While reading the paper "Perturbation analysis for subspace decomposition with applications in subspace-based algorithms" by Zhengyuan Xu, I came across the decomposition technique called Subspace decomposition, where the data matrix $X$ (noise free) is decomposed as: Journal: https://sci-hub.st/10.1109/TSP.2002.804084

$$X = U_s \Lambda_s V_s^H + U_n \Lambda_n V_n^H$$

where $U_s$ and $V_s$ span the column spaces of $X$ and $X^H$ respectively, whereas $U_n$ and $V_n$ span their orthogonal spaces. $\Lambda_s$ and $\Lambda_n$ are corresponding singular values (or eigenvalues).

Could someone explain me this decomposition? I understand it is related to SVD, but I am not sure about the additive funtion here.

  • $\begingroup$ How can one read the paper, without paying IEEE for access? ieeexplore.ieee.org/document/1041039 $\endgroup$ Aug 6, 2021 at 17:00
  • $\begingroup$ I am using my academic institution login, and it is accessible. $\endgroup$
    – Neuling
    Aug 6, 2021 at 23:19
  • 1
    $\begingroup$ @Neuling we are eager to help, but not all of us have your academic institution login. $\endgroup$
    – AlexTP
    Aug 7, 2021 at 14:45
  • 1
    $\begingroup$ journals are lame $\endgroup$ Aug 7, 2021 at 16:55
  • $\begingroup$ Thank you @OverLordGoldDragon .. The Journal is linked in the previous comment. I will include it also in my post. $\endgroup$
    – Neuling
    Aug 8, 2021 at 19:07

1 Answer 1


You can rewrite the equation in a block matrix form:

\begin{eqnarray}X &=& U_s \Lambda_s V_s^H + U_n \Lambda_n V_n^H \\ &=& \left[ \begin{matrix} U_s & U_n\end{matrix} \right] \left[ \begin{matrix} \Lambda_s & 0 \\ 0 & \Lambda_n\end{matrix} \right] \left[ \begin{matrix} V_s^H \\ V_n^H\end{matrix} \right] \\ &=& U_t \Lambda_t V_t^H \end{eqnarray}

where $U_t$, $\Lambda_t$, $V_t$ represent the total/complete SVD decomposition of $X$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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