0
$\begingroup$

I am reading this article, in which the authors state that the problem

\begin{equation*} \begin{aligned} & \min_{\mathbf w_1 \cdots \mathbf w_K} \sum_{k = 1}^K ||\mathbf w_k||_2^2\\ & \text{s.t.} && \frac{|\mathbf w_k^H\mathbf h_k|^2}{\sum_{l \ne k} |\mathbf w_l^H\mathbf h_k|^2 + \sigma^2_k} \ge c_k,\ \forall k\in \{1, \cdots, K\} \end{aligned} \end{equation*}

, where $\mathbf w_k \in \mathbb C^n$, is an SOCP. I would like to prove this myself, and here's my attempt.

Obviously, we can move the quadratic term in objective function into the constraints by introducing a new variable $\mathbf t \in \mathbb R^K$, and rewrite the problem into:

\begin{equation*} \begin{aligned} & \min_{\mathbf w_1 \cdots \mathbf w_K, \mathbf t} \mathbf 1^T \mathbf t\\ & \text{s.t.} && \frac{|\mathbf w_k^H\mathbf h_k|^2}{\sum_{l \ne k} |\mathbf w_l^H\mathbf h_k|^2 + \sigma^2_k} \ge c_k,\ \forall k\in \{1, \cdots, K\}\\ &&& ||\mathbf w_k||_2 \le \sqrt{t_k},\ \forall k \in \{1, \cdots, K\}. \end{aligned} \end{equation*}

The next step should be to rewrite the other constraint into second-order cone form, but I have no idea how to proceed.


Update:

I've read this reference(page 24, 18.29) but a few details still bother me.

1) I thought an SOCP must have its objective in the $f^Tx$ form, but in the reference, the objective function is the sum of $d$ second-order terms.

2) Why is it an SOCP with the constraint $Im [\mathbf w_i^H \mathbf h_i] = 0$? I can't find a way to rewrite it to the standard $Fx = g$ form.

$\endgroup$
  • $\begingroup$ Why is the constraint with $\sqrt{t_k}$ a SOCO constraint? $\endgroup$ – LinAlg Oct 7 '18 at 13:01
1
$\begingroup$

Multiply with the denominator to get: $$c_k \left( \sum_{l \ne k} |\mathbf w_l^H\mathbf h_k|^2 + \sigma^2_k\right) \le |\mathbf w_k^H\mathbf h_k|^2$$ This constraint is of the form $||x||^2 \leq |t|^2$, which is not equivalent to $||x|| \leq t$. Your problem in the stated form is not convex and not SOCO.

If you read reference 2, they add the extra constraints that $t$ is nonnegative and that the imaginary part of $t$ is $0$. Under those conditions, $||x||^2 \leq |t|^2$ is equivalent to the SOCO constraint $||x|| \leq t$.


  1. The objective can be written as $\sum t_k$ with $||w_k||^2 \leq t_k$. Use that $x^T x \leq t$ is equal to $||[2x; t-1]|| \leq t+1$.

  2. For a complex number $x+yi$, the optimization variables are $x$ and $y$.

$\endgroup$
  • $\begingroup$ Thank you for your answer! I've read the reference but a few details still bother me. Could you take a look at the edited question? $\endgroup$ – hklel Oct 7 '18 at 13:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.