Assume that a discrete-time system can be described by the following state-space equations


where the input signal $u(k)$ is stationary and ergodic with $\mathbf{E}[u(k)]=0$.

Define then the covariance matrix

$ R(k) := \mathbf{E}\Bigg[ \begin{bmatrix} x(k)\\u(k) \end{bmatrix}\begin{bmatrix} x(k)\\u(k) \end{bmatrix}^\top\Bigg] = \begin{bmatrix} r_{xx}(k) & r_{xu}(k)\\ r_{xu}^\top (k) & r_{uu}(k) \end{bmatrix} $

In particular, if $u$ is persistently exciting of order $n$ then $R(k)>0$ and in particular for the Sylvester Theorem,

$R_{uu}(k)= \mathbf{E}\Bigg[ \begin{bmatrix} u(k)\\ \vdots \\u(k+n-1) \end{bmatrix}\begin{bmatrix} u(k)\\ \vdots \\u(k+n-1) \end{bmatrix}^\top\Bigg]>0$.

I have two statements/questions:

1) if $u$ is PE(n) then it is also PE(n-1) so is it true that

$r_{uu}(k)=\mathbf{E}[u(k)u^\top (k)] >0 \quad ?$

2) Knowing that $r_{uu}(k)>0$, is it possible to verify that also $r_{xx}(k)=\mathbf{E}[x(k)x^\top (k)]>0$?


If your covariance matrix is PD, i.e. $R(k)>0$, for Sylvester criterion $r_{uu}(k)>0$ and so also $r_{xx}(k)>0$.


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.