},{r_p}} \big) = {\boldsymbol{d}}\big( {{r_p}} \big) \odot {{\boldsymbol{a}}_t}\big( {{\theta _p}} \big) $$ (8) $$ {\boldsymbol{d}}\big( {{r_p}} \big) = {\left[ {1,{{\mathrm{e}}^{\tfrac{{ - {\mathrm{j}}4 {\text π} \Delta f{r_p}}}{c}}}, \cdots ,{{\mathrm{e}}^{\tfrac{{ - {\mathrm{j}}4 {\text π} ( {{N_t} - 1} )\Delta f{r_p}}}{c}}}} \right]^{\rm{T}}} $$ (9) $$ {{\boldsymbol{a}}_t}\big( {{\theta _p}} \big) = {\left[ {1,{{\mathrm{e}}^{{\mathrm{j}} {\text π} \sin ( {{\theta _p}} )}}, \cdots ,{{\mathrm{e}}^{{\mathrm{j}} {\text π} ( {{N_t} - 1} )\sin ( {{\theta _p}} )}}} \right]^{\rm{T}}} $$ (10) 式中,
$ \otimes $ 和$ \odot $ 分别表示Kronecker积和Hadamard积。假设目标散射特性满足SwerlingⅡ起伏模型,不同目标的回波强度以及噪声之间相互独立,
$ {{\boldsymbol{R}}_t} = {\mathrm{E}}\left[ {{\boldsymbol{\alpha }}( l ){{\boldsymbol{\alpha }}^{\mathrm{H}}}( l )} \right] = {\mathrm{diag}}\left[ {\sigma _1^2,\sigma _2^2, \cdots ,\sigma _P^2} \right] $ ,其中,${\mathrm{H}}$ 表示共轭转置,$ {\mathrm{diag}}\left[ {} \right] $ 表示对角矩阵,$ \sigma _p^2 $ 表示目标$p$ 的回波功率。噪声协方差矩阵满足$ {\mathrm{E}}\left[ {{\boldsymbol{w}}( l ){{\boldsymbol{w}}^{\mathrm{H}}}( l )} \right] = \sigma _w^2{{\boldsymbol{I}}_{{N_t}{M_r}}} $ ,其中,$ {\boldsymbol{I}} $ 表示单位矩阵,$ \sigma _w^2 $ 表示噪声功率。因此接收信号$ {\boldsymbol{y}}( l ) $ 的协方差矩阵为:$$ {\boldsymbol{R}} = {\mathrm{E}}\left[ {{\boldsymbol{y}}( l ){{\boldsymbol{y}}^{\mathrm{H}}}( l )} \right] = {\boldsymbol{A}}{{\boldsymbol{R}}_t}{{\boldsymbol{A}}^{\mathrm{H}}} + \sigma _w^2{{\boldsymbol{I}}_{{N_t}{M_r}}} $$ (11) -
由式(6)可知,阵元空间的数据维数为
${N_t}{M_r}$ ,对于大型阵列,大的数据维数将造成计算复杂和工程实现困难。本节首先推导用于FDA-MIMO雷达参数估计的卷积波束空间传播算子算法(CBS PM),包括卷积波束空间变换、角度/距离相位信息估计和角度/距离解模糊处理3个步骤。然后给出估计的克拉美-罗界(Cramer-Rao Bounds, CRB),分析CBS PM算法的性能。 -
卷积波束空间处理包括卷积处理和抽取处理,如图1所示,覆盖
$\omega \in \left[ { - {\text π} , {\text π} } \right]$ 的滤波器组由$I$ 个滤波器构成($i = 0,1, \cdots ,I - 1$ )。第$i$ 个滤波器的频率响应为${H_i}\big( {{{\mathrm{e}}^{{\mathrm{j}}\omega }}} \big)$ ,通带为$\left[ {\dfrac{{ - ( {i + 1} ) {\text π} }}{I},\dfrac{{ - i {\text π} }}{I}} \right] \cup \left[ {\dfrac{{i {\text π} }}{I},\dfrac{{( {i + 1} ) {\text π} }}{I}} \right]$ ,其单位冲激响应${h_i}( n )$ ($n = 0,1, \cdots ,{Q_i} - 1$ )的自相关函数为:$$ {\varphi _i}( k ) = \sum\limits_{n = 0}^{{Q_i} - 1} {{h_i}( n )} h_i^ * ( {n - k} ) $$ (12) 且满足奈奎斯特性质[13] :
$$ \left[ {{\varphi _i}( 0 ),\varphi _i^*( I ),\varphi _i^*( {( {M_{{\mathrm{dec}}}^i - 1} )I} )} \right] = \delta ( k ) $$ (13) 式中,
$M_{{\mathrm{dec}}}^i = \left\lfloor {{{( {{M_r} - {Q_i} + 1} )} \mathord{\left/ {\vphantom {{( {{M_r} - {Q_i} + 1} )} I}} \right. } I}} \right\rfloor $ ;$ \delta ( k ) $ 为单位冲激序列。式(4)所给目标
$p$ 的反射信号$ {{\boldsymbol{y}}_{n,p}}( l ) $ 为一个空间角频率$ {\omega _{s,p}} = {\text π} \sin \big( {{\theta _p}} \big) $ 的复正弦序列。不失一般性,假设$ {\omega _{s,p}} $ 位于${H_i}\big( {{{\mathrm{e}}^{{\mathrm{j}}\omega }}} \big)$ 的通带内,将$ {{\boldsymbol{y}}_{n,p}}( l ) $ 与${h_i}( n )$ 进行卷积处理,所得稳态响应为:$$ \begin{split} &\qquad \qquad {\boldsymbol{z}}_{n,p}^i( l ) = {{\boldsymbol{H}}_i}{{\boldsymbol{y}}_{n,p}}( l )= \\ & \beta _p^i( l ){{\mathrm{e}}^{{\mathrm{j}}( {n - 1} ) {\text π} \sin ( {{\theta _p}} )}}{{\mathrm{e}}^{\tfrac{{ - {\mathrm{j}}4 {\text π} ( {n - 1} )\Delta f{r_p}}}{c}}}{{\boldsymbol{a}}_{{\mathrm{rc}}}}\big( {{\theta _p}} \big) \end{split} $$ (14) 式中,
$ {{\boldsymbol{H}}_i} $ 为$( {{M_r} - {Q_i} + 1} ) \times {M_r}$ 维矩阵:$${\begin{split} &\qquad \qquad \qquad \qquad \qquad \boldsymbol{H}_i= \\ & {\left[ \begin{array}{cccccc} h_i(Q_i-1) & \cdots & h_i(0) & 0 & \cdots & 0 \\ 0 & h_i(Q_i-1) & \cdots & h_i(0) & \cdots & 0 \\ \vdots & \vdots & \text{} & \vdots & \text{} & \vdots \\ 0 & 0 & \cdots & h_i(Q_i-1) & \cdots & h_i(0) \end{array} \right]} \end{split}} $$ (15) $$ \beta _p^i( l ) = {\alpha _p}( l ){H_i}\big( {{{\mathrm{e}}^{{\mathrm{j}}{\omega _{s,p}}}}} \big){{\mathrm{e}}^{{\mathrm{j}}( {{Q_i} - 1} ){\omega _{s,p}}}} $$ (16) $$ {{\boldsymbol{a}}_{{\mathrm{rc}}}}\big( {{\theta _p}} \big) = {{\boldsymbol{J}}_{( {{M_r} - {Q_i} + 1} ) \times {M_r}}}{{\boldsymbol{a}}_r}\big( {{\theta _p}} \big) $$ (17) $ {{\boldsymbol{J}}_{( {{M_r} - {Q_i} + 1} ) \times {M_r}}} $ 表示$( {{M_r} - {Q_i} + 1} ) \times {M_r}$ 维选择矩阵:$$ {{\boldsymbol{J}}_{( {{M_r} - {Q_i} + 1} ) \times {M_r}}} = \left[ {\begin{array}{*{20}{c}} 1&0& \cdots &0 \\ 0&1& \cdots &0 \\ \vdots & \vdots & \text{} & \vdots \\ 0&0& \cdots &1 \end{array}} \right] $$ (18) 令
$ {{\boldsymbol{y}}_p}( l ) = {\left[ {{{( {{{\boldsymbol{y}}_{1,p}}( l )} )}^{\rm{T}}},{{( {{{\boldsymbol{y}}_{2,p}}( l )} )}^{\rm{T}}} \cdots ,{{( {{{\boldsymbol{y}}_{{N_t},p}}( l )} )}^{\rm{T}}}} \right]^{\rm{T}}} $ 表示阵列收到目标$p$ 对所有发射信号的反射回波,$ {\boldsymbol{z}}_p^i( l ) = {\left[ {{{( {{\boldsymbol{z}}_{1,p}^i( l )} )}^{\rm{T}}},{{( {{\boldsymbol{z}}_{2,p}^i( l )} )}^{\rm{T}}} \cdots ,{{( {{\boldsymbol{z}}_{{N_t},p}^i( l )} )}^{\rm{T}}}} \right]^{\rm{T}}} $ 表示将$ {{\boldsymbol{y}}_p}( l ) $ 变换到卷积波束空间的数据:$$ {\boldsymbol{z}}_p^i( l ) = {{\boldsymbol{\varPsi }}_i}{{\boldsymbol{y}}_p}( l ) = \beta _p^i( l ){{\boldsymbol{a}}_{{\mathrm{con}}}}\big( {{\theta _p},{r_p}} \big) $$ (19) 式中,
${\boldsymbol{\varPsi}}_i $ 为$( {{M_r} - {Q_i} + 1} ){N_t} \times {M_r}{N_t}$ 维块对角阵:$$ {{\boldsymbol{\varPsi }}_i} = \left[ {\begin{array}{*{20}{c}} {{{\boldsymbol{H}}_i}}&{\boldsymbol{0}}& \cdots &{\boldsymbol{0}} \\ {\boldsymbol{0}}&{{{\boldsymbol{H}}_i}}& \cdots &{\boldsymbol{0}} \\ \vdots & \vdots & \text{} & \vdots \\ {\boldsymbol{0}}&{\boldsymbol{0}}& \cdots &{{{\boldsymbol{H}}_i}} \end{array}} \right] $$ (20) $$ {{\boldsymbol{a}}_{{\mathrm{con}}}}\big( {{\theta _p},{r_p}} \big) = {\boldsymbol{b}}\big( {{\theta _p},{r_p}} \big) \otimes {{\boldsymbol{a}}_{{\mathrm{rc}}}}\big( {{\theta _p}} \big) $$ (21) -
令
${{\boldsymbol{e}}_m}( {m = 0,1, \cdots ,I - 1} )$ 表示第$m + 1$ 元素为1、其他元素为0的$ {M_r} - {Q_i} + 1 $ 维列向量,定义抽取矩阵$ {{\boldsymbol{E}}_m} = {\left[ {{{\boldsymbol{e}}_m},{{\boldsymbol{e}}_{m + I}}, \cdots ;{{\boldsymbol{e}}_{m + ( {M_{{\mathrm{dec}}}^i - 1} )I}}} \right]^{\rm{T}}} $ ,用正整数$I$ 对卷积结果$ {\boldsymbol{z}}_{n,p}^i( l ) $ 进行抽取,可获得$I$ 组抽取数据:$$ \begin{split} &\qquad\qquad {\boldsymbol{x}}_{n,p}^{i,m}( l ) = {{\boldsymbol{E}}_m}{\boldsymbol{z}}_{n,p}^i( l )= \\ & \gamma _{p,m}^i( l ){{\mathrm{e}}^{{\mathrm{j}}( {n - 1} ) {\text π} \sin ( {{\theta _p}} )}}{{\mathrm{e}}^{\tfrac{{ - {\mathrm{j}}4 {\text π} ( {n - 1} )\Delta f{r_p}}}{c}}}{{\boldsymbol{a}}_{{\mathrm{rdec}}}}\big( {{\theta _p}} \big) \end{split} $$ (22) 式中,
$m = 0,1, \cdots ,I - 1$ ,$$ \gamma _{p,m}^i( l ) = \beta _p^i( l ){{\mathrm{e}}^{{\mathrm{j}}m{\omega _{s,p}}}} $$ (23) $$ {{\boldsymbol{a}}_{{\mathrm{rdec}}}}\big( {{\theta _p}} \big) = {\left[ {1,{{\mathrm{e}}^{{\mathrm{j}}I {\text π} \sin ( {{\theta _p}} )}}, \cdots ,{{\mathrm{e}}^{{\mathrm{j}}( {M_{dec}^i - 1} )I {\text π} \sin ( {{\theta _p}} )}}} \right]^{\rm{T}}} $$ (24) 由式(22)可将对应
${N_{\mathrm{t}}}$ 个发射波形的抽取数据$ {\boldsymbol{x}}_p^{i,m}( l ) = {\left[ {{{( {{\boldsymbol{x}}_{1,p}^{i,m}( l )} )}^{\rm{T}}},{{( {{\boldsymbol{x}}_{2,p}^{i,m}( l )} )}^{\rm{T}}}, \cdots ,{{( {{\boldsymbol{x}}_{{N_t},p}^{i,m}( l )} )}^{\rm{T}}}} \right]^{\rm{T}}} $ 表示成:$$ {\boldsymbol{x}}_p^{i,m}( l ) = {{\boldsymbol{D}}_m}{\boldsymbol{z}}_p^i( l ) = \gamma _{p,m}^i( l ){{\boldsymbol{a}}_{{\mathrm{cd}}}}\big( {{\theta _p},{r_p}} \big) $$ (25) 式中,
${\boldsymbol{D}}_m $ 为$M_{{\mathrm{dec}}}^i{N_t} \times ( {{M_r} - {Q_i} + 1} ){N_t}$ 维块对角阵:$$ {{\boldsymbol{D}}_m} = \left[ {\begin{array}{*{20}{c}} {{{\boldsymbol{E}}_m}}&{\boldsymbol{0}}& \cdots &{\boldsymbol{0}} \\ {\boldsymbol{0}}&{{{\boldsymbol{E}}_m}}& \cdots &{\boldsymbol{0}} \\ \vdots & \vdots & \text{} & \vdots \\ {\boldsymbol{0}}&{\boldsymbol{0}}& \cdots &{{{\boldsymbol{E}}_m}} \end{array}} \right] $$ (26) $$ {{\boldsymbol{a}}_{{\mathrm{cd}}}}( {{\theta _p},{r_p}} ) = {\boldsymbol{b}}( {{\theta _p},{r_p}} ) \otimes {{\boldsymbol{a}}_{{\mathrm{rdec}}}}( {{\theta _p}} ) $$ (27) 由式(25)可知,
$ {\boldsymbol{x}}_p^{i,m}( l ) $ 的数据维数降为Nt ×$ ( {{M_r} - {Q_i} + 1} )/I$ 。当${M_r} \gg {Q_i}$ ,$ {\boldsymbol{x}}_p^{i,m}( l ) $ 的维数是阵元空间数据$ {{\boldsymbol{y}}_p}( l ) $ 的$1/I$ 。因此高斯白噪声背景下,当滤波器${H_i}\big( {{e^{j\omega }}} \big)$ 通带内有${P_i}(P = \displaystyle\sum\limits_{i = 0}^{I - 1} {{P_i}}) $ 个目标时,目标的回波经卷积和抽取处理可表示为:$$ {{\boldsymbol{x}}^{i,m}}( l ) = {{\boldsymbol{A}}_{{\mathrm{cd}}}}{\boldsymbol{\gamma }}_m^i( l ) + {\boldsymbol{w}}_{{\mathrm{cd}}}^{i,m}( l )\quad m = 0,1, \cdots ,I - 1 $$ (28) 式中,
$ {{\boldsymbol{A}}_{{\mathrm{cd}}}} = \left[ {{{\boldsymbol{a}}_{{\mathrm{cd}}}} ( {{\theta _1},{r_1}} ) ,{{\boldsymbol{a}}_{{\mathrm{cd}}}} ( {{\theta _2},{r_2}} ), \cdots ,{{\boldsymbol{a}}_{{\mathrm{cd}}}}( {{\theta _{{P_i}}},{r_{{P_i}}}} )} \right] $ ;$ {\boldsymbol{\gamma }}_m^i( l ) = {\left[ {\gamma _{1,m}^i( l ),\gamma _{2,m}^i( l ) ,\cdots ,\gamma _{{P_i},m}^i( l )} \right]^{\rm{T}}} $ ;$ {\boldsymbol{w}}_{{\mathrm{cd}}}^{i,m}( l ) = {{\boldsymbol{D}}_m}{{\boldsymbol{\varPsi }}_i}{\boldsymbol{w}}( l ) $ 。由式(28)可知噪声
$ {\boldsymbol{w}}_{{\mathrm{cd}}}^{i,m}( l ) $ 的协方差矩阵$ {{\boldsymbol{D}}_m}{{\boldsymbol{\varPsi }}_i}{\mathrm{E}}\left[ {{\boldsymbol{w}}( l ){{\boldsymbol{w}}^{\mathrm{H}}}( l )} \right]{\boldsymbol{\varPsi }}_i^{\mathrm{H}}{{\boldsymbol{D}}_m} $ 可表示为:$$ \sigma _w^2\left[ {\begin{array}{*{20}{c}} {{{\boldsymbol{E}}_m}{{\boldsymbol{{{H}}}}_i}{\boldsymbol{H}}_i^{\mathrm{H}}{{\boldsymbol{E}}_m}}&{\boldsymbol{0}}& \cdots &{\boldsymbol{0}} \\ {\boldsymbol{0}}&{{{\boldsymbol{E}}_m}{{\boldsymbol{H}}_i}{\boldsymbol{H}}_i^{\mathrm{H}}{{\boldsymbol{E}}_m}}& \cdots &{\boldsymbol{0}} \\ \vdots & \vdots & \text{} & \vdots \\ {\boldsymbol{0}}&{\boldsymbol{0}}& \cdots &{{{\boldsymbol{E}}_m}{{\boldsymbol{H}}_i}{\boldsymbol{H}}_i^{\mathrm{H}}{{\boldsymbol{E}}_m}} \end{array}} \right] $$ (29) 式中,
$ {{\boldsymbol{E}}_m}{{\boldsymbol{H}}_i}{\boldsymbol{H}}_i^{\mathrm{H}}{{\boldsymbol{E}}_m} $ 是一个Hermitian Toeplitz矩阵,其第一行为$ \left[ {{\varphi _i}( 0 ),\varphi _i^*( I ),\varphi _i^*\big( {\big( {M_{{\mathrm{dec}}}^i - 1} \big)I} \big)} \right] $ 。当${H_i}\big( {{{\mathrm{e}}^{{\mathrm{j}}\omega }}} \big)$ 取为仿酉滤波器组中的滤波器,或其在通带内具有良好带内起伏特性时,滤波器${H_i}\big( {{{\mathrm{e}}^{{\mathrm{j}}\omega }}} \big)$ 满足式(13)所给的奈奎斯特性质[13],此时:$$ {{\boldsymbol{E}}_m}{{\boldsymbol{H}}_i}{\boldsymbol{H}}_i^{\mathrm{H}}{{\boldsymbol{E}}_m} = {{\boldsymbol{I}}_{M_{{\mathrm{dec}}}^i}} $$ (30) $$ {\mathrm{E}}\left[ {{\boldsymbol{w}}_{{\mathrm{cd}}}^{i,m}( l ){{( {{\boldsymbol{w}}_{{\mathrm{cd}}}^{i,m}( l )} )}^{\mathrm{H}}}} \right] = \sigma _w^2{{\boldsymbol{I}}_{{N_t}M_{{\mathrm{dec}}}^i}} $$ (31) 故噪声
$ {\boldsymbol{w}}_{{\mathrm{cd}}}^{i,m}( l ) $ 仍然为高斯白噪声。 -
令矩阵
$ {{\boldsymbol{A}}_{{\mathrm{cd}},1}} $ 和$ {{\boldsymbol{A}}_{{\mathrm{cd}},2}} $ 分别表示矩阵$ {{\boldsymbol{A}}_{{\mathrm{cd}}}} $ 的前$ {P_i} $ 行和后${N_t}M_{{\mathrm{dec}}}^i - {P_i}$ 行,即$ {{\boldsymbol{A}}_{{\mathrm{cd}}}} = {\left[ {{\boldsymbol{A}}_{{\mathrm{cd}},1}^{\rm{T}},{\boldsymbol{A}}_{{\mathrm{cd}},2}^{\rm{T}}} \right]^{\rm{T}}} $ 。由于$ {{\boldsymbol{A}}_{{\mathrm{cd}},1}} $ 列满秩,所以存在线性算子$ {{\boldsymbol{P}}_{{\mathrm{cd}}}} $ 满足[18-19]:$$ {\boldsymbol{P}}_{{\mathrm{cd}}}^{\mathrm{H}}{{\boldsymbol{A}}_{{\mathrm{cd}},1}} = {{\boldsymbol{A}}_{{\mathrm{cd}},2}} $$ (32) 令传播算子
$ {{\boldsymbol{P}}_{{\mathrm{CBS}}}} = {\left[ {{{\boldsymbol{I}}_{{P_i}}},{\boldsymbol{P}}_{{\mathrm{cd}}}^*} \right]^{\rm{T}}} $ ,则由式(32)可得:$$ {{\boldsymbol{P}}_{{\mathrm{CBS}}}}{{\boldsymbol{A}}_{{\mathrm{cd}},1}} = {{\boldsymbol{A}}_{{\mathrm{cd}}}} $$ (33) 定义选择矩阵:
$$ {{\boldsymbol{J}}_{r,1}} = {{\boldsymbol{I}}_{{N_t}}} \otimes \left[ {{{\boldsymbol{I}}_{M_{{\mathrm{dec}}}^i - 1}},{{\boldsymbol{0}}_{( {M_{{\mathrm{dec}}}^i - 1} ) \times 1}}} \right] $$ (34) $$ {{\boldsymbol{J}}_{r,2}} = {{\boldsymbol{I}}_{{N_t}}} \otimes \left[ {{{\boldsymbol{0}}_{( {M_{{\mathrm{dec}}}^i - 1} ) \times 1}},{{\boldsymbol{I}}_{M_{{\mathrm{dec}}}^i - 1}}} \right] $$ (35) $$ {{\boldsymbol{J}}_{t,1}} = \left[ {{{\boldsymbol{I}}_{{N_t} - 1}},{{\boldsymbol{0}}_{( {{N_t} - 1} ) \times 1}}} \right] \otimes {{\boldsymbol{I}}_{M_{{\mathrm{dec}}}^i}} $$ (36) $$ {{\boldsymbol{J}}_{t,2}} = \left[ {{{\boldsymbol{0}}_{( {{N_t} - 1} ) \times 1}},{{\boldsymbol{I}}_{{N_t} - 1}}} \right] \otimes {{\boldsymbol{I}}_{M_{{\mathrm{dec}}}^i}} $$ (37) 令
$ {{\boldsymbol{P}}_{r,1}} = {{\boldsymbol{J}}_{r,1}}{{\boldsymbol{P}}_{{\mathrm{CBS}}}} $ ,$ \boldsymbol{P}_{r,2}=\boldsymbol{J}_{r,2}P_{\mathrm{CBS}} $ ,$ {{\boldsymbol{P}}_{t,1}} = {{\boldsymbol{J}}_{t,1}}{{\boldsymbol{P}}_{{\mathrm{CBS}}}} $ ,$ {{\boldsymbol{P}}_{t,2}} = {{\boldsymbol{J}}_{t,2}}{{\boldsymbol{P}}_{{\mathrm{CBS}}}} $ ,由式(33)~式(37)可得:$$ {{\boldsymbol{P}}_{r,2}}{{\boldsymbol{A}}_{{\mathrm{cd}},1}} = {{\boldsymbol{P}}_{r,1}}{{\boldsymbol{A}}_{{\mathrm{cd}},1}}{{\boldsymbol{\varphi }}_r} $$ (38) $$ {{\boldsymbol{P}}_{t,2}}{{\boldsymbol{A}}_{{\mathrm{cd}},1}} = {{\boldsymbol{P}}_{t,1}}{{\boldsymbol{A}}_{{\mathrm{cd}},1}}{{\boldsymbol{\varphi }}_t} $$ (39) 其中,
$$ {{\boldsymbol{\varphi }}_r} = {\mathrm{diag}}\left[ {{{\mathrm{e}}^{{\mathrm{j}}I {\text π} \sin ( {{\theta _1}} )}},{{\mathrm{e}}^{{\mathrm{j}}I {\text π} \sin ( {{\theta _2}} )}}, \cdots ,{{\mathrm{e}}^{{\mathrm{j}}I {\text π} \sin ( {{\theta _{{P_i}}}} )}}} \right] $$ (40) $$ \begin{split} {{\boldsymbol{\varphi }}_t} =& {\mathrm{diag}}\left[ {{\mathrm{e}}^{{\mathrm{j}} {\text π} \sin ( {{\theta _1}} ) - \tfrac{{{\mathrm{j}}4 {\text π} \Delta f{r_1}}}{c}}},{{\mathrm{e}}^{{\mathrm{j}} {\text π} \sin ( {{\theta _2}} ) - \tfrac{{{\mathrm{j}}4 {\text π} \Delta f{r_2}}}{c}}}, \cdots ,\right.\\ &\qquad \qquad \left.{{\mathrm{e}}^{{\mathrm{j}} {\text π} \sin ( {{\theta _{{P_i}}}} ) - \tfrac{{{\mathrm{j}}4 {\text π} \Delta f{r_{p_i}}}}{c}}} \right] \end{split} $$ (41) 由式(38)和式(39)可得:
$$ {( {{{\boldsymbol{P}}_{r,1}}} )^\dagger }{{\boldsymbol{P}}_{r,2}} = {{\boldsymbol{A}}_{{\mathrm{cd}},1}}{{\boldsymbol{\varphi }}_r}{\boldsymbol{A}}_{{\mathrm{cd}},1}^{ - 1} $$ (42) $$ {( {{{\boldsymbol{P}}_{t,1}}} )^\dagger }{{\boldsymbol{P}}_{t,2}} = {{\boldsymbol{A}}_{{\mathrm{cd}},1}}{{\boldsymbol{\varphi }}_t}{\boldsymbol{A}}_{{\mathrm{cd}},1}^{ - 1} $$ (43) 式中,
$ \dagger $ 表示伪逆。由式(42)和式(43)可知,$ {( {{{\boldsymbol{P}}_{r,1}}} )^\dagger }{{\boldsymbol{P}}_{r,2}} $ 和$ {( {{{\boldsymbol{P}}_{t,1}}} )^\dagger }{{\boldsymbol{P}}_{t,2}} $ 的特征值分别对应$ {{\boldsymbol{\varphi }}_r} $ 和$ {{\boldsymbol{\varphi }}_t} $ 主对角线元素。因此求出算子$ {{\boldsymbol{P}}_{{\mathrm{CBS}}}} $ 的估计值,再由2.3节所给的解模糊和配对处理即可估计出${H_i}\big( {{{\mathrm{e}}^{{\mathrm{j}}\omega }}} \big)$ 通带内${P_i}$ 个目标的角度和距离。同理可得其他$I - 1$ 个滤波器通带内目标的角度、距离估计值。以下给出基于
$ I $ 组抽取数据$ {{\boldsymbol{x}}^{i,m}}( l ) $ ($ m = 0,1, \cdots ,I - 1 $ )求算子$ {{\boldsymbol{P}}_{{\mathrm{CBS}}}} $ 估计值的方法。在无噪声的条件下,由式(28)可得$ {{\boldsymbol{x}}^{i,m}}( l ) $ 的协方差矩阵:$$ {\boldsymbol{R}}_m^i = {\mathrm{E}}\left[ {{{\boldsymbol{x}}^{i,m}}( l ){{( {{{\boldsymbol{x}}^{i,m}}( l )} )}^{\mathrm{H}}}} \right] = {{\boldsymbol{A}}_{{\mathrm{cd}}}}{\boldsymbol{R}}_s^i{\boldsymbol{A}}_{{\mathrm{cd}}}^{\mathrm{H}} $$ (44) 其中,
$$ \begin{split} &\qquad \qquad{\boldsymbol{R}}_s^i = {\mathrm{E}}\left\{ {{\boldsymbol{\gamma }}_m^i( l ){{\left[ {{\boldsymbol{\gamma }}_m^i( l )} \right]}^{\mathrm{H}}}} \right\}= \\ & {\mathrm{diag}}\left[ {{{\left| {{H_i}\big( {{{\mathrm{e}}^{{\mathrm{j}}{\omega _{s,1}}}}} \big)} \right|}^2}\sigma _1^2, \cdots ,{{\left| {{H_i}\big( {{{\mathrm{e}}^{{\mathrm{j}}{\omega _{s,{P_i}}}}}} \big)} \right|}^2}\sigma _{{P_i}}^2} \right] \end{split} $$ (45) 定义
$ I $ 组抽取数据的平均协方差矩阵$ {\boldsymbol{R}}_{{\mathrm{av}}}^i = \dfrac{1}{I}\displaystyle\sum\limits_{m = 0}^{I - 1} {{\boldsymbol{R}}_m^i} $ ,由式(44)可知$ {\boldsymbol{R}}_{{\mathrm{av}}}^i = {{\boldsymbol{A}}_{{\mathrm{cd}}}}{\boldsymbol{R}}_s^i{\boldsymbol{A}}_{{\mathrm{cd}}}^{\mathrm{H}} $ ,令$ {\boldsymbol{R}}_{{\mathrm{av}},1}^i $ 和$ {\boldsymbol{R}}_{{\mathrm{av}},2}^i $ 分别表示$ {\boldsymbol{R}}_{{\mathrm{av}}}^i $ 的前${P_i}$ 行和后${N_t}M_{{\mathrm{dec}}}^i - {P_i}$ 行,则:$$ {\boldsymbol{R}}_{{\mathrm{av}},1}^i = {{\boldsymbol{A}}_{{\mathrm{cd}},1}}{\boldsymbol{R}}_s^i{\boldsymbol{A}}_{{\mathrm{cd}}}^{\mathrm{H}} $$ (46) $$ {\boldsymbol{R}}_{{\mathrm{av}},2}^i = {{\boldsymbol{A}}_{{\mathrm{cd}},2}}{\boldsymbol{R}}_s^i{\boldsymbol{A}}_{{\mathrm{cd}}}^{\mathrm{H}} $$ (47) 由式(32)、式(46)和式(47)可得:
$$ {\boldsymbol{R}}_{{\mathrm{av}},2}^i = {\boldsymbol{P}}_{{\mathrm{cd}}}^{\mathrm{H}}{\boldsymbol{R}}_{{\mathrm{av}},1}^i $$ (48) 当雷达工作在高斯白噪声背景,且只有
$ L $ 组采样数据时,线性算子$ {{\boldsymbol{P}}_{{\mathrm{cd}}}} $ 的估计值为:$$ {{\boldsymbol{\hat P}}_{{\mathrm{cd}}}} = {\left[ {{\boldsymbol{\hat R}}_{{\mathrm{av}},1}^i{{( {{\boldsymbol{\hat R}}_{{\mathrm{av}},1}^i} )}^{\mathrm{H}}}} \right]^{ - 1}}{\boldsymbol{\hat R}}_{{\mathrm{av}},1}^i{( {{\boldsymbol{\hat R}}_{{\mathrm{av}},2}^i} )^{\mathrm{H}}} $$ (49) 式中,
$ {\boldsymbol{\hat R}}_{{\mathrm{av}},1}^i $ 和$ {\boldsymbol{\hat R}}_{{\mathrm{av}},2}^i $ 分别表示矩阵$ {\boldsymbol{\hat R}}_{{\mathrm{av}}}^i $ 的前${P_i}$ 行和后${N_t}M_{{\mathrm{dec}}}^i - {P_i}$ 行,$ {\boldsymbol{\hat R}}_{{\mathrm{av}}}^i $ 为$ {\boldsymbol{R}}_{{\mathrm{av}}}^i $ 的估计值:$$ {\boldsymbol{\hat R}}_{{\mathrm{av}}}^i = \frac{1}{I}\sum\limits_{m = 0}^{I - 1} {{\boldsymbol{\hat R}}_m^i} = \frac{1}{I}\sum\limits_{m = 0}^{I - 1} {\left[ {\frac{1}{L}\sum\limits_{l = 1}^L {{{\boldsymbol{x}}^{i,m}}( l ){{\big( {{{\boldsymbol{x}}^{i,m}}( l )} \big)}^{\mathrm{H}}}} } \right]} $$ (50) 由式(49)可得传播算子估计值:
$$ {{\boldsymbol{\hat P}}_{{\mathrm{CBS}}}} = {\left[ {{{\boldsymbol{I}}_{{P_i}}},{\boldsymbol{\hat P}}_{{\mathrm{cd}}}^*} \right]^{\rm{T}}} $$ (51) 故
$ {{\boldsymbol{P}}_{r,1}} $ 、$ {{\boldsymbol{P}}_{r,2}} $ 、$ {{\boldsymbol{P}}_{t,1}} $ 和$ {{\boldsymbol{P}}_{t,2}} $ 的估计值分别为$ {{\boldsymbol{\hat P}}_{r,1}} = {{\boldsymbol{J}}_{r,1}}{{\boldsymbol{\hat P}}_{{\mathrm{CBS}}}} $ 、$ {{\boldsymbol{\hat P}}_{r,2}} = {{\boldsymbol{J}}_{r,2}}{{\boldsymbol{\hat P}}_{{\mathrm{CBS}}}} $ 、$ {{\boldsymbol{\hat P}}_{t,1}} = {{\boldsymbol{J}}_{t,1}}{{\boldsymbol{\hat P}}_{{\mathrm{CBS}}}} $ 和$ {{\boldsymbol{\hat P}}_{t,2}} = {{\boldsymbol{J}}_{t,2}}{{\boldsymbol{\hat P}}_{{\mathrm{CBS}}}} $ 。 -
令
$ \arg ( z ) $ 表示复数$ z $ 的相位在主值区间$[ { - {\text π} , {\text π} } ) $ 的值,$ \arg \big( {{\boldsymbol{\hat \varphi }}_{{p_i},{p_i}}^r} \big) $ 表示$ {{\boldsymbol{\varphi }}_r} $ 主对角线元素$ {{\mathrm{e}}^{{\mathrm{j}}I {\text π} \sin ( {{\theta _{{p_i}}}} )}} $ 相位主值的估计值。由于$ {\omega _{s,{p_i}}} = {\text π} \sin \big( {{\theta _{{p_i}}}} \big) $ 位于${H_i}\big( {{{\mathrm{e}}^{{\mathrm{j}}\omega }}} \big)$ 的通带内,故式(40)中元素$ {{\mathrm{e}}^{{\mathrm{j}}I {\text π} \sin ( {{\theta _{{p_i}}}} )}} $ 的相位位于区间$\left[ { - ( {i + 1} ) {\text π} , - i {\text π} } \right] \cup \left[ {i {\text π} ,( {i + 1} ) {\text π} } \right]$ ,由$ {\big( {{{{\boldsymbol{\hat P}}}_{r,1}}} \big)^\dagger }{{\boldsymbol{\hat P}}_{r,2}} $ 的特征值估计出的$ \arg \big( {{\boldsymbol{\hat \varphi }}_{{p_i},{p_i}}^r} \big) $ 与$ I {\text π} \sin \big( {{\theta _{{p_i}}}} \big) $ 存在$ 2 {\text π} $ 整数倍的模糊:1)当
$ {\omega _{s,{p_i}}} = {\text π} \sin \big( {{\theta _{{p_i}}}} \big) \in \left[ {\dfrac{{i {\text π} }}{I},\dfrac{{( {i + 1} ) {\text π} }}{I}} \right] $ ,可得:$$ \arg \big(\hat{\boldsymbol{\varphi}}_{p_i, p_i}^r\big) = \left\{ \begin{array}{cl} I {\text π} \sin \big(\theta_{p_i}\big)-{\text π} i \geqslant 0 & i \text { 为偶数 } \\ I {\text π} \sin \big(\theta_{p_i}\big)-{\text π}(i+1)< 0 & i \text { 为奇数 } \end{array}\right. $$ (52) 2)当
$ {\omega _{s,{p_i}}} = {\text π} \sin \big( {{\theta _{{p_i}}}} \big) \in \left[ { - \dfrac{{( {i + 1} ) {\text π} }}{I}, - \dfrac{{i {\text π} }}{I}} \right] $ ,可得:$$ \arg \big(\hat{\boldsymbol{\varphi}}_{p_i, p_i}^r\big) = \left\{ \begin{array}{cl} I {\text π} \sin \big(\theta_{p_i}\big)+ {\text π} i < 0 & i \text { 为偶数 } \\ I {\text π} \sin \big(\theta_{p_i}\big)+ {\text π}(i+1) \geqslant 0 & i \text { 为奇数 } \end{array}\right. $$ (53) 因此,根据
$i$ 的奇偶性和$ \arg \big( {{\boldsymbol{\hat \varphi }}_{{p_i},{p_i}}^r} \big) $ 值的正负可获取${H_i}\big( {{e^{j\omega }}} \big)$ 通带内各目标无模糊的角度估计值:1)当
$i$ 为偶数时,目标角度估计值:$$ {\hat \theta _{{p_i}}} = \left\{ {\begin{array}{*{20}{c}} {a\sin \left[ {\dfrac{{\arg ( {{\boldsymbol{\hat \varphi }}_{{p_i},{p_i}}^r} ) + {\text π} i}}{{I {\text π} }}} \right]}&{\arg \big( {{\boldsymbol{\hat \varphi }}_{{p_i},{p_i}}^r} \big) \geqslant 0} \\ {a\sin \left[ {\dfrac{{\arg ( {{\boldsymbol{\hat \varphi }}_{{p_i},{p_i}}^r} ) - {\text π} i}}{{I {\text π} }}} \right]}&{\arg \big( {{\boldsymbol{\hat \varphi }}_{{p_i},{p_i}}^r} \big) < 0} \end{array}} \right. $$ (54) 2)当
$i$ 为奇数时,目标角度估计值:$$ {\hat \theta _{{p_i}}} = \left\{ {\begin{array}{*{20}{c}} {a\sin \left[ {\dfrac{{\arg ( {{\boldsymbol{\hat \varphi }}_{{p_i},{p_i}}^r} ) - {\text π} ( {i + 1} )}}{{I {\text π} }}} \right]}& {\arg ( {{\boldsymbol{\hat \varphi }}_{{p_i},{p_i}}^r} ) \geqslant 0} \\ {a\sin \left[ {\dfrac{{\arg ( {{\boldsymbol{\hat \varphi }}_{{p_i},{p_i}}^r} ) + {\text π} ( {i + 1} )}}{{I {\text π} }}} \right]}& {\arg ( {{\boldsymbol{\hat \varphi }}_{{p_i},{p_i}}^r} ) < 0} \end{array}} \right. $$ (55) -
令
$ \arg \big( {{\boldsymbol{\hat \varphi }}_{{p_i},{p_i}}^t} \big) $ 表示$ {{\boldsymbol{\varphi }}_t} $ 主对角线上元素$ {\mathrm{e}}^{{\mathrm{j}} {\text π} \sin ( {{\theta _{{p_i}}}} ) - \frac{{{\mathrm{j}}4 {\text π} \Delta f{r_{{p_i}}}}}{c}} $ 相位在主值区间的估计值。由${r_{{p_i}}} \in ( {0,{c \mathord{\left/ {\vphantom {c {2\Delta f}}} \right. } {2\Delta f}}} )$ 和${\theta _{{p_i}}} \in ( { - 0.5 {\text π} ,0.5 {\text π} } )$ 可得${{ - 4 {\text π} \Delta f{r_{{p_i}}}} \mathord{\left/ {\vphantom {{ - 4 {\text π} \Delta f{r_{{p_i}}}} c}} \right. } c} \in ( { - 2 {\text π} ,0} )$ ,$ {\text π} \sin \big( {{\theta _{{p_i}}}} \big) \in ( { - {\text π} , {\text π} } )$ ,故$ {\text π} \sin \big( {{\theta _{{p_i}}}} \big){{ - 4 {\text π} \Delta f{r_{p_i}}} \mathord{\left/ {\vphantom {{ - 4 {\text π} \Delta f{r_p}} c}} \right. } c} \in $ $( { - 3 {\text π} , {\text π} } )$ 。因此当$ {\text π} \sin \big( {{\theta _{{p_i}}}} \big){{ - 4 {\text π} \Delta f{r_{{p_i}}}} \mathord{\left/ {\vphantom {{ - 4 {\text π} \Delta f{r_{{p_i}}}} {c \in [ { - {\text π} , {\text π} } )}}} \right. } {c \in [ { - {\text π} , {\text π} } )}} $ ,由$ {\big( {{{{\boldsymbol{\hat P}}}_{t,1}}} \big)^\dagger }{{\boldsymbol{\hat P}}_{t,2}} $ 的特征值估计出的$ \arg \big( {{\boldsymbol{\hat \varphi }}_{{p_i},{p_i}}^t} \big) $ 满足:$$ \arg \big( {{\boldsymbol{\hat \varphi }}_{{p_i},{p_i}}^t} \big) = {\text π} \sin \big( {{\theta _{{p_i}}}} \big){{ - 4 {\text π} \Delta f{r_{{p_i}}}} \mathord{\left/ {\vphantom {{ - 4 {\text π} \Delta f{r_{{p_i}}}} c}} \right. } c} $$ (56) 当
$ {\text π} \sin \big( {{\theta _{{p_i}}}} \big){{ - 4 {\text π} \Delta f{r_{{p_i}}}} \mathord{\left/ {\vphantom {{ - 4 {\text π} \Delta f{r_{{p_i}}}} {c \in }}} \right. } {c \in }}( { - 3 {\text π} , - {\text π} } ) $ ,$ \arg \big( {{\boldsymbol{\hat \varphi }}_{{p_i},{p_i}}^t} \big) $ 满足:$$ \arg \big( {{\boldsymbol{\hat \varphi }}_{{p_i},{p_i}}^t} \big) = {\text π} \sin \big( {{\theta _{{p_i}}}} \big){{ - 4 {\text π} \Delta f{r_{{p_i}}}} \mathord{\left/ {\vphantom {{ - 4 {\text π} \Delta f{r_{{p_i}}}} c}} \right. } c} + 2 {\text π} $$ (57) 由式(54)和(55)所得的角度估计值
$ {\hat \theta _{{p_i}}} $ 可求出$ {\text π} \sin \big( {{{\hat \theta }_{{p_i}}}} \big) $ 。根据式(56)和(57)可知:1)若
$ {\text π} \sin \big( {{{\hat \theta }_{{p_i}}}} \big) \geqslant $ $ \arg \big( {{\boldsymbol{\hat \varphi }}_{{p_i},{p_i}}^t} \big) $ ,目标距离估计值:$$ {\hat r_{{p_i}}} = \left[ { {\text π} \sin \big( {{{\hat \theta }_{{p_i}}}} \big) - \arg \big( {{\boldsymbol{\hat \varphi }}_{{p_i},{p_i}}^t} \big)} \right]{c \mathord{\left/ {\vphantom {c {4 {\text π} \Delta f}}} \right. } {4 {\text π} \Delta f}} $$ (58) 2) 若
$ {\text π} \sin \big( {{{\hat \theta }_{{p_i}}}} \big) < $ $ \arg \big( {{\boldsymbol{\hat \varphi }}_{{p_i},{p_i}}^t} \big) $ ,目标距离估计值:$$ {\hat r_{{p_i}}} = \left[ { {\text π} \sin \big( {{{\hat \theta }_{{p_i}}}} \big) - \arg \big( {{\boldsymbol{\hat \varphi }}_{{p_i},{p_i}}^t} \big) + 2 {\text π} } \right]{c \mathord{\left/ {\vphantom {c {4 {\text π} \Delta f}}} \right. } {4 {\text π} \Delta f}} $$ (59) 由式(42)和式(43)可知
$ {( {{{\boldsymbol{P}}_{r,1}}} )^\dagger }{{\boldsymbol{P}}_{r,2}} $ 和$ {( {{{\boldsymbol{P}}_{t,1}}} )^\dagger } {{\boldsymbol{P}}_{t,2}} $ 的特征向量相同,因此同一目标的距离和角度与同一个特征向量相对应。当只有$ L $ 组采样数据的情况下,可以采用如下方法进行$ {\hat \theta _{{p_i}}} $ 和$ {\hat r_{{p_i}}} $ 的配对[19]:令$ {\hat {\boldsymbol{A}}}_{{\mathrm{cd}},1}^r $ 和$ {\hat {\boldsymbol{A}}}_{{\mathrm{cd}},1}^t $ 分别表示由$ {\big( {{{{\boldsymbol{\hat P}}}_{r,1}}} \big)^\dagger }{{\boldsymbol{\hat P}}_{r,2}} $ 和$ {\big( {{{{\boldsymbol{\hat P}}}_{t,1}}} \big)^\dagger }{{\boldsymbol{\hat P}}_{t,2}} $ 获得的矩阵$ {{\boldsymbol{A}}_{{\mathrm{cd}},1}} $ 的估计值,$ {\big( {{\hat {\boldsymbol{A}}}_{{\mathrm{cd}},1}^t} \big)^{ - 1}}{\hat {\boldsymbol{A}}}_{{\mathrm{cd}},1}^r $ 第$q$ 列模值最大的元素位于第${i_q}$ 行,则$ {\hat {\boldsymbol{A}}}_{{\mathrm{cd}},1}^r $ 第$q$ 列和$ {\hat {\boldsymbol{A}}}_{{\mathrm{cd}},1}^t $ 第${i_q}$ 列分别对应的特征值,来自于同一个目标。图2给出了基于CBS PM算法的目标角度、距离参数估计流程图。 -
定义
$ {{\boldsymbol{\eta }}_d} = {\left[ {{\theta _1}, \cdots ,{\theta _{P,}}{r_1}, \cdots ,{r_P}} \right]^{\rm{T}}} $ 表示式(6)中需要估计的角度和距离参数,类似文献[22]可推导出其估计的CRB可表示为:$$ \frac{{\sigma _w^2}}{{2L}}{\left\{ {{{\mathrm{Re}}} \left[ {( {{{\boldsymbol{D}}^{\mathrm{H}}}{\boldsymbol{\varDelta D}}} ) \odot {{( {\left[ {\begin{array}{*{20}{c}} 1 & 1 \\ 1 & 1 \end{array}} \right] \otimes ( {{{\boldsymbol{R}}_t}{{\boldsymbol{A}}^{\mathrm{H}}}{{\boldsymbol{R}}^{ - 1}}{\boldsymbol{A}}{{\boldsymbol{R}}_t}} )} )}^{\rm{T}}}} \right]} \right\}^{ - 1}} $$ (60) 其中,
$$ {\boldsymbol{D}} = \left[ {{{\boldsymbol{d}}_\theta }( {{\theta _1},{r_1}} ), \cdots ,{{\boldsymbol{d}}_\theta }( {{\theta _P},{r_P}} ),{{\boldsymbol{d}}_r}( {{\theta _1},{r_1}} ), \cdots ,{{\boldsymbol{d}}_r}( {{\theta _P},{r_P}} )} \right] $$ (61) $$ {{\boldsymbol{d}}_\theta }( {{\theta _p},{r_p}} ) = \frac{{\partial {\boldsymbol{a}}( {\theta ,r} )}}{{\partial \theta }}\left| {_{\theta = {\theta _p},r = {r_p}}} \right.\;\;\;p = 1,2, \cdots ,P $$ (62) $$ {{\boldsymbol{d}}_r}( {{\theta _p},{r_p}} ) = \frac{{\partial {\boldsymbol{a}}( {\theta ,r} )}}{{\partial r}}\left| {_{\theta = {\theta _p},r = {r_p}}} \right.\;\;\; p = 1,2, \cdots ,P $$ (63) $$ {\boldsymbol{\varDelta }} = \sigma _w^2{{\boldsymbol{I}}_{{N_t}{M_r}}} - {\boldsymbol{A}}{( {{{\boldsymbol{A}}^{\mathrm{H}}}{\boldsymbol{A}}} )^{ - 1}}{{\boldsymbol{A}}^{\mathrm{H}}} $$ (64)
Angle Range Estimation Algorithm for FDA-MIMO Radar Based on Convolutional Beamspace Propagator Method
-
摘要: 针对FDA-MIMO雷达接收数据维数高和工程实现复杂的问题,提出一种FDA-MIMO雷达卷积波束空间目标角度、距离估计算法。首先基于发射波形间的正交性,将对应各发射波形的回波信号通过一组满足奈奎斯特性质的滤波器组,然后对各滤波器的稳态卷积输出进行抽取处理以降低数据维数;对抽取数据应用传播算子法,再基于卷积滤波器的通带特性和FDA阵列角度、距离间的耦合关系,实现目标角度和距离参数的无模糊估计。仿真实验验证了该参数估计算法的有效性。Abstract: Addressing the issues of high dimensional received data and complex engineering implementation for FDA-MIMO (Frequency Diverse Array - Multiple Input Multiple Output) radar, this paper proposes an algorithm to estimate angle and range of target in the convolutional beamspace. Firstly, based on the orthogonality between the transmitted waveforms, the echo signal corresponding to each transmitted waveform is passed through a set of filters that satisfy the Nyquist property. Secondly, the output data of the convolutional steady state of each filter is decimated to reduce the data dimensionality. Then, an angle and range estimation algorithm using propagator method is developed. The unambiguous angle and range estimates are obtained based on the passband characteristics of the convolutional filters and the coupling relationship between the angle and range of FDA. Finally, numerical simulations are conducted to validate the effectiveness of the proposed convolutional beamspace algorithm.
-
Key words:
- convolutional beamspace /
- disambiguation /
- frequency diverse array /
- MIMO radar /
- parameter estimation
-
[1] LAN L, MARINO A, AUBRY A, et al. GLRT-based adaptive target detection in FDA-MIMO radar[J]. IEEE Transactions on Aerospace and Electronic Systems, 2021, 57(1): 597-613. doi: 10.1109/TAES.2020.3028485 [2] HUANG B, JIAN J W, BASIT A, et al. Adaptive distributed target detection for FDA-MIMO radar in Gaussian clutter without training data[J]. IEEE Transactions on Aerospace and Electronic Systems, 2022, 58(4): 2961-2972. doi: 10.1109/TAES.2022.3145781 [3] HUANG B, ORLANDO D, Wang W Q, et al. Adaptive multiple targets detection for FDA-MIMO radar with Gaussian clutter[J]. Signal Processing, 2023, 205: 108893. [4] HU Y M, DENG W B, ZHANG X, et al. FDA-MIMO radar with long-baseline transmit array using ESPRIT[J]. IEEE Signal Processing Letters, 2021, 28: 1530-1534. doi: 10.1109/LSP.2021.3095612 [5] LIU Y B, WANG C Y, ZHENG G M. Joint range and angle estimation of low-elevation target with bistatic meter-wave FDA-MIMO radar[J]. Digital Signal Processing, 2022, 127: 103556. [6] ZHENG G M, SONG Y W, LIU Y B, et al. Search-free range and angle estimation for bistatic VHF-FDA-MIMO radar in complex terrain[J]. Signal Processing, 2023, 212: 109163. doi: 10.1016/j.sigpro.2023.109163 [7] LAN L, XU J W, LIAO G S, et al. Suppression of mainbeam deceptive jammer with FDA-MIMO radar[J]. IEEE Transactions on Vehicular Technology, 2020, 69(10): 11584-11598. doi: 10.1109/TVT.2020.3014689 [8] 孙晨伟. 用于大型阵列的自适应主旁瓣干扰抑制算法[D]. 西安: 西安电子科技大学, 2021. SUN C W. Adaptive mainlobe and sidelobe jammers suppression algorithms for large scale phased array[D]. Xi’an: Xidian University, 2021. [9] GAO X Y, DAI L L, SAYEED A M. Low RF-complexity technologies to enable millimeter-wave MIMO with large antenna array for 5G wireless communications[J]. IEEE Communications Magazine, 2018, 56(4): 211-217. doi: 10.1109/MCOM.2018.1600727 [10] LIU D H, ZHAO Y B, ZHANG T X. DOA estimation in beamspace for monostatic MIMO radar utilizing real polynomial rooting[J]. Digital Signal Processing, 2021, 118: 103245. doi: 10.1016/j.dsp.2021.103245 [11] XU S Q, CHEN B X, CHEN X Y, et al. A beamspace dimension reduction technique with application to doa estimation in Low-angle tracking[C]//Proceedings of the CIE International Conference on Radar. New York: IEEE, 2021: 1585-1588. [12] LIU D H, ZHAO Y B. Real-valued sparse Bayesian learning algorithm for off-grid DOA estimation in the beamspace[J]. Digital Signal Processing, 2022, 121: 103322. [13] CHEN P C, VAIDYANATHAN P P. Convolutional beamspace for linear arrays[J]. IEEE Transactions on Signal Processing, 2020, 68: 5395-5410. doi: 10.1109/TSP.2020.3021670 [14] CHEN P C, VAIDYANATHAN P P. Convolutional beamspace using IIR filters[C]//Proceedings of the ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing. New York: IEEE, 2022: 5003-5007. [15] CHEN P C, VAIDYANATHAN P P. Sliding-capon based convolutional beamspace for linear arrays[C]//Proceedings of the ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing. New York: IEEE, 2021: 4565-4569. [16] CHEN P C, VAIDYANATHAN P P. Hybrid convolutional beamspace for DOA estimation of millimeter wave sources[C]//Proceedings of the 56th Asilomar Conference on Signals, Systems, and Computers. New York: IEEE, 2022: 86-90. [17] 张贤达. 现代信号处理[M]. 2版. 北京: 清华大学出版社, 2002. ZHANG X D. Modern signal processing[M]. 2nd ed. Beijing: Tsinghua University Press, 2002. [18] MARCOS S, MARSAL A, BENIDIR M. The propagator method for source bearing estimation[J]. Signal Processing, 1995, 42(2): 121-138. doi: 10.1016/0165-1684(94)00122-G [19] HE J, LIU Z. Extended aperture 2-D direction finding with a two-parallel-shape-array using propagator method[J]. IEEE Antennas and Wireless Propagation Letters, 2009, 8: 323-327. doi: 10.1109/LAWP.2009.2015224 [20] LI J F, JIANG D F. Performance analysis of propagator-based ESPRIT for direction of arrival estimation[J]. IET Signal Processing, 2018, 12(4): 481-486. doi: 10.1049/iet-spr.2017.0033 [21] XU J W, LIAO G S, ZHU S Q, et al. Joint range and angle estimation using MIMO radar with frequency diverse array[J]. IEEE Transactions on Signal Processing, 2015, 63(13): 3396-3410. doi: 10.1109/TSP.2015.2422680 [22] STOICA P, LARSSON E G, GERSHMAN A B. The stochastic CRB for array processing: A textbook derivation[J]. IEEE Signal Processing Letters, 2001, 8(5): 148-150. doi: 10.1109/97.917699 [23] VAIDYANATHAN P P. Multirate systems and filter banks[M]. Englewood Cliffs, NJ: Prentice Hall, 1993.