bisection search method二分搜索法的使用

Optimizing the Matrices $V$ and $V_E$

In this subsection, the TPC matrix $\mathbf{V}$ and matrix ${\mathbf{V}}_E$ are optimized by fixing $\mathbf{\Phi }$. Specifically, the unit modulus constraint on the phase shifts $\mathbf{\Phi }$ is removed, and the updated optimization problem reduced from Problem is given by
$$\begin{array}{rl}& \underset{\mathbf{V},{\mathbf{V}}_E}{\min }-\text{Tr}({\mathbf{W}}_I{\mathbf{V}}^H{{\hat{\mathbf{H}}}_I}^H{\mathbf{U}}_I)-\text{Tr}({\mathbf{W}}_I{\mathbf{U}}_I^H{\hat{\mathbf{H}}}_I\mathbf{V})+\text{Tr}({\mathbf{V}}^H{\mathbf{H}}_V\mathbf{V})\\ & -\text{Tr}({\mathbf{W}}_E{\mathbf{V}}_E^H{{\hat{\mathbf{H}}}_E}^H{\mathbf{U}}_E)-\text{Tr}({\mathbf{W}}_E{\mathbf{U}}_E^H{\hat{\mathbf{H}}}_E{\mathbf{V}}_E)+\text{Tr}({\mathbf{V}}_E^H{\mathbf{H}}_{VE}{\mathbf{V}}_E)\\ & \text{s.t.}\mathrm{Tr}(\mathbf{V}{\mathbf{V}}^H+{\mathbf{V}}_E{\mathbf{V}}_E^H)\le P_T.\end{array}$$

The above problem is a convex QCQP problem, and the standard optimization packages, such as CVX can be exploited to solve it. However, the calculation burden is heavy. To reduce the complexity, the near-optimal closed form expressions of the TPC matrix and AN covariance matrix are provided by applying the Lagrangian multiplier method.

Since Problem is a convex problem, the Slater’s condition is satisfied, where the duality gap between Problem and its dual problem is zero. Thus, Problem can be solved by addressing its dual problem if the dual problem is easier. For this purpose, by introducing Lagrange multiplier $\lambda$ to combine the the constraint and OF of Problem, the Lagrangian function of Problem is obtained as
$$\begin{array}{rl}\mathcal{L}\left(\mathbf{V},{\mathbf{V}}_E,\lambda \right)& \triangleq -\mathrm{Tr}\left({\mathbf{W}}_I{\mathbf{V}}^H{\hat{\mathbf{H}}}_I^H{\mathbf{U}}_I\right)-\mathrm{Tr}\left({\mathbf{W}}_I{\mathbf{U}}_I^H{\hat{\mathbf{H}}}_I\mathbf{V}\right)+\mathrm{Tr}\left({\mathbf{V}}^H{\mathbf{H}}_V\mathbf{V}\right)-\mathrm{Tr}\left({\mathbf{W}}_E{\mathbf{V}}_E^H{\hat{\mathbf{H}}}_E^H{\mathbf{U}}_E\right)\\ & -\mathrm{Tr}\left({\mathbf{W}}_E{\mathbf{U}}_E^H{\hat{\mathbf{H}}}_E{\mathbf{V}}_E\right)+\mathrm{Tr}\left({\mathbf{V}}_E^H{\mathbf{H}}_{VE}{\mathbf{V}}_E\right)+\lambda [\mathrm{Tr}\left(\mathbf{V}{\mathbf{V}}^H+{\mathbf{V}}_E{\mathbf{V}}_E^H\right)-P_T]\\ & =-\mathrm{Tr}\left({\mathbf{W}}_I{\mathbf{V}}^H{\hat{\mathbf{H}}}_I^H{\mathbf{U}}_I\right)-\mathrm{Tr}\left({\mathbf{W}}_I{\mathbf{U}}_I^H{\hat{\mathbf{H}}}_I\mathbf{V}\right)+\mathrm{Tr}\left[{\mathbf{V}}^H\left({\mathbf{H}}_V+\lambda \mathbf{I}\right)\mathbf{V}\right]\\ & -\mathrm{Tr}\left({\mathbf{W}}_E{\mathbf{V}}_E^H{\hat{\mathbf{H}}}_E^H{\mathbf{U}}_E\right)-\mathrm{Tr}\left({\mathbf{W}}_E{\mathbf{U}}_E^H{\hat{\mathbf{H}}}_E{\mathbf{V}}_E\right)+\mathrm{Tr}\left[{\mathbf{V}}_E^H\left({\mathbf{H}}_{VE}+\lambda \mathbf{I}\right){\mathbf{V}}_E\right]-\lambda P_T.\end{array}$$

Then the dual problem of Problem is
$$\begin{array}{r}\underset{\lambda }{\max }h\left(\lambda \right)\\ \text{s.t.}\lambda \ge 0,\end{array}$$

where $h\left(\lambda \right)$ is the dual function given by
$$\begin{array}{r}h\left(\lambda \right)\triangleq \underset{\mathbf{V},{\mathbf{V}}_E}{\min }\mathcal{L}\left(\mathbf{V},{\mathbf{V}}_E,\lambda \right).\end{array}$$

Note that Problem is a convex quadratic optimization problem with no constraint, which can be solved in closed form. The optimal solution ${\mathbf{V}}^{\star },{{\mathbf{V}}^{\star }}_E$ for Proble is
$$\begin{array}{r}[{\mathbf{V}}^{\star },{{\mathbf{V}}^{\star }}_E]=\text{arg}\underset{\mathbf{V},{\mathbf{V}}_E}{\min }\mathcal{L}\left(\mathbf{V},{\mathbf{V}}_E,\lambda \right).\end{array}$$

By setting the first-order derivative of $\mathcal{L}\left(\mathbf{V},{\mathbf{V}}_E,\lambda \right)$ w.r.t. $\mathbf{V}$ to zero matrix, we can obtain the optimal solution of $\mathbf{V}$ as follows:
$$\begin{array}{r}\frac{\partial \mathcal{L}\left(\mathbf{V},{\mathbf{V}}_E,\lambda \right)}{\partial \mathbf{V}}=0,\\ \frac{\partial \mathcal{L}\left(\mathbf{V},{\mathbf{V}}_E,\lambda \right)}{\partial {\mathbf{V}}_E}=0.\end{array}$$
The left hand side of Equation can be expanded as

$$\begin{array}{rl}\frac{\partial \mathcal{L}\left(\mathbf{V},{\mathbf{V}}_E,\lambda \right)}{\partial \mathbf{V}}& =\frac{\partial \mathrm{Tr}\left[{\mathbf{V}}^H\left({\mathbf{H}}_V+\lambda \mathbf{I}\right)\mathbf{V}\right]}{\partial \mathbf{V}}-{\left({\mathbf{W}}_I{\mathbf{U}}_I^H{\hat{\mathbf{H}}}_I\right)}^H-\left({\hat{\mathbf{H}}}_I^H{\mathbf{U}}_I{\mathbf{W}}_I\right)\\ & =2\left({\mathbf{H}}_V+\lambda \mathbf{I}\right)\mathbf{V}-2\left({\hat{\mathbf{H}}}_I^H{\mathbf{U}}_I{\mathbf{W}}_I\right).\end{array}$$

这里理论上需要使用复数求导的公式，但是这里实际上将函数当作普通的实数变量来求导了，根据 matrix-cookbook 当中的这个公式 $$\frac{\partial }{\partial \mathbf{X}}\mathrm{Tr}\left({\mathbf{X}}^T\mathbf{BX}\right)=\mathbf{BX}+{\mathbf{B}}^T\mathbf{X}$$，并且 ${\mathbf{H}}_V+\lambda \mathbf{I}$ 实际上是共轭对称的，因此可以写成$$\frac{\partial \mathrm{Tr}\left[{\mathbf{V}}^H\left({\mathbf{H}}_V+\lambda \mathbf{I}\right)\mathbf{V}\right]}{\partial \mathbf{V}}=2\left({\mathbf{H}}_V+\lambda \mathbf{I}\right)\mathbf{V}.$$，我们可以将所有的共轭转置符号 $H$ 当作 $T$ 来理解。同理，因为 $$\frac{\partial }{\partial \mathbf{X}}\mathrm{Tr}(\mathbf{XA})={\mathbf{A}}^T,$$我们有 $$-\frac{\partial }{\partial \mathbf{X}}\mathrm{Tr}\left({\mathbf{W}}_I{\mathbf{U}}_I^H{\hat{\mathbf{H}}}_I\mathbf{V}\right)=-{\left({\mathbf{W}}_I{\mathbf{U}}_I^H{\hat{\mathbf{H}}}_I\right)}^H.$$目前这一系列的结论都是观察得到的，至于是不是确实这样计算的有待商榷。

The equation becomes
$$\begin{array}{r}\left({\mathbf{H}}_V+\lambda \mathbf{I}\right)\mathbf{V}=\left({\hat{\mathbf{H}}}_I^H{\mathbf{U}}_I{\mathbf{W}}_I\right).\end{array}$$

Then the optimal solution ${\mathbf{V}}^{\star }$ for Problem is
$$\begin{array}{rl}{\mathbf{V}}^{\star }& ={\left({\mathbf{H}}_V+\lambda \mathbf{I}\right)}^{†}\left({\hat{\mathbf{H}}}_I^H{\mathbf{U}}_I{\mathbf{W}}_I\right)\\ & \triangleq {\mathbf{\Theta }}_V\left(\lambda \right)\left({\hat{\mathbf{H}}}_I^H{\mathbf{U}}_I{\mathbf{W}}_I\right).\end{array}$$

Similarly, we solve Problem by setting the first-order derivative of $\mathcal{L}\left(\mathbf{V},{\mathbf{V}}_E,\lambda \right)$ w.r.t. ${\mathbf{V}}_E$ to zero matrix, which becomes
$$\begin{array}{r}2\left({\mathbf{H}}_{VE}+\lambda \mathbf{I}\right){\mathbf{V}}_E-2{\hat{\mathbf{H}}}_E^H{\mathbf{U}}_E{\mathbf{W}}_E^H=0.\end{array}$$
Then the optimal solution ${\mathbf{V}}_E^{\star }$ for Problem is
$$\begin{array}{rl}{\mathbf{V}}_E^{\star }& ={\left({\mathbf{H}}_{VE}+\lambda \mathbf{I}\right)}^{†}{\hat{\mathbf{H}}}_E^H{\mathbf{U}}_E{\mathbf{W}}_E^H\\ & \triangleq {\mathbf{\Theta }}_{VE}\left(\lambda \right){\hat{\mathbf{H}}}_E^H{\mathbf{U}}_E{\mathbf{W}}_E^H.\end{array}$$
Once the optimal solution ${\lambda }^{\star }$ for Problem is found, the final optimal ${\mathbf{V}}^{\star },{\mathbf{V}}_E^{\star }$ can be obtained. The value of ${\lambda }^{\star }$ should be chosen in order to guarantee the complementary slackness condition as
$$\begin{array}{r}\lambda [\mathrm{Tr}({\mathbf{V}}^{\star }{\mathbf{V}}^{\star H}+{\mathbf{V}}_E^{\star }{\mathbf{V}}_E^{\star H})-P_T]=0.\end{array}$$

We define
$$\begin{array}{rl}P(\lambda )& \triangleq \mathrm{Tr}({\mathbf{V}}^{\star }{\mathbf{V}}^{\star H}+{\mathbf{V}}_E^{\star }{\mathbf{V}}_E^{\star H})=\mathrm{Tr}({\mathbf{V}}^{\star }{\mathbf{V}}^{\star H})+\mathrm{Tr}({\mathbf{V}}_E^{\star }{\mathbf{V}}_E^{\star H}),\end{array}$$

where
$$\begin{array}{rl}\mathrm{Tr}\left({\mathbf{V}}^{\star }{\mathbf{V}}^{\star H}\right)& =\mathrm{Tr}\left({\mathbf{\Theta }}_V\left(\lambda \right)({\hat{\mathbf{H}}}_I^H{\mathbf{U}}_I{\mathbf{W}}_I^H)({\hat{\mathbf{H}}}_I^H{\mathbf{U}}_I{\mathbf{W}}_I^H{)}^H{\mathbf{\Theta }}_V^H\left(\lambda \right)\right)\\ & =\mathrm{Tr}\left({\mathbf{\Theta }}_V^H\left(\lambda \right){\mathbf{\Theta }}_V\left(\lambda \right)({\hat{\mathbf{H}}}_I^H{\mathbf{U}}_I{\mathbf{W}}_I^H)({\hat{\mathbf{H}}}_I^H{\mathbf{U}}_I{\mathbf{W}}_I^H{)}^H\right),\end{array}$$
$$\begin{array}{rl}\mathrm{Tr}\left({\mathbf{V}}_E^{\star H}{\mathbf{V}}_E^{\star }\right)& =\mathrm{Tr}\left({\mathbf{\Theta }}_{VE}\left(\lambda \right)({\hat{\mathbf{H}}}_E^H{\mathbf{U}}_E{\mathbf{W}}_E^H)({\hat{\mathbf{H}}}_E^H{\mathbf{U}}_E{\mathbf{W}}_E^H{)}^H{\mathbf{\Theta }}_{VE}^H\left(\lambda \right)\right)\\ & =\mathrm{Tr}\left({\mathbf{\Theta }}_{VE}^H\left(\lambda \right){\mathbf{\Theta }}_{VE}\left(\lambda \right)({\hat{\mathbf{H}}}_E^H{\mathbf{U}}_E{\mathbf{W}}_E^H)({\hat{\mathbf{H}}}_E^H{\mathbf{U}}_E{\mathbf{W}}_E^H{)}^H\right).\end{array}$$
Then $P(\lambda )$ becomes
$$\begin{array}{r}P(\lambda )=\mathrm{Tr}\left({\mathbf{\Theta }}_V^n({\hat{\mathbf{H}}}_I^H{\mathbf{U}}_I{\mathbf{W}}_I^H)({\hat{\mathbf{H}}}_I^H{\mathbf{U}}_I{\mathbf{W}}_I^H{)}^H\right)+\mathrm{Tr}\left({\mathbf{\Theta }}_{VE}^n({\hat{\mathbf{H}}}_E^H{\mathbf{U}}_E{\mathbf{W}}_E^H)({\hat{\mathbf{H}}}_E^H{\mathbf{U}}_E{\mathbf{W}}_E^H{)}^H\right),\end{array}$$
where
$$\begin{array}{rl}{\mathbf{\Theta }}_V^n& ={\mathbf{\Theta }}_V^H\left(\lambda \right){\mathbf{\Theta }}_V\left(\lambda \right)={\left({\mathbf{H}}_V+\lambda \mathbf{I}\right)}^{†H}{\left({\mathbf{H}}_V+\lambda \mathbf{I}\right)}^{†},\end{array}$$
$$\begin{array}{rl}{\mathbf{\Theta }}_{VE}^n& ={\mathbf{\Theta }}_{VE}^H\left(\lambda \right){\mathbf{\Theta }}_{VE}\left(\lambda \right)={\left({\mathbf{H}}_{VE}+\lambda \mathbf{I}\right)}^{†H}{\left({\mathbf{H}}_{VE}+\lambda \mathbf{I}\right)}^{†}.\end{array}$$
To find the optimal ${\lambda }^{\star }\ge 0$, we first check whether $\lambda =0$ is the optimal solution or not. If
$$\begin{array}{r}P(0)=\mathrm{Tr}\left({\mathbf{V}}^{\star H}(0){\mathbf{V}}^{\star }(0)\right)+\mathrm{Tr}\left({\mathbf{V}}_E^{\star H}(0){{\mathbf{V}}_E}^{\star }(0)\right)\le P_T,\end{array}$$
then the optimal solutions are given by ${\mathbf{V}}^{\star }=\mathbf{V}(0)$ and ${\mathbf{V}}_E^{\star }={\mathbf{V}}_E(0)$. Otherwise, the optimal ${\lambda }^{\star }>0$ is the solution of the equation $P(\lambda )=0$.

It is ready to verify that ${\mathbf{H}}_V$ and ${\mathbf{H}}_{VE}$ is a positive semidefinite matrix. Let us define the rank of ${\mathbf{H}}_V$ and ${\mathbf{H}}_{VE}$ as $r_V=\mathrm{rank}({\mathbf{H}}_V)\le N_T$ and $r_{VE}=\mathrm{rank}({\mathbf{H}}_{VE})\le N_T$ respectively. By decomposing ${\mathbf{H}}_V$ and ${\mathbf{H}}_{VE}$ by using the singular value decomposition (SVD), we have ${\mathbf{H}}_V=\left[{\mathbf{P}}_{V,1},{\mathbf{P}}_{V,2}\right]{\mathbf{\Sigma }}_V{\left[{\mathbf{P}}_{V,1},{\mathbf{P}}_{V,2}\right]}^{\mathrm{H}},{\mathbf{H}}_{VE}=\left[{\mathbf{P}}_{VE,1},{\mathbf{P}}_{VE,2}\right]{\mathbf{\Sigma }}_{VE}{\left[{\mathbf{P}}_{VE,1},{\mathbf{P}}_{VE,2}\right]}^{\mathrm{H}},$ where ${\mathbf{P}}_{V,1}$ comprises the first $r_V$ singular vectors associated with the $r_V$ positive eigenvalues of ${\mathbf{H}}_V$, and ${\mathbf{P}}_{V,2}$ includes the last $N_T-r_V$ singular vectors associated with the $N_T-r_V$ zero-valued eigenvalues of ${\mathbf{H}}_V$, ${\mathbf{\Sigma }}_V=\mathrm{diag}\left\{{\mathbf{\Sigma }}_{V,1},0_{\left(N_T-r_V\right)\times \left(N_T-r_V\right)}\right\}$ with ${\mathbf{\Sigma }}_{V,1}$ representing the diagonal submatrix collecting the first $r_V$ positive eigenvalues. Similarly, the first $r_{VE}$ singular vectors corresponding to the $r_{VE}$ positive eigenvalues of ${\mathbf{H}}_{VE}$ are contained in ${\mathbf{P}}_{VE,1}$, while the last $N_T-r_{VE}$ singular vectors corresponding to the $N_T-r_{VE}$ zero-valued eigenvalues of ${\mathbf{H}}_{VE}$ are held in ${\mathbf{P}}_{VE,2}$. ${\mathbf{\Sigma }}_{VE}=\mathrm{diag}\left\{{\mathbf{\Sigma }}_{VE,1},0_{\left(N_T-r_{VE}\right)\times \left(N_T-r_{VE}\right)}\right\}$ is a diagonal matrix with ${\mathbf{\Sigma }}_{VE,1}$ representing the diagonal submatrix gathering the first $r_{VE}$ positive eigenvalues. By defining ${\mathbf{P}}_V\triangleq \left[{\mathbf{P}}_{V,1},{\mathbf{P}}_{V,2}\right]$ and ${\mathbf{P}}_{VE}\triangleq \left[{\mathbf{P}}_{VE,1},{\mathbf{P}}_{VE,2}\right]$, and substituting \eqref{svdforHVHVE} into and , $P(\lambda )$ becomes
$$\begin{array}{rl}& P(\lambda )=\mathrm{Tr}\left([{\left({\mathbf{P}}_V{\mathbf{\Sigma }}_V{\mathbf{P}}_V^H+\lambda {\mathbf{P}}_V{\mathbf{P}}_V^H\right)}^{-1}{\left({\mathbf{P}}_V{\mathbf{\Sigma }}_V{\mathbf{P}}_V^H+\lambda {\mathbf{P}}_V{\mathbf{P}}_V^H\right)}^{-1}]({\hat{\mathbf{H}}}_I^H{\mathbf{U}}_I{\mathbf{W}}_I^H)({\hat{\mathbf{H}}}_I^H{\mathbf{U}}_I{\mathbf{W}}_I^H{)}^H\right)\\ & +\mathrm{Tr}\left([{\left({\mathbf{P}}_{VE}{\mathbf{\Sigma }}_{VE}{\mathbf{P}}_{VE}^H+\lambda {\mathbf{P}}_{VE}{\mathbf{P}}_{VE}^H\right)}^{-1}{\left({\mathbf{P}}_{VE}{\mathbf{\Sigma }}_{VE}{\mathbf{P}}_{VE}^H+\lambda {\mathbf{P}}_{VE}{\mathbf{P}}_{VE}^H\right)}^{-1}]({\hat{\mathbf{H}}}_E^H{\mathbf{U}}_E{\mathbf{W}}_E^H)({\hat{\mathbf{H}}}_E^H{\mathbf{U}}_E{\mathbf{W}}_E^H{)}^H\right)\\ & =\mathrm{Tr}\left([{\left({\mathbf{\Sigma }}_V+\lambda \mathbf{I}\right)}^{-2}]{\mathbf{Z}}_V\right)+\mathrm{Tr}\left([{\left({\mathbf{\Sigma }}_{VE}+\lambda \mathbf{I}\right)}^{-2}]{\mathbf{Z}}_{VE}\right)\\ & =\sum _{i=1}^{r_V}\left[\frac{{\left[{\mathbf{Z}}_V\right]}_{i,i}}{{\left({\left[{\Sigma }_V\right]}_{i,i}+\lambda \right)}^2}\right]+\sum _{i=1}^{r_{VE}}\left[\frac{{\left[{\mathbf{Z}}_{VE}\right]}_{i,i}}{{\left({\left[{\Sigma }_{VE}\right]}_{i,i}+\lambda \right)}^2}\right]+\sum _{i=r_V+1}^{N_T}\left[\frac{{\left[{\mathbf{Z}}_V\right]}_{i,i}}{{\left(\lambda \right)}^2}\right]+\sum _{i=r_{VE}+1}^{N_T}\left[\frac{{\left[{\mathbf{Z}}_{VE}\right]}_{i,i}}{{\left(\lambda \right)}^2}\right],\end{array}$$
where ${\mathbf{Z}}_V={\mathbf{P}}_V^H({\hat{\mathbf{H}}}_I^H{\mathbf{U}}_I{\mathbf{W}}_I^H)({\hat{\mathbf{H}}}_I^H{\mathbf{U}}_I{\mathbf{W}}_I^H{)}^H{\mathbf{P}}_V$ and ${\mathbf{Z}}_{VE}={\mathbf{P}}_{VE}^H({\hat{\mathbf{H}}}_E^H{\mathbf{U}}_E{\mathbf{W}}_E^H)({\hat{\mathbf{H}}}_E^H{\mathbf{U}}_E{\mathbf{W}}_E^H{)}^H{\mathbf{P}}_{VE}$. ${\left[{\mathbf{Z}}_V\right]}_{i,i}$, ${\left[{\mathbf{Z}}_{VE}\right]}_{i,i}$, ${\left[{\Sigma }_V\right]}_{i,i}$, and ${\left[{\Sigma }_{VE}\right]}_{i,i}$ represent the $i$th diagonal element of matrices ${\mathbf{Z}}_V$, ${\mathbf{Z}}_{VE}$, ${\Sigma }_V$, and ${\Sigma }_{VE}$, respectively. The first line of (\ref{Plamda}) is obtained by substituting (\ref{svdforHVHVE}) into the expression of $P(\lambda )$ in (\ref{Plambdanew}). It can be verified from the last line of (\ref{Plamda}) that $P(\lambda )$ is a monotonically decreasing function.

Then, the optimal ${\lambda }^{\star }$ can be obtained by solving the following equation,
$$\begin{array}{r}\sum _{i=1}^{r_V}\left[\frac{{\left[{\mathbf{Z}}_V\right]}_{i,i}}{{\left({\left[{\Sigma }_V\right]}_{i,i}+\lambda \right)}^2}\right]+\sum _{i=1}^{r_{VE}}\left[\frac{{\left[{\mathbf{Z}}_{VE}\right]}_{i,i}}{{\left({\left[{\Sigma }_{VE}\right]}_{i,i}+\lambda \right)}^2}\right]+\sum _{i=r_V+1}^{N_T}\left[\frac{{\left[{\mathbf{Z}}_V\right]}_{i,i}}{{\left(\lambda \right)}^2}\right]+\sum _{i=r_{VE}+1}^{N_T}\left[\frac{{\left[{\mathbf{Z}}_{VE}\right]}_{i,i}}{{\left(\lambda \right)}^2}\right]=P_T.\end{array}$$

To solve it, the bisection search method is utilized. Since $P(\infty )=0$, the solution to Equation (\ref{lambdaequat}) must exist. The lower bound of ${\lambda }^{\star }$ is a positive value approaching zero, while the upper bound of ${\lambda }^{\star }$ is given by

$${\lambda }^{\star }<\sqrt{\frac{\sum _{i=1}^{N_T}{\left[{\mathbf{Z}}_V\right]}_{i,i}+\sum _{i=1}^{N_T}{\left[{\mathbf{Z}}_{VE}\right]}_{i,i}}{P_T}}\triangleq {\lambda }^{\mathrm{ub}}.$$
which can be proved as
$$\begin{array}{rl}P({\lambda }^{\mathrm{ub}})& =\sum _{i=1}^{r_V}\frac{{\left[{\mathbf{Z}}_V\right]}_{i,i}}{{\left({\left[{\Sigma }_V\right]}_{i,i}+{\lambda }^{\mathrm{ub}}\right)}^2}+\sum _{i=1}^{r_{VE}}\frac{{\left[{\mathbf{Z}}_{VE}\right]}_{i,i}}{{\left({\left[{\Sigma }_{VE}\right]}_{i,i}+{\lambda }^{\mathrm{ub}}\right)}^2}+\sum _{i=r_V+1}^{N_T}\left[\frac{{\left[{\mathbf{Z}}_V\right]}_{i,i}}{{\left({\lambda }^{\mathrm{ub}}\right)}^2}\right]+\sum _{i=r_{VE}+1}^{N_T}\left[\frac{{\left[{\mathbf{Z}}_{VE}\right]}_{i,i}}{{\left({\lambda }^{\mathrm{ub}}\right)}^2}\right]\\ & <\sum _{i=1}^{N_T}\frac{{\left[{\mathbf{Z}}_V\right]}_{i,i}}{{\left({\lambda }^{\mathrm{ub}}\right)}^2}+\sum _{i=1}^{N_T}\frac{{\left[{\mathbf{Z}}_{VE}\right]}_{i,i}}{{\left({\lambda }^{\mathrm{ub}}\right)}^2}=P_T.\end{array}$$

When the optimal ${\lambda }^{\star }$ is found, the optimal matrices ${\mathbf{V}}^{\star }$ and ${\mathbf{V}}_E^{\star }$ can be obtained by substituting ${\lambda }^{\star }$ into (10) and (12).