In block-fading transmission, the discrete-time taps of time-varying channels can be generally modelled as a stationary random vector process with unknown and rank-deficient correlation matrix. The model applies to single/ multi-antenna systems for single/multi-carrier modulation. In this paper, we propose a hybrid estimator that sequentially estimates the (deterministic) rank-deficient channel correlation matrix and the (random) channel taps. In addition, we derive a bound on the corresponding estimation error by adapting the Cramer Rao bound (CRB) to the case of mixed deterministic and random parameters (Hybrid CRB). Finally, we assess that the proposed estimator is statistically efficient through comparison with the Hybrid CRB.