The problem of controlling a time varying finite-dimensional linear stochastic system \[ \begin{aligned} dx & +(Ax+B_ 2 u)dt+B_ 1 dw\\ dy & =C_ 2 xdt+D_{21} dw\end{aligned} \] in order to minimize the ergodic cost \(\limsup_{T\to\infty} {1\over T}\int^ T_ 0 E| z(t)|^ 2 dt\), where \(z=C_ 1 x+D_{12} u\), is considered. The originality of the paper lies in the fact that the control is assumed to be generated by a linear time varying compensator \(dx_ c=A_ c xdt+B_ c dy\), \(u=C_ c x_ c\) such that the controlled evolution in absence of noise is exponentially stable. Under appropriate nondegeneracy, stabilizability and detectability conditions the coefficients of the optimal compensator are computed by solving the usual pair of time varying Riccati equations for estimation and control, respectively. The a priori restriction on the class of admissible controls allows the author to simplify the proof. Of course the result he obtains is weaker than the separation principle between estimation and control.